2
0
mirror of https://gitlab.isc.org/isc-projects/bind9 synced 2025-08-22 18:19:42 +00:00
bind/lib/ns/query.c
Petr Špaček 750d8a61b6 Convert DNS_RDATASETATTR_ bitfield manipulation to struct of bools
RRset ordering is now an enum inside struct rdataset attributes. This
was done to keep size to of the structure to its original value before
this MR.

I expect zero performance impact but it should be easier to deal with
attributes in debuggers and language servers.
2025-07-10 11:17:19 +02:00

11853 lines
319 KiB
C

/*
* Copyright (C) Internet Systems Consortium, Inc. ("ISC")
*
* SPDX-License-Identifier: MPL-2.0
*
* 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.
*/
/*! \file */
#include <ctype.h>
#include <inttypes.h>
#include <stdbool.h>
#include <stdint.h>
#include <string.h>
#include <isc/async.h>
#include <isc/counter.h>
#include <isc/hex.h>
#include <isc/list.h>
#include <isc/log.h>
#include <isc/mem.h>
#include <isc/once.h>
#include <isc/random.h>
#include <isc/result.h>
#include <isc/rwlock.h>
#include <isc/serial.h>
#include <isc/stats.h>
#include <isc/string.h>
#include <isc/thread.h>
#include <isc/util.h>
#include <dns/adb.h>
#include <dns/badcache.h>
#include <dns/byaddr.h>
#include <dns/cache.h>
#include <dns/db.h>
#include <dns/dlz.h>
#include <dns/dns64.h>
#include <dns/dnssec.h>
#include <dns/ede.h>
#include <dns/keytable.h>
#include <dns/message.h>
#include <dns/ncache.h>
#include <dns/nsec.h>
#include <dns/nsec3.h>
#include <dns/order.h>
#include <dns/rcode.h>
#include <dns/rdata.h>
#include <dns/rdataclass.h>
#include <dns/rdatalist.h>
#include <dns/rdataset.h>
#include <dns/rdatasetiter.h>
#include <dns/rdatastruct.h>
#include <dns/rdatatype.h>
#include <dns/resolver.h>
#include <dns/result.h>
#include <dns/stats.h>
#include <dns/tkey.h>
#include <dns/types.h>
#include <dns/view.h>
#include <dns/zone.h>
#include <dns/zt.h>
#include <ns/client.h>
#include <ns/hooks.h>
#include <ns/interfacemgr.h>
#include <ns/server.h>
#include <ns/stats.h>
#include <ns/xfrout.h>
#include "probes.h"
#if 0
/*
* It has been recommended that DNS64 be changed to return excluded
* AAAA addresses if DNS64 synthesis does not occur. This minimises
* the impact on the lookup results. While most DNS AAAA lookups are
* done to send IP packets to a host, not all of them are and filtering
* excluded addresses has a negative impact on those uses.
*/
#define dns64_bis_return_excluded_addresses 1
#endif /* if 0 */
#define QUERY_ERROR(qctx, r) \
do { \
(qctx)->result = r; \
(qctx)->want_restart = false; \
(qctx)->line = __LINE__; \
} while (0)
/*% Partial answer? */
#define PARTIALANSWER(c) \
(((c)->query.attributes & NS_QUERYATTR_PARTIALANSWER) != 0)
/*% Use Cache? */
#define USECACHE(c) (((c)->query.attributes & NS_QUERYATTR_CACHEOK) != 0)
/*% Recursion OK? */
#define RECURSIONOK(c) (((c)->query.attributes & NS_QUERYATTR_RECURSIONOK) != 0)
/*% Recursing? */
#define RECURSING(c) (((c)->query.attributes & NS_QUERYATTR_RECURSING) != 0)
/*% Want Recursion? */
#define WANTRECURSION(c) \
(((c)->query.attributes & NS_QUERYATTR_WANTRECURSION) != 0)
/*% Is TCP? */
#define TCP(c) (((c)->inner.attributes & NS_CLIENTATTR_TCP) != 0)
/*% This query needs to have been sent over TCP. Return TC=1. */
#define NEEDTCP(c) (((c)->inner.attributes & NS_CLIENTATTR_NEEDTCP) != 0)
/*% Want DNSSEC? */
#define WANTDNSSEC(c) (((c)->inner.attributes & NS_CLIENTATTR_WANTDNSSEC) != 0)
/*% Want WANTAD? */
#define WANTAD(c) (((c)->inner.attributes & NS_CLIENTATTR_WANTAD) != 0)
/*% Client presented a bad COOKIE. */
#define BADCOOKIE(c) (((c)->inner.attributes & NS_CLIENTATTR_BADCOOKIE) != 0)
/*% Client presented a valid COOKIE. */
#define HAVECOOKIE(c) (((c)->inner.attributes & NS_CLIENTATTR_HAVECOOKIE) != 0)
/*% Client presented a COOKIE. */
#define WANTCOOKIE(c) (((c)->inner.attributes & NS_CLIENTATTR_WANTCOOKIE) != 0)
/*% Client presented a CLIENT-SUBNET option. */
#define HAVEECS(c) (((c)->inner.attributes & NS_CLIENTATTR_HAVEECS) != 0)
/*% No authority? */
#define NOAUTHORITY(c) (((c)->query.attributes & NS_QUERYATTR_NOAUTHORITY) != 0)
/*% No additional? */
#define NOADDITIONAL(c) \
(((c)->query.attributes & NS_QUERYATTR_NOADDITIONAL) != 0)
/*% Secure? */
#define SECURE(c) (((c)->query.attributes & NS_QUERYATTR_SECURE) != 0)
/*% DNS64 A lookup? */
#define DNS64(c) (((c)->query.attributes & NS_QUERYATTR_DNS64) != 0)
#define DNS64EXCLUDE(c) \
(((c)->query.attributes & NS_QUERYATTR_DNS64EXCLUDE) != 0)
#define REDIRECT(c) (((c)->query.attributes & NS_QUERYATTR_REDIRECT) != 0)
/*% Was the client already sent a response? */
#define QUERY_ANSWERED(q) (((q)->attributes & NS_QUERYATTR_ANSWERED) != 0)
/*% Does the query allow stale data in the response? */
#define QUERY_STALEOK(q) (((q)->attributes & NS_QUERYATTR_STALEOK) != 0)
/*% Does the query wants to check for stale RRset due to a timeout? */
#define QUERY_STALETIMEOUT(q) (((q)->dboptions & DNS_DBFIND_STALETIMEOUT) != 0)
/*% Does the rdataset 'r' have an attached 'No QNAME Proof'? */
#define NOQNAME(r) (((r)->attributes.noqname))
/*% Does the rdataset 'r' contain a stale answer? */
#define STALE(r) (((r)->attributes.stale))
/*% Does the rdataset 'r' is stale and within stale-refresh-time? */
#define STALE_WINDOW(r) (((r)->attributes.stale_window))
#ifdef WANT_QUERYTRACE
static void
client_trace(ns_client_t *client, int level, const char *message) {
if (client != NULL && client->query.qname != NULL) {
if (isc_log_wouldlog(level)) {
char qbuf[DNS_NAME_FORMATSIZE];
char tbuf[DNS_RDATATYPE_FORMATSIZE];
dns_name_format(client->query.qname, qbuf,
sizeof(qbuf));
dns_rdatatype_format(client->query.qtype, tbuf,
sizeof(tbuf));
isc_log_write(NS_LOGCATEGORY_CLIENT, NS_LOGMODULE_QUERY,
level,
"query client=%p thread=0x%" PRIxPTR
"(%s/%s): %s",
client, isc_thread_self(), qbuf, tbuf,
message);
}
} else {
isc_log_write(NS_LOGCATEGORY_CLIENT, NS_LOGMODULE_QUERY, level,
"query client=%p thread=0x%" PRIxPTR
"(<unknown-query>): %s",
client, isc_thread_self(), message);
}
}
#define CTRACE(l, m) client_trace(client, l, m)
#define CCTRACE(l, m) client_trace(qctx->client, l, m)
#else /* ifdef WANT_QUERYTRACE */
#define CTRACE(l, m) ((void)m)
#define CCTRACE(l, m) ((void)m)
#endif /* WANT_QUERYTRACE */
#define PENDINGOK(x) (((x) & DNS_DBFIND_PENDINGOK) != 0)
#define SFCACHE_CDFLAG 0x1
/*
* SAVE and RESTORE have the same semantics as:
*
* foo_attach(b, &a);
* foo_detach(&b);
*
* without the locking and magic testing.
*
* We use the names SAVE and RESTORE to show the operation being performed,
* even though the two macros are identical.
*/
#define SAVE(a, b) \
do { \
INSIST(a == NULL); \
a = b; \
b = NULL; \
} while (0)
#define RESTORE(a, b) SAVE(a, b)
static bool
validate(ns_client_t *client, dns_db_t *db, dns_name_t *name,
dns_rdataset_t *rdataset, dns_rdataset_t *sigrdataset);
static void
query_findclosestnsec3(dns_name_t *qname, dns_db_t *db,
dns_dbversion_t *version, ns_client_t *client,
dns_rdataset_t *rdataset, dns_rdataset_t *sigrdataset,
dns_name_t *fname, bool exact, dns_name_t *found);
static void
log_queryerror(ns_client_t *client, isc_result_t result, int line, int level);
static void
rpz_st_clear(ns_client_t *client);
static bool
rpz_ck_dnssec(ns_client_t *client, isc_result_t qresult,
dns_rdataset_t *rdataset, dns_rdataset_t *sigrdataset);
static void
log_noexistnodata(void *val, int level, const char *fmt, ...)
ISC_FORMAT_PRINTF(3, 4);
static isc_result_t
query_addanswer(query_ctx_t *qctx);
static isc_result_t
query_prepare_delegation_response(query_ctx_t *qctx);
static isc_result_t
acquire_recursionquota(ns_client_t *client);
static void
release_recursionquota(ns_client_t *client);
/*
* Return the hooktable in use with 'qctx', or if there isn't one
* set, return the default hooktable.
*/
static ns_hooktable_t *
get_hooktab(query_ctx_t *qctx) {
if (qctx == NULL || qctx->view == NULL || qctx->view->hooktable == NULL)
{
return ns__hook_table;
}
return qctx->view->hooktable;
}
/*
* Call the specified hook function in every configured module that implements
* that function. If any hook function returns NS_HOOK_RETURN, we
* set 'result' and terminate processing by jumping to the 'cleanup' tag.
*
* (Note that a hook function may set the 'result' to ISC_R_SUCCESS but
* still terminate processing within the calling function. That's why this
* is a macro instead of a static function; it needs to be able to use
* 'goto cleanup' regardless of the return value.)
*/
#define CALL_HOOK(_id, _qctx) \
do { \
isc_result_t _res = result; \
ns_hooktable_t *_tab = get_hooktab(_qctx); \
ns_hook_t *_hook; \
_hook = ISC_LIST_HEAD((*_tab)[_id]); \
while (_hook != NULL) { \
ns_hook_action_t _func = _hook->action; \
void *_data = _hook->action_data; \
INSIST(_func != NULL); \
switch (_func(_qctx, _data, &_res)) { \
case NS_HOOK_CONTINUE: \
_hook = ISC_LIST_NEXT(_hook, link); \
break; \
case NS_HOOK_RETURN: \
result = _res; \
goto cleanup; \
default: \
UNREACHABLE(); \
} \
} \
} while (false)
/*
* Call the specified hook function in every configured module that
* implements that function. All modules are called; hook function return
* codes are ignored. This is intended for use with initialization and
* destruction calls which *must* run in every configured module.
*
* (This could be implemented as a static void function, but is left as a
* macro for symmetry with CALL_HOOK above.)
*/
#define CALL_HOOK_NORETURN(_id, _qctx) \
do { \
isc_result_t _res; \
ns_hooktable_t *_tab = get_hooktab(_qctx); \
ns_hook_t *_hook; \
_hook = ISC_LIST_HEAD((*_tab)[_id]); \
while (_hook != NULL) { \
ns_hook_action_t _func = _hook->action; \
void *_data = _hook->action_data; \
INSIST(_func != NULL); \
_func(_qctx, _data, &_res); \
_hook = ISC_LIST_NEXT(_hook, link); \
} \
} while (false)
/*
* The functions defined below implement the query logic that previously lived
* in the single very complex function query_find(). The query_ctx_t structure
* defined in <ns/query.h> maintains state from function to function. The call
* flow for the general query processing algorithm is described below:
*
* 1. Set up query context and other resources for a client
* query (query_setup())
*
* 2. Start the search (ns__query_start())
*
* 3. Identify authoritative data sources which may have an answer;
* search them (query_lookup()). If an answer is found, go to 7.
*
* 4. If recursion or cache access are allowed, search the cache
* (query_lookup() again, using the cache database) to find a better
* answer. If an answer is found, go to 7.
*
* 5. If recursion is allowed, begin recursion (ns_query_recurse()).
* Go to 15 to clean up this phase of the query. When recursion
* is complete, processing will resume at 6.
*
* 6. Resume from recursion; set up query context for resumed processing.
*
* 7. Determine what sort of answer we've found (query_gotanswer())
* and call other functions accordingly:
* - not found (auth or cache), go to 8
* - delegation, go to 9
* - no such domain (auth), go to 10
* - empty answer (auth), go to 11
* - negative response (cache), go to 12
* - answer found, go to 13
*
* 8. The answer was not found in the database (query_notfound().
* Set up a referral and go to 9.
*
* 9. Handle a delegation response (query_delegation()). If we need
* to and are allowed to recurse (query_delegation_recurse()), go to 5,
* otherwise go to 15 to clean up and return the delegation to the client.
*
* 10. No such domain (query_nxdomain()). Attempt redirection; if
* unsuccessful, add authority section records (query_addsoa(),
* query_addauth()), then go to 15 to return NXDOMAIN to client.
*
* 11. Empty answer (query_nodata()). Add authority section records
* (query_addsoa(), query_addauth()) and signatures if authoritative
* (query_sign_nodata()) then go to 15 and return
* NOERROR/ANCOUNT=0 to client.
*
* 12. No such domain or empty answer returned from cache (query_ncache()).
* Set response code appropriately, go to 11.
*
* 13. Prepare a response (query_prepresponse()) and then fill it
* appropriately (query_respond(), or for type ANY,
* query_respond_any()).
*
* 14. If a restart is needed due to CNAME/DNAME chaining, go to 2.
*
* 15. Clean up resources. If recursing, stop and wait for the event
* handler to be called back (step 6). If an answer is ready,
* return it to the client.
*
* (XXX: This description omits several special cases including
* DNS64, RPZ, RRL, and the SERVFAIL cache. It also doesn't discuss
* plugins.)
*/
static void
query_trace(query_ctx_t *qctx);
static void
qctx_init(ns_client_t *client, dns_fetchresponse_t **respp,
dns_rdatatype_t qtype, query_ctx_t *qctx);
static isc_result_t
qctx_prepare_buffers(query_ctx_t *qctx, isc_buffer_t *buffer);
static void
qctx_freedata(query_ctx_t *qctx);
static void
qctx_destroy(query_ctx_t *qctx);
static void
query_setup(ns_client_t *client, dns_rdatatype_t qtype);
static isc_result_t
query_lookup(query_ctx_t *qctx);
static void
fetch_callback(void *arg);
static void
recparam_update(ns_query_recparam_t *param, dns_rdatatype_t qtype,
const dns_name_t *qname, const dns_name_t *qdomain);
static isc_result_t
query_resume(query_ctx_t *qctx);
static isc_result_t
query_checkrrl(query_ctx_t *qctx, isc_result_t result);
static isc_result_t
query_checkrpz(query_ctx_t *qctx, isc_result_t result);
static isc_result_t
query_rpzcname(query_ctx_t *qctx, dns_name_t *cname);
static isc_result_t
query_gotanswer(query_ctx_t *qctx, isc_result_t result);
static void
query_addnoqnameproof(query_ctx_t *qctx);
static isc_result_t
query_respond_any(query_ctx_t *qctx);
static isc_result_t
query_respond(query_ctx_t *qctx);
static isc_result_t
query_dns64(query_ctx_t *qctx);
static void
query_filter64(query_ctx_t *qctx);
static isc_result_t
query_notfound(query_ctx_t *qctx);
static isc_result_t
query_zone_delegation(query_ctx_t *qctx);
static isc_result_t
query_delegation(query_ctx_t *qctx);
static isc_result_t
query_delegation_recurse(query_ctx_t *qctx);
static void
query_addds(query_ctx_t *qctx);
static isc_result_t
query_nodata(query_ctx_t *qctx, isc_result_t result);
static isc_result_t
query_sign_nodata(query_ctx_t *qctx);
static void
query_addnxrrsetnsec(query_ctx_t *qctx);
static isc_result_t
query_nxdomain(query_ctx_t *qctx, isc_result_t result);
static isc_result_t
query_redirect(query_ctx_t *qctx, isc_result_t result);
static isc_result_t
query_ncache(query_ctx_t *qctx, isc_result_t result);
static isc_result_t
query_coveringnsec(query_ctx_t *qctx);
static isc_result_t
query_zerottl_refetch(query_ctx_t *qctx);
static isc_result_t
query_cname(query_ctx_t *qctx);
static isc_result_t
query_dname(query_ctx_t *qctx);
static void
query_addcname(query_ctx_t *qctx, dns_trust_t trust, dns_ttl_t ttl);
static isc_result_t
query_prepresponse(query_ctx_t *qctx);
static isc_result_t
query_addsoa(query_ctx_t *qctx, unsigned int override_ttl,
dns_section_t section);
static isc_result_t
query_addns(query_ctx_t *qctx);
static void
query_addbestns(query_ctx_t *qctx);
static void
query_addwildcardproof(query_ctx_t *qctx, bool ispositive, bool nodata);
static void
query_addauth(query_ctx_t *qctx);
/*
* Increment query statistics counters.
*/
static void
inc_stats(ns_client_t *client, isc_statscounter_t counter) {
dns_zone_t *zone = client->query.authzone;
dns_rdatatype_t qtype;
dns_rdataset_t *rdataset;
isc_stats_t *zonestats;
dns_stats_t *querystats = NULL;
ns_stats_increment(client->manager->sctx->nsstats, counter);
if (zone == NULL) {
return;
}
/* Do regular response type stats */
zonestats = dns_zone_getrequeststats(zone);
if (zonestats != NULL) {
isc_stats_increment(zonestats, counter);
}
/* Do query type statistics
*
* We only increment per-type if we're using the authoritative
* answer counter, preventing double-counting.
*/
if (counter == ns_statscounter_authans) {
querystats = dns_zone_getrcvquerystats(zone);
if (querystats != NULL) {
rdataset = ISC_LIST_HEAD(client->query.qname->list);
if (rdataset != NULL) {
qtype = rdataset->type;
dns_rdatatypestats_increment(querystats, qtype);
}
}
}
}
#define NS_CLIENT_FLAGS_FORMATSIZE sizeof("+E(255)STDCV")
static inline void
ns_client_log_flags(ns_client_t *client, unsigned int flags,
unsigned int extflags, char *buf, size_t len) {
isc_buffer_t b;
isc_buffer_init(&b, buf, len);
isc_buffer_putuint8(&b, WANTRECURSION(client) ? '+' : '-');
if (client->inner.ednsversion >= 0) {
char ednsbuf[sizeof("E(255)")] = { 0 };
snprintf(ednsbuf, sizeof(ednsbuf), "E(%hhu)",
(unsigned char)client->inner.ednsversion);
isc_buffer_putstr(&b, ednsbuf);
}
if (client->inner.signer != NULL) {
isc_buffer_putuint8(&b, 'S');
}
if (TCP(client)) {
isc_buffer_putuint8(&b, 'T');
}
if ((extflags & DNS_MESSAGEEXTFLAG_DO) != 0) {
isc_buffer_putuint8(&b, 'D');
}
if ((flags & DNS_MESSAGEFLAG_CD) != 0) {
isc_buffer_putuint8(&b, 'C');
}
if (HAVECOOKIE(client)) {
isc_buffer_putuint8(&b, 'V');
} else if (WANTCOOKIE(client)) {
isc_buffer_putuint8(&b, 'K');
}
isc_buffer_putuint8(&b, 0);
}
#define NS_CLIENT_ECS_FORMATSIZE (DNS_ECS_FORMATSIZE + sizeof(" [ECS ]") - 1)
static inline void
ns_client_log_ecs(ns_client_t *client, char *ecsbuf, size_t len) {
strlcpy(ecsbuf, " [ECS ", len);
dns_ecs_format(&client->inner.ecs, ecsbuf + 6, len - 6);
strlcat(ecsbuf, "]", len);
}
static inline void
log_response(ns_client_t *client, dns_rcode_t rcode) {
char namebuf[DNS_NAME_FORMATSIZE];
char typebuf[DNS_RDATATYPE_FORMATSIZE];
char classbuf[DNS_RDATACLASS_FORMATSIZE];
char rcodebuf[20];
char onbuf[ISC_NETADDR_FORMATSIZE];
char ecsbuf[NS_CLIENT_ECS_FORMATSIZE] = { 0 };
char flagsbuf[NS_CLIENT_FLAGS_FORMATSIZE] = { 0 };
isc_buffer_t b;
int level = ISC_LOG_INFO;
if (!isc_log_wouldlog(level)) {
return;
}
dns_name_format(client->query.origqname, namebuf, sizeof(namebuf));
dns_rdataclass_format(client->message->rdclass, classbuf,
sizeof(classbuf));
dns_rdatatype_format(client->query.qtype, typebuf, sizeof(typebuf));
isc_buffer_init(&b, rcodebuf, sizeof(rcodebuf));
dns_rcode_totext(rcode, &b);
isc_buffer_putuint8(&b, 0);
isc_netaddr_format(&client->inner.destaddr, onbuf, sizeof(onbuf));
if (HAVEECS(client)) {
ns_client_log_ecs(client, ecsbuf, sizeof(ecsbuf));
}
ns_client_log_flags(client, client->message->flags,
client->inner.extflags, flagsbuf, sizeof(flagsbuf));
ns_client_log(client, NS_LOGCATEGORY_RESPONSES, NS_LOGMODULE_QUERY,
level, "response: %s %s %s %s %u %u %u %s (%s)%s",
namebuf, classbuf, typebuf, rcodebuf,
client->message->counts[DNS_SECTION_ANSWER],
client->message->counts[DNS_SECTION_AUTHORITY],
client->message->counts[DNS_SECTION_ADDITIONAL], flagsbuf,
onbuf, ecsbuf);
}
static void
query_send(ns_client_t *client) {
isc_statscounter_t counter;
if ((client->message->flags & DNS_MESSAGEFLAG_AA) == 0) {
inc_stats(client, ns_statscounter_nonauthans);
} else {
inc_stats(client, ns_statscounter_authans);
}
if (client->message->rcode == dns_rcode_noerror) {
dns_section_t answer = DNS_SECTION_ANSWER;
if (ISC_LIST_EMPTY(client->message->sections[answer])) {
if (client->query.isreferral) {
counter = ns_statscounter_referral;
} else {
counter = ns_statscounter_nxrrset;
}
} else {
counter = ns_statscounter_success;
}
} else if (client->message->rcode == dns_rcode_nxdomain) {
counter = ns_statscounter_nxdomain;
} else if (client->message->rcode == dns_rcode_badcookie) {
counter = ns_statscounter_badcookie;
} else { /* We end up here in case of YXDOMAIN, and maybe others */
counter = ns_statscounter_failure;
}
inc_stats(client, counter);
ns_client_send(client);
if ((client->manager->sctx->options & NS_SERVER_LOGRESPONSES) != 0) {
log_response(client, client->message->rcode);
}
isc_nmhandle_detach(&client->inner.reqhandle);
}
static void
query_error(ns_client_t *client, isc_result_t result, int line) {
int loglevel = ISC_LOG_DEBUG(3);
dns_rcode_t rcode;
rcode = dns_result_torcode(result);
switch (rcode) {
case dns_rcode_servfail:
loglevel = ISC_LOG_DEBUG(1);
inc_stats(client, ns_statscounter_servfail);
break;
case dns_rcode_formerr:
inc_stats(client, ns_statscounter_formerr);
break;
default:
inc_stats(client, ns_statscounter_failure);
break;
}
if ((client->manager->sctx->options & NS_SERVER_LOGQUERIES) != 0) {
loglevel = ISC_LOG_INFO;
}
log_queryerror(client, result, line, loglevel);
ns_client_error(client, result);
if (client->query.origqname != NULL &&
(client->manager->sctx->options & NS_SERVER_LOGRESPONSES) != 0)
{
log_response(client, rcode);
}
isc_nmhandle_detach(&client->inner.reqhandle);
}
static void
query_next(ns_client_t *client, isc_result_t result) {
if (result == DNS_R_DUPLICATE) {
inc_stats(client, ns_statscounter_duplicate);
} else if (result == DNS_R_DROP) {
inc_stats(client, ns_statscounter_dropped);
} else {
inc_stats(client, ns_statscounter_failure);
}
ns_client_drop(client, result);
isc_nmhandle_detach(&client->inner.reqhandle);
}
static void
query_freefreeversions(ns_client_t *client, bool everything) {
unsigned int i = 0;
ISC_LIST_FOREACH (client->query.freeversions, dbversion, link) {
/*
* If we're not freeing everything, we keep the first three
* dbversions structures around.
*/
if (i++ > 3 || everything) {
ISC_LIST_UNLINK(client->query.freeversions, dbversion,
link);
isc_mem_put(client->manager->mctx, dbversion,
sizeof(*dbversion));
}
}
}
void
ns_query_cancel(ns_client_t *client) {
REQUIRE(NS_CLIENT_VALID(client));
LOCK(&client->query.fetchlock);
for (int i = 0; i < RECTYPE_COUNT; i++) {
dns_fetch_t **fetchp = &client->query.recursions[i].fetch;
if (*fetchp != NULL) {
dns_resolver_cancelfetch(*fetchp);
*fetchp = NULL;
}
}
if (client->query.hookactx != NULL) {
client->query.hookactx->cancel(client->query.hookactx);
client->query.hookactx = NULL;
}
UNLOCK(&client->query.fetchlock);
}
static void
query_reset(ns_client_t *client, bool everything) {
CTRACE(ISC_LOG_DEBUG(3), "query_reset");
/*%
* Reset the query state of a client to its default state.
*/
/*
* Cancel the fetch if it's running.
*/
ns_query_cancel(client);
/*
* Cleanup any active versions.
*/
ISC_LIST_FOREACH (client->query.activeversions, dbversion, link) {
dns_db_closeversion(dbversion->db, &dbversion->version, false);
dns_db_detach(&dbversion->db);
ISC_LIST_INITANDAPPEND(client->query.freeversions, dbversion,
link);
}
ISC_LIST_INIT(client->query.activeversions);
if (client->query.authdb != NULL) {
dns_db_detach(&client->query.authdb);
}
if (client->query.authzone != NULL) {
dns_zone_detach(&client->query.authzone);
}
if (client->query.dns64_aaaa != NULL) {
ns_client_putrdataset(client, &client->query.dns64_aaaa);
}
if (client->query.dns64_sigaaaa != NULL) {
ns_client_putrdataset(client, &client->query.dns64_sigaaaa);
}
if (client->query.dns64_aaaaok != NULL) {
isc_mem_cput(client->manager->mctx, client->query.dns64_aaaaok,
client->query.dns64_aaaaoklen, sizeof(bool));
client->query.dns64_aaaaoklen = 0;
}
ns_client_putrdataset(client, &client->query.redirect.rdataset);
ns_client_putrdataset(client, &client->query.redirect.sigrdataset);
if (client->query.redirect.db != NULL) {
if (client->query.redirect.node != NULL) {
dns_db_detachnode(client->query.redirect.db,
&client->query.redirect.node);
}
dns_db_detach(&client->query.redirect.db);
}
if (client->query.redirect.zone != NULL) {
dns_zone_detach(&client->query.redirect.zone);
}
query_freefreeversions(client, everything);
ISC_LIST_FOREACH (client->query.namebufs, dbuf, link) {
if (ISC_LIST_NEXT(dbuf, link) != NULL || everything) {
ISC_LIST_UNLINK(client->query.namebufs, dbuf, link);
isc_buffer_free(&dbuf);
}
}
if (client->query.restarts > 0) {
/*
* client->query.qname was dynamically allocated.
*/
dns_message_puttempname(client->message, &client->query.qname);
}
client->query.qname = NULL;
client->query.attributes = (NS_QUERYATTR_RECURSIONOK |
NS_QUERYATTR_CACHEOK | NS_QUERYATTR_SECURE);
client->query.restarts = 0;
client->query.timerset = false;
if (client->query.rpz_st != NULL) {
rpz_st_clear(client);
if (everything) {
isc_mem_put(client->manager->mctx, client->query.rpz_st,
sizeof(*client->query.rpz_st));
}
}
if (client->query.qc != NULL) {
isc_counter_detach(&client->query.qc);
}
client->query.origqname = NULL;
client->query.dboptions = 0;
client->query.fetchoptions = 0;
client->query.gluedb = NULL;
client->query.authdbset = false;
client->query.isreferral = false;
client->query.dns64_options = 0;
client->query.dns64_ttl = UINT32_MAX;
recparam_update(&client->query.recparam, 0, NULL, NULL);
client->query.root_key_sentinel_keyid = 0;
client->query.root_key_sentinel_is_ta = false;
client->query.root_key_sentinel_not_ta = false;
}
static void
query_cleanup(ns_client_t *client) {
query_reset(client, false);
}
void
ns_query_free(ns_client_t *client) {
REQUIRE(NS_CLIENT_VALID(client));
query_reset(client, true);
}
void
ns_query_init(ns_client_t *client) {
REQUIRE(NS_CLIENT_VALID(client));
client->query = (ns_query_t){ 0 };
ISC_LIST_INIT(client->query.namebufs);
ISC_LIST_INIT(client->query.activeversions);
ISC_LIST_INIT(client->query.freeversions);
/*
* This mutex is destroyed when the client is destroyed in
* exit_check().
*/
isc_mutex_init(&client->query.fetchlock);
client->query.redirect.fname =
dns_fixedname_initname(&client->query.redirect.fixed);
query_reset(client, false);
ns_client_newdbversion(client, 3);
ns_client_newnamebuf(client);
}
/*%
* Check if 'client' is allowed to query the cache of its associated view.
* Unless 'options' has the 'nolog' flag set, log the result of cache ACL
* evaluation using the appropriate level, along with 'name' and 'qtype'.
*
* The cache ACL is only evaluated once for each client and then the result is
* cached: if NS_QUERYATTR_CACHEACLOKVALID is set in client->query.attributes,
* cache ACL evaluation has already been performed. The evaluation result is
* also stored in client->query.attributes: if NS_QUERYATTR_CACHEACLOK is set,
* the client is allowed cache access.
*
* Returns:
*
*\li #ISC_R_SUCCESS 'client' is allowed to access cache
*\li #DNS_R_REFUSED 'client' is not allowed to access cache
*/
static isc_result_t
query_checkcacheaccess(ns_client_t *client, const dns_name_t *name,
dns_rdatatype_t qtype, dns_getdb_options_t options) {
isc_result_t result;
if ((client->query.attributes & NS_QUERYATTR_CACHEACLOKVALID) == 0) {
enum refusal_reasons {
ALLOW_QUERY_CACHE,
ALLOW_QUERY_CACHE_ON
};
static const char *acl_desc[] = {
"allow-query-cache did not match",
"allow-query-cache-on did not match",
};
/*
* The view's cache ACLs have not yet been evaluated.
