2
0
mirror of https://gitlab.isc.org/isc-projects/bind9 synced 2025-08-22 18:19:42 +00:00
bind/lib/isc/netmgr/netmgr-int.h

1545 lines
45 KiB
C
Raw Normal View History

/*
* Copyright (C) Internet Systems Consortium, Inc. ("ISC")
*
* 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
* file, you can obtain one at https://mozilla.org/MPL/2.0/.
*
* See the COPYRIGHT file distributed with this work for additional
* information regarding copyright ownership.
*/
#pragma once
#include <unistd.h>
#include <uv.h>
#include <openssl/err.h>
#include <openssl/ssl.h>
#include <isc/astack.h>
#include <isc/atomic.h>
#include <isc/buffer.h>
#include <isc/condition.h>
#include <isc/magic.h>
#include <isc/mem.h>
#include <isc/netmgr.h>
#include <isc/queue.h>
#include <isc/quota.h>
#include <isc/random.h>
#include <isc/refcount.h>
#include <isc/region.h>
#include <isc/result.h>
#include <isc/sockaddr.h>
#include <isc/stats.h>
#include <isc/thread.h>
#include <isc/util.h>
#include "uv-compat.h"
#define ISC_NETMGR_TID_UNKNOWN -1
#if !defined(WIN32)
/*
* New versions of libuv support recvmmsg on unices.
* Since recvbuf is only allocated per worker allocating a bigger one is not
* that wasteful.
* 20 here is UV__MMSG_MAXWIDTH taken from the current libuv source, nothing
* will break if the original value changes.
*/
#define ISC_NETMGR_RECVBUF_SIZE (20 * 65536)
#else
#define ISC_NETMGR_RECVBUF_SIZE (65536)
#endif
#define ISC_NETMGR_SENDBUF_SIZE (sizeof(uint16_t) + UINT16_MAX)
/*%
* Regular TCP buffer size.
*/
#define NM_REG_BUF 4096
/*%
* Larger buffer for when the regular one isn't enough; this will
* hold two full DNS packets with lengths. netmgr receives 64k at
* most in TCPDNS or TLSDNS connections, so there's no risk of overrun
* when using a buffer this size.
*/
#define NM_BIG_BUF (65535 + 2) * 2
#if defined(SO_REUSEPORT_LB) || (defined(SO_REUSEPORT) && defined(__linux__))
#define HAVE_SO_REUSEPORT_LB 1
#endif
/*
* Define NETMGR_TRACE to activate tracing of handles and sockets.
* This will impair performance but enables us to quickly determine,
* if netmgr resources haven't been cleaned up on shutdown, which ones
* are still in use.
*/
#ifdef NETMGR_TRACE
#define TRACE_SIZE 8
void
isc__nm_dump_active(isc_nm_t *nm);
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
#if defined(__linux__)
#include <syscall.h>
#define gettid() (uint32_t) syscall(SYS_gettid)
#elif defined(_WIN32)
#define gettid() (uint32_t) GetCurrentThreadId()
#else
#define gettid() (uint32_t) pthread_self()
#endif
#ifdef NETMGR_TRACE_VERBOSE
#define NETMGR_TRACE_LOG(format, ...) \
fprintf(stderr, "%" PRIu32 ":%d:%s:%u:%s:" format, gettid(), \
isc_nm_tid(), file, line, func, __VA_ARGS__)
#else
#define NETMGR_TRACE_LOG(format, ...) \
(void)file; \
(void)line; \
(void)func;
#endif
#define FLARG_PASS , file, line, func
#define FLARG \
, const char *file __attribute__((unused)), \
unsigned int line __attribute__((unused)), \
const char *func __attribute__((unused))
#define FLARG_IEVENT(ievent) \
const char *file = ievent->file; \
unsigned int line = ievent->line; \
const char *func = ievent->func;
#define FLARG_IEVENT_PASS(ievent) \
ievent->file = file; \
ievent->line = line; \
ievent->func = func;
#define isc__nm_uvreq_get(req, sock) \
isc___nm_uvreq_get(req, sock, __FILE__, __LINE__, __func__)
#define isc__nm_uvreq_put(req, sock) \
isc___nm_uvreq_put(req, sock, __FILE__, __LINE__, __func__)
#define isc__nmsocket_init(sock, mgr, type, iface) \
isc___nmsocket_init(sock, mgr, type, iface, __FILE__, __LINE__, \
__func__)
#define isc__nmsocket_put(sockp) \
isc___nmsocket_put(sockp, __FILE__, __LINE__, __func__)
#define isc__nmsocket_attach(sock, target) \
isc___nmsocket_attach(sock, target, __FILE__, __LINE__, __func__)
#define isc__nmsocket_detach(socketp) \
isc___nmsocket_detach(socketp, __FILE__, __LINE__, __func__)
#define isc__nmsocket_close(socketp) \
isc___nmsocket_close(socketp, __FILE__, __LINE__, __func__)
#define isc__nmhandle_get(sock, peer, local) \
isc___nmhandle_get(sock, peer, local, __FILE__, __LINE__, __func__)
#define isc__nmsocket_prep_destroy(sock) \
isc___nmsocket_prep_destroy(sock, __FILE__, __LINE__, __func__)
#else
#define NETMGR_TRACE_LOG(format, ...)
#define FLARG_PASS
#define FLARG
#define FLARG_IEVENT(ievent)
#define FLARG_IEVENT_PASS(ievent)
#define isc__nm_uvreq_get(req, sock) isc___nm_uvreq_get(req, sock)
#define isc__nm_uvreq_put(req, sock) isc___nm_uvreq_put(req, sock)
#define isc__nmsocket_init(sock, mgr, type, iface) \
isc___nmsocket_init(sock, mgr, type, iface)
#define isc__nmsocket_put(sockp) isc___nmsocket_put(sockp)
#define isc__nmsocket_attach(sock, target) isc___nmsocket_attach(sock, target)
#define isc__nmsocket_detach(socketp) isc___nmsocket_detach(socketp)
#define isc__nmsocket_close(socketp) isc___nmsocket_close(socketp)
#define isc__nmhandle_get(sock, peer, local) \
isc___nmhandle_get(sock, peer, local)
#define isc__nmsocket_prep_destroy(sock) isc___nmsocket_prep_destroy(sock)
#endif
/*
* Single network event loop worker.
*/
typedef struct isc__networker {
2020-02-14 08:14:03 +01:00
isc_nm_t *mgr;
int id; /* thread id */
uv_loop_t loop; /* libuv loop structure */
uv_async_t async; /* async channel to send
* data to this networker */
2020-02-14 08:14:03 +01:00
isc_mutex_t lock;
isc_condition_t cond;
2020-02-14 08:14:03 +01:00
bool paused;
bool finished;
isc_thread_t thread;
isc_queue_t *ievents; /* incoming async events */
isc_queue_t *ievents_prio; /* priority async events
* used for listening etc.
* can be processed while
* worker is paused */
isc_refcount_t references;
atomic_int_fast64_t pktcount;
char *recvbuf;
char *sendbuf;
2020-02-14 08:14:03 +01:00
bool recvbuf_inuse;
} isc__networker_t;
/*
* A general handle for a connection bound to a networker. For UDP
* connections we have peer address here, so both TCP and UDP can be
* handled with a simple send-like function
*/
#define NMHANDLE_MAGIC ISC_MAGIC('N', 'M', 'H', 'D')
#define VALID_NMHANDLE(t) \
(ISC_MAGIC_VALID(t, NMHANDLE_MAGIC) && \
atomic_load(&(t)->references) > 0)
typedef void (*isc__nm_closecb)(isc_nmhandle_t *);
struct isc_nmhandle {
2020-02-14 08:14:03 +01:00
int magic;
isc_refcount_t references;
/*
* The socket is not 'attached' in the traditional
* reference-counting sense. Instead, we keep all handles in an
* array in the socket object. This way, we don't have circular
* dependencies and we can close all handles when we're destroying
* the socket.
