mir/bind
2
0
Fork 0
mirror of https://gitlab.isc.org/isc-projects/bind9 synced 2025-08-22 10:10:06 +00:00
Code Issues Releases Wiki Activity
bind/lib/dns/dispatch.c

1977 lines
46 KiB
C
Raw Normal View History

checkpoint
1999-06-16 01:32:31 +00:00
/*
Update license headers to not include years in copyright in all applicable files
2018-02-23 09:53:12 +01:00
* Copyright (C) Internet Systems Consortium, Inc. ("ISC")
Trailing whitespace trimmed. Perhaps running "perl util/spacewhack.pl in your own CVS tree will help minimize CVS conflicts. Maybe not. Blame Graff for getting me to trim all trailing whitespace.
2000-08-01 01:33:37 +00:00
*
4401. [misc] Change LICENSE to MPL 2.0.
2016-06-27 14:56:38 +10:00
* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
update all copyright headers to eliminate the typo
2020-09-14 16:20:40 -07:00
* file, you can obtain one at https://mozilla.org/MPL/2.0/.
Update license headers to not include years in copyright in all applicable files
2018-02-23 09:53:12 +01:00
*
* See the COPYRIGHT file distributed with this work for additional
* information regarding copyright ownership.
checkpoint
1999-06-16 01:32:31 +00:00
*/
1851. [doc] Doxygen comment markup. [RT #11398]
2005-04-27 04:57:32 +00:00
/*! \file */
add RCS id string
2000-06-22 22:00:42 +00:00
Replace custom isc_u?intNN_t types with C99 u?intNN_t types
2018-03-28 14:19:37 +02:00
#include <inttypes.h>
Replace custom isc_boolean_t with C standard bool type
2018-04-17 08:29:14 -07:00
#include <stdbool.h>
checkpoint
1999-06-16 01:32:31 +00:00
#include <stdlib.h>
Use the new sorting rules to regroup #include headers
2020-03-09 16:17:26 +01:00
#include <sys/types.h>
2203. [security] Query id generation was cryptographically weak. [RT # 16915]
2007-06-26 02:52:15 +00:00
#include <unistd.h>
checkpoint
1999-06-16 01:32:31 +00:00
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
#include <isc/atomic.h>
checkpoint
1999-06-18 02:01:42 +00:00
#include <isc/mem.h>
implement dns_dispatchmgr_t, replacing dns_dispatchlist_t. Use it throughout the library/server.
2000-05-10 21:34:50 +00:00
#include <isc/mutex.h>
Use system ephemeral ports for default portset In dispatch, use system ephemeral ports for default portset instead of hardcoded <1024,65535> range.
2021-10-04 14:11:57 +02:00
#include <isc/net.h>
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
#include <isc/netmgr.h>
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
#include <isc/portset.h>
Megacommit of dozens of files. Cleanup of redundant/useless header file inclusion. ISC style lint, primarily for function declarations and standalone comments -- ie, those that appear on a line without any code, which should be written as follows: /* * This is a comment. */
2000-05-08 14:38:29 +00:00
#include <isc/print.h>
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
#include <isc/random.h>
2537. [func] Added more statistics counters including those on socket I/O events and query RTT histograms. [RT #18802]
2009-01-27 22:30:00 +00:00
#include <isc/stats.h>
needs string.h for memset() prototype.
2000-05-13 21:57:02 +00:00
#include <isc/string.h>
replace gettimeofday with isc_time_now
2007-06-26 06:02:37 +00:00
#include <isc/time.h>
move util.h to <isc/util.h>
1999-12-16 22:24:22 +00:00
#include <isc/util.h>
checkpoint
1999-06-16 01:32:31 +00:00
539. [func] Support the blackhole option.
2000-11-03 02:45:55 +00:00
#include <dns/acl.h>
checkpoint
1999-06-16 01:32:31 +00:00
#include <dns/dispatch.h>
add dispatcher logging functions
2000-04-29 00:45:26 +00:00
#include <dns/log.h>
add dns_dispatch_attach/detach(), and start on serious dispatcher hacking.
1999-06-30 01:33:11 +00:00
#include <dns/message.h>
2355. [func] Extend the number statistics counters available. [RT #17590]
2008-04-03 05:55:52 +00:00
#include <dns/stats.h>
implement dns_dispatchmgr_t, replacing dns_dispatchlist_t. Use it throughout the library/server.
2000-05-10 21:34:50 +00:00
#include <dns/types.h>
Use clang-format to reformat the source files
2020-02-12 13:59:18 +01:00
typedef ISC_LIST(dns_dispentry_t) dns_displist_t;
469. [bug] "query-source address * port 53;" now works.
2000-09-18 04:50:05 +00:00
typedef struct dns_qid {
apply the modified style
2020-02-13 14:44:37 -08:00
unsigned int magic;
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
isc_mutex_t lock;
apply the modified style
2020-02-13 14:44:37 -08:00
unsigned int qid_nbuckets; /*%< hash table size */
unsigned int qid_increment; /*%< id increment on collision */
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
dns_displist_t *qid_table; /*%< the table itself */
469. [bug] "query-source address * port 53;" now works.
2000-09-18 04:50:05 +00:00
} dns_qid_t;
implement dns_dispatchmgr_t, replacing dns_dispatchlist_t. Use it throughout the library/server.
2000-05-10 21:34:50 +00:00
struct dns_dispatchmgr {
/* Unlocked. */
apply the modified style
2020-02-13 14:44:37 -08:00
unsigned int magic;
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
isc_refcount_t references;
apply the modified style
2020-02-13 14:44:37 -08:00
isc_mem_t *mctx;
dns_acl_t *blackhole;
isc_stats_t *stats;
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
isc_nm_t *nm;
implement dns_dispatchmgr_t, replacing dns_dispatchlist_t. Use it throughout the library/server.
2000-05-10 21:34:50 +00:00
/* Locked by "lock". */
apply the modified style
2020-02-13 14:44:37 -08:00
isc_mutex_t lock;
Use clang-format to reformat the source files
2020-02-12 13:59:18 +01:00
unsigned int state;
ISC_LIST(dns_dispatch_t) list;
Move the UDP buffer pool to the dispatchmgr. UDP parameters are now set by the first call to dns_dispatch_getudp(), subsecquent calls can raise the maximum number of buffers. dns_dispatchmgr_setudp() could be made public. Tidy dispatch_allocate(). entropy and mctx are now gone from dns_qid_t.
2000-09-19 06:59:28 +00:00
apply the modified style
2020-02-13 14:44:37 -08:00
dns_qid_t *qid;
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
apply the modified style
2020-02-13 14:44:37 -08:00
in_port_t *v4ports; /*%< available ports for IPv4 */
Use clang-format to reformat the source files
2020-02-12 13:59:18 +01:00
unsigned int nv4ports; /*%< # of available ports for IPv4 */
apply the modified style
2020-02-13 14:44:37 -08:00
in_port_t *v6ports; /*%< available ports for IPv4 */
Use clang-format to reformat the source files
2020-02-12 13:59:18 +01:00
unsigned int nv6ports; /*%< # of available ports for IPv4 */
implement dns_dispatchmgr_t, replacing dns_dispatchlist_t. Use it throughout the library/server.
2000-05-10 21:34:50 +00:00
};
apply the modified style
2020-02-13 14:44:37 -08:00
#define MGR_SHUTTINGDOWN 0x00000001U
Use clang-format to reformat the source files
2020-02-12 13:59:18 +01:00
#define MGR_IS_SHUTTINGDOWN(l) (((l)->state & MGR_SHUTTINGDOWN) != 0)
implement dns_dispatchmgr_t, replacing dns_dispatchlist_t. Use it throughout the library/server.
2000-05-10 21:34:50 +00:00
checkpoint
1999-07-06 19:32:40 +00:00
struct dns_dispentry {
apply the modified style
2020-02-13 14:44:37 -08:00
unsigned int magic;
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
isc_refcount_t references;
apply the modified style
2020-02-13 14:44:37 -08:00
dns_dispatch_t *disp;
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
isc_nmhandle_t *handle; /*%< netmgr handle for UDP connection */
apply the modified style
2020-02-13 14:44:37 -08:00
unsigned int bucket;
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
unsigned int retries;
dispatch: Enforce original timeout when calling _getnext() udp_recv() will call dispatch_getnext() if the message received is invalid or doesn't match; we need to reduce the timeout each time this happens so we can't be starved forever by someone sending garbage packets.
2021-10-01 12:53:31 -07:00
unsigned int timeout;
isc_time_t start;
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
isc_sockaddr_t local;
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
isc_sockaddr_t peer;
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
in_port_t port;
dns_messageid_t id;
dispatch: Remove 'timeout' callback - It is no longer necessary to pass a 'timeout' callback to dns_dispatch_addresponse(); timeouts are handled directly by the 'response' callback instead. - The netmgr handle is no longer passed to dispatch callbacks, since they don't (and can't) use it. instead, dispatch_cb_t now takes a result code, region, and argument. - Cleaned up timeout-related tests in dispatch_test.c
2021-08-03 15:27:06 +02:00
dispatch_cb_t connected;
dispatch_cb_t sent;
dispatch_cb_t response;
apply the modified style
2020-02-13 14:44:37 -08:00
void *arg;
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
bool canceled;
Use clang-format to reformat the source files
2020-02-12 13:59:18 +01:00
ISC_LINK(dns_dispentry_t) link;
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
ISC_LINK(dns_dispentry_t) alink;
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
ISC_LINK(dns_dispentry_t) plink;
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
ISC_LINK(dns_dispentry_t) rlink;
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
};
Simplify dns_dispatchmgr_create with fixed buffersize - UDP buffersize is now established when creating dispatch manager and is always set to 4096. - Set up the default port range in dispatchmgr before setting the magic number. - Magic is not set until dispatchmgr is fully created.
2020-12-09 15:45:13 -08:00
/*%
* Fixed UDP buffer size.
*/
#ifndef DNS_DISPATCH_UDPBUFSIZE
#define DNS_DISPATCH_UDPBUFSIZE 4096
#endif /* ifndef DNS_DISPATCH_UDPBUFSIZE */
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
typedef enum {
DNS_DISPATCHSTATE_NONE = 0UL,
DNS_DISPATCHSTATE_CONNECTING,
DNS_DISPATCHSTATE_CONNECTED
} dns_dispatchstate_t;
checkpoint
1999-06-16 01:32:31 +00:00
struct dns_dispatch {
/* Unlocked. */
apply the modified style
2020-02-13 14:44:37 -08:00
unsigned int magic; /*%< magic */
dns_dispatchmgr_t *mgr; /*%< dispatch manager */
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
isc_nmhandle_t *handle; /*%< netmgr handle for TCP connection */
isc_sockaddr_t local; /*%< local address */
in_port_t localport; /*%< local UDP port */
isc_sockaddr_t peer; /*%< peer address (TCP) */
Improve locking performance in dispatch.c 3316. [tuning] Improved locking performance when recursing. [RT #28836] - Use one lock per memory pool instead of associating them all with a single 'pool_lock' in the dispatch manager. - Reduce the critical sections for qid->lock in get_dispsocket(), deref_portentry() and dns_dispatch_addresponse2(). - Added a memory pool for socket events in dns_dispatch_t. - Add an isc_socketevent_t member in the resquery_t object, and use it with isc_socket_sendto2() instead of using isc_socket_sendto() - Tuned the memory pools in dispatch.c for better performance under load
2012-04-28 14:52:28 -07:00
1851. [doc] Doxygen comment markup. [RT #11398]
2005-04-27 04:57:32 +00:00
/*% Locked by mgr->lock. */
implement dns_dispatchmgr_t, replacing dns_dispatchlist_t. Use it throughout the library/server.
2000-05-10 21:34:50 +00:00
ISC_LINK(dns_dispatch_t) link;
/* Locked by "lock". */
apply the modified style
2020-02-13 14:44:37 -08:00
isc_mutex_t lock; /*%< locks all below */
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
isc_socktype_t socktype;
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
atomic_uint_fast32_t state;
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
isc_refcount_t references;
dispatch: Refactor to eliminate dns_dispatchevent - Responses received by the dispatch are no longer sent to the caller via a task event, but via a netmgr-style recv callback. the 'action' parameter to dns_dispatch_addresponse() is now called 'response' and is called directly from udp_recv() or tcp_recv() when a valid response has been received. - All references to isc_task and isc_taskmgr have been removed from dispatch functions. - All references to dns_dispatchevent_t have been removed and the type has been deleted. - Added a task to the resolver response context, to be used for fctx events. - When the caller cancels an operation, the response handler will be called with ISC_R_CANCELED; it can abort immediately since the caller will presumably have taken care of cleanup already. - Cleaned up attach/detach in resquery and request.
2021-07-26 20:23:18 -07:00
unsigned int shutdown_out : 1;
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
dns_displist_t pending;
dns_displist_t active;
apply the modified style
2020-02-13 14:44:37 -08:00
unsigned int nsockets;
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
apply the modified style
2020-02-13 14:44:37 -08:00
unsigned int requests; /*%< how many requests we have */
unsigned int tcpbuffers; /*%< allocated buffers */
checkpoint
1999-06-16 01:32:31 +00:00
};
apply the modified style
2020-02-13 14:44:37 -08:00
#define QID_MAGIC ISC_MAGIC('Q', 'i', 'd', ' ')
Use clang-format to reformat the source files
2020-02-12 13:59:18 +01:00
#define VALID_QID(e) ISC_MAGIC_VALID((e), QID_MAGIC)
469. [bug] "query-source address * port 53;" now works.
2000-09-18 04:50:05 +00:00
apply the modified style
2020-02-13 14:44:37 -08:00
#define RESPONSE_MAGIC ISC_MAGIC('D', 'r', 's', 'p')
Use clang-format to reformat the source files
2020-02-12 13:59:18 +01:00
#define VALID_RESPONSE(e) ISC_MAGIC_VALID((e), RESPONSE_MAGIC)
checkpoint
1999-06-18 02:01:42 +00:00
apply the modified style
2020-02-13 14:44:37 -08:00
#define DISPSOCK_MAGIC ISC_MAGIC('D', 's', 'o', 'c')
Use clang-format to reformat the source files
2020-02-12 13:59:18 +01:00
#define VALID_DISPSOCK(e) ISC_MAGIC_VALID((e), DISPSOCK_MAGIC)
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
apply the modified style
2020-02-13 14:44:37 -08:00
#define DISPATCH_MAGIC ISC_MAGIC('D', 'i', 's', 'p')
Use clang-format to reformat the source files
2020-02-12 13:59:18 +01:00
#define VALID_DISPATCH(e) ISC_MAGIC_VALID((e), DISPATCH_MAGIC)
checkpoint
1999-06-18 02:01:42 +00:00
Use clang-format to reformat the source files
2020-02-12 13:59:18 +01:00
#define DNS_DISPATCHMGR_MAGIC ISC_MAGIC('D', 'M', 'g', 'r')
apply the modified style
2020-02-13 14:44:37 -08:00
#define VALID_DISPATCHMGR(e) ISC_MAGIC_VALID((e), DNS_DISPATCHMGR_MAGIC)
checkpoint
1999-06-18 02:01:42 +00:00
Dispatch API simplification - Many dispatch attributes can be set implicitly instead of being passed in. we can infer whether to set DNS_DISPATCHATTR_TCP or _UDP from whether we're calling dns_dispatch_createtcp() or _createudp(). we can also infer DNS_DISPATCHATTR_IPV4 or _IPV6 from the addresses or the socket that were passed in. - We no longer use dup'd sockets in UDP dispatches, so the 'dup_socket' parameter has been removed from dns_dispatch_createudp(), along with the code implementing it. also removed isc_socket_dup() since it no longer has any callers. - The 'buffersize' parameter was ignored and has now been removed; buffersize is now fixed at 4096. - Maxbuffers and maxrequests don't need to be passed in on every call to dns_dispatch_createtcp() and _createudp(). In all current uses, the value for mgr->maxbuffers will either be raised once from its default of 20000 to 32768, or else left alone. (passing in a value lower than 20000 does not lower it.) there isn't enough difference between these values for there to be any need to configure this. The value for disp->maxrequests controls both the quota of concurrent requests for a dispatch and also the size of the dispatch socket memory pool. it's not clear that this quota is necessary at all. the memory pool size currently starts at 32768, but is sometimes lowered to 4096, which is definitely unnecessary. This commit sets both values permanently to 32768. - Previously TCP dispatches allocated their own separate QID table, which didn't incorporate a port table. this commit removes per-dispatch QID tables and shares the same table between all dispatches. since dispatches are created for each TCP socket, this may speed up the dispatch allocation process. there may be a slight increase in lock contention since all dispatches are sharing a single QID table, but since TCP sockets are used less often than UDP sockets (which were already sharing a QID table), it should not be a substantial change. - The dispatch port table was being used to determine whether a port was already in use; if so, then a UDP socket would be bound with REUSEADDR. this commit removes the port table, and always binds UDP sockets that way.
2020-12-17 00:43:00 -08:00
/*%
* Quota to control the number of UDP dispatch sockets. If a dispatch has
* more than the quota of sockets, new queries will purge oldest ones, so
* that a massive number of outstanding queries won't prevent subsequent
* queries (especially if the older ones take longer time and result in
* timeout).
*/
#ifndef DNS_DISPATCH_SOCKSQUOTA
#define DNS_DISPATCH_SOCKSQUOTA 3072
#endif /* ifndef DNS_DISPATCH_SOCKSQUOTA */
/*%
* Quota to control the number of concurrent requests that can be handled
* by each TCP dispatch. (UDP dispatches do not currently support socket
* sharing.)
*/
#ifndef DNS_DISPATCH_MAXREQUESTS
#define DNS_DISPATCH_MAXREQUESTS 32768
#endif /* ifndef DNS_DISPATCH_MAXREQUESTS */
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
/*%
Dispatch API simplification - Many dispatch attributes can be set implicitly instead of being passed in. we can infer whether to set DNS_DISPATCHATTR_TCP or _UDP from whether we're calling dns_dispatch_createtcp() or _createudp(). we can also infer DNS_DISPATCHATTR_IPV4 or _IPV6 from the addresses or the socket that were passed in. - We no longer use dup'd sockets in UDP dispatches, so the 'dup_socket' parameter has been removed from dns_dispatch_createudp(), along with the code implementing it. also removed isc_socket_dup() since it no longer has any callers. - The 'buffersize' parameter was ignored and has now been removed; buffersize is now fixed at 4096. - Maxbuffers and maxrequests don't need to be passed in on every call to dns_dispatch_createtcp() and _createudp(). In all current uses, the value for mgr->maxbuffers will either be raised once from its default of 20000 to 32768, or else left alone. (passing in a value lower than 20000 does not lower it.) there isn't enough difference between these values for there to be any need to configure this. The value for disp->maxrequests controls both the quota of concurrent requests for a dispatch and also the size of the dispatch socket memory pool. it's not clear that this quota is necessary at all. the memory pool size currently starts at 32768, but is sometimes lowered to 4096, which is definitely unnecessary. This commit sets both values permanently to 32768. - Previously TCP dispatches allocated their own separate QID table, which didn't incorporate a port table. this commit removes per-dispatch QID tables and shares the same table between all dispatches. since dispatches are created for each TCP socket, this may speed up the dispatch allocation process. there may be a slight increase in lock contention since all dispatches are sharing a single QID table, but since TCP sockets are used less often than UDP sockets (which were already sharing a QID table), it should not be a substantial change. - The dispatch port table was being used to determine whether a port was already in use; if so, then a UDP socket would be bound with REUSEADDR. this commit removes the port table, and always binds UDP sockets that way.
2020-12-17 00:43:00 -08:00
* Number of buckets in the QID hash table, and the value to
* increment the QID by when attempting to avoid collisions.
* The number of buckets should be prime, and the increment
* should be the next higher prime number.
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
*/
Dispatch API simplification - Many dispatch attributes can be set implicitly instead of being passed in. we can infer whether to set DNS_DISPATCHATTR_TCP or _UDP from whether we're calling dns_dispatch_createtcp() or _createudp(). we can also infer DNS_DISPATCHATTR_IPV4 or _IPV6 from the addresses or the socket that were passed in. - We no longer use dup'd sockets in UDP dispatches, so the 'dup_socket' parameter has been removed from dns_dispatch_createudp(), along with the code implementing it. also removed isc_socket_dup() since it no longer has any callers. - The 'buffersize' parameter was ignored and has now been removed; buffersize is now fixed at 4096. - Maxbuffers and maxrequests don't need to be passed in on every call to dns_dispatch_createtcp() and _createudp(). In all current uses, the value for mgr->maxbuffers will either be raised once from its default of 20000 to 32768, or else left alone. (passing in a value lower than 20000 does not lower it.) there isn't enough difference between these values for there to be any need to configure this. The value for disp->maxrequests controls both the quota of concurrent requests for a dispatch and also the size of the dispatch socket memory pool. it's not clear that this quota is necessary at all. the memory pool size currently starts at 32768, but is sometimes lowered to 4096, which is definitely unnecessary. This commit sets both values permanently to 32768. - Previously TCP dispatches allocated their own separate QID table, which didn't incorporate a port table. this commit removes per-dispatch QID tables and shares the same table between all dispatches. since dispatches are created for each TCP socket, this may speed up the dispatch allocation process. there may be a slight increase in lock contention since all dispatches are sharing a single QID table, but since TCP sockets are used less often than UDP sockets (which were already sharing a QID table), it should not be a substantial change. - The dispatch port table was being used to determine whether a port was already in use; if so, then a UDP socket would be bound with REUSEADDR. this commit removes the port table, and always binds UDP sockets that way.
2020-12-17 00:43:00 -08:00
#ifndef DNS_QID_BUCKETS
#define DNS_QID_BUCKETS 16411
#endif /* ifndef DNS_QID_BUCKETS */
#ifndef DNS_QID_INCREMENT
#define DNS_QID_INCREMENT 16433
#endif /* ifndef DNS_QID_INCREMENT */
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
add dns_dispatch_attach/detach(), and start on serious dispatcher hacking.
1999-06-30 01:33:11 +00:00
/*
needs string.h for memset() prototype.
2000-05-13 21:57:02 +00:00
* Statics.
add dns_dispatch_attach/detach(), and start on serious dispatcher hacking.
1999-06-30 01:33:11 +00:00
*/
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
static void
dispatchmgr_destroy(dns_dispatchmgr_t *mgr);
Reformat using the new rules
2020-02-14 08:14:03 +01:00
static dns_dispentry_t *
entry_search(dns_qid_t *, const isc_sockaddr_t *, dns_messageid_t, in_port_t,
unsigned int);
static void
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
udp_recv(isc_nmhandle_t *handle, isc_result_t eresult, isc_region_t *region,
void *arg);
Reformat using the new rules
2020-02-14 08:14:03 +01:00
static void
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
tcp_recv(isc_nmhandle_t *handle, isc_result_t eresult, isc_region_t *region,
void *arg);
Reformat using the new rules
2020-02-14 08:14:03 +01:00
static uint32_t
dns_hash(dns_qid_t *, const isc_sockaddr_t *, dns_messageid_t, in_port_t);
static void
dispatch_free(dns_dispatch_t **dispp);
static isc_result_t
dispatch: Refactor to eliminate dns_dispatchevent - Responses received by the dispatch are no longer sent to the caller via a task event, but via a netmgr-style recv callback. the 'action' parameter to dns_dispatch_addresponse() is now called 'response' and is called directly from udp_recv() or tcp_recv() when a valid response has been received. - All references to isc_task and isc_taskmgr have been removed from dispatch functions. - All references to dns_dispatchevent_t have been removed and the type has been deleted. - Added a task to the resolver response context, to be used for fctx events. - When the caller cancels an operation, the response handler will be called with ISC_R_CANCELED; it can abort immediately since the caller will presumably have taken care of cleanup already. - Cleaned up attach/detach in resquery and request.
2021-07-26 20:23:18 -07:00
dispatch_createudp(dns_dispatchmgr_t *mgr, const isc_sockaddr_t *localaddr,
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
dns_dispatch_t **dispp);
Dispatch API simplification - Many dispatch attributes can be set implicitly instead of being passed in. we can infer whether to set DNS_DISPATCHATTR_TCP or _UDP from whether we're calling dns_dispatch_createtcp() or _createudp(). we can also infer DNS_DISPATCHATTR_IPV4 or _IPV6 from the addresses or the socket that were passed in. - We no longer use dup'd sockets in UDP dispatches, so the 'dup_socket' parameter has been removed from dns_dispatch_createudp(), along with the code implementing it. also removed isc_socket_dup() since it no longer has any callers. - The 'buffersize' parameter was ignored and has now been removed; buffersize is now fixed at 4096. - Maxbuffers and maxrequests don't need to be passed in on every call to dns_dispatch_createtcp() and _createudp(). In all current uses, the value for mgr->maxbuffers will either be raised once from its default of 20000 to 32768, or else left alone. (passing in a value lower than 20000 does not lower it.) there isn't enough difference between these values for there to be any need to configure this. The value for disp->maxrequests controls both the quota of concurrent requests for a dispatch and also the size of the dispatch socket memory pool. it's not clear that this quota is necessary at all. the memory pool size currently starts at 32768, but is sometimes lowered to 4096, which is definitely unnecessary. This commit sets both values permanently to 32768. - Previously TCP dispatches allocated their own separate QID table, which didn't incorporate a port table. this commit removes per-dispatch QID tables and shares the same table between all dispatches. since dispatches are created for each TCP socket, this may speed up the dispatch allocation process. there may be a slight increase in lock contention since all dispatches are sharing a single QID table, but since TCP sockets are used less often than UDP sockets (which were already sharing a QID table), it should not be a substantial change. - The dispatch port table was being used to determine whether a port was already in use; if so, then a UDP socket would be bound with REUSEADDR. this commit removes the port table, and always binds UDP sockets that way.
2020-12-17 00:43:00 -08:00
static void
qid_allocate(dns_dispatchmgr_t *mgr, dns_qid_t **qidp);
Reformat using the new rules
2020-02-14 08:14:03 +01:00
static void
qid_destroy(isc_mem_t *mctx, dns_qid_t **qidp);
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
static void
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
startrecv(isc_nmhandle_t *handle, dns_dispatch_t *disp, dns_dispentry_t *resp);
void
dispatch_getnext(dns_dispatch_t *disp, dns_dispentry_t *resp, int32_t timeout);
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
#define LVL(x) ISC_LOG_DEBUG(x)
checkpoint
1999-07-08 02:50:00 +00:00
Reformat using the new rules
2020-02-14 08:14:03 +01:00
static void
mgr_log(dns_dispatchmgr_t *mgr, int level, const char *fmt, ...)
Use clang-format to reformat the source files
2020-02-12 13:59:18 +01:00
ISC_FORMAT_PRINTF(3, 4);
format string bugs and improved format string checking [RT #1578]
2001-08-08 22:54:55 +00:00
add dispatcher logging functions
2000-04-29 00:45:26 +00:00
static void
apply the modified style
2020-02-13 14:44:37 -08:00
mgr_log(dns_dispatchmgr_t *mgr, int level, const char *fmt, ...) {
char msgbuf[2048];
add dispatcher logging functions
2000-04-29 00:45:26 +00:00
va_list ap;
Use clang-tidy to add curly braces around one-line statements The command used to reformat the files in this commit was: ./util/run-clang-tidy \ -clang-tidy-binary clang-tidy-11 -clang-apply-replacements-binary clang-apply-replacements-11 \ -checks=-*,readability-braces-around-statements \ -j 9 \ -fix \ -format \ -style=file \ -quiet clang-format -i --style=format $(git ls-files '*.c' '*.h') uncrustify -c .uncrustify.cfg --replace --no-backup $(git ls-files '*.c' '*.h') clang-format -i --style=format $(git ls-files '*.c' '*.h')
2020-02-13 21:48:23 +01:00
if (!isc_log_wouldlog(dns_lctx, level)) {
Added calls to isc_log_wouldlog() in additional logging functions.
2000-07-17 23:25:35 +00:00
return;
Use clang-tidy to add curly braces around one-line statements The command used to reformat the files in this commit was: ./util/run-clang-tidy \ -clang-tidy-binary clang-tidy-11 -clang-apply-replacements-binary clang-apply-replacements-11 \ -checks=-*,readability-braces-around-statements \ -j 9 \ -fix \ -format \ -style=file \ -quiet clang-format -i --style=format $(git ls-files '*.c' '*.h') uncrustify -c .uncrustify.cfg --replace --no-backup $(git ls-files '*.c' '*.h') clang-format -i --style=format $(git ls-files '*.c' '*.h')
2020-02-13 21:48:23 +01:00
}
Added calls to isc_log_wouldlog() in additional logging functions.