* Do it now. Both allow-query-cache and
* allow-query-cache-on must be satisfied.
*/
char msg[NS_CLIENT_ACLMSGSIZE("query (cache)")];
enum refusal_reasons refusal_reason = ALLOW_QUERY_CACHE;
result = ns_client_checkaclsilent(
client, NULL, client->inner.view->cacheacl, true);
if (result == ISC_R_SUCCESS) {
refusal_reason = ALLOW_QUERY_CACHE_ON;
result = ns_client_checkaclsilent(
client, &client->inner.destaddr,
client->inner.view->cacheonacl, true);
}
if (result == ISC_R_SUCCESS) {
/*
* We were allowed by the "allow-query-cache" ACL.
*/
client->query.attributes |= NS_QUERYATTR_CACHEACLOK;
if (!options.nolog &&
isc_log_wouldlog(ISC_LOG_DEBUG(3)))
{
ns_client_aclmsg("query (cache)", name, qtype,
client->inner.view->rdclass,
msg, sizeof(msg));
ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
NS_LOGMODULE_QUERY,
ISC_LOG_DEBUG(3), "%s approved",
msg);
}
} else {
/*
* We were denied by the "allow-query-cache" ACL.
* There is no need to clear NS_QUERYATTR_CACHEACLOK
* since it is cleared by query_reset(), before query
* processing starts.
*/
dns_ede_add(&client->edectx, DNS_EDE_PROHIBITED, NULL);
if (!options.nolog) {
ns_client_aclmsg("query (cache)", name, qtype,
client->inner.view->rdclass,
msg, sizeof(msg));
ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
NS_LOGMODULE_QUERY, ISC_LOG_INFO,
"%s denied (%s)", msg,
acl_desc[refusal_reason]);
}
}
/*
* Evaluation has been finished; make sure we will just consult
* NS_QUERYATTR_CACHEACLOK for this client from now on.
*/
client->query.attributes |= NS_QUERYATTR_CACHEACLOKVALID;
}
return (client->query.attributes & NS_QUERYATTR_CACHEACLOK) != 0
? ISC_R_SUCCESS
: DNS_R_REFUSED;
}
static isc_result_t
query_validatezonedb(ns_client_t *client, const dns_name_t *name,
dns_rdatatype_t qtype, dns_getdb_options_t options,
dns_zone_t *zone, dns_db_t *db,
dns_dbversion_t **versionp) {
isc_result_t result;
dns_acl_t *queryacl, *queryonacl;
ns_dbversion_t *dbversion;
REQUIRE(zone != NULL);
REQUIRE(db != NULL);
/*
* Mirror zone data is treated as cache data.
*/
if (dns_zone_gettype(zone) == dns_zone_mirror) {
return query_checkcacheaccess(client, name, qtype, options);
}
/*
* This limits our searching to the zone where the first name
* (the query target) was looked for. This prevents following
* CNAMES or DNAMES into other zones and prevents returning
* additional data from other zones. This does not apply if we're
* answering a query where recursion is requested and allowed.
*/
if (client->query.rpz_st == NULL &&
!(WANTRECURSION(client) && RECURSIONOK(client)) &&
client->query.authdbset && db != client->query.authdb)
{
return DNS_R_REFUSED;
}
/*
* Non recursive query to a static-stub zone is prohibited; its
* zone content is not public data, but a part of local configuration
* and should not be disclosed.
*/
if (dns_zone_gettype(zone) == dns_zone_staticstub &&
!RECURSIONOK(client))
{
return DNS_R_REFUSED;
}
/*
* If the zone has an ACL, we'll check it, otherwise
* we use the view's "allow-query" ACL. Each ACL is only checked
* once per query.
*
* Also, get the database version to use.
*/
/*
* Get the current version of this database.
*/
dbversion = ns_client_findversion(client, db);
if (dbversion == NULL) {
CTRACE(ISC_LOG_ERROR, "unable to get db version");
return DNS_R_SERVFAIL;
}
if (options.ignoreacl) {
goto approved;
}
if (dbversion->acl_checked) {
if (!dbversion->queryok) {
return DNS_R_REFUSED;
}
goto approved;
}
queryacl = dns_zone_getqueryacl(zone);
if (queryacl == NULL) {
queryacl = client->inner.view->queryacl;
if ((client->query.attributes & NS_QUERYATTR_QUERYOKVALID) != 0)
{
/*
* We've evaluated the view's queryacl already. If
* NS_QUERYATTR_QUERYOK is set, then the client is
* allowed to make queries, otherwise the query should
* be refused.
*/
dbversion->acl_checked = true;
if ((client->query.attributes & NS_QUERYATTR_QUERYOK) ==
0)
{
dbversion->queryok = false;
return DNS_R_REFUSED;
}
dbversion->queryok = true;
goto approved;
}
}
result = ns_client_checkaclsilent(client, NULL, queryacl, true);
if (!options.nolog) {
char msg[NS_CLIENT_ACLMSGSIZE("query")];
if (result == ISC_R_SUCCESS) {
if (isc_log_wouldlog(ISC_LOG_DEBUG(3))) {
ns_client_aclmsg("query", name, qtype,
client->inner.view->rdclass,
msg, sizeof(msg));
ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
NS_LOGMODULE_QUERY,
ISC_LOG_DEBUG(3), "%s approved",
msg);
}
} else {
ns_client_aclmsg("query", name, qtype,
client->inner.view->rdclass, msg,
sizeof(msg));
ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
NS_LOGMODULE_QUERY, ISC_LOG_INFO,
"%s denied", msg);
dns_ede_add(&client->edectx, DNS_EDE_PROHIBITED, NULL);
}
}
if (queryacl == client->inner.view->queryacl) {
if (result == ISC_R_SUCCESS) {
/*
* We were allowed by the default
* "allow-query" ACL. Remember this so we
* don't have to check again.
*/
client->query.attributes |= NS_QUERYATTR_QUERYOK;
}
/*
* We've now evaluated the view's query ACL, and
* the NS_QUERYATTR_QUERYOK attribute is now valid.
*/
client->query.attributes |= NS_QUERYATTR_QUERYOKVALID;
}
/* If and only if we've gotten this far, check allow-query-on too */
if (result == ISC_R_SUCCESS) {
queryonacl = dns_zone_getqueryonacl(zone);
if (queryonacl == NULL) {
queryonacl = client->inner.view->queryonacl;
}
result = ns_client_checkaclsilent(
client, &client->inner.destaddr, queryonacl, true);
if (result != ISC_R_SUCCESS) {
dns_ede_add(&client->edectx, DNS_EDE_PROHIBITED, NULL);
}
if (!options.nolog && result != ISC_R_SUCCESS) {
ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
NS_LOGMODULE_QUERY, ISC_LOG_INFO,
"query-on denied");
}
}
dbversion->acl_checked = true;
if (result != ISC_R_SUCCESS) {
dbversion->queryok = false;
return DNS_R_REFUSED;
}
dbversion->queryok = true;
approved:
/* Transfer ownership, if necessary. */
SET_IF_NOT_NULL(versionp, dbversion->version);
return ISC_R_SUCCESS;
}
static isc_result_t
query_getzonedb(ns_client_t *client, const dns_name_t *name,
dns_rdatatype_t qtype, dns_getdb_options_t options,
dns_zone_t **zonep, dns_db_t **dbp,
dns_dbversion_t **versionp) {
isc_result_t result;
unsigned int ztoptions;
dns_zone_t *zone = NULL;
dns_db_t *db = NULL;
bool partial = false;
REQUIRE(zonep != NULL && *zonep == NULL);
REQUIRE(dbp != NULL && *dbp == NULL);
/*%
* Find a zone database to answer the query.
*/
ztoptions = DNS_ZTFIND_MIRROR;
if (options.noexact) {
ztoptions |= DNS_ZTFIND_NOEXACT;
}
result = dns_view_findzone(client->inner.view, name, ztoptions, &zone);
if (result == DNS_R_PARTIALMATCH) {
partial = true;
}
if (result == ISC_R_SUCCESS || result == DNS_R_PARTIALMATCH) {
result = dns_zone_getdb(zone, &db);
}
if (result != ISC_R_SUCCESS) {
goto fail;
}
result = query_validatezonedb(client, name, qtype, options, zone, db,
versionp);
if (result != ISC_R_SUCCESS) {
goto fail;
}
/* Transfer ownership. */
*zonep = zone;
*dbp = db;
if (partial && options.partial) {
return DNS_R_PARTIALMATCH;
}
return ISC_R_SUCCESS;
fail:
if (zone != NULL) {
dns_zone_detach(&zone);
}
if (db != NULL) {
dns_db_detach(&db);
}
return result;
}
static void
rpz_log_rewrite(ns_client_t *client, bool disabled, dns_rpz_policy_t policy,
dns_rpz_type_t type, dns_zone_t *p_zone, dns_name_t *p_name,
dns_name_t *cname, dns_rpz_num_t rpz_num) {
char cname_buf[DNS_NAME_FORMATSIZE] = { 0 };
char p_name_buf[DNS_NAME_FORMATSIZE];
char qname_buf[DNS_NAME_FORMATSIZE];
char classbuf[DNS_RDATACLASS_FORMATSIZE];
char typebuf[DNS_RDATATYPE_FORMATSIZE];
const char *s1 = cname_buf, *s2 = cname_buf;
dns_rdataset_t *rdataset;
dns_rpz_st_t *st;
isc_stats_t *zonestats;
/*
* Count enabled rewrites in the global counter.
* Count both enabled and disabled rewrites for each zone.
*/
if (!disabled && policy != DNS_RPZ_POLICY_PASSTHRU) {
ns_stats_increment(client->manager->sctx->nsstats,
ns_statscounter_rpz_rewrites);
}
if (p_zone != NULL) {
zonestats = dns_zone_getrequeststats(p_zone);
if (zonestats != NULL) {
isc_stats_increment(zonestats,
ns_statscounter_rpz_rewrites);
}
}
if (!isc_log_wouldlog(DNS_RPZ_INFO_LEVEL)) {
return;
}
st = client->query.rpz_st;
if ((st->popt.no_log & DNS_RPZ_ZBIT(rpz_num)) != 0) {
return;
}
dns_name_format(client->query.qname, qname_buf, sizeof(qname_buf));
dns_name_format(p_name, p_name_buf, sizeof(p_name_buf));
if (cname != NULL) {
s1 = " (CNAME to: ";
dns_name_format(cname, cname_buf, sizeof(cname_buf));
s2 = ")";
}
/*
* Log Qclass and Qtype in addition to existing
* fields.
*/
rdataset = ISC_LIST_HEAD(client->query.origqname->list);
INSIST(rdataset != NULL);
dns_rdataclass_format(rdataset->rdclass, classbuf, sizeof(classbuf));
dns_rdatatype_format(rdataset->type, typebuf, sizeof(typebuf));
/* It's possible to have a separate log channel for rpz passthru. */
isc_logcategory_t log_cat = (policy == DNS_RPZ_POLICY_PASSTHRU)
? DNS_LOGCATEGORY_RPZ_PASSTHRU
: DNS_LOGCATEGORY_RPZ;
ns_client_log(client, log_cat, NS_LOGMODULE_QUERY, DNS_RPZ_INFO_LEVEL,
"%srpz %s %s rewrite %s/%s/%s via %s%s%s%s",
disabled ? "disabled " : "", dns_rpz_type2str(type),
dns_rpz_policy2str(policy), qname_buf, typebuf, classbuf,
p_name_buf, s1, cname_buf, s2);
}
static void
rpz_log_fail_helper(ns_client_t *client, int level, dns_name_t *p_name,
dns_rpz_type_t rpz_type1, dns_rpz_type_t rpz_type2,
const char *str, isc_result_t result) {
char qnamebuf[DNS_NAME_FORMATSIZE];
char p_namebuf[DNS_NAME_FORMATSIZE];
const char *failed, *via, *slash, *str_blank;
const char *rpztypestr1;
const char *rpztypestr2;
if (!isc_log_wouldlog(level)) {
return;
}
/*
* bin/tests/system/rpz/tests.sh looks for "rpz.*failed" for problems.
*/
if (level <= DNS_RPZ_DEBUG_LEVEL1) {
failed = " failed: ";
} else {
failed = ": ";
}
rpztypestr1 = dns_rpz_type2str(rpz_type1);
if (rpz_type2 != DNS_RPZ_TYPE_BAD) {
slash = "/";
rpztypestr2 = dns_rpz_type2str(rpz_type2);
} else {
slash = "";
rpztypestr2 = "";
}
str_blank = (*str != ' ' && *str != '\0') ? " " : "";
dns_name_format(client->query.qname, qnamebuf, sizeof(qnamebuf));
if (p_name != NULL) {
via = " via ";
dns_name_format(p_name, p_namebuf, sizeof(p_namebuf));
} else {
via = "";
p_namebuf[0] = '\0';
}
ns_client_log(client, NS_LOGCATEGORY_QUERY_ERRORS, NS_LOGMODULE_QUERY,
level, "rpz %s%s%s rewrite %s%s%s%s%s%s%s", rpztypestr1,
slash, rpztypestr2, qnamebuf, via, p_namebuf, str_blank,
str, failed, isc_result_totext(result));
}
static void
rpz_log_fail(ns_client_t *client, int level, dns_name_t *p_name,
dns_rpz_type_t rpz_type, const char *str, isc_result_t result) {
rpz_log_fail_helper(client, level, p_name, rpz_type, DNS_RPZ_TYPE_BAD,
str, result);
}
/*
* Get a policy rewrite zone database.
*/
static isc_result_t
rpz_getdb(ns_client_t *client, dns_name_t *p_name, dns_rpz_type_t rpz_type,
dns_zone_t **zonep, dns_db_t **dbp, dns_dbversion_t **versionp) {
char qnamebuf[DNS_NAME_FORMATSIZE];
char p_namebuf[DNS_NAME_FORMATSIZE];
dns_dbversion_t *rpz_version = NULL;
isc_result_t result;
CTRACE(ISC_LOG_DEBUG(3), "rpz_getdb");
dns_getdb_options_t options = { .ignoreacl = true };
result = query_getzonedb(client, p_name, dns_rdatatype_any, options,
zonep, dbp, &rpz_version);
if (result == ISC_R_SUCCESS) {
dns_rpz_st_t *st = client->query.rpz_st;
/*
* It isn't meaningful to log this message when
* logging is disabled for some policy zones.
*/
if (st->popt.no_log == 0 &&
isc_log_wouldlog(DNS_RPZ_DEBUG_LEVEL2))
{
dns_name_format(client->query.qname, qnamebuf,
sizeof(qnamebuf));
dns_name_format(p_name, p_namebuf, sizeof(p_namebuf));
ns_client_log(client, DNS_LOGCATEGORY_RPZ,
NS_LOGMODULE_QUERY, DNS_RPZ_DEBUG_LEVEL2,
"try rpz %s rewrite %s via %s",
dns_rpz_type2str(rpz_type), qnamebuf,
p_namebuf);
}
*versionp = rpz_version;
return ISC_R_SUCCESS;
}
rpz_log_fail(client, DNS_RPZ_ERROR_LEVEL, p_name, rpz_type,
"query_getzonedb()", result);
return result;
}
/*%
* Find a cache database to answer the query. This may fail with DNS_R_REFUSED
* if the client is not allowed to use the cache.
*/
static isc_result_t
query_getcachedb(ns_client_t *client, const dns_name_t *name,
dns_rdatatype_t qtype, dns_db_t **dbp,
dns_getdb_options_t options) {
isc_result_t result;
dns_db_t *db = NULL;
REQUIRE(dbp != NULL && *dbp == NULL);
if (!USECACHE(client)) {
return DNS_R_REFUSED;
}
dns_db_attach(client->inner.view->cachedb, &db);
result = query_checkcacheaccess(client, name, qtype, options);
if (result != ISC_R_SUCCESS) {
dns_db_detach(&db);
}
/*
* If query_checkcacheaccess() succeeded, transfer ownership of 'db'.
* Otherwise, 'db' will be NULL due to the dns_db_detach() call above.
*/
*dbp = db;
return result;
}
static isc_result_t
query_getdb(ns_client_t *client, dns_name_t *name, dns_rdatatype_t qtype,
dns_getdb_options_t options, dns_zone_t **zonep, dns_db_t **dbp,
dns_dbversion_t **versionp, bool *is_zonep) {
isc_result_t result;
isc_result_t tresult;
unsigned int namelabels;
unsigned int zonelabels;
dns_zone_t *zone = NULL;
REQUIRE(zonep != NULL && *zonep == NULL);
/* Calculate how many labels are in name. */
namelabels = dns_name_countlabels(name);
zonelabels = 0;
/* Try to find name in bind's standard database. */
result = query_getzonedb(client, name, qtype, options, &zone, dbp,
versionp);
/* See how many labels are in the zone's name. */
if (result == ISC_R_SUCCESS && zone != NULL) {
zonelabels = dns_name_countlabels(dns_zone_getorigin(zone));
}
/*
* If # zone labels < # name labels, try to find an even better match
* Only try if DLZ drivers are loaded for this view
*/
if (zonelabels < namelabels &&
!ISC_LIST_EMPTY(client->inner.view->dlz_searched))
{
dns_clientinfomethods_t cm;
dns_clientinfo_t ci;
dns_db_t *tdbp;
dns_clientinfomethods_init(&cm, ns_client_sourceip);
dns_clientinfo_init(&ci, client, NULL);
dns_clientinfo_setecs(&ci, &client->inner.ecs);
tdbp = NULL;
tresult = dns_view_searchdlz(client->inner.view, name,
zonelabels, &cm, &ci, &tdbp);
/* If we successful, we found a better match. */
if (tresult == ISC_R_SUCCESS) {
ns_dbversion_t *dbversion;
/*
* If the previous search returned a zone, detach it.
*/
if (zone != NULL) {
dns_zone_detach(&zone);
}
/*
* If the previous search returned a database,
* detach it.
*/
if (*dbp != NULL) {
dns_db_detach(dbp);
}
/*
* If the previous search returned a version, clear it.
*/
*versionp = NULL;
dbversion = ns_client_findversion(client, tdbp);
/*
* Be sure to return our database.
*/
*dbp = tdbp;
*versionp = dbversion->version;
/*
* We return a null zone, No stats for DLZ zones.
*/
zone = NULL;
result = tresult;
}
}
/* If successful, Transfer ownership of zone. */
if (result == ISC_R_SUCCESS) {
*zonep = zone;
/*
* If neither attempt above succeeded, return the cache instead
*/
*is_zonep = true;
} else {
if (result == ISC_R_NOTFOUND) {
result = query_getcachedb(client, name, qtype, dbp,
options);
}
*is_zonep = false;
}
return result;
}
static bool
query_isduplicate(ns_client_t *client, dns_name_t *name, dns_rdatatype_t type,
dns_name_t **mnamep) {
dns_section_t section;
dns_name_t *mname = NULL;
isc_result_t result;
CTRACE(ISC_LOG_DEBUG(3), "query_isduplicate");
for (section = DNS_SECTION_ANSWER; section <= DNS_SECTION_ADDITIONAL;
section++)
{
result = dns_message_findname(client->message, section, name,
type, 0, &mname, NULL);
if (result == ISC_R_SUCCESS) {
/*
* We've already got this RRset in the response.
*/
CTRACE(ISC_LOG_DEBUG(3), "query_isduplicate: true: "
"done");
return true;
} else if (result == DNS_R_NXRRSET) {
/*
* The name exists, but the rdataset does not.
*/
if (section == DNS_SECTION_ADDITIONAL) {
break;
}
} else {
RUNTIME_CHECK(result == DNS_R_NXDOMAIN);
}
mname = NULL;
}
SET_IF_NOT_NULL(mnamep, mname);
CTRACE(ISC_LOG_DEBUG(3), "query_isduplicate: false: done");
return false;
}
/*
* Look up data for given 'name' and 'type' in given 'version' of 'db' for
* 'client'. Called from query_additionalauth().
*
* If the lookup is successful:
*
* - store the node containing the result at 'nodep',
*
* - store the owner name of the returned node in 'fname',
*
* - if 'type' is not ANY, dns_db_findext() will put the exact rdataset being
* looked for in 'rdataset' and its signatures (if any) in 'sigrdataset',
*
* - if 'type' is ANY, dns_db_findext() will leave 'rdataset' and
* 'sigrdataset' disassociated and the returned node will be iterated in
* query_additional_cb().
*
* If the lookup is not successful:
*
* - 'nodep' will not be written to,
* - 'fname' may still be modified as it is passed to dns_db_findext(),
* - 'rdataset' and 'sigrdataset' will remain disassociated.
*/
static isc_result_t
query_additionalauthfind(dns_db_t *db, dns_dbversion_t *version,
const dns_name_t *name, dns_rdatatype_t type,
ns_client_t *client, dns_dbnode_t **nodep,
dns_name_t *fname, dns_rdataset_t *rdataset,
dns_rdataset_t *sigrdataset) {
dns_clientinfomethods_t cm;
dns_dbnode_t *node = NULL;
dns_clientinfo_t ci;
isc_result_t result;
dns_clientinfomethods_init(&cm, ns_client_sourceip);
dns_clientinfo_init(&ci, client, NULL);
/*
* Since we are looking for authoritative data, we do not set
* the GLUEOK flag. Glue will be looked for later, but not
* necessarily in the same database.
*/
result = dns_db_findext(db, name, version, type,
client->query.dboptions, client->inner.now,
&node, fname, &cm, &ci, rdataset, sigrdataset);
if (result != ISC_R_SUCCESS) {
if (dns_rdataset_isassociated(rdataset)) {
dns_rdataset_disassociate(rdataset);
}
if (sigrdataset != NULL &&
dns_rdataset_isassociated(sigrdataset))
{
dns_rdataset_disassociate(sigrdataset);
}
if (node != NULL) {
dns_db_detachnode(db, &node);
}
return result;
}
/*
* Do not return signatures if the zone is not fully signed.
*/
if (sigrdataset != NULL && !dns_db_issecure(db) &&
dns_rdataset_isassociated(sigrdataset))
{
dns_rdataset_disassociate(sigrdataset);
}
*nodep = node;
return ISC_R_SUCCESS;
}
/*
* For query context 'qctx', try finding authoritative additional data for
* given 'name' and 'type'. Called from query_additional_cb().
*
* If successful:
*
* - store pointers to the database and node which contain the result in
* 'dbp' and 'nodep', respectively,
*
* - store the owner name of the returned node in 'fname',
*
* - potentially bind 'rdataset' and 'sigrdataset', as explained in the
* comment for query_additionalauthfind().
*
* If unsuccessful:
*
* - 'dbp' and 'nodep' will not be written to,
* - 'fname' may still be modified as it is passed to dns_db_findext(),
* - 'rdataset' and 'sigrdataset' will remain disassociated.
*/
static isc_result_t
query_additionalauth(query_ctx_t *qctx, const dns_name_t *name,
dns_rdatatype_t type, dns_db_t **dbp, dns_dbnode_t **nodep,
dns_name_t *fname, dns_rdataset_t *rdataset,
dns_rdataset_t *sigrdataset) {
ns_client_t *client = qctx->client;
ns_dbversion_t *dbversion = NULL;
dns_dbversion_t *version = NULL;
dns_dbnode_t *node = NULL;
dns_zone_t *zone = NULL;
dns_db_t *db = NULL;
isc_result_t result;
/*
* First, look within the same zone database for authoritative
* additional data.
*/
if (!client->query.authdbset || client->query.authdb == NULL) {
return ISC_R_NOTFOUND;
}
dbversion = ns_client_findversion(client, client->query.authdb);
if (dbversion == NULL) {
return ISC_R_NOTFOUND;
}
dns_db_attach(client->query.authdb, &db);
version = dbversion->version;
CTRACE(ISC_LOG_DEBUG(3), "query_additionalauth: same zone");
result = query_additionalauthfind(db, version, name, type, client,
&node, fname, rdataset, sigrdataset);
if (result != ISC_R_SUCCESS &&
qctx->view->minimalresponses == dns_minimal_no &&
RECURSIONOK(client))
{
/*
* If we aren't doing response minimization and recursion is
* allowed, we can try and see if any other zone matches.
*/
version = NULL;
dns_db_detach(&db);
dns_getdb_options_t options = { .nolog = true };
result = query_getzonedb(client, name, type, options, &zone,
&db, &version);
if (result != ISC_R_SUCCESS) {
return result;
}
dns_zone_detach(&zone);
CTRACE(ISC_LOG_DEBUG(3), "query_additionalauth: other zone");
result = query_additionalauthfind(db, version, name, type,
client, &node, fname,
rdataset, sigrdataset);
}
if (result != ISC_R_SUCCESS) {
dns_db_detach(&db);
} else {
*nodep = node;
node = NULL;
*dbp = db;
db = NULL;
}
return result;
}
static isc_result_t
query_additional_cb(void *arg, const dns_name_t *name, dns_rdatatype_t qtype,
dns_rdataset_t *found DNS__DB_FLARG) {
query_ctx_t *qctx = arg;
ns_client_t *client = qctx->client;
isc_result_t result, eresult = ISC_R_SUCCESS;
dns_dbnode_t *node = NULL;
dns_db_t *db = NULL;
dns_name_t *fname = NULL, *mname = NULL;
dns_rdataset_t *rdataset = NULL, *sigrdataset = NULL;
dns_rdataset_t *trdataset = NULL;
isc_buffer_t *dbuf = NULL;
isc_buffer_t b;
ns_dbversion_t *dbversion = NULL;
dns_dbversion_t *version = NULL;
bool added_something = false, need_addname = false;
dns_rdatatype_t type;
dns_clientinfomethods_t cm;
dns_clientinfo_t ci;
dns_rdatasetadditional_t additionaltype =
dns_rdatasetadditional_fromauth;
REQUIRE(NS_CLIENT_VALID(client));
REQUIRE(qtype != dns_rdatatype_any);
if (!WANTDNSSEC(client) && dns_rdatatype_isdnssec(qtype)) {
return ISC_R_SUCCESS;
}
CTRACE(ISC_LOG_DEBUG(3), "query_additional_cb");
dns_clientinfomethods_init(&cm, ns_client_sourceip);
dns_clientinfo_init(&ci, client, NULL);
/*
* We treat type A additional section processing as if it
* were "any address type" additional section processing.
* To avoid multiple lookups, we do an 'any' database
* lookup and iterate over the node.
*/
if (qtype == dns_rdatatype_a) {
type = dns_rdatatype_any;
} else {
type = qtype;
}
/*
* Get some resources.
*/
dbuf = ns_client_getnamebuf(client);
fname = ns_client_newname(client, dbuf, &b);
rdataset = ns_client_newrdataset(client);
if (WANTDNSSEC(client)) {
sigrdataset = ns_client_newrdataset(client);
}
/*
* If we want only minimal responses and are here, then it must
* be for glue.
*/
if (qctx->view->minimalresponses == dns_minimal_yes &&
client->query.qtype != dns_rdatatype_ns)
{
goto try_glue;
}
/*
* First, look for authoritative additional data.
*/
result = query_additionalauth(qctx, name, type, &db, &node, fname,
rdataset, sigrdataset);
if (result == ISC_R_SUCCESS) {
goto found;
}
/*
* No authoritative data was found. The cache is our next best bet.
*/
if (!qctx->view->recursion) {
goto try_glue;
}
additionaltype = dns_rdatasetadditional_fromcache;
dns_getdb_options_t options = { .nolog = true };
result = query_getcachedb(client, name, qtype, &db, options);
if (result != ISC_R_SUCCESS) {
/*
* Most likely the client isn't allowed to query the cache.
*/
goto try_glue;
}
/*
* Attempt to validate glue.
*/
if (sigrdataset == NULL) {
sigrdataset = ns_client_newrdataset(client);
}
version = NULL;
result = dns_db_findext(db, name, version, type,
client->query.dboptions | DNS_DBFIND_GLUEOK |
DNS_DBFIND_ADDITIONALOK,
client->inner.now, &node, fname, &cm, &ci,
rdataset, sigrdataset);
dns_cache_updatestats(qctx->view->cache, result);
if (!WANTDNSSEC(client)) {
ns_client_putrdataset(client, &sigrdataset);
}
if (result == ISC_R_SUCCESS) {
goto found;
}
if (dns_rdataset_isassociated(rdataset)) {
dns_rdataset_disassociate(rdataset);
}
if (sigrdataset != NULL && dns_rdataset_isassociated(sigrdataset)) {
dns_rdataset_disassociate(sigrdataset);
}
if (node != NULL) {
dns_db_detachnode(db, &node);
}
dns_db_detach(&db);
try_glue:
/*
* No cached data was found. Glue is our last chance.
* RFC1035 sayeth:
*
* NS records cause both the usual additional section
* processing to locate a type A record, and, when used
* in a referral, a special search of the zone in which
* they reside for glue information.
*
* This is the "special search". Note that we must search
* the zone where the NS record resides, not the zone it
* points to, and that we only do the search in the delegation
* case (identified by client->query.gluedb being set).
*/
if (client->query.gluedb == NULL) {
goto cleanup;
}
/*
* Don't poison caches using the bailiwick protection model.
*/
if (!dns_name_issubdomain(name, dns_db_origin(client->query.gluedb))) {
goto cleanup;
}
dbversion = ns_client_findversion(client, client->query.gluedb);
if (dbversion == NULL) {
goto cleanup;
}
dns_db_attach(client->query.gluedb, &db);
version = dbversion->version;
additionaltype = dns_rdatasetadditional_fromglue;
result = dns_db_findext(db, name, version, type,
client->query.dboptions | DNS_DBFIND_GLUEOK,
client->inner.now, &node, fname, &cm, &ci,
rdataset, sigrdataset);
if (result != ISC_R_SUCCESS && result != DNS_R_ZONECUT &&
result != DNS_R_GLUE)
{
goto cleanup;
}
found:
/*
* We have found a potential additional data rdataset, or
* at least a node to iterate over.
*/
ns_client_keepname(client, fname, dbuf);
/*
* Does the caller want the found rdataset?
*/
if (found != NULL && dns_rdataset_isassociated(rdataset)) {
dns_rdataset_clone(rdataset, found);
}
/*
* If we have an rdataset, add it to the additional data
* section.
*/
mname = NULL;
if (dns_rdataset_isassociated(rdataset) &&
!query_isduplicate(client, fname, type, &mname))
{
if (mname != NULL) {
INSIST(mname != fname);
ns_client_releasename(client, &fname);
fname = mname;
} else {
need_addname = true;
}
ISC_LIST_APPEND(fname->list, rdataset, link);
trdataset = rdataset;
rdataset = NULL;
added_something = true;
/*
* Note: we only add SIGs if we've added the type they cover,
* so we do not need to check if the SIG rdataset is already
* in the response.
*/
if (sigrdataset != NULL &&
dns_rdataset_isassociated(sigrdataset))
{
ISC_LIST_APPEND(fname->list, sigrdataset, link);
sigrdataset = NULL;
}
}
if (qtype == dns_rdatatype_a) {
/*
* We now go looking for A and AAAA records, along with
* their signatures.
*
* XXXRTH This code could be more efficient.