*/
isc_nmsocket_t *sock;
size_t ah_pos; /* Position in the socket's 'active handles' array */
2020-02-14 08:14:03 +01:00
isc_sockaddr_t peer;
isc_sockaddr_t local;
isc_nm_opaquecb_t doreset; /* reset extra callback, external */
isc_nm_opaquecb_t dofree; /* free extra callback, external */
#ifdef NETMGR_TRACE
void *backtrace[TRACE_SIZE];
int backtrace_size;
LINK(isc_nmhandle_t) active_link;
#endif
2020-02-14 08:14:03 +01:00
void *opaque;
char extra[];
};
/*
* An interface - an address we can listen on.
*/
struct isc_nmiface {
isc_sockaddr_t addr;
};
typedef enum isc__netievent_type {
netievent_udpconnect,
netievent_udpsend,
netievent_udpread,
netievent_udpstop,
netievent_udpcancel,
netievent_udpclose,
netievent_tcpconnect,
netievent_tcpsend,
netievent_tcpstartread,
netievent_tcppauseread,
netievent_tcpaccept,
netievent_tcpstop,
netievent_tcpcancel,
netievent_tcpclose,
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
netievent_tcpdnsaccept,
netievent_tcpdnsconnect,
netievent_tcpdnssend,
netievent_tcpdnsread,
netievent_tcpdnscancel,
netievent_tcpdnsclose,
netievent_tcpdnsstop,
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
netievent_tlsdnsaccept,
netievent_tlsdnsconnect,
netievent_tlsdnssend,
netievent_tlsdnsread,
netievent_tlsdnscancel,
netievent_tlsdnsclose,
netievent_tlsdnsstop,
netievent_tlsdnscycle,
netievent_tlsdnsshutdown,
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
netievent_close,
netievent_shutdown,
netievent_stop,
netievent_pause,
netievent_connectcb,
netievent_readcb,
netievent_sendcb,
netievent_prio = 0xff, /* event type values higher than this
* will be treated as high-priority
* events, which can be processed
* while the netmgr is paused.
*/
netievent_udplisten,
netievent_tcplisten,
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
netievent_tcpdnslisten,
netievent_tlsdnslisten,
netievent_resume,
netievent_detach,
} isc__netievent_type;
typedef union {
isc_nm_recv_cb_t recv;
2020-02-14 08:14:03 +01:00
isc_nm_cb_t send;
isc_nm_cb_t connect;
isc_nm_accept_cb_t accept;
} isc__nm_cb_t;
/*
* Wrapper around uv_req_t with 'our' fields in it. req->data should
* always point to its parent. Note that we always allocate more than
* sizeof(struct) because we make room for different req types;
*/
2020-02-13 14:44:37 -08:00
#define UVREQ_MAGIC ISC_MAGIC('N', 'M', 'U', 'R')
#define VALID_UVREQ(t) ISC_MAGIC_VALID(t, UVREQ_MAGIC)
typedef struct isc__nm_uvreq isc__nm_uvreq_t;
struct isc__nm_uvreq {
2020-02-14 08:14:03 +01:00
int magic;
isc_nmsocket_t *sock;
isc_nmhandle_t *handle;
char tcplen[2]; /* The TCP DNS message length */
2020-02-14 08:14:03 +01:00
uv_buf_t uvbuf; /* translated isc_region_t, to be
* sent or received */
isc_sockaddr_t local; /* local address */
isc_sockaddr_t peer; /* peer address */
isc__nm_cb_t cb; /* callback */
void *cbarg; /* callback argument */
uv_pipe_t ipc; /* used for sending socket
* uv_handles to other threads */
union {
uv_handle_t handle;
2020-02-14 08:14:03 +01:00
uv_req_t req;
uv_getaddrinfo_t getaddrinfo;
uv_getnameinfo_t getnameinfo;
2020-02-14 08:14:03 +01:00
uv_shutdown_t shutdown;
uv_write_t write;
uv_connect_t connect;
uv_udp_send_t udp_send;
uv_fs_t fs;
uv_work_t work;
} uv_req;
ISC_LINK(isc__nm_uvreq_t) link;
};
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
void *
isc__nm_get_netievent(isc_nm_t *mgr, isc__netievent_type type);
/*%<
* Allocate an ievent and set the type.
*/
void
isc__nm_put_netievent(isc_nm_t *mgr, void *ievent);
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
/*
* The macros here are used to simulate the "inheritance" in C, there's the base
* netievent structure that contains just its own type and socket, and there are
* extended netievent types that also have handles or requests or other data.
*
* The macros here ensure that:
*
* 1. every netievent type has matching definition, declaration and
* implementation
*
* 2. we handle all the netievent types of same subclass the same, e.g. if the
* extended netievent contains handle, we always attach to the handle in
* the ctor and detach from the handle in dtor.
*
* There are three macros here for each netievent subclass:
*
* 1. NETIEVENT_*_TYPE(type) creates the typedef for each type; used below in
* this header
*
* 2. NETIEVENT_*_DECL(type) generates the declaration of the get and put
* functions (isc__nm_get_netievent_* and isc__nm_put_netievent_*); used
* below in this header
*
* 3. NETIEVENT_*_DEF(type) generates the definition of the functions; used
* either in netmgr.c or matching protocol file (e.g. udp.c, tcp.c, etc.)
*/
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
#define NETIEVENT__SOCKET \
isc__netievent_type type; \
isc_nmsocket_t *sock; \
const char *file; \
unsigned int line; \
const char *func
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
typedef struct isc__netievent__socket {
NETIEVENT__SOCKET;
} isc__netievent__socket_t;
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
#define NETIEVENT_SOCKET_TYPE(type) \
typedef isc__netievent__socket_t isc__netievent_##type##_t;
#define NETIEVENT_SOCKET_DECL(type) \
isc__netievent_##type##_t *isc__nm_get_netievent_##type( \
isc_nm_t *nm, isc_nmsocket_t *sock); \
void isc__nm_put_netievent_##type(isc_nm_t *nm, \
isc__netievent_##type##_t *ievent);
#define NETIEVENT_SOCKET_DEF(type) \
isc__netievent_##type##_t *isc__nm_get_netievent_##type( \
isc_nm_t *nm, isc_nmsocket_t *sock) { \
isc__netievent_##type##_t *ievent = \
isc__nm_get_netievent(nm, netievent_##type); \
isc__nmsocket_attach(sock, &ievent->sock); \
\
return (ievent); \
} \
\
void isc__nm_put_netievent_##type(isc_nm_t *nm, \
isc__netievent_##type##_t *ievent) { \
isc__nmsocket_detach(&ievent->sock); \
isc__nm_put_netievent(nm, ievent); \
}
typedef struct isc__netievent__socket_req {
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
NETIEVENT__SOCKET;
2020-02-14 08:14:03 +01:00
isc__nm_uvreq_t *req;
} isc__netievent__socket_req_t;
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
#define NETIEVENT_SOCKET_REQ_TYPE(type) \
typedef isc__netievent__socket_req_t isc__netievent_##type##_t;
#define NETIEVENT_SOCKET_REQ_DECL(type) \
isc__netievent_##type##_t *isc__nm_get_netievent_##type( \
isc_nm_t *nm, isc_nmsocket_t *sock, isc__nm_uvreq_t *req); \
void isc__nm_put_netievent_##type(isc_nm_t *nm, \
isc__netievent_##type##_t *ievent);
#define NETIEVENT_SOCKET_REQ_DEF(type) \