2000-07-17 23:25:35 +00:00
add dispatcher logging functions
2000-04-29 00:45:26 +00:00
va_start(ap, fmt);
vsnprintf(msgbuf, sizeof(msgbuf), fmt, ap);
va_end(ap);
Use clang-format to reformat the source files
2020-02-12 13:59:18 +01:00
isc_log_write(dns_lctx, DNS_LOGCATEGORY_DISPATCH,
DNS_LOGMODULE_DISPATCH, level, "dispatchmgr %p: %s", mgr,
msgbuf);
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
}
2541. [bug] Conditionally update dispatch manager statistics. [RT #19247]
2009-01-31 00:10:24 +00:00
static inline void
apply the modified style
2020-02-13 14:44:37 -08:00
inc_stats(dns_dispatchmgr_t *mgr, isc_statscounter_t counter) {
Use clang-tidy to add curly braces around one-line statements The command used to reformat the files in this commit was: ./util/run-clang-tidy \ -clang-tidy-binary clang-tidy-11 -clang-apply-replacements-binary clang-apply-replacements-11 \ -checks=-*,readability-braces-around-statements \ -j 9 \ -fix \ -format \ -style=file \ -quiet clang-format -i --style=format $(git ls-files '*.c' '*.h') uncrustify -c .uncrustify.cfg --replace --no-backup $(git ls-files '*.c' '*.h') clang-format -i --style=format $(git ls-files '*.c' '*.h')
2020-02-13 21:48:23 +01:00
if (mgr->stats != NULL) {
2541. [bug] Conditionally update dispatch manager statistics. [RT #19247]
2009-01-31 00:10:24 +00:00
isc_stats_increment(mgr->stats, counter);
Use clang-tidy to add curly braces around one-line statements The command used to reformat the files in this commit was: ./util/run-clang-tidy \ -clang-tidy-binary clang-tidy-11 -clang-apply-replacements-binary clang-apply-replacements-11 \ -checks=-*,readability-braces-around-statements \ -j 9 \ -fix \ -format \ -style=file \ -quiet clang-format -i --style=format $(git ls-files '*.c' '*.h') uncrustify -c .uncrustify.cfg --replace --no-backup $(git ls-files '*.c' '*.h') clang-format -i --style=format $(git ls-files '*.c' '*.h')
2020-02-13 21:48:23 +01:00
}
2541. [bug] Conditionally update dispatch manager statistics. [RT #19247]
2009-01-31 00:10:24 +00:00
}
Merge statistics code (ATT SoW, rt24117) This includes the following changes: 3326. [func] Added task list statistics: task model, worker threads, quantum, tasks running, tasks ready. [RT #27678] 3325. [func] Report cache statistics: memory use, number of nodes, number of hash buckets, hit and miss counts. [RT #27056] 3324. [test] Add better tests for ADB stats [RT #27057] 3323. [func] Report the number of buckets the resolver is using. [RT #27020] 3322. [func] Monitor the number of active TCP and UDP dispatches. [RT #27055] 3321. [func] Monitor the number of recursive fetches and the number of open sockets, and report these values in the statistics channel. [RT #27054] 3320. [func] Added support for monitoring of recursing client count. [RT #27009] 3319. [func] Added support for monitoring of ADB entry count and hash size. [RT #27057]
2012-05-14 10:06:05 -07:00
static inline void
apply the modified style
2020-02-13 14:44:37 -08:00
dec_stats(dns_dispatchmgr_t *mgr, isc_statscounter_t counter) {
Use clang-tidy to add curly braces around one-line statements The command used to reformat the files in this commit was: ./util/run-clang-tidy \ -clang-tidy-binary clang-tidy-11 -clang-apply-replacements-binary clang-apply-replacements-11 \ -checks=-*,readability-braces-around-statements \ -j 9 \ -fix \ -format \ -style=file \ -quiet clang-format -i --style=format $(git ls-files '*.c' '*.h') uncrustify -c .uncrustify.cfg --replace --no-backup $(git ls-files '*.c' '*.h') clang-format -i --style=format $(git ls-files '*.c' '*.h')
2020-02-13 21:48:23 +01:00
if (mgr->stats != NULL) {
Merge statistics code (ATT SoW, rt24117) This includes the following changes: 3326. [func] Added task list statistics: task model, worker threads, quantum, tasks running, tasks ready. [RT #27678] 3325. [func] Report cache statistics: memory use, number of nodes, number of hash buckets, hit and miss counts. [RT #27056] 3324. [test] Add better tests for ADB stats [RT #27057] 3323. [func] Report the number of buckets the resolver is using. [RT #27020] 3322. [func] Monitor the number of active TCP and UDP dispatches. [RT #27055] 3321. [func] Monitor the number of recursive fetches and the number of open sockets, and report these values in the statistics channel. [RT #27054] 3320. [func] Added support for monitoring of recursing client count. [RT #27009] 3319. [func] Added support for monitoring of ADB entry count and hash size. [RT #27057]
2012-05-14 10:06:05 -07:00
isc_stats_decrement(mgr->stats, counter);
Use clang-tidy to add curly braces around one-line statements The command used to reformat the files in this commit was: ./util/run-clang-tidy \ -clang-tidy-binary clang-tidy-11 -clang-apply-replacements-binary clang-apply-replacements-11 \ -checks=-*,readability-braces-around-statements \ -j 9 \ -fix \ -format \ -style=file \ -quiet clang-format -i --style=format $(git ls-files '*.c' '*.h') uncrustify -c .uncrustify.cfg --replace --no-backup $(git ls-files '*.c' '*.h') clang-format -i --style=format $(git ls-files '*.c' '*.h')
2020-02-13 21:48:23 +01:00
}
Merge statistics code (ATT SoW, rt24117) This includes the following changes: 3326. [func] Added task list statistics: task model, worker threads, quantum, tasks running, tasks ready. [RT #27678] 3325. [func] Report cache statistics: memory use, number of nodes, number of hash buckets, hit and miss counts. [RT #27056] 3324. [test] Add better tests for ADB stats [RT #27057] 3323. [func] Report the number of buckets the resolver is using. [RT #27020] 3322. [func] Monitor the number of active TCP and UDP dispatches. [RT #27055] 3321. [func] Monitor the number of recursive fetches and the number of open sockets, and report these values in the statistics channel. [RT #27054] 3320. [func] Added support for monitoring of recursing client count. [RT #27009] 3319. [func] Added support for monitoring of ADB entry count and hash size. [RT #27057]
2012-05-14 10:06:05 -07:00
}
Reformat using the new rules
2020-02-14 08:14:03 +01:00
static void
dispatch_log(dns_dispatch_t *disp, int level, const char *fmt, ...)
Use clang-format to reformat the source files
2020-02-12 13:59:18 +01:00
ISC_FORMAT_PRINTF(3, 4);
format string bugs and improved format string checking [RT #1578]
2001-08-08 22:54:55 +00:00
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
static void
apply the modified style
2020-02-13 14:44:37 -08:00
dispatch_log(dns_dispatch_t *disp, int level, const char *fmt, ...) {
char msgbuf[2048];
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
va_list ap;
Use clang-tidy to add curly braces around one-line statements The command used to reformat the files in this commit was: ./util/run-clang-tidy \ -clang-tidy-binary clang-tidy-11 -clang-apply-replacements-binary clang-apply-replacements-11 \ -checks=-*,readability-braces-around-statements \ -j 9 \ -fix \ -format \ -style=file \ -quiet clang-format -i --style=format $(git ls-files '*.c' '*.h') uncrustify -c .uncrustify.cfg --replace --no-backup $(git ls-files '*.c' '*.h') clang-format -i --style=format $(git ls-files '*.c' '*.h')
2020-02-13 21:48:23 +01:00
if (!isc_log_wouldlog(dns_lctx, level)) {
use isc_log_wouldlog()
2000-07-13 01:16:22 +00:00
return;
Use clang-tidy to add curly braces around one-line statements The command used to reformat the files in this commit was: ./util/run-clang-tidy \ -clang-tidy-binary clang-tidy-11 -clang-apply-replacements-binary clang-apply-replacements-11 \ -checks=-*,readability-braces-around-statements \ -j 9 \ -fix \ -format \ -style=file \ -quiet clang-format -i --style=format $(git ls-files '*.c' '*.h') uncrustify -c .uncrustify.cfg --replace --no-backup $(git ls-files '*.c' '*.h') clang-format -i --style=format $(git ls-files '*.c' '*.h')
2020-02-13 21:48:23 +01:00
}
Trailing whitespace trimmed. Perhaps running "perl util/spacewhack.pl in your own CVS tree will help minimize CVS conflicts. Maybe not. Blame Graff for getting me to trim all trailing whitespace.
2000-08-01 01:33:37 +00:00
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
va_start(ap, fmt);
vsnprintf(msgbuf, sizeof(msgbuf), fmt, ap);
va_end(ap);
Use clang-format to reformat the source files
2020-02-12 13:59:18 +01:00
isc_log_write(dns_lctx, DNS_LOGCATEGORY_DISPATCH,
DNS_LOGMODULE_DISPATCH, level, "dispatch %p: %s", disp,
msgbuf);
add dispatcher logging functions
2000-04-29 00:45:26 +00:00
}
Start moving the dispatcher code to allow it to be used for wire format as well as for lightweight resolver daemon stuff.
2000-01-07 01:17:47 +00:00
/*
* Return a hash of the destination and message id.
*/
Replace custom isc_u?intNN_t types with C99 u?intNN_t types
2018-03-28 14:19:37 +02:00
static uint32_t
4546. [func] Extend the use of const declarations. [RT #43379]
2016-12-30 15:45:08 +11:00
dns_hash(dns_qid_t *qid, const isc_sockaddr_t *dest, dns_messageid_t id,
apply the modified style
2020-02-13 14:44:37 -08:00
in_port_t port) {
Replace custom isc_u?intNN_t types with C99 u?intNN_t types
2018-03-28 14:19:37 +02:00
uint32_t ret;
Start moving the dispatcher code to allow it to be used for wire format as well as for lightweight resolver daemon stuff.
2000-01-07 01:17:47 +00:00
Replace custom isc_boolean_t with C standard bool type
2018-04-17 08:29:14 -07:00
ret = isc_sockaddr_hash(dest, true);
Replace custom isc_u?intNN_t types with C99 u?intNN_t types
2018-03-28 14:19:37 +02:00
ret ^= ((uint32_t)id << 16) | port;
469. [bug] "query-source address * port 53;" now works.
2000-09-18 04:50:05 +00:00
ret %= qid->qid_nbuckets;
Start moving the dispatcher code to allow it to be used for wire format as well as for lightweight resolver daemon stuff.
2000-01-07 01:17:47 +00:00
469. [bug] "query-source address * port 53;" now works.
2000-09-18 04:50:05 +00:00
INSIST(ret < qid->qid_nbuckets);
Start moving the dispatcher code to allow it to be used for wire format as well as for lightweight resolver daemon stuff.
2000-01-07 01:17:47 +00:00
return (ret);
}
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
/*%
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
* Choose a random port number for a dispatch entry.
Improve locking performance in dispatch.c 3316. [tuning] Improved locking performance when recursing. [RT #28836] - Use one lock per memory pool instead of associating them all with a single 'pool_lock' in the dispatch manager. - Reduce the critical sections for qid->lock in get_dispsocket(), deref_portentry() and dns_dispatch_addresponse2(). - Added a memory pool for socket events in dns_dispatch_t. - Add an isc_socketevent_t member in the resquery_t object, and use it with isc_socket_sendto2() instead of using isc_socket_sendto() - Tuned the memory pools in dispatch.c for better performance under load
2012-04-28 14:52:28 -07:00
* The caller must hold the disp->lock
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
*/
static isc_result_t
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
setup_socket(dns_dispatch_t *disp, dns_dispentry_t *resp,
const isc_sockaddr_t *dest, in_port_t *portp) {
apply the modified style
2020-02-13 14:44:37 -08:00
dns_dispatchmgr_t *mgr = disp->mgr;
unsigned int nports;
General code refactoring - style cleanup - removed NULL checks in places where they are not currently needed - use isc_refcount for dispatch reference counting - revised code flow for readability - remove some #ifdefs that are no longer relevant - remove unused struct members - removed unnecessary function parameters - use C99 struct initialization
2020-12-09 19:44:41 -08:00
in_port_t *ports = NULL;
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
in_port_t port;
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
if (resp->retries++ > 5) {
return (ISC_R_FAILURE);
}
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
if (isc_sockaddr_pf(&disp->local) == AF_INET) {
General code refactoring - style cleanup - removed NULL checks in places where they are not currently needed - use isc_refcount for dispatch reference counting - revised code flow for readability - remove some #ifdefs that are no longer relevant - remove unused struct members - removed unnecessary function parameters - use C99 struct initialization
2020-12-09 19:44:41 -08:00
nports = mgr->nv4ports;
ports = mgr->v4ports;
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
} else {
General code refactoring - style cleanup - removed NULL checks in places where they are not currently needed - use isc_refcount for dispatch reference counting - revised code flow for readability - remove some #ifdefs that are no longer relevant - remove unused struct members - removed unnecessary function parameters - use C99 struct initialization
2020-12-09 19:44:41 -08:00
nports = mgr->nv6ports;
ports = mgr->v6ports;
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
}
Use clang-tidy to add curly braces around one-line statements The command used to reformat the files in this commit was: ./util/run-clang-tidy \ -clang-tidy-binary clang-tidy-11 -clang-apply-replacements-binary clang-apply-replacements-11 \ -checks=-*,readability-braces-around-statements \ -j 9 \ -fix \ -format \ -style=file \ -quiet clang-format -i --style=format $(git ls-files '*.c' '*.h') uncrustify -c .uncrustify.cfg --replace --no-backup $(git ls-files '*.c' '*.h') clang-format -i --style=format $(git ls-files '*.c' '*.h')
2020-02-13 21:48:23 +01:00
if (nports == 0) {
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
return (ISC_R_ADDRNOTAVAIL);
Use clang-tidy to add curly braces around one-line statements The command used to reformat the files in this commit was: ./util/run-clang-tidy \ -clang-tidy-binary clang-tidy-11 -clang-apply-replacements-binary clang-apply-replacements-11 \ -checks=-*,readability-braces-around-statements \ -j 9 \ -fix \ -format \ -style=file \ -quiet clang-format -i --style=format $(git ls-files '*.c' '*.h') uncrustify -c .uncrustify.cfg --replace --no-backup $(git ls-files '*.c' '*.h') clang-format -i --style=format $(git ls-files '*.c' '*.h')
2020-02-13 21:48:23 +01:00
}
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
disp->nsockets++;
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
resp->local = disp->local;
resp->peer = *dest;
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
port = ports[isc_random_uniform(nports)];
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
isc_sockaddr_setport(&resp->local, port);
resp->port = port;
Dispatch API simplification - Many dispatch attributes can be set implicitly instead of being passed in. we can infer whether to set DNS_DISPATCHATTR_TCP or _UDP from whether we're calling dns_dispatch_createtcp() or _createudp(). we can also infer DNS_DISPATCHATTR_IPV4 or _IPV6 from the addresses or the socket that were passed in. - We no longer use dup'd sockets in UDP dispatches, so the 'dup_socket' parameter has been removed from dns_dispatch_createudp(), along with the code implementing it. also removed isc_socket_dup() since it no longer has any callers. - The 'buffersize' parameter was ignored and has now been removed; buffersize is now fixed at 4096. - Maxbuffers and maxrequests don't need to be passed in on every call to dns_dispatch_createtcp() and _createudp(). In all current uses, the value for mgr->maxbuffers will either be raised once from its default of 20000 to 32768, or else left alone. (passing in a value lower than 20000 does not lower it.) there isn't enough difference between these values for there to be any need to configure this. The value for disp->maxrequests controls both the quota of concurrent requests for a dispatch and also the size of the dispatch socket memory pool. it's not clear that this quota is necessary at all. the memory pool size currently starts at 32768, but is sometimes lowered to 4096, which is definitely unnecessary. This commit sets both values permanently to 32768. - Previously TCP dispatches allocated their own separate QID table, which didn't incorporate a port table. this commit removes per-dispatch QID tables and shares the same table between all dispatches. since dispatches are created for each TCP socket, this may speed up the dispatch allocation process. there may be a slight increase in lock contention since all dispatches are sharing a single QID table, but since TCP sockets are used less often than UDP sockets (which were already sharing a QID table), it should not be a substantial change. - The dispatch port table was being used to determine whether a port was already in use; if so, then a UDP socket would be bound with REUSEADDR. this commit removes the port table, and always binds UDP sockets that way.
2020-12-17 00:43:00 -08:00
Remove support for shared UDP dispatch sockets Currently the netmgr doesn't support unconnected, shared UDP sockets, so there's no reason to retain that functionality in the dispatcher prior to porting to the netmgr. In this commit, the DNS_DISPATCHATTR_EXCLUSIVE attribute has been removed as it is now non-optional; UDP dispatches are alwasy exclusive. Code implementing non-exclusive UDP dispatches has been removed. dns_dispatch_getentrysocket() now always returns the dispsocket for UDP dispatches and the dispatch socket for TCP dispatches. There is no longer any need to search for existing dispatches from dns_dispatch_getudp(), so the 'mask' option has been removed, and the function renamed to the more descriptive dns_dispatch_createudp().
2020-12-16 01:32:06 -08:00
*portp = port;
return (ISC_R_SUCCESS);
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
}
/*%
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
* Deactivate the socket for a dispatch entry.
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
* The dispatch must be locked.
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
*/
static void
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
deactivate_dispentry(dns_dispatch_t *disp, dns_dispentry_t *resp) {
if (ISC_LINK_LINKED(resp, alink)) {
ISC_LIST_UNLINK(disp->active, resp, alink);
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
}
dispatch: Refactor to eliminate dns_dispatchevent - Responses received by the dispatch are no longer sent to the caller via a task event, but via a netmgr-style recv callback. the 'action' parameter to dns_dispatch_addresponse() is now called 'response' and is called directly from udp_recv() or tcp_recv() when a valid response has been received. - All references to isc_task and isc_taskmgr have been removed from dispatch functions. - All references to dns_dispatchevent_t have been removed and the type has been deleted. - Added a task to the resolver response context, to be used for fctx events. - When the caller cancels an operation, the response handler will be called with ISC_R_CANCELED; it can abort immediately since the caller will presumably have taken care of cleanup already. - Cleaned up attach/detach in resquery and request.
2021-07-26 20:23:18 -07:00
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
if (resp->handle != NULL) {
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
INSIST(disp->socktype == isc_socktype_udp);
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
isc_nm_cancelread(resp->handle);
isc_nmhandle_detach(&resp->handle);
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
}
dispatch: Refactor to eliminate dns_dispatchevent - Responses received by the dispatch are no longer sent to the caller via a task event, but via a netmgr-style recv callback. the 'action' parameter to dns_dispatch_addresponse() is now called 'response' and is called directly from udp_recv() or tcp_recv() when a valid response has been received. - All references to isc_task and isc_taskmgr have been removed from dispatch functions. - All references to dns_dispatchevent_t have been removed and the type has been deleted. - Added a task to the resolver response context, to be used for fctx events. - When the caller cancels an operation, the response handler will be called with ISC_R_CANCELED; it can abort immediately since the caller will presumably have taken care of cleanup already. - Cleaned up attach/detach in resquery and request.
2021-07-26 20:23:18 -07:00
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
disp->nsockets--;
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
}
add dns_dispatch_attach/detach(), and start on serious dispatcher hacking.
1999-06-30 01:33:11 +00:00
commented and removed dead code
2001-01-25 13:47:59 +00:00
/*
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
* Find an entry for query ID 'id', socket address 'dest', and port number
2417. [bug] Connecting UDP sockets for outgoing queries could unexpectedly fail with an 'address already in use' error. [RT #18411]
2008-08-15 17:29:52 +00:00
* 'port'.
commented and removed dead code
2001-01-25 13:47:59 +00:00
* Return NULL if no such entry exists.
*/
checkpoint
1999-07-06 19:32:40 +00:00
static dns_dispentry_t *
4546. [func] Extend the use of const declarations. [RT #43379]
2016-12-30 15:45:08 +11:00
entry_search(dns_qid_t *qid, const isc_sockaddr_t *dest, dns_messageid_t id,
apply the modified style
2020-02-13 14:44:37 -08:00
in_port_t port, unsigned int bucket) {
General code refactoring - style cleanup - removed NULL checks in places where they are not currently needed - use isc_refcount for dispatch reference counting - revised code flow for readability - remove some #ifdefs that are no longer relevant - remove unused struct members - removed unnecessary function parameters - use C99 struct initialization
2020-12-09 19:44:41 -08:00
dns_dispentry_t *res = NULL;
add dns_dispatch_attach/detach(), and start on serious dispatcher hacking.
1999-06-30 01:33:11 +00:00
[master] dispatch.c race 3695. [bug] Address a possible race in dispatch.c. [RT #35107]
2013-12-23 09:50:18 -08:00
REQUIRE(VALID_QID(qid));
469. [bug] "query-source address * port 53;" now works.
2000-09-18 04:50:05 +00:00
REQUIRE(bucket < qid->qid_nbuckets);
add another REQUIRE() where it might help
1999-07-09 20:34:26 +00:00
2417. [bug] Connecting UDP sockets for outgoing queries could unexpectedly fail with an 'address already in use' error. [RT #18411]
2008-08-15 17:29:52 +00:00
res = ISC_LIST_HEAD(qid->qid_table[bucket]);
add dns_dispatch_attach/detach(), and start on serious dispatcher hacking.
1999-06-30 01:33:11 +00:00
while (res != NULL) {
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
if (res->id == id && isc_sockaddr_equal(dest, &res->peer) &&
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
res->port == port) {
add dns_dispatch_attach/detach(), and start on serious dispatcher hacking.
1999-06-30 01:33:11 +00:00
return (res);
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
}
add dns_dispatch_attach/detach(), and start on serious dispatcher hacking.
1999-06-30 01:33:11 +00:00
res = ISC_LIST_NEXT(res, link);
}
return (NULL);
}
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
static void
Remove DISPATH_TRACE, RESOLVER_TRACE and REQ_TRACE logging Remove the debugging printfs. (leaving this as a separate commit rather than squashing it so we can restore it in the future if we ever need it again.)
2021-09-09 15:25:08 -07:00
dispentry_attach(dns_dispentry_t *resp, dns_dispentry_t **respp) {
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
REQUIRE(VALID_RESPONSE(resp));
REQUIRE(respp != NULL && *respp == NULL);
Remove DISPATH_TRACE, RESOLVER_TRACE and REQ_TRACE logging Remove the debugging printfs. (leaving this as a separate commit rather than squashing it so we can restore it in the future if we ever need it again.)
2021-09-09 15:25:08 -07:00
isc_refcount_increment(&resp->references);
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
*respp = resp;
}
static void
Remove DISPATH_TRACE, RESOLVER_TRACE and REQ_TRACE logging Remove the debugging printfs. (leaving this as a separate commit rather than squashing it so we can restore it in the future if we ever need it again.)
2021-09-09 15:25:08 -07:00
dispentry_destroy(dns_dispentry_t *resp) {
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
dns_dispatch_t *disp = resp->disp;
resp->magic = 0;
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
if (ISC_LINK_LINKED(resp, plink)) {
ISC_LIST_UNLINK(disp->pending, resp, plink);
}
INSIST(!ISC_LINK_LINKED(resp, alink));
INSIST(!ISC_LINK_LINKED(resp, rlink));
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
if (resp->handle != NULL) {
isc_nmhandle_detach(&resp->handle);
}
isc_refcount_destroy(&resp->references);
isc_mem_put(disp->mgr->mctx, resp, sizeof(*resp));
dns_dispatch_detach(&disp);
}
static void
Remove DISPATH_TRACE, RESOLVER_TRACE and REQ_TRACE logging Remove the debugging printfs. (leaving this as a separate commit rather than squashing it so we can restore it in the future if we ever need it again.)
2021-09-09 15:25:08 -07:00
dispentry_detach(dns_dispentry_t **respp) {
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
dns_dispentry_t *resp = NULL;
uint_fast32_t ref;
REQUIRE(respp != NULL && VALID_RESPONSE(*respp));
resp = *respp;
*respp = NULL;
ref = isc_refcount_decrement(&resp->references);
if (ref == 1) {
Remove DISPATH_TRACE, RESOLVER_TRACE and REQ_TRACE logging Remove the debugging printfs. (leaving this as a separate commit rather than squashing it so we can restore it in the future if we ever need it again.)
2021-09-09 15:25:08 -07:00
dispentry_destroy(resp);
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
}
}
dispatch: Enforce original timeout when calling _getnext() udp_recv() will call dispatch_getnext() if the message received is invalid or doesn't match; we need to reduce the timeout each time this happens so we can't be starved forever by someone sending garbage packets.
2021-10-01 12:53:31 -07:00
/*
* How long in milliseconds has it been since this dispentry
* started reading? (Only used for UDP, to adjust the timeout
* downward when running getnext.)
*/
static unsigned int
dispentry_runtime(dns_dispentry_t *resp) {
isc_time_t now;
if (isc_time_isepoch(&resp->start)) {
return (0);
}
TIME_NOW(&now);
return (isc_time_microdiff(&now, &resp->start) / 1000);
}
checkpoint
1999-07-08 02:50:00 +00:00
/*
* General flow:
*
don't retry UDP receives after they return an error - an error return from the socket library means the library has already decided that the error is not retryable and likely to throw us into an infinite loop if retried
2001-05-14 23:10:19 +00:00
* If I/O result == CANCELED or error, free the buffer.
checkpoint
1999-07-08 02:50:00 +00:00
*
fixed a bug that named crashes with an assertion failure on exit when sharing the same port for listening and querying, and changing listening addresses several times. [RT# 3509] additionally, + limited the canceled socket tasks in dispatch.c + made dns_dispatch_changeattributes() care about the NOLISTEN mask + described side effects of dns_dispatch_changeattributes() in its description comment
2002-09-04 02:26:13 +00:00
* If query, free the buffer, restart.
checkpoint
1999-07-08 02:50:00 +00:00
*
* If response:
* Allocate event, fill in details.
* If cannot allocate, free buffer, restart.
* find target. If not found, free buffer, restart.
* if event queue is not empty, queue. else, send.
* restart.
*/
add dns_dispatch_attach/detach(), and start on serious dispatcher hacking.
1999-06-30 01:33:11 +00:00
static void
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
udp_recv(isc_nmhandle_t *handle, isc_result_t eresult, isc_region_t *region,
void *arg) {
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
dns_dispentry_t *resp = (dns_dispentry_t *)arg;
Remove support for shared UDP dispatch sockets Currently the netmgr doesn't support unconnected, shared UDP sockets, so there's no reason to retain that functionality in the dispatcher prior to porting to the netmgr. In this commit, the DNS_DISPATCHATTR_EXCLUSIVE attribute has been removed as it is now non-optional; UDP dispatches are alwasy exclusive. Code implementing non-exclusive UDP dispatches has been removed. dns_dispatch_getentrysocket() now always returns the dispsocket for UDP dispatches and the dispatch socket for TCP dispatches. There is no longer any need to search for existing dispatches from dns_dispatch_getudp(), so the 'mask' option has been removed, and the function renamed to the more descriptive dns_dispatch_createudp().
2020-12-16 01:32:06 -08:00
dns_dispatch_t *disp = NULL;
apply the modified style
2020-02-13 14:44:37 -08:00
dns_messageid_t id;
isc_result_t dres;
isc_buffer_t source;
unsigned int flags;
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
isc_sockaddr_t peer;
apply the modified style
2020-02-13 14:44:37 -08:00
isc_netaddr_t netaddr;
dispatch: Enforce original timeout when calling _getnext() udp_recv() will call dispatch_getnext() if the message received is invalid or doesn't match; we need to reduce the timeout each time this happens so we can't be starved forever by someone sending garbage packets.
2021-10-01 12:53:31 -07:00
int match, timeout;
dispatch: Remove 'timeout' callback - It is no longer necessary to pass a 'timeout' callback to dns_dispatch_addresponse(); timeouts are handled directly by the 'response' callback instead. - The netmgr handle is no longer passed to dispatch callbacks, since they don't (and can't) use it. instead, dispatch_cb_t now takes a result code, region, and argument. - Cleaned up timeout-related tests in dispatch_test.c
2021-08-03 15:27:06 +02:00
dispatch_cb_t response = NULL;
Remove support for shared UDP dispatch sockets Currently the netmgr doesn't support unconnected, shared UDP sockets, so there's no reason to retain that functionality in the dispatcher prior to porting to the netmgr. In this commit, the DNS_DISPATCHATTR_EXCLUSIVE attribute has been removed as it is now non-optional; UDP dispatches are alwasy exclusive. Code implementing non-exclusive UDP dispatches has been removed. dns_dispatch_getentrysocket() now always returns the dispsocket for UDP dispatches and the dispatch socket for TCP dispatches. There is no longer any need to search for existing dispatches from dns_dispatch_getudp(), so the 'mask' option has been removed, and the function renamed to the more descriptive dns_dispatch_createudp().
2020-12-16 01:32:06 -08:00
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
REQUIRE(VALID_RESPONSE(resp));
dispatch: Refactor to eliminate dns_dispatchevent - Responses received by the dispatch are no longer sent to the caller via a task event, but via a netmgr-style recv callback. the 'action' parameter to dns_dispatch_addresponse() is now called 'response' and is called directly from udp_recv() or tcp_recv() when a valid response has been received. - All references to isc_task and isc_taskmgr have been removed from dispatch functions. - All references to dns_dispatchevent_t have been removed and the type has been deleted. - Added a task to the resolver response context, to be used for fctx events. - When the caller cancels an operation, the response handler will be called with ISC_R_CANCELED; it can abort immediately since the caller will presumably have taken care of cleanup already. - Cleaned up attach/detach in resquery and request.