*/
if (rdataset != NULL) {
if (dns_rdataset_isassociated(rdataset)) {
dns_rdataset_disassociate(rdataset);
}
} else {
rdataset = ns_client_newrdataset(client);
}
if (sigrdataset != NULL) {
if (dns_rdataset_isassociated(sigrdataset)) {
dns_rdataset_disassociate(sigrdataset);
}
} else if (WANTDNSSEC(client)) {
sigrdataset = ns_client_newrdataset(client);
}
if (query_isduplicate(client, fname, dns_rdatatype_a, NULL)) {
goto aaaa_lookup;
}
result = dns_db_findrdataset(db, node, version, dns_rdatatype_a,
0, client->inner.now, rdataset,
sigrdataset);
if (result == DNS_R_NCACHENXDOMAIN) {
goto addname;
} else if (result == DNS_R_NCACHENXRRSET) {
dns_rdataset_disassociate(rdataset);
if (sigrdataset != NULL &&
dns_rdataset_isassociated(sigrdataset))
{
dns_rdataset_disassociate(sigrdataset);
}
} else if (result == ISC_R_SUCCESS) {
bool invalid = false;
mname = NULL;
if (additionaltype ==
dns_rdatasetadditional_fromcache &&
(DNS_TRUST_PENDING(rdataset->trust) ||
DNS_TRUST_GLUE(rdataset->trust)))
{
/* validate() may change rdataset->trust */
invalid = !validate(client, db, fname, rdataset,
sigrdataset);
}
if (invalid && DNS_TRUST_PENDING(rdataset->trust)) {
dns_rdataset_disassociate(rdataset);
if (sigrdataset != NULL &&
dns_rdataset_isassociated(sigrdataset))
{
dns_rdataset_disassociate(sigrdataset);
}
} else if (!query_isduplicate(client, fname,
dns_rdatatype_a, &mname))
{
if (mname != fname) {
if (mname != NULL) {
ns_client_releasename(client,
&fname);
fname = mname;
} else {
need_addname = true;
}
}
ISC_LIST_APPEND(fname->list, rdataset, link);
added_something = true;
if (sigrdataset != NULL &&
dns_rdataset_isassociated(sigrdataset))
{
ISC_LIST_APPEND(fname->list,
sigrdataset, link);
sigrdataset =
ns_client_newrdataset(client);
}
rdataset = ns_client_newrdataset(client);
} else {
dns_rdataset_disassociate(rdataset);
if (sigrdataset != NULL &&
dns_rdataset_isassociated(sigrdataset))
{
dns_rdataset_disassociate(sigrdataset);
}
}
}
aaaa_lookup:
if (query_isduplicate(client, fname, dns_rdatatype_aaaa, NULL))
{
goto addname;
}
result = dns_db_findrdataset(
db, node, version, dns_rdatatype_aaaa, 0,
client->inner.now, rdataset, sigrdataset);
if (result == DNS_R_NCACHENXDOMAIN) {
goto addname;
} else if (result == DNS_R_NCACHENXRRSET) {
dns_rdataset_disassociate(rdataset);
if (sigrdataset != NULL &&
dns_rdataset_isassociated(sigrdataset))
{
dns_rdataset_disassociate(sigrdataset);
}
} else if (result == ISC_R_SUCCESS) {
bool invalid = false;
mname = NULL;
if (additionaltype ==
dns_rdatasetadditional_fromcache &&
(DNS_TRUST_PENDING(rdataset->trust) ||
DNS_TRUST_GLUE(rdataset->trust)))
{
/* validate() may change rdataset->trust */
invalid = !validate(client, db, fname, rdataset,
sigrdataset);
}
if (invalid && DNS_TRUST_PENDING(rdataset->trust)) {
dns_rdataset_disassociate(rdataset);
if (sigrdataset != NULL &&
dns_rdataset_isassociated(sigrdataset))
{
dns_rdataset_disassociate(sigrdataset);
}
} else if (!query_isduplicate(client, fname,
dns_rdatatype_aaaa,
&mname))
{
if (mname != fname) {
if (mname != NULL) {
ns_client_releasename(client,
&fname);
fname = mname;
} else {
need_addname = true;
}
}
ISC_LIST_APPEND(fname->list, rdataset, link);
added_something = true;
if (sigrdataset != NULL &&
dns_rdataset_isassociated(sigrdataset))
{
ISC_LIST_APPEND(fname->list,
sigrdataset, link);
sigrdataset = NULL;
}
rdataset = NULL;
}
}
}
addname:
CTRACE(ISC_LOG_DEBUG(3), "query_additional_cb: addname");
/*
* If we haven't added anything, then we're done.
*/
if (!added_something) {
goto cleanup;
}
/*
* We may have added our rdatasets to an existing name, if so, then
* need_addname will be false. Whether we used an existing name
* or a new one, we must set fname to NULL to prevent cleanup.
*/
if (need_addname) {
dns_message_addname(client->message, fname,
DNS_SECTION_ADDITIONAL);
}
/*
* In some cases, a record that has been added as additional
* data may *also* trigger the addition of additional data.
* This cannot go more than 'max-restarts' levels deep.
*/
if (trdataset != NULL && dns_rdatatype_followadditional(type)) {
if (client->inner.additionaldepth++ <
client->inner.view->max_restarts)
{
eresult = dns_rdataset_additionaldata(
trdataset, fname, query_additional_cb, qctx,
DNS_RDATASET_MAXADDITIONAL);
}
client->inner.additionaldepth--;
}
/*
* Don't release fname.
*/
fname = NULL;
cleanup:
CTRACE(ISC_LOG_DEBUG(3), "query_additional_cb: cleanup");
ns_client_putrdataset(client, &rdataset);
if (sigrdataset != NULL) {
ns_client_putrdataset(client, &sigrdataset);
}
if (fname != NULL) {
ns_client_releasename(client, &fname);
}
if (node != NULL) {
dns_db_detachnode(db, &node);
}
if (db != NULL) {
dns_db_detach(&db);
}
CTRACE(ISC_LOG_DEBUG(3), "query_additional_cb: done");
return eresult;
}
/*
* Add 'rdataset' to 'name'.
*/
static void
query_addtoname(dns_name_t *name, dns_rdataset_t *rdataset) {
ISC_LIST_APPEND(name->list, rdataset, link);
}
/*
* Set the ordering for 'rdataset'.
*/
static void
query_setorder(query_ctx_t *qctx, dns_name_t *name, dns_rdataset_t *rdataset) {
ns_client_t *client = qctx->client;
dns_order_t *order = client->inner.view->order;
CTRACE(ISC_LOG_DEBUG(3), "query_setorder");
UNUSED(client);
if (order != NULL) {
rdataset->attributes.order = dns_order_find(
order, name, rdataset->type, rdataset->rdclass);
}
}
/*
* Handle glue and fetch any other needed additional data for 'rdataset'.
*/
static void
query_additional(query_ctx_t *qctx, dns_name_t *name,
dns_rdataset_t *rdataset) {
ns_client_t *client = qctx->client;
isc_result_t result;
CTRACE(ISC_LOG_DEBUG(3), "query_additional");
if (NOADDITIONAL(client)) {
return;
}
/*
* Try to process glue directly.
*/
if (rdataset->type == dns_rdatatype_ns &&
client->query.gluedb != NULL && dns_db_iszone(client->query.gluedb))
{
ns_dbversion_t *dbversion = NULL;
dbversion = ns_client_findversion(client, client->query.gluedb);
if (dbversion == NULL) {
goto regular;
}
result = dns_db_addglue(qctx->db, dbversion->version, rdataset,
client->message);
if (result == ISC_R_SUCCESS) {
return;
}
}
regular:
/*
* Add other additional data if needed.
* We don't care if dns_rdataset_additionaldata() fails.
*/
(void)dns_rdataset_additionaldata(rdataset, name, query_additional_cb,
qctx, DNS_RDATASET_MAXADDITIONAL);
CTRACE(ISC_LOG_DEBUG(3), "query_additional: done");
}
static void
query_addrrset(query_ctx_t *qctx, dns_name_t **namep,
dns_rdataset_t **rdatasetp, dns_rdataset_t **sigrdatasetp,
isc_buffer_t *dbuf, dns_section_t section) {
isc_result_t result;
ns_client_t *client = qctx->client;
dns_name_t *name = *namep, *mname = NULL;
dns_rdataset_t *rdataset = *rdatasetp, *mrdataset = NULL;
dns_rdataset_t *sigrdataset = NULL;
CTRACE(ISC_LOG_DEBUG(3), "query_addrrset");
REQUIRE(name != NULL);
if (sigrdatasetp != NULL) {
sigrdataset = *sigrdatasetp;
}
/*%
* To the current response for 'client', add the answer RRset
* '*rdatasetp' and an optional signature set '*sigrdatasetp', with
* owner name '*namep', to section 'section', unless they are
* already there. Also add any pertinent additional data.
*
* If 'dbuf' is not NULL, then '*namep' is the name whose data is
* stored in 'dbuf'. In this case, query_addrrset() guarantees that
* when it returns the name will either have been kept or released.
*/
result = dns_message_findname(client->message, section, name,
rdataset->type, rdataset->covers, &mname,
&mrdataset);
if (result == ISC_R_SUCCESS) {
/*
* We've already got an RRset of the given name and type.
*/
CTRACE(ISC_LOG_DEBUG(3), "query_addrrset: dns_message_findname "
"succeeded: done");
if (dbuf != NULL) {
ns_client_releasename(client, namep);
}
if (rdataset->attributes.required) {
mrdataset->attributes.required = true;
}
if (rdataset->attributes.stale_added) {
mrdataset->attributes.stale_added = true;
}
return;
} else if (result == DNS_R_NXDOMAIN) {
/*
* The name doesn't exist.
*/
if (dbuf != NULL) {
ns_client_keepname(client, name, dbuf);
}
dns_message_addname(client->message, name, section);
*namep = NULL;
mname = name;
} else {
RUNTIME_CHECK(result == DNS_R_NXRRSET);
if (dbuf != NULL) {
ns_client_releasename(client, namep);
}
}
if (rdataset->trust != dns_trust_secure &&
(section == DNS_SECTION_ANSWER || section == DNS_SECTION_AUTHORITY))
{
client->query.attributes &= ~NS_QUERYATTR_SECURE;
}
/*
* Update message name, set rdataset order, and do additional
* section processing if needed.
*/
query_addtoname(mname, rdataset);
query_setorder(qctx, mname, rdataset);
query_additional(qctx, mname, rdataset);
/*
* Note: we only add SIGs if we've added the type they cover, so
* we do not need to check if the SIG rdataset is already in the
* response.
*/
*rdatasetp = NULL;
if (sigrdataset != NULL && dns_rdataset_isassociated(sigrdataset)) {
/*
* We have a signature. Add it to the response.
*/
ISC_LIST_APPEND(mname->list, sigrdataset, link);
*sigrdatasetp = NULL;
}
CTRACE(ISC_LOG_DEBUG(3), "query_addrrset: done");
}
/*
* Mark the RRsets as secure. Update the cache (db) to reflect the
* change in trust level.
*/
static void
mark_secure(ns_client_t *client, dns_db_t *db, dns_name_t *name,
dns_rdata_rrsig_t *rrsig, dns_rdataset_t *rdataset,
dns_rdataset_t *sigrdataset) {
isc_result_t result;
dns_dbnode_t *node = NULL;
dns_clientinfomethods_t cm;
dns_clientinfo_t ci;
isc_stdtime_t now;
rdataset->trust = dns_trust_secure;
sigrdataset->trust = dns_trust_secure;
dns_clientinfomethods_init(&cm, ns_client_sourceip);
dns_clientinfo_init(&ci, client, NULL);
/*
* Save the updated secure state. Ignore failures.
*/
result = dns_db_findnodeext(db, name, true, &cm, &ci, &node);
if (result != ISC_R_SUCCESS) {
return;
}
now = isc_stdtime_now();
dns_rdataset_trimttl(rdataset, sigrdataset, rrsig, now,
client->inner.view->acceptexpired);
(void)dns_db_addrdataset(db, node, NULL, client->inner.now, rdataset, 0,
NULL);
(void)dns_db_addrdataset(db, node, NULL, client->inner.now, sigrdataset,
0, NULL);
dns_db_detachnode(db, &node);
}
/*
* Find the secure key that corresponds to rrsig.
* Note: 'keyrdataset' maintains state between successive calls,
* there may be multiple keys with the same keyid.
* Return false if we have exhausted all the possible keys.
*/
static bool
get_key(ns_client_t *client, dns_db_t *db, dns_rdata_rrsig_t *rrsig,
dns_rdataset_t *keyrdataset, dst_key_t **keyp) {
isc_result_t result;
dns_dbnode_t *node = NULL;
bool secure = false;
dns_clientinfomethods_t cm;
dns_clientinfo_t ci;
dst_algorithm_t sigalg, keyalg;
dns_clientinfomethods_init(&cm, ns_client_sourceip);
dns_clientinfo_init(&ci, client, NULL);
if (!dns_rdataset_isassociated(keyrdataset)) {
result = dns_db_findnodeext(db, &rrsig->signer, false, &cm, &ci,
&node);
if (result != ISC_R_SUCCESS) {
return false;
}
result = dns_db_findrdataset(
db, node, NULL, dns_rdatatype_dnskey, 0,
client->inner.now, keyrdataset, NULL);
dns_db_detachnode(db, &node);
if (result != ISC_R_SUCCESS) {
return false;
}
if (keyrdataset->trust != dns_trust_secure) {
return false;
}
result = dns_rdataset_first(keyrdataset);
} else {
result = dns_rdataset_next(keyrdataset);
}
sigalg = dst_algorithm_fromdata(rrsig->algorithm, rrsig->signature,
rrsig->siglen);
for (; result == ISC_R_SUCCESS; result = dns_rdataset_next(keyrdataset))
{
dns_rdata_t rdata = DNS_RDATA_INIT;
dns_rdata_dnskey_t key;
isc_region_t r;
dns_rdataset_current(keyrdataset, &rdata);
dns_rdata_tostruct(&rdata, &key, NULL); /* can't fail */
keyalg = dst_algorithm_fromdata(key.algorithm, key.data,
key.datalen);
if (sigalg != keyalg || !dns_dnssec_iszonekey(&key)) {
continue;
}
dns_rdata_toregion(&rdata, &r);
if (dst_region_computeid(&r) != rrsig->keyid) {
continue;
}
result = dns_dnssec_keyfromrdata(&rrsig->signer, &rdata,
client->manager->mctx, keyp);
if (result == ISC_R_SUCCESS) {
secure = true;
break;
}
}
return secure;
}
static bool
verify(dst_key_t *key, dns_name_t *name, dns_rdataset_t *rdataset,
dns_rdata_t *rdata, ns_client_t *client) {
isc_result_t result;
dns_fixedname_t fixed;
bool ignore = false;
dns_fixedname_init(&fixed);
again:
result = dns_dnssec_verify(name, rdataset, key, ignore,
client->manager->mctx, rdata, NULL);
if (result == DNS_R_SIGEXPIRED && client->inner.view->acceptexpired) {
ignore = true;
goto again;
}
if (result == ISC_R_SUCCESS || result == DNS_R_FROMWILDCARD) {
return true;
}
return false;
}
/*
* Validate the rdataset if possible with available records.
*/
static bool
validate(ns_client_t *client, dns_db_t *db, dns_name_t *name,
dns_rdataset_t *rdataset, dns_rdataset_t *sigrdataset) {
isc_result_t result;
dns_rdata_rrsig_t rrsig;
dst_key_t *key = NULL;
dns_rdataset_t keyrdataset;
if (sigrdataset == NULL || !dns_rdataset_isassociated(sigrdataset)) {
return false;
}
DNS_RDATASET_FOREACH (sigrdataset) {
dns_rdata_t rdata = DNS_RDATA_INIT;
dns_rdataset_current(sigrdataset, &rdata);
result = dns_rdata_tostruct(&rdata, &rrsig, NULL);
RUNTIME_CHECK(result == ISC_R_SUCCESS);
if (!dns_resolver_algorithm_supported(
client->inner.view->resolver, name, rrsig.algorithm,
rrsig.signature, rrsig.siglen))
{
char txt[DNS_NAME_FORMATSIZE + 32];
isc_buffer_t buffer;
dst_algorithm_t alg;
alg = dst_algorithm_fromdata(
rrsig.algorithm, rrsig.signature, rrsig.siglen);
isc_buffer_init(&buffer, txt, sizeof(txt));
dst_algorithm_totext(alg, &buffer);
isc_buffer_putstr(&buffer, " ");
dns_name_totext(name, DNS_NAME_OMITFINALDOT, &buffer);
isc_buffer_putstr(&buffer, " (cached)");
isc_buffer_putuint8(&buffer, 0);
dns_ede_add(&client->edectx, DNS_EDE_DNSKEYALG,
isc_buffer_base(&buffer));
continue;
}
if (!dns_name_issubdomain(name, &rrsig.signer)) {
continue;
}
dns_rdataset_init(&keyrdataset);
do {
if (!get_key(client, db, &rrsig, &keyrdataset, &key)) {
break;
}
if (verify(key, name, rdataset, &rdata, client)) {
dst_key_free(&key);
dns_rdataset_disassociate(&keyrdataset);
mark_secure(client, db, name, &rrsig, rdataset,
sigrdataset);
return true;
}
dst_key_free(&key);
} while (1);
if (dns_rdataset_isassociated(&keyrdataset)) {
dns_rdataset_disassociate(&keyrdataset);
}
}
return false;
}
static void
fixrdataset(ns_client_t *client, dns_rdataset_t **rdataset) {
if (*rdataset == NULL) {
*rdataset = ns_client_newrdataset(client);
} else if (dns_rdataset_isassociated(*rdataset)) {
dns_rdataset_disassociate(*rdataset);
}
}
static void
fixfname(ns_client_t *client, dns_name_t **fname, isc_buffer_t **dbuf,
isc_buffer_t *nbuf) {
if (*fname == NULL) {
*dbuf = ns_client_getnamebuf(client);
*fname = ns_client_newname(client, *dbuf, nbuf);
}
}
static void
free_fresp(ns_client_t *client, dns_fetchresponse_t **frespp) {
dns_fetchresponse_t *fresp = *frespp;
CTRACE(ISC_LOG_DEBUG(3), "free_fresp");
if (fresp->fetch != NULL) {
dns_resolver_destroyfetch(&fresp->fetch);
}
if (fresp->node != NULL) {
dns_db_detachnode(fresp->db, &fresp->node);
}
if (fresp->db != NULL) {
dns_db_detach(&fresp->db);
}
if (fresp->rdataset != NULL) {
ns_client_putrdataset(client, &fresp->rdataset);
}
if (fresp->sigrdataset != NULL) {
ns_client_putrdataset(client, &fresp->sigrdataset);
}
dns_resolver_freefresp(frespp);
}
static isc_result_t
recursionquotatype_attach(ns_client_t *client, bool soft_limit) {
isc_statscounter_t recurscount;
isc_result_t result;
result = isc_quota_acquire(&client->manager->sctx->recursionquota);
switch (result) {
case ISC_R_SUCCESS:
break;
case ISC_R_SOFTQUOTA:
if (soft_limit) {
/*
* Exceeding soft quota was allowed, so continue as if
* 'result' was ISC_R_SUCCESS while retaining the
* original result code.
*/
break;
}
isc_quota_release(&client->manager->sctx->recursionquota);
FALLTHROUGH;
default:
return result;
}
recurscount = ns_stats_increment(client->manager->sctx->nsstats,
ns_statscounter_recursclients);
ns_stats_update_if_greater(client->manager->sctx->nsstats,
ns_statscounter_recurshighwater,
recurscount + 1);
return result;
}
static isc_result_t
recursionquotatype_attach_hard(ns_client_t *client) {
return recursionquotatype_attach(client, false);
}
static isc_result_t
recursionquotatype_attach_soft(ns_client_t *client) {
return recursionquotatype_attach(client, true);
}
static void
recursionquotatype_detach(ns_client_t *client) {
isc_quota_release(&client->manager->sctx->recursionquota);
ns_stats_decrement(client->manager->sctx->nsstats,
ns_statscounter_recursclients);
}
static void
stale_refresh_aftermath(ns_client_t *client, isc_result_t result) {
dns_db_t *db = NULL;
unsigned int dboptions;
isc_buffer_t buffer;
query_ctx_t qctx;
dns_clientinfomethods_t cm;
dns_clientinfo_t ci;
char namebuf[DNS_NAME_FORMATSIZE];
char typebuf[DNS_RDATATYPE_FORMATSIZE];
/*
* If refreshing a stale RRset failed, we need to set the
* stale-refresh-time window, so that on future requests for this
* RRset the stale entry may be used immediately.
*/
switch (result) {
case ISC_R_SUCCESS:
case DNS_R_GLUE:
case DNS_R_ZONECUT:
case ISC_R_NOTFOUND:
case DNS_R_DELEGATION:
case DNS_R_EMPTYNAME:
case DNS_R_NXRRSET:
case DNS_R_EMPTYWILD:
case DNS_R_NXDOMAIN:
case DNS_R_COVERINGNSEC:
case DNS_R_NCACHENXDOMAIN:
case DNS_R_NCACHENXRRSET:
case DNS_R_CNAME:
case DNS_R_DNAME:
break;
default:
dns_name_format(client->query.qname, namebuf, sizeof(namebuf));
dns_rdatatype_format(client->query.qtype, typebuf,
sizeof(typebuf));
ns_client_log(client, NS_LOGCATEGORY_SERVE_STALE,
NS_LOGMODULE_QUERY, ISC_LOG_NOTICE,
"%s/%s stale refresh failed: timed out", namebuf,
typebuf);
/*
* Set up a short lived query context, solely to set the
* last refresh failure time on the RRset in the cache
* database, starting the stale-refresh-time window for it.
* This is a condensed form of query_lookup().
*/
client->inner.now = isc_stdtime_now();
client->query.attributes &= ~NS_QUERYATTR_RECURSIONOK;
qctx_init(client, NULL, 0, &qctx);
dns_clientinfomethods_init(&cm, ns_client_sourceip);
dns_clientinfo_init(&ci, qctx.client, NULL);
if (HAVEECS(qctx.client)) {
dns_clientinfo_setecs(&ci, &qctx.client->inner.ecs);
}
result = qctx_prepare_buffers(&qctx, &buffer);
if (result != ISC_R_SUCCESS) {
goto cleanup;
}
dboptions = qctx.client->query.dboptions;
dboptions |= DNS_DBFIND_STALEOK;
dboptions |= DNS_DBFIND_STALESTART;
dns_db_attach(qctx.client->inner.view->cachedb, &db);
(void)dns_db_findext(db, qctx.client->query.qname, NULL,
qctx.client->query.qtype, dboptions,
qctx.client->inner.now, &qctx.node,
qctx.fname, &cm, &ci, qctx.rdataset,
qctx.sigrdataset);
if (qctx.node != NULL) {
dns_db_detachnode(db, &qctx.node);
}
dns_db_detach(&db);
cleanup:
qctx_freedata(&qctx);
qctx_destroy(&qctx);
}
}
static void
cleanup_after_fetch(dns_fetchresponse_t *resp, const char *ctracestr,
ns_query_rectype_t recursion_type) {
ns_client_t *client = resp->arg;
isc_nmhandle_t **handlep = NULL;
dns_fetch_t **fetchp = NULL;
isc_result_t result;
REQUIRE(NS_CLIENT_VALID(client));
CTRACE(ISC_LOG_DEBUG(3), ctracestr);
handlep = &client->query.recursions[recursion_type].handle;
fetchp = &client->query.recursions[recursion_type].fetch;
result = resp->result;
LOCK(&client->query.fetchlock);
if (*fetchp != NULL) {
INSIST(resp->fetch == *fetchp);
*fetchp = NULL;
}
UNLOCK(&client->query.fetchlock);
/* Some type of recursions require a bit of aftermath. */
if (recursion_type == RECTYPE_STALE_REFRESH) {
stale_refresh_aftermath(client, result);
}
recursionquotatype_detach(client);
free_fresp(client, &resp);
isc_nmhandle_detach(handlep);
}
static void
prefetch_done(void *arg) {
cleanup_after_fetch(arg, "prefetch_done", RECTYPE_PREFETCH);
}
static void
rpzfetch_done(void *arg) {
cleanup_after_fetch(arg, "rpzfetch_done", RECTYPE_RPZ);
}
static void
stale_refresh_done(void *arg) {
cleanup_after_fetch(arg, "stale_refresh_done", RECTYPE_STALE_REFRESH);
}
/*
* Try initiating a fetch for the given 'qname' and 'qtype' (using the slot in
* the 'recursions' array indicated by 'recursion_type') that will be
* associated with 'client'. If the recursive clients quota (or even soft
* quota) is reached or some other error occurs, just return without starting
* the fetch. If a fetch is successfully created, its results will be cached
* upon successful completion, but no further actions will be taken afterwards.
*/
static void
fetch_and_forget(ns_client_t *client, dns_name_t *qname, dns_rdatatype_t qtype,
ns_query_rectype_t recursion_type) {
dns_rdataset_t *tmprdataset;
isc_sockaddr_t *peeraddr;
unsigned int options;
isc_job_cb cb;
isc_nmhandle_t **handlep;
dns_fetch_t **fetchp;
isc_result_t result;
result = recursionquotatype_attach_hard(client);
if (result != ISC_R_SUCCESS) {
return;
}
tmprdataset = ns_client_newrdataset(client);
if (!TCP(client)) {
peeraddr = &client->inner.peeraddr;
} else {
peeraddr = NULL;
}
switch (recursion_type) {
case RECTYPE_PREFETCH:
options = client->query.fetchoptions | DNS_FETCHOPT_PREFETCH;
cb = prefetch_done;
break;
case RECTYPE_RPZ:
options = client->query.fetchoptions;
cb = rpzfetch_done;
break;
case RECTYPE_STALE_REFRESH:
options = client->query.fetchoptions;
cb = stale_refresh_done;
break;
default:
UNREACHABLE();
}
handlep = &client->query.recursions[recursion_type].handle;
fetchp = &client->query.recursions[recursion_type].fetch;
isc_nmhandle_attach(client->inner.handle, handlep);
result = dns_resolver_createfetch(
client->inner.view->resolver, qname, qtype, NULL, NULL, NULL,
peeraddr, client->message->id, options, 0, NULL,
client->query.qc, client->manager->loop, cb, client, NULL,
tmprdataset, NULL, fetchp);
if (result != ISC_R_SUCCESS) {
ns_client_putrdataset(client, &tmprdataset);
isc_nmhandle_detach(handlep);
recursionquotatype_detach(client);
}
}
static void
query_prefetch(ns_client_t *client, dns_name_t *qname,
dns_rdataset_t *rdataset) {
CTRACE(ISC_LOG_DEBUG(3), "query_prefetch");
if (FETCH_RECTYPE_PREFETCH(client) != NULL ||
client->inner.view->prefetch_trigger == 0U ||
rdataset->ttl > client->inner.view->prefetch_trigger ||
!rdataset->attributes.prefetch)
{
return;
}
fetch_and_forget(client, qname, rdataset->type, RECTYPE_PREFETCH);
dns_rdataset_clearprefetch(rdataset);
ns_stats_increment(client->manager->sctx->nsstats,
ns_statscounter_prefetch);
}
static void
query_stale_refresh(ns_client_t *client) {
dns_name_t *qname;
CTRACE(ISC_LOG_DEBUG(3), "query_stale_refresh");
if (FETCH_RECTYPE_STALE_REFRESH(client) != NULL) {
return;
}
client->query.dboptions &= ~(DNS_DBFIND_STALETIMEOUT |
DNS_DBFIND_STALEOK |
DNS_DBFIND_STALEENABLED);
if (client->query.origqname != NULL) {
qname = client->query.origqname;
} else {
qname = client->query.qname;
}
fetch_and_forget(client, qname, client->query.qtype,
RECTYPE_STALE_REFRESH);
}
static void
rpz_clean(dns_zone_t **zonep, dns_db_t **dbp, dns_dbnode_t **nodep,
dns_rdataset_t **rdatasetp) {
if (nodep != NULL && *nodep != NULL) {
REQUIRE(dbp != NULL && *dbp != NULL);
dns_db_detachnode(*dbp, nodep);
}
if (dbp != NULL && *dbp != NULL) {
dns_db_detach(dbp);
}
if (zonep != NULL && *zonep != NULL) {
dns_zone_detach(zonep);
}
if (rdatasetp != NULL && *rdatasetp != NULL &&
dns_rdataset_isassociated(*rdatasetp))
{
dns_rdataset_disassociate(*rdatasetp);
}
}
static void
rpz_match_clear(dns_rpz_st_t *st) {
rpz_clean(&st->m.zone, &st->m.db, &st->m.node, &st->m.rdataset);
st->m.version = NULL;
}
static isc_result_t
rpz_ready(ns_client_t *client, dns_rdataset_t **rdatasetp) {
REQUIRE(rdatasetp != NULL);
CTRACE(ISC_LOG_DEBUG(3), "rpz_ready");
if (*rdatasetp == NULL) {
*rdatasetp = ns_client_newrdataset(client);
} else if (dns_rdataset_isassociated(*rdatasetp)) {
dns_rdataset_disassociate(*rdatasetp);
}
return ISC_R_SUCCESS;
}
static void
rpz_st_clear(ns_client_t *client) {
dns_rpz_st_t *st = client->query.rpz_st;
CTRACE(ISC_LOG_DEBUG(3), "rpz_st_clear");
if (st->m.rdataset != NULL) {
ns_client_putrdataset(client, &st->m.rdataset);
}
rpz_match_clear(st);
rpz_clean(NULL, &st->r.db, NULL, NULL);
if (st->r.ns_rdataset != NULL) {
ns_client_putrdataset(client, &st->r.ns_rdataset);
}
if (st->r.r_rdataset != NULL) {
ns_client_putrdataset(client, &st->r.r_rdataset);
}
rpz_clean(&st->q.zone, &st->q.db, &st->q.node, NULL);
if (st->q.rdataset != NULL) {
ns_client_putrdataset(client, &st->q.rdataset);
}
if (st->q.sigrdataset != NULL) {
ns_client_putrdataset(client, &st->q.sigrdataset);
}
st->state = 0;
st->m.type = DNS_RPZ_TYPE_BAD;
st->m.policy = DNS_RPZ_POLICY_MISS;
}
static dns_rpz_zbits_t
rpz_get_zbits(ns_client_t *client, dns_rdatatype_t ip_type,
dns_rpz_type_t rpz_type) {
dns_rpz_st_t *st;
dns_rpz_zbits_t zbits = 0;
REQUIRE(client != NULL);
REQUIRE(client->query.rpz_st != NULL);
st = client->query.rpz_st;
switch (rpz_type) {
case DNS_RPZ_TYPE_CLIENT_IP:
zbits = st->have.client_ip;
break;
case DNS_RPZ_TYPE_QNAME:
zbits = st->have.qname;
break;
case DNS_RPZ_TYPE_IP:
if (ip_type == dns_rdatatype_a) {
zbits = st->have.ipv4;
} else if (ip_type == dns_rdatatype_aaaa) {
zbits = st->have.ipv6;
} else {
zbits = st->have.ip;
}
break;
case DNS_RPZ_TYPE_NSDNAME:
zbits = st->have.nsdname;
break;
case DNS_RPZ_TYPE_NSIP:
if (ip_type == dns_rdatatype_a) {
zbits = st->have.nsipv4;
} else if (ip_type == dns_rdatatype_aaaa) {
zbits = st->have.nsipv6;
} else {
zbits = st->have.nsip;
}
break;
default:
UNREACHABLE();
}
/*
* Choose
* the earliest configured policy zone (rpz->num)
* QNAME over IP over NSDNAME over NSIP (rpz_type)
* the smallest name,
* the longest IP address prefix,
* the lexically smallest address.