isc__netievent_##type##_t *isc__nm_get_netievent_##type( \
isc_nm_t *nm, isc_nmsocket_t *sock, isc__nm_uvreq_t *req) { \
isc__netievent_##type##_t *ievent = \
isc__nm_get_netievent(nm, netievent_##type); \
isc__nmsocket_attach(sock, &ievent->sock); \
ievent->req = req; \
\
return (ievent); \
} \
\
void isc__nm_put_netievent_##type(isc_nm_t *nm, \
isc__netievent_##type##_t *ievent) { \
isc__nmsocket_detach(&ievent->sock); \
isc__nm_put_netievent(nm, ievent); \
}
typedef struct isc__netievent__socket_req_result {
isc__netievent_type type;
isc_nmsocket_t *sock;
isc__nm_uvreq_t *req;
isc_result_t result;
} isc__netievent__socket_req_result_t;
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
#define NETIEVENT_SOCKET_REQ_RESULT_TYPE(type) \
typedef isc__netievent__socket_req_result_t isc__netievent_##type##_t;
#define NETIEVENT_SOCKET_REQ_RESULT_DECL(type) \
isc__netievent_##type##_t *isc__nm_get_netievent_##type( \
isc_nm_t *nm, isc_nmsocket_t *sock, isc__nm_uvreq_t *req, \
isc_result_t result); \
void isc__nm_put_netievent_##type(isc_nm_t *nm, \
isc__netievent_##type##_t *ievent);
#define NETIEVENT_SOCKET_REQ_RESULT_DEF(type) \
isc__netievent_##type##_t *isc__nm_get_netievent_##type( \
isc_nm_t *nm, isc_nmsocket_t *sock, isc__nm_uvreq_t *req, \
isc_result_t result) { \
isc__netievent_##type##_t *ievent = \
isc__nm_get_netievent(nm, netievent_##type); \
isc__nmsocket_attach(sock, &ievent->sock); \
ievent->req = req; \
ievent->result = result; \
\
return (ievent); \
} \
\
void isc__nm_put_netievent_##type(isc_nm_t *nm, \
isc__netievent_##type##_t *ievent) { \
isc__nmsocket_detach(&ievent->sock); \
isc__nm_put_netievent(nm, ievent); \
}
typedef struct isc__netievent__socket_handle {
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
NETIEVENT__SOCKET;
2020-02-14 08:14:03 +01:00
isc_nmhandle_t *handle;
} isc__netievent__socket_handle_t;
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
#define NETIEVENT_SOCKET_HANDLE_TYPE(type) \
typedef isc__netievent__socket_handle_t isc__netievent_##type##_t;
#define NETIEVENT_SOCKET_HANDLE_DECL(type) \
isc__netievent_##type##_t *isc__nm_get_netievent_##type( \
isc_nm_t *nm, isc_nmsocket_t *sock, isc_nmhandle_t *handle); \
void isc__nm_put_netievent_##type(isc_nm_t *nm, \
isc__netievent_##type##_t *ievent);
#define NETIEVENT_SOCKET_HANDLE_DEF(type) \
isc__netievent_##type##_t *isc__nm_get_netievent_##type( \
isc_nm_t *nm, isc_nmsocket_t *sock, isc_nmhandle_t *handle) { \
isc__netievent_##type##_t *ievent = \
isc__nm_get_netievent(nm, netievent_##type); \
isc__nmsocket_attach(sock, &ievent->sock); \
isc_nmhandle_attach(handle, &ievent->handle); \
\
return (ievent); \
} \
\
void isc__nm_put_netievent_##type(isc_nm_t *nm, \
isc__netievent_##type##_t *ievent) { \
isc__nmsocket_detach(&ievent->sock); \
isc_nmhandle_detach(&ievent->handle); \
isc__nm_put_netievent(nm, ievent); \
}
typedef struct isc__netievent__socket_quota {
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
NETIEVENT__SOCKET;
isc_quota_t *quota;
} isc__netievent__socket_quota_t;
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
#define NETIEVENT_SOCKET_QUOTA_TYPE(type) \
typedef isc__netievent__socket_quota_t isc__netievent_##type##_t;
#define NETIEVENT_SOCKET_QUOTA_DECL(type) \
isc__netievent_##type##_t *isc__nm_get_netievent_##type( \
isc_nm_t *nm, isc_nmsocket_t *sock, isc_quota_t *quota); \
void isc__nm_put_netievent_##type(isc_nm_t *nm, \
isc__netievent_##type##_t *ievent);
#define NETIEVENT_SOCKET_QUOTA_DEF(type) \
isc__netievent_##type##_t *isc__nm_get_netievent_##type( \
isc_nm_t *nm, isc_nmsocket_t *sock, isc_quota_t *quota) { \
isc__netievent_##type##_t *ievent = \
isc__nm_get_netievent(nm, netievent_##type); \
isc__nmsocket_attach(sock, &ievent->sock); \
ievent->quota = quota; \
\
return (ievent); \
} \
\
void isc__nm_put_netievent_##type(isc_nm_t *nm, \
isc__netievent_##type##_t *ievent) { \
isc__nmsocket_detach(&ievent->sock); \
isc__nm_put_netievent(nm, ievent); \
}
typedef struct isc__netievent_udpsend {
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
NETIEVENT__SOCKET;
2020-02-14 08:14:03 +01:00
isc_sockaddr_t peer;
isc__nm_uvreq_t *req;
} isc__netievent_udpsend_t;
typedef struct isc__netievent_tlsconnect {
isc__netievent_type type;
isc_nmsocket_t *sock;
SSL_CTX *ctx;
isc_sockaddr_t local; /* local address */
isc_sockaddr_t peer; /* peer address */
} isc__netievent_tlsconnect_t;
typedef struct isc__netievent {
isc__netievent_type type;
} isc__netievent_t;
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
#define NETIEVENT_TYPE(type) typedef isc__netievent_t isc__netievent_##type##_t;
#define NETIEVENT_DECL(type) \
isc__netievent_##type##_t *isc__nm_get_netievent_##type(isc_nm_t *nm); \
void isc__nm_put_netievent_##type(isc_nm_t *nm, \
isc__netievent_##type##_t *ievent);
#define NETIEVENT_DEF(type) \
isc__netievent_##type##_t *isc__nm_get_netievent_##type( \
isc_nm_t *nm) { \
isc__netievent_##type##_t *ievent = \
isc__nm_get_netievent(nm, netievent_##type); \
\
return (ievent); \
} \
\
void isc__nm_put_netievent_##type(isc_nm_t *nm, \
isc__netievent_##type##_t *ievent) { \
isc__nm_put_netievent(nm, ievent); \
}
typedef union {
2020-02-14 08:14:03 +01:00
isc__netievent_t ni;
isc__netievent__socket_t nis;
isc__netievent__socket_req_t nisr;
isc__netievent_udpsend_t nius;
isc__netievent__socket_quota_t nisq;
isc__netievent_tlsconnect_t nitc;
} isc__netievent_storage_t;
/*
* Network manager
*/
2020-02-13 14:44:37 -08:00
#define NM_MAGIC ISC_MAGIC('N', 'E', 'T', 'M')
#define VALID_NM(t) ISC_MAGIC_VALID(t, NM_MAGIC)
struct isc_nm {
2020-02-14 08:14:03 +01:00
int magic;
isc_refcount_t references;
isc_mem_t *mctx;
uint32_t nworkers;
isc_mutex_t lock;
isc_condition_t wkstatecond;
isc__networker_t *workers;
isc_stats_t *stats;
isc_mempool_t *reqpool;
2020-02-14 08:14:03 +01:00
isc_mutex_t reqlock;
isc_mempool_t *evpool;
2020-02-14 08:14:03 +01:00
isc_mutex_t evlock;
uint_fast32_t workers_running;
uint_fast32_t workers_paused;
atomic_uint_fast32_t maxudp;
/*
* Active connections are being closed and new connections are
* no longer allowed.
*/
atomic_bool closing;
/*
* A worker is actively waiting for other workers, for example to
* stop listening; that means no other thread can do the same thing
* or pause, or we'll deadlock. We have to either re-enqueue our
* event or wait for the other one to finish if we want to pause.