2021-07-26 20:23:18 -07:00
REQUIRE(VALID_DISPATCH(resp->disp));
Remove support for shared UDP dispatch sockets Currently the netmgr doesn't support unconnected, shared UDP sockets, so there's no reason to retain that functionality in the dispatcher prior to porting to the netmgr. In this commit, the DNS_DISPATCHATTR_EXCLUSIVE attribute has been removed as it is now non-optional; UDP dispatches are alwasy exclusive. Code implementing non-exclusive UDP dispatches has been removed. dns_dispatch_getentrysocket() now always returns the dispsocket for UDP dispatches and the dispatch socket for TCP dispatches. There is no longer any need to search for existing dispatches from dns_dispatch_getudp(), so the 'mask' option has been removed, and the function renamed to the more descriptive dns_dispatch_createudp().
2020-12-16 01:32:06 -08:00
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
disp = resp->disp;
add dns_dispatch_attach/detach(), and start on serious dispatcher hacking.
1999-06-30 01:33:11 +00:00
LOCK(&disp->lock);
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
dispatch_log(disp, LVL(90), "UDP response %p:%s:requests %d", resp,
isc_result_totext(eresult), disp->requests);
/*
* The resp may have been deactivated by shutdown; if
* so, we can skip the response callback.
*/
if (ISC_LINK_LINKED(resp, alink)) {
response = resp->response;
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
}
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
if (eresult != ISC_R_SUCCESS) {
add missing exit checks to udp_recv and tcp_recv
2000-02-02 23:29:47 +00:00
/*
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
* This is most likely a network error on a connected
* socket, a timeout, or the query has been canceled.
* It makes no sense to check the address or parse the
* packet, but we can return the error to the caller.
add missing exit checks to udp_recv and tcp_recv
2000-02-02 23:29:47 +00:00
*/
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
goto done;
add missing exit checks to udp_recv and tcp_recv
2000-02-02 23:29:47 +00:00
}
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
INSIST(ISC_LINK_LINKED(resp, alink));
Remove support for shared UDP dispatch sockets Currently the netmgr doesn't support unconnected, shared UDP sockets, so there's no reason to retain that functionality in the dispatcher prior to porting to the netmgr. In this commit, the DNS_DISPATCHATTR_EXCLUSIVE attribute has been removed as it is now non-optional; UDP dispatches are alwasy exclusive. Code implementing non-exclusive UDP dispatches has been removed. dns_dispatch_getentrysocket() now always returns the dispsocket for UDP dispatches and the dispatch socket for TCP dispatches. There is no longer any need to search for existing dispatches from dns_dispatch_getudp(), so the 'mask' option has been removed, and the function renamed to the more descriptive dns_dispatch_createudp().
2020-12-16 01:32:06 -08:00
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
peer = isc_nmhandle_peeraddr(handle);
isc_netaddr_fromsockaddr(&netaddr, &peer);
539. [func] Support the blackhole option.
2000-11-03 02:45:55 +00:00
/*
* If this is from a blackholed address, drop it.
*/
there is no need for each dispatch to have a separate reference to the blackhole list; they can use the blackhole list of the dispatch manager directly. This eliminates some unsafe unlocked accesses to the reference count of the blackhole list dns_acl_t object
2001-02-08 18:25:09 +00:00
if (disp->mgr->blackhole != NULL &&
Use clang-format to reformat the source files
2020-02-12 13:59:18 +01:00
dns_acl_match(&netaddr, NULL, disp->mgr->blackhole, NULL, &match,
NULL) == ISC_R_SUCCESS &&
apply the modified style
2020-02-13 14:44:37 -08:00
match > 0)
{
539. [func] Support the blackhole option.
2000-11-03 02:45:55 +00:00
if (isc_log_wouldlog(dns_lctx, LVL(10))) {
char netaddrstr[ISC_NETADDR_FORMATSIZE];
isc_netaddr_format(&netaddr, netaddrstr,
sizeof(netaddrstr));
Use clang-format to reformat the source files
2020-02-12 13:59:18 +01:00
dispatch_log(disp, LVL(10), "blackholed packet from %s",
539. [func] Support the blackhole option.
2000-11-03 02:45:55 +00:00
netaddrstr);
}
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
goto next;
539. [func] Support the blackhole option.
2000-11-03 02:45:55 +00:00
}
add dns_dispatch_attach/detach(), and start on serious dispatcher hacking.
1999-06-30 01:33:11 +00:00
/*
* Peek into the buffer to see what we can see.
*/
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
id = resp->id;
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
isc_buffer_init(&source, region->base, region->length);
isc_buffer_add(&source, region->length);
add dns_dispatch_attach/detach(), and start on serious dispatcher hacking.
1999-06-30 01:33:11 +00:00
dres = dns_message_peekheader(&source, &id, &flags);
s/DNS_R_/ISC_R_/ change for some codes.
2000-04-06 22:03:35 +00:00
if (dres != ISC_R_SUCCESS) {
do not use initial caps in log messages
2000-06-16 15:51:58 +00:00
dispatch_log(disp, LVL(10), "got garbage packet");
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
goto next;
add dns_dispatch_attach/detach(), and start on serious dispatcher hacking.
1999-06-30 01:33:11 +00:00
}
add dispatcher logging functions
2000-04-29 00:45:26 +00:00
dispatch_log(disp, LVL(92),
do not use initial caps in log messages
2000-06-16 15:51:58 +00:00
"got valid DNS message header, /QR %c, id %u",
Turn (int & flag) into (int & flag) != 0 when implicitly typed to bool
2018-10-11 11:57:57 +02:00
(((flags & DNS_MESSAGEFLAG_QR) != 0) ? '1' : '0'), id);
checkpoint
1999-07-08 22:12:37 +00:00
add dns_dispatch_attach/detach(), and start on serious dispatcher hacking.
1999-06-30 01:33:11 +00:00
/*
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
* Look at the message flags. If it's a query, ignore it.
add dns_dispatch_attach/detach(), and start on serious dispatcher hacking.
1999-06-30 01:33:11 +00:00
*/
checkpoint
1999-07-06 19:32:40 +00:00
if ((flags & DNS_MESSAGEFLAG_QR) == 0) {
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
goto next;
1687. [bug] Race condition in dispatch. [RT #10272]
2004-07-21 00:48:19 +00:00
}
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
/*
Remove support for shared UDP dispatch sockets Currently the netmgr doesn't support unconnected, shared UDP sockets, so there's no reason to retain that functionality in the dispatcher prior to porting to the netmgr. In this commit, the DNS_DISPATCHATTR_EXCLUSIVE attribute has been removed as it is now non-optional; UDP dispatches are alwasy exclusive. Code implementing non-exclusive UDP dispatches has been removed. dns_dispatch_getentrysocket() now always returns the dispsocket for UDP dispatches and the dispatch socket for TCP dispatches. There is no longer any need to search for existing dispatches from dns_dispatch_getudp(), so the 'mask' option has been removed, and the function renamed to the more descriptive dns_dispatch_createudp().
2020-12-16 01:32:06 -08:00
* The QID and the address must match the expected ones.
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
*/
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
if (resp->id != id || !isc_sockaddr_equal(&peer, &resp->peer)) {
Remove support for shared UDP dispatch sockets Currently the netmgr doesn't support unconnected, shared UDP sockets, so there's no reason to retain that functionality in the dispatcher prior to porting to the netmgr. In this commit, the DNS_DISPATCHATTR_EXCLUSIVE attribute has been removed as it is now non-optional; UDP dispatches are alwasy exclusive. Code implementing non-exclusive UDP dispatches has been removed. dns_dispatch_getentrysocket() now always returns the dispsocket for UDP dispatches and the dispatch socket for TCP dispatches. There is no longer any need to search for existing dispatches from dns_dispatch_getudp(), so the 'mask' option has been removed, and the function renamed to the more descriptive dns_dispatch_createudp().
2020-12-16 01:32:06 -08:00
dispatch_log(disp, LVL(90), "response doesn't match");
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
inc_stats(disp->mgr, dns_resstatscounter_mismatch);
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
goto next;
Silence cppcheck 2.2 false positive in udp_recv() cppcheck 2.2 reports the following false positive: lib/dns/dispatch.c:1239:14: warning: Either the condition 'resp==NULL' is redundant or there is possible null pointer dereference: resp. [nullPointerRedundantCheck] if (disp != resp->disp) { ^ lib/dns/dispatch.c:1210:11: note: Assuming that condition 'resp==NULL' is not redundant if (resp == NULL) { ^ lib/dns/dispatch.c:1239:14: note: Null pointer dereference if (disp != resp->disp) { ^ Apparently this version of cppcheck gets confused about conditional "goto" statements because line 1239 can never be reached if 'resp' is NULL. Move a code block to prevent the above false positive from being reported without affecting the processing logic.
2020-11-25 12:45:47 +01:00
}
checkpoint
1999-07-08 22:12:37 +00:00
/*
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
* We have the right resp, so call the caller back.
checkpoint
1999-07-08 22:12:37 +00:00
*/
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
goto done;
checkpoint
1999-07-08 22:12:37 +00:00
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
next:
/*
Fix handling of mismatched responses past timeout When a UDP dispatch receives a mismatched response, it checks whether there is still enough time to wait for the correct one to arrive before the timeout fires. If there is not, the result code is set to ISC_R_TIMEDOUT, but it is not subsequently used anywhere as 'response' is set to NULL a few lines earlier. This results in the higher-level read callback (resquery_response() in case of resolver code) not being called. However, shortly afterwards, a few levels up the call chain, isc__nm_udp_read_cb() calls isc__nmsocket_timer_stop() on the dispatch socket, effectively disabling read timeout handling for that socket. Combined with the fact that reading is not restarted in such a case (e.g. by calling dispatch_getnext() from udp_recv()), this leads to the higher-level query structure remaining referenced indefinitely because the dispatch socket it uses will neither be read from nor closed due to a timeout. This in turn causes fetch contexts to linger around indefinitely, which in turn i.a. prevents certain cache nodes (those containing rdatasets used by fetch contexts, like fctx->nameservers) from being cleaned. Fix by making sure the higher-level callback does get invoked with the ISC_R_TIMEDOUT result code when udp_recv() determines there is no more time left to receive the correct UDP response before the timeout fires. This allows the higher-level callback to clean things up, preventing the reference leak described above.
2021-11-23 15:35:39 +01:00
* This is the wrong response. Check whether there is still enough
* time to wait for the correct one to arrive before the timeout fires.
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
*/
dispatch: Enforce original timeout when calling _getnext() udp_recv() will call dispatch_getnext() if the message received is invalid or doesn't match; we need to reduce the timeout each time this happens so we can't be starved forever by someone sending garbage packets.
2021-10-01 12:53:31 -07:00
timeout = resp->timeout - dispentry_runtime(resp);
if (timeout <= 0) {
Fix handling of mismatched responses past timeout When a UDP dispatch receives a mismatched response, it checks whether there is still enough time to wait for the correct one to arrive before the timeout fires. If there is not, the result code is set to ISC_R_TIMEDOUT, but it is not subsequently used anywhere as 'response' is set to NULL a few lines earlier. This results in the higher-level read callback (resquery_response() in case of resolver code) not being called. However, shortly afterwards, a few levels up the call chain, isc__nm_udp_read_cb() calls isc__nmsocket_timer_stop() on the dispatch socket, effectively disabling read timeout handling for that socket. Combined with the fact that reading is not restarted in such a case (e.g. by calling dispatch_getnext() from udp_recv()), this leads to the higher-level query structure remaining referenced indefinitely because the dispatch socket it uses will neither be read from nor closed due to a timeout. This in turn causes fetch contexts to linger around indefinitely, which in turn i.a. prevents certain cache nodes (those containing rdatasets used by fetch contexts, like fctx->nameservers) from being cleaned. Fix by making sure the higher-level callback does get invoked with the ISC_R_TIMEDOUT result code when udp_recv() determines there is no more time left to receive the correct UDP response before the timeout fires. This allows the higher-level callback to clean things up, preventing the reference leak described above.
2021-11-23 15:35:39 +01:00
/*
* The time window for receiving the correct response is
* already closed, libuv has just not processed the socket
* timer yet. Invoke the read callback, indicating a timeout.
*/
dispatch: Enforce original timeout when calling _getnext() udp_recv() will call dispatch_getnext() if the message received is invalid or doesn't match; we need to reduce the timeout each time this happens so we can't be starved forever by someone sending garbage packets.
2021-10-01 12:53:31 -07:00
eresult = ISC_R_TIMEDOUT;
goto done;
}
Fix handling of mismatched responses past timeout When a UDP dispatch receives a mismatched response, it checks whether there is still enough time to wait for the correct one to arrive before the timeout fires. If there is not, the result code is set to ISC_R_TIMEDOUT, but it is not subsequently used anywhere as 'response' is set to NULL a few lines earlier. This results in the higher-level read callback (resquery_response() in case of resolver code) not being called. However, shortly afterwards, a few levels up the call chain, isc__nm_udp_read_cb() calls isc__nmsocket_timer_stop() on the dispatch socket, effectively disabling read timeout handling for that socket. Combined with the fact that reading is not restarted in such a case (e.g. by calling dispatch_getnext() from udp_recv()), this leads to the higher-level query structure remaining referenced indefinitely because the dispatch socket it uses will neither be read from nor closed due to a timeout. This in turn causes fetch contexts to linger around indefinitely, which in turn i.a. prevents certain cache nodes (those containing rdatasets used by fetch contexts, like fctx->nameservers) from being cleaned. Fix by making sure the higher-level callback does get invoked with the ISC_R_TIMEDOUT result code when udp_recv() determines there is no more time left to receive the correct UDP response before the timeout fires. This allows the higher-level callback to clean things up, preventing the reference leak described above.
2021-11-23 15:35:39 +01:00
/*
* Do not invoke the read callback just yet and instead wait for the
* proper response to arrive until the original timeout fires.
*/
response = NULL;
dispatch: Enforce original timeout when calling _getnext() udp_recv() will call dispatch_getnext() if the message received is invalid or doesn't match; we need to reduce the timeout each time this happens so we can't be starved forever by someone sending garbage packets.
2021-10-01 12:53:31 -07:00
dispatch_getnext(disp, resp, resp->timeout - dispentry_runtime(resp));
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
done:
3913. [bug] Address race issue in dispatch. [RT #36731]
2014-08-06 18:49:53 +10:00
UNLOCK(&disp->lock);
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
dispatch: Refactor to eliminate dns_dispatchevent - Responses received by the dispatch are no longer sent to the caller via a task event, but via a netmgr-style recv callback. the 'action' parameter to dns_dispatch_addresponse() is now called 'response' and is called directly from udp_recv() or tcp_recv() when a valid response has been received. - All references to isc_task and isc_taskmgr have been removed from dispatch functions. - All references to dns_dispatchevent_t have been removed and the type has been deleted. - Added a task to the resolver response context, to be used for fctx events. - When the caller cancels an operation, the response handler will be called with ISC_R_CANCELED; it can abort immediately since the caller will presumably have taken care of cleanup already. - Cleaned up attach/detach in resquery and request.
2021-07-26 20:23:18 -07:00
if (response != NULL) {
dispatch: Remove 'timeout' callback - It is no longer necessary to pass a 'timeout' callback to dns_dispatch_addresponse(); timeouts are handled directly by the 'response' callback instead. - The netmgr handle is no longer passed to dispatch callbacks, since they don't (and can't) use it. instead, dispatch_cb_t now takes a result code, region, and argument. - Cleaned up timeout-related tests in dispatch_test.c
2021-08-03 15:27:06 +02:00
response(eresult, region, resp->arg);
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
}
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
dispentry_detach(&resp);
add dns_dispatch_attach/detach(), and start on serious dispatcher hacking.
1999-06-30 01:33:11 +00:00
}
make this (mostly) work with tcp
1999-07-12 23:44:31 +00:00
/*
* General flow:
*
fixed a bug that named crashes with an assertion failure on exit when sharing the same port for listening and querying, and changing listening addresses several times. [RT# 3509] additionally, + limited the canceled socket tasks in dispatch.c + made dns_dispatch_changeattributes() care about the NOLISTEN mask + described side effects of dns_dispatch_changeattributes() in its description comment
2002-09-04 02:26:13 +00:00
* If I/O result == CANCELED, EOF, or error, notify everyone as the
* various queues drain.
make this (mostly) work with tcp
1999-07-12 23:44:31 +00:00
*
fixed a bug that named crashes with an assertion failure on exit when sharing the same port for listening and querying, and changing listening addresses several times. [RT# 3509] additionally, + limited the canceled socket tasks in dispatch.c + made dns_dispatch_changeattributes() care about the NOLISTEN mask + described side effects of dns_dispatch_changeattributes() in its description comment
2002-09-04 02:26:13 +00:00
* If query, restart.
make this (mostly) work with tcp
1999-07-12 23:44:31 +00:00
*
* If response:
* Allocate event, fill in details.
fixed a bug that named crashes with an assertion failure on exit when sharing the same port for listening and querying, and changing listening addresses several times. [RT# 3509] additionally, + limited the canceled socket tasks in dispatch.c + made dns_dispatch_changeattributes() care about the NOLISTEN mask + described side effects of dns_dispatch_changeattributes() in its description comment
2002-09-04 02:26:13 +00:00
* If cannot allocate, restart.
* find target. If not found, restart.
make this (mostly) work with tcp
1999-07-12 23:44:31 +00:00
* if event queue is not empty, queue. else, send.
* restart.
*/
static void
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
tcp_recv(isc_nmhandle_t *handle, isc_result_t eresult, isc_region_t *region,
void *arg) {
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
dns_dispatch_t *disp = (dns_dispatch_t *)arg;
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
dns_dispentry_t *resp = NULL, *next = NULL;
apply the modified style
2020-02-13 14:44:37 -08:00
dns_messageid_t id;
isc_result_t dres;
unsigned int flags;
unsigned int bucket;
General code refactoring - style cleanup - removed NULL checks in places where they are not currently needed - use isc_refcount for dispatch reference counting - revised code flow for readability - remove some #ifdefs that are no longer relevant - remove unused struct members - removed unnecessary function parameters - use C99 struct initialization
2020-12-09 19:44:41 -08:00
dns_qid_t *qid = NULL;
apply the modified style
2020-02-13 14:44:37 -08:00
int level;
char buf[ISC_SOCKADDR_FORMATSIZE];
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
isc_buffer_t source;
isc_sockaddr_t peer;
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
dns_displist_t resps;
make this (mostly) work with tcp
1999-07-12 23:44:31 +00:00
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
REQUIRE(VALID_DISPATCH(disp));
In tcp_recv(), free the event before calling destroy(), since the event is a statically allocated part of the tcpmsg.
1999-11-16 21:05:09 +00:00
Dispatch API simplification - Many dispatch attributes can be set implicitly instead of being passed in. we can infer whether to set DNS_DISPATCHATTR_TCP or _UDP from whether we're calling dns_dispatch_createtcp() or _createudp(). we can also infer DNS_DISPATCHATTR_IPV4 or _IPV6 from the addresses or the socket that were passed in. - We no longer use dup'd sockets in UDP dispatches, so the 'dup_socket' parameter has been removed from dns_dispatch_createudp(), along with the code implementing it. also removed isc_socket_dup() since it no longer has any callers. - The 'buffersize' parameter was ignored and has now been removed; buffersize is now fixed at 4096. - Maxbuffers and maxrequests don't need to be passed in on every call to dns_dispatch_createtcp() and _createudp(). In all current uses, the value for mgr->maxbuffers will either be raised once from its default of 20000 to 32768, or else left alone. (passing in a value lower than 20000 does not lower it.) there isn't enough difference between these values for there to be any need to configure this. The value for disp->maxrequests controls both the quota of concurrent requests for a dispatch and also the size of the dispatch socket memory pool. it's not clear that this quota is necessary at all. the memory pool size currently starts at 32768, but is sometimes lowered to 4096, which is definitely unnecessary. This commit sets both values permanently to 32768. - Previously TCP dispatches allocated their own separate QID table, which didn't incorporate a port table. this commit removes per-dispatch QID tables and shares the same table between all dispatches. since dispatches are created for each TCP socket, this may speed up the dispatch allocation process. there may be a slight increase in lock contention since all dispatches are sharing a single QID table, but since TCP sockets are used less often than UDP sockets (which were already sharing a QID table), it should not be a substantial change. - The dispatch port table was being used to determine whether a port was already in use; if so, then a UDP socket would be bound with REUSEADDR. this commit removes the port table, and always binds UDP sockets that way.
2020-12-17 00:43:00 -08:00
qid = disp->mgr->qid;
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
lock dispatch before reporting state
2019-11-22 10:49:40 +11:00
LOCK(&disp->lock);
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
dispatch_log(disp, LVL(90), "TCP read:%s:requests %d, buffers %d",
isc_result_totext(eresult), disp->requests,
disp->tcpbuffers);
make this (mostly) work with tcp
1999-07-12 23:44:31 +00:00
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
peer = isc_nmhandle_peeraddr(handle);
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
ISC_LIST_INIT(resps);
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
switch (eresult) {
case ISC_R_SUCCESS:
/* got our answer */
break;
2353. [func] Add support for Name Server ID (RFC 5001). 'dig +nsid' requests NSID from server. 'request-nsid yes;' causes recursive server to send NSID requests to upstream servers. Server responds to NSID requests with the string configured by 'server-id' option. [RT #17091]
2008-04-03 02:01:08 +00:00
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
case ISC_R_SHUTTINGDOWN:
case ISC_R_CANCELED:
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
case ISC_R_EOF:
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
dispatch_log(disp, LVL(90), "shutting down: %s",
isc_result_totext(eresult));
/*
* If there are any active responses, shut them all down.
*/
Shutdown all TCP connection on invalid DNS message Previously, when invalid DNS message is received over TCP we throw the garbage DNS message away and continued looking for valid DNS message that would match our outgoing queries. This logic makes sense for UDP, because anyone can send DNS message over UDP. Change the logic that the TCP connection is closed when we receive garbage, because the other side is acting malicious.
2021-11-26 09:14:58 +01:00
goto shutdown;
TCP receives should never be retried
2000-09-08 22:02:21 +00:00
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
case ISC_R_TIMEDOUT:
/*
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
* Time out the oldest response in the active queue,
* and move it to the end. (We don't remove it from the
* active queue immediately, though, because the callback
* might decide to keep waiting and leave it active.)
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
*/
resp = ISC_LIST_HEAD(disp->active);
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
if (resp != NULL) {
dispentry_attach(resp, &(dns_dispentry_t *){ NULL });
ISC_LIST_UNLINK(disp->active, resp, alink);
ISC_LIST_APPEND(disp->active, resp, alink);
}
goto done;
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
default:
if (eresult == ISC_R_CONNECTIONRESET) {
level = ISC_LOG_INFO;
} else {
level = ISC_LOG_ERROR;
TCP receives should never be retried
2000-09-08 22:02:21 +00:00
}
add dispatcher logging functions
2000-04-29 00:45:26 +00:00
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
isc_sockaddr_format(&peer, buf, sizeof(buf));
dispatch_log(disp, level,
"shutting down due to TCP "
"receive error: %s: %s",
buf, isc_result_totext(eresult));
Shutdown all active TCP connections on error When outgoing TCP connection was prematurely terminated (f.e. with connection reset), the dispatch code would not cleanup the resources used by such connection leading to dangling dns_dispentry_t entries.
2021-11-26 09:14:58 +01:00
/*
* If there are any active responses, shut them all down.
*/
for (resp = ISC_LIST_HEAD(disp->active); resp != NULL;
resp = next) {
next = ISC_LIST_NEXT(resp, alink);
dispentry_attach(resp, &(dns_dispentry_t *){ NULL });
ISC_LIST_UNLINK(disp->active, resp, alink);
ISC_LIST_APPEND(resps, resp, rlink);
}
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
goto done;
make this (mostly) work with tcp
1999-07-12 23:44:31 +00:00
}
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
dispatch_log(disp, LVL(90), "success, length == %d, addr = %p",
region->length, region->base);
make this (mostly) work with tcp
1999-07-12 23:44:31 +00:00
/*
* Peek into the buffer to see what we can see.
*/
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
isc_buffer_init(&source, region->base, region->length);
isc_buffer_add(&source, region->length);
dres = dns_message_peekheader(&source, &id, &flags);
s/DNS_R_/ISC_R_/ change for some codes.
2000-04-06 22:03:35 +00:00
if (dres != ISC_R_SUCCESS) {
do not use initial caps in log messages
2000-06-16 15:51:58 +00:00
dispatch_log(disp, LVL(10), "got garbage packet");
Shutdown all TCP connection on invalid DNS message Previously, when invalid DNS message is received over TCP we throw the garbage DNS message away and continued looking for valid DNS message that would match our outgoing queries. This logic makes sense for UDP, because anyone can send DNS message over UDP. Change the logic that the TCP connection is closed when we receive garbage, because the other side is acting malicious.
2021-11-26 09:14:58 +01:00
eresult = ISC_R_UNEXPECTED;
goto shutdown;
make this (mostly) work with tcp
1999-07-12 23:44:31 +00:00
}
add dispatcher logging functions
2000-04-29 00:45:26 +00:00
dispatch_log(disp, LVL(92),
do not use initial caps in log messages
2000-06-16 15:51:58 +00:00
"got valid DNS message header, /QR %c, id %u",
Turn (int & flag) into (int & flag) != 0 when implicitly typed to bool
2018-10-11 11:57:57 +02:00
(((flags & DNS_MESSAGEFLAG_QR) != 0) ? '1' : '0'), id);
make this (mostly) work with tcp
1999-07-12 23:44:31 +00:00
/*
Shutdown all TCP connection on invalid DNS message Previously, when invalid DNS message is received over TCP we throw the garbage DNS message away and continued looking for valid DNS message that would match our outgoing queries. This logic makes sense for UDP, because anyone can send DNS message over UDP. Change the logic that the TCP connection is closed when we receive garbage, because the other side is acting malicious.
2021-11-26 09:14:58 +01:00
* Look at the message flags. If it's a query, stop reading.
make this (mostly) work with tcp
1999-07-12 23:44:31 +00:00
*/
if ((flags & DNS_MESSAGEFLAG_QR) == 0) {
Shutdown all TCP connection on invalid DNS message Previously, when invalid DNS message is received over TCP we throw the garbage DNS message away and continued looking for valid DNS message that would match our outgoing queries. This logic makes sense for UDP, because anyone can send DNS message over UDP. Change the logic that the TCP connection is closed when we receive garbage, because the other side is acting malicious.
2021-11-26 09:14:58 +01:00
dispatch_log(disp, LVL(10), "got DNS query instead of answer");
eresult = ISC_R_UNEXPECTED;
goto shutdown;
make this (mostly) work with tcp
1999-07-12 23:44:31 +00:00
}
1687. [bug] Race condition in dispatch. [RT #10272]
2004-07-21 00:48:19 +00:00
/*
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
* We have a valid response; find the associated dispentry object
* and call the caller back.
1687. [bug] Race condition in dispatch. [RT #10272]
2004-07-21 00:48:19 +00:00
*/
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
bucket = dns_hash(qid, &peer, id, disp->localport);
1687. [bug] Race condition in dispatch. [RT #10272]
2004-07-21 00:48:19 +00:00
LOCK(&qid->lock);
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
resp = entry_search(qid, &peer, id, disp->localport, bucket);
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
if (resp != NULL) {
dispentry_attach(resp, &(dns_dispentry_t *){ NULL });
}
Use clang-format to reformat the source files
2020-02-12 13:59:18 +01:00
dispatch_log(disp, LVL(90), "search for response in bucket %d: %s",
1687. [bug] Race condition in dispatch. [RT #10272]
2004-07-21 00:48:19 +00:00
bucket, (resp == NULL ? "not found" : "found"));
Rewrite dns_resolver and dns_request to use netmgr timeouts - The `timeout_action` parameter to dns_dispatch_addresponse() been replaced with a netmgr callback that is called when a dispatch read times out. this callback may optionally reset the read timer and resume reading. - Added a function to convert isc_interval to milliseconds; this is used to translate fctx->interval into a value that can be passed to dns_dispatch_addresponse() as the timeout. - Note that netmgr timeouts are accurate to the millisecond, so code to check whether a timeout has been reached cannot rely on microsecond accuracy. - If serve-stale is configured, then a timeout received by the resolver may trigger it to return stale data, and then resume waiting for the read timeout. this is no longer based on a separate stale timer. - The code for canceling requests in request.c has been altered so that it can run asynchronously. - TCP timeout events apply to the dispatch, which may be shared by multiple queries. since in the event of a timeout we have no query ID to use to identify the resp we wanted, we now just send the timeout to the oldest query that was pending. - There was some additional refactoring in the resolver: combining fctx_join() and fctx_try_events() into one function to reduce code duplication, and using fixednames in fetchctx and fetchevent. - Incidental fix: new_adbaddrinfo() can't return NULL anymore, so the code can be simplified.