*/
if (st->m.policy != DNS_RPZ_POLICY_MISS) {
if (st->m.type >= rpz_type) {
zbits &= DNS_RPZ_ZMASK(st->m.rpz->num);
} else {
zbits &= DNS_RPZ_ZMASK(st->m.rpz->num) >> 1;
}
}
/*
* If the client wants recursion, allow only compatible policies.
*/
if (!RECURSIONOK(client)) {
zbits &= st->popt.no_rd_ok;
}
return zbits;
}
static void
query_rpzfetch(ns_client_t *client, dns_name_t *qname, dns_rdatatype_t type) {
CTRACE(ISC_LOG_DEBUG(3), "query_rpzfetch");
if (FETCH_RECTYPE_RPZ(client) != NULL) {
return;
}
fetch_and_forget(client, qname, type, RECTYPE_RPZ);
}
/*
* Get an NS, A, or AAAA rrset related to the response for the client
* to check the contents of that rrset for hits by eligible policy zones.
*/
static isc_result_t
rpz_rrset_find(ns_client_t *client, dns_name_t *name, dns_rdatatype_t type,
unsigned int options, dns_rpz_type_t rpz_type, dns_db_t **dbp,
dns_dbversion_t *version, dns_rdataset_t **rdatasetp,
bool resuming) {
dns_rpz_st_t *st;
bool is_zone;
dns_dbnode_t *node;
dns_fixedname_t fixed;
dns_name_t *found;
isc_result_t result;
dns_clientinfomethods_t cm;
dns_clientinfo_t ci;
CTRACE(ISC_LOG_DEBUG(3), "rpz_rrset_find");
st = client->query.rpz_st;
if ((st->state & DNS_RPZ_RECURSING) != 0) {
INSIST(st->r.r_type == type);
INSIST(dns_name_equal(name, st->r_name));
INSIST(*rdatasetp == NULL ||
!dns_rdataset_isassociated(*rdatasetp));
st->state &= ~DNS_RPZ_RECURSING;
RESTORE(*dbp, st->r.db);
if (*rdatasetp != NULL) {
ns_client_putrdataset(client, rdatasetp);
}
RESTORE(*rdatasetp, st->r.r_rdataset);
result = st->r.r_result;
if (result == DNS_R_DELEGATION) {
CTRACE(ISC_LOG_ERROR, "RPZ recursing");
rpz_log_fail(client, DNS_RPZ_ERROR_LEVEL, name,
rpz_type, "rpz_rrset_find(1)", result);
st->m.policy = DNS_RPZ_POLICY_ERROR;
result = DNS_R_SERVFAIL;
}
return result;
}
result = rpz_ready(client, rdatasetp);
if (result != ISC_R_SUCCESS) {
st->m.policy = DNS_RPZ_POLICY_ERROR;
return result;
}
if (*dbp != NULL) {
is_zone = false;
} else {
dns_zone_t *zone;
version = NULL;
zone = NULL;
result = query_getdb(client, name, type,
(dns_getdb_options_t){ 0 }, &zone, dbp,
&version, &is_zone);
if (result != ISC_R_SUCCESS) {
rpz_log_fail(client, DNS_RPZ_ERROR_LEVEL, name,
rpz_type, "rpz_rrset_find(2)", result);
st->m.policy = DNS_RPZ_POLICY_ERROR;
if (zone != NULL) {
dns_zone_detach(&zone);
}
return result;
}
if (zone != NULL) {
dns_zone_detach(&zone);
}
}
node = NULL;
found = dns_fixedname_initname(&fixed);
dns_clientinfomethods_init(&cm, ns_client_sourceip);
dns_clientinfo_init(&ci, client, NULL);
result = dns_db_findext(*dbp, name, version, type, options,
client->inner.now, &node, found, &cm, &ci,
*rdatasetp, NULL);
if (result == DNS_R_DELEGATION && is_zone && USECACHE(client)) {
/*
* Try the cache if we're authoritative for an
* ancestor but not the domain itself.
*/
rpz_clean(NULL, dbp, &node, rdatasetp);
version = NULL;
dns_db_attach(client->inner.view->cachedb, dbp);
result = dns_db_findext(*dbp, name, version, type, 0,
client->inner.now, &node, found, &cm,
&ci, *rdatasetp, NULL);
}
rpz_clean(NULL, dbp, &node, NULL);
if (result == DNS_R_DELEGATION) {
rpz_clean(NULL, NULL, NULL, rdatasetp);
/*
* Recurse for NS rrset or A or AAAA rrset for an NS.
* Do not recurse for addresses for the query name.
*/
if (rpz_type == DNS_RPZ_TYPE_IP) {
result = DNS_R_NXRRSET;
} else if (!client->inner.view->rpzs->p.nsip_wait_recurse ||
(!client->inner.view->rpzs->p.nsdname_wait_recurse &&
rpz_type == DNS_RPZ_TYPE_NSDNAME))
{
query_rpzfetch(client, name, type);
result = DNS_R_NXRRSET;
} else {
dns_name_copy(name, st->r_name);
result = ns_query_recurse(client, type, st->r_name,
NULL, NULL, resuming);
if (result == ISC_R_SUCCESS) {
st->state |= DNS_RPZ_RECURSING;
result = DNS_R_DELEGATION;
}
}
}
return result;
}
/*
* Compute a policy owner name, p_name, in a policy zone given the needed
* policy type and the trigger name.
*/
static isc_result_t
rpz_get_p_name(ns_client_t *client, dns_name_t *p_name, dns_rpz_zone_t *rpz,
dns_rpz_type_t rpz_type, dns_name_t *trig_name) {
dns_name_t prefix, *suffix;
unsigned int first, labels;
isc_result_t result;
CTRACE(ISC_LOG_DEBUG(3), "rpz_get_p_name");
/*
* The policy owner name consists of a suffix depending on the type
* and policy zone and a prefix that is the longest possible string
* from the trigger name that keesp the resulting policy owner name
* from being too long.
*/
switch (rpz_type) {
case DNS_RPZ_TYPE_CLIENT_IP:
suffix = &rpz->client_ip;
break;
case DNS_RPZ_TYPE_QNAME:
suffix = &rpz->origin;
break;
case DNS_RPZ_TYPE_IP:
suffix = &rpz->ip;
break;
case DNS_RPZ_TYPE_NSDNAME:
suffix = &rpz->nsdname;
break;
case DNS_RPZ_TYPE_NSIP:
suffix = &rpz->nsip;
break;
default:
UNREACHABLE();
}
/*
* Start with relative version of the full trigger name,
* and trim enough allow the addition of the suffix.
*/
dns_name_init(&prefix);
labels = dns_name_countlabels(trig_name);
first = 0;
for (;;) {
dns_name_getlabelsequence(trig_name, first, labels - first - 1,
&prefix);
result = dns_name_concatenate(&prefix, suffix, p_name);
if (result == ISC_R_SUCCESS) {
break;
}
INSIST(result == DNS_R_NAMETOOLONG);
/*
* Trim the trigger name until the combination is not too long.
*/
if (labels - first < 2) {
rpz_log_fail(client, DNS_RPZ_ERROR_LEVEL, suffix,
rpz_type, "concatenate()", result);
return ISC_R_FAILURE;
}
/*
* Complain once about trimming the trigger name.
*/
if (first == 0) {
rpz_log_fail(client, DNS_RPZ_DEBUG_LEVEL1, suffix,
rpz_type, "concatenate()", result);
}
++first;
}
return ISC_R_SUCCESS;
}
/*
* Look in policy zone rpz for a policy of rpz_type by p_name.
* The self-name (usually the client qname or an NS name) is compared with
* the target of a CNAME policy for the old style passthru encoding.
* If found, the policy is recorded in *zonep, *dbp, *versionp, *nodep,
* *rdatasetp, and *policyp.
* The target DNS type, qtype, chooses the best rdataset for *rdatasetp.
* The caller must decide if the found policy is most suitable, including
* better than a previously found policy.
* If it is best, the caller records it in client->query.rpz_st->m.
*/
static isc_result_t
rpz_find_p(ns_client_t *client, dns_name_t *self_name, dns_rdatatype_t qtype,
dns_name_t *p_name, dns_rpz_zone_t *rpz, dns_rpz_type_t rpz_type,
dns_zone_t **zonep, dns_db_t **dbp, dns_dbversion_t **versionp,
dns_dbnode_t **nodep, dns_rdataset_t **rdatasetp,
dns_rpz_policy_t *policyp) {
dns_fixedname_t foundf;
dns_name_t *found;
isc_result_t result;
dns_clientinfomethods_t cm;
dns_clientinfo_t ci;
bool found_a = false;
REQUIRE(nodep != NULL);
CTRACE(ISC_LOG_DEBUG(3), "rpz_find_p");
dns_clientinfomethods_init(&cm, ns_client_sourceip);
dns_clientinfo_init(&ci, client, NULL);
/*
* Try to find either a CNAME or the type of record demanded by the
* request from the policy zone.
*/
rpz_clean(zonep, dbp, nodep, rdatasetp);
result = rpz_ready(client, rdatasetp);
if (result != ISC_R_SUCCESS) {
CTRACE(ISC_LOG_ERROR, "rpz_ready() failed");
return DNS_R_SERVFAIL;
}
*versionp = NULL;
result = rpz_getdb(client, p_name, rpz_type, zonep, dbp, versionp);
if (result != ISC_R_SUCCESS) {
return DNS_R_NXDOMAIN;
}
found = dns_fixedname_initname(&foundf);
result = dns_db_findext(*dbp, p_name, *versionp, dns_rdatatype_any, 0,
client->inner.now, nodep, found, &cm, &ci,
*rdatasetp, NULL);
/*
* Choose the best rdataset if we found something.
*/
if (result == ISC_R_SUCCESS) {
dns_rdatasetiter_t *rdsiter = NULL;
bool match = false;
result = dns_db_allrdatasets(*dbp, *nodep, *versionp, 0, 0,
&rdsiter);
if (result != ISC_R_SUCCESS) {
rpz_log_fail(client, DNS_RPZ_ERROR_LEVEL, p_name,
rpz_type, "allrdatasets()", result);
CTRACE(ISC_LOG_ERROR,
"rpz_find_p: allrdatasets failed");
return DNS_R_SERVFAIL;
}
if (qtype == dns_rdatatype_aaaa &&
!ISC_LIST_EMPTY(client->inner.view->dns64))
{
DNS_RDATASETITER_FOREACH (rdsiter) {
dns_rdatasetiter_current(rdsiter, *rdatasetp);
if ((*rdatasetp)->type == dns_rdatatype_a) {
found_a = true;
}
dns_rdataset_disassociate(*rdatasetp);
}
}
DNS_RDATASETITER_FOREACH (rdsiter) {
dns_rdatasetiter_current(rdsiter, *rdatasetp);
if ((*rdatasetp)->type == dns_rdatatype_cname ||
(*rdatasetp)->type == qtype)
{
match = true;
break;
}
dns_rdataset_disassociate(*rdatasetp);
}
dns_rdatasetiter_destroy(&rdsiter);
if (!match) {
/*
* Ask again to get the right DNS_R_DNAME/NXRRSET/...
* result if there is neither a CNAME nor target type.
*/
if (dns_rdataset_isassociated(*rdatasetp)) {
dns_rdataset_disassociate(*rdatasetp);
}
dns_db_detachnode(*dbp, nodep);
if (dns_rdatatype_issig(qtype)) {
result = DNS_R_NXRRSET;
} else {
result = dns_db_findext(
*dbp, p_name, *versionp, qtype, 0,
client->inner.now, nodep, found, &cm,
&ci, *rdatasetp, NULL);
}
}
}
switch (result) {
case ISC_R_SUCCESS:
if ((*rdatasetp)->type != dns_rdatatype_cname) {
*policyp = DNS_RPZ_POLICY_RECORD;
} else {
*policyp = dns_rpz_decode_cname(rpz, *rdatasetp,
self_name);
if ((*policyp == DNS_RPZ_POLICY_RECORD ||
*policyp == DNS_RPZ_POLICY_WILDCNAME) &&
qtype != dns_rdatatype_cname &&
qtype != dns_rdatatype_any)
{
return DNS_R_CNAME;
}
}
return ISC_R_SUCCESS;
case DNS_R_NXRRSET:
if (found_a) {
*policyp = DNS_RPZ_POLICY_DNS64;
} else {
*policyp = DNS_RPZ_POLICY_NODATA;
}
return result;
case DNS_R_DNAME:
/*
* DNAME policy RRs have very few if any uses that are not
* better served with simple wildcards. Making them work would
* require complications to get the number of labels matched
* in the name or the found name to the main DNS_R_DNAME case
* in query_dname(). The domain also does not appear in the
* summary database at the right level, so this happens only
* with a single policy zone when we have no summary database.
* Treat it as a miss.
*/
case DNS_R_NXDOMAIN:
case DNS_R_EMPTYNAME:
return DNS_R_NXDOMAIN;
default:
rpz_log_fail(client, DNS_RPZ_ERROR_LEVEL, p_name, rpz_type, "",
result);
CTRACE(ISC_LOG_ERROR, "rpz_find_p: unexpected result");
return DNS_R_SERVFAIL;
}
}
static void
rpz_save_p(dns_rpz_st_t *st, dns_rpz_zone_t *rpz, dns_rpz_type_t rpz_type,
dns_rpz_policy_t policy, dns_name_t *p_name, dns_rpz_prefix_t prefix,
isc_result_t result, dns_zone_t **zonep, dns_db_t **dbp,
dns_dbnode_t **nodep, dns_rdataset_t **rdatasetp,
dns_dbversion_t *version) {
dns_rdataset_t *trdataset = NULL;
rpz_match_clear(st);
st->m.rpz = rpz;
st->m.type = rpz_type;
st->m.policy = policy;
dns_name_copy(p_name, st->p_name);
st->m.prefix = prefix;
st->m.result = result;
SAVE(st->m.zone, *zonep);
SAVE(st->m.db, *dbp);
SAVE(st->m.node, *nodep);
if (*rdatasetp != NULL && dns_rdataset_isassociated(*rdatasetp)) {
/*
* Save the replacement rdataset from the policy
* and make the previous replacement rdataset scratch.
*/
SAVE(trdataset, st->m.rdataset);
SAVE(st->m.rdataset, *rdatasetp);
SAVE(*rdatasetp, trdataset);
st->m.ttl = ISC_MIN(st->m.rdataset->ttl, rpz->max_policy_ttl);
} else {
st->m.ttl = ISC_MIN(DNS_RPZ_TTL_DEFAULT, rpz->max_policy_ttl);
}
SAVE(st->m.version, version);
}
/*
* Check this address in every eligible policy zone.
*/
static isc_result_t
rpz_rewrite_ip(ns_client_t *client, const isc_netaddr_t *netaddr,
dns_rdatatype_t qtype, dns_rpz_type_t rpz_type,
dns_rpz_zbits_t zbits, dns_rdataset_t **p_rdatasetp) {
dns_rpz_zones_t *rpzs;
dns_rpz_st_t *st;
dns_rpz_zone_t *rpz;
dns_rpz_prefix_t prefix;
dns_rpz_num_t rpz_num;
dns_fixedname_t ip_namef, p_namef;
dns_name_t *ip_name, *p_name;
dns_zone_t *p_zone;
dns_db_t *p_db;
dns_dbversion_t *p_version;
dns_dbnode_t *p_node;
dns_rpz_policy_t policy;
isc_result_t result;
CTRACE(ISC_LOG_DEBUG(3), "rpz_rewrite_ip");
rpzs = client->inner.view->rpzs;
st = client->query.rpz_st;
ip_name = dns_fixedname_initname(&ip_namef);
p_zone = NULL;
p_db = NULL;
p_node = NULL;
while (zbits != 0) {
rpz_num = dns_rpz_find_ip(rpzs, rpz_type, zbits, netaddr,
ip_name, &prefix);
if (rpz_num == DNS_RPZ_INVALID_NUM) {
break;
}
zbits &= (DNS_RPZ_ZMASK(rpz_num) >> 1);
/*
* Do not try applying policy zones that cannot replace a
* previously found policy zone.
* Stop looking if the next best choice cannot
* replace what we already have.
*/
rpz = rpzs->zones[rpz_num];
if (st->m.policy != DNS_RPZ_POLICY_MISS) {
if (st->m.rpz->num < rpz->num) {
break;
}
if (st->m.rpz->num == rpz->num &&
(st->m.type < rpz_type || st->m.prefix > prefix))
{
break;
}
}
/*
* Get the policy for a prefix at least as long
* as the prefix of the entry we had before.
*/
p_name = dns_fixedname_initname(&p_namef);
result = rpz_get_p_name(client, p_name, rpz, rpz_type, ip_name);
if (result != ISC_R_SUCCESS) {
continue;
}
result = rpz_find_p(client, ip_name, qtype, p_name, rpz,
rpz_type, &p_zone, &p_db, &p_version,
&p_node, p_rdatasetp, &policy);
switch (result) {
case DNS_R_NXDOMAIN:
/*
* Continue after a policy record that is missing
* contrary to the summary data. The summary
* data can out of date during races with and among
* policy zone updates.
*/
CTRACE(ISC_LOG_ERROR, "rpz_rewrite_ip: mismatched "
"summary data; "
"continuing");
continue;
case DNS_R_SERVFAIL:
rpz_clean(&p_zone, &p_db, &p_node, p_rdatasetp);
st->m.policy = DNS_RPZ_POLICY_ERROR;
return DNS_R_SERVFAIL;
default:
/*
* Forget this policy if it is not preferable
* to the previously found policy.
* If this policy is not good, then stop looking
* because none of the later policy zones would work.
*
* With more than one applicable policy, prefer
* the earliest configured policy,
* client-IP over QNAME over IP over NSDNAME over NSIP,
* the longest prefix
* the lexically smallest address.
* dns_rpz_find_ip() ensures st->m.rpz->num >= rpz->num.
* We can compare new and current p_name because
* both are of the same type and in the same zone.
* The tests above eliminate other reasons to
* reject this policy. If this policy can't work,
* then neither can later zones.
*/
if (st->m.policy != DNS_RPZ_POLICY_MISS &&
rpz->num == st->m.rpz->num &&
(st->m.type == rpz_type && st->m.prefix == prefix &&
0 > dns_name_rdatacompare(st->p_name, p_name)))
{
break;
}
/*
* Stop checking after saving an enabled hit in this
* policy zone. The radix tree in the policy zone
* ensures that we found the longest match.
*/
if (rpz->policy != DNS_RPZ_POLICY_DISABLED) {
CTRACE(ISC_LOG_DEBUG(3), "rpz_rewrite_ip: "
"rpz_save_p");
rpz_save_p(st, rpz, rpz_type, policy, p_name,
prefix, result, &p_zone, &p_db,
&p_node, p_rdatasetp, p_version);
break;
}
/*
* Log DNS_RPZ_POLICY_DISABLED zones
* and try the next eligible policy zone.
*/
rpz_log_rewrite(client, true, policy, rpz_type, p_zone,
p_name, NULL, rpz_num);
}
}
rpz_clean(&p_zone, &p_db, &p_node, p_rdatasetp);
return ISC_R_SUCCESS;
}
/*
* Check the IP addresses in the A or AAAA rrsets for name against
* all eligible rpz_type (IP or NSIP) response policy rewrite rules.
*/
static isc_result_t
rpz_rewrite_ip_rrset(ns_client_t *client, dns_name_t *name,
dns_rdatatype_t qtype, dns_rpz_type_t rpz_type,
dns_rdatatype_t ip_type, dns_db_t **ip_dbp,
dns_dbversion_t *ip_version, dns_rdataset_t **ip_rdatasetp,
dns_rdataset_t **p_rdatasetp, bool resuming) {
dns_rpz_zbits_t zbits;
isc_netaddr_t netaddr;
struct in_addr ina;
struct in6_addr in6a;
isc_result_t result;
unsigned int options = client->query.dboptions | DNS_DBFIND_GLUEOK;
bool done = false;
CTRACE(ISC_LOG_DEBUG(3), "rpz_rewrite_ip_rrset");
do {
zbits = rpz_get_zbits(client, ip_type, rpz_type);
if (zbits == 0) {
return ISC_R_SUCCESS;
}
/*
* Get the A or AAAA rdataset.
*/
result = rpz_rrset_find(client, name, ip_type, options,
rpz_type, ip_dbp, ip_version,
ip_rdatasetp, resuming);
switch (result) {
case ISC_R_SUCCESS:
case DNS_R_GLUE:
case DNS_R_ZONECUT:
break;
case DNS_R_EMPTYNAME:
case DNS_R_EMPTYWILD:
case DNS_R_NXDOMAIN:
case DNS_R_NCACHENXDOMAIN:
case DNS_R_NXRRSET:
case DNS_R_NCACHENXRRSET:
case ISC_R_NOTFOUND:
return ISC_R_SUCCESS;
case DNS_R_DELEGATION:
case DNS_R_DUPLICATE:
case DNS_R_DROP:
return result;
case DNS_R_CNAME:
case DNS_R_DNAME:
rpz_log_fail(client, DNS_RPZ_DEBUG_LEVEL1, name,
rpz_type, "NS address rewrite rrset",
result);
return ISC_R_SUCCESS;
default:
if (client->query.rpz_st->m.policy !=
DNS_RPZ_POLICY_ERROR)
{
client->query.rpz_st->m.policy =
DNS_RPZ_POLICY_ERROR;
rpz_log_fail(client, DNS_RPZ_ERROR_LEVEL, name,
rpz_type,
"NS address rewrite rrset",
result);
}
CTRACE(ISC_LOG_ERROR,
"rpz_rewrite_ip_rrset: unexpected "
"result");
return DNS_R_SERVFAIL;
}
/*
* If we are processing glue setup for the next loop
* otherwise we are done.
*/
if (result == DNS_R_GLUE) {
options = client->query.dboptions;
} else {
options = client->query.dboptions | DNS_DBFIND_GLUEOK;
done = true;
}
/*
* Check all of the IP addresses in the rdataset.
*/
DNS_RDATASET_FOREACH (*ip_rdatasetp) {
dns_rdata_t rdata = DNS_RDATA_INIT;
dns_rdataset_current(*ip_rdatasetp, &rdata);
switch (rdata.type) {
case dns_rdatatype_a:
INSIST(rdata.length == 4);
memmove(&ina.s_addr, rdata.data, 4);
isc_netaddr_fromin(&netaddr, &ina);
break;
case dns_rdatatype_aaaa:
INSIST(rdata.length == 16);
memmove(in6a.s6_addr, rdata.data, 16);
isc_netaddr_fromin6(&netaddr, &in6a);
break;
default:
continue;
}
result = rpz_rewrite_ip(client, &netaddr, qtype,
rpz_type, zbits, p_rdatasetp);
if (result != ISC_R_SUCCESS) {
return result;
}
}
} while (!done &&
client->query.rpz_st->m.policy == DNS_RPZ_POLICY_MISS);
return ISC_R_SUCCESS;
}
/*
* Look for IP addresses in A and AAAA rdatasets
* that trigger all eligible IP or NSIP policy rules.
*/
static isc_result_t
rpz_rewrite_ip_rrsets(ns_client_t *client, dns_name_t *name,
dns_rdatatype_t qtype, dns_rpz_type_t rpz_type,
dns_rdataset_t **ip_rdatasetp, bool resuming) {
dns_rpz_st_t *st;
dns_dbversion_t *ip_version;
dns_db_t *ip_db;
dns_rdataset_t *p_rdataset;
isc_result_t result;
CTRACE(ISC_LOG_DEBUG(3), "rpz_rewrite_ip_rrsets");
st = client->query.rpz_st;
ip_version = NULL;
ip_db = NULL;
p_rdataset = NULL;
if ((st->state & DNS_RPZ_DONE_IPv4) == 0 &&
(qtype == dns_rdatatype_a || qtype == dns_rdatatype_any ||
rpz_type == DNS_RPZ_TYPE_NSIP))
{
/*
* Rewrite based on an IPv4 address that will appear
* in the ANSWER section or if we are checking IP addresses.
*/
result = rpz_rewrite_ip_rrset(
client, name, qtype, rpz_type, dns_rdatatype_a, &ip_db,
ip_version, ip_rdatasetp, &p_rdataset, resuming);
if (result == ISC_R_SUCCESS) {
st->state |= DNS_RPZ_DONE_IPv4;
}
} else {
result = ISC_R_SUCCESS;
}
if (result == ISC_R_SUCCESS &&
(qtype == dns_rdatatype_aaaa || qtype == dns_rdatatype_any ||
rpz_type == DNS_RPZ_TYPE_NSIP))
{
/*
* Rewrite based on IPv6 addresses that will appear
* in the ANSWER section or if we are checking IP addresses.
*/
result = rpz_rewrite_ip_rrset(client, name, qtype, rpz_type,
dns_rdatatype_aaaa, &ip_db,
ip_version, ip_rdatasetp,
&p_rdataset, resuming);
}
if (ip_db != NULL) {
dns_db_detach(&ip_db);
}
ns_client_putrdataset(client, &p_rdataset);
return result;
}
/*
* Try to rewrite a request for a qtype rdataset based on the trigger name
* trig_name and rpz_type (DNS_RPZ_TYPE_QNAME or DNS_RPZ_TYPE_NSDNAME).
* Record the results including the replacement rdataset if any
* in client->query.rpz_st.
* *rdatasetp is a scratch rdataset.
*/
static isc_result_t
rpz_rewrite_name(ns_client_t *client, dns_name_t *trig_name,
dns_rdatatype_t qtype, dns_rpz_type_t rpz_type,
dns_rpz_zbits_t allowed_zbits, dns_rdataset_t **rdatasetp) {
dns_rpz_zones_t *rpzs;
dns_rpz_zone_t *rpz;
dns_rpz_st_t *st;
dns_fixedname_t p_namef;
dns_name_t *p_name;
dns_rpz_zbits_t zbits;
dns_rpz_num_t rpz_num;
dns_zone_t *p_zone;
dns_db_t *p_db;
dns_dbversion_t *p_version;
dns_dbnode_t *p_node;
dns_rpz_policy_t policy;
isc_result_t result;
CTRACE(ISC_LOG_DEBUG(3), "rpz_rewrite_name");
rpzs = client->inner.view->rpzs;
st = client->query.rpz_st;
zbits = rpz_get_zbits(client, qtype, rpz_type);
zbits &= allowed_zbits;
if (zbits == 0) {
return ISC_R_SUCCESS;
}
/*
* Use the summary database to find the bit mask of policy zones
* with policies for this trigger name. We do this even if there
* is only one eligible policy zone so that wildcard triggers
* are matched correctly, and not into their parent.
*/
zbits = dns_rpz_find_name(rpzs, rpz_type, zbits, trig_name);
if (zbits == 0) {
return ISC_R_SUCCESS;
}
p_name = dns_fixedname_initname(&p_namef);
p_zone = NULL;
p_db = NULL;
p_node = NULL;
/*
* Check the trigger name in every policy zone that the summary data
* says has a hit for the trigger name.
* Most of the time there are no eligible zones and the summary data
* keeps us from getting this far.
* We check the most eligible zone first and so usually check only
* one policy zone.
*/
for (rpz_num = 0; zbits != 0; ++rpz_num, zbits >>= 1) {
if ((zbits & 1) == 0) {
continue;
}
/*
* Do not check policy zones that cannot replace a previously
* found policy.
*/
rpz = rpzs->zones[rpz_num];
if (st->m.policy != DNS_RPZ_POLICY_MISS) {
if (st->m.rpz->num < rpz->num) {
break;
}
if (st->m.rpz->num == rpz->num && st->m.type < rpz_type)
{
break;
}
}
/*
* Get the next policy zone's record for this trigger name.
*/
result = rpz_get_p_name(client, p_name, rpz, rpz_type,
trig_name);
if (result != ISC_R_SUCCESS) {
continue;
}
result = rpz_find_p(client, trig_name, qtype, p_name, rpz,
rpz_type, &p_zone, &p_db, &p_version,
&p_node, rdatasetp, &policy);
switch (result) {
case DNS_R_NXDOMAIN:
/*
* Continue after a missing policy record
* contrary to the summary data. The summary
* data can out of date during races with and among
* policy zone updates.
*/
CTRACE(ISC_LOG_ERROR, "rpz_rewrite_name: mismatched "
"summary data; "
"continuing");
continue;
case DNS_R_SERVFAIL:
rpz_clean(&p_zone, &p_db, &p_node, rdatasetp);
st->m.policy = DNS_RPZ_POLICY_ERROR;
return DNS_R_SERVFAIL;
default:
/*
* With more than one applicable policy, prefer
* the earliest configured policy,
* client-IP over QNAME over IP over NSDNAME over NSIP,
* and the smallest name.
* We known st->m.rpz->num >= rpz->num and either
* st->m.rpz->num > rpz->num or st->m.type >= rpz_type
*/
if (st->m.policy != DNS_RPZ_POLICY_MISS &&
rpz->num == st->m.rpz->num &&
(st->m.type < rpz_type ||
(st->m.type == rpz_type &&
0 >= dns_name_compare(p_name, st->p_name))))
{
continue;
}
if (rpz->policy != DNS_RPZ_POLICY_DISABLED) {
CTRACE(ISC_LOG_DEBUG(3), "rpz_rewrite_name: "
"rpz_save_p");
rpz_save_p(st, rpz, rpz_type, policy, p_name, 0,
result, &p_zone, &p_db, &p_node,
rdatasetp, p_version);
/*
* After a hit, higher numbered policy zones
* are irrelevant
*/
rpz_clean(&p_zone, &p_db, &p_node, rdatasetp);
return ISC_R_SUCCESS;
}
/*
* Log DNS_RPZ_POLICY_DISABLED zones
* and try the next eligible policy zone.
*/
rpz_log_rewrite(client, true, policy, rpz_type, p_zone,
p_name, NULL, rpz_num);
break;
}
}
rpz_clean(&p_zone, &p_db, &p_node, rdatasetp);
return ISC_R_SUCCESS;
}
static void
rpz_rewrite_ns_skip(ns_client_t *client, dns_name_t *nsname,
isc_result_t result, int level, const char *str) {
dns_rpz_st_t *st;
CTRACE(ISC_LOG_DEBUG(3), "rpz_rewrite_ns_skip");
st = client->query.rpz_st;
if (str != NULL) {
rpz_log_fail_helper(client, level, nsname, DNS_RPZ_TYPE_NSIP,
DNS_RPZ_TYPE_NSDNAME, str, result);
}
if (st->r.ns_rdataset != NULL &&
dns_rdataset_isassociated(st->r.ns_rdataset))
{
dns_rdataset_disassociate(st->r.ns_rdataset);
}
st->r.label--;
}
/*
* RPZ query result types
*/
typedef enum {
qresult_type_done = 0,
qresult_type_restart = 1,
qresult_type_recurse = 2
} qresult_type_t;
/*
* Look for response policy zone QNAME, NSIP, and NSDNAME rewriting.