*/
atomic_bool interlocked;
/*
* Timeout values for TCP connections, corresponding to
* tcp-intiial-timeout, tcp-idle-timeout, tcp-keepalive-timeout,
* and tcp-advertised-timeout. Note that these are stored in
* milliseconds so they can be used directly with the libuv timer,
* but they are configured in tenths of seconds.
*/
Fix the data race in accessing the isc_nm_t timers The following TSAN report about accessing the mgr timers (mgr->init, mgr->idle, mgr->keepalive and mgr->advertised) has been fixed in this commit: ================== WARNING: ThreadSanitizer: data race (pid=2746) Read of size 4 at 0x7b440008a948 by thread T18: #0 isc__nm_tcpdns_read /home/ondrej/Projects/bind9/lib/isc/netmgr/tcpdns.c:849:25 (libisc.so.1706+0x2ba0f) #1 isc_nm_read /home/ondrej/Projects/bind9/lib/isc/netmgr/netmgr.c:1679:3 (libisc.so.1706+0x22258) #2 tcpdns_connect_connect_cb /home/ondrej/Projects/bind9/lib/isc/tests/tcpdns_test.c:363:2 (tcpdns_test+0x4bc5fb) #3 isc__nm_async_connectcb /home/ondrej/Projects/bind9/lib/isc/netmgr/netmgr.c:1816:2 (libisc.so.1706+0x228c9) #4 isc__nm_connectcb /home/ondrej/Projects/bind9/lib/isc/netmgr/netmgr.c:1791:3 (libisc.so.1706+0x22713) #5 tcpdns_connect_cb /home/ondrej/Projects/bind9/lib/isc/netmgr/tcpdns.c:343:2 (libisc.so.1706+0x2d89d) #6 uv__stream_connect /home/ondrej/Projects/tsan/libuv/src/unix/stream.c:1381:5 (libuv.so.1+0x27c18) #7 uv__stream_io /home/ondrej/Projects/tsan/libuv/src/unix/stream.c:1298:5 (libuv.so.1+0x25977) #8 uv__io_poll /home/ondrej/Projects/tsan/libuv/src/unix/linux-core.c:462:11 (libuv.so.1+0x2e795) #9 uv_run /home/ondrej/Projects/tsan/libuv/src/unix/core.c:385:5 (libuv.so.1+0x158ec) #10 nm_thread /home/ondrej/Projects/bind9/lib/isc/netmgr/netmgr.c:530:11 (libisc.so.1706+0x1c94a) Previous write of size 4 at 0x7b440008a948 by main thread: #0 isc_nm_settimeouts /home/ondrej/Projects/bind9/lib/isc/netmgr/netmgr.c:490:12 (libisc.so.1706+0x1dda5) #1 tcpdns_recv_two /home/ondrej/Projects/bind9/lib/isc/tests/tcpdns_test.c:601:2 (tcpdns_test+0x4bad0e) #2 cmocka_run_one_test_or_fixture <null> (libcmocka.so.0+0x70be) #3 __libc_start_main /build/glibc-vjB4T1/glibc-2.28/csu/../csu/libc-start.c:308:16 (libc.so.6+0x2409a) Location is heap block of size 281 at 0x7b440008a840 allocated by main thread: #0 malloc <null> (tcpdns_test+0x42864b) #1 default_memalloc /home/ondrej/Projects/bind9/lib/isc/mem.c:713:8 (libisc.so.1706+0x6d261) #2 mem_get /home/ondrej/Projects/bind9/lib/isc/mem.c:622:8 (libisc.so.1706+0x69b9c) #3 isc___mem_get /home/ondrej/Projects/bind9/lib/isc/mem.c:1044:9 (libisc.so.1706+0x6d379) #4 isc__mem_get /home/ondrej/Projects/bind9/lib/isc/mem.c:2432:10 (libisc.so.1706+0x6889e) #5 isc_nm_start /home/ondrej/Projects/bind9/lib/isc/netmgr/netmgr.c:203:8 (libisc.so.1706+0x1c219) #6 nm_setup /home/ondrej/Projects/bind9/lib/isc/tests/tcpdns_test.c:244:11 (tcpdns_test+0x4baaa4) #7 cmocka_run_one_test_or_fixture <null> (libcmocka.so.0+0x70fd) #8 __libc_start_main /build/glibc-vjB4T1/glibc-2.28/csu/../csu/libc-start.c:308:16 (libc.so.6+0x2409a) Thread T18 'isc-net-0000' (tid=3513, running) created by main thread at: #0 pthread_create <null> (tcpdns_test+0x429e7b) #1 isc_thread_create /home/ondrej/Projects/bind9/lib/isc/pthreads/thread.c:73:8 (libisc.so.1706+0x8476a) #2 isc_nm_start /home/ondrej/Projects/bind9/lib/isc/netmgr/netmgr.c:271:3 (libisc.so.1706+0x1c66a) #3 nm_setup /home/ondrej/Projects/bind9/lib/isc/tests/tcpdns_test.c:244:11 (tcpdns_test+0x4baaa4) #4 cmocka_run_one_test_or_fixture <null> (libcmocka.so.0+0x70fd) #5 __libc_start_main /build/glibc-vjB4T1/glibc-2.28/csu/../csu/libc-start.c:308:16 (libc.so.6+0x2409a) SUMMARY: ThreadSanitizer: data race /home/ondrej/Projects/bind9/lib/isc/netmgr/tcpdns.c:849:25 in isc__nm_tcpdns_read ================== ThreadSanitizer: reported 1 warnings
2020-12-02 09:52:39 +01:00
atomic_uint_fast32_t init;
atomic_uint_fast32_t idle;
atomic_uint_fast32_t keepalive;
atomic_uint_fast32_t advertised;
#ifdef NETMGR_TRACE
ISC_LIST(isc_nmsocket_t) active_sockets;
#endif
};
typedef enum isc_nmsocket_type {
isc_nm_udpsocket,
isc_nm_udplistener, /* Aggregate of nm_udpsocks */
isc_nm_tcpsocket,
isc_nm_tcplistener,
isc_nm_tcpdnslistener,
isc_nm_tcpdnssocket,
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
isc_nm_tlsdnslistener,
isc_nm_tlsdnssocket
} isc_nmsocket_type;
/*%
* A universal structure for either a single socket or a group of
* dup'd/SO_REUSE_PORT-using sockets listening on the same interface.
*/
2020-02-13 14:44:37 -08:00
#define NMSOCK_MAGIC ISC_MAGIC('N', 'M', 'S', 'K')
#define VALID_NMSOCK(t) ISC_MAGIC_VALID(t, NMSOCK_MAGIC)
/*%
* Index into socket stat counter arrays.
*/
2020-08-24 09:23:19 +02:00
enum {
STATID_OPEN = 0,
STATID_OPENFAIL = 1,
STATID_CLOSE = 2,
STATID_BINDFAIL = 3,
STATID_CONNECTFAIL = 4,
STATID_CONNECT = 5,
STATID_ACCEPTFAIL = 6,
STATID_ACCEPT = 7,
STATID_SENDFAIL = 8,
STATID_RECVFAIL = 9,
STATID_ACTIVE = 10
};
struct isc_nmsocket {
/*% Unlocked, RO */
2020-02-14 08:14:03 +01:00
int magic;
int tid;
isc_nmsocket_type type;
2020-02-14 08:14:03 +01:00
isc_nm_t *mgr;
/*% Parent socket for multithreaded listeners */
isc_nmsocket_t *parent;
/*% Listener socket this connection was accepted on */
isc_nmsocket_t *listener;
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
/*% Self socket */
isc_nmsocket_t *self;
/*% TLS stuff */
struct tls {
SSL *ssl;
SSL_CTX *ctx;
BIO *app_rbio;
BIO *app_wbio;
BIO *ssl_rbio;
BIO *ssl_wbio;
2020-08-24 09:23:19 +02:00
enum {
TLS_STATE_NONE,
TLS_STATE_HANDSHAKE,
TLS_STATE_IO,
TLS_STATE_ERROR,
TLS_STATE_CLOSING
2020-08-24 09:23:19 +02:00
} state;
uv_buf_t senddata;
bool cycle;
isc_result_t pending_error;
/* List of active send requests. */
isc__nm_uvreq_t *pending_req;
} tls;
2019-12-03 00:07:59 -08:00
/*%
* quota is the TCP client, attached when a TCP connection
* is established. pquota is a non-attached pointer to the
* TCP client quota, stored in listening sockets but only
* attached in connected sockets.