2021-04-22 18:58:01 -07:00
UNLOCK(&qid->lock);
1687. [bug] Race condition in dispatch. [RT #10272]
2004-07-21 00:48:19 +00:00
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
dispatch_getnext(disp, NULL, -1);
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
Shutdown all TCP connection on invalid DNS message Previously, when invalid DNS message is received over TCP we throw the garbage DNS message away and continued looking for valid DNS message that would match our outgoing queries. This logic makes sense for UDP, because anyone can send DNS message over UDP. Change the logic that the TCP connection is closed when we receive garbage, because the other side is acting malicious.
2021-11-26 09:14:58 +01:00
goto done;
shutdown:
for (resp = ISC_LIST_HEAD(disp->active); resp != NULL; resp = next) {
next = ISC_LIST_NEXT(resp, alink);
dispentry_attach(resp, &(dns_dispentry_t *){ NULL });
ISC_LIST_UNLINK(disp->active, resp, alink);
ISC_LIST_APPEND(resps, resp, rlink);
}
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
done:
3913. [bug] Address race issue in dispatch. [RT #36731]
2014-08-06 18:49:53 +10:00
UNLOCK(&disp->lock);
make this (mostly) work with tcp
1999-07-12 23:44:31 +00:00
dispatch: Refactor to eliminate dns_dispatchevent - Responses received by the dispatch are no longer sent to the caller via a task event, but via a netmgr-style recv callback. the 'action' parameter to dns_dispatch_addresponse() is now called 'response' and is called directly from udp_recv() or tcp_recv() when a valid response has been received. - All references to isc_task and isc_taskmgr have been removed from dispatch functions. - All references to dns_dispatchevent_t have been removed and the type has been deleted. - Added a task to the resolver response context, to be used for fctx events. - When the caller cancels an operation, the response handler will be called with ISC_R_CANCELED; it can abort immediately since the caller will presumably have taken care of cleanup already. - Cleaned up attach/detach in resquery and request.
2021-07-26 20:23:18 -07:00
if (resp != NULL) {
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
/* We got a matching response, or timed out */
dispatch: Remove 'timeout' callback - It is no longer necessary to pass a 'timeout' callback to dns_dispatch_addresponse(); timeouts are handled directly by the 'response' callback instead. - The netmgr handle is no longer passed to dispatch callbacks, since they don't (and can't) use it. instead, dispatch_cb_t now takes a result code, region, and argument. - Cleaned up timeout-related tests in dispatch_test.c
2021-08-03 15:27:06 +02:00
resp->response(eresult, region, resp->arg);
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
dispentry_detach(&resp);
} else {
/* We're being shut down; cancel all outstanding resps */
for (resp = ISC_LIST_HEAD(resps); resp != NULL; resp = next) {
next = ISC_LIST_NEXT(resp, rlink);
ISC_LIST_UNLINK(resps, resp, rlink);
resp->response(ISC_R_SHUTTINGDOWN, region, resp->arg);
dispentry_detach(&resp);
}
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
}
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
dns_dispatch_detach(&disp);
implement dns_dispatchmgr_t, replacing dns_dispatchlist_t. Use it throughout the library/server.
2000-05-10 21:34:50 +00:00
}
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
/*%
* Create a temporary port list to set the initial default set of dispatch
Use system ephemeral ports for default portset In dispatch, use system ephemeral ports for default portset instead of hardcoded <1024,65535> range.
2021-10-04 14:11:57 +02:00
* ephemeral ports. This is almost meaningless as the application will
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
* normally set the ports explicitly, but is provided to fill some minor corner
* cases.
implement dns_dispatchmgr_t, replacing dns_dispatchlist_t. Use it throughout the library/server.
2000-05-10 21:34:50 +00:00
*/
static void
Use system ephemeral ports for default portset In dispatch, use system ephemeral ports for default portset instead of hardcoded <1024,65535> range.
2021-10-04 14:11:57 +02:00
create_default_portset(isc_mem_t *mctx, int family, isc_portset_t **portsetp) {
in_port_t low, high;
isc_net_getudpportrange(family, &low, &high);
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
isc_portset_create(mctx, portsetp);
Use system ephemeral ports for default portset In dispatch, use system ephemeral ports for default portset instead of hardcoded <1024,65535> range.
2021-10-04 14:11:57 +02:00
isc_portset_addrange(*portsetp, low, high);
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
}
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
Simplify dns_dispatchmgr_create with fixed buffersize - UDP buffersize is now established when creating dispatch manager and is always set to 4096. - Set up the default port range in dispatchmgr before setting the magic number. - Magic is not set until dispatchmgr is fully created.
2020-12-09 15:45:13 -08:00
static isc_result_t
setavailports(dns_dispatchmgr_t *mgr, isc_portset_t *v4portset,
isc_portset_t *v6portset) {
in_port_t *v4ports, *v6ports, p = 0;
unsigned int nv4ports, nv6ports, i4 = 0, i6 = 0;
nv4ports = isc_portset_nports(v4portset);
nv6ports = isc_portset_nports(v6portset);
v4ports = NULL;
if (nv4ports != 0) {
v4ports = isc_mem_get(mgr->mctx, sizeof(in_port_t) * nv4ports);
Use clang-tidy to add curly braces around one-line statements The command used to reformat the files in this commit was: ./util/run-clang-tidy \ -clang-tidy-binary clang-tidy-11 -clang-apply-replacements-binary clang-apply-replacements-11 \ -checks=-*,readability-braces-around-statements \ -j 9 \ -fix \ -format \ -style=file \ -quiet clang-format -i --style=format $(git ls-files '*.c' '*.h') uncrustify -c .uncrustify.cfg --replace --no-backup $(git ls-files '*.c' '*.h') clang-format -i --style=format $(git ls-files '*.c' '*.h')
2020-02-13 21:48:23 +01:00
}
Simplify dns_dispatchmgr_create with fixed buffersize - UDP buffersize is now established when creating dispatch manager and is always set to 4096. - Set up the default port range in dispatchmgr before setting the magic number. - Magic is not set until dispatchmgr is fully created.
2020-12-09 15:45:13 -08:00
v6ports = NULL;
if (nv6ports != 0) {
v6ports = isc_mem_get(mgr->mctx, sizeof(in_port_t) * nv6ports);
}
do {
if (isc_portset_isset(v4portset, p)) {
INSIST(i4 < nv4ports);
v4ports[i4++] = p;
}
if (isc_portset_isset(v6portset, p)) {
INSIST(i6 < nv6ports);
v6ports[i6++] = p;
}
} while (p++ < 65535);
INSIST(i4 == nv4ports && i6 == nv6ports);
if (mgr->v4ports != NULL) {
isc_mem_put(mgr->mctx, mgr->v4ports,
mgr->nv4ports * sizeof(in_port_t));
}
mgr->v4ports = v4ports;
mgr->nv4ports = nv4ports;
if (mgr->v6ports != NULL) {
isc_mem_put(mgr->mctx, mgr->v6ports,
mgr->nv6ports * sizeof(in_port_t));
}
mgr->v6ports = v6ports;
mgr->nv6ports = nv6ports;
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
return (ISC_R_SUCCESS);
}
add dns_dispatch_attach/detach(), and start on serious dispatcher hacking.
1999-06-30 01:33:11 +00:00
/*
* Publics.
*/
fix many bugs.
1999-07-22 01:34:31 +00:00
isc_result_t
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
dns_dispatchmgr_create(isc_mem_t *mctx, isc_nm_t *nm,
dns_dispatchmgr_t **mgrp) {
Simplify dns_dispatchmgr_create with fixed buffersize - UDP buffersize is now established when creating dispatch manager and is always set to 4096. - Set up the default port range in dispatchmgr before setting the magic number. - Magic is not set until dispatchmgr is fully created.
2020-12-09 15:45:13 -08:00
dns_dispatchmgr_t *mgr = NULL;
apply the modified style
2020-02-13 14:44:37 -08:00
isc_portset_t *v4portset = NULL;
isc_portset_t *v6portset = NULL;
implement dns_dispatchmgr_t, replacing dns_dispatchlist_t. Use it throughout the library/server.
2000-05-10 21:34:50 +00:00
REQUIRE(mctx != NULL);
REQUIRE(mgrp != NULL && *mgrp == NULL);
mgr = isc_mem_get(mctx, sizeof(dns_dispatchmgr_t));
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
*mgr = (dns_dispatchmgr_t){ .magic = 0 };
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
isc_refcount_init(&mgr->references, 1);
implement dns_dispatchmgr_t, replacing dns_dispatchlist_t. Use it throughout the library/server.
2000-05-10 21:34:50 +00:00
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
isc_mem_attach(mctx, &mgr->mctx);
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
isc_nm_attach(nm, &mgr->nm);
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
isc_mutex_init returns 'void'
2018-11-16 15:33:22 +01:00
isc_mutex_init(&mgr->lock);
Replace locked mempools with memory contexts Current mempools are kind of hybrid structures - they serve two purposes: 1. mempool with a lock is basically static sized allocator with pre-allocated free items 2. mempool without a lock is a doubly-linked list of preallocated items The first kind of usage could be easily replaced with jemalloc small sized arena objects and thread-local caches. The second usage not-so-much and we need to keep this (in libdns:message.c) for performance reasons.
2021-05-12 21:16:17 +02:00
implement dns_dispatchmgr_t, replacing dns_dispatchlist_t. Use it throughout the library/server.
2000-05-10 21:34:50 +00:00
ISC_LIST_INIT(mgr->list);
Replace locked mempools with memory contexts Current mempools are kind of hybrid structures - they serve two purposes: 1. mempool with a lock is basically static sized allocator with pre-allocated free items 2. mempool without a lock is a doubly-linked list of preallocated items The first kind of usage could be easily replaced with jemalloc small sized arena objects and thread-local caches. The second usage not-so-much and we need to keep this (in libdns:message.c) for performance reasons.
2021-05-12 21:16:17 +02:00
Use system ephemeral ports for default portset In dispatch, use system ephemeral ports for default portset instead of hardcoded <1024,65535> range.
2021-10-04 14:11:57 +02:00
create_default_portset(mctx, AF_INET, &v4portset);
create_default_portset(mctx, AF_INET6, &v6portset);
implement dns_dispatchmgr_t, replacing dns_dispatchlist_t. Use it throughout the library/server.
2000-05-10 21:34:50 +00:00
Simplify dns_dispatchmgr_create with fixed buffersize - UDP buffersize is now established when creating dispatch manager and is always set to 4096. - Set up the default port range in dispatchmgr before setting the magic number. - Magic is not set until dispatchmgr is fully created.
2020-12-09 15:45:13 -08:00
setavailports(mgr, v4portset, v6portset);
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
Simplify dns_dispatchmgr_create with fixed buffersize - UDP buffersize is now established when creating dispatch manager and is always set to 4096. - Set up the default port range in dispatchmgr before setting the magic number. - Magic is not set until dispatchmgr is fully created.
2020-12-09 15:45:13 -08:00
isc_portset_destroy(mctx, &v4portset);
isc_portset_destroy(mctx, &v6portset);
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
Dispatch API simplification - Many dispatch attributes can be set implicitly instead of being passed in. we can infer whether to set DNS_DISPATCHATTR_TCP or _UDP from whether we're calling dns_dispatch_createtcp() or _createudp(). we can also infer DNS_DISPATCHATTR_IPV4 or _IPV6 from the addresses or the socket that were passed in. - We no longer use dup'd sockets in UDP dispatches, so the 'dup_socket' parameter has been removed from dns_dispatch_createudp(), along with the code implementing it. also removed isc_socket_dup() since it no longer has any callers. - The 'buffersize' parameter was ignored and has now been removed; buffersize is now fixed at 4096. - Maxbuffers and maxrequests don't need to be passed in on every call to dns_dispatch_createtcp() and _createudp(). In all current uses, the value for mgr->maxbuffers will either be raised once from its default of 20000 to 32768, or else left alone. (passing in a value lower than 20000 does not lower it.) there isn't enough difference between these values for there to be any need to configure this. The value for disp->maxrequests controls both the quota of concurrent requests for a dispatch and also the size of the dispatch socket memory pool. it's not clear that this quota is necessary at all. the memory pool size currently starts at 32768, but is sometimes lowered to 4096, which is definitely unnecessary. This commit sets both values permanently to 32768. - Previously TCP dispatches allocated their own separate QID table, which didn't incorporate a port table. this commit removes per-dispatch QID tables and shares the same table between all dispatches. since dispatches are created for each TCP socket, this may speed up the dispatch allocation process. there may be a slight increase in lock contention since all dispatches are sharing a single QID table, but since TCP sockets are used less often than UDP sockets (which were already sharing a QID table), it should not be a substantial change. - The dispatch port table was being used to determine whether a port was already in use; if so, then a UDP socket would be bound with REUSEADDR. this commit removes the port table, and always binds UDP sockets that way.
2020-12-17 00:43:00 -08:00
qid_allocate(mgr, &mgr->qid);
Simplify dns_dispatchmgr_create with fixed buffersize - UDP buffersize is now established when creating dispatch manager and is always set to 4096. - Set up the default port range in dispatchmgr before setting the magic number. - Magic is not set until dispatchmgr is fully created.
2020-12-09 15:45:13 -08:00
mgr->magic = DNS_DISPATCHMGR_MAGIC;
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
Simplify dns_dispatchmgr_create with fixed buffersize - UDP buffersize is now established when creating dispatch manager and is always set to 4096. - Set up the default port range in dispatchmgr before setting the magic number. - Magic is not set until dispatchmgr is fully created.
2020-12-09 15:45:13 -08:00
*mgrp = mgr;
return (ISC_R_SUCCESS);
implement dns_dispatchmgr_t, replacing dns_dispatchlist_t. Use it throughout the library/server.
2000-05-10 21:34:50 +00:00
}
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
void
Remove DISPATH_TRACE, RESOLVER_TRACE and REQ_TRACE logging Remove the debugging printfs. (leaving this as a separate commit rather than squashing it so we can restore it in the future if we ever need it again.)
2021-09-09 15:25:08 -07:00
dns_dispatchmgr_attach(dns_dispatchmgr_t *mgr, dns_dispatchmgr_t **mgrp) {
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
REQUIRE(VALID_DISPATCHMGR(mgr));
REQUIRE(mgrp != NULL && *mgrp == NULL);
Remove DISPATH_TRACE, RESOLVER_TRACE and REQ_TRACE logging Remove the debugging printfs. (leaving this as a separate commit rather than squashing it so we can restore it in the future if we ever need it again.)
2021-09-09 15:25:08 -07:00
isc_refcount_increment(&mgr->references);
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
*mgrp = mgr;
}
void
Remove DISPATH_TRACE, RESOLVER_TRACE and REQ_TRACE logging Remove the debugging printfs. (leaving this as a separate commit rather than squashing it so we can restore it in the future if we ever need it again.)
2021-09-09 15:25:08 -07:00
dns_dispatchmgr_detach(dns_dispatchmgr_t **mgrp) {
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
dns_dispatchmgr_t *mgr = NULL;
uint_fast32_t ref;
REQUIRE(mgrp != NULL && VALID_DISPATCHMGR(*mgrp));
mgr = *mgrp;
*mgrp = NULL;
ref = isc_refcount_decrement(&mgr->references);
if (ref == 1) {
dispatchmgr_destroy(mgr);
}
}
539. [func] Support the blackhole option.
2000-11-03 02:45:55 +00:00
void
apply the modified style
2020-02-13 14:44:37 -08:00
dns_dispatchmgr_setblackhole(dns_dispatchmgr_t *mgr, dns_acl_t *blackhole) {
539. [func] Support the blackhole option.
2000-11-03 02:45:55 +00:00
REQUIRE(VALID_DISPATCHMGR(mgr));
Use clang-tidy to add curly braces around one-line statements The command used to reformat the files in this commit was: ./util/run-clang-tidy \ -clang-tidy-binary clang-tidy-11 -clang-apply-replacements-binary clang-apply-replacements-11 \ -checks=-*,readability-braces-around-statements \ -j 9 \ -fix \ -format \ -style=file \ -quiet clang-format -i --style=format $(git ls-files '*.c' '*.h') uncrustify -c .uncrustify.cfg --replace --no-backup $(git ls-files '*.c' '*.h') clang-format -i --style=format $(git ls-files '*.c' '*.h')
2020-02-13 21:48:23 +01:00
if (mgr->blackhole != NULL) {
635. [bug] Reloading a server with a configured blackhole list would cause an assertion. [RT #590]
2000-12-26 09:48:41 +00:00
dns_acl_detach(&mgr->blackhole);
Use clang-tidy to add curly braces around one-line statements The command used to reformat the files in this commit was: ./util/run-clang-tidy \ -clang-tidy-binary clang-tidy-11 -clang-apply-replacements-binary clang-apply-replacements-11 \ -checks=-*,readability-braces-around-statements \ -j 9 \ -fix \ -format \ -style=file \ -quiet clang-format -i --style=format $(git ls-files '*.c' '*.h') uncrustify -c .uncrustify.cfg --replace --no-backup $(git ls-files '*.c' '*.h') clang-format -i --style=format $(git ls-files '*.c' '*.h')
2020-02-13 21:48:23 +01:00
}
539. [func] Support the blackhole option.
2000-11-03 02:45:55 +00:00
dns_acl_attach(blackhole, &mgr->blackhole);
}
removed more needless attaching and detaching of the blackhole ACL
2001-02-09 00:23:16 +00:00
dns_acl_t *
apply the modified style
2020-02-13 14:44:37 -08:00
dns_dispatchmgr_getblackhole(dns_dispatchmgr_t *mgr) {
539. [func] Support the blackhole option.
2000-11-03 02:45:55 +00:00
REQUIRE(VALID_DISPATCHMGR(mgr));
removed more needless attaching and detaching of the blackhole ACL
2001-02-09 00:23:16 +00:00
return (mgr->blackhole);
539. [func] Support the blackhole option.
2000-11-03 02:45:55 +00:00
}
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
isc_result_t
dns_dispatchmgr_setavailports(dns_dispatchmgr_t *mgr, isc_portset_t *v4portset,
apply the modified style
2020-02-13 14:44:37 -08:00
isc_portset_t *v6portset) {
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
REQUIRE(VALID_DISPATCHMGR(mgr));
Simplify dns_dispatchmgr_create with fixed buffersize - UDP buffersize is now established when creating dispatch manager and is always set to 4096. - Set up the default port range in dispatchmgr before setting the magic number. - Magic is not set until dispatchmgr is fully created.
2020-12-09 15:45:13 -08:00
return (setavailports(mgr, v4portset, v6portset));
1442. [func] New fuctions for manipulating port lists: dns_portlist_create(), dns_portlist_add(), dns_portlist_remove(), dns_portlist_match(), dns_portlist_attach() and dns_portlist_detach(). 1441. [func] It is now possible to tell dig to bind to a specific source port. 1440. [func] It is now possible to tell named to avoid using certian source ports (avoid-v4-udp-ports, avoid-v6-udp-ports). developer: marka reviewer: explorer
2003-02-26 05:05:16 +00:00
}
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
static void
dispatchmgr_destroy(dns_dispatchmgr_t *mgr) {
REQUIRE(VALID_DISPATCHMGR(mgr));
implement dns_dispatchmgr_t, replacing dns_dispatchlist_t. Use it throughout the library/server.
2000-05-10 21:34:50 +00:00
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
isc_refcount_destroy(&mgr->references);
implement dns_dispatchmgr_t, replacing dns_dispatchlist_t. Use it throughout the library/server.
2000-05-10 21:34:50 +00:00
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
mgr->magic = 0;
isc_mutex_destroy(&mgr->lock);
mgr->state = 0;
implement dns_dispatchmgr_t, replacing dns_dispatchlist_t. Use it throughout the library/server.
2000-05-10 21:34:50 +00:00
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
qid_destroy(mgr->mctx, &mgr->qid);
if (mgr->blackhole != NULL) {
dns_acl_detach(&mgr->blackhole);
}
implement dns_dispatchmgr_t, replacing dns_dispatchlist_t. Use it throughout the library/server.
2000-05-10 21:34:50 +00:00
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
if (mgr->stats != NULL) {
isc_stats_detach(&mgr->stats);
}
Trailing whitespace trimmed. Perhaps running "perl util/spacewhack.pl in your own CVS tree will help minimize CVS conflicts. Maybe not. Blame Graff for getting me to trim all trailing whitespace.
2000-08-01 01:33:37 +00:00
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
if (mgr->v4ports != NULL) {
isc_mem_put(mgr->mctx, mgr->v4ports,
mgr->nv4ports * sizeof(in_port_t));
Use clang-tidy to add curly braces around one-line statements The command used to reformat the files in this commit was: ./util/run-clang-tidy \ -clang-tidy-binary clang-tidy-11 -clang-apply-replacements-binary clang-apply-replacements-11 \ -checks=-*,readability-braces-around-statements \ -j 9 \ -fix \ -format \ -style=file \ -quiet clang-format -i --style=format $(git ls-files '*.c' '*.h') uncrustify -c .uncrustify.cfg --replace --no-backup $(git ls-files '*.c' '*.h') clang-format -i --style=format $(git ls-files '*.c' '*.h')
2020-02-13 21:48:23 +01:00
}
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
if (mgr->v6ports != NULL) {
isc_mem_put(mgr->mctx, mgr->v6ports,
mgr->nv6ports * sizeof(in_port_t));
}
isc_nm_detach(&mgr->nm);
isc_mem_putanddetach(&mgr->mctx, mgr, sizeof(dns_dispatchmgr_t));
implement dns_dispatchmgr_t, replacing dns_dispatchlist_t. Use it throughout the library/server.
2000-05-10 21:34:50 +00:00
}
2355. [func] Extend the number statistics counters available. [RT #17590]
2008-04-03 05:55:52 +00:00
void
apply the modified style
2020-02-13 14:44:37 -08:00
dns_dispatchmgr_setstats(dns_dispatchmgr_t *mgr, isc_stats_t *stats) {
2355. [func] Extend the number statistics counters available. [RT #17590]
2008-04-03 05:55:52 +00:00
REQUIRE(VALID_DISPATCHMGR(mgr));
REQUIRE(ISC_LIST_EMPTY(mgr->list));
REQUIRE(mgr->stats == NULL);
2537. [func] Added more statistics counters including those on socket I/O events and query RTT histograms. [RT #18802]
2009-01-27 22:30:00 +00:00
isc_stats_attach(stats, &mgr->stats);
2355. [func] Extend the number statistics counters available. [RT #17590]
2008-04-03 05:55:52 +00:00
}
Dispatch API simplification - Many dispatch attributes can be set implicitly instead of being passed in. we can infer whether to set DNS_DISPATCHATTR_TCP or _UDP from whether we're calling dns_dispatch_createtcp() or _createudp(). we can also infer DNS_DISPATCHATTR_IPV4 or _IPV6 from the addresses or the socket that were passed in. - We no longer use dup'd sockets in UDP dispatches, so the 'dup_socket' parameter has been removed from dns_dispatch_createudp(), along with the code implementing it. also removed isc_socket_dup() since it no longer has any callers. - The 'buffersize' parameter was ignored and has now been removed; buffersize is now fixed at 4096. - Maxbuffers and maxrequests don't need to be passed in on every call to dns_dispatch_createtcp() and _createudp(). In all current uses, the value for mgr->maxbuffers will either be raised once from its default of 20000 to 32768, or else left alone. (passing in a value lower than 20000 does not lower it.) there isn't enough difference between these values for there to be any need to configure this. The value for disp->maxrequests controls both the quota of concurrent requests for a dispatch and also the size of the dispatch socket memory pool. it's not clear that this quota is necessary at all. the memory pool size currently starts at 32768, but is sometimes lowered to 4096, which is definitely unnecessary. This commit sets both values permanently to 32768. - Previously TCP dispatches allocated their own separate QID table, which didn't incorporate a port table. this commit removes per-dispatch QID tables and shares the same table between all dispatches. since dispatches are created for each TCP socket, this may speed up the dispatch allocation process. there may be a slight increase in lock contention since all dispatches are sharing a single QID table, but since TCP sockets are used less often than UDP sockets (which were already sharing a QID table), it should not be a substantial change. - The dispatch port table was being used to determine whether a port was already in use; if so, then a UDP socket would be bound with REUSEADDR. this commit removes the port table, and always binds UDP sockets that way.
2020-12-17 00:43:00 -08:00
static void
qid_allocate(dns_dispatchmgr_t *mgr, dns_qid_t **qidp) {
General code refactoring - style cleanup - removed NULL checks in places where they are not currently needed - use isc_refcount for dispatch reference counting - revised code flow for readability - remove some #ifdefs that are no longer relevant - remove unused struct members - removed unnecessary function parameters - use C99 struct initialization
2020-12-09 19:44:41 -08:00
dns_qid_t *qid = NULL;
469. [bug] "query-source address * port 53;" now works.
2000-09-18 04:50:05 +00:00
unsigned int i;
REQUIRE(qidp != NULL && *qidp == NULL);
Move the UDP buffer pool to the dispatchmgr. UDP parameters are now set by the first call to dns_dispatch_getudp(), subsecquent calls can raise the maximum number of buffers. dns_dispatchmgr_setudp() could be made public. Tidy dispatch_allocate(). entropy and mctx are now gone from dns_qid_t.
2000-09-19 06:59:28 +00:00
qid = isc_mem_get(mgr->mctx, sizeof(*qid));
Dispatch API simplification - Many dispatch attributes can be set implicitly instead of being passed in. we can infer whether to set DNS_DISPATCHATTR_TCP or _UDP from whether we're calling dns_dispatch_createtcp() or _createudp(). we can also infer DNS_DISPATCHATTR_IPV4 or _IPV6 from the addresses or the socket that were passed in. - We no longer use dup'd sockets in UDP dispatches, so the 'dup_socket' parameter has been removed from dns_dispatch_createudp(), along with the code implementing it. also removed isc_socket_dup() since it no longer has any callers. - The 'buffersize' parameter was ignored and has now been removed; buffersize is now fixed at 4096. - Maxbuffers and maxrequests don't need to be passed in on every call to dns_dispatch_createtcp() and _createudp(). In all current uses, the value for mgr->maxbuffers will either be raised once from its default of 20000 to 32768, or else left alone. (passing in a value lower than 20000 does not lower it.) there isn't enough difference between these values for there to be any need to configure this. The value for disp->maxrequests controls both the quota of concurrent requests for a dispatch and also the size of the dispatch socket memory pool. it's not clear that this quota is necessary at all. the memory pool size currently starts at 32768, but is sometimes lowered to 4096, which is definitely unnecessary. This commit sets both values permanently to 32768. - Previously TCP dispatches allocated their own separate QID table, which didn't incorporate a port table. this commit removes per-dispatch QID tables and shares the same table between all dispatches. since dispatches are created for each TCP socket, this may speed up the dispatch allocation process. there may be a slight increase in lock contention since all dispatches are sharing a single QID table, but since TCP sockets are used less often than UDP sockets (which were already sharing a QID table), it should not be a substantial change. - The dispatch port table was being used to determine whether a port was already in use; if so, then a UDP socket would be bound with REUSEADDR. this commit removes the port table, and always binds UDP sockets that way.
2020-12-17 00:43:00 -08:00
*qid = (dns_qid_t){ .qid_nbuckets = DNS_QID_BUCKETS,
.qid_increment = DNS_QID_INCREMENT };
469. [bug] "query-source address * port 53;" now works.
2000-09-18 04:50:05 +00:00
apply the modified style
2020-02-13 14:44:37 -08:00
qid->qid_table = isc_mem_get(mgr->mctx,
Dispatch API simplification - Many dispatch attributes can be set implicitly instead of being passed in. we can infer whether to set DNS_DISPATCHATTR_TCP or _UDP from whether we're calling dns_dispatch_createtcp() or _createudp(). we can also infer DNS_DISPATCHATTR_IPV4 or _IPV6 from the addresses or the socket that were passed in. - We no longer use dup'd sockets in UDP dispatches, so the 'dup_socket' parameter has been removed from dns_dispatch_createudp(), along with the code implementing it. also removed isc_socket_dup() since it no longer has any callers. - The 'buffersize' parameter was ignored and has now been removed; buffersize is now fixed at 4096. - Maxbuffers and maxrequests don't need to be passed in on every call to dns_dispatch_createtcp() and _createudp(). In all current uses, the value for mgr->maxbuffers will either be raised once from its default of 20000 to 32768, or else left alone. (passing in a value lower than 20000 does not lower it.) there isn't enough difference between these values for there to be any need to configure this. The value for disp->maxrequests controls both the quota of concurrent requests for a dispatch and also the size of the dispatch socket memory pool. it's not clear that this quota is necessary at all. the memory pool size currently starts at 32768, but is sometimes lowered to 4096, which is definitely unnecessary. This commit sets both values permanently to 32768. - Previously TCP dispatches allocated their own separate QID table, which didn't incorporate a port table. this commit removes per-dispatch QID tables and shares the same table between all dispatches. since dispatches are created for each TCP socket, this may speed up the dispatch allocation process. there may be a slight increase in lock contention since all dispatches are sharing a single QID table, but since TCP sockets are used less often than UDP sockets (which were already sharing a QID table), it should not be a substantial change. - The dispatch port table was being used to determine whether a port was already in use; if so, then a UDP socket would be bound with REUSEADDR. this commit removes the port table, and always binds UDP sockets that way.