*/
static isc_result_t
rpz_rewrite(ns_client_t *client, dns_rdatatype_t qtype, isc_result_t qresult,
bool resuming, dns_rdataset_t *ordataset, dns_rdataset_t *osigset) {
dns_rpz_zones_t *rpzs;
dns_rpz_st_t *st;
dns_rdataset_t *rdataset;
dns_fixedname_t nsnamef;
dns_name_t *nsname;
qresult_type_t qresult_type;
dns_rpz_zbits_t zbits;
isc_result_t result = ISC_R_SUCCESS;
dns_rpz_have_t have;
dns_rpz_popt_t popt;
int rpz_ver;
unsigned int options;
CTRACE(ISC_LOG_DEBUG(3), "rpz_rewrite");
rpzs = client->inner.view->rpzs;
st = client->query.rpz_st;
if (rpzs == NULL) {
return ISC_R_NOTFOUND;
}
if (st != NULL && (st->state & DNS_RPZ_REWRITTEN) != 0) {
return DNS_R_DISALLOWED;
}
if (RECURSING(client)) {
return DNS_R_DISALLOWED;
}
RWLOCK(&rpzs->search_lock, isc_rwlocktype_read);
if ((rpzs->p.num_zones == 0) ||
(!RECURSIONOK(client) && rpzs->p.no_rd_ok == 0) ||
!rpz_ck_dnssec(client, qresult, ordataset, osigset))
{
RWUNLOCK(&rpzs->search_lock, isc_rwlocktype_read);
return DNS_R_DISALLOWED;
}
have = rpzs->have;
popt = rpzs->p;
rpz_ver = rpzs->rpz_ver;
RWUNLOCK(&rpzs->search_lock, isc_rwlocktype_read);
if (st == NULL) {
st = isc_mem_get(client->manager->mctx, sizeof(*st));
st->state = 0;
}
if (st->state == 0) {
st->state |= DNS_RPZ_ACTIVE;
memset(&st->m, 0, sizeof(st->m));
st->m.type = DNS_RPZ_TYPE_BAD;
st->m.policy = DNS_RPZ_POLICY_MISS;
st->m.ttl = ~0;
memset(&st->r, 0, sizeof(st->r));
memset(&st->q, 0, sizeof(st->q));
st->p_name = dns_fixedname_initname(&st->_p_namef);
st->r_name = dns_fixedname_initname(&st->_r_namef);
st->fname = dns_fixedname_initname(&st->_fnamef);
st->have = have;
st->popt = popt;
st->rpz_ver = rpz_ver;
client->query.rpz_st = st;
}
/*
* There is nothing to rewrite if the main query failed.
*/
switch (qresult) {
case ISC_R_SUCCESS:
case DNS_R_GLUE:
case DNS_R_ZONECUT:
qresult_type = qresult_type_done;
break;
case DNS_R_EMPTYNAME:
case DNS_R_NXRRSET:
case DNS_R_NXDOMAIN:
case DNS_R_EMPTYWILD:
case DNS_R_NCACHENXDOMAIN:
case DNS_R_NCACHENXRRSET:
case DNS_R_COVERINGNSEC:
case DNS_R_CNAME:
case DNS_R_DNAME:
qresult_type = qresult_type_restart;
break;
case DNS_R_DELEGATION:
case ISC_R_NOTFOUND:
/*
* If recursion is on, do only tentative rewriting.
* If recursion is off, this the normal and only time we
* can rewrite.
*/
if (RECURSIONOK(client)) {
qresult_type = qresult_type_recurse;
} else {
qresult_type = qresult_type_restart;
}
break;
case ISC_R_FAILURE:
case ISC_R_TIMEDOUT:
case ISC_R_CANCELED:
case DNS_R_BROKENCHAIN:
rpz_log_fail(client, DNS_RPZ_DEBUG_LEVEL3, NULL,
DNS_RPZ_TYPE_QNAME,
"stop on qresult in rpz_rewrite()", qresult);
return ISC_R_SUCCESS;
default:
rpz_log_fail(client, DNS_RPZ_DEBUG_LEVEL1, NULL,
DNS_RPZ_TYPE_QNAME,
"stop on unrecognized qresult in rpz_rewrite()",
qresult);
return ISC_R_SUCCESS;
}
rdataset = NULL;
if ((st->state & (DNS_RPZ_DONE_CLIENT_IP | DNS_RPZ_DONE_QNAME)) !=
(DNS_RPZ_DONE_CLIENT_IP | DNS_RPZ_DONE_QNAME))
{
isc_netaddr_t netaddr;
dns_rpz_zbits_t allowed;
if (qresult_type == qresult_type_recurse) {
/*
* This request needs recursion that has not been done.
* Get bits for the policy zones that do not need
* to wait for the results of recursion.
*/
allowed = st->have.qname_skip_recurse;
if (allowed == 0) {
return ISC_R_SUCCESS;
}
} else {
allowed = DNS_RPZ_ALL_ZBITS;
}
/*
* Check once for triggers for the client IP address.
*/
if ((st->state & DNS_RPZ_DONE_CLIENT_IP) == 0) {
zbits = rpz_get_zbits(client, dns_rdatatype_none,
DNS_RPZ_TYPE_CLIENT_IP);
zbits &= allowed;
if (zbits != 0) {
isc_netaddr_fromsockaddr(
&netaddr, &client->inner.peeraddr);
result = rpz_rewrite_ip(client, &netaddr, qtype,
DNS_RPZ_TYPE_CLIENT_IP,
zbits, &rdataset);
if (result != ISC_R_SUCCESS) {
goto cleanup;
}
}
}
/*
* Check triggers for the query name if this is the first time
* for the current qname.
* There is a first time for each name in a CNAME chain
*/
if ((st->state & DNS_RPZ_DONE_QNAME) == 0) {
result = rpz_rewrite_name(client, client->query.qname,
qtype, DNS_RPZ_TYPE_QNAME,
allowed, &rdataset);
if (result != ISC_R_SUCCESS) {
goto cleanup;
}
/*
* Check IPv4 addresses in A RRs next.
* Reset to the start of the NS names.
*/
st->r.label = dns_name_countlabels(client->query.qname);
st->state &= ~(DNS_RPZ_DONE_QNAME_IP |
DNS_RPZ_DONE_IPv4);
}
/*
* Quit if this was an attempt to find a qname or
* client-IP trigger before recursion.
* We will be back if no pre-recursion triggers hit.
* For example, consider 2 policy zones, both with qname and
* IP address triggers. If the qname misses the 1st zone,
* then we cannot know whether a hit for the qname in the
* 2nd zone matters until after recursing to get the A RRs and
* testing them in the first zone.
* Do not bother saving the work from this attempt,
* because recursion is so slow.
*/
if (qresult_type == qresult_type_recurse) {
goto cleanup;
}
/*
* DNS_RPZ_DONE_QNAME but not DNS_RPZ_DONE_CLIENT_IP
* is reset at the end of dealing with each CNAME.
*/
st->state |= (DNS_RPZ_DONE_CLIENT_IP | DNS_RPZ_DONE_QNAME);
}
/*
* Check known IP addresses for the query name if the database lookup
* resulted in some addresses (qresult_type == qresult_type_done)
* and if we have not already checked them.
* Any recursion required for the query has already happened.
* Do not check addresses that will not be in the ANSWER section.
*/
if ((st->state & DNS_RPZ_DONE_QNAME_IP) == 0 &&
qresult_type == qresult_type_done &&
rpz_get_zbits(client, qtype, DNS_RPZ_TYPE_IP) != 0)
{
result = rpz_rewrite_ip_rrsets(client, client->query.qname,
qtype, DNS_RPZ_TYPE_IP,
&rdataset, resuming);
if (result != ISC_R_SUCCESS) {
goto cleanup;
}
/*
* We are finished checking the IP addresses for the qname.
* Start with IPv4 if we will check NS IP addresses.
*/
st->state |= DNS_RPZ_DONE_QNAME_IP;
st->state &= ~DNS_RPZ_DONE_IPv4;
}
/*
* Stop looking for rules if there are none of the other kinds
* that could override what we already have.
*/
if (rpz_get_zbits(client, dns_rdatatype_any, DNS_RPZ_TYPE_NSDNAME) ==
0 &&
rpz_get_zbits(client, dns_rdatatype_any, DNS_RPZ_TYPE_NSIP) == 0)
{
result = ISC_R_SUCCESS;
goto cleanup;
}
dns_fixedname_init(&nsnamef);
dns_name_clone(client->query.qname, dns_fixedname_name(&nsnamef));
options = client->query.dboptions | DNS_DBFIND_GLUEOK;
while (st->r.label > st->popt.min_ns_labels) {
bool was_glue = false;
/*
* Get NS rrset for each domain in the current qname.
*/
if (st->r.label == dns_name_countlabels(client->query.qname)) {
nsname = client->query.qname;
} else {
nsname = dns_fixedname_name(&nsnamef);
dns_name_split(client->query.qname, st->r.label, NULL,
nsname);
}
if (st->r.ns_rdataset == NULL ||
!dns_rdataset_isassociated(st->r.ns_rdataset))
{
dns_db_t *db = NULL;
result = rpz_rrset_find(client, nsname,
dns_rdatatype_ns, options,
DNS_RPZ_TYPE_NSDNAME, &db, NULL,
&st->r.ns_rdataset, resuming);
if (db != NULL) {
dns_db_detach(&db);
}
if (st->m.policy == DNS_RPZ_POLICY_ERROR) {
goto cleanup;
}
switch (result) {
case DNS_R_GLUE:
was_glue = true;
FALLTHROUGH;
case ISC_R_SUCCESS:
result = dns_rdataset_first(st->r.ns_rdataset);
if (result != ISC_R_SUCCESS) {
goto cleanup;
}
st->state &= ~(DNS_RPZ_DONE_NSDNAME |
DNS_RPZ_DONE_IPv4);
break;
case DNS_R_DELEGATION:
case DNS_R_DUPLICATE:
case DNS_R_DROP:
goto cleanup;
case DNS_R_EMPTYNAME:
case DNS_R_NXRRSET:
case DNS_R_EMPTYWILD:
case DNS_R_NXDOMAIN:
case DNS_R_NCACHENXDOMAIN:
case DNS_R_NCACHENXRRSET:
case ISC_R_NOTFOUND:
case DNS_R_CNAME:
case DNS_R_DNAME:
rpz_rewrite_ns_skip(client, nsname, result, 0,
NULL);
continue;
case ISC_R_TIMEDOUT:
case DNS_R_BROKENCHAIN:
case ISC_R_FAILURE:
rpz_rewrite_ns_skip(client, nsname, result,
DNS_RPZ_DEBUG_LEVEL3,
" NS rpz_rrset_find()");
continue;
default:
rpz_rewrite_ns_skip(client, nsname, result,
DNS_RPZ_INFO_LEVEL,
" unrecognized NS"
" rpz_rrset_find()");
continue;
}
}
/*
* Check all NS names.
*/
do {
dns_rdata_ns_t ns;
dns_rdata_t nsrdata = DNS_RDATA_INIT;
dns_rdataset_current(st->r.ns_rdataset, &nsrdata);
result = dns_rdata_tostruct(&nsrdata, &ns, NULL);
RUNTIME_CHECK(result == ISC_R_SUCCESS);
dns_rdata_reset(&nsrdata);
/*
* Do nothing about "NS ."
*/
if (dns_name_equal(&ns.name, dns_rootname)) {
dns_rdata_freestruct(&ns);
result = dns_rdataset_next(st->r.ns_rdataset);
continue;
}
/*
* Check this NS name if we did not handle it
* during a previous recursion.
*/
if ((st->state & DNS_RPZ_DONE_NSDNAME) == 0) {
result = rpz_rewrite_name(
client, &ns.name, qtype,
DNS_RPZ_TYPE_NSDNAME, DNS_RPZ_ALL_ZBITS,
&rdataset);
if (result != ISC_R_SUCCESS) {
dns_rdata_freestruct(&ns);
goto cleanup;
}
st->state |= DNS_RPZ_DONE_NSDNAME;
}
/*
* Check all IP addresses for this NS name.
*/
result = rpz_rewrite_ip_rrsets(client, &ns.name, qtype,
DNS_RPZ_TYPE_NSIP,
&rdataset, resuming);
dns_rdata_freestruct(&ns);
if (result != ISC_R_SUCCESS) {
goto cleanup;
}
st->state &= ~(DNS_RPZ_DONE_NSDNAME |
DNS_RPZ_DONE_IPv4);
result = dns_rdataset_next(st->r.ns_rdataset);
} while (result == ISC_R_SUCCESS);
dns_rdataset_disassociate(st->r.ns_rdataset);
/*
* If we just checked a glue NS RRset retry without allowing
* glue responses, otherwise setup for the next name.
*/
if (was_glue) {
options = client->query.dboptions;
} else {
options = client->query.dboptions | DNS_DBFIND_GLUEOK;
st->r.label--;
}
if (rpz_get_zbits(client, dns_rdatatype_any,
DNS_RPZ_TYPE_NSDNAME) == 0 &&
rpz_get_zbits(client, dns_rdatatype_any,
DNS_RPZ_TYPE_NSIP) == 0)
{
break;
}
}
/*
* Use the best hit, if any.
*/
result = ISC_R_SUCCESS;
cleanup:
if (st->m.policy != DNS_RPZ_POLICY_MISS &&
st->m.policy != DNS_RPZ_POLICY_ERROR &&
st->m.rpz->policy != DNS_RPZ_POLICY_GIVEN)
{
st->m.policy = st->m.rpz->policy;
}
if (st->m.policy == DNS_RPZ_POLICY_MISS ||
st->m.policy == DNS_RPZ_POLICY_PASSTHRU ||
st->m.policy == DNS_RPZ_POLICY_ERROR)
{
if (st->m.policy == DNS_RPZ_POLICY_PASSTHRU &&
result != DNS_R_DELEGATION)
{
rpz_log_rewrite(client, false, st->m.policy, st->m.type,
st->m.zone, st->p_name, NULL,
st->m.rpz->num);
}
rpz_match_clear(st);
}
if (st->m.policy == DNS_RPZ_POLICY_ERROR) {
CTRACE(ISC_LOG_ERROR, "SERVFAIL due to RPZ policy");
st->m.type = DNS_RPZ_TYPE_BAD;
result = DNS_R_SERVFAIL;
}
ns_client_putrdataset(client, &rdataset);
if ((st->state & DNS_RPZ_RECURSING) == 0) {
rpz_clean(NULL, &st->r.db, NULL, &st->r.ns_rdataset);
}
return result;
}
/*
* See if response policy zone rewriting is allowed by a lack of interest
* by the client in DNSSEC or a lack of signatures.
*/
static bool
rpz_ck_dnssec(ns_client_t *client, isc_result_t qresult,
dns_rdataset_t *rdataset, dns_rdataset_t *sigrdataset) {
dns_fixedname_t fixed;
dns_name_t *found = NULL;
dns_rdataset_t trdataset = DNS_RDATASET_INIT;
dns_rdatatype_t type;
CTRACE(ISC_LOG_DEBUG(3), "rpz_ck_dnssec");
if (client->inner.view->rpzs->p.break_dnssec || !WANTDNSSEC(client)) {
return true;
}
/*
* We do not know if there are signatures if we have not recursed
* for them.
*/
if (qresult == DNS_R_DELEGATION || qresult == ISC_R_NOTFOUND) {
return false;
}
if (sigrdataset == NULL) {
return true;
}
if (dns_rdataset_isassociated(sigrdataset)) {
return false;
}
/*
* We are happy to rewrite nothing.
*/
if (rdataset == NULL || !dns_rdataset_isassociated(rdataset)) {
return true;
}
/*
* Do not rewrite if there is any sign of signatures.
*/
if (rdataset->type == dns_rdatatype_nsec ||
rdataset->type == dns_rdatatype_nsec3 ||
rdataset->type == dns_rdatatype_rrsig)
{
return false;
}
/*
* Look for a signature in a negative cache rdataset.
*/
if (!rdataset->attributes.negative) {
return true;
}
found = dns_fixedname_initname(&fixed);
dns_rdataset_init(&trdataset);
DNS_RDATASET_FOREACH (rdataset) {
dns_ncache_current(rdataset, found, &trdataset);
type = trdataset.type;
dns_rdataset_disassociate(&trdataset);
if (type == dns_rdatatype_nsec || type == dns_rdatatype_nsec3 ||
type == dns_rdatatype_rrsig)
{
return false;
}
}
return true;
}
static unsigned char inaddr10[] = "\00210\007IN-ADDR\004ARPA";
static unsigned char inaddr16172[] = "\00216\003172\007IN-ADDR\004ARPA";
static unsigned char inaddr17172[] = "\00217\003172\007IN-ADDR\004ARPA";
static unsigned char inaddr18172[] = "\00218\003172\007IN-ADDR\004ARPA";
static unsigned char inaddr19172[] = "\00219\003172\007IN-ADDR\004ARPA";
static unsigned char inaddr20172[] = "\00220\003172\007IN-ADDR\004ARPA";
static unsigned char inaddr21172[] = "\00221\003172\007IN-ADDR\004ARPA";
static unsigned char inaddr22172[] = "\00222\003172\007IN-ADDR\004ARPA";
static unsigned char inaddr23172[] = "\00223\003172\007IN-ADDR\004ARPA";
static unsigned char inaddr24172[] = "\00224\003172\007IN-ADDR\004ARPA";
static unsigned char inaddr25172[] = "\00225\003172\007IN-ADDR\004ARPA";
static unsigned char inaddr26172[] = "\00226\003172\007IN-ADDR\004ARPA";
static unsigned char inaddr27172[] = "\00227\003172\007IN-ADDR\004ARPA";
static unsigned char inaddr28172[] = "\00228\003172\007IN-ADDR\004ARPA";
static unsigned char inaddr29172[] = "\00229\003172\007IN-ADDR\004ARPA";
static unsigned char inaddr30172[] = "\00230\003172\007IN-ADDR\004ARPA";
static unsigned char inaddr31172[] = "\00231\003172\007IN-ADDR\004ARPA";
static unsigned char inaddr168192[] = "\003168\003192\007IN-ADDR\004ARPA";
static dns_name_t rfc1918names[] = {
DNS_NAME_INITABSOLUTE(inaddr10), DNS_NAME_INITABSOLUTE(inaddr16172),
DNS_NAME_INITABSOLUTE(inaddr17172), DNS_NAME_INITABSOLUTE(inaddr18172),
DNS_NAME_INITABSOLUTE(inaddr19172), DNS_NAME_INITABSOLUTE(inaddr20172),
DNS_NAME_INITABSOLUTE(inaddr21172), DNS_NAME_INITABSOLUTE(inaddr22172),
DNS_NAME_INITABSOLUTE(inaddr23172), DNS_NAME_INITABSOLUTE(inaddr24172),
DNS_NAME_INITABSOLUTE(inaddr25172), DNS_NAME_INITABSOLUTE(inaddr26172),
DNS_NAME_INITABSOLUTE(inaddr27172), DNS_NAME_INITABSOLUTE(inaddr28172),
DNS_NAME_INITABSOLUTE(inaddr29172), DNS_NAME_INITABSOLUTE(inaddr30172),
DNS_NAME_INITABSOLUTE(inaddr31172), DNS_NAME_INITABSOLUTE(inaddr168192)
};
static unsigned char prisoner_data[] = "\010prisoner\004iana\003org";
static unsigned char hostmaster_data[] = "\012hostmaster\014root-"
"servers\003org";
static dns_name_t const prisoner = DNS_NAME_INITABSOLUTE(prisoner_data);
static dns_name_t const hostmaster = DNS_NAME_INITABSOLUTE(hostmaster_data);
static void
warn_rfc1918(ns_client_t *client, dns_name_t *fname, dns_rdataset_t *rdataset) {
unsigned int i;
dns_rdata_t rdata = DNS_RDATA_INIT;
dns_rdata_soa_t soa;
dns_rdataset_t found;
isc_result_t result;
for (i = 0; i < (sizeof(rfc1918names) / sizeof(*rfc1918names)); i++) {
if (dns_name_issubdomain(fname, &rfc1918names[i])) {
dns_rdataset_init(&found);
result = dns_ncache_getrdataset(
rdataset, &rfc1918names[i], dns_rdatatype_soa,
&found);
if (result != ISC_R_SUCCESS) {
return;
}
result = dns_rdataset_first(&found);
RUNTIME_CHECK(result == ISC_R_SUCCESS);
dns_rdataset_current(&found, &rdata);
result = dns_rdata_tostruct(&rdata, &soa, NULL);
RUNTIME_CHECK(result == ISC_R_SUCCESS);
if (dns_name_equal(&soa.origin, &prisoner) &&
dns_name_equal(&soa.contact, &hostmaster))
{
char buf[DNS_NAME_FORMATSIZE];
dns_name_format(fname, buf, sizeof(buf));
ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
NS_LOGMODULE_QUERY,
ISC_LOG_WARNING,
"RFC 1918 response from "
"Internet for %s",
buf);
}
dns_rdataset_disassociate(&found);
return;
}
}
}
static void
query_findclosestnsec3(dns_name_t *qname, dns_db_t *db,
dns_dbversion_t *version, ns_client_t *client,
dns_rdataset_t *rdataset, dns_rdataset_t *sigrdataset,
dns_name_t *fname, bool exact, dns_name_t *found) {
unsigned char salt[256];
size_t salt_length;
uint16_t iterations;
isc_result_t result;
unsigned int dboptions;
dns_fixedname_t fixed;
dns_hash_t hash;
dns_name_t name;
unsigned int skip = 0, labels;
dns_rdata_nsec3_t nsec3;
dns_rdata_t rdata = DNS_RDATA_INIT;
bool optout;
dns_clientinfomethods_t cm;
dns_clientinfo_t ci;
salt_length = sizeof(salt);
result = dns_db_getnsec3parameters(db, version, &hash, NULL,
&iterations, salt, &salt_length);
if (result != ISC_R_SUCCESS) {
return;
}
dns_name_init(&name);
dns_name_clone(qname, &name);
labels = dns_name_countlabels(&name);
dns_clientinfomethods_init(&cm, ns_client_sourceip);
dns_clientinfo_init(&ci, client, NULL);
/*
* Map unknown algorithm to known value.
*/
if (hash == DNS_NSEC3_UNKNOWNALG) {
hash = 1;
}
again:
dns_fixedname_init(&fixed);
result = dns_nsec3_hashname(&fixed, NULL, NULL, &name,
dns_db_origin(db), hash, iterations, salt,
salt_length);
if (result != ISC_R_SUCCESS) {
return;
}
dboptions = client->query.dboptions | DNS_DBFIND_FORCENSEC3;
result = dns_db_findext(db, dns_fixedname_name(&fixed), version,
dns_rdatatype_nsec3, dboptions,
client->inner.now, NULL, fname, &cm, &ci,
rdataset, sigrdataset);
if (result == DNS_R_NXDOMAIN) {
if (!dns_rdataset_isassociated(rdataset)) {
return;
}
result = dns_rdataset_first(rdataset);
INSIST(result == ISC_R_SUCCESS);
dns_rdataset_current(rdataset, &rdata);
result = dns_rdata_tostruct(&rdata, &nsec3, NULL);
RUNTIME_CHECK(result == ISC_R_SUCCESS);
dns_rdata_reset(&rdata);
optout = ((nsec3.flags & DNS_NSEC3FLAG_OPTOUT) != 0);
if (found != NULL && optout &&
dns_name_issubdomain(&name, dns_db_origin(db)))
{
dns_rdataset_disassociate(rdataset);
if (dns_rdataset_isassociated(sigrdataset)) {
dns_rdataset_disassociate(sigrdataset);
}
skip++;
dns_name_getlabelsequence(qname, skip, labels - skip,
&name);
ns_client_log(client, DNS_LOGCATEGORY_DNSSEC,
NS_LOGMODULE_QUERY, ISC_LOG_DEBUG(3),
"looking for closest provable encloser");
goto again;
}
if (exact) {
ns_client_log(client, DNS_LOGCATEGORY_DNSSEC,
NS_LOGMODULE_QUERY, ISC_LOG_WARNING,
"expected a exact match NSEC3, got "
"a covering record");
}
} else if (result != ISC_R_SUCCESS) {
return;
} else if (!exact) {
ns_client_log(client, DNS_LOGCATEGORY_DNSSEC,
NS_LOGMODULE_QUERY, ISC_LOG_WARNING,
"expected covering NSEC3, got an exact match");
}
if (found == qname) {
if (skip != 0U) {
dns_name_getlabelsequence(qname, skip, labels - skip,
found);
}
} else if (found != NULL) {
dns_name_copy(&name, found);
}
return;
}
static uint32_t
dns64_ttl(dns_db_t *db, dns_dbversion_t *version) {
dns_dbnode_t *node = NULL;
dns_rdata_soa_t soa;
dns_rdata_t rdata = DNS_RDATA_INIT;
dns_rdataset_t rdataset;
isc_result_t result;
uint32_t ttl = UINT32_MAX;
dns_rdataset_init(&rdataset);
result = dns_db_getoriginnode(db, &node);
if (result != ISC_R_SUCCESS) {
goto cleanup;
}
result = dns_db_findrdataset(db, node, version, dns_rdatatype_soa, 0, 0,
&rdataset, NULL);
if (result != ISC_R_SUCCESS) {
goto cleanup;
}
result = dns_rdataset_first(&rdataset);
if (result != ISC_R_SUCCESS) {
goto cleanup;
}
dns_rdataset_current(&rdataset, &rdata);
result = dns_rdata_tostruct(&rdata, &soa, NULL);
RUNTIME_CHECK(result == ISC_R_SUCCESS);
ttl = ISC_MIN(rdataset.ttl, soa.minimum);
cleanup:
if (dns_rdataset_isassociated(&rdataset)) {
dns_rdataset_disassociate(&rdataset);
}
if (node != NULL) {
dns_db_detachnode(db, &node);
}
return ttl;
}
static bool
dns64_aaaaok(ns_client_t *client, dns_rdataset_t *rdataset,
dns_rdataset_t *sigrdataset) {
isc_netaddr_t netaddr;
dns_aclenv_t *env = client->manager->aclenv;
dns_dns64_t *dns64 = ISC_LIST_HEAD(client->inner.view->dns64);
unsigned int flags = 0;
unsigned int i, count;
bool *aaaaok;
INSIST(client->query.dns64_aaaaok == NULL);
INSIST(client->query.dns64_aaaaoklen == 0);
INSIST(client->query.dns64_aaaa == NULL);
INSIST(client->query.dns64_sigaaaa == NULL);
if (dns64 == NULL) {
return true;
}
if (RECURSIONOK(client)) {
flags |= DNS_DNS64_RECURSIVE;
}
if (WANTDNSSEC(client) && sigrdataset != NULL &&
dns_rdataset_isassociated(sigrdataset))
{
flags |= DNS_DNS64_DNSSEC;
}
count = dns_rdataset_count(rdataset);
aaaaok = isc_mem_cget(client->manager->mctx, count, sizeof(bool));
isc_netaddr_fromsockaddr(&netaddr, &client->inner.peeraddr);
if (dns_dns64_aaaaok(dns64, &netaddr, client->inner.signer, env, flags,
rdataset, aaaaok, count))
{
for (i = 0; i < count; i++) {
if (aaaaok != NULL && !aaaaok[i]) {
SAVE(client->query.dns64_aaaaok, aaaaok);
client->query.dns64_aaaaoklen = count;
break;
}
}
if (aaaaok != NULL) {
isc_mem_cput(client->manager->mctx, aaaaok, count,
sizeof(bool));
}
return true;
}
if (aaaaok != NULL) {
isc_mem_cput(client->manager->mctx, aaaaok, count,
sizeof(bool));
}
return false;
}
/*
* Look for the name and type in the redirection zone. If found update
* the arguments as appropriate. Return true if a update was
* performed.
*
* Only perform the update if the client is in the allow query acl and
* returning the update would not cause a DNSSEC validation failure.
*/
static isc_result_t
redirect(ns_client_t *client, dns_name_t *name, dns_rdataset_t *rdataset,
dns_dbnode_t **nodep, dns_db_t **dbp, dns_dbversion_t **versionp,
dns_rdatatype_t qtype) {
dns_db_t *db = NULL;
dns_dbnode_t *node = NULL;
dns_fixedname_t fixed;
dns_name_t *found = NULL;
dns_rdataset_t trdataset = DNS_RDATASET_INIT;
isc_result_t result;
dns_rdatatype_t type;
dns_clientinfomethods_t cm;
dns_clientinfo_t ci;
ns_dbversion_t *dbversion = NULL;
CTRACE(ISC_LOG_DEBUG(3), "redirect");
if (client->inner.view->redirect == NULL) {
return ISC_R_NOTFOUND;
}
found = dns_fixedname_initname(&fixed);
dns_clientinfomethods_init(&cm, ns_client_sourceip);
dns_clientinfo_init(&ci, client, NULL);
dns_clientinfo_setecs(&ci, &client->inner.ecs);
if (WANTDNSSEC(client) && dns_db_iszone(*dbp) && dns_db_issecure(*dbp))
{
return ISC_R_NOTFOUND;
}
if (WANTDNSSEC(client) && dns_rdataset_isassociated(rdataset)) {
if (rdataset->trust == dns_trust_secure) {
return ISC_R_NOTFOUND;
}
if (rdataset->trust == dns_trust_ultimate &&
(rdataset->type == dns_rdatatype_nsec ||
rdataset->type == dns_rdatatype_nsec3))
{
return ISC_R_NOTFOUND;
}
if (rdataset->attributes.negative) {
DNS_RDATASET_FOREACH (rdataset) {
dns_ncache_current(rdataset, found, &trdataset);
type = trdataset.type;
dns_rdataset_disassociate(&trdataset);
if (type == dns_rdatatype_nsec ||
type == dns_rdatatype_nsec3 ||
type == dns_rdatatype_rrsig)
{
return ISC_R_NOTFOUND;
}
}
}
}
result = ns_client_checkaclsilent(
client, NULL,
dns_zone_getqueryacl(client->inner.view->redirect), true);
if (result != ISC_R_SUCCESS) {
return ISC_R_NOTFOUND;
}
result = dns_zone_getdb(client->inner.view->redirect, &db);
if (result != ISC_R_SUCCESS) {
return ISC_R_NOTFOUND;
}
dbversion = ns_client_findversion(client, db);
if (dbversion == NULL) {
dns_db_detach(&db);
return ISC_R_NOTFOUND;
}
/*
* Lookup the requested data in the redirect zone.