*/
isc_quota_t *quota;
isc_quota_t *pquota;
isc_quota_cb_t quotacb;
/*%
* Socket statistics
*/
const isc_statscounter_t *statsindex;
2019-12-03 00:07:59 -08:00
/*%
* TCP read/connect timeout timers.
*/
uv_timer_t timer;
2020-02-14 08:14:03 +01:00
bool timer_initialized;
bool timer_running;
2020-02-14 08:14:03 +01:00
uint64_t read_timeout;
uint64_t connect_timeout;
/*% outer socket is for 'wrapped' sockets - e.g. tcpdns in tcp */
isc_nmsocket_t *outer;
/*% server socket for connections */
isc_nmsocket_t *server;
2019-12-03 00:07:59 -08:00
/*% Child sockets for multi-socket setups */
isc_nmsocket_t *children;
uint_fast32_t nchildren;
2020-02-14 08:14:03 +01:00
isc_nmiface_t *iface;
isc_nmhandle_t *statichandle;
isc_nmhandle_t *outerhandle;
2019-12-03 00:07:59 -08:00
/*% Extra data allocated at the end of each isc_nmhandle_t */
size_t extrahandlesize;
/*% TCP backlog */
int backlog;
/*% libuv data */
2020-02-14 08:14:03 +01:00
uv_os_sock_t fd;
union uv_any_handle uv_handle;
2019-12-03 00:07:59 -08:00
/*% Peer address */
isc_sockaddr_t peer;
/* Atomic */
2019-12-03 00:07:59 -08:00
/*% Number of running (e.g. listening) child sockets */
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
uint_fast32_t rchildren;
/*%
2019-12-03 00:07:59 -08:00
* Socket is active if it's listening, working, etc. If it's
* closing, then it doesn't make a sense, for example, to
* push handles or reqs for reuse.
*/
atomic_bool active;
atomic_bool destroying;
/*%
* Socket is closed if it's not active and all the possible
* callbacks were fired, there are no active handles, etc.
2019-12-03 00:07:59 -08:00
* If active==false but closed==false, that means the socket
* is closing.
*/
atomic_bool closing;
2020-02-14 08:14:03 +01:00
atomic_bool closed;
atomic_bool listening;
atomic_bool connecting;
atomic_bool connected;
bool accepting;
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
bool reading;
isc_refcount_t references;
/*%
* Established an outgoing connection, as client not server.
*/
atomic_bool client;
/*%
* TCPDNS socket has been set not to pipeline.
*/
atomic_bool sequential;
/*%
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
* The socket is processing read callback, this is guard to not read
* data before the readcb is back.
*/
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
bool processing;
/*%
* A TCP socket has had isc_nm_pauseread() called.
*/
atomic_bool readpaused;
/*%
* A TCP or TCPDNS socket has been set to use the keepalive
* timeout instead of the default idle timeout.
*/
atomic_bool keepalive;
/*%
* 'spare' handles for that can be reused to avoid allocations,
* for UDP.
*/
isc_astack_t *inactivehandles;
isc_astack_t *inactivereqs;
2019-12-03 00:07:59 -08:00
/*%
* Used to wait for TCP listening events to complete, and
* for the number of running children to reach zero during
* shutdown.
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
*
* We use two condition variables to prevent the race where the netmgr
* threads would be able to finish and destroy the socket before it's
* unlocked by the isc_nm_listen<proto>() function. So, the flow is as
* follows:
*
* 1. parent thread creates all children sockets and passes then to
* netthreads, looks at the signaling variable and WAIT(cond) until
* the childrens are done initializing
*
* 2. the events get picked by netthreads, calls the libuv API (and
* either succeeds or fails) and WAIT(scond) until all other
* children sockets in netthreads are initialized and the listening
* socket lock is unlocked
*
* 3. the control is given back to the parent thread which now either
* returns success or shutdowns the listener if an error has
* occured in the children netthread
*
* NOTE: The other approach would be doing an extra attach to the parent
* listening socket, and then detach it in the parent thread, but that
* breaks the promise that once the libuv socket is initialized on the
* nmsocket, the nmsocket needs to be handled only by matching
* netthread, so in fact that would add a complexity in a way that
* isc__nmsocket_detach would have to be converted to use an
* asynchrounous netievent.
*/
2020-02-14 08:14:03 +01:00
isc_mutex_t lock;
isc_condition_t cond;
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
isc_condition_t scond;
2019-12-03 00:07:59 -08:00
/*%
* Used to pass a result back from listen or connect events.
2019-12-03 00:07:59 -08:00
*/
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
isc_result_t result;
/*%
* List of active handles.
* ah - current position in 'ah_frees'; this represents the
* current number of active handles;
* ah_size - size of the 'ah_frees' and 'ah_handles' arrays
* ah_handles - array pointers to active handles
*
* Adding a handle
* - if ah == ah_size, reallocate
* - x = ah_frees[ah]
* - ah_frees[ah++] = 0;
* - ah_handles[x] = handle
* - x must be stored with the handle!
* Removing a handle:
* - ah_frees[--ah] = x
* - ah_handles[x] = NULL;
*
2019-11-15 13:22:13 -08:00
* XXX: for now this is locked with socket->lock, but we
* might want to change it to something lockless in the
* future.
*/
atomic_int_fast32_t ah;
2020-02-14 08:14:03 +01:00
size_t ah_size;
size_t *ah_frees;
isc_nmhandle_t **ah_handles;
2019-12-03 00:07:59 -08:00
/*% Buffer for TCPDNS processing */
2020-02-14 08:14:03 +01:00
size_t buf_size;
size_t buf_len;
unsigned char *buf;
2019-12-03 00:07:59 -08:00
/*%
* This function will be called with handle->sock
* as the argument whenever a handle's references drop
* to zero, after its reset callback has been called.
*/
isc_nm_opaquecb_t closehandle_cb;
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
isc_nmhandle_t *recv_handle;
isc_nm_recv_cb_t recv_cb;
void *recv_cbarg;
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
bool recv_read;
isc_nm_cb_t connect_cb;
void *connect_cbarg;
isc_nm_accept_cb_t accept_cb;
2020-02-14 08:14:03 +01:00
void *accept_cbarg;
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
atomic_int_fast32_t active_child_connections;
#ifdef NETMGR_TRACE
void *backtrace[TRACE_SIZE];
int backtrace_size;
LINK(isc_nmsocket_t) active_link;
ISC_LIST(isc_nmhandle_t) active_handles;
#endif
};
bool
isc__nm_in_netthread(void);
/*%
* Returns 'true' if we're in the network thread.
*/
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
void
isc__nm_maybe_enqueue_ievent(isc__networker_t *worker, isc__netievent_t *event);
/*%<
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
* If the caller is already in the matching nmthread, process the netievent
* directly, if not enqueue using isc__nm_enqueue_ievent().
*/
void
isc__nm_enqueue_ievent(isc__networker_t *worker, isc__netievent_t *event);
/*%<
* Enqueue an ievent onto a specific worker queue. (This the only safe
* way to use an isc__networker_t from another thread.)
*/
void
isc__nm_free_uvbuf(isc_nmsocket_t *sock, const uv_buf_t *buf);
/*%<
* Free a buffer allocated for a receive operation.
*
* Note that as currently implemented, this doesn't actually
* free anything, marks the isc__networker's UDP receive buffer
* as "not in use".