2020-12-17 00:43:00 -08:00
DNS_QID_BUCKETS * sizeof(dns_displist_t));
for (i = 0; i < qid->qid_nbuckets; i++) {
469. [bug] "query-source address * port 53;" now works.
2000-09-18 04:50:05 +00:00
ISC_LIST_INIT(qid->qid_table[i]);
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
}
469. [bug] "query-source address * port 53;" now works.
2000-09-18 04:50:05 +00:00
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
isc_mutex_init(&qid->lock);
469. [bug] "query-source address * port 53;" now works.
2000-09-18 04:50:05 +00:00
qid->magic = QID_MAGIC;
*qidp = qid;
}
static void
apply the modified style
2020-02-13 14:44:37 -08:00
qid_destroy(isc_mem_t *mctx, dns_qid_t **qidp) {
General code refactoring - style cleanup - removed NULL checks in places where they are not currently needed - use isc_refcount for dispatch reference counting - revised code flow for readability - remove some #ifdefs that are no longer relevant - remove unused struct members - removed unnecessary function parameters - use C99 struct initialization
2020-12-09 19:44:41 -08:00
dns_qid_t *qid = NULL;
469. [bug] "query-source address * port 53;" now works.
2000-09-18 04:50:05 +00:00
REQUIRE(qidp != NULL);
qid = *qidp;
Clear the pointer to destroyed object early using the semantic patch Also disable the semantic patch as the code needs tweaks here and there because some destroy functions might not destroy the object and return early if the object is still in use.
2020-02-08 04:37:54 -08:00
*qidp = NULL;
469. [bug] "query-source address * port 53;" now works.
2000-09-18 04:50:05 +00:00
REQUIRE(VALID_QID(qid));
qid->magic = 0;
Move the UDP buffer pool to the dispatchmgr. UDP parameters are now set by the first call to dns_dispatch_getudp(), subsecquent calls can raise the maximum number of buffers. dns_dispatchmgr_setudp() could be made public. Tidy dispatch_allocate(). entropy and mctx are now gone from dns_qid_t.
2000-09-19 06:59:28 +00:00
isc_mem_put(mctx, qid->qid_table,
469. [bug] "query-source address * port 53;" now works.
2000-09-18 04:50:05 +00:00
qid->qid_nbuckets * sizeof(dns_displist_t));
- Make isc_mutex_destroy return void - Make isc_mutexblock_init/destroy return void - Minor cleanups
2018-11-19 10:31:09 +00:00
isc_mutex_destroy(&qid->lock);
Move the UDP buffer pool to the dispatchmgr. UDP parameters are now set by the first call to dns_dispatch_getudp(), subsecquent calls can raise the maximum number of buffers. dns_dispatchmgr_setudp() could be made public. Tidy dispatch_allocate(). entropy and mctx are now gone from dns_qid_t.
2000-09-19 06:59:28 +00:00
isc_mem_put(mctx, qid, sizeof(*qid));
469. [bug] "query-source address * port 53;" now works.
2000-09-18 04:50:05 +00:00
}
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
/*
* Allocate and set important limits.
*/
Creating TCP dispatch now creates/binds the socket Previously, creation of TCP dispatches differed from UDP in that a TCP dispatch was created to attach to an existing socket, whereas a UDP dispatch would be created in a vacuum and sockets would be opened on demand when a transaction was initiated. We are moving as much socket code as possible into the dispatch module, so that it can be replaced with a netmgr version as easily as possible. (This will also have the side effect of making TCP and UDP dispatches more similar.) As a step in that direction, this commit changes dns_dispatch_createtcp() so that it creates the TCP socket.
2020-12-18 14:59:50 -08:00
static void
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
dispatch_allocate(dns_dispatchmgr_t *mgr, isc_socktype_t type,
dns_dispatch_t **dispp) {
General code refactoring - style cleanup - removed NULL checks in places where they are not currently needed - use isc_refcount for dispatch reference counting - revised code flow for readability - remove some #ifdefs that are no longer relevant - remove unused struct members - removed unnecessary function parameters - use C99 struct initialization
2020-12-09 19:44:41 -08:00
dns_dispatch_t *disp = NULL;
checkpoint
1999-06-18 02:01:42 +00:00
implement dns_dispatchmgr_t, replacing dns_dispatchlist_t. Use it throughout the library/server.
2000-05-10 21:34:50 +00:00
REQUIRE(VALID_DISPATCHMGR(mgr));
checkpoint
1999-06-18 23:54:59 +00:00
REQUIRE(dispp != NULL && *dispp == NULL);
checkpoint
1999-06-18 02:01:42 +00:00
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
/*
* Set up the dispatcher, mostly. Don't bother setting some of
* the options that are controlled by tcp vs. udp, etc.
*/
checkpoint
1999-06-18 02:01:42 +00:00
Replace locked mempools with memory contexts Current mempools are kind of hybrid structures - they serve two purposes: 1. mempool with a lock is basically static sized allocator with pre-allocated free items 2. mempool without a lock is a doubly-linked list of preallocated items The first kind of usage could be easily replaced with jemalloc small sized arena objects and thread-local caches. The second usage not-so-much and we need to keep this (in libdns:message.c) for performance reasons.
2021-05-12 21:16:17 +02:00
disp = isc_mem_get(mgr->mctx, sizeof(*disp));
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
*disp = (dns_dispatch_t){ .socktype = type };
checkpoint
1999-06-18 02:01:42 +00:00
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
dns_dispatchmgr_attach(mgr, &disp->mgr);
isc_refcount_init(&disp->references, 1);
implement dns_dispatchmgr_t, replacing dns_dispatchlist_t. Use it throughout the library/server.
2000-05-10 21:34:50 +00:00
ISC_LINK_INIT(disp, link);
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
ISC_LIST_INIT(disp->active);
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
ISC_LIST_INIT(disp->pending);
Creating TCP dispatch now creates/binds the socket Previously, creation of TCP dispatches differed from UDP in that a TCP dispatch was created to attach to an existing socket, whereas a UDP dispatch would be created in a vacuum and sockets would be opened on demand when a transaction was initiated. We are moving as much socket code as possible into the dispatch module, so that it can be replaced with a netmgr version as easily as possible. (This will also have the side effect of making TCP and UDP dispatches more similar.) As a step in that direction, this commit changes dns_dispatch_createtcp() so that it creates the TCP socket.
2020-12-18 14:59:50 -08:00
isc_mutex_init returns 'void'
2018-11-16 15:33:22 +01:00
isc_mutex_init(&disp->lock);
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
checkpoint
1999-06-18 23:54:59 +00:00
disp->magic = DISPATCH_MAGIC;
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
*dispp = disp;
}
/*
spelling
2009-01-17 14:18:27 +00:00
* MUST be unlocked, and not used by anything.
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
*/
static void
apply the modified style
2020-02-13 14:44:37 -08:00
dispatch_free(dns_dispatch_t **dispp) {
General code refactoring - style cleanup - removed NULL checks in places where they are not currently needed - use isc_refcount for dispatch reference counting - revised code flow for readability - remove some #ifdefs that are no longer relevant - remove unused struct members - removed unnecessary function parameters - use C99 struct initialization
2020-12-09 19:44:41 -08:00
dns_dispatch_t *disp = NULL;
dns_dispatchmgr_t *mgr = NULL;
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
REQUIRE(VALID_DISPATCH(*dispp));
disp = *dispp;
*dispp = NULL;
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
disp->magic = 0;
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
mgr = disp->mgr;
REQUIRE(VALID_DISPATCHMGR(mgr));
INSIST(disp->requests == 0);
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
INSIST(ISC_LIST_EMPTY(disp->active));
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
- Make isc_mutex_destroy return void - Make isc_mutexblock_init/destroy return void - Minor cleanups
2018-11-19 10:31:09 +00:00
isc_mutex_destroy(&disp->lock);
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
Replace locked mempools with memory contexts Current mempools are kind of hybrid structures - they serve two purposes: 1. mempool with a lock is basically static sized allocator with pre-allocated free items 2. mempool without a lock is a doubly-linked list of preallocated items The first kind of usage could be easily replaced with jemalloc small sized arena objects and thread-local caches. The second usage not-so-much and we need to keep this (in libdns:message.c) for performance reasons.
2021-05-12 21:16:17 +02:00
isc_mem_put(mgr->mctx, disp, sizeof(*disp));
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
}
isc_result_t
dispatch: Refactor to eliminate dns_dispatchevent - Responses received by the dispatch are no longer sent to the caller via a task event, but via a netmgr-style recv callback. the 'action' parameter to dns_dispatch_addresponse() is now called 'response' and is called directly from udp_recv() or tcp_recv() when a valid response has been received. - All references to isc_task and isc_taskmgr have been removed from dispatch functions. - All references to dns_dispatchevent_t have been removed and the type has been deleted. - Added a task to the resolver response context, to be used for fctx events. - When the caller cancels an operation, the response handler will be called with ISC_R_CANCELED; it can abort immediately since the caller will presumably have taken care of cleanup already. - Cleaned up attach/detach in resquery and request.
2021-07-26 20:23:18 -07:00
dns_dispatch_createtcp(dns_dispatchmgr_t *mgr, const isc_sockaddr_t *localaddr,
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
const isc_sockaddr_t *destaddr, isc_dscp_t dscp,
dns_dispatch_t **dispp) {
General code refactoring - style cleanup - removed NULL checks in places where they are not currently needed - use isc_refcount for dispatch reference counting - revised code flow for readability - remove some #ifdefs that are no longer relevant - remove unused struct members - removed unnecessary function parameters - use C99 struct initialization
2020-12-09 19:44:41 -08:00
dns_dispatch_t *disp = NULL;
Move the UDP buffer pool to the dispatchmgr. UDP parameters are now set by the first call to dns_dispatch_getudp(), subsecquent calls can raise the maximum number of buffers. dns_dispatchmgr_setudp() could be made public. Tidy dispatch_allocate(). entropy and mctx are now gone from dns_qid_t.
2000-09-19 06:59:28 +00:00
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
REQUIRE(VALID_DISPATCHMGR(mgr));
Creating TCP dispatch now creates/binds the socket Previously, creation of TCP dispatches differed from UDP in that a TCP dispatch was created to attach to an existing socket, whereas a UDP dispatch would be created in a vacuum and sockets would be opened on demand when a transaction was initiated. We are moving as much socket code as possible into the dispatch module, so that it can be replaced with a netmgr version as easily as possible. (This will also have the side effect of making TCP and UDP dispatches more similar.) As a step in that direction, this commit changes dns_dispatch_createtcp() so that it creates the TCP socket.
2020-12-18 14:59:50 -08:00
REQUIRE(destaddr != NULL);
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
UNUSED(dscp);
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
LOCK(&mgr->lock);
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
dispatch_allocate(mgr, isc_socktype_tcp, &disp);
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
Creating TCP dispatch now creates/binds the socket Previously, creation of TCP dispatches differed from UDP in that a TCP dispatch was created to attach to an existing socket, whereas a UDP dispatch would be created in a vacuum and sockets would be opened on demand when a transaction was initiated. We are moving as much socket code as possible into the dispatch module, so that it can be replaced with a netmgr version as easily as possible. (This will also have the side effect of making TCP and UDP dispatches more similar.) As a step in that direction, this commit changes dns_dispatch_createtcp() so that it creates the TCP socket.
2020-12-18 14:59:50 -08:00
disp->peer = *destaddr;
if (localaddr != NULL) {
disp->local = *localaddr;
} else {
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
int pf;
pf = isc_sockaddr_pf(destaddr);
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
isc_sockaddr_anyofpf(&disp->local, pf);
isc_sockaddr_setport(&disp->local, 0);
Remove support for shared UDP dispatch sockets Currently the netmgr doesn't support unconnected, shared UDP sockets, so there's no reason to retain that functionality in the dispatcher prior to porting to the netmgr. In this commit, the DNS_DISPATCHATTR_EXCLUSIVE attribute has been removed as it is now non-optional; UDP dispatches are alwasy exclusive. Code implementing non-exclusive UDP dispatches has been removed. dns_dispatch_getentrysocket() now always returns the dispsocket for UDP dispatches and the dispatch socket for TCP dispatches. There is no longer any need to search for existing dispatches from dns_dispatch_getudp(), so the 'mask' option has been removed, and the function renamed to the more descriptive dns_dispatch_createudp().
2020-12-16 01:32:06 -08:00
}
implement dns_dispatchmgr_t, replacing dns_dispatchlist_t. Use it throughout the library/server.
2000-05-10 21:34:50 +00:00
/*
* Append it to the dispatcher list.
*/
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
/* FIXME: There should be a lookup hashtable here */
implement dns_dispatchmgr_t, replacing dns_dispatchlist_t. Use it throughout the library/server.
2000-05-10 21:34:50 +00:00
ISC_LIST_APPEND(mgr->list, disp, link);
UNLOCK(&mgr->lock);
make this (mostly) work with tcp
1999-07-12 23:44:31 +00:00
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
mgr_log(mgr, LVL(90), "dns_dispatch_createtcp: created TCP dispatch %p",
disp);
checkpoint
1999-06-18 23:54:59 +00:00
*dispp = disp;
add dns_dispatch_attach/detach(), and start on serious dispatcher hacking.
1999-06-30 01:33:11 +00:00
s/DNS_R_/ISC_R_/ change for some codes.
2000-04-06 22:03:35 +00:00
return (ISC_R_SUCCESS);
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
}
3939. [func] Improve UPDATE forwarding performance by allowing TCP connections to be shared. [RT #37039]
2014-09-04 10:37:45 +10:00
isc_result_t
4546. [func] Extend the use of const declarations. [RT #43379]
2016-12-30 15:45:08 +11:00
dns_dispatch_gettcp(dns_dispatchmgr_t *mgr, const isc_sockaddr_t *destaddr,
Replace custom isc_boolean_t with C standard bool type
2018-04-17 08:29:14 -07:00
const isc_sockaddr_t *localaddr, bool *connected,
apply the modified style
2020-02-13 14:44:37 -08:00
dns_dispatch_t **dispp) {
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
dns_dispatch_t *disp_connected = NULL;
dns_dispatch_t *disp_fallback = NULL;
isc_result_t result = ISC_R_NOTFOUND;
[master] allow shared TCP sockets when connecting 4041. [func] TCP sockets can now be shared while connecting. (This will be used to enable client-side support of pipelined queries.) [RT #38231]
2015-01-20 17:22:31 -08:00
REQUIRE(VALID_DISPATCHMGR(mgr));
REQUIRE(destaddr != NULL);
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
REQUIRE(connected != NULL);
[master] allow shared TCP sockets when connecting 4041. [func] TCP sockets can now be shared while connecting. (This will be used to enable client-side support of pipelined queries.) [RT #38231]
2015-01-20 17:22:31 -08:00
REQUIRE(dispp != NULL && *dispp == NULL);
LOCK(&mgr->lock);
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
for (dns_dispatch_t *disp = ISC_LIST_HEAD(mgr->list); disp != NULL;
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
disp = ISC_LIST_NEXT(disp, link))
{
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
isc_sockaddr_t sockname;
isc_sockaddr_t peeraddr;
[master] allow shared TCP sockets when connecting 4041. [func] TCP sockets can now be shared while connecting. (This will be used to enable client-side support of pipelined queries.) [RT #38231]
2015-01-20 17:22:31 -08:00
LOCK(&disp->lock);
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
if (disp->handle != NULL) {
sockname = isc_nmhandle_localaddr(disp->handle);
peeraddr = isc_nmhandle_peeraddr(disp->handle);
} else {
sockname = disp->local;
peeraddr = disp->peer;
}
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
/*
* The conditions match:
* 1. socktype is TCP
* 2. destination address is same
* 3. local address is either NULL or same
*/
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
if (disp->socktype == isc_socktype_tcp &&
isc_sockaddr_equal(destaddr, &peeraddr) &&
(localaddr == NULL ||
isc_sockaddr_eqaddr(localaddr, &sockname)))
{
if (atomic_load(&disp->state) ==
DNS_DISPATCHSTATE_CONNECTED) {
/* We found connected dispatch */
disp_connected = disp;
UNLOCK(&disp->lock);
break;
}
/* We found "a" dispatch, store it for later */
if (disp_fallback == NULL) {
disp_fallback = disp;
}
UNLOCK(&disp->lock);
continue;
}
[master] allow shared TCP sockets when connecting 4041. [func] TCP sockets can now be shared while connecting. (This will be used to enable client-side support of pipelined queries.) [RT #38231]
2015-01-20 17:22:31 -08:00
UNLOCK(&disp->lock);
}
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
if (disp_connected != NULL) {
/* We found connected dispatch */
INSIST(disp_connected->handle != NULL);
[master] allow shared TCP sockets when connecting 4041. [func] TCP sockets can now be shared while connecting. (This will be used to enable client-side support of pipelined queries.) [RT #38231]
2015-01-20 17:22:31 -08:00
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
*connected = true;
dns_dispatch_attach(disp_connected, dispp);
result = ISC_R_SUCCESS;
} else if (disp_fallback != NULL) {
/* We found matching dispatch */
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
*connected = false;
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
dns_dispatch_attach(disp_fallback, dispp);
result = ISC_R_SUCCESS;
[master] allow shared TCP sockets when connecting 4041. [func] TCP sockets can now be shared while connecting. (This will be used to enable client-side support of pipelined queries.) [RT #38231]
2015-01-20 17:22:31 -08:00
}
Creating TCP dispatch now creates/binds the socket Previously, creation of TCP dispatches differed from UDP in that a TCP dispatch was created to attach to an existing socket, whereas a UDP dispatch would be created in a vacuum and sockets would be opened on demand when a transaction was initiated. We are moving as much socket code as possible into the dispatch module, so that it can be replaced with a netmgr version as easily as possible. (This will also have the side effect of making TCP and UDP dispatches more similar.) As a step in that direction, this commit changes dns_dispatch_createtcp() so that it creates the TCP socket.
2020-12-18 14:59:50 -08:00
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
UNLOCK(&mgr->lock);
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
return (result);
[master] allow shared TCP sockets when connecting 4041. [func] TCP sockets can now be shared while connecting. (This will be used to enable client-side support of pipelined queries.) [RT #38231]
2015-01-20 17:22:31 -08:00
}
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
isc_result_t
dispatch: Refactor to eliminate dns_dispatchevent - Responses received by the dispatch are no longer sent to the caller via a task event, but via a netmgr-style recv callback. the 'action' parameter to dns_dispatch_addresponse() is now called 'response' and is called directly from udp_recv() or tcp_recv() when a valid response has been received. - All references to isc_task and isc_taskmgr have been removed from dispatch functions. - All references to dns_dispatchevent_t have been removed and the type has been deleted. - Added a task to the resolver response context, to be used for fctx events. - When the caller cancels an operation, the response handler will be called with ISC_R_CANCELED; it can abort immediately since the caller will presumably have taken care of cleanup already. - Cleaned up attach/detach in resquery and request.
2021-07-26 20:23:18 -07:00
dns_dispatch_createudp(dns_dispatchmgr_t *mgr, const isc_sockaddr_t *localaddr,
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
dns_dispatch_t **dispp) {
apply the modified style
2020-02-13 14:44:37 -08:00
isc_result_t result;
2129. [func] Provide a pool of UDP sockets for queries to be made over. See use-queryport-pool, queryport-pool-ports and queryport-pool-updateinterval. [RT #16415]
2007-02-02 02:18:06 +00:00
dns_dispatch_t *disp = NULL;
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
REQUIRE(VALID_DISPATCHMGR(mgr));
REQUIRE(localaddr != NULL);
REQUIRE(dispp != NULL && *dispp == NULL);
LOCK(&mgr->lock);
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
result = dispatch_createudp(mgr, localaddr, &disp);
Remove support for shared UDP dispatch sockets Currently the netmgr doesn't support unconnected, shared UDP sockets, so there's no reason to retain that functionality in the dispatcher prior to porting to the netmgr. In this commit, the DNS_DISPATCHATTR_EXCLUSIVE attribute has been removed as it is now non-optional; UDP dispatches are alwasy exclusive. Code implementing non-exclusive UDP dispatches has been removed. dns_dispatch_getentrysocket() now always returns the dispsocket for UDP dispatches and the dispatch socket for TCP dispatches. There is no longer any need to search for existing dispatches from dns_dispatch_getudp(), so the 'mask' option has been removed, and the function renamed to the more descriptive dns_dispatch_createudp().
2020-12-16 01:32:06 -08:00
if (result == ISC_R_SUCCESS) {
*dispp = disp;
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
}
UNLOCK(&mgr->lock);
Creating TCP dispatch now creates/binds the socket Previously, creation of TCP dispatches differed from UDP in that a TCP dispatch was created to attach to an existing socket, whereas a UDP dispatch would be created in a vacuum and sockets would be opened on demand when a transaction was initiated. We are moving as much socket code as possible into the dispatch module, so that it can be replaced with a netmgr version as easily as possible. (This will also have the side effect of making TCP and UDP dispatches more similar.) As a step in that direction, this commit changes dns_dispatch_createtcp() so that it creates the TCP socket.
2020-12-18 14:59:50 -08:00
return (result);
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
}
static isc_result_t
dispatch: Refactor to eliminate dns_dispatchevent - Responses received by the dispatch are no longer sent to the caller via a task event, but via a netmgr-style recv callback. the 'action' parameter to dns_dispatch_addresponse() is now called 'response' and is called directly from udp_recv() or tcp_recv() when a valid response has been received. - All references to isc_task and isc_taskmgr have been removed from dispatch functions. - All references to dns_dispatchevent_t have been removed and the type has been deleted. - Added a task to the resolver response context, to be used for fctx events. - When the caller cancels an operation, the response handler will be called with ISC_R_CANCELED; it can abort immediately since the caller will presumably have taken care of cleanup already. - Cleaned up attach/detach in resquery and request.
2021-07-26 20:23:18 -07:00
dispatch_createudp(dns_dispatchmgr_t *mgr, const isc_sockaddr_t *localaddr,
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
dns_dispatch_t **dispp) {
Creating TCP dispatch now creates/binds the socket Previously, creation of TCP dispatches differed from UDP in that a TCP dispatch was created to attach to an existing socket, whereas a UDP dispatch would be created in a vacuum and sockets would be opened on demand when a transaction was initiated. We are moving as much socket code as possible into the dispatch module, so that it can be replaced with a netmgr version as easily as possible. (This will also have the side effect of making TCP and UDP dispatches more similar.) As a step in that direction, this commit changes dns_dispatch_createtcp() so that it creates the TCP socket.
2020-12-18 14:59:50 -08:00
isc_result_t result = ISC_R_SUCCESS;
General code refactoring - style cleanup - removed NULL checks in places where they are not currently needed - use isc_refcount for dispatch reference counting - revised code flow for readability - remove some #ifdefs that are no longer relevant - remove unused struct members - removed unnecessary function parameters - use C99 struct initialization
2020-12-09 19:44:41 -08:00
dns_dispatch_t *disp = NULL;
Remove support for shared UDP dispatch sockets Currently the netmgr doesn't support unconnected, shared UDP sockets, so there's no reason to retain that functionality in the dispatcher prior to porting to the netmgr. In this commit, the DNS_DISPATCHATTR_EXCLUSIVE attribute has been removed as it is now non-optional; UDP dispatches are alwasy exclusive. Code implementing non-exclusive UDP dispatches has been removed. dns_dispatch_getentrysocket() now always returns the dispsocket for UDP dispatches and the dispatch socket for TCP dispatches. There is no longer any need to search for existing dispatches from dns_dispatch_getudp(), so the 'mask' option has been removed, and the function renamed to the more descriptive dns_dispatch_createudp().
2020-12-16 01:32:06 -08:00
isc_sockaddr_t sa_any;
style: one-line statement braces, line length, etc
2018-11-08 19:34:51 -08:00
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
dispatch_allocate(mgr, isc_socktype_udp, &disp);
[master] fix PRNG selection in dispatch.c 4058. [bug] UDP dispatches could use the wrong psuedorandom number generator context. [RT #38578]
2015-02-10 13:54:03 -08:00
Remove support for shared UDP dispatch sockets Currently the netmgr doesn't support unconnected, shared UDP sockets, so there's no reason to retain that functionality in the dispatcher prior to porting to the netmgr. In this commit, the DNS_DISPATCHATTR_EXCLUSIVE attribute has been removed as it is now non-optional; UDP dispatches are alwasy exclusive. Code implementing non-exclusive UDP dispatches has been removed. dns_dispatch_getentrysocket() now always returns the dispsocket for UDP dispatches and the dispatch socket for TCP dispatches. There is no longer any need to search for existing dispatches from dns_dispatch_getudp(), so the 'mask' option has been removed, and the function renamed to the more descriptive dns_dispatch_createudp().
2020-12-16 01:32:06 -08:00
/*
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
* Check whether this address/port is available locally.
Remove support for shared UDP dispatch sockets Currently the netmgr doesn't support unconnected, shared UDP sockets, so there's no reason to retain that functionality in the dispatcher prior to porting to the netmgr. In this commit, the DNS_DISPATCHATTR_EXCLUSIVE attribute has been removed as it is now non-optional; UDP dispatches are alwasy exclusive. Code implementing non-exclusive UDP dispatches has been removed. dns_dispatch_getentrysocket() now always returns the dispsocket for UDP dispatches and the dispatch socket for TCP dispatches. There is no longer any need to search for existing dispatches from dns_dispatch_getudp(), so the 'mask' option has been removed, and the function renamed to the more descriptive dns_dispatch_createudp().
2020-12-16 01:32:06 -08:00
*/
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
isc_sockaddr_anyofpf(&sa_any, isc_sockaddr_pf(localaddr));
Remove support for shared UDP dispatch sockets Currently the netmgr doesn't support unconnected, shared UDP sockets, so there's no reason to retain that functionality in the dispatcher prior to porting to the netmgr. In this commit, the DNS_DISPATCHATTR_EXCLUSIVE attribute has been removed as it is now non-optional; UDP dispatches are alwasy exclusive. Code implementing non-exclusive UDP dispatches has been removed. dns_dispatch_getentrysocket() now always returns the dispsocket for UDP dispatches and the dispatch socket for TCP dispatches. There is no longer any need to search for existing dispatches from dns_dispatch_getudp(), so the 'mask' option has been removed, and the function renamed to the more descriptive dns_dispatch_createudp().
2020-12-16 01:32:06 -08:00
if (!isc_sockaddr_eqaddr(&sa_any, localaddr)) {
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
result = isc_nm_checkaddr(localaddr, isc_socktype_udp);
style: one-line statement braces, line length, etc
2018-11-08 19:34:51 -08:00
if (result != ISC_R_SUCCESS) {
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
goto cleanup;
style: one-line statement braces, line length, etc
2018-11-08 19:34:51 -08:00
}
Remove support for shared UDP dispatch sockets Currently the netmgr doesn't support unconnected, shared UDP sockets, so there's no reason to retain that functionality in the dispatcher prior to porting to the netmgr. In this commit, the DNS_DISPATCHATTR_EXCLUSIVE attribute has been removed as it is now non-optional; UDP dispatches are alwasy exclusive. Code implementing non-exclusive UDP dispatches has been removed. dns_dispatch_getentrysocket() now always returns the dispsocket for UDP dispatches and the dispatch socket for TCP dispatches. There is no longer any need to search for existing dispatches from dns_dispatch_getudp(), so the 'mask' option has been removed, and the function renamed to the more descriptive dns_dispatch_createudp().
2020-12-16 01:32:06 -08:00
}
3137. [func] Improve hardware scalability by allowing multiple worker threads to process incoming UDP packets. This can significantly increase query throughput on some systems. [RT #22992]
2011-07-28 04:04:37 +00:00
Remove support for shared UDP dispatch sockets Currently the netmgr doesn't support unconnected, shared UDP sockets, so there's no reason to retain that functionality in the dispatcher prior to porting to the netmgr. In this commit, the DNS_DISPATCHATTR_EXCLUSIVE attribute has been removed as it is now non-optional; UDP dispatches are alwasy exclusive. Code implementing non-exclusive UDP dispatches has been removed. dns_dispatch_getentrysocket() now always returns the dispsocket for UDP dispatches and the dispatch socket for TCP dispatches. There is no longer any need to search for existing dispatches from dns_dispatch_getudp(), so the 'mask' option has been removed, and the function renamed to the more descriptive dns_dispatch_createudp().
2020-12-16 01:32:06 -08:00
if (isc_log_wouldlog(dns_lctx, 90)) {
char addrbuf[ISC_SOCKADDR_FORMATSIZE];
2491. [func] Attempt to re-use a local port if we are already using the port. [RT #18548]
2008-11-12 23:10:57 +00:00
Remove support for shared UDP dispatch sockets Currently the netmgr doesn't support unconnected, shared UDP sockets, so there's no reason to retain that functionality in the dispatcher prior to porting to the netmgr. In this commit, the DNS_DISPATCHATTR_EXCLUSIVE attribute has been removed as it is now non-optional; UDP dispatches are alwasy exclusive. Code implementing non-exclusive UDP dispatches has been removed. dns_dispatch_getentrysocket() now always returns the dispsocket for UDP dispatches and the dispatch socket for TCP dispatches. There is no longer any need to search for existing dispatches from dns_dispatch_getudp(), so the 'mask' option has been removed, and the function renamed to the more descriptive dns_dispatch_createudp().