*/
result = dns_db_findext(db, client->query.qname, dbversion->version,
qtype, DNS_DBFIND_NOZONECUT, client->inner.now,
&node, found, &cm, &ci, &trdataset, NULL);
if (result == DNS_R_NXRRSET || result == DNS_R_NCACHENXRRSET) {
if (dns_rdataset_isassociated(rdataset)) {
dns_rdataset_disassociate(rdataset);
}
if (dns_rdataset_isassociated(&trdataset)) {
dns_rdataset_disassociate(&trdataset);
}
goto nxrrset;
} else if (result != ISC_R_SUCCESS) {
if (dns_rdataset_isassociated(&trdataset)) {
dns_rdataset_disassociate(&trdataset);
}
if (node != NULL) {
dns_db_detachnode(db, &node);
}
dns_db_detach(&db);
return ISC_R_NOTFOUND;
}
CTRACE(ISC_LOG_DEBUG(3), "redirect: found data: done");
dns_name_copy(found, name);
if (dns_rdataset_isassociated(rdataset)) {
dns_rdataset_disassociate(rdataset);
}
if (dns_rdataset_isassociated(&trdataset)) {
dns_rdataset_clone(&trdataset, rdataset);
dns_rdataset_disassociate(&trdataset);
}
nxrrset:
if (*nodep != NULL) {
dns_db_detachnode(*dbp, nodep);
}
dns_db_detach(dbp);
dns_db_attachnode(db, node, nodep);
dns_db_attach(db, dbp);
dns_db_detachnode(db, &node);
dns_db_detach(&db);
*versionp = dbversion->version;
client->query.attributes |= (NS_QUERYATTR_NOAUTHORITY |
NS_QUERYATTR_NOADDITIONAL);
return result;
}
static isc_result_t
redirect2(ns_client_t *client, dns_name_t *name, dns_rdataset_t *rdataset,
dns_dbnode_t **nodep, dns_db_t **dbp, dns_dbversion_t **versionp,
dns_rdatatype_t qtype, bool *is_zonep) {
dns_db_t *db = NULL;
dns_dbnode_t *node = NULL;
dns_fixedname_t fixed;
dns_fixedname_t fixedredirect;
dns_name_t *found = NULL, *redirectname = NULL;
dns_rdataset_t trdataset = DNS_RDATASET_INIT;
isc_result_t result;
dns_rdatatype_t type;
dns_clientinfomethods_t cm;
dns_clientinfo_t ci;
dns_dbversion_t *version = NULL;
dns_zone_t *zone = NULL;
bool is_zone;
unsigned int labels;
CTRACE(ISC_LOG_DEBUG(3), "redirect2");
if (client->inner.view->redirectzone == NULL) {
return ISC_R_NOTFOUND;
}
if (dns_name_issubdomain(name, client->inner.view->redirectzone)) {
return ISC_R_NOTFOUND;
}
found = dns_fixedname_initname(&fixed);
dns_clientinfomethods_init(&cm, ns_client_sourceip);
dns_clientinfo_init(&ci, client, NULL);
dns_clientinfo_setecs(&ci, &client->inner.ecs);
if (WANTDNSSEC(client) && dns_db_iszone(*dbp) && dns_db_issecure(*dbp))
{
return ISC_R_NOTFOUND;
}
if (WANTDNSSEC(client) && dns_rdataset_isassociated(rdataset)) {
if (rdataset->trust == dns_trust_secure) {
return ISC_R_NOTFOUND;
}
if (rdataset->trust == dns_trust_ultimate &&
(rdataset->type == dns_rdatatype_nsec ||
rdataset->type == dns_rdatatype_nsec3))
{
return ISC_R_NOTFOUND;
}
if (rdataset->attributes.negative) {
DNS_RDATASET_FOREACH (rdataset) {
dns_ncache_current(rdataset, found, &trdataset);
type = trdataset.type;
dns_rdataset_disassociate(&trdataset);
if (type == dns_rdatatype_nsec ||
type == dns_rdatatype_nsec3 ||
type == dns_rdatatype_rrsig)
{
return ISC_R_NOTFOUND;
}
}
}
}
redirectname = dns_fixedname_initname(&fixedredirect);
labels = dns_name_countlabels(client->query.qname);
if (labels > 1U) {
dns_name_t prefix;
dns_name_init(&prefix);
dns_name_getlabelsequence(client->query.qname, 0, labels - 1,
&prefix);
result = dns_name_concatenate(&prefix,
client->inner.view->redirectzone,
redirectname);
if (result != ISC_R_SUCCESS) {
return ISC_R_NOTFOUND;
}
} else {
dns_name_copy(redirectname, client->inner.view->redirectzone);
}
result = query_getdb(client, redirectname, qtype,
(dns_getdb_options_t){ 0 }, &zone, &db, &version,
&is_zone);
if (result != ISC_R_SUCCESS) {
return ISC_R_NOTFOUND;
}
if (zone != NULL) {
dns_zone_detach(&zone);
}
/*
* Lookup the requested data in the redirect zone.
*/
result = dns_db_findext(db, redirectname, version, qtype, 0,
client->inner.now, &node, found, &cm, &ci,
&trdataset, NULL);
if (result == DNS_R_NXRRSET || result == DNS_R_NCACHENXRRSET) {
if (dns_rdataset_isassociated(rdataset)) {
dns_rdataset_disassociate(rdataset);
}
if (dns_rdataset_isassociated(&trdataset)) {
dns_rdataset_disassociate(&trdataset);
}
goto nxrrset;
} else if (result == ISC_R_NOTFOUND || result == DNS_R_DELEGATION) {
/*
* Cleanup.
*/
if (dns_rdataset_isassociated(&trdataset)) {
dns_rdataset_disassociate(&trdataset);
}
if (node != NULL) {
dns_db_detachnode(db, &node);
}
dns_db_detach(&db);
/*
* Don't loop forever if the lookup failed last time.
*/
if (!REDIRECT(client)) {
result = ns_query_recurse(client, qtype, redirectname,
NULL, NULL, true);
if (result == ISC_R_SUCCESS) {
client->query.attributes |=
NS_QUERYATTR_RECURSING;
client->query.attributes |=
NS_QUERYATTR_REDIRECT;
return DNS_R_CONTINUE;
}
}
return ISC_R_NOTFOUND;
} else if (result != ISC_R_SUCCESS) {
if (dns_rdataset_isassociated(&trdataset)) {
dns_rdataset_disassociate(&trdataset);
}
if (node != NULL) {
dns_db_detachnode(db, &node);
}
dns_db_detach(&db);
return ISC_R_NOTFOUND;
}
CTRACE(ISC_LOG_DEBUG(3), "redirect2: found data: done");
/*
* Adjust the found name to not include the redirectzone suffix.
*/
dns_name_split(found,
dns_name_countlabels(client->inner.view->redirectzone),
found, NULL);
/*
* Make the name absolute.
*/
result = dns_name_concatenate(found, dns_rootname, found);
RUNTIME_CHECK(result == ISC_R_SUCCESS);
dns_name_copy(found, name);
if (dns_rdataset_isassociated(rdataset)) {
dns_rdataset_disassociate(rdataset);
}
if (dns_rdataset_isassociated(&trdataset)) {
dns_rdataset_clone(&trdataset, rdataset);
dns_rdataset_disassociate(&trdataset);
}
nxrrset:
if (*nodep != NULL) {
dns_db_detachnode(*dbp, nodep);
}
dns_db_detach(dbp);
dns_db_attachnode(db, node, nodep);
dns_db_attach(db, dbp);
dns_db_detachnode(db, &node);
dns_db_detach(&db);
*is_zonep = is_zone;
*versionp = version;
client->query.attributes |= (NS_QUERYATTR_NOAUTHORITY |
NS_QUERYATTR_NOADDITIONAL);
return result;
}
/*%
* Initialize query context 'qctx'. Run by query_setup() when
* first handling a client query, and by query_resume() when
* returning from recursion.
*
* Whenever this function is called, qctx_destroy() must be called
* when leaving the scope or freeing the qctx.
*/
static void
qctx_init(ns_client_t *client, dns_fetchresponse_t **frespp,
dns_rdatatype_t qtype, query_ctx_t *qctx) {
REQUIRE(qctx != NULL);
REQUIRE(client != NULL);
memset(qctx, 0, sizeof(*qctx));
/* Set this first so CCTRACE will work */
qctx->client = client;
dns_view_attach(client->inner.view, &qctx->view);
CCTRACE(ISC_LOG_DEBUG(3), "qctx_init");
if (frespp != NULL) {
qctx->fresp = *frespp;
*frespp = NULL;
} else {
qctx->fresp = NULL;
}
qctx->qtype = qctx->type = qtype;
qctx->result = ISC_R_SUCCESS;
qctx->findcoveringnsec = qctx->view->synthfromdnssec;
/*
* If it's an RRSIG or SIG query, we'll iterate the node.
*/
if (dns_rdatatype_issig(qctx->qtype)) {
qctx->type = dns_rdatatype_any;
}
CALL_HOOK_NORETURN(NS_QUERY_QCTX_INITIALIZED, qctx);
}
/*%
* Clean up and disassociate the rdataset and node pointers in qctx.
*/
static void
qctx_clean(query_ctx_t *qctx) {
if (qctx->rdataset != NULL && dns_rdataset_isassociated(qctx->rdataset))
{
dns_rdataset_disassociate(qctx->rdataset);
}
if (qctx->sigrdataset != NULL &&
dns_rdataset_isassociated(qctx->sigrdataset))
{
dns_rdataset_disassociate(qctx->sigrdataset);
}
if (qctx->db != NULL && qctx->node != NULL) {
dns_db_detachnode(qctx->db, &qctx->node);
}
if (qctx->client != NULL && qctx->client->query.gluedb != NULL) {
dns_db_detach(&qctx->client->query.gluedb);
}
}
/*%
* Free any allocated memory associated with qctx.
*/
static void
qctx_freedata(query_ctx_t *qctx) {
if (qctx->rdataset != NULL) {
ns_client_putrdataset(qctx->client, &qctx->rdataset);
}
if (qctx->sigrdataset != NULL) {
ns_client_putrdataset(qctx->client, &qctx->sigrdataset);
}
if (qctx->fname != NULL) {
ns_client_releasename(qctx->client, &qctx->fname);
}
if (qctx->db != NULL) {
INSIST(qctx->node == NULL);
dns_db_detach(&qctx->db);
}
if (qctx->zone != NULL) {
dns_zone_detach(&qctx->zone);
}
if (qctx->zdb != NULL) {
ns_client_putrdataset(qctx->client, &qctx->zsigrdataset);
ns_client_putrdataset(qctx->client, &qctx->zrdataset);
ns_client_releasename(qctx->client, &qctx->zfname);
dns_db_detachnode(qctx->zdb, &qctx->znode);
dns_db_detach(&qctx->zdb);
qctx->zversion = NULL;
}
if (qctx->fresp != NULL) {
free_fresp(qctx->client, &qctx->fresp);
}
}
static void
qctx_destroy(query_ctx_t *qctx) {
CALL_HOOK_NORETURN(NS_QUERY_QCTX_DESTROYED, qctx);
dns_view_detach(&qctx->view);
}
/*
* Call SAVE but set 'a' to NULL first so as not to assert.
*/
#define INITANDSAVE(a, b) \
do { \
a = NULL; \
SAVE(a, b); \
} while (0)
/*
* "save" qctx data from 'src' to 'tgt'.
* It essentially moves ownership of the data from src to tgt, so the former
* becomes unusable except for final cleanup (such as by qctx_destroy).
* Note: this function doesn't attach to the client's handle. It's the caller's
* responsibility to do it if it's necessary.
*/
static void
qctx_save(query_ctx_t *src, query_ctx_t *tgt) {
/* First copy all fields in a straightforward way */
*tgt = *src;
/* Then "move" pointers (except client and view) */
INITANDSAVE(tgt->dbuf, src->dbuf);
INITANDSAVE(tgt->fname, src->fname);
INITANDSAVE(tgt->tname, src->tname);
INITANDSAVE(tgt->rdataset, src->rdataset);
INITANDSAVE(tgt->sigrdataset, src->sigrdataset);
INITANDSAVE(tgt->noqname, src->noqname);
INITANDSAVE(tgt->fresp, src->fresp);
INITANDSAVE(tgt->db, src->db);
INITANDSAVE(tgt->version, src->version);
INITANDSAVE(tgt->node, src->node);
INITANDSAVE(tgt->zdb, src->zdb);
INITANDSAVE(tgt->znode, src->znode);
INITANDSAVE(tgt->zfname, src->zfname);
INITANDSAVE(tgt->zversion, src->zversion);
INITANDSAVE(tgt->zrdataset, src->zrdataset);
INITANDSAVE(tgt->zsigrdataset, src->zsigrdataset);
INITANDSAVE(tgt->rpz_st, src->rpz_st);
INITANDSAVE(tgt->zone, src->zone);
/* View has to stay in 'src' for qctx_destroy. */
tgt->view = NULL;
dns_view_attach(src->view, &tgt->view);
}
/*%
* Log detailed information about the query immediately after
* the client request or a return from recursion.
*/
static void
query_trace(query_ctx_t *qctx) {
#ifdef WANT_QUERYTRACE
char mbuf[2 * DNS_NAME_FORMATSIZE];
char qbuf[DNS_NAME_FORMATSIZE];
if (qctx->client->query.origqname != NULL) {
dns_name_format(qctx->client->query.origqname, qbuf,
sizeof(qbuf));
} else {
snprintf(qbuf, sizeof(qbuf), "<unset>");
}
snprintf(mbuf, sizeof(mbuf) - 1,
"client attr:0x%x, query attr:0x%X, restarts:%u, "
"origqname:%s, timer:%d, authdb:%d, referral:%d",
qctx->client->inner.attributes, qctx->client->query.attributes,
qctx->client->query.restarts, qbuf,
(int)qctx->client->query.timerset,
(int)qctx->client->query.authdbset,
(int)qctx->client->query.isreferral);
CCTRACE(ISC_LOG_DEBUG(3), mbuf);
#else /* ifdef WANT_QUERYTRACE */
UNUSED(qctx);
#endif /* ifdef WANT_QUERYTRACE */
}
/*
* Set up query processing for the current query of 'client'.
* Calls qctx_init() to initialize a query context, checks
* the SERVFAIL cache, then hands off processing to ns__query_start().
*
* This is called only from ns_query_start(), to begin a query
* for the first time. Restarting an existing query (for
* instance, to handle CNAME lookups), is done by calling
* ns__query_start() again with the same query context. Resuming from
* recursion is handled by query_resume().
*/
static void
query_setup(ns_client_t *client, dns_rdatatype_t qtype) {
isc_result_t result = ISC_R_UNSET;
query_ctx_t qctx;
qctx_init(client, NULL, qtype, &qctx);
query_trace(&qctx);
CALL_HOOK(NS_QUERY_SETUP, &qctx);
/*
* Check SERVFAIL cache
*/
result = ns__query_sfcache(&qctx);
if (result != ISC_R_COMPLETE) {
goto cleanup;
}
(void)ns__query_start(&qctx);
cleanup:
qctx_destroy(&qctx);
}
static bool
get_root_key_sentinel_id(query_ctx_t *qctx, const char *ndata) {
unsigned int v = 0;
int i;
for (i = 0; i < 5; i++) {
if (!isdigit((unsigned char)ndata[i])) {
return false;
}
v *= 10;
v += ndata[i] - '0';
}
if (v > 65535U) {
return false;
}
qctx->client->query.root_key_sentinel_keyid = v;
return true;
}
/*%
* Find out if the query is for a root key sentinel and if so, record the type
* of root key sentinel query and the key id that is being checked for.
*
* The code is assuming a zero padded decimal field of width 5.
*/
static void
root_key_sentinel_detect(query_ctx_t *qctx) {
const char *ndata = (const char *)qctx->client->query.qname->ndata;
if (qctx->client->query.qname->length > 30 && ndata[0] == 29 &&
strncasecmp(ndata + 1, "root-key-sentinel-is-ta-", 24) == 0)
{
if (!get_root_key_sentinel_id(qctx, ndata + 25)) {
return;
}
qctx->client->query.root_key_sentinel_is_ta = true;
/*
* Simplify processing by disabling aggressive
* negative caching.
*/
qctx->findcoveringnsec = false;
ns_client_log(qctx->client, NS_LOGCATEGORY_TAT,
NS_LOGMODULE_QUERY, ISC_LOG_INFO,
"root-key-sentinel-is-ta query label found");
} else if (qctx->client->query.qname->length > 31 && ndata[0] == 30 &&
strncasecmp(ndata + 1, "root-key-sentinel-not-ta-", 25) == 0)
{
if (!get_root_key_sentinel_id(qctx, ndata + 26)) {
return;
}
qctx->client->query.root_key_sentinel_not_ta = true;
/*
* Simplify processing by disabling aggressive
* negative caching.
*/
qctx->findcoveringnsec = false;
ns_client_log(qctx->client, NS_LOGCATEGORY_TAT,
NS_LOGMODULE_QUERY, ISC_LOG_INFO,
"root-key-sentinel-not-ta query label found");
}
}
static void
qctx_reportquery(query_ctx_t *qctx) {
ns_client_t *client = qctx->client;
client->inner.attributes |= NS_CLIENTATTR_WANTRC;
/* If this isn't a report-logging zone, there's no more to do */
dns_zoneopt_t opts = dns_zone_getoptions(qctx->zone);
if ((opts & DNS_ZONEOPT_LOGREPORTS) == 0) {
return;
}
/*
* Suppress EDNS Report-Channel in responses from report-
* logging zones; this prevents infinite loops.
*/
client->inner.attributes &= ~NS_CLIENTATTR_WANTRC;
/* If this isn't an error-report query, there's nothing more to do */
if (client->query.qtype != dns_rdatatype_txt ||
!dns_name_israd(client->query.qname,
dns_zone_getorigin(qctx->zone)))
{
return;
}
/*
* Check for TCP or a good server cookie. If neither, send
* back BADCOOKIE or TC=1.
*/
if (!TCP(client) && !HAVECOOKIE(client)) {
if (WANTCOOKIE(client)) {
client->inner.attributes |= NS_CLIENTATTR_BADCOOKIE;
} else {
client->inner.attributes |= NS_CLIENTATTR_NEEDTCP;
}
}
if (isc_log_wouldlog(ISC_LOG_INFO)) {
char classbuf[DNS_RDATACLASS_FORMATSIZE];
char namebuf[DNS_NAME_FORMATSIZE];
dns_name_format(client->query.qname, namebuf, sizeof(namebuf));
dns_rdataclass_format(client->inner.view->rdclass, classbuf,
sizeof(classbuf));
isc_log_write(NS_LOGCATEGORY_DRA, NS_LOGMODULE_QUERY,
ISC_LOG_INFO, "dns-reporting-agent '%s/%s'",
namebuf, classbuf);
}
}
static void
qctx_setrad(query_ctx_t *qctx) {
ns_client_t *client = qctx->client;
/* Set the client to send a Report-Channel option when replying */
if ((client->inner.attributes & NS_CLIENTATTR_WANTRC) != 0) {
dns_fixedname_t fixed;
dns_name_t *rad = dns_fixedname_initname(&fixed);
if (!dns_name_dynamic(&client->inner.rad) &&
dns_zone_getrad(qctx->zone, rad) == ISC_R_SUCCESS)
{
dns_name_dup(rad, client->manager->mctx,
&client->inner.rad);
}
}
}
/*%
* Starting point for a client query or a chaining query.
*
* Called first by query_setup(), and then again as often as needed to
* follow a CNAME chain. Determines which authoritative database to
* search, then hands off processing to query_lookup().
*/
isc_result_t
ns__query_start(query_ctx_t *qctx) {
isc_result_t result = ISC_R_UNSET;
CCTRACE(ISC_LOG_DEBUG(3), "ns__query_start");
qctx->want_restart = false;
qctx->authoritative = false;
qctx->version = NULL;
qctx->zversion = NULL;
qctx->need_wildcardproof = false;
qctx->rpz = false;
CALL_HOOK(NS_QUERY_START_BEGIN, qctx);
/*
* If we require a server cookie or the presented server
* cookie was bad then send back BADCOOKIE before we have
* done too much work.
*/
if (!TCP(qctx->client) &&
(BADCOOKIE(qctx->client) ||
(qctx->view->requireservercookie && WANTCOOKIE(qctx->client) &&
!HAVECOOKIE(qctx->client))))
{
qctx->client->message->flags &= ~DNS_MESSAGEFLAG_AA;
qctx->client->message->flags &= ~DNS_MESSAGEFLAG_AD;
qctx->client->message->rcode = dns_rcode_badcookie;
qctx->client->inner.attributes &= ~NS_CLIENTATTR_WANTRC;
return ns_query_done(qctx);
}
/*
* Respond with TC=1 if we need TCP for this request.
*/
if (!TCP(qctx->client) && NEEDTCP(qctx->client)) {
qctx->client->message->flags &= ~DNS_MESSAGEFLAG_AA;
qctx->client->message->flags &= ~DNS_MESSAGEFLAG_AD;
qctx->client->message->flags |= DNS_MESSAGEFLAG_TC;
return ns_query_done(qctx);
}
if (qctx->view->checknames &&
!dns_rdata_checkowner(qctx->client->query.qname,
qctx->client->message->rdclass, qctx->qtype,
false))
{
char namebuf[DNS_NAME_FORMATSIZE];
char typebuf[DNS_RDATATYPE_FORMATSIZE];
char classbuf[DNS_RDATACLASS_FORMATSIZE];
dns_name_format(qctx->client->query.qname, namebuf,
sizeof(namebuf));
dns_rdatatype_format(qctx->qtype, typebuf, sizeof(typebuf));
dns_rdataclass_format(qctx->client->message->rdclass, classbuf,
sizeof(classbuf));
ns_client_log(qctx->client, DNS_LOGCATEGORY_SECURITY,
NS_LOGMODULE_QUERY, ISC_LOG_ERROR,
"check-names failure %s/%s/%s", namebuf, typebuf,
classbuf);
QUERY_ERROR(qctx, DNS_R_REFUSED);
return ns_query_done(qctx);
}
/*
* Setup for root key sentinel processing.
*/
if (qctx->view->root_key_sentinel &&
qctx->client->query.restarts == 0 &&
(dns_rdatatype_isaddr(qctx->qtype)) &&
(qctx->client->message->flags & DNS_MESSAGEFLAG_CD) == 0)
{
root_key_sentinel_detect(qctx);
}
/*
* First we must find the right database. Reset the options but preserve
* the 'nolog' flag.
*/
qctx->options = (dns_getdb_options_t){ .nolog = qctx->options.nolog };
if (dns_rdatatype_atparent(qctx->qtype) &&
!dns_name_equal(qctx->client->query.qname, dns_rootname))
{
/*
* If authoritative data for this QTYPE is supposed to live in
* the parent zone, do not look for an exact match for QNAME,
* but rather for its containing zone (unless the QNAME is
* root).
*/
qctx->options.noexact = true;
}
result = query_getdb(qctx->client, qctx->client->query.qname,
qctx->qtype, qctx->options, &qctx->zone, &qctx->db,
&qctx->version, &qctx->is_zone);
if ((result != ISC_R_SUCCESS || !qctx->is_zone) &&
qctx->qtype == dns_rdatatype_ds && !RECURSIONOK(qctx->client) &&
qctx->options.noexact)
{
/*
* This is a non-recursive QTYPE=DS query with QNAME whose
* parent we are not authoritative for. Check whether we are
* authoritative for QNAME, because if so, we need to send a
* "no data" response as required by RFC 4035, section 3.1.4.1.
*/
dns_db_t *tdb = NULL;
dns_zone_t *tzone = NULL;
dns_dbversion_t *tversion = NULL;
isc_result_t tresult;
dns_getdb_options_t options = { .partial = true };
tresult = query_getzonedb(
qctx->client, qctx->client->query.qname, qctx->qtype,
options, &tzone, &tdb, &tversion);
if (tresult == ISC_R_SUCCESS) {
/*
* We are authoritative for QNAME. Attach the relevant
* zone to query context, set result to ISC_R_SUCCESS.
*/
qctx->options.noexact = false;
ns_client_putrdataset(qctx->client, &qctx->rdataset);
if (qctx->db != NULL) {
dns_db_detach(&qctx->db);
}
if (qctx->zone != NULL) {
dns_zone_detach(&qctx->zone);
}
qctx->version = NULL;
RESTORE(qctx->version, tversion);
RESTORE(qctx->db, tdb);
RESTORE(qctx->zone, tzone);
qctx->is_zone = true;
result = ISC_R_SUCCESS;
} else {
/*
* We are not authoritative for QNAME. Clean up and
* leave result as it was.
*/
if (tdb != NULL) {
dns_db_detach(&tdb);
}
if (tzone != NULL) {
dns_zone_detach(&tzone);
}
}
}
/*
* If we did not find a database from which we can answer the query,
* respond with either REFUSED or SERVFAIL, depending on what the
* result of query_getdb() was.
*/
if (result != ISC_R_SUCCESS) {
if (result == DNS_R_REFUSED) {
if (WANTRECURSION(qctx->client)) {
dns_ede_add(&qctx->client->edectx,
DNS_EDE_NOTAUTH,
"recursion disabled");
inc_stats(qctx->client,
ns_statscounter_recurserej);
} else {
inc_stats(qctx->client,
ns_statscounter_authrej);
}
if (!PARTIALANSWER(qctx->client)) {
QUERY_ERROR(qctx, DNS_R_REFUSED);
}
} else {
CCTRACE(ISC_LOG_ERROR, "ns__query_start: query_getdb "
"failed");
QUERY_ERROR(qctx, result);
}
return ns_query_done(qctx);
}
/*
* We found a database from which we can answer the query. Update
* relevant query context flags if the answer is to be prepared using
* authoritative data.
*/
qctx->is_staticstub_zone = false;
if (qctx->is_zone) {
qctx->authoritative = true;
if (qctx->zone != NULL) {
switch (dns_zone_gettype(qctx->zone)) {
case dns_zone_mirror:
qctx->authoritative = false;
break;
case dns_zone_staticstub:
qctx->is_staticstub_zone = true;
break;
case dns_zone_primary:
case dns_zone_secondary:
qctx_reportquery(qctx);
qctx_setrad(qctx);
break;
default:
break;
}
}
}
/*
* Attach to the database which will be used to prepare the answer.
* Update query statistics.
*/
if (qctx->fresp == NULL && qctx->client->query.restarts == 0) {
if (qctx->is_zone) {
if (qctx->zone != NULL) {
/*
* if is_zone = true, zone = NULL then this is
* a DLZ zone. Don't attempt to attach zone.
*/
dns_zone_attach(qctx->zone,
&qctx->client->query.authzone);
}
dns_db_attach(qctx->db, &qctx->client->query.authdb);
}
qctx->client->query.authdbset = true;
isc_nmhandle_t *handle = qctx->client->inner.handle;
/* Track protocol stats per zone */
switch (isc_nm_socket_type(handle)) {
case isc_nm_httpsocket:
switch (isc_nmhandle_proxy_type(handle)) {
case ISC_NM_PROXY_ENCRYPTED:
/* Encrypted PROXYv2 cannot carry plain DoH */
INSIST(isc_nm_has_encryption(handle));
inc_stats(qctx->client,
ns_statscounter_encryptedproxydoh);
break;
case ISC_NM_PROXY_PLAIN:
if (isc_nm_has_encryption(handle)) {
inc_stats(qctx->client,
ns_statscounter_proxydoh);
} else {
inc_stats(
qctx->client,
ns_statscounter_proxydohplain);
}
break;
case ISC_NM_PROXY_NONE:
if (isc_nm_has_encryption(handle)) {
inc_stats(qctx->client,
ns_statscounter_doh);
} else {
inc_stats(qctx->client,
ns_statscounter_dohplain);
}
break;
}
break;
case isc_nm_streamdnssocket:
switch (isc_nmhandle_proxy_type(handle)) {
case ISC_NM_PROXY_ENCRYPTED:
inc_stats(qctx->client,
ns_statscounter_encryptedproxydot);
break;
case ISC_NM_PROXY_PLAIN:
if (isc_nm_has_encryption(handle)) {
inc_stats(qctx->client,
ns_statscounter_proxydot);
} else {
/*
* If the StreamDNS socket doesn't have
* encryption, it has to be plain TCP
* DNS.
*/
inc_stats(qctx->client,
ns_statscounter_proxytcp);
}
break;
case ISC_NM_PROXY_NONE:
if (isc_nm_has_encryption(handle)) {
inc_stats(qctx->client,
ns_statscounter_dot);
} else {
/*
* If the StreamDNS socket doesn't have
* encryption, it has to be plain TCP
* DNS.
*/
inc_stats(qctx->client,
ns_statscounter_tcp);
}
break;
}
break;
case isc_nm_proxyudpsocket:
inc_stats(qctx->client, ns_statscounter_proxyudp);
break;
case isc_nm_udpsocket:
inc_stats(qctx->client, ns_statscounter_udp);
break;
default:
UNREACHABLE();
}
}
if (!qctx->is_zone && qctx->view->staleanswerclienttimeout == 0 &&
dns_view_staleanswerenabled(qctx->view))
{
/*
* If stale answers are enabled and
* stale-answer-client-timeout is zero, then we can promptly
* answer with a stale RRset if one is available in cache.
*/
qctx->options.stalefirst = true;
}
result = query_lookup(qctx);
/*
* Clear "look-also-for-stale-data" flag.
* If a fetch is created to resolve this query, then,
* when it completes, this option is not expected to be set.
*/
qctx->options.stalefirst = false;
cleanup:
return result;
}
static void
async_restart(void *arg) {
query_ctx_t *qctx = arg;
ns_client_t *client = qctx->client;
isc_nmhandle_t *handle = client->inner.restarthandle;
client->inner.restarthandle = NULL;
ns__query_start(qctx);
qctx_clean(qctx);
qctx_freedata(qctx);
qctx_destroy(qctx);
isc_mem_put(client->manager->mctx, qctx, sizeof(*qctx));
isc_nmhandle_detach(&handle);
}
/*
* Allocate buffers in 'qctx' used to store query results.
*
* 'buffer' must be a pointer to an object whose lifetime
* doesn't expire while 'qctx' is in use.
*/
static isc_result_t
qctx_prepare_buffers(query_ctx_t *qctx, isc_buffer_t *buffer) {
REQUIRE(qctx != NULL);
REQUIRE(qctx->client != NULL);
REQUIRE(buffer != NULL);
qctx->dbuf = ns_client_getnamebuf(qctx->client);
qctx->fname = ns_client_newname(qctx->client, qctx->dbuf, buffer);
qctx->rdataset = ns_client_newrdataset(qctx->client);
if ((WANTDNSSEC(qctx->client) || qctx->findcoveringnsec) &&
(!qctx->is_zone || dns_db_issecure(qctx->db)))
{
qctx->sigrdataset = ns_client_newrdataset(qctx->client);
}
return ISC_R_SUCCESS;
}
/*%
* Depending on the db lookup result, we can respond to the
* client this stale answer.
*/
static bool
stale_client_answer(isc_result_t result) {
switch (result) {
case ISC_R_SUCCESS:
case DNS_R_EMPTYNAME:
case DNS_R_NXRRSET:
case DNS_R_NCACHENXRRSET:
case DNS_R_CNAME:
case DNS_R_DNAME:
return true;
default:
return false;
}
UNREACHABLE();
}
/*%
* Perform a local database lookup, in either an authoritative or
* cache database. If unable to answer, call ns_query_done(); otherwise
* hand off processing to query_gotanswer().
*/
static isc_result_t
query_lookup(query_ctx_t *qctx) {
isc_buffer_t buffer;
isc_result_t result = ISC_R_UNSET;
dns_clientinfomethods_t cm;
dns_clientinfo_t ci;
dns_name_t *rpzqname = NULL;
char namebuf[DNS_NAME_FORMATSIZE];
char typebuf[DNS_RDATATYPE_FORMATSIZE];
unsigned int dboptions;
dns_ttl_t stale_refresh = 0;
bool dbfind_stale = false;
bool stale_timeout = false;
bool answer_found = false;
bool stale_found = false;
bool stale_refresh_window = false;
uint16_t ede = 0;
CCTRACE(ISC_LOG_DEBUG(3), "query_lookup");
CALL_HOOK(NS_QUERY_LOOKUP_BEGIN, qctx);
dns_clientinfomethods_init(&cm, ns_client_sourceip);
dns_clientinfo_init(&ci, qctx->client, NULL);
if (HAVEECS(qctx->client)) {
dns_clientinfo_setecs(&ci, &qctx->client->inner.ecs);
}
/*
* We'll need some resources...
*/
result = qctx_prepare_buffers(qctx, &buffer);
if (result != ISC_R_SUCCESS) {
QUERY_ERROR(qctx, result);
return ns_query_done(qctx);
}
/*
* Now look for an answer in the database.