*/
isc_nmhandle_t *
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
isc___nmhandle_get(isc_nmsocket_t *sock, isc_sockaddr_t *peer,
isc_sockaddr_t *local FLARG);
/*%<
* Get a handle for the socket 'sock', allocating a new one
* if there isn't one available in 'sock->inactivehandles'.
*
* If 'peer' is not NULL, set the handle's peer address to 'peer',
* otherwise set it to 'sock->peer'.
*
* If 'local' is not NULL, set the handle's local address to 'local',
* otherwise set it to 'sock->iface->addr'.
*
* 'sock' will be attached to 'handle->sock'. The caller may need
* to detach the socket afterward.
*/
isc__nm_uvreq_t *
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
isc___nm_uvreq_get(isc_nm_t *mgr, isc_nmsocket_t *sock FLARG);
/*%<
* Get a UV request structure for the socket 'sock', allocating a
* new one if there isn't one available in 'sock->inactivereqs'.
*/
void
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
isc___nm_uvreq_put(isc__nm_uvreq_t **req, isc_nmsocket_t *sock FLARG);
/*%<
* Completes the use of a UV request structure, setting '*req' to NULL.
*
* The UV request is pushed onto the 'sock->inactivereqs' stack or,
* if that doesn't work, freed.
*/
void
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
isc___nmsocket_init(isc_nmsocket_t *sock, isc_nm_t *mgr, isc_nmsocket_type type,
isc_nmiface_t *iface FLARG);
/*%<
* Initialize socket 'sock', attach it to 'mgr', and set it to type 'type'
* and its interface to 'iface'.
*/
void
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
isc___nmsocket_attach(isc_nmsocket_t *sock, isc_nmsocket_t **target FLARG);
/*%<
* Attach to a socket, increasing refcount
*/
void
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
isc___nmsocket_detach(isc_nmsocket_t **socketp FLARG);
/*%<
* Detach from socket, decreasing refcount and possibly destroying the
* socket if it's no longer referenced.
*/
void
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
isc___nmsocket_prep_destroy(isc_nmsocket_t *sock FLARG);
/*%<
* Market 'sock' as inactive, close it if necessary, and destroy it
* if there are no remaining references or active handles.
*/
bool
isc__nmsocket_active(isc_nmsocket_t *sock);
/*%<
* Determine whether 'sock' is active by checking 'sock->active'
* or, for child sockets, 'sock->parent->active'.
*/
bool
isc__nmsocket_deactivate(isc_nmsocket_t *sock);
/*%<
* @brief Deactivate active socket
*
* Atomically deactive the socket by setting @p sock->active or, for child
* sockets, @p sock->parent->active to @c false
*
* @param[in] sock - valid nmsocket
* @return @c false if the socket was already inactive, @c true otherwise
*/
void
isc__nmsocket_clearcb(isc_nmsocket_t *sock);
/*%<
* Clear the recv and accept callbacks in 'sock'.
*/
void
isc__nm_connectcb(isc_nmsocket_t *sock, isc__nm_uvreq_t *uvreq,
isc_result_t eresult);
void
isc__nm_async_connectcb(isc__networker_t *worker, isc__netievent_t *ev0);
/*%<
* Issue a connect callback on the socket, used to call the callback
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
*/
void
isc__nm_readcb(isc_nmsocket_t *sock, isc__nm_uvreq_t *uvreq,
isc_result_t eresult);
void
isc__nm_async_readcb(isc__networker_t *worker, isc__netievent_t *ev0);
/*%<
* Issue a read callback on the socket, used to call the callback
* on failed conditions when the event can't be scheduled on the uv loop.
*
*/
void
isc__nm_sendcb(isc_nmsocket_t *sock, isc__nm_uvreq_t *uvreq,
isc_result_t eresult);
void
isc__nm_async_sendcb(isc__networker_t *worker, isc__netievent_t *ev0);
/*%<
* Issue a write callback on the socket, used to call the callback
* on failed conditions when the event can't be scheduled on the uv loop.
*/
void
isc__nm_async_shutdown(isc__networker_t *worker, isc__netievent_t *ev0);
/*%<
* Walk through all uv handles, get the underlying sockets and issue
* close on them.
*/
void
isc__nm_udp_send(isc_nmhandle_t *handle, isc_region_t *region, isc_nm_cb_t cb,
void *cbarg);
/*%<
* Back-end implementation of isc_nm_send() for UDP handles.
*/
void
isc__nm_udp_read(isc_nmhandle_t *handle, isc_nm_recv_cb_t cb, void *cbarg);
/*
* Back-end implementation of isc_nm_read() for UDP handles.
*/
void
isc__nm_udp_close(isc_nmsocket_t *sock);
/*%<
* Close a UDP socket.
*/
void
isc__nm_udp_cancelread(isc_nmhandle_t *handle);
/*%<
* Stop reading on a connected UDP handle.
*/
void
isc__nm_udp_shutdown(isc_nmsocket_t *sock);
/*%<
* Called during the shutdown process to close and clean up connected
* sockets.
*/
void
isc__nm_udp_stoplistening(isc_nmsocket_t *sock);
/*%<
* Stop listening on 'sock'.
*/
void
isc__nm_udp_settimeout(isc_nmhandle_t *handle, uint32_t timeout);
/*%<
* Set the recv timeout for the UDP socket associated with 'handle'.
*/
void
isc__nm_async_udplisten(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_async_udpconnect(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_async_udpstop(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_async_udpsend(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_async_udpread(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_async_udpcancel(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_async_udpclose(isc__networker_t *worker, isc__netievent_t *ev0);
/*%<
* Callback handlers for asynchronous UDP events (listen, stoplisten, send).
*/
void
isc__nm_tcp_send(isc_nmhandle_t *handle, isc_region_t *region, isc_nm_cb_t cb,
void *cbarg);
/*%<
* Back-end implementation of isc_nm_send() for TCP handles.
*/
void
isc__nm_tcp_read(isc_nmhandle_t *handle, isc_nm_recv_cb_t cb, void *cbarg);
/*
* Back-end implementation of isc_nm_read() for TCP handles.
*/
void
isc__nm_tcp_close(isc_nmsocket_t *sock);
/*%<
* Close a TCP socket.
*/
void
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
isc__nm_tcp_pauseread(isc_nmhandle_t *handle);
/*%<
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
* Pause reading on this handle, while still remembering the callback.
*/
void
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
isc__nm_tcp_resumeread(isc_nmhandle_t *handle);
/*%<
* Resume reading from socket.
*
*/
void
isc__nm_tcp_shutdown(isc_nmsocket_t *sock);
/*%<
* Called during the shutdown process to close and clean up connected
* sockets.
*/
void
isc__nm_tcp_cancelread(isc_nmhandle_t *handle);
/*%<
* Stop reading on a connected TCP handle.
*/
void
isc__nm_tcp_stoplistening(isc_nmsocket_t *sock);
/*%<
* Stop listening on 'sock'.
*/
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
int_fast32_t
isc__nm_tcp_listener_nactive(isc_nmsocket_t *sock);
/*%<
* Returns the number of active connections for the TCP listener socket.
*/
void
isc__nm_tcp_settimeout(isc_nmhandle_t *handle, uint32_t timeout);
/*%<
* Set the read timeout for the TCP socket associated with 'handle'.
*/
void
isc__nm_async_tcpconnect(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_async_tcplisten(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_async_tcpaccept(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_async_tcpstop(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_async_tcpsend(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_async_startread(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_async_pauseread(isc__networker_t *worker, isc__netievent_t *ev0);
void
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
isc__nm_async_tcpstartread(isc__networker_t *worker, isc__netievent_t *ev0);
void
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
isc__nm_async_tcppauseread(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_async_tcpcancel(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_async_tcpclose(isc__networker_t *worker, isc__netievent_t *ev0);
/*%<
* Callback handlers for asynchronous TCP events (connect, listen,
* stoplisten, send, read, pause, close).