2020-12-16 01:32:06 -08:00
isc_sockaddr_format(localaddr, addrbuf,
ISC_SOCKADDR_FORMATSIZE);
mgr_log(mgr, LVL(90),
"dispatch_createudp: created UDP dispatch for %s",
addrbuf);
1442. [func] New fuctions for manipulating port lists: dns_portlist_create(), dns_portlist_add(), dns_portlist_remove(), dns_portlist_match(), dns_portlist_attach() and dns_portlist_detach(). 1441. [func] It is now possible to tell dig to bind to a specific source port. 1440. [func] It is now possible to tell named to avoid using certian source ports (avoid-v4-udp-ports, avoid-v6-udp-ports). developer: marka reviewer: explorer
2003-02-26 05:05:16 +00:00
}
Remove support for shared UDP dispatch sockets Currently the netmgr doesn't support unconnected, shared UDP sockets, so there's no reason to retain that functionality in the dispatcher prior to porting to the netmgr. In this commit, the DNS_DISPATCHATTR_EXCLUSIVE attribute has been removed as it is now non-optional; UDP dispatches are alwasy exclusive. Code implementing non-exclusive UDP dispatches has been removed. dns_dispatch_getentrysocket() now always returns the dispsocket for UDP dispatches and the dispatch socket for TCP dispatches. There is no longer any need to search for existing dispatches from dns_dispatch_getudp(), so the 'mask' option has been removed, and the function renamed to the more descriptive dns_dispatch_createudp().
2020-12-16 01:32:06 -08:00
when code is duplicated (which it shoudn't be), it should at least be duplicated consistently
2001-01-07 22:02:48 +00:00
disp->local = *localaddr;
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
/*
* Append it to the dispatcher list.
*/
ISC_LIST_APPEND(mgr->list, disp, link);
mgr_log(mgr, LVL(90), "created UDP dispatcher %p", disp);
*dispp = disp;
3137. [func] Improve hardware scalability by allowing multiple worker threads to process incoming UDP packets. This can significantly increase query throughput on some systems. [RT #22992]
2011-07-28 04:04:37 +00:00
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
return (result);
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
/*
* Error returns.
*/
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
cleanup:
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
dispatch_free(&disp);
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
return (result);
checkpoint
1999-06-16 01:32:31 +00:00
}
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
static void
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
dispatch_destroy(dns_dispatch_t *disp) {
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
dns_dispatchmgr_t *mgr = disp->mgr;
LOCK(&mgr->lock);
ISC_LIST_UNLINK(mgr->list, disp, link);
UNLOCK(&mgr->lock);
dispatch_log(disp, LVL(90), "shutting down; detaching from handle %p",
disp->handle);
if (disp->handle != NULL) {
isc_nmhandle_detach(&disp->handle);
}
dispatch_free(&disp);
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
/*
* Because dispatch uses mgr->mctx, we must detach after freeing
* dispatch, not before.
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
*/
dns_dispatchmgr_detach(&mgr);
}
checkpoint
1999-06-18 23:54:59 +00:00
void
Remove DISPATH_TRACE, RESOLVER_TRACE and REQ_TRACE logging Remove the debugging printfs. (leaving this as a separate commit rather than squashing it so we can restore it in the future if we ever need it again.)
2021-09-09 15:25:08 -07:00
dns_dispatch_attach(dns_dispatch_t *disp, dns_dispatch_t **dispp) {
s/dns_dispatch_destroy()/dns_dispatch_detach()/g
1999-07-13 00:25:21 +00:00
REQUIRE(VALID_DISPATCH(disp));
REQUIRE(dispp != NULL && *dispp == NULL);
Remove DISPATH_TRACE, RESOLVER_TRACE and REQ_TRACE logging Remove the debugging printfs. (leaving this as a separate commit rather than squashing it so we can restore it in the future if we ever need it again.)
2021-09-09 15:25:08 -07:00
isc_refcount_increment(&disp->references);
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
s/dns_dispatch_destroy()/dns_dispatch_detach()/g
1999-07-13 00:25:21 +00:00
*dispp = disp;
}
void
Remove DISPATH_TRACE, RESOLVER_TRACE and REQ_TRACE logging Remove the debugging printfs. (leaving this as a separate commit rather than squashing it so we can restore it in the future if we ever need it again.)
2021-09-09 15:25:08 -07:00
dns_dispatch_detach(dns_dispatch_t **dispp) {
General code refactoring - style cleanup - removed NULL checks in places where they are not currently needed - use isc_refcount for dispatch reference counting - revised code flow for readability - remove some #ifdefs that are no longer relevant - remove unused struct members - removed unnecessary function parameters - use C99 struct initialization
2020-12-09 19:44:41 -08:00
dns_dispatch_t *disp = NULL;
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
uint_fast32_t ref;
checkpoint
1999-06-18 23:54:59 +00:00
REQUIRE(dispp != NULL && VALID_DISPATCH(*dispp));
disp = *dispp;
*dispp = NULL;
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
ref = isc_refcount_decrement(&disp->references);
dispatch_log(disp, LVL(90), "detach: refcount %" PRIuFAST32, ref - 1);
if (ref == 1) {
LOCK(&disp->lock);
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
INSIST(ISC_LIST_EMPTY(disp->pending));
INSIST(ISC_LIST_EMPTY(disp->active));
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
UNLOCK(&disp->lock);
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
dispatch_destroy(disp);
Use clang-tidy to add curly braces around one-line statements The command used to reformat the files in this commit was: ./util/run-clang-tidy \ -clang-tidy-binary clang-tidy-11 -clang-apply-replacements-binary clang-apply-replacements-11 \ -checks=-*,readability-braces-around-statements \ -j 9 \ -fix \ -format \ -style=file \ -quiet clang-format -i --style=format $(git ls-files '*.c' '*.h') uncrustify -c .uncrustify.cfg --replace --no-backup $(git ls-files '*.c' '*.h') clang-format -i --style=format $(git ls-files '*.c' '*.h')
2020-02-13 21:48:23 +01:00
}
checkpoint
1999-06-16 01:32:31 +00:00
}
fix many bugs.
1999-07-22 01:34:31 +00:00
isc_result_t
refactor dispatch cancellation Renamed some functions for clarity and readability: - dns_dispatch_addresponse() -> dns_dispatch_add() - dns_dispatch_removeresponse() -> dns_dispatch_done() The dns_dispatch_cancel() function now calls dns_dispatch_done() directly, so it is no longer ever necessary to call both functions. dns_dispatch_cancel() is used to terminate dispatch connections that are still pending, while dns_dispatch_done() is used when they are complete.
2021-10-03 15:15:50 -07:00
dns_dispatch_add(dns_dispatch_t *disp, unsigned int options,
unsigned int timeout, const isc_sockaddr_t *dest,
dispatch_cb_t connected, dispatch_cb_t sent,
dispatch_cb_t response, void *arg, dns_messageid_t *idp,
dns_dispentry_t **resp) {
General code refactoring - style cleanup - removed NULL checks in places where they are not currently needed - use isc_refcount for dispatch reference counting - revised code flow for readability - remove some #ifdefs that are no longer relevant - remove unused struct members - removed unnecessary function parameters - use C99 struct initialization
2020-12-09 19:44:41 -08:00
dns_dispentry_t *res = NULL;
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
dns_qid_t *qid = NULL;
apply the modified style
2020-02-13 14:44:37 -08:00
in_port_t localport = 0;
dns_messageid_t id;
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
unsigned int bucket;
General code refactoring - style cleanup - removed NULL checks in places where they are not currently needed - use isc_refcount for dispatch reference counting - revised code flow for readability - remove some #ifdefs that are no longer relevant - remove unused struct members - removed unnecessary function parameters - use C99 struct initialization
2020-12-09 19:44:41 -08:00
bool ok = false;
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
int i = 0;
dispatch: Remove 'timeout' callback - It is no longer necessary to pass a 'timeout' callback to dns_dispatch_addresponse(); timeouts are handled directly by the 'response' callback instead. - The netmgr handle is no longer passed to dispatch callbacks, since they don't (and can't) use it. instead, dispatch_cb_t now takes a result code, region, and argument. - Cleaned up timeout-related tests in dispatch_test.c
2021-08-03 15:27:06 +02:00
dispatch_cb_t oldest_response = NULL;
checkpoint
1999-06-18 02:01:42 +00:00
REQUIRE(VALID_DISPATCH(disp));
REQUIRE(dest != NULL);
REQUIRE(resp != NULL && *resp == NULL);
REQUIRE(idp != NULL);
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
REQUIRE(disp->socktype == isc_socktype_tcp ||
disp->socktype == isc_socktype_udp);
checkpoint
1999-06-18 02:01:42 +00:00
LOCK(&disp->lock);
Dispatch API simplification - Many dispatch attributes can be set implicitly instead of being passed in. we can infer whether to set DNS_DISPATCHATTR_TCP or _UDP from whether we're calling dns_dispatch_createtcp() or _createudp(). we can also infer DNS_DISPATCHATTR_IPV4 or _IPV6 from the addresses or the socket that were passed in. - We no longer use dup'd sockets in UDP dispatches, so the 'dup_socket' parameter has been removed from dns_dispatch_createudp(), along with the code implementing it. also removed isc_socket_dup() since it no longer has any callers. - The 'buffersize' parameter was ignored and has now been removed; buffersize is now fixed at 4096. - Maxbuffers and maxrequests don't need to be passed in on every call to dns_dispatch_createtcp() and _createudp(). In all current uses, the value for mgr->maxbuffers will either be raised once from its default of 20000 to 32768, or else left alone. (passing in a value lower than 20000 does not lower it.) there isn't enough difference between these values for there to be any need to configure this. The value for disp->maxrequests controls both the quota of concurrent requests for a dispatch and also the size of the dispatch socket memory pool. it's not clear that this quota is necessary at all. the memory pool size currently starts at 32768, but is sometimes lowered to 4096, which is definitely unnecessary. This commit sets both values permanently to 32768. - Previously TCP dispatches allocated their own separate QID table, which didn't incorporate a port table. this commit removes per-dispatch QID tables and shares the same table between all dispatches. since dispatches are created for each TCP socket, this may speed up the dispatch allocation process. there may be a slight increase in lock contention since all dispatches are sharing a single QID table, but since TCP sockets are used less often than UDP sockets (which were already sharing a QID table), it should not be a substantial change. - The dispatch port table was being used to determine whether a port was already in use; if so, then a UDP socket would be bound with REUSEADDR. this commit removes the port table, and always binds UDP sockets that way.
2020-12-17 00:43:00 -08:00
if (disp->requests >= DNS_DISPATCH_MAXREQUESTS) {
checkpoint
1999-07-09 02:47:55 +00:00
UNLOCK(&disp->lock);
cleanup a few return codes
1999-07-10 00:15:41 +00:00
return (ISC_R_QUOTA);
checkpoint
1999-07-09 02:47:55 +00:00
}
Dispatch API simplification - Many dispatch attributes can be set implicitly instead of being passed in. we can infer whether to set DNS_DISPATCHATTR_TCP or _UDP from whether we're calling dns_dispatch_createtcp() or _createudp(). we can also infer DNS_DISPATCHATTR_IPV4 or _IPV6 from the addresses or the socket that were passed in. - We no longer use dup'd sockets in UDP dispatches, so the 'dup_socket' parameter has been removed from dns_dispatch_createudp(), along with the code implementing it. also removed isc_socket_dup() since it no longer has any callers. - The 'buffersize' parameter was ignored and has now been removed; buffersize is now fixed at 4096. - Maxbuffers and maxrequests don't need to be passed in on every call to dns_dispatch_createtcp() and _createudp(). In all current uses, the value for mgr->maxbuffers will either be raised once from its default of 20000 to 32768, or else left alone. (passing in a value lower than 20000 does not lower it.) there isn't enough difference between these values for there to be any need to configure this. The value for disp->maxrequests controls both the quota of concurrent requests for a dispatch and also the size of the dispatch socket memory pool. it's not clear that this quota is necessary at all. the memory pool size currently starts at 32768, but is sometimes lowered to 4096, which is definitely unnecessary. This commit sets both values permanently to 32768. - Previously TCP dispatches allocated their own separate QID table, which didn't incorporate a port table. this commit removes per-dispatch QID tables and shares the same table between all dispatches. since dispatches are created for each TCP socket, this may speed up the dispatch allocation process. there may be a slight increase in lock contention since all dispatches are sharing a single QID table, but since TCP sockets are used less often than UDP sockets (which were already sharing a QID table), it should not be a substantial change. - The dispatch port table was being used to determine whether a port was already in use; if so, then a UDP socket would be bound with REUSEADDR. this commit removes the port table, and always binds UDP sockets that way.
2020-12-17 00:43:00 -08:00
qid = disp->mgr->qid;
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
if (disp->socktype == isc_socktype_udp &&
apply the modified style
2020-02-13 14:44:37 -08:00
disp->nsockets > DNS_DISPATCH_SOCKSQUOTA)
{
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
dns_dispentry_t *oldest = NULL;
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
/*
* Kill oldest outstanding query if the number of sockets
* exceeds the quota to keep the room for new queries.
*/
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
oldest = ISC_LIST_HEAD(disp->active);
if (oldest != NULL) {
dispatch: Refactor to eliminate dns_dispatchevent - Responses received by the dispatch are no longer sent to the caller via a task event, but via a netmgr-style recv callback. the 'action' parameter to dns_dispatch_addresponse() is now called 'response' and is called directly from udp_recv() or tcp_recv() when a valid response has been received. - All references to isc_task and isc_taskmgr have been removed from dispatch functions. - All references to dns_dispatchevent_t have been removed and the type has been deleted. - Added a task to the resolver response context, to be used for fctx events. - When the caller cancels an operation, the response handler will be called with ISC_R_CANCELED; it can abort immediately since the caller will presumably have taken care of cleanup already. - Cleaned up attach/detach in resquery and request.
2021-07-26 20:23:18 -07:00
oldest_response = oldest->response;
General code refactoring - style cleanup - removed NULL checks in places where they are not currently needed - use isc_refcount for dispatch reference counting - revised code flow for readability - remove some #ifdefs that are no longer relevant - remove unused struct members - removed unnecessary function parameters - use C99 struct initialization
2020-12-09 19:44:41 -08:00
inc_stats(disp->mgr, dns_resstatscounter_dispabort);
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
}
}
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
res = isc_mem_get(disp->mgr->mctx, sizeof(*res));
*res = (dns_dispentry_t){ .port = localport,
.timeout = timeout,
.peer = *dest,
.connected = connected,
.sent = sent,
dispatch: Refactor to eliminate dns_dispatchevent - Responses received by the dispatch are no longer sent to the caller via a task event, but via a netmgr-style recv callback. the 'action' parameter to dns_dispatch_addresponse() is now called 'response' and is called directly from udp_recv() or tcp_recv() when a valid response has been received. - All references to isc_task and isc_taskmgr have been removed from dispatch functions. - All references to dns_dispatchevent_t have been removed and the type has been deleted. - Added a task to the resolver response context, to be used for fctx events. - When the caller cancels an operation, the response handler will be called with ISC_R_CANCELED; it can abort immediately since the caller will presumably have taken care of cleanup already. - Cleaned up attach/detach in resquery and request.
2021-07-26 20:23:18 -07:00
.response = response,
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
.arg = arg };
isc_refcount_init(&res->references, 1);
ISC_LINK_INIT(res, link);
ISC_LINK_INIT(res, alink);
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
ISC_LINK_INIT(res, plink);
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
ISC_LINK_INIT(res, rlink);
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
if (disp->socktype == isc_socktype_udp) {
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
isc_result_t result = setup_socket(disp, res, dest, &localport);
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
if (result != ISC_R_SUCCESS) {
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
isc_mem_put(disp->mgr->mctx, res, sizeof(*res));
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
UNLOCK(&disp->lock);
2541. [bug] Conditionally update dispatch manager statistics. [RT #19247]
2009-01-31 00:10:24 +00:00
inc_stats(disp->mgr, dns_resstatscounter_dispsockfail);
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
return (result);
}
}
checkpoint
1999-06-18 02:01:42 +00:00
/*
General code refactoring - style cleanup - removed NULL checks in places where they are not currently needed - use isc_refcount for dispatch reference counting - revised code flow for readability - remove some #ifdefs that are no longer relevant - remove unused struct members - removed unnecessary function parameters - use C99 struct initialization
2020-12-09 19:44:41 -08:00
* Try somewhat hard to find a unique ID. Start with
* a random number unless DNS_DISPATCHOPT_FIXEDID is set,
* in which case we start with the ID passed in via *idp.
checkpoint
1999-06-18 02:01:42 +00:00
*/
Replace all random functions with isc_random, isc_random_buf and isc_random_uniform API. The three functions has been modeled after the arc4random family of functions, and they will always return random bytes. The isc_random family of functions internally use these CSPRNG (if available): 1. getrandom() libc call (might be available on Linux and Solaris) 2. SYS_getrandom syscall (might be available on Linux, detected at runtime) 3. arc4random(), arc4random_buf() and arc4random_uniform() (available on BSDs and Mac OS X) 4. crypto library function: 4a. RAND_bytes in case OpenSSL 4b. pkcs_C_GenerateRandom() in case PKCS#11 library
2018-04-22 14:56:28 +02:00
if ((options & DNS_DISPATCHOPT_FIXEDID) != 0) {
3961. [bug] Forwarding of SIG(0) signed UPDATE messages failed with BADSIG. [RT #37216]
2014-10-01 07:24:16 +10:00
id = *idp;
Replace all random functions with isc_random, isc_random_buf and isc_random_uniform API. The three functions has been modeled after the arc4random family of functions, and they will always return random bytes. The isc_random family of functions internally use these CSPRNG (if available): 1. getrandom() libc call (might be available on Linux and Solaris) 2. SYS_getrandom syscall (might be available on Linux, detected at runtime) 3. arc4random(), arc4random_buf() and arc4random_uniform() (available on BSDs and Mac OS X) 4. crypto library function: 4a. RAND_bytes in case OpenSSL 4b. pkcs_C_GenerateRandom() in case PKCS#11 library
2018-04-22 14:56:28 +02:00
} else {
Change isc_random() to be just PRNG, and add isc_nonce_buf() that uses CSPRNG This commit reverts the previous change to use system provided entropy, as (SYS_)getrandom is very slow on Linux because it is a syscall. The change introduced in this commit adds a new call isc_nonce_buf that uses CSPRNG from cryptographic library provider to generate secure data that can be and must be used for generating nonces. Example usage would be DNS cookies. The isc_random() API has been changed to use fast PRNG that is not cryptographically secure, but runs entirely in user space. Two contestants have been considered xoroshiro family of the functions by Villa&Blackman and PCG by O'Neill. After a consideration the xoshiro128starstar function has been used as uint32_t random number provider because it is very fast and has good enough properties for our usage pattern. The other change introduced in the commit is the more extensive usage of isc_random_uniform in places where the usage pattern was isc_random() % n to prevent modulo bias. For usage patterns where only 16 or 8 bits are needed (DNS Message ID), the isc_random() functions has been renamed to isc_random32(), and isc_random16() and isc_random8() functions have been introduced by &-ing the isc_random32() output with 0xffff and 0xff. Please note that the functions that uses stripped down bit count doesn't pass our NIST SP 800-22 based random test.
2018-05-28 15:22:23 +02:00
id = (dns_messageid_t)isc_random16();
Replace all random functions with isc_random, isc_random_buf and isc_random_uniform API. The three functions has been modeled after the arc4random family of functions, and they will always return random bytes. The isc_random family of functions internally use these CSPRNG (if available): 1. getrandom() libc call (might be available on Linux and Solaris) 2. SYS_getrandom syscall (might be available on Linux, detected at runtime) 3. arc4random(), arc4random_buf() and arc4random_uniform() (available on BSDs and Mac OS X) 4. crypto library function: 4a. RAND_bytes in case OpenSSL 4b. pkcs_C_GenerateRandom() in case PKCS#11 library
2018-04-22 14:56:28 +02:00
}
General code refactoring - style cleanup - removed NULL checks in places where they are not currently needed - use isc_refcount for dispatch reference counting - revised code flow for readability - remove some #ifdefs that are no longer relevant - remove unused struct members - removed unnecessary function parameters - use C99 struct initialization
2020-12-09 19:44:41 -08:00
LOCK(&qid->lock);
silence compiler warning
2014-01-09 15:57:59 +11:00
do {
General code refactoring - style cleanup - removed NULL checks in places where they are not currently needed - use isc_refcount for dispatch reference counting - revised code flow for readability - remove some #ifdefs that are no longer relevant - remove unused struct members - removed unnecessary function parameters - use C99 struct initialization
2020-12-09 19:44:41 -08:00
dns_dispentry_t *entry = NULL;
[master] dispatch.c race 3695. [bug] Address a possible race in dispatch.c. [RT #35107]
2013-12-23 09:50:18 -08:00
bucket = dns_hash(qid, dest, id, localport);
General code refactoring - style cleanup - removed NULL checks in places where they are not currently needed - use isc_refcount for dispatch reference counting - revised code flow for readability - remove some #ifdefs that are no longer relevant - remove unused struct members - removed unnecessary function parameters - use C99 struct initialization
2020-12-09 19:44:41 -08:00
entry = entry_search(qid, dest, id, localport, bucket);
if (entry == NULL) {
Replace custom isc_boolean_t with C standard bool type
2018-04-17 08:29:14 -07:00
ok = true;
checkpoint
1999-06-18 02:01:42 +00:00
break;
}
469. [bug] "query-source address * port 53;" now works.
2000-09-18 04:50:05 +00:00
id += qid->qid_increment;
make dispatcher hash sizes be primes, and provide an increment to help resolve hash collisions
1999-12-15 17:14:52 +00:00
id &= 0x0000ffff;
silence compiler warning
2014-01-09 15:57:59 +11:00
} while (i++ < 64);
Improve locking performance in dispatch.c 3316. [tuning] Improved locking performance when recursing. [RT #28836] - Use one lock per memory pool instead of associating them all with a single 'pool_lock' in the dispatch manager. - Reduce the critical sections for qid->lock in get_dispsocket(), deref_portentry() and dns_dispatch_addresponse2(). - Added a memory pool for socket events in dns_dispatch_t. - Add an isc_socketevent_t member in the resquery_t object, and use it with isc_socket_sendto2() instead of using isc_socket_sendto() - Tuned the memory pools in dispatch.c for better performance under load
2012-04-28 14:52:28 -07:00
UNLOCK(&qid->lock);
checkpoint
1999-06-18 02:01:42 +00:00
if (!ok) {
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
isc_mem_put(disp->mgr->mctx, res, sizeof(*res));
checkpoint
1999-06-18 02:01:42 +00:00
UNLOCK(&disp->lock);
s/DNS_R_/ISC_R_/ change for some codes.
2000-04-06 22:03:35 +00:00
return (ISC_R_NOMORE);
checkpoint
1999-06-18 02:01:42 +00:00
}
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
dns_dispatch_attach(disp, &res->disp);
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
res->id = id;
res->bucket = bucket;
General code refactoring - style cleanup - removed NULL checks in places where they are not currently needed - use isc_refcount for dispatch reference counting - revised code flow for readability - remove some #ifdefs that are no longer relevant - remove unused struct members - removed unnecessary function parameters - use C99 struct initialization
2020-12-09 19:44:41 -08:00
res->magic = RESPONSE_MAGIC;
checkpoint
1999-06-18 02:01:42 +00:00
snapshot
1999-07-09 00:51:08 +00:00
disp->requests++;
General code refactoring - style cleanup - removed NULL checks in places where they are not currently needed - use isc_refcount for dispatch reference counting - revised code flow for readability - remove some #ifdefs that are no longer relevant - remove unused struct members - removed unnecessary function parameters - use C99 struct initialization
2020-12-09 19:44:41 -08:00
Improve locking performance in dispatch.c 3316. [tuning] Improved locking performance when recursing. [RT #28836] - Use one lock per memory pool instead of associating them all with a single 'pool_lock' in the dispatch manager. - Reduce the critical sections for qid->lock in get_dispsocket(), deref_portentry() and dns_dispatch_addresponse2(). - Added a memory pool for socket events in dns_dispatch_t. - Add an isc_socketevent_t member in the resquery_t object, and use it with isc_socket_sendto2() instead of using isc_socket_sendto() - Tuned the memory pools in dispatch.c for better performance under load
2012-04-28 14:52:28 -07:00
LOCK(&qid->lock);
469. [bug] "query-source address * port 53;" now works.
2000-09-18 04:50:05 +00:00
ISC_LIST_APPEND(qid->qid_table[bucket], res, link);
UNLOCK(&qid->lock);
checkpoint
1999-06-18 02:01:42 +00:00
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
inc_stats(disp->mgr, (disp->socktype == isc_socktype_udp)
? dns_resstatscounter_disprequdp
: dns_resstatscounter_dispreqtcp);
2375. [security] Fully randomize UDP query ports to improve forgery resilience. [RT #17949, #18098]
2008-06-23 19:41:20 +00:00
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
ISC_LIST_APPEND(disp->active, res, alink);
snapshot
1999-07-09 00:51:08 +00:00
checkpoint
1999-06-18 02:01:42 +00:00
UNLOCK(&disp->lock);
dispatch: Refactor to eliminate dns_dispatchevent - Responses received by the dispatch are no longer sent to the caller via a task event, but via a netmgr-style recv callback. the 'action' parameter to dns_dispatch_addresponse() is now called 'response' and is called directly from udp_recv() or tcp_recv() when a valid response has been received. - All references to isc_task and isc_taskmgr have been removed from dispatch functions. - All references to dns_dispatchevent_t have been removed and the type has been deleted. - Added a task to the resolver response context, to be used for fctx events. - When the caller cancels an operation, the response handler will be called with ISC_R_CANCELED; it can abort immediately since the caller will presumably have taken care of cleanup already. - Cleaned up attach/detach in resquery and request.
2021-07-26 20:23:18 -07:00
if (oldest_response != NULL) {
dispatch: Remove 'timeout' callback - It is no longer necessary to pass a 'timeout' callback to dns_dispatch_addresponse(); timeouts are handled directly by the 'response' callback instead. - The netmgr handle is no longer passed to dispatch callbacks, since they don't (and can't) use it. instead, dispatch_cb_t now takes a result code, region, and argument. - Cleaned up timeout-related tests in dispatch_test.c
2021-08-03 15:27:06 +02:00
oldest_response(ISC_R_CANCELED, NULL, res->arg);
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
}
Remove support for shared UDP dispatch sockets Currently the netmgr doesn't support unconnected, shared UDP sockets, so there's no reason to retain that functionality in the dispatcher prior to porting to the netmgr. In this commit, the DNS_DISPATCHATTR_EXCLUSIVE attribute has been removed as it is now non-optional; UDP dispatches are alwasy exclusive. Code implementing non-exclusive UDP dispatches has been removed. dns_dispatch_getentrysocket() now always returns the dispsocket for UDP dispatches and the dispatch socket for TCP dispatches. There is no longer any need to search for existing dispatches from dns_dispatch_getudp(), so the 'mask' option has been removed, and the function renamed to the more descriptive dns_dispatch_createudp().
2020-12-16 01:32:06 -08:00
checkpoint
1999-06-18 02:01:42 +00:00
*idp = id;
*resp = res;
s/DNS_R_/ISC_R_/ change for some codes.
2000-04-06 22:03:35 +00:00
return (ISC_R_SUCCESS);
checkpoint
1999-06-16 01:32:31 +00:00
}
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
void
dispatch_getnext(dns_dispatch_t *disp, dns_dispentry_t *resp, int32_t timeout) {
silence compiler warning Silence a warning about a signed/unsigned integer comparison.
2021-10-18 12:49:40 -07:00
REQUIRE(timeout <= (int32_t)UINT16_MAX);
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
switch (disp->socktype) {
case isc_socktype_udp:
dispentry_attach(resp, &(dns_dispentry_t *){ NULL });
if (timeout > 0) {
isc_nmhandle_settimeout(resp->handle, timeout);
}
isc_nm_read(resp->handle, udp_recv, resp);
break;
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
case isc_socktype_tcp:
dns_dispatch_attach(disp, &(dns_dispatch_t *){ NULL });
if (timeout > 0) {
isc_nmhandle_settimeout(disp->handle, timeout);
}
isc_nm_read(disp->handle, tcp_recv, disp);
break;
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
default:
INSIST(0);
ISC_UNREACHABLE();
}
}
4408. [func] Continue waiting for expected response when we the response we get does not match the request. [RT #41026]
2016-07-11 13:36:16 +10:00
isc_result_t
dispatch: Refactor to eliminate dns_dispatchevent - Responses received by the dispatch are no longer sent to the caller via a task event, but via a netmgr-style recv callback. the 'action' parameter to dns_dispatch_addresponse() is now called 'response' and is called directly from udp_recv() or tcp_recv() when a valid response has been received. - All references to isc_task and isc_taskmgr have been removed from dispatch functions. - All references to dns_dispatchevent_t have been removed and the type has been deleted. - Added a task to the resolver response context, to be used for fctx events. - When the caller cancels an operation, the response handler will be called with ISC_R_CANCELED; it can abort immediately since the caller will presumably have taken care of cleanup already. - Cleaned up attach/detach in resquery and request.