*/
if (qctx->dns64 && qctx->rpz) {
rpzqname = qctx->client->query.rpz_st->p_name;
} else {
rpzqname = qctx->client->query.qname;
}
if (qctx->options.stalefirst) {
/*
* If the 'stalefirst' flag is set, it means that a stale
* RRset may be returned as part of this lookup. An attempt
* to refresh the RRset will still take place if an
* active RRset is not available.
*/
qctx->client->query.dboptions |= DNS_DBFIND_STALETIMEOUT;
}
dboptions = qctx->client->query.dboptions;
if (!qctx->is_zone && qctx->findcoveringnsec &&
(qctx->type != dns_rdatatype_null || !dns_name_istat(rpzqname)))
{
dboptions |= DNS_DBFIND_COVERINGNSEC;
}
(void)dns_db_getservestalerefresh(qctx->client->inner.view->cachedb,
&stale_refresh);
if (stale_refresh > 0 &&
dns_view_staleanswerenabled(qctx->client->inner.view))
{
dboptions |= DNS_DBFIND_STALEENABLED;
}
result = dns_db_findext(qctx->db, rpzqname, qctx->version, qctx->type,
dboptions, qctx->client->inner.now, &qctx->node,
qctx->fname, &cm, &ci, qctx->rdataset,
qctx->sigrdataset);
/*
* Fixup fname and sigrdataset.
*/
if (qctx->dns64 && qctx->rpz) {
dns_name_copy(qctx->client->query.qname, qctx->fname);
if (qctx->sigrdataset != NULL &&
dns_rdataset_isassociated(qctx->sigrdataset))
{
dns_rdataset_disassociate(qctx->sigrdataset);
}
}
if (!qctx->is_zone) {
dns_cache_updatestats(qctx->view->cache, result);
}
/*
* If DNS_DBFIND_STALEOK is set this means we are dealing with a
* lookup following a failed lookup and it is okay to serve a stale
* answer. This will (re)start the 'stale-refresh-time' window in
* rbtdb, tracking the last time the RRset lookup failed.
*/
dbfind_stale = ((dboptions & DNS_DBFIND_STALEOK) != 0);
/*
* If DNS_DBFIND_STALEENABLED is set, this may be a normal lookup, but
* we are allowed to immediately respond with a stale answer if the
* request is within the 'stale-refresh-time' window.
*/
stale_refresh_window = (STALE_WINDOW(qctx->rdataset) &&
(dboptions & DNS_DBFIND_STALEENABLED) != 0);
/*
* If DNS_DBFIND_STALETIMEOUT is set, a stale answer is requested.
* This can happen if 'stale-answer-client-timeout' is enabled.
*
* If a stale answer is found, send it to the client, and try to refresh
* the RRset.
*/
stale_timeout = ((dboptions & DNS_DBFIND_STALETIMEOUT) != 0);
if (dns_rdataset_isassociated(qctx->rdataset) &&
dns_rdataset_count(qctx->rdataset) > 0 && !STALE(qctx->rdataset))
{
/* Found non-stale usable rdataset. */
answer_found = true;
}
if (dbfind_stale || stale_refresh_window || stale_timeout) {
dns_name_format(qctx->client->query.qname, namebuf,
sizeof(namebuf));
dns_rdatatype_format(qctx->qtype, typebuf, sizeof(typebuf));
inc_stats(qctx->client, ns_statscounter_trystale);
if (dns_rdataset_isassociated(qctx->rdataset) &&
dns_rdataset_count(qctx->rdataset) > 0 &&
STALE(qctx->rdataset))
{
stale_found = true;
if (result == DNS_R_NCACHENXDOMAIN ||
result == DNS_R_NXDOMAIN)
{
ede = DNS_EDE_STALENXANSWER;
} else {
ede = DNS_EDE_STALEANSWER;
}
qctx->rdataset->ttl = qctx->view->staleanswerttl;
inc_stats(qctx->client, ns_statscounter_usedstale);
} else {
stale_found = false;
}
}
if (dbfind_stale) {
isc_log_write(NS_LOGCATEGORY_SERVE_STALE, NS_LOGMODULE_QUERY,
ISC_LOG_INFO,
"%s %s resolver failure, stale answer %s (%s)",
namebuf, typebuf,
stale_found ? "used" : "unavailable",
isc_result_totext(result));
if (stale_found) {
dns_ede_add(&qctx->client->edectx, ede,
"resolver failure");
} else if (!answer_found) {
/*
* Resolver failure, no stale data, nothing more we
* can do, return SERVFAIL.
*/
QUERY_ERROR(qctx, DNS_R_SERVFAIL);
return ns_query_done(qctx);
}
} else if (stale_refresh_window) {
/*
* A recent lookup failed, so during this time window we are
* allowed to return stale data immediately.
*/
isc_log_write(NS_LOGCATEGORY_SERVE_STALE, NS_LOGMODULE_QUERY,
ISC_LOG_INFO,
"%s %s query within stale refresh time, stale "
"answer %s (%s)",
namebuf, typebuf,
stale_found ? "used" : "unavailable",
isc_result_totext(result));
if (stale_found) {
dns_ede_add(&qctx->client->edectx, ede,
"query within stale refresh time window");
} else if (!answer_found) {
/*
* During the stale refresh window explicitly do not try
* to refresh the data, because a recent lookup failed.
*/
QUERY_ERROR(qctx, DNS_R_SERVFAIL);
return ns_query_done(qctx);
}
} else if (stale_timeout) {
if (qctx->options.stalefirst) {
/*
* If 'qctx->zdb' is set, this was a cache lookup after
* an authoritative lookup returned a delegation (in
* order to find a better answer). But we still can
* return without getting any usable answer here, as
* query_notfound() should handle it from here.
* Otherwise, if nothing useful was found in cache then
* recursively call query_lookup() again without the
* 'stalefirst' option set.
*/
if (!stale_found && !answer_found && qctx->zdb == NULL)
{
qctx_clean(qctx);
qctx_freedata(qctx);
dns_db_attach(qctx->client->inner.view->cachedb,
&qctx->db);
qctx->client->query.dboptions &=
~DNS_DBFIND_STALETIMEOUT;
qctx->options.stalefirst = false;
if (FETCH_RECTYPE_NORMAL(qctx->client) != NULL)
{
dns_resolver_destroyfetch(
&FETCH_RECTYPE_NORMAL(
qctx->client));
}
return query_lookup(qctx);
} else if (stale_client_answer(result)) {
/*
* Immediately return the stale answer, start a
* resolver fetch to refresh the data in cache.
*/
isc_log_write(
NS_LOGCATEGORY_SERVE_STALE,
NS_LOGMODULE_QUERY, ISC_LOG_INFO,
"%s %s stale answer used, an attempt "
"to refresh the RRset will still be "
"made",
namebuf, typebuf);
qctx->refresh_rrset = STALE(qctx->rdataset);
if (stale_found) {
dns_ede_add(
&qctx->client->edectx, ede,
"stale data prioritized over "
"lookup");
}
}
} else {
UNREACHABLE();
}
}
if (stale_timeout && (answer_found || stale_found)) {
/*
* Mark RRsets that we are adding to the client message on a
* lookup during 'stale-answer-client-timeout', so we can
* clean it up if needed when we resume from recursion.
*/
qctx->client->query.attributes |= NS_QUERYATTR_STALEOK;
qctx->rdataset->attributes.stale_added = true;
}
result = query_gotanswer(qctx, result);
cleanup:
return result;
}
/*
* Clear all rdatasets from the message, or only those with stale_added
* attribute set.
*/
static void
message_clearrdataset(dns_message_t *msg, bool stale_only) {
unsigned int i;
/*
* Clean up name lists by calling the rdataset disassociate function.
*/
for (i = DNS_SECTION_ANSWER; i < DNS_SECTION_MAX; i++) {
ISC_LIST_FOREACH (msg->sections[i], name, link) {
ISC_LIST_FOREACH (name->list, rds, link) {
if (stale_only && !rds->attributes.stale_added)
{
continue;
}
ISC_LIST_UNLINK(name->list, rds, link);
INSIST(dns_rdataset_isassociated(rds));
dns_rdataset_disassociate(rds);
isc_mempool_put(msg->rdspool, rds);
}
if (ISC_LIST_EMPTY(name->list)) {
ISC_LIST_UNLINK(msg->sections[i], name, link);
if (dns_name_dynamic(name)) {
dns_name_free(name, msg->mctx);
}
isc_mempool_put(msg->namepool, name);
}
}
}
}
/*
* Event handler to resume processing a query after recursion, or when a
* client timeout is triggered. If the query has timed out or been cancelled
* or the system is shutting down, clean up and exit. If a client timeout is
* triggered, see if we can respond with a stale answer from cache. Otherwise,
* call query_resume() to continue the ongoing work.
*/
static void
fetch_callback(void *arg) {
dns_fetchresponse_t *resp = (dns_fetchresponse_t *)arg;
ns_client_t *client = resp->arg;
dns_fetch_t *fetch = NULL;
bool fetch_canceled = false;
isc_logcategory_t logcategory = NS_LOGCATEGORY_QUERY_ERRORS;
isc_result_t result;
int errorloglevel;
query_ctx_t qctx;
REQUIRE(NS_CLIENT_VALID(client));
REQUIRE(RECURSING(client));
CTRACE(ISC_LOG_DEBUG(3), "fetch_callback");
/*
* We are resuming from recursion. Reset any attributes, options
* that a lookup due to stale-answer-client-timeout may have set.
*/
if (client->inner.view->cachedb != NULL &&
client->inner.view->recursion)
{
client->query.attributes |= NS_QUERYATTR_RECURSIONOK;
}
client->query.dboptions &= ~DNS_DBFIND_STALETIMEOUT;
LOCK(&client->query.fetchlock);
INSIST(FETCH_RECTYPE_NORMAL(client) == resp->fetch ||
FETCH_RECTYPE_NORMAL(client) == NULL);
if (FETCH_RECTYPE_NORMAL(client) != NULL) {
/*
* This is the fetch we've been waiting for.
*/
INSIST(FETCH_RECTYPE_NORMAL(client) == resp->fetch);
FETCH_RECTYPE_NORMAL(client) = NULL;
/*
* Update client->now.
*/
client->inner.now = isc_stdtime_now();
} else {
/*
* This is a fetch completion event for a canceled fetch.
* Clean up and don't resume the find.
*/
fetch_canceled = true;
}
UNLOCK(&client->query.fetchlock);
SAVE(fetch, resp->fetch);
/*
* We're done recursing, detach from quota and unlink from
* the manager's recursing-clients list.
*/
release_recursionquota(client);
isc_nmhandle_detach(&HANDLE_RECTYPE_NORMAL(client));
client->query.attributes &= ~NS_QUERYATTR_RECURSING;
client->inner.state = NS_CLIENTSTATE_WORKING;
/*
* Initialize a new qctx and use it to either resume from
* recursion or clean up after cancelation. Transfer
* ownership of resp to the new qctx in the process.
*/
qctx_init(client, &resp, 0, &qctx);
if (fetch_canceled) {
/*
* We've timed out or are shutting down. We can now
* free the event and other resources held by qctx, but
* don't call qctx_destroy() yet: it might destroy the
* client, which we still need for a moment.
*/
qctx_freedata(&qctx);
/*
* Return an error to the client.
*/
CTRACE(ISC_LOG_ERROR, "fetch cancelled");
query_error(client, DNS_R_SERVFAIL, __LINE__);
/*
* Free any persistent plugin data that was allocated to
* service the client, then detach the client object.
*/
qctx.detach_client = true;
qctx_destroy(&qctx);
} else {
/*
* Resume the find process.
*/
query_trace(&qctx);
result = query_resume(&qctx);
if (result != ISC_R_SUCCESS) {
if (result == DNS_R_SERVFAIL) {
errorloglevel = ISC_LOG_DEBUG(2);
} else {
errorloglevel = ISC_LOG_DEBUG(4);
}
if (isc_log_wouldlog(errorloglevel)) {
dns_resolver_logfetch(fetch, logcategory,
NS_LOGMODULE_QUERY,
errorloglevel, false);
}
}
qctx_destroy(&qctx);
}
dns_resolver_destroyfetch(&fetch);
}
/*%
* Check whether the recursion parameters in 'param' match the current query's
* recursion parameters provided in 'qtype', 'qname', and 'qdomain'.
*/
static bool
recparam_match(const ns_query_recparam_t *param, dns_rdatatype_t qtype,
const dns_name_t *qname, const dns_name_t *qdomain) {
REQUIRE(param != NULL);
return param->qtype == qtype && param->qname != NULL && qname != NULL &&
param->qdomain != NULL && qdomain != NULL &&
dns_name_equal(param->qname, qname) &&
dns_name_equal(param->qdomain, qdomain);
}
/*%
* Update 'param' with current query's recursion parameters provided in
* 'qtype', 'qname', and 'qdomain'.
*/
static void
recparam_update(ns_query_recparam_t *param, dns_rdatatype_t qtype,
const dns_name_t *qname, const dns_name_t *qdomain) {
REQUIRE(param != NULL);
param->qtype = qtype;
if (qname == NULL) {
param->qname = NULL;
} else {
param->qname = dns_fixedname_initname(&param->fqname);
dns_name_copy(qname, param->qname);
}
if (qdomain == NULL) {
param->qdomain = NULL;
} else {
param->qdomain = dns_fixedname_initname(&param->fqdomain);
dns_name_copy(qdomain, param->qdomain);
}
}
static void
recursionquota_log(ns_client_t *client, atomic_uint_fast32_t *last_log_time,
const char *format, isc_quota_t *quota) {
isc_stdtime_t now = isc_stdtime_now();
if (now == atomic_load_relaxed(last_log_time)) {
return;
}
atomic_store_relaxed(last_log_time, now);
ns_client_log(client, NS_LOGCATEGORY_CLIENT, NS_LOGMODULE_QUERY,
ISC_LOG_WARNING, format, isc_quota_getused(quota),
isc_quota_getsoft(quota), isc_quota_getmax(quota));
}
static atomic_uint_fast32_t last_soft, last_hard;
/*%
* Acquire recursion quota before making the current client "recursing".
*/
static isc_result_t
acquire_recursionquota(ns_client_t *client) {
isc_result_t result;
result = recursionquotatype_attach_soft(client);
switch (result) {
case ISC_R_SOFTQUOTA:
recursionquota_log(client, &last_soft,
"recursive-clients soft limit exceeded "
"(%u/%u/%u), aborting oldest query",
&client->manager->sctx->recursionquota);
ns_client_killoldestquery(client);
FALLTHROUGH;
case ISC_R_SUCCESS:
break;
case ISC_R_QUOTA:
recursionquota_log(client, &last_hard,
"no more recursive clients (%u/%u/%u)",
&client->manager->sctx->recursionquota);
ns_client_killoldestquery(client);
return result;
default:
UNREACHABLE();
}
dns_message_clonebuffer(client->message);
ns_client_recursing(client);
return ISC_R_SUCCESS;
}
/*%
* Release recursion quota and remove the client from the "recursing" list.
*/
static void
release_recursionquota(ns_client_t *client) {
recursionquotatype_detach(client);
LOCK(&client->manager->reclock);
if (ISC_LINK_LINKED(client, inner.rlink)) {
ISC_LIST_UNLINK(client->manager->recursing, client,
inner.rlink);
}
UNLOCK(&client->manager->reclock);
}
isc_result_t
ns_query_recurse(ns_client_t *client, dns_rdatatype_t qtype, dns_name_t *qname,
dns_name_t *qdomain, dns_rdataset_t *nameservers,
bool resuming) {
isc_result_t result;
dns_rdataset_t *rdataset, *sigrdataset;
isc_sockaddr_t *peeraddr = NULL;
CTRACE(ISC_LOG_DEBUG(3), "ns_query_recurse");
/*
* Check recursion parameters from the previous query to see if they
* match. If not, update recursion parameters and proceed.
*/
if (recparam_match(&client->query.recparam, qtype, qname, qdomain)) {
ns_client_log(client, NS_LOGCATEGORY_CLIENT, NS_LOGMODULE_QUERY,
ISC_LOG_INFO, "recursion loop detected");
return ISC_R_FAILURE;
}
recparam_update(&client->query.recparam, qtype, qname, qdomain);
if (!resuming) {
inc_stats(client, ns_statscounter_recursion);
}
result = acquire_recursionquota(client);
if (result != ISC_R_SUCCESS) {
return result;
}
/*
* Invoke the resolver.
*/
REQUIRE(nameservers == NULL || nameservers->type == dns_rdatatype_ns);
REQUIRE(FETCH_RECTYPE_NORMAL(client) == NULL);
rdataset = ns_client_newrdataset(client);
if (WANTDNSSEC(client)) {
sigrdataset = ns_client_newrdataset(client);
} else {
sigrdataset = NULL;
}
if (!client->query.timerset) {
ns_client_settimeout(client, 60);
}
if (!TCP(client)) {
peeraddr = &client->inner.peeraddr;
}
isc_nmhandle_attach(client->inner.handle,
&HANDLE_RECTYPE_NORMAL(client));
result = dns_resolver_createfetch(
client->inner.view->resolver, qname, qtype, qdomain,
nameservers, NULL, peeraddr, client->message->id,
client->query.fetchoptions, 0, NULL, client->query.qc,
client->manager->loop, fetch_callback, client, &client->edectx,
rdataset, sigrdataset, &FETCH_RECTYPE_NORMAL(client));
if (result != ISC_R_SUCCESS) {
release_recursionquota(client);
ns_client_putrdataset(client, &rdataset);
if (sigrdataset != NULL) {
ns_client_putrdataset(client, &sigrdataset);
}
isc_nmhandle_detach(&HANDLE_RECTYPE_NORMAL(client));
}
/*
* We're now waiting for a fetch event. A client which is
* shutting down will not be destroyed until all the events
* have been received.
*/
return result;
}
/*%
* Restores the query context after resuming from recursion, and
* continues the query processing if needed.
*/
static isc_result_t
query_resume(query_ctx_t *qctx) {
isc_result_t result = ISC_R_UNSET;
dns_name_t *tname;
isc_buffer_t b;
#ifdef WANT_QUERYTRACE
char mbuf[4 * DNS_NAME_FORMATSIZE];
char qbuf[DNS_NAME_FORMATSIZE];
char tbuf[DNS_RDATATYPE_FORMATSIZE];
#endif /* ifdef WANT_QUERYTRACE */
CCTRACE(ISC_LOG_DEBUG(3), "query_resume");
CALL_HOOK(NS_QUERY_RESUME_BEGIN, qctx);
qctx->want_restart = false;
qctx->rpz_st = qctx->client->query.rpz_st;
if (qctx->rpz_st != NULL &&
(qctx->rpz_st->state & DNS_RPZ_RECURSING) != 0)
{
CCTRACE(ISC_LOG_DEBUG(3), "resume from RPZ recursion");
#ifdef WANT_QUERYTRACE
{
char pbuf[DNS_NAME_FORMATSIZE] = "<unset>";
char fbuf[DNS_NAME_FORMATSIZE] = "<unset>";
if (qctx->rpz_st->r_name != NULL) {
dns_name_format(qctx->rpz_st->r_name, qbuf,
sizeof(qbuf));
} else {
snprintf(qbuf, sizeof(qbuf), "<unset>");
}
if (qctx->rpz_st->p_name != NULL) {
dns_name_format(qctx->rpz_st->p_name, pbuf,
sizeof(pbuf));
}
if (qctx->rpz_st->fname != NULL) {
dns_name_format(qctx->rpz_st->fname, fbuf,
sizeof(fbuf));
}
snprintf(mbuf, sizeof(mbuf) - 1,
"rpz rname:%s, pname:%s, qctx->fname:%s", qbuf,
pbuf, fbuf);
CCTRACE(ISC_LOG_DEBUG(3), mbuf);
}
#endif /* ifdef WANT_QUERYTRACE */
qctx->is_zone = qctx->rpz_st->q.is_zone;
qctx->authoritative = qctx->rpz_st->q.authoritative;
RESTORE(qctx->zone, qctx->rpz_st->q.zone);
RESTORE(qctx->node, qctx->rpz_st->q.node);
RESTORE(qctx->db, qctx->rpz_st->q.db);
RESTORE(qctx->rdataset, qctx->rpz_st->q.rdataset);
RESTORE(qctx->sigrdataset, qctx->rpz_st->q.sigrdataset);
qctx->qtype = qctx->rpz_st->q.qtype;
if (qctx->fresp->node != NULL) {
dns_db_detachnode(qctx->fresp->db, &qctx->fresp->node);
}
SAVE(qctx->rpz_st->r.db, qctx->fresp->db);
qctx->rpz_st->r.r_type = qctx->fresp->qtype;
SAVE(qctx->rpz_st->r.r_rdataset, qctx->fresp->rdataset);
ns_client_putrdataset(qctx->client, &qctx->fresp->sigrdataset);
} else if (REDIRECT(qctx->client)) {
/*
* Restore saved state.
*/
CCTRACE(ISC_LOG_DEBUG(3), "resume from redirect recursion");
#ifdef WANT_QUERYTRACE
dns_name_format(qctx->client->query.redirect.fname, qbuf,
sizeof(qbuf));
dns_rdatatype_format(qctx->client->query.redirect.qtype, tbuf,
sizeof(tbuf));
snprintf(mbuf, sizeof(mbuf) - 1,
"redirect qctx->fname:%s, qtype:%s, auth:%d", qbuf,
tbuf, qctx->client->query.redirect.authoritative);
CCTRACE(ISC_LOG_DEBUG(3), mbuf);
#endif /* ifdef WANT_QUERYTRACE */
qctx->qtype = qctx->client->query.redirect.qtype;
INSIST(qctx->client->query.redirect.rdataset != NULL);
RESTORE(qctx->rdataset, qctx->client->query.redirect.rdataset);
RESTORE(qctx->sigrdataset,
qctx->client->query.redirect.sigrdataset);
RESTORE(qctx->db, qctx->client->query.redirect.db);
RESTORE(qctx->node, qctx->client->query.redirect.node);
RESTORE(qctx->zone, qctx->client->query.redirect.zone);
qctx->authoritative =
qctx->client->query.redirect.authoritative;
/*
* Free resources used while recursing.
*/
ns_client_putrdataset(qctx->client, &qctx->fresp->rdataset);
ns_client_putrdataset(qctx->client, &qctx->fresp->sigrdataset);
if (qctx->fresp->node != NULL) {
dns_db_detachnode(qctx->fresp->db, &qctx->fresp->node);
}
if (qctx->fresp->db != NULL) {
dns_db_detach(&qctx->fresp->db);
}
} else {
CCTRACE(ISC_LOG_DEBUG(3), "resume from normal recursion");
qctx->authoritative = false;
qctx->qtype = qctx->fresp->qtype;
SAVE(qctx->db, qctx->fresp->db);
SAVE(qctx->node, qctx->fresp->node);
SAVE(qctx->rdataset, qctx->fresp->rdataset);
SAVE(qctx->sigrdataset, qctx->fresp->sigrdataset);
}
INSIST(qctx->rdataset != NULL);
if (dns_rdatatype_issig(qctx->qtype)) {
qctx->type = dns_rdatatype_any;
} else {
qctx->type = qctx->qtype;
}
CALL_HOOK(NS_QUERY_RESUME_RESTORED, qctx);
if (DNS64(qctx->client)) {
qctx->client->query.attributes &= ~NS_QUERYATTR_DNS64;
qctx->dns64 = true;
}
if (DNS64EXCLUDE(qctx->client)) {
qctx->client->query.attributes &= ~NS_QUERYATTR_DNS64EXCLUDE;
qctx->dns64_exclude = true;
}
if (qctx->rpz_st != NULL &&
(qctx->rpz_st->state & DNS_RPZ_RECURSING) != 0)
{
/*
* Has response policy changed out from under us?
*/
if (qctx->view->rpzs == NULL ||
qctx->rpz_st->rpz_ver != qctx->view->rpzs->rpz_ver)
{
ns_client_log(qctx->client, NS_LOGCATEGORY_CLIENT,
NS_LOGMODULE_QUERY, DNS_RPZ_INFO_LEVEL,
"query_resume: RPZ settings out of date "
"after of a reconfiguration");
QUERY_ERROR(qctx, DNS_R_SERVFAIL);
return ns_query_done(qctx);
}
}
/*
* We'll need some resources...
*/
qctx->dbuf = ns_client_getnamebuf(qctx->client);
qctx->fname = ns_client_newname(qctx->client, qctx->dbuf, &b);
if (qctx->rpz_st != NULL &&
(qctx->rpz_st->state & DNS_RPZ_RECURSING) != 0)
{
tname = qctx->rpz_st->fname;
} else if (REDIRECT(qctx->client)) {
tname = qctx->client->query.redirect.fname;
} else {
tname = qctx->fresp->foundname;
}
dns_name_copy(tname, qctx->fname);
if (qctx->rpz_st != NULL &&
(qctx->rpz_st->state & DNS_RPZ_RECURSING) != 0)
{
qctx->rpz_st->r.r_result = qctx->fresp->result;
result = qctx->rpz_st->q.result;
free_fresp(qctx->client, &qctx->fresp);
} else if (REDIRECT(qctx->client)) {
result = qctx->client->query.redirect.result;
} else {
result = qctx->fresp->result;
}
qctx->resuming = true;
return query_gotanswer(qctx, result);
cleanup:
return result;
}
static void
query_hookresume(void *arg) {
ns_hook_resume_t *rev = (ns_hook_resume_t *)arg;
ns_hookasync_t *hctx = NULL;
ns_client_t *client = rev->arg;
query_ctx_t *qctx = rev->saved_qctx;
bool canceled;
CTRACE(ISC_LOG_DEBUG(3), "query_hookresume");
REQUIRE(NS_CLIENT_VALID(client));
LOCK(&client->query.fetchlock);
if (client->query.hookactx != NULL) {
INSIST(rev->ctx == client->query.hookactx);
client->query.hookactx = NULL;
canceled = false;
client->inner.now = isc_stdtime_now();
} else {
canceled = true;
}
UNLOCK(&client->query.fetchlock);
SAVE(hctx, rev->ctx);
release_recursionquota(client);
/*
* The fetch handle should be detached before resuming query processing
* below, since that may trigger another recursion or asynchronous hook
* event.
*/
isc_nmhandle_detach(&HANDLE_RECTYPE_HOOK(client));
client->inner.state = NS_CLIENTSTATE_WORKING;
if (canceled) {
/*
* Note: unlike fetch_callback, this function doesn't bother
* to check the 'shutdown' condition, as that doesn't seem to
* happen in the latest implementation.
*/
query_error(client, DNS_R_SERVFAIL, __LINE__);
/*
* There's no other place to free/release any data maintained
* in qctx. We need to do it here to prevent leak.
*/
qctx_clean(qctx);
qctx_freedata(qctx);
/*
* As we're almost done with this client, make sure any internal
* resource for hooks will be released (if necessary) via the
* QCTX_DESTROYED hook.
*/
qctx->detach_client = true;
} else {
switch (rev->hookpoint) {
case NS_QUERY_SETUP:
query_setup(client, qctx->qtype);
break;
case NS_QUERY_START_BEGIN:
(void)ns__query_start(qctx);
break;
case NS_QUERY_LOOKUP_BEGIN:
(void)query_lookup(qctx);
break;
case NS_QUERY_RESUME_BEGIN:
case NS_QUERY_RESUME_RESTORED:
(void)query_resume(qctx);
break;
case NS_QUERY_GOT_ANSWER_BEGIN:
(void)query_gotanswer(qctx, rev->origresult);
break;
case NS_QUERY_RESPOND_ANY_BEGIN:
(void)query_respond_any(qctx);
break;
case NS_QUERY_ADDANSWER_BEGIN:
(void)query_addanswer(qctx);
break;
case NS_QUERY_NOTFOUND_BEGIN:
(void)query_notfound(qctx);
break;
case NS_QUERY_PREP_DELEGATION_BEGIN:
(void)query_prepare_delegation_response(qctx);
break;
case NS_QUERY_ZONE_DELEGATION_BEGIN:
(void)query_zone_delegation(qctx);
break;
case NS_QUERY_DELEGATION_BEGIN:
(void)query_delegation(qctx);
break;
case NS_QUERY_DELEGATION_RECURSE_BEGIN:
(void)query_delegation_recurse(qctx);
break;
case NS_QUERY_NODATA_BEGIN:
(void)query_nodata(qctx, rev->origresult);
break;
case NS_QUERY_NXDOMAIN_BEGIN:
(void)query_nxdomain(qctx, rev->origresult);
break;
case NS_QUERY_NCACHE_BEGIN:
(void)query_ncache(qctx, rev->origresult);
break;
case NS_QUERY_CNAME_BEGIN:
(void)query_cname(qctx);
break;
case NS_QUERY_DNAME_BEGIN:
(void)query_dname(qctx);
break;
case NS_QUERY_RESPOND_BEGIN:
(void)query_respond(qctx);
break;
case NS_QUERY_PREP_RESPONSE_BEGIN:
(void)query_prepresponse(qctx);
break;
case NS_QUERY_DONE_BEGIN:
case NS_QUERY_DONE_SEND:
(void)ns_query_done(qctx);
break;
/* Not all hookpoints can use recursion. Catch violations */
case NS_QUERY_RESPOND_ANY_FOUND: /* due to side effect */
case NS_QUERY_NOTFOUND_RECURSE: /* in recursion */
case NS_QUERY_ZEROTTL_RECURSE: /* in recursion */
default: /* catch-all just in case */
INSIST(false);
}
}
isc_mem_put(hctx->mctx, rev, sizeof(*rev));
hctx->destroy(&hctx);
qctx_destroy(qctx);
isc_mem_put(client->manager->mctx, qctx, sizeof(*qctx));
}
isc_result_t
ns_query_hookasync(query_ctx_t *qctx, ns_query_starthookasync_t runasync,
void *arg) {
isc_result_t result;
ns_client_t *client = qctx->client;
query_ctx_t *saved_qctx = NULL;
CTRACE(ISC_LOG_DEBUG(3), "ns_query_hookasync");
REQUIRE(NS_CLIENT_VALID(client));
REQUIRE(client->query.hookactx == NULL);
REQUIRE(FETCH_RECTYPE_NORMAL(client) == NULL);
result = acquire_recursionquota(client);
if (result != ISC_R_SUCCESS) {
goto cleanup;
}
saved_qctx = isc_mem_get(client->manager->mctx, sizeof(*saved_qctx));
qctx_save(qctx, saved_qctx);
result = runasync(saved_qctx, client->manager->mctx, arg,
client->manager->loop, query_hookresume, client,
&client->query.hookactx);
if (result != ISC_R_SUCCESS) {
goto cleanup_and_detach_from_quota;
}
/* Record that an asynchronous copy of the qctx has been started */
qctx->async = true;
/*
* Typically the runasync() function will trigger recursion, but
* there is no need to set NS_QUERYATTR_RECURSING. The calling hook
* is expected to return NS_HOOK_RETURN, and the RECURSING
* attribute won't be checked anywhere.
*
* Hook-based asynchronous processing cannot coincide with normal
* recursion. Unlike in ns_query_recurse(), we attach to the handle
* only if 'runasync' succeeds. It should be safe since we're either in
* the client task or pausing it.
*/
isc_nmhandle_attach(client->inner.handle, &HANDLE_RECTYPE_HOOK(client));
return ISC_R_SUCCESS;
cleanup_and_detach_from_quota:
release_recursionquota(client);
cleanup:
/*
* If we fail, send SERVFAIL now. It may be better to let the caller
* decide what to do on failure of this function, but hooks don't have
* access to query_error().