*/
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
void
isc__nm_async_tcpdnsaccept(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_async_tcpdnsconnect(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_async_tcpdnslisten(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_tcpdns_send(isc_nmhandle_t *handle, isc_region_t *region,
isc_nm_cb_t cb, void *cbarg);
/*%<
* Back-end implementation of isc_nm_send() for TCPDNS handles.
*/
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
void
isc__nm_tcpdns_shutdown(isc_nmsocket_t *sock);
void
isc__nm_tcpdns_close(isc_nmsocket_t *sock);
/*%<
* Close a TCPDNS socket.
*/
void
isc__nm_tcpdns_stoplistening(isc_nmsocket_t *sock);
/*%<
* Stop listening on 'sock'.
*/
void
isc__nm_tcpdns_settimeout(isc_nmhandle_t *handle, uint32_t timeout);
/*%<
* Set the read timeout and reset the timer for the TCPDNS socket
* associated with 'handle', and the TCP socket it wraps around.
*/
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
void
isc__nm_async_tcpdnslisten(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_async_tcpdnsaccept(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_async_tcpdnscancel(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_async_tcpdnsclose(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_async_tcpdnssend(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_async_tcpdnsstop(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_async_tcpdnsread(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_tcpdns_read(isc_nmhandle_t *handle, isc_nm_recv_cb_t cb, void *cbarg);
void
isc__nm_tcpdns_cancelread(isc_nmhandle_t *handle);
/*%<
* Stop reading on a connected TCPDNS handle.
*/
void
isc__nm_async_tlsdnscycle(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_async_tlsdnsaccept(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_async_tlsdnsconnect(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_async_tlsdnslisten(isc__networker_t *worker, isc__netievent_t *ev0);
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
void
isc__nm_tlsdns_send(isc_nmhandle_t *handle, isc_region_t *region,
isc_nm_cb_t cb, void *cbarg);
/*%<
* Back-end implementation of isc_nm_send() for TLSDNS handles.
*/
void
isc__nm_tlsdns_shutdown(isc_nmsocket_t *sock);
void
isc__nm_tlsdns_close(isc_nmsocket_t *sock);
/*%<
* Close a TLSDNS socket.
*/
void
isc__nm_tlsdns_stoplistening(isc_nmsocket_t *sock);
/*%<
* Stop listening on 'sock'.
*/
void
isc__nm_tlsdns_settimeout(isc_nmhandle_t *handle, uint32_t timeout);
/*%<
* Set the read timeout and reset the timer for the TLSDNS socket
* associated with 'handle', and the TCP socket it wraps around.
*/
void
isc__nm_async_tlsdnslisten(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_async_tlsdnsaccept(isc__networker_t *worker, isc__netievent_t *ev0);
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
void
isc__nm_async_tlsdnscancel(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_async_tlsdnsclose(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_async_tlsdnssend(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_async_tlsdnsstop(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_async_tlsdnsshutdown(isc__networker_t *worker, isc__netievent_t *ev0);
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
void
isc__nm_async_tlsdnsread(isc__networker_t *worker, isc__netievent_t *ev0);
void
isc__nm_tlsdns_read(isc_nmhandle_t *handle, isc_nm_recv_cb_t cb, void *cbarg);
void
isc__nm_tlsdns_cancelread(isc_nmhandle_t *handle);
/*%<
* Stop reading on a connected TLSDNS handle.
*/
#define isc__nm_uverr2result(x) \
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
isc___nm_uverr2result(x, true, __FILE__, __LINE__, __func__)
isc_result_t
isc___nm_uverr2result(int uverr, bool dolog, const char *file,
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
unsigned int line, const char *func);
/*%<
* Convert a libuv error value into an isc_result_t. The
* list of supported error values is not complete; new users
* of this function should add any expected errors that are
* not already there.
*/
bool
isc__nm_acquire_interlocked(isc_nm_t *mgr);
/*%<
* Try to acquire interlocked state; return true if successful.
*/
void
isc__nm_drop_interlocked(isc_nm_t *mgr);
/*%<
* Drop interlocked state; signal waiters.
*/
void
isc__nm_acquire_interlocked_force(isc_nm_t *mgr);
/*%<
* Actively wait for interlocked state.
*/
void
isc__nm_incstats(isc_nm_t *mgr, isc_statscounter_t counterid);
/*%<
* Increment socket-related statistics counters.
*/
void
isc__nm_decstats(isc_nm_t *mgr, isc_statscounter_t counterid);
/*%<
* Decrement socket-related statistics counters.
*/
isc_result_t
isc__nm_socket(int domain, int type, int protocol, uv_os_sock_t *sockp);
/*%<
* Platform independent socket() version
*/
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
void
isc__nm_closesocket(uv_os_sock_t sock);
/*%<
* Platform independent closesocket() version
*/
isc_result_t
isc__nm_socket_freebind(uv_os_sock_t fd, sa_family_t sa_family);
/*%<
* Set the IP_FREEBIND (or equivalent) socket option on the uv_handle
*/
isc_result_t
isc__nm_socket_reuse(uv_os_sock_t fd);
/*%<
* Set the SO_REUSEADDR or SO_REUSEPORT (or equivalent) socket option on the fd
*/
isc_result_t
isc__nm_socket_reuse_lb(uv_os_sock_t fd);
/*%<
* Set the SO_REUSEPORT_LB (or equivalent) socket option on the fd
*/
isc_result_t
isc__nm_socket_incoming_cpu(uv_os_sock_t fd);
/*%<
* Set the SO_INCOMING_CPU socket option on the fd if available
*/
isc_result_t
isc__nm_socket_dontfrag(uv_os_sock_t fd, sa_family_t sa_family);
/*%<
* Set the SO_IP_DONTFRAG (or equivalent) socket option of the fd if available
*/
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
isc_result_t
isc__nm_socket_connectiontimeout(uv_os_sock_t fd, int timeout_ms);
/*%<
* Set the connection timeout in miliseconds, on non-Linux platforms,
* the minimum value must be at least 1000 (1 second).