2021-07-26 20:23:18 -07:00
dns_dispatch_getnext(dns_dispentry_t *resp) {
General code refactoring - style cleanup - removed NULL checks in places where they are not currently needed - use isc_refcount for dispatch reference counting - revised code flow for readability - remove some #ifdefs that are no longer relevant - remove unused struct members - removed unnecessary function parameters - use C99 struct initialization
2020-12-09 19:44:41 -08:00
dns_dispatch_t *disp = NULL;
dispatch: Enforce original timeout when calling _getnext() udp_recv() will call dispatch_getnext() if the message received is invalid or doesn't match; we need to reduce the timeout each time this happens so we can't be starved forever by someone sending garbage packets.
2021-10-01 12:53:31 -07:00
int32_t timeout;
4408. [func] Continue waiting for expected response when we the response we get does not match the request. [RT #41026]
2016-07-11 13:36:16 +10:00
REQUIRE(VALID_RESPONSE(resp));
disp = resp->disp;
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
4408. [func] Continue waiting for expected response when we the response we get does not match the request. [RT #41026]
2016-07-11 13:36:16 +10:00
REQUIRE(VALID_DISPATCH(disp));
dispatch: Enforce original timeout when calling _getnext() udp_recv() will call dispatch_getnext() if the message received is invalid or doesn't match; we need to reduce the timeout each time this happens so we can't be starved forever by someone sending garbage packets.
2021-10-01 12:53:31 -07:00
if (disp->socktype == isc_socktype_udp) {
timeout = resp->timeout - dispentry_runtime(resp);
if (timeout <= 0) {
return (ISC_R_TIMEDOUT);
}
} else {
timeout = -1;
}
4408. [func] Continue waiting for expected response when we the response we get does not match the request. [RT #41026]
2016-07-11 13:36:16 +10:00
LOCK(&disp->lock);
dispatch: Enforce original timeout when calling _getnext() udp_recv() will call dispatch_getnext() if the message received is invalid or doesn't match; we need to reduce the timeout each time this happens so we can't be starved forever by someone sending garbage packets.
2021-10-01 12:53:31 -07:00
dispatch_getnext(disp, resp, timeout);
4408. [func] Continue waiting for expected response when we the response we get does not match the request. [RT #41026]
2016-07-11 13:36:16 +10:00
UNLOCK(&disp->lock);
return (ISC_R_SUCCESS);
}
checkpoint
1999-06-18 02:01:42 +00:00
void
refactor dispatch cancellation Renamed some functions for clarity and readability: - dns_dispatch_addresponse() -> dns_dispatch_add() - dns_dispatch_removeresponse() -> dns_dispatch_done() The dns_dispatch_cancel() function now calls dns_dispatch_done() directly, so it is no longer ever necessary to call both functions. dns_dispatch_cancel() is used to terminate dispatch connections that are still pending, while dns_dispatch_done() is used when they are complete.
2021-10-03 15:15:50 -07:00
dns_dispatch_cancel(dns_dispentry_t **respp) {
dns_dispentry_t *resp = NULL;
Lock dispatch when canceling connect When canceling pending connections, the disp->pending list was accessed unlocked.
2021-10-18 11:49:56 +02:00
dns_dispatch_t *disp = NULL;
refactor dispatch cancellation Renamed some functions for clarity and readability: - dns_dispatch_addresponse() -> dns_dispatch_add() - dns_dispatch_removeresponse() -> dns_dispatch_done() The dns_dispatch_cancel() function now calls dns_dispatch_done() directly, so it is no longer ever necessary to call both functions. dns_dispatch_cancel() is used to terminate dispatch connections that are still pending, while dns_dispatch_done() is used when they are complete.
2021-10-03 15:15:50 -07:00
REQUIRE(respp != NULL);
resp = *respp;
Correctly call dns_dispatch_done() in dns_dispatch_cancel() Pass '&resp' rather than 'respp' as we have already cleared '*respp' when we took ownership of 'resp'.
2021-10-06 18:32:19 +11:00
*respp = NULL;
refactor dispatch cancellation Renamed some functions for clarity and readability: - dns_dispatch_addresponse() -> dns_dispatch_add() - dns_dispatch_removeresponse() -> dns_dispatch_done() The dns_dispatch_cancel() function now calls dns_dispatch_done() directly, so it is no longer ever necessary to call both functions. dns_dispatch_cancel() is used to terminate dispatch connections that are still pending, while dns_dispatch_done() is used when they are complete.
2021-10-03 15:15:50 -07:00
REQUIRE(VALID_RESPONSE(resp));
Lock dispatch when canceling connect When canceling pending connections, the disp->pending list was accessed unlocked.
2021-10-18 11:49:56 +02:00
disp = resp->disp;
refactor dispatch cancellation Renamed some functions for clarity and readability: - dns_dispatch_addresponse() -> dns_dispatch_add() - dns_dispatch_removeresponse() -> dns_dispatch_done() The dns_dispatch_cancel() function now calls dns_dispatch_done() directly, so it is no longer ever necessary to call both functions. dns_dispatch_cancel() is used to terminate dispatch connections that are still pending, while dns_dispatch_done() is used when they are complete.
2021-10-03 15:15:50 -07:00
resp->canceled = true;
/* Connected UDP. */
if (resp->handle != NULL) {
isc_nm_cancelread(resp->handle);
goto done;
}
Lock dispatch when canceling connect When canceling pending connections, the disp->pending list was accessed unlocked.
2021-10-18 11:49:56 +02:00
LOCK(&disp->lock);
refactor dispatch cancellation Renamed some functions for clarity and readability: - dns_dispatch_addresponse() -> dns_dispatch_add() - dns_dispatch_removeresponse() -> dns_dispatch_done() The dns_dispatch_cancel() function now calls dns_dispatch_done() directly, so it is no longer ever necessary to call both functions. dns_dispatch_cancel() is used to terminate dispatch connections that are still pending, while dns_dispatch_done() is used when they are complete.
2021-10-03 15:15:50 -07:00
/* TCP pending connection. */
if (ISC_LINK_LINKED(resp, plink)) {
dns_dispentry_t *copy = resp;
Lock dispatch when canceling connect When canceling pending connections, the disp->pending list was accessed unlocked.
2021-10-18 11:49:56 +02:00
ISC_LIST_UNLINK(disp->pending, resp, plink);
refactor dispatch cancellation Renamed some functions for clarity and readability: - dns_dispatch_addresponse() -> dns_dispatch_add() - dns_dispatch_removeresponse() -> dns_dispatch_done() The dns_dispatch_cancel() function now calls dns_dispatch_done() directly, so it is no longer ever necessary to call both functions. dns_dispatch_cancel() is used to terminate dispatch connections that are still pending, while dns_dispatch_done() is used when they are complete.
2021-10-03 15:15:50 -07:00
if (resp->connected != NULL) {
resp->connected(ISC_R_CANCELED, NULL, resp->arg);
}
/*
* We need to detach twice if we were pending
* connection - once to take the place of the
* detach in tcp_connected() or udp_connected()
* that we won't reach, and again later in
* dns_dispatch_done().
*/
dispentry_detach(&copy);
Lock dispatch when canceling connect When canceling pending connections, the disp->pending list was accessed unlocked.
2021-10-18 11:49:56 +02:00
UNLOCK(&disp->lock);
refactor dispatch cancellation Renamed some functions for clarity and readability: - dns_dispatch_addresponse() -> dns_dispatch_add() - dns_dispatch_removeresponse() -> dns_dispatch_done() The dns_dispatch_cancel() function now calls dns_dispatch_done() directly, so it is no longer ever necessary to call both functions. dns_dispatch_cancel() is used to terminate dispatch connections that are still pending, while dns_dispatch_done() is used when they are complete.
2021-10-03 15:15:50 -07:00
goto done;
}
/*
* Connected TCP, or unconnected UDP.
*
* If TCP, we don't want to cancel the dispatch
* unless this is the last resp waiting.
*/
if (ISC_LINK_LINKED(resp, alink)) {
Lock dispatch when canceling connect When canceling pending connections, the disp->pending list was accessed unlocked.
2021-10-18 11:49:56 +02:00
ISC_LIST_UNLINK(disp->active, resp, alink);
if (ISC_LIST_EMPTY(disp->active) && disp->handle != NULL) {
isc_nm_cancelread(disp->handle);
refactor dispatch cancellation Renamed some functions for clarity and readability: - dns_dispatch_addresponse() -> dns_dispatch_add() - dns_dispatch_removeresponse() -> dns_dispatch_done() The dns_dispatch_cancel() function now calls dns_dispatch_done() directly, so it is no longer ever necessary to call both functions. dns_dispatch_cancel() is used to terminate dispatch connections that are still pending, while dns_dispatch_done() is used when they are complete.
2021-10-03 15:15:50 -07:00
} else if (resp->response != NULL) {
resp->response(ISC_R_CANCELED, NULL, resp->arg);
}
}
Lock dispatch when canceling connect When canceling pending connections, the disp->pending list was accessed unlocked.
2021-10-18 11:49:56 +02:00
UNLOCK(&disp->lock);
refactor dispatch cancellation Renamed some functions for clarity and readability: - dns_dispatch_addresponse() -> dns_dispatch_add() - dns_dispatch_removeresponse() -> dns_dispatch_done() The dns_dispatch_cancel() function now calls dns_dispatch_done() directly, so it is no longer ever necessary to call both functions. dns_dispatch_cancel() is used to terminate dispatch connections that are still pending, while dns_dispatch_done() is used when they are complete.
2021-10-03 15:15:50 -07:00
done:
Correctly call dns_dispatch_done() in dns_dispatch_cancel() Pass '&resp' rather than 'respp' as we have already cleared '*respp' when we took ownership of 'resp'.
2021-10-06 18:32:19 +11:00
dns_dispatch_done(&resp);
refactor dispatch cancellation Renamed some functions for clarity and readability: - dns_dispatch_addresponse() -> dns_dispatch_add() - dns_dispatch_removeresponse() -> dns_dispatch_done() The dns_dispatch_cancel() function now calls dns_dispatch_done() directly, so it is no longer ever necessary to call both functions. dns_dispatch_cancel() is used to terminate dispatch connections that are still pending, while dns_dispatch_done() is used when they are complete.
2021-10-03 15:15:50 -07:00
}
void
dns_dispatch_done(dns_dispentry_t **respp) {
General code refactoring - style cleanup - removed NULL checks in places where they are not currently needed - use isc_refcount for dispatch reference counting - revised code flow for readability - remove some #ifdefs that are no longer relevant - remove unused struct members - removed unnecessary function parameters - use C99 struct initialization
2020-12-09 19:44:41 -08:00
dns_dispatchmgr_t *mgr = NULL;
dns_dispatch_t *disp = NULL;
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
dns_dispentry_t *resp = NULL;
General code refactoring - style cleanup - removed NULL checks in places where they are not currently needed - use isc_refcount for dispatch reference counting - revised code flow for readability - remove some #ifdefs that are no longer relevant - remove unused struct members - removed unnecessary function parameters - use C99 struct initialization
2020-12-09 19:44:41 -08:00
dns_qid_t *qid = NULL;
checkpoint
1999-06-18 02:01:42 +00:00
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
REQUIRE(respp != NULL);
checkpoint
1999-06-18 02:01:42 +00:00
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
resp = *respp;
REQUIRE(VALID_RESPONSE(resp));
disp = resp->disp;
checkpoint
1999-06-18 02:01:42 +00:00
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
REQUIRE(VALID_DISPATCH(disp));
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
mgr = disp->mgr;
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
Merge the mlg-20000518 branch onto the mainline. Change summary: dns_dispatch_create() no longer exists. dns_dispatch_createtcp() and dns_dispatch_getudp() are the replacements. _createtcp() takes a bound, connected TCP socket, while _getudp() will search for a sharable UDP socket, and if found, attach to it and return a pointer to it. If one is not found, it will create a udp socket, bind it to a supplied local address, and create a new dispatcher around it. dns_dispatch_remove{request,response}() no longer take the dispatch as an argument. query-source can now be set per view. The dispatch manager holds onto three memory pools, one for allocating dispatchers from, one for events, and one for requests/replies. The free list on these pools is hard-coded, but set to 1024. This keeps us from having to dig into the isc_mem_t the pools draw from as often. dns_resolver_create() and dns_view_createresolver() require that valid dispatchers be passed in; dispatchers are no longer created for the caller.
2000-05-19 21:46:46 +00:00
REQUIRE(VALID_DISPATCHMGR(mgr));
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
Dispatch API simplification - Many dispatch attributes can be set implicitly instead of being passed in. we can infer whether to set DNS_DISPATCHATTR_TCP or _UDP from whether we're calling dns_dispatch_createtcp() or _createudp(). we can also infer DNS_DISPATCHATTR_IPV4 or _IPV6 from the addresses or the socket that were passed in. - We no longer use dup'd sockets in UDP dispatches, so the 'dup_socket' parameter has been removed from dns_dispatch_createudp(), along with the code implementing it. also removed isc_socket_dup() since it no longer has any callers. - The 'buffersize' parameter was ignored and has now been removed; buffersize is now fixed at 4096. - Maxbuffers and maxrequests don't need to be passed in on every call to dns_dispatch_createtcp() and _createudp(). In all current uses, the value for mgr->maxbuffers will either be raised once from its default of 20000 to 32768, or else left alone. (passing in a value lower than 20000 does not lower it.) there isn't enough difference between these values for there to be any need to configure this. The value for disp->maxrequests controls both the quota of concurrent requests for a dispatch and also the size of the dispatch socket memory pool. it's not clear that this quota is necessary at all. the memory pool size currently starts at 32768, but is sometimes lowered to 4096, which is definitely unnecessary. This commit sets both values permanently to 32768. - Previously TCP dispatches allocated their own separate QID table, which didn't incorporate a port table. this commit removes per-dispatch QID tables and shares the same table between all dispatches. since dispatches are created for each TCP socket, this may speed up the dispatch allocation process. there may be a slight increase in lock contention since all dispatches are sharing a single QID table, but since TCP sockets are used less often than UDP sockets (which were already sharing a QID table), it should not be a substantial change. - The dispatch port table was being used to determine whether a port was already in use; if so, then a UDP socket would be bound with REUSEADDR. this commit removes the port table, and always binds UDP sockets that way.
2020-12-17 00:43:00 -08:00
qid = mgr->qid;
469. [bug] "query-source address * port 53;" now works.
2000-09-18 04:50:05 +00:00
checkpoint
1999-06-18 02:01:42 +00:00
LOCK(&disp->lock);
snapshot
1999-07-09 00:51:08 +00:00
INSIST(disp->requests > 0);
disp->requests--;
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
dec_stats(disp->mgr, (disp->socktype == isc_socktype_udp)
Use clang-format to reformat the source files
2020-02-12 13:59:18 +01:00
? dns_resstatscounter_disprequdp
: dns_resstatscounter_dispreqtcp);
General code refactoring - style cleanup - removed NULL checks in places where they are not currently needed - use isc_refcount for dispatch reference counting - revised code flow for readability - remove some #ifdefs that are no longer relevant - remove unused struct members - removed unnecessary function parameters - use C99 struct initialization
2020-12-09 19:44:41 -08:00
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
deactivate_dispentry(disp, resp);
checkpoint
1999-06-18 02:01:42 +00:00
469. [bug] "query-source address * port 53;" now works.
2000-09-18 04:50:05 +00:00
LOCK(&qid->lock);
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
ISC_LIST_UNLINK(qid->qid_table[resp->bucket], resp, link);
469. [bug] "query-source address * port 53;" now works.
2000-09-18 04:50:05 +00:00
UNLOCK(&qid->lock);
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
UNLOCK(&disp->lock);
snapshot
1999-07-09 00:51:08 +00:00
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
dispentry_detach(respp);
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
}
/*
* disp must be locked.
*/
static void
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
startrecv(isc_nmhandle_t *handle, dns_dispatch_t *disp, dns_dispentry_t *resp) {
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
switch (disp->socktype) {
case isc_socktype_udp:
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
REQUIRE(resp != NULL && resp->handle == NULL);
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
dispatch: Enforce original timeout when calling _getnext() udp_recv() will call dispatch_getnext() if the message received is invalid or doesn't match; we need to reduce the timeout each time this happens so we can't be starved forever by someone sending garbage packets.
2021-10-01 12:53:31 -07:00
TIME_NOW(&resp->start);
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
isc_nmhandle_attach(handle, &resp->handle);
dispatch: Refactor to eliminate dns_dispatchevent - Responses received by the dispatch are no longer sent to the caller via a task event, but via a netmgr-style recv callback. the 'action' parameter to dns_dispatch_addresponse() is now called 'response' and is called directly from udp_recv() or tcp_recv() when a valid response has been received. - All references to isc_task and isc_taskmgr have been removed from dispatch functions. - All references to dns_dispatchevent_t have been removed and the type has been deleted. - Added a task to the resolver response context, to be used for fctx events. - When the caller cancels an operation, the response handler will be called with ISC_R_CANCELED; it can abort immediately since the caller will presumably have taken care of cleanup already. - Cleaned up attach/detach in resquery and request.
2021-07-26 20:23:18 -07:00
dispentry_attach(resp, &(dns_dispentry_t *){ NULL });
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
isc_nm_read(resp->handle, udp_recv, resp);
break;
dispatch: Refactor to eliminate dns_dispatchevent - Responses received by the dispatch are no longer sent to the caller via a task event, but via a netmgr-style recv callback. the 'action' parameter to dns_dispatch_addresponse() is now called 'response' and is called directly from udp_recv() or tcp_recv() when a valid response has been received. - All references to isc_task and isc_taskmgr have been removed from dispatch functions. - All references to dns_dispatchevent_t have been removed and the type has been deleted. - Added a task to the resolver response context, to be used for fctx events. - When the caller cancels an operation, the response handler will be called with ISC_R_CANCELED; it can abort immediately since the caller will presumably have taken care of cleanup already. - Cleaned up attach/detach in resquery and request.
2021-07-26 20:23:18 -07:00
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
case isc_socktype_tcp:
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
REQUIRE(disp != NULL);
LOCK(&disp->lock);
REQUIRE(disp->handle == NULL);
don't set DNS_DISPATCHSTATE_CONNECTED until attaching handle there was a race possible in which a dispatch was put into the 'connected' state before it had a TCP handle attached, which could cause an assertion failure in dns_dispatch_gettcp().
2021-10-20 11:39:13 -07:00
REQUIRE(atomic_compare_exchange_strong(
&disp->state,
&(uint_fast32_t){ DNS_DISPATCHSTATE_CONNECTING },
DNS_DISPATCHSTATE_CONNECTED));
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
isc_nmhandle_attach(handle, &disp->handle);
dns_dispatch_attach(disp, &(dns_dispatch_t *){ NULL });
isc_nm_read(disp->handle, tcp_recv, disp);
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
UNLOCK(&disp->lock);
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
break;
dispatch: Refactor to eliminate dns_dispatchevent - Responses received by the dispatch are no longer sent to the caller via a task event, but via a netmgr-style recv callback. the 'action' parameter to dns_dispatch_addresponse() is now called 'response' and is called directly from udp_recv() or tcp_recv() when a valid response has been received. - All references to isc_task and isc_taskmgr have been removed from dispatch functions. - All references to dns_dispatchevent_t have been removed and the type has been deleted. - Added a task to the resolver response context, to be used for fctx events. - When the caller cancels an operation, the response handler will be called with ISC_R_CANCELED; it can abort immediately since the caller will presumably have taken care of cleanup already. - Cleaned up attach/detach in resquery and request.
2021-07-26 20:23:18 -07:00
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
default:
Move isc_socket_connect() calls into dispatch dns_dispatch_connect() connects a dispatch socket (for TCP) or a dispatch entry socket (for UDP). This is the next step in moving all uses of the isc_socket code into the dispatch module. This API is temporary; it needs to be cleaned up further so that it can be called the same way for both TCP and UDP.
2020-12-19 01:34:41 -08:00
INSIST(0);
ISC_UNREACHABLE();
}
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
}
static void
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
tcp_connected(isc_nmhandle_t *handle, isc_result_t eresult, void *arg) {
dns_dispatch_t *disp = (dns_dispatch_t *)arg;
dns_dispentry_t *resp = NULL, *next = NULL;
dns_displist_t resps;
dispatch_log(disp, LVL(90), "TCP connected (%p): %s", disp,
isc_result_totext(eresult));
ISC_LIST_INIT(resps);
if (MGR_IS_SHUTTINGDOWN(disp->mgr)) {
eresult = ISC_R_SHUTTINGDOWN;
}
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
if (eresult == ISC_R_SUCCESS) {
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
startrecv(handle, disp, NULL);
}
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
/*
* If there are pending responses, call the connect
* callbacks for all of them.
*/
LOCK(&disp->lock);
for (resp = ISC_LIST_HEAD(disp->pending); resp != NULL; resp = next) {
next = ISC_LIST_NEXT(resp, plink);
ISC_LIST_UNLINK(disp->pending, resp, plink);
ISC_LIST_APPEND(resps, resp, plink);
}
UNLOCK(&disp->lock);
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
for (resp = ISC_LIST_HEAD(resps); resp != NULL; resp = next) {
next = ISC_LIST_NEXT(resp, plink);
ISC_LIST_UNLINK(resps, resp, plink);
if (resp->connected != NULL) {
resp->connected(eresult, NULL, resp->arg);
}
dispentry_detach(&resp);
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
}
Move isc_socket_connect() calls into dispatch dns_dispatch_connect() connects a dispatch socket (for TCP) or a dispatch entry socket (for UDP). This is the next step in moving all uses of the isc_socket code into the dispatch module. This API is temporary; it needs to be cleaned up further so that it can be called the same way for both TCP and UDP.
2020-12-19 01:34:41 -08:00
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
dns_dispatch_detach(&disp);
}
static void
udp_connected(isc_nmhandle_t *handle, isc_result_t eresult, void *arg) {
dns_dispentry_t *resp = (dns_dispentry_t *)arg;
dns_dispatch_t *disp = resp->disp;
dispatch_log(disp, LVL(90), "UDP connected (%p): %s", resp,
isc_result_totext(eresult));
dispatch: Refactor to eliminate dns_dispatchevent - Responses received by the dispatch are no longer sent to the caller via a task event, but via a netmgr-style recv callback. the 'action' parameter to dns_dispatch_addresponse() is now called 'response' and is called directly from udp_recv() or tcp_recv() when a valid response has been received. - All references to isc_task and isc_taskmgr have been removed from dispatch functions. - All references to dns_dispatchevent_t have been removed and the type has been deleted. - Added a task to the resolver response context, to be used for fctx events. - When the caller cancels an operation, the response handler will be called with ISC_R_CANCELED; it can abort immediately since the caller will presumably have taken care of cleanup already. - Cleaned up attach/detach in resquery and request.
2021-07-26 20:23:18 -07:00
if (MGR_IS_SHUTTINGDOWN(disp->mgr)) {
eresult = ISC_R_SHUTTINGDOWN;
}
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
if (eresult == ISC_R_SUCCESS && resp->canceled) {
eresult = ISC_R_CANCELED;
} else if (eresult == ISC_R_SUCCESS) {
startrecv(handle, disp, resp);
} else if (eresult == ISC_R_ADDRINUSE) {
in_port_t localport = 0;
isc_result_t result;
/* probably a port collision; try a different one */
disp->nsockets--;
result = setup_socket(disp, resp, &resp->peer, &localport);
if (result == ISC_R_SUCCESS) {
dns_dispatch_connect(resp);
goto detach;
}
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
}
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
if (resp->connected != NULL) {
resp->connected(eresult, NULL, resp->arg);
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
}
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
detach:
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
dispentry_detach(&resp);
Move isc_socket_connect() calls into dispatch dns_dispatch_connect() connects a dispatch socket (for TCP) or a dispatch entry socket (for UDP). This is the next step in moving all uses of the isc_socket code into the dispatch module. This API is temporary; it needs to be cleaned up further so that it can be called the same way for both TCP and UDP.
2020-12-19 01:34:41 -08:00
}
Move isc_socket_sendto2() calls into dispatch We now use dns_dispatch_send() for this purpose.
2021-01-04 23:03:50 -08:00
isc_result_t
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
dns_dispatch_connect(dns_dispentry_t *resp) {
dns_dispatch_t *disp = NULL;
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
uint_fast32_t state = DNS_DISPATCHSTATE_NONE;
Move isc_socket_sendto2() calls into dispatch We now use dns_dispatch_send() for this purpose.
2021-01-04 23:03:50 -08:00
REQUIRE(VALID_RESPONSE(resp));
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
disp = resp->disp;
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
/* This will be detached once we've connected. */
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
dispentry_attach(resp, &(dns_dispentry_t *){ NULL });
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
switch (disp->socktype) {
case isc_socktype_tcp:
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
/*
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
* Check whether the dispatch is already connecting
* or connected.
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
*/
atomic_compare_exchange_strong(&disp->state,
(uint_fast32_t *)&state,
DNS_DISPATCHSTATE_CONNECTING);
switch (state) {
case DNS_DISPATCHSTATE_NONE:
/* First connection, continue with connecting */
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
LOCK(&disp->lock);
INSIST(ISC_LIST_EMPTY(disp->pending));
ISC_LIST_APPEND(disp->pending, resp, plink);
UNLOCK(&disp->lock);
dns_dispatch_attach(disp, &(dns_dispatch_t *){ NULL });
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
isc_nm_tcpdnsconnect(disp->mgr->nm, &disp->local,
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
&disp->peer, tcp_connected, disp,
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
resp->timeout, 0);
break;
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
case DNS_DISPATCHSTATE_CONNECTING:
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
/* Connection pending; add resp to the list */
LOCK(&disp->lock);
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
ISC_LIST_APPEND(disp->pending, resp, plink);
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
UNLOCK(&disp->lock);
break;
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
case DNS_DISPATCHSTATE_CONNECTED:
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
/* We are already connected; call the connected cb */
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
if (resp->connected != NULL) {
resp->connected(ISC_R_SUCCESS, NULL, resp->arg);
}
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
dispentry_detach(&resp);
break;
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
default:
INSIST(0);
ISC_UNREACHABLE();
}
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
break;
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
case isc_socktype_udp:
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
isc_nm_udpconnect(disp->mgr->nm, &resp->local, &resp->peer,
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
udp_connected, resp, resp->timeout, 0);
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
break;
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
default:
return (ISC_R_NOTIMPLEMENTED);
}
Move isc_socket_sendto2() calls into dispatch We now use dns_dispatch_send() for this purpose.
2021-01-04 23:03:50 -08:00
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
return (ISC_R_SUCCESS);
}
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
static void
send_done(isc_nmhandle_t *handle, isc_result_t result, void *cbarg) {
dns_dispentry_t *resp = (dns_dispentry_t *)cbarg;
REQUIRE(VALID_RESPONSE(resp));
dispatch: Remove 'timeout' callback - It is no longer necessary to pass a 'timeout' callback to dns_dispatch_addresponse(); timeouts are handled directly by the 'response' callback instead. - The netmgr handle is no longer passed to dispatch callbacks, since they don't (and can't) use it. instead, dispatch_cb_t now takes a result code, region, and argument. - Cleaned up timeout-related tests in dispatch_test.c
2021-08-03 15:27:06 +02:00
resp->sent(result, NULL, resp->arg);
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
if (result != ISC_R_SUCCESS) {
isc_nm_cancelread(handle);
}
dispentry_detach(&resp);
}
dispatch: Remove 'timeout' callback - It is no longer necessary to pass a 'timeout' callback to dns_dispatch_addresponse(); timeouts are handled directly by the 'response' callback instead. - The netmgr handle is no longer passed to dispatch callbacks, since they don't (and can't) use it. instead, dispatch_cb_t now takes a result code, region, and argument. - Cleaned up timeout-related tests in dispatch_test.c
2021-08-03 15:27:06 +02:00
void
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
dns_dispatch_resume(dns_dispentry_t *resp, uint16_t timeout) {
dns_dispatch_t *disp = NULL;
dispatch: Clean up connect and recv callbacks - disp_connected() has been split into two functions, udp_connected() (which takes 'resp' as an argument) and tcp_connected() (which takes 'disp', and calls the connect callbacks for all pending resps). - In dns_dispatch_connect(), if a connection is already open, we need to detach the dispentry immediately because we won't be running tcp_connected(). - dns_disptach_cancel() also now calls the connect callbacks for pending TCP responses, and the response callbacks for open TCP connections waiting on read. - If udp_connected() runs after dns_dispatch_cancel() has been called, ensure that the caller's connect callback is run. - If a UDP connection fails with EADDRINUSE, we try again up to five times with a different local port number before giving up. - If a TCP connection is canceled while still pending connection, the connect timeout may still fire. we attach the dispatch before connecting to ensure that it won't be detached too soon in this case. - The dispentry is no longer removed from the pending list when deactivating, so that the connect callback can still be run if dns_dispatch_removeresponse() was run while the connecting was pending. - Rewrote dns_dispatch_gettcp() to avoid a data race. - startrecv() and dispatch_getnext() can be called with a NULL resp when using TCP. - Refactored udp_recv() and tcp_recv() and added result logging. - EOF is now treated the same as CANCELED in response callbacks. - ISC_R_SHUTTINGDOWN is sent to the reponse callbacks for all resps if tcp_recv() is triggered by a netmgr shutdown. (response callbacks are *not* sent by udp_recv() in this case.)