*/
/*
* typedef all the netievent types
*/
NETIEVENT_SOCKET_TYPE(close);
NETIEVENT_SOCKET_TYPE(tcpclose);
NETIEVENT_SOCKET_TYPE(tcplisten);
NETIEVENT_SOCKET_TYPE(tcppauseread);
NETIEVENT_SOCKET_TYPE(tcpstop);
NETIEVENT_SOCKET_TYPE(udpclose);
NETIEVENT_SOCKET_TYPE(udplisten);
NETIEVENT_SOCKET_TYPE(udpread);
/* NETIEVENT_SOCKET_TYPE(udpsend); */ /* unique type, defined independently */
NETIEVENT_SOCKET_TYPE(udpstop);
NETIEVENT_SOCKET_TYPE(tcpdnsclose);
NETIEVENT_SOCKET_TYPE(tcpdnsread);
NETIEVENT_SOCKET_TYPE(tcpdnsstop);
NETIEVENT_SOCKET_TYPE(tcpdnslisten);
NETIEVENT_SOCKET_REQ_TYPE(tcpdnsconnect);
NETIEVENT_SOCKET_REQ_TYPE(tcpdnssend);
NETIEVENT_SOCKET_HANDLE_TYPE(tcpdnscancel);
NETIEVENT_SOCKET_QUOTA_TYPE(tcpdnsaccept);
NETIEVENT_SOCKET_TYPE(tlsdnsclose);
NETIEVENT_SOCKET_TYPE(tlsdnsread);
NETIEVENT_SOCKET_TYPE(tlsdnsstop);
NETIEVENT_SOCKET_TYPE(tlsdnsshutdown);
NETIEVENT_SOCKET_TYPE(tlsdnslisten);
NETIEVENT_SOCKET_REQ_TYPE(tlsdnsconnect);
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
NETIEVENT_SOCKET_REQ_TYPE(tlsdnssend);
NETIEVENT_SOCKET_HANDLE_TYPE(tlsdnscancel);
NETIEVENT_SOCKET_QUOTA_TYPE(tlsdnsaccept);
NETIEVENT_SOCKET_TYPE(tlsdnscycle);
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
NETIEVENT_SOCKET_REQ_TYPE(tcpconnect);
NETIEVENT_SOCKET_REQ_TYPE(tcpsend);
NETIEVENT_SOCKET_TYPE(tcpstartread);
NETIEVENT_SOCKET_REQ_TYPE(udpconnect);
NETIEVENT_SOCKET_REQ_RESULT_TYPE(connectcb);
NETIEVENT_SOCKET_REQ_RESULT_TYPE(readcb);
NETIEVENT_SOCKET_REQ_RESULT_TYPE(sendcb);
NETIEVENT_SOCKET_HANDLE_TYPE(detach);
NETIEVENT_SOCKET_HANDLE_TYPE(tcpcancel);
NETIEVENT_SOCKET_HANDLE_TYPE(udpcancel);
NETIEVENT_SOCKET_QUOTA_TYPE(tcpaccept);
NETIEVENT_TYPE(pause);
NETIEVENT_TYPE(resume);
NETIEVENT_TYPE(shutdown);
NETIEVENT_TYPE(stop);
/* Now declared the helper functions */
NETIEVENT_SOCKET_DECL(close);
NETIEVENT_SOCKET_DECL(tcpclose);
NETIEVENT_SOCKET_DECL(tcplisten);
NETIEVENT_SOCKET_DECL(tcppauseread);
NETIEVENT_SOCKET_DECL(tcpstartread);
NETIEVENT_SOCKET_DECL(tcpstop);
NETIEVENT_SOCKET_DECL(udpclose);
NETIEVENT_SOCKET_DECL(udplisten);
NETIEVENT_SOCKET_DECL(udpread);
NETIEVENT_SOCKET_DECL(udpsend);
NETIEVENT_SOCKET_DECL(udpstop);
NETIEVENT_SOCKET_DECL(tcpdnsclose);
NETIEVENT_SOCKET_DECL(tcpdnsread);
NETIEVENT_SOCKET_DECL(tcpdnsstop);
NETIEVENT_SOCKET_DECL(tcpdnslisten);
NETIEVENT_SOCKET_REQ_DECL(tcpdnsconnect);
NETIEVENT_SOCKET_REQ_DECL(tcpdnssend);
NETIEVENT_SOCKET_HANDLE_DECL(tcpdnscancel);
NETIEVENT_SOCKET_QUOTA_DECL(tcpdnsaccept);
NETIEVENT_SOCKET_DECL(tlsdnsclose);
NETIEVENT_SOCKET_DECL(tlsdnsread);
NETIEVENT_SOCKET_DECL(tlsdnsstop);
NETIEVENT_SOCKET_DECL(tlsdnsshutdown);
NETIEVENT_SOCKET_DECL(tlsdnslisten);
NETIEVENT_SOCKET_REQ_DECL(tlsdnsconnect);
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
NETIEVENT_SOCKET_REQ_DECL(tlsdnssend);
NETIEVENT_SOCKET_HANDLE_DECL(tlsdnscancel);
NETIEVENT_SOCKET_QUOTA_DECL(tlsdnsaccept);
NETIEVENT_SOCKET_DECL(tlsdnscycle);
Refactor netmgr and add more unit tests This is a part of the works that intends to make the netmgr stable, testable, maintainable and tested. It contains a numerous changes to the netmgr code and unfortunately, it was not possible to split this into smaller chunks as the work here needs to be committed as a complete works. NOTE: There's a quite a lot of duplicated code between udp.c, tcp.c and tcpdns.c and it should be a subject to refactoring in the future. The changes that are included in this commit are listed here (extensively, but not exclusively): * The netmgr_test unit test was split into individual tests (udp_test, tcp_test, tcpdns_test and newly added tcp_quota_test) * The udp_test and tcp_test has been extended to allow programatic failures from the libuv API. Unfortunately, we can't use cmocka mock() and will_return(), so we emulate the behaviour with #define and including the netmgr/{udp,tcp}.c source file directly. * The netievents that we put on the nm queue have variable number of members, out of these the isc_nmsocket_t and isc_nmhandle_t always needs to be attached before enqueueing the netievent_<foo> and detached after we have called the isc_nm_async_<foo> to ensure that the socket (handle) doesn't disappear between scheduling the event and actually executing the event. * Cancelling the in-flight TCP connection using libuv requires to call uv_close() on the original uv_tcp_t handle which just breaks too many assumptions we have in the netmgr code. Instead of using uv_timer for TCP connection timeouts, we use platform specific socket option. * Fix the synchronization between {nm,async}_{listentcp,tcpconnect} When isc_nm_listentcp() or isc_nm_tcpconnect() is called it was waiting for socket to either end up with error (that path was fine) or to be listening or connected using condition variable and mutex. Several things could happen: 0. everything is ok 1. the waiting thread would miss the SIGNAL() - because the enqueued event would be processed faster than we could start WAIT()ing. In case the operation would end up with error, it would be ok, as the error variable would be unchanged. 2. the waiting thread miss the sock->{connected,listening} = `true` would be set to `false` in the tcp_{listen,connect}close_cb() as the connection would be so short lived that the socket would be closed before we could even start WAIT()ing * The tcpdns has been converted to using libuv directly. Previously, the tcpdns protocol used tcp protocol from netmgr, this proved to be very complicated to understand, fix and make changes to. The new tcpdns protocol is modeled in a similar way how tcp netmgr protocol. Closes: #2194, #2283, #2318, #2266, #2034, #1920 * The tcp and tcpdns is now not using isc_uv_import/isc_uv_export to pass accepted TCP sockets between netthreads, but instead (similar to UDP) uses per netthread uv_loop listener. This greatly reduces the complexity as the socket is always run in the associated nm and uv loops, and we are also not touching the libuv internals. There's an unfortunate side effect though, the new code requires support for load-balanced sockets from the operating system for both UDP and TCP (see #2137). If the operating system doesn't support the load balanced sockets (either SO_REUSEPORT on Linux or SO_REUSEPORT_LB on FreeBSD 12+), the number of netthreads is limited to 1. * The netmgr has now two debugging #ifdefs: 1. Already existing NETMGR_TRACE prints any dangling nmsockets and nmhandles before triggering assertion failure. This options would reduce performance when enabled, but in theory, it could be enabled on low-performance systems. 2. New NETMGR_TRACE_VERBOSE option has been added that enables extensive netmgr logging that allows the software engineer to precisely track any attach/detach operations on the nmsockets and nmhandles. This is not suitable for any kind of production machine, only for debugging. * The tlsdns netmgr protocol has been split from the tcpdns and it still uses the old method of stacking the netmgr boxes on top of each other. We will have to refactor the tlsdns netmgr protocol to use the same approach - build the stack using only libuv and openssl. * Limit but not assert the tcp buffer size in tcp_alloc_cb Closes: #2061
2020-11-12 10:32:18 +01:00
NETIEVENT_SOCKET_REQ_DECL(tcpconnect);
NETIEVENT_SOCKET_REQ_DECL(tcpsend);
NETIEVENT_SOCKET_REQ_DECL(udpconnect);
NETIEVENT_SOCKET_REQ_RESULT_DECL(connectcb);
NETIEVENT_SOCKET_REQ_RESULT_DECL(readcb);
NETIEVENT_SOCKET_REQ_RESULT_DECL(sendcb);
NETIEVENT_SOCKET_HANDLE_DECL(udpcancel);
NETIEVENT_SOCKET_HANDLE_DECL(tcpcancel);
NETIEVENT_SOCKET_DECL(detach);
NETIEVENT_SOCKET_QUOTA_DECL(tcpaccept);
NETIEVENT_DECL(pause);
NETIEVENT_DECL(resume);
NETIEVENT_DECL(shutdown);
NETIEVENT_DECL(stop);