2021-08-04 13:14:11 -07:00
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
REQUIRE(VALID_RESPONSE(resp));
dispatch: Remove 'timeout' callback - It is no longer necessary to pass a 'timeout' callback to dns_dispatch_addresponse(); timeouts are handled directly by the 'response' callback instead. - The netmgr handle is no longer passed to dispatch callbacks, since they don't (and can't) use it. instead, dispatch_cb_t now takes a result code, region, and argument. - Cleaned up timeout-related tests in dispatch_test.c
2021-08-03 15:27:06 +02:00
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
disp = resp->disp;
dispatch: Remove 'timeout' callback - It is no longer necessary to pass a 'timeout' callback to dns_dispatch_addresponse(); timeouts are handled directly by the 'response' callback instead. - The netmgr handle is no longer passed to dispatch callbacks, since they don't (and can't) use it. instead, dispatch_cb_t now takes a result code, region, and argument. - Cleaned up timeout-related tests in dispatch_test.c
2021-08-03 15:27:06 +02:00
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
REQUIRE(VALID_DISPATCH(disp));
dispatch: Remove 'timeout' callback - It is no longer necessary to pass a 'timeout' callback to dns_dispatch_addresponse(); timeouts are handled directly by the 'response' callback instead. - The netmgr handle is no longer passed to dispatch callbacks, since they don't (and can't) use it. instead, dispatch_cb_t now takes a result code, region, and argument. - Cleaned up timeout-related tests in dispatch_test.c
2021-08-03 15:27:06 +02:00
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
dispatch_getnext(disp, resp, timeout);
dispatch: Remove 'timeout' callback - It is no longer necessary to pass a 'timeout' callback to dns_dispatch_addresponse(); timeouts are handled directly by the 'response' callback instead. - The netmgr handle is no longer passed to dispatch callbacks, since they don't (and can't) use it. instead, dispatch_cb_t now takes a result code, region, and argument. - Cleaned up timeout-related tests in dispatch_test.c
2021-08-03 15:27:06 +02:00
}
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
void
dns_dispatch_send(dns_dispentry_t *resp, isc_region_t *r, isc_dscp_t dscp) {
isc_nmhandle_t *handle = NULL;
REQUIRE(VALID_RESPONSE(resp));
UNUSED(dscp);
#if 0
/* XXX: no DSCP support */
Move isc_socket_sendto2() calls into dispatch We now use dns_dispatch_send() for this purpose.
2021-01-04 23:03:50 -08:00
if (dscp == -1) {
sendevent->attributes &= ~ISC_SOCKEVENTATTR_DSCP;
sendevent->dscp = 0;
} else {
sendevent->attributes |= ISC_SOCKEVENTATTR_DSCP;
sendevent->dscp = dscp;
if (tcp) {
isc_socket_dscp(sock, dscp);
}
}
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
#endif
Move isc_socket_sendto2() calls into dispatch We now use dns_dispatch_send() for this purpose.
2021-01-04 23:03:50 -08:00
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
if (resp->disp->socktype == isc_socktype_tcp) {
handle = resp->disp->handle;
} else {
handle = resp->handle;
}
dispatch: Refactor to eliminate dns_dispatchevent - Responses received by the dispatch are no longer sent to the caller via a task event, but via a netmgr-style recv callback. the 'action' parameter to dns_dispatch_addresponse() is now called 'response' and is called directly from udp_recv() or tcp_recv() when a valid response has been received. - All references to isc_task and isc_taskmgr have been removed from dispatch functions. - All references to dns_dispatchevent_t have been removed and the type has been deleted. - Added a task to the resolver response context, to be used for fctx events. - When the caller cancels an operation, the response handler will be called with ISC_R_CANCELED; it can abort immediately since the caller will presumably have taken care of cleanup already. - Cleaned up attach/detach in resquery and request.
2021-07-26 20:23:18 -07:00
dispentry_attach(resp, &(dns_dispentry_t *){ NULL });
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
isc_nm_send(handle, r, send_done, resp);
Move isc_socket_sendto2() calls into dispatch We now use dns_dispatch_send() for this purpose.
2021-01-04 23:03:50 -08:00
}
add dns_dispatch_getlocaladdress().
2001-03-13 05:48:41 +00:00
isc_result_t
apply the modified style
2020-02-13 14:44:37 -08:00
dns_dispatch_getlocaladdress(dns_dispatch_t *disp, isc_sockaddr_t *addrp) {
add dns_dispatch_getlocaladdress().
2001-03-13 05:48:41 +00:00
REQUIRE(VALID_DISPATCH(disp));
REQUIRE(addrp != NULL);
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
if (disp->socktype == isc_socktype_udp) {
add dns_dispatch_getlocaladdress().
2001-03-13 05:48:41 +00:00
*addrp = disp->local;
return (ISC_R_SUCCESS);
}
return (ISC_R_NOTIMPLEMENTED);
}
Move isc_socket_getsockname() calls into dispatch We now use dns_dispentry_getlocaladdress(). (this API is likely to be cleaned up further later.)
2021-01-04 14:38:35 -08:00
isc_result_t
dns_dispentry_getlocaladdress(dns_dispentry_t *resp, isc_sockaddr_t *addrp) {
REQUIRE(VALID_RESPONSE(resp));
REQUIRE(addrp != NULL);
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
if (resp->disp->socktype == isc_socktype_tcp) {
*addrp = resp->disp->local;
return (ISC_R_SUCCESS);
Move isc_socket_getsockname() calls into dispatch We now use dns_dispentry_getlocaladdress(). (this API is likely to be cleaned up further later.)
2021-01-04 14:38:35 -08:00
}
Refactor dispatch, resolver and request Since every dispsock was associated with a dispentry anyway (though not always vice versa), the members of dispsock have been combined into dispentry, which is now reference-counted. dispentry objects are now attached before connecting and detached afterward to prevent races between the connect callback and dns_dispatch_removeresponse(). Dispatch and dispatchmgr objects are now reference counted as well, and the shutdown process has been simplified. reference counting of resquery and request objects has also been cleaned up significantly. dns_dispatch_cancel() now flags a dispentry as having been canceled, so that if the connect callback runs after cancellation, it will not initiate a read. The isblackholed() function has been simplified.
2021-05-25 22:54:17 -07:00
if (resp->handle != NULL) {
*addrp = isc_nmhandle_localaddr(resp->handle);
Convert dispatch to netmgr The flow of operations in dispatch is changing and will now be similar for both UDP and TCP queries: 1) Call dns_dispatch_addresponse() to assign a query ID and register that we'll be listening for a response with that ID soon. the parameters for this function include callback functions to inform the caller when the socket is connected and when the message has been sent, as well as a task action that will be sent when the response arrives. (later this could become a netmgr callback, but at this stage to minimize disruption to the calling code, we continue to use isc_task for the response event.) on successful completion of this function, a dispatch entry object will be instantiated. 2) Call dns_dispatch_connect() on the dispatch entry. this runs isc_nm_udpconnect() or isc_nm_tcpdnsconnect(), as needed, and begins listening for responses. the caller is informed via a callback function when the connection is established. 3) Call dns_dispatch_send() on the dispatch entry. this runs isc_nm_send() to send a request. 4) Call dns_dispatch_removeresponse() to terminate listening and close the connection. Implementation comments below: - As we will be using netmgr buffers now. code to send the length in TCP queries has also been removed as that is handled by the netmgr. - TCP dispatches can be used by multiple simultaneous queries, so dns_dispatch_connect() now checks whether the dispatch is already connected before calling isc_nm_tcpdnsconnect() again. - Running dns_dispatch_getnext() from a non-network thread caused a crash due to assertions in the netmgr read functions that appear to be unnecessary now. the assertions have been removed. - fctx->nqueries was formerly incremented when the connection was successful, but is now incremented when the query is started and decremented if the connection fails. - It's no longer necessary for each dispatch to have a pool of tasks, so there's now a single task per dispatch. - Dispatch code to avoid UDP ports already in use has been removed. - dns_resolver and dns_request have been modified to use netmgr callback functions instead of task events. some additional changes were needed to handle shutdown processing correctly. - Timeout processing is not yet fully converted to use netmgr timeouts. - Fixed a lock order cycle reported by TSAN (view -> zone-> adb -> view) by by calling dns_zt functions without holding the view lock.
2021-01-14 13:02:57 -08:00
return (ISC_R_SUCCESS);
Move isc_socket_getsockname() calls into dispatch We now use dns_dispentry_getlocaladdress(). (this API is likely to be cleaned up further later.)
2021-01-04 14:38:35 -08:00
}
return (ISC_R_NOTIMPLEMENTED);
}
create and use multiple fetch dispatches Added API to create a set of UDP dispatches which can be shared round-robin style when making upstream queries for authoritative data; this should reduce lock contention in the query source dispatch.
2012-04-27 16:07:24 -07:00
dns_dispatch_t *
apply the modified style
2020-02-13 14:44:37 -08:00
dns_dispatchset_get(dns_dispatchset_t *dset) {
General code refactoring - style cleanup - removed NULL checks in places where they are not currently needed - use isc_refcount for dispatch reference counting - revised code flow for readability - remove some #ifdefs that are no longer relevant - remove unused struct members - removed unnecessary function parameters - use C99 struct initialization
2020-12-09 19:44:41 -08:00
dns_dispatch_t *disp = NULL;
create and use multiple fetch dispatches Added API to create a set of UDP dispatches which can be shared round-robin style when making upstream queries for authoritative data; this should reduce lock contention in the query source dispatch.
2012-04-27 16:07:24 -07:00
/* check that dispatch set is configured */
Use clang-tidy to add curly braces around one-line statements The command used to reformat the files in this commit was: ./util/run-clang-tidy \ -clang-tidy-binary clang-tidy-11 -clang-apply-replacements-binary clang-apply-replacements-11 \ -checks=-*,readability-braces-around-statements \ -j 9 \ -fix \ -format \ -style=file \ -quiet clang-format -i --style=format $(git ls-files '*.c' '*.h') uncrustify -c .uncrustify.cfg --replace --no-backup $(git ls-files '*.c' '*.h') clang-format -i --style=format $(git ls-files '*.c' '*.h')
2020-02-13 21:48:23 +01:00
if (dset == NULL || dset->ndisp == 0) {
create and use multiple fetch dispatches Added API to create a set of UDP dispatches which can be shared round-robin style when making upstream queries for authoritative data; this should reduce lock contention in the query source dispatch.
2012-04-27 16:07:24 -07:00
return (NULL);
Use clang-tidy to add curly braces around one-line statements The command used to reformat the files in this commit was: ./util/run-clang-tidy \ -clang-tidy-binary clang-tidy-11 -clang-apply-replacements-binary clang-apply-replacements-11 \ -checks=-*,readability-braces-around-statements \ -j 9 \ -fix \ -format \ -style=file \ -quiet clang-format -i --style=format $(git ls-files '*.c' '*.h') uncrustify -c .uncrustify.cfg --replace --no-backup $(git ls-files '*.c' '*.h') clang-format -i --style=format $(git ls-files '*.c' '*.h')
2020-02-13 21:48:23 +01:00
}
create and use multiple fetch dispatches Added API to create a set of UDP dispatches which can be shared round-robin style when making upstream queries for authoritative data; this should reduce lock contention in the query source dispatch.
2012-04-27 16:07:24 -07:00
LOCK(&dset->lock);
disp = dset->dispatches[dset->cur];
dset->cur++;
Use clang-tidy to add curly braces around one-line statements The command used to reformat the files in this commit was: ./util/run-clang-tidy \ -clang-tidy-binary clang-tidy-11 -clang-apply-replacements-binary clang-apply-replacements-11 \ -checks=-*,readability-braces-around-statements \ -j 9 \ -fix \ -format \ -style=file \ -quiet clang-format -i --style=format $(git ls-files '*.c' '*.h') uncrustify -c .uncrustify.cfg --replace --no-backup $(git ls-files '*.c' '*.h') clang-format -i --style=format $(git ls-files '*.c' '*.h')
2020-02-13 21:48:23 +01:00
if (dset->cur == dset->ndisp) {
create and use multiple fetch dispatches Added API to create a set of UDP dispatches which can be shared round-robin style when making upstream queries for authoritative data; this should reduce lock contention in the query source dispatch.
2012-04-27 16:07:24 -07:00
dset->cur = 0;
Use clang-tidy to add curly braces around one-line statements The command used to reformat the files in this commit was: ./util/run-clang-tidy \ -clang-tidy-binary clang-tidy-11 -clang-apply-replacements-binary clang-apply-replacements-11 \ -checks=-*,readability-braces-around-statements \ -j 9 \ -fix \ -format \ -style=file \ -quiet clang-format -i --style=format $(git ls-files '*.c' '*.h') uncrustify -c .uncrustify.cfg --replace --no-backup $(git ls-files '*.c' '*.h') clang-format -i --style=format $(git ls-files '*.c' '*.h')
2020-02-13 21:48:23 +01:00
}
create and use multiple fetch dispatches Added API to create a set of UDP dispatches which can be shared round-robin style when making upstream queries for authoritative data; this should reduce lock contention in the query source dispatch.
2012-04-27 16:07:24 -07:00
UNLOCK(&dset->lock);
return (disp);
}
isc_result_t
dispatch: Refactor to eliminate dns_dispatchevent - Responses received by the dispatch are no longer sent to the caller via a task event, but via a netmgr-style recv callback. the 'action' parameter to dns_dispatch_addresponse() is now called 'response' and is called directly from udp_recv() or tcp_recv() when a valid response has been received. - All references to isc_task and isc_taskmgr have been removed from dispatch functions. - All references to dns_dispatchevent_t have been removed and the type has been deleted. - Added a task to the resolver response context, to be used for fctx events. - When the caller cancels an operation, the response handler will be called with ISC_R_CANCELED; it can abort immediately since the caller will presumably have taken care of cleanup already. - Cleaned up attach/detach in resquery and request.
2021-07-26 20:23:18 -07:00
dns_dispatchset_create(isc_mem_t *mctx, dns_dispatch_t *source,
dns_dispatchset_t **dsetp, int n) {
apply the modified style
2020-02-13 14:44:37 -08:00
isc_result_t result;
General code refactoring - style cleanup - removed NULL checks in places where they are not currently needed - use isc_refcount for dispatch reference counting - revised code flow for readability - remove some #ifdefs that are no longer relevant - remove unused struct members - removed unnecessary function parameters - use C99 struct initialization
2020-12-09 19:44:41 -08:00
dns_dispatchset_t *dset = NULL;
dns_dispatchmgr_t *mgr = NULL;
apply the modified style
2020-02-13 14:44:37 -08:00
int i, j;
create and use multiple fetch dispatches Added API to create a set of UDP dispatches which can be shared round-robin style when making upstream queries for authoritative data; this should reduce lock contention in the query source dispatch.
2012-04-27 16:07:24 -07:00
REQUIRE(VALID_DISPATCH(source));
dispatch: Fix several connect-related issues - startrecv() and getnext() have been rewritten. - Don't set TCP flag when connecting a UDP dispatch. - Prevent TCP connections from trying to connect twice. - dns_dispatch_gettcp() can now find a matching TCP dispatch that has not yet fully connected, and attach to it. when the connection is completed, the connect callbacks are run for all of the pending entries. - An atomic 'state' variable is now used for connection state instead of attributes. - When dns_dispatch_cancel() is called on a TCP dispatch entry, only that one entry is canceled. the dispatch itself should not be shut down until there are no dispatch entries left associated with it. - Other incidental cleanup, including removing DNS_DISPATCHATTR_IPV4 and _IPV6 (they were being set in the dispatch attributes but never used), cleaning up dns_requestmgr_create(), and renaming dns_dispatch_read() to the more descriptive dns_dispatch_resume().
2021-08-03 18:24:27 -07:00
REQUIRE(source->socktype == isc_socktype_udp);
create and use multiple fetch dispatches Added API to create a set of UDP dispatches which can be shared round-robin style when making upstream queries for authoritative data; this should reduce lock contention in the query source dispatch.
2012-04-27 16:07:24 -07:00
REQUIRE(dsetp != NULL && *dsetp == NULL);
mgr = source->mgr;
dset = isc_mem_get(mctx, sizeof(dns_dispatchset_t));
General code refactoring - style cleanup - removed NULL checks in places where they are not currently needed - use isc_refcount for dispatch reference counting - revised code flow for readability - remove some #ifdefs that are no longer relevant - remove unused struct members - removed unnecessary function parameters - use C99 struct initialization
2020-12-09 19:44:41 -08:00
*dset = (dns_dispatchset_t){ .ndisp = n };
create and use multiple fetch dispatches Added API to create a set of UDP dispatches which can be shared round-robin style when making upstream queries for authoritative data; this should reduce lock contention in the query source dispatch.
2012-04-27 16:07:24 -07:00
isc_mutex_init returns 'void'
2018-11-16 15:33:22 +01:00
isc_mutex_init(&dset->lock);
create and use multiple fetch dispatches Added API to create a set of UDP dispatches which can be shared round-robin style when making upstream queries for authoritative data; this should reduce lock contention in the query source dispatch.
2012-04-27 16:07:24 -07:00
dset->dispatches = isc_mem_get(mctx, sizeof(dns_dispatch_t *) * n);
isc_mem_attach(mctx, &dset->mctx);
dset->dispatches[0] = NULL;
dns_dispatch_attach(source, &dset->dispatches[0]);
LOCK(&mgr->lock);
for (i = 1; i < n; i++) {
dset->dispatches[i] = NULL;
dispatch: Refactor to eliminate dns_dispatchevent - Responses received by the dispatch are no longer sent to the caller via a task event, but via a netmgr-style recv callback. the 'action' parameter to dns_dispatch_addresponse() is now called 'response' and is called directly from udp_recv() or tcp_recv() when a valid response has been received. - All references to isc_task and isc_taskmgr have been removed from dispatch functions. - All references to dns_dispatchevent_t have been removed and the type has been deleted. - Added a task to the resolver response context, to be used for fctx events. - When the caller cancels an operation, the response handler will be called with ISC_R_CANCELED; it can abort immediately since the caller will presumably have taken care of cleanup already. - Cleaned up attach/detach in resquery and request.
2021-07-26 20:23:18 -07:00
result = dispatch_createudp(mgr, &source->local,
Remove support for shared UDP dispatch sockets Currently the netmgr doesn't support unconnected, shared UDP sockets, so there's no reason to retain that functionality in the dispatcher prior to porting to the netmgr. In this commit, the DNS_DISPATCHATTR_EXCLUSIVE attribute has been removed as it is now non-optional; UDP dispatches are alwasy exclusive. Code implementing non-exclusive UDP dispatches has been removed. dns_dispatch_getentrysocket() now always returns the dispsocket for UDP dispatches and the dispatch socket for TCP dispatches. There is no longer any need to search for existing dispatches from dns_dispatch_getudp(), so the 'mask' option has been removed, and the function renamed to the more descriptive dns_dispatch_createudp().
2020-12-16 01:32:06 -08:00
&dset->dispatches[i]);
Use clang-tidy to add curly braces around one-line statements The command used to reformat the files in this commit was: ./util/run-clang-tidy \ -clang-tidy-binary clang-tidy-11 -clang-apply-replacements-binary clang-apply-replacements-11 \ -checks=-*,readability-braces-around-statements \ -j 9 \ -fix \ -format \ -style=file \ -quiet clang-format -i --style=format $(git ls-files '*.c' '*.h') uncrustify -c .uncrustify.cfg --replace --no-backup $(git ls-files '*.c' '*.h') clang-format -i --style=format $(git ls-files '*.c' '*.h')
2020-02-13 21:48:23 +01:00
if (result != ISC_R_SUCCESS) {
create and use multiple fetch dispatches Added API to create a set of UDP dispatches which can be shared round-robin style when making upstream queries for authoritative data; this should reduce lock contention in the query source dispatch.
2012-04-27 16:07:24 -07:00
goto fail;
Use clang-tidy to add curly braces around one-line statements The command used to reformat the files in this commit was: ./util/run-clang-tidy \ -clang-tidy-binary clang-tidy-11 -clang-apply-replacements-binary clang-apply-replacements-11 \ -checks=-*,readability-braces-around-statements \ -j 9 \ -fix \ -format \ -style=file \ -quiet clang-format -i --style=format $(git ls-files '*.c' '*.h') uncrustify -c .uncrustify.cfg --replace --no-backup $(git ls-files '*.c' '*.h') clang-format -i --style=format $(git ls-files '*.c' '*.h')
2020-02-13 21:48:23 +01:00
}
create and use multiple fetch dispatches Added API to create a set of UDP dispatches which can be shared round-robin style when making upstream queries for authoritative data; this should reduce lock contention in the query source dispatch.
2012-04-27 16:07:24 -07:00
}
UNLOCK(&mgr->lock);
*dsetp = dset;
return (ISC_R_SUCCESS);
Use clang-format to reformat the source files
2020-02-12 13:59:18 +01:00
fail:
create and use multiple fetch dispatches Added API to create a set of UDP dispatches which can be shared round-robin style when making upstream queries for authoritative data; this should reduce lock contention in the query source dispatch.
2012-04-27 16:07:24 -07:00
UNLOCK(&mgr->lock);
update copyright notice
2012-04-28 23:45:42 +00:00
Use clang-tidy to add curly braces around one-line statements The command used to reformat the files in this commit was: ./util/run-clang-tidy \ -clang-tidy-binary clang-tidy-11 -clang-apply-replacements-binary clang-apply-replacements-11 \ -checks=-*,readability-braces-around-statements \ -j 9 \ -fix \ -format \ -style=file \ -quiet clang-format -i --style=format $(git ls-files '*.c' '*.h') uncrustify -c .uncrustify.cfg --replace --no-backup $(git ls-files '*.c' '*.h') clang-format -i --style=format $(git ls-files '*.c' '*.h')
2020-02-13 21:48:23 +01:00
for (j = 0; j < i; j++) {
create and use multiple fetch dispatches Added API to create a set of UDP dispatches which can be shared round-robin style when making upstream queries for authoritative data; this should reduce lock contention in the query source dispatch.
2012-04-27 16:07:24 -07:00
dns_dispatch_detach(&(dset->dispatches[j]));
Use clang-tidy to add curly braces around one-line statements The command used to reformat the files in this commit was: ./util/run-clang-tidy \ -clang-tidy-binary clang-tidy-11 -clang-apply-replacements-binary clang-apply-replacements-11 \ -checks=-*,readability-braces-around-statements \ -j 9 \ -fix \ -format \ -style=file \ -quiet clang-format -i --style=format $(git ls-files '*.c' '*.h') uncrustify -c .uncrustify.cfg --replace --no-backup $(git ls-files '*.c' '*.h') clang-format -i --style=format $(git ls-files '*.c' '*.h')
2020-02-13 21:48:23 +01:00
}
create and use multiple fetch dispatches Added API to create a set of UDP dispatches which can be shared round-robin style when making upstream queries for authoritative data; this should reduce lock contention in the query source dispatch.
2012-04-27 16:07:24 -07:00
isc_mem_put(mctx, dset->dispatches, sizeof(dns_dispatch_t *) * n);
Use clang-tidy to add curly braces around one-line statements The command used to reformat the files in this commit was: ./util/run-clang-tidy \ -clang-tidy-binary clang-tidy-11 -clang-apply-replacements-binary clang-apply-replacements-11 \ -checks=-*,readability-braces-around-statements \ -j 9 \ -fix \ -format \ -style=file \ -quiet clang-format -i --style=format $(git ls-files '*.c' '*.h') uncrustify -c .uncrustify.cfg --replace --no-backup $(git ls-files '*.c' '*.h') clang-format -i --style=format $(git ls-files '*.c' '*.h')
2020-02-13 21:48:23 +01:00
if (dset->mctx == mctx) {
create and use multiple fetch dispatches Added API to create a set of UDP dispatches which can be shared round-robin style when making upstream queries for authoritative data; this should reduce lock contention in the query source dispatch.
2012-04-27 16:07:24 -07:00
isc_mem_detach(&dset->mctx);
Use clang-tidy to add curly braces around one-line statements The command used to reformat the files in this commit was: ./util/run-clang-tidy \ -clang-tidy-binary clang-tidy-11 -clang-apply-replacements-binary clang-apply-replacements-11 \ -checks=-*,readability-braces-around-statements \ -j 9 \ -fix \ -format \ -style=file \ -quiet clang-format -i --style=format $(git ls-files '*.c' '*.h') uncrustify -c .uncrustify.cfg --replace --no-backup $(git ls-files '*.c' '*.h') clang-format -i --style=format $(git ls-files '*.c' '*.h')
2020-02-13 21:48:23 +01:00
}
create and use multiple fetch dispatches Added API to create a set of UDP dispatches which can be shared round-robin style when making upstream queries for authoritative data; this should reduce lock contention in the query source dispatch.
2012-04-27 16:07:24 -07:00
- Make isc_mutex_destroy return void - Make isc_mutexblock_init/destroy return void - Minor cleanups
2018-11-19 10:31:09 +00:00
isc_mutex_destroy(&dset->lock);
create and use multiple fetch dispatches Added API to create a set of UDP dispatches which can be shared round-robin style when making upstream queries for authoritative data; this should reduce lock contention in the query source dispatch.
2012-04-27 16:07:24 -07:00
isc_mem_put(mctx, dset, sizeof(dns_dispatchset_t));
return (result);
}
void
apply the modified style
2020-02-13 14:44:37 -08:00
dns_dispatchset_destroy(dns_dispatchset_t **dsetp) {
General code refactoring - style cleanup - removed NULL checks in places where they are not currently needed - use isc_refcount for dispatch reference counting - revised code flow for readability - remove some #ifdefs that are no longer relevant - remove unused struct members - removed unnecessary function parameters - use C99 struct initialization
2020-12-09 19:44:41 -08:00
dns_dispatchset_t *dset = NULL;
apply the modified style
2020-02-13 14:44:37 -08:00
int i;
create and use multiple fetch dispatches Added API to create a set of UDP dispatches which can be shared round-robin style when making upstream queries for authoritative data; this should reduce lock contention in the query source dispatch.
2012-04-27 16:07:24 -07:00
REQUIRE(dsetp != NULL && *dsetp != NULL);
dset = *dsetp;
Clear the pointer to destroyed object early using the semantic patch Also disable the semantic patch as the code needs tweaks here and there because some destroy functions might not destroy the object and return early if the object is still in use.
2020-02-08 04:37:54 -08:00
*dsetp = NULL;
Use clang-tidy to add curly braces around one-line statements The command used to reformat the files in this commit was: ./util/run-clang-tidy \ -clang-tidy-binary clang-tidy-11 -clang-apply-replacements-binary clang-apply-replacements-11 \ -checks=-*,readability-braces-around-statements \ -j 9 \ -fix \ -format \ -style=file \ -quiet clang-format -i --style=format $(git ls-files '*.c' '*.h') uncrustify -c .uncrustify.cfg --replace --no-backup $(git ls-files '*.c' '*.h') clang-format -i --style=format $(git ls-files '*.c' '*.h')
2020-02-13 21:48:23 +01:00
for (i = 0; i < dset->ndisp; i++) {
create and use multiple fetch dispatches Added API to create a set of UDP dispatches which can be shared round-robin style when making upstream queries for authoritative data; this should reduce lock contention in the query source dispatch.
2012-04-27 16:07:24 -07:00
dns_dispatch_detach(&(dset->dispatches[i]));
Use clang-tidy to add curly braces around one-line statements The command used to reformat the files in this commit was: ./util/run-clang-tidy \ -clang-tidy-binary clang-tidy-11 -clang-apply-replacements-binary clang-apply-replacements-11 \ -checks=-*,readability-braces-around-statements \ -j 9 \ -fix \ -format \ -style=file \ -quiet clang-format -i --style=format $(git ls-files '*.c' '*.h') uncrustify -c .uncrustify.cfg --replace --no-backup $(git ls-files '*.c' '*.h') clang-format -i --style=format $(git ls-files '*.c' '*.h')
2020-02-13 21:48:23 +01:00
}
create and use multiple fetch dispatches Added API to create a set of UDP dispatches which can be shared round-robin style when making upstream queries for authoritative data; this should reduce lock contention in the query source dispatch.
2012-04-27 16:07:24 -07:00
isc_mem_put(dset->mctx, dset->dispatches,
sizeof(dns_dispatch_t *) * dset->ndisp);
- Make isc_mutex_destroy return void - Make isc_mutexblock_init/destroy return void - Minor cleanups
2018-11-19 10:31:09 +00:00
isc_mutex_destroy(&dset->lock);
create and use multiple fetch dispatches Added API to create a set of UDP dispatches which can be shared round-robin style when making upstream queries for authoritative data; this should reduce lock contention in the query source dispatch.
2012-04-27 16:07:24 -07:00
isc_mem_putanddetach(&dset->mctx, dset, sizeof(dns_dispatchset_t));
}
Reference in New Issue Copy Permalink