*/
query_error(client, DNS_R_SERVFAIL, __LINE__);
/*
* Free all resource related to the query and set detach_client,
* similar to the cancel case of query_hookresume; the callers will
* simply return on failure of this function, so there's no other
* place for this to prevent leak.
*/
if (saved_qctx != NULL) {
qctx_clean(saved_qctx);
qctx_freedata(saved_qctx);
qctx_destroy(saved_qctx);
isc_mem_put(client->manager->mctx, saved_qctx,
sizeof(*saved_qctx));
}
qctx->detach_client = true;
return result;
}
/*%
* If the query is recursive, check the SERVFAIL cache to see whether
* identical queries have failed recently. If we find a match, and it was
* from a query with CD=1, *or* if the current query has CD=0, then we just
* return SERVFAIL again. This prevents a validation failure from eliciting a
* SERVFAIL response to a CD=1 query.
*/
isc_result_t
ns__query_sfcache(query_ctx_t *qctx) {
isc_result_t failcache;
uint32_t flags;
/*
* The SERVFAIL cache doesn't apply to authoritative queries.
*/
if (!RECURSIONOK(qctx->client)) {
return ISC_R_COMPLETE;
}
flags = 0;
#ifdef ENABLE_AFL
if (qctx->client->manager->sctx->fuzztype == isc_fuzz_resolver) {
failcache = ISC_R_NOTFOUND;
} else
#endif /* ifdef ENABLE_AFL */
{
failcache = dns_badcache_find(
qctx->view->failcache, qctx->client->query.qname,
qctx->qtype, &flags,
isc_time_seconds(&qctx->client->inner.tnow));
}
if (failcache != ISC_R_SUCCESS) {
return ISC_R_COMPLETE;
}
if (((flags & NS_FAILCACHE_CD) != 0) ||
((qctx->client->message->flags & DNS_MESSAGEFLAG_CD) == 0))
{
if (isc_log_wouldlog(ISC_LOG_DEBUG(1))) {
char namebuf[DNS_NAME_FORMATSIZE];
char typebuf[DNS_RDATATYPE_FORMATSIZE];
dns_name_format(qctx->client->query.qname, namebuf,
sizeof(namebuf));
dns_rdatatype_format(qctx->qtype, typebuf,
sizeof(typebuf));
ns_client_log(qctx->client, NS_LOGCATEGORY_CLIENT,
NS_LOGMODULE_QUERY, ISC_LOG_DEBUG(1),
"servfail cache hit %s/%s (%s)", namebuf,
typebuf,
((flags & NS_FAILCACHE_CD) != 0) ? "CD=1"
: "CD="
"0");
}
qctx->client->inner.attributes |= NS_CLIENTATTR_NOSETFC;
QUERY_ERROR(qctx, DNS_R_SERVFAIL);
return ns_query_done(qctx);
}
return ISC_R_COMPLETE;
}
static void
query_trace_rrldrop(query_ctx_t *qctx,
dns_rrl_result_t rrl_result ISC_ATTR_UNUSED) {
if (!LIBNS_RRL_DROP_ENABLED()) {
return;
}
char peerbuf[ISC_SOCKADDR_FORMATSIZE];
isc_netaddr_t peer;
isc_netaddr_fromsockaddr(&peer, &qctx->client->inner.peeraddr);
isc_netaddr_format(&peer, peerbuf, sizeof(peerbuf));
char qnamebuf[DNS_NAME_FORMATSIZE];
char fnamebuf[DNS_NAME_FORMATSIZE];
dns_name_format(qctx->client->query.qname, qnamebuf, sizeof(qnamebuf));
dns_name_format(qctx->fname, fnamebuf, sizeof(fnamebuf));
LIBNS_RRL_DROP(peerbuf, qnamebuf, fnamebuf, rrl_result);
}
/*%
* Handle response rate limiting (RRL).
*/
static isc_result_t
query_checkrrl(query_ctx_t *qctx, isc_result_t result) {
/*
* Rate limit these responses to this client.
* Do not delay counting and handling obvious referrals,
* since those won't come here again.
* Delay handling delegations for which we are certain to recurse and
* return here (DNS_R_DELEGATION, not a child of one of our
* own zones, and recursion enabled)
* Don't mess with responses rewritten by RPZ
* Count each response at most once.
*/
/*
* XXXMPA the rrl system tests fails sometimes and RRL_CHECKED
* is set when we are called the second time preventing the
* response being dropped.
*/
ns_client_log(
qctx->client, DNS_LOGCATEGORY_RRL, NS_LOGMODULE_QUERY,
ISC_LOG_DEBUG(99),
"rrl=%p, HAVECOOKIE=%u, result=%s, "
"fname=%p(%u), is_zone=%u, RECURSIONOK=%u, "
"query.rpz_st=%p(%u), RRL_CHECKED=%u",
qctx->client->inner.view->rrl, HAVECOOKIE(qctx->client),
isc_result_toid(result), qctx->fname,
qctx->fname != NULL ? dns_name_isabsolute(qctx->fname) : 0,
qctx->is_zone, RECURSIONOK(qctx->client),
qctx->client->query.rpz_st,
qctx->client->query.rpz_st != NULL
? ((qctx->client->query.rpz_st->state &
DNS_RPZ_REWRITTEN) != 0)
: 0,
(qctx->client->query.attributes & NS_QUERYATTR_RRL_CHECKED) !=
0);
if (qctx->view->rrl != NULL && !HAVECOOKIE(qctx->client) &&
((qctx->fname != NULL && dns_name_isabsolute(qctx->fname)) ||
(result == ISC_R_NOTFOUND && !RECURSIONOK(qctx->client))) &&
!(result == DNS_R_DELEGATION && !qctx->is_zone &&
RECURSIONOK(qctx->client)) &&
(qctx->client->query.rpz_st == NULL ||
(qctx->client->query.rpz_st->state & DNS_RPZ_REWRITTEN) == 0) &&
(qctx->client->query.attributes & NS_QUERYATTR_RRL_CHECKED) == 0)
{
bool wouldlog;
const dns_name_t *constname;
char log_buf[DNS_RRL_LOG_BUF_LEN];
isc_result_t resp_result;
dns_rrl_result_t rrl_result;
qctx->client->query.attributes |= NS_QUERYATTR_RRL_CHECKED;
wouldlog = isc_log_wouldlog(DNS_RRL_LOG_DROP);
constname = qctx->fname;
if (result == DNS_R_NXDOMAIN) {
/*
* Use the database origin name to rate limit NXDOMAIN
*/
if (qctx->db != NULL) {
constname = dns_db_origin(qctx->db);
}
resp_result = result;
} else if (result == DNS_R_NCACHENXDOMAIN &&
qctx->rdataset != NULL &&
dns_rdataset_isassociated(qctx->rdataset) &&
qctx->rdataset->attributes.negative)
{
/*
* Try to use owner name in the negative cache SOA.
*/
dns_rdataset_t nc_rdataset = DNS_RDATASET_INIT;
dns_fixedname_t fixed;
dns_name_t *n = dns_fixedname_initname(&fixed);
DNS_RDATASET_FOREACH (qctx->rdataset) {
dns_ncache_current(qctx->rdataset, n,
&nc_rdataset);
if (nc_rdataset.type == dns_rdatatype_soa) {
dns_rdataset_disassociate(&nc_rdataset);
constname = n;
break;
}
dns_rdataset_disassociate(&nc_rdataset);
}
resp_result = DNS_R_NXDOMAIN;
} else if (result == DNS_R_NXRRSET || result == DNS_R_EMPTYNAME)
{
resp_result = DNS_R_NXRRSET;
} else if (result == DNS_R_DELEGATION) {
resp_result = result;
} else if (result == ISC_R_NOTFOUND) {
/*
* Handle referral to ".", including when recursion
* is off or not requested and the hints have not
* been loaded.
*/
constname = dns_rootname;
resp_result = DNS_R_DELEGATION;
} else {
resp_result = ISC_R_SUCCESS;
}
rrl_result = dns_rrl(
qctx->view, qctx->zone, &qctx->client->inner.peeraddr,
TCP(qctx->client), qctx->client->message->rdclass,
qctx->qtype, constname, resp_result,
qctx->client->inner.now, wouldlog, log_buf,
sizeof(log_buf));
if (rrl_result != DNS_RRL_RESULT_OK) {
/*
* Log dropped or slipped responses in the query-errors
* category so that requests are not silently lost.
* Starts of rate-limited bursts are logged in
* DNS_LOGCATEGORY_RRL.
*
* Dropped responses are counted with dropped queries
* in QryDropped while slipped responses are counted
* with other truncated responses in RespTruncated.
*/
if (wouldlog) {
ns_client_log(qctx->client,
NS_LOGCATEGORY_QUERY_ERRORS,
NS_LOGMODULE_QUERY,
DNS_RRL_LOG_DROP, "%s", log_buf);
}
/*
* If tracing is enabled, format some extra information
* to pass along.
*/
query_trace_rrldrop(qctx, rrl_result);
if (!qctx->view->rrl->log_only) {
if (rrl_result == DNS_RRL_RESULT_DROP) {
/*
* These will also be counted in
* ns_statscounter_dropped
*/
inc_stats(qctx->client,
ns_statscounter_ratedropped);
QUERY_ERROR(qctx, DNS_R_DROP);
} else {
/*
* These will also be counted in
* ns_statscounter_truncatedresp
*/
inc_stats(qctx->client,
ns_statscounter_rateslipped);
if (WANTCOOKIE(qctx->client)) {
qctx->client->message->flags &=
~DNS_MESSAGEFLAG_AA;
qctx->client->message->flags &=
~DNS_MESSAGEFLAG_AD;
qctx->client->message->rcode =
dns_rcode_badcookie;
qctx->client->inner.attributes &=
~NS_CLIENTATTR_WANTRC;
} else {
qctx->client->message->flags |=
DNS_MESSAGEFLAG_TC;
if (resp_result ==
DNS_R_NXDOMAIN)
{
qctx->client->message
->rcode =
dns_rcode_nxdomain;
}
}
}
return DNS_R_DROP;
}
}
}
return ISC_R_SUCCESS;
}
/*%
* Do any RPZ rewriting that may be needed for this query.
*/
static isc_result_t
query_checkrpz(query_ctx_t *qctx, isc_result_t result) {
isc_result_t rresult;
CCTRACE(ISC_LOG_DEBUG(3), "query_checkrpz");
rresult = rpz_rewrite(qctx->client, qctx->qtype, result, qctx->resuming,
qctx->rdataset, qctx->sigrdataset);
qctx->rpz_st = qctx->client->query.rpz_st;
switch (rresult) {
case ISC_R_SUCCESS:
break;
case ISC_R_NOTFOUND:
case DNS_R_DISALLOWED:
return result;
case DNS_R_DELEGATION:
/*
* recursing for NS names or addresses,
* so save the main query state
*/
INSIST(!RECURSING(qctx->client));
qctx->rpz_st->q.qtype = qctx->qtype;
qctx->rpz_st->q.is_zone = qctx->is_zone;
qctx->rpz_st->q.authoritative = qctx->authoritative;
SAVE(qctx->rpz_st->q.zone, qctx->zone);
SAVE(qctx->rpz_st->q.db, qctx->db);
SAVE(qctx->rpz_st->q.node, qctx->node);
SAVE(qctx->rpz_st->q.rdataset, qctx->rdataset);
SAVE(qctx->rpz_st->q.sigrdataset, qctx->sigrdataset);
dns_name_copy(qctx->fname, qctx->rpz_st->fname);
qctx->rpz_st->q.result = result;
qctx->client->query.attributes |= NS_QUERYATTR_RECURSING;
return ISC_R_COMPLETE;
default:
QUERY_ERROR(qctx, rresult);
return ISC_R_COMPLETE;
}
if (qctx->rpz_st->m.policy != DNS_RPZ_POLICY_MISS) {
qctx->rpz_st->state |= DNS_RPZ_REWRITTEN;
}
if (qctx->rpz_st->m.policy != DNS_RPZ_POLICY_MISS &&
qctx->rpz_st->m.policy != DNS_RPZ_POLICY_PASSTHRU &&
(qctx->rpz_st->m.policy != DNS_RPZ_POLICY_TCP_ONLY ||
!TCP(qctx->client)) &&
qctx->rpz_st->m.policy != DNS_RPZ_POLICY_ERROR)
{
/*
* We got a hit and are going to answer with our
* fiction. Ensure that we answer with the name
* we looked up even if we were stopped short
* in recursion or for a deferral.
*/
dns_name_copy(qctx->client->query.qname, qctx->fname);
rpz_clean(&qctx->zone, &qctx->db, &qctx->node, NULL);
if (qctx->rpz_st->m.rdataset != NULL) {
ns_client_putrdataset(qctx->client, &qctx->rdataset);
RESTORE(qctx->rdataset, qctx->rpz_st->m.rdataset);
} else {
qctx_clean(qctx);
}
qctx->version = NULL;
RESTORE(qctx->node, qctx->rpz_st->m.node);
RESTORE(qctx->db, qctx->rpz_st->m.db);
RESTORE(qctx->version, qctx->rpz_st->m.version);
RESTORE(qctx->zone, qctx->rpz_st->m.zone);
/*
* Add SOA record to additional section
*/
if (qctx->rpz_st->m.rpz->addsoa) {
rresult = query_addsoa(qctx, UINT32_MAX,
DNS_SECTION_ADDITIONAL);
if (rresult != ISC_R_SUCCESS) {
QUERY_ERROR(qctx, result);
return ISC_R_COMPLETE;
}
}
switch (qctx->rpz_st->m.policy) {
case DNS_RPZ_POLICY_TCP_ONLY:
qctx->client->message->flags |= DNS_MESSAGEFLAG_TC;
if (result == DNS_R_NXDOMAIN ||
result == DNS_R_NCACHENXDOMAIN)
{
qctx->client->message->rcode =
dns_rcode_nxdomain;
}
rpz_log_rewrite(qctx->client, false,
qctx->rpz_st->m.policy,
qctx->rpz_st->m.type, qctx->zone,
qctx->rpz_st->p_name, NULL,
qctx->rpz_st->m.rpz->num);
return ISC_R_COMPLETE;
case DNS_RPZ_POLICY_DROP:
QUERY_ERROR(qctx, DNS_R_DROP);
rpz_log_rewrite(qctx->client, false,
qctx->rpz_st->m.policy,
qctx->rpz_st->m.type, qctx->zone,
qctx->rpz_st->p_name, NULL,
qctx->rpz_st->m.rpz->num);
return ISC_R_COMPLETE;
case DNS_RPZ_POLICY_NXDOMAIN:
result = DNS_R_NXDOMAIN;
qctx->nxrewrite = true;
qctx->rpz = true;
break;
case DNS_RPZ_POLICY_NODATA:
qctx->nxrewrite = true;
FALLTHROUGH;
case DNS_RPZ_POLICY_DNS64:
result = DNS_R_NXRRSET;
qctx->rpz = true;
break;
case DNS_RPZ_POLICY_RECORD:
result = qctx->rpz_st->m.result;
if (qctx->qtype == dns_rdatatype_any &&
result != DNS_R_CNAME)
{
/*
* We will add all of the rdatasets of
* the node by iterating later,
* and set the TTL then.
*/
if (dns_rdataset_isassociated(qctx->rdataset)) {
dns_rdataset_disassociate(
qctx->rdataset);
}
} else {
/*
* We will add this rdataset.
*/
qctx->rdataset->ttl =
ISC_MIN(qctx->rdataset->ttl,
qctx->rpz_st->m.ttl);
}
qctx->rpz = true;
break;
case DNS_RPZ_POLICY_WILDCNAME: {
dns_rdata_t rdata = DNS_RDATA_INIT;
dns_rdata_cname_t cname;
result = dns_rdataset_first(qctx->rdataset);
RUNTIME_CHECK(result == ISC_R_SUCCESS);
dns_rdataset_current(qctx->rdataset, &rdata);
result = dns_rdata_tostruct(&rdata, &cname, NULL);
RUNTIME_CHECK(result == ISC_R_SUCCESS);
dns_rdata_reset(&rdata);
result = query_rpzcname(qctx, &cname.cname);
if (result != ISC_R_SUCCESS) {
return ISC_R_COMPLETE;
}
qctx->fname = NULL;
qctx->want_restart = true;
return ISC_R_COMPLETE;
}
case DNS_RPZ_POLICY_CNAME:
/*
* Add overriding CNAME from a named.conf
* response-policy statement
*/
result = query_rpzcname(qctx,
&qctx->rpz_st->m.rpz->cname);
if (result != ISC_R_SUCCESS) {
return ISC_R_COMPLETE;
}
qctx->fname = NULL;
qctx->want_restart = true;
return ISC_R_COMPLETE;
default:
UNREACHABLE();
}
if (qctx->rpz_st->m.rpz->ede != 0 &&
qctx->rpz_st->m.rpz->ede != UINT16_MAX)
{
dns_ede_add(&qctx->client->edectx,
qctx->rpz_st->m.rpz->ede, NULL);
}
/*
* Turn off DNSSEC because the results of a
* response policy zone cannot verify.
*/
qctx->client->inner.attributes &= ~(NS_CLIENTATTR_WANTDNSSEC |
NS_CLIENTATTR_WANTAD);
qctx->client->message->flags &= ~DNS_MESSAGEFLAG_AD;
ns_client_putrdataset(qctx->client, &qctx->sigrdataset);
qctx->rpz_st->q.is_zone = qctx->is_zone;
qctx->is_zone = true;
rpz_log_rewrite(qctx->client, false, qctx->rpz_st->m.policy,
qctx->rpz_st->m.type, qctx->zone,
qctx->rpz_st->p_name, NULL,
qctx->rpz_st->m.rpz->num);
}
return result;
}
/*%
* Add a CNAME to a query response, including translating foo.evil.com and
* *.evil.com CNAME *.example.com
* to
* foo.evil.com CNAME foo.evil.com.example.com
*/
static isc_result_t
query_rpzcname(query_ctx_t *qctx, dns_name_t *cname) {
ns_client_t *client;
dns_fixedname_t prefix, suffix;
unsigned int labels;
isc_result_t result;
REQUIRE(qctx != NULL && qctx->client != NULL);
client = qctx->client;
CTRACE(ISC_LOG_DEBUG(3), "query_rpzcname");
labels = dns_name_countlabels(cname);
if (labels > 2 && dns_name_iswildcard(cname)) {
dns_fixedname_init(&prefix);
dns_name_split(client->query.qname, 1,
dns_fixedname_name(&prefix), NULL);
dns_fixedname_init(&suffix);
dns_name_split(cname, labels - 1, NULL,
dns_fixedname_name(&suffix));
result = dns_name_concatenate(dns_fixedname_name(&prefix),
dns_fixedname_name(&suffix),
qctx->fname);
if (result == DNS_R_NAMETOOLONG) {
client->message->rcode = dns_rcode_yxdomain;
} else if (result != ISC_R_SUCCESS) {
return result;
}
} else {
dns_name_copy(cname, qctx->fname);
}
ns_client_keepname(client, qctx->fname, qctx->dbuf);
query_addcname(qctx, dns_trust_authanswer, qctx->rpz_st->m.ttl);
rpz_log_rewrite(client, false, qctx->rpz_st->m.policy,
qctx->rpz_st->m.type, qctx->rpz_st->m.zone,
qctx->rpz_st->p_name, qctx->fname,
qctx->rpz_st->m.rpz->num);
ns_client_qnamereplace(client, qctx->fname);
/*
* Turn off DNSSEC because the results of a
* response policy zone cannot verify.
*/
client->inner.attributes &= ~(NS_CLIENTATTR_WANTDNSSEC |
NS_CLIENTATTR_WANTAD);
return ISC_R_SUCCESS;
}
/*%
* Check the configured trust anchors for a root zone trust anchor
* with a key id that matches qctx->client->query.root_key_sentinel_keyid.
*
* Return true when found, otherwise return false.
*/
static bool
has_ta(query_ctx_t *qctx) {
dns_keytable_t *keytable = NULL;
dns_keynode_t *keynode = NULL;
dns_rdataset_t dsset;
dns_keytag_t sentinel = qctx->client->query.root_key_sentinel_keyid;
isc_result_t result;
result = dns_view_getsecroots(qctx->view, &keytable);
if (result != ISC_R_SUCCESS) {
return false;
}
result = dns_keytable_find(keytable, dns_rootname, &keynode);
if (result != ISC_R_SUCCESS) {
if (keynode != NULL) {
dns_keynode_detach(&keynode);
}
dns_keytable_detach(&keytable);
return false;
}
dns_rdataset_init(&dsset);
if (dns_keynode_dsset(keynode, &dsset)) {
DNS_RDATASET_FOREACH (&dsset) {
dns_rdata_t rdata = DNS_RDATA_INIT;
dns_rdata_ds_t ds;
dns_rdata_reset(&rdata);
dns_rdataset_current(&dsset, &rdata);
result = dns_rdata_tostruct(&rdata, &ds, NULL);
RUNTIME_CHECK(result == ISC_R_SUCCESS);
if (ds.key_tag == sentinel) {
dns_keynode_detach(&keynode);
dns_keytable_detach(&keytable);
dns_rdataset_disassociate(&dsset);
return true;
}
}
dns_rdataset_disassociate(&dsset);
}
if (keynode != NULL) {
dns_keynode_detach(&keynode);
}
dns_keytable_detach(&keytable);
return false;
}
/*%
* Check if a root key sentinel SERVFAIL should be returned.
*/
static bool
root_key_sentinel_return_servfail(query_ctx_t *qctx, isc_result_t result) {
/*
* Are we looking at a "root-key-sentinel" query?
*/
if (!qctx->client->query.root_key_sentinel_is_ta &&
!qctx->client->query.root_key_sentinel_not_ta)
{
return false;
}
/*
* We only care about the query if 'result' indicates we have a cached
* answer.
*/
switch (result) {
case ISC_R_SUCCESS:
case DNS_R_CNAME:
case DNS_R_DNAME:
case DNS_R_NCACHENXDOMAIN:
case DNS_R_NCACHENXRRSET:
break;
default:
return false;
}
/*
* Do we meet the specified conditions to return SERVFAIL?
*/
if (!qctx->is_zone && qctx->rdataset->trust == dns_trust_secure &&
((qctx->client->query.root_key_sentinel_is_ta && !has_ta(qctx)) ||
(qctx->client->query.root_key_sentinel_not_ta && has_ta(qctx))))
{
return true;
}
/*
* As special processing may only be triggered by the original QNAME,
* disable it after following a CNAME/DNAME.
*/
qctx->client->query.root_key_sentinel_is_ta = false;
qctx->client->query.root_key_sentinel_not_ta = false;
return false;
}
/*%
* If serving stale answers is allowed, set up 'qctx' to look for one and
* return true; otherwise, return false.
*/
static bool
query_usestale(query_ctx_t *qctx, isc_result_t result) {
if ((qctx->client->query.dboptions & DNS_DBFIND_STALEOK) != 0) {
/*
* Query was already using stale, if that didn't work the
* last time, it won't work this time either.
*/
return false;
}
if (result == DNS_R_DUPLICATE || result == DNS_R_DROP) {
/*
* Don't enable serve-stale if the result signals a duplicate
* query or query that is being dropped.
*/
return false;
}
qctx_clean(qctx);
qctx_freedata(qctx);
if (dns_view_staleanswerenabled(qctx->client->inner.view)) {
isc_result_t ret;
ret = query_getdb(qctx->client, qctx->client->query.qname,
qctx->client->query.qtype, qctx->options,
&qctx->zone, &qctx->db, &qctx->version,
&qctx->is_zone);
if (ret != ISC_R_SUCCESS) {
/*
* Failed to get the database, unexpected, but let us
* at least abandon serve-stale.
*/
return false;
}
qctx->client->query.dboptions |= DNS_DBFIND_STALEOK;
if (FETCH_RECTYPE_NORMAL(qctx->client) != NULL) {
dns_resolver_destroyfetch(
&FETCH_RECTYPE_NORMAL(qctx->client));
}
/*
* Start the stale-refresh-time window in case there was a
* resolver query timeout.
*/
if (qctx->resuming && result == ISC_R_TIMEDOUT) {
qctx->client->query.dboptions |= DNS_DBFIND_STALESTART;
}
return true;
}
return false;
}
/*%
* Continue after doing a database lookup or returning from
* recursion, and call out to the next function depending on the
* result from the search.
*/
static isc_result_t
query_gotanswer(query_ctx_t *qctx, isc_result_t result) {
char errmsg[256];
CCTRACE(ISC_LOG_DEBUG(3), "query_gotanswer");
CALL_HOOK(NS_QUERY_GOT_ANSWER_BEGIN, qctx);
if (query_checkrrl(qctx, result) != ISC_R_SUCCESS) {
return ns_query_done(qctx);
}
if (!dns_name_equal(qctx->client->query.qname, dns_rootname)) {
result = query_checkrpz(qctx, result);
if (result == ISC_R_NOTFOUND) {
/*
* RPZ not configured for this view.
*/
goto root_key_sentinel;
}
if (RECURSING(qctx->client) && result == DNS_R_DISALLOWED) {
/*
* We are recursing, and thus RPZ processing is not
* allowed at the moment. This could happen on a
* "stale-answer-client-timeout" lookup. In this case,
* bail out and wait for recursion to complete, as we
* we can't perform the RPZ rewrite rules.
*/
return result;
}
if (result == ISC_R_COMPLETE) {
return ns_query_done(qctx);
}
}
root_key_sentinel:
/*
* If required, handle special "root-key-sentinel-is-ta-<keyid>" and
* "root-key-sentinel-not-ta-<keyid>" labels by returning SERVFAIL.
*/
if (root_key_sentinel_return_servfail(qctx, result)) {
/*
* Don't record this response in the SERVFAIL cache.
*/
qctx->client->inner.attributes |= NS_CLIENTATTR_NOSETFC;
QUERY_ERROR(qctx, DNS_R_SERVFAIL);
return ns_query_done(qctx);
}
switch (result) {
case ISC_R_SUCCESS:
return query_prepresponse(qctx);
case DNS_R_GLUE:
case DNS_R_ZONECUT:
INSIST(qctx->is_zone);
qctx->authoritative = false;
return query_prepresponse(qctx);
case ISC_R_NOTFOUND:
return query_notfound(qctx);
case DNS_R_DELEGATION:
return query_delegation(qctx);
case DNS_R_EMPTYNAME:
case DNS_R_NXRRSET:
return query_nodata(qctx, result);
case DNS_R_EMPTYWILD:
case DNS_R_NXDOMAIN:
return query_nxdomain(qctx, result);
case DNS_R_COVERINGNSEC:
return query_coveringnsec(qctx);
case DNS_R_NCACHENXDOMAIN:
result = query_redirect(qctx, result);
if (result != ISC_R_COMPLETE) {
return result;
}
return query_ncache(qctx, DNS_R_NCACHENXDOMAIN);
case DNS_R_NCACHENXRRSET:
return query_ncache(qctx, DNS_R_NCACHENXRRSET);
case DNS_R_CNAME:
return query_cname(qctx);
case DNS_R_DNAME:
return query_dname(qctx);
default:
/*
* Something has gone wrong.
*/
snprintf(errmsg, sizeof(errmsg) - 1,
"query_gotanswer: unexpected error: %s",
isc_result_totext(result));
CCTRACE(ISC_LOG_ERROR, errmsg);
if (query_usestale(qctx, result)) {
/*
* If serve-stale is enabled, query_usestale() already
* set up 'qctx' for looking up a stale response.
*/
return query_lookup(qctx);
}
/*
* Regardless of the triggering result, we definitely
* want to return SERVFAIL from here.
*/
qctx->client->inner.rcode_override = dns_rcode_servfail;
QUERY_ERROR(qctx, result);
return ns_query_done(qctx);
}
cleanup:
return result;
}
static void
query_addnoqnameproof(query_ctx_t *qctx) {
ns_client_t *client = qctx->client;
isc_buffer_t *dbuf, b;
dns_name_t *fname = NULL;
dns_rdataset_t *neg = NULL, *negsig = NULL;
isc_result_t result;
CTRACE(ISC_LOG_DEBUG(3), "query_addnoqnameproof");
if (qctx->noqname == NULL) {
return;
}
dbuf = ns_client_getnamebuf(client);
fname = ns_client_newname(client, dbuf, &b);
neg = ns_client_newrdataset(client);
negsig = ns_client_newrdataset(client);
result = dns_rdataset_getnoqname(qctx->noqname, fname, neg, negsig);
RUNTIME_CHECK(result == ISC_R_SUCCESS);
query_addrrset(qctx, &fname, &neg, &negsig, dbuf,
DNS_SECTION_AUTHORITY);
if (!qctx->noqname->attributes.closest) {
goto cleanup;
}
if (fname == NULL) {
dbuf = ns_client_getnamebuf(client);
fname = ns_client_newname(client, dbuf, &b);
}
if (neg == NULL) {
neg = ns_client_newrdataset(client);
} else if (dns_rdataset_isassociated(neg)) {
dns_rdataset_disassociate(neg);
}
if (negsig == NULL) {
negsig = ns_client_newrdataset(client);
} else if (dns_rdataset_isassociated(negsig)) {
dns_rdataset_disassociate(negsig);
}
result = dns_rdataset_getclosest(qctx->noqname, fname, neg, negsig);
RUNTIME_CHECK(result == ISC_R_SUCCESS);
query_addrrset(qctx, &fname, &neg, &negsig, dbuf,
DNS_SECTION_AUTHORITY);
cleanup:
if (neg != NULL) {
ns_client_putrdataset(client, &neg);
}
if (negsig != NULL) {
ns_client_putrdataset(client, &negsig);
}
if (fname != NULL) {
ns_client_releasename(client, &fname);
}
}
/*%
* Build the response for a query for type ANY.
*/
static isc_result_t
query_respond_any(query_ctx_t *qctx) {
bool found = false, hidden = false;
dns_rdatasetiter_t *rdsiter = NULL;
isc_result_t result = ISC_R_UNSET;
dns_rdatatype_t onetype = 0; /* type to use for minimal-any */
isc_buffer_t b;
CCTRACE(ISC_LOG_DEBUG(3), "query_respond_any");
CALL_HOOK(NS_QUERY_RESPOND_ANY_BEGIN, qctx);
result = dns_db_allrdatasets(qctx->db, qctx->node, qctx->version, 0, 0,
&rdsiter);
if (result != ISC_R_SUCCESS) {
CCTRACE(ISC_LOG_ERROR, "query_respond_any: allrdatasets "
"failed");
QUERY_ERROR(qctx, result);
return ns_query_done(qctx);
}
/*
* Calling query_addrrset() with a non-NULL dbuf is going
* to either keep or release the name. We don't want it to
* release fname, since we may have to call query_addrrset()
* more than once. That means we have to call ns_client_keepname()
* now, and pass a NULL dbuf to query_addrrset().
*
* If we do a query_addrrset() below, we must set qctx->fname to
* NULL before leaving this block, otherwise we might try to
* cleanup qctx->fname even though we're using it!
*/
ns_client_keepname(qctx->client, qctx->fname, qctx->dbuf);
qctx->tname = qctx->fname;
DNS_RDATASETITER_FOREACH (rdsiter) {
dns_rdatasetiter_current(rdsiter, qctx->rdataset);
/*
* We found an NS RRset; no need to add one later.
*/
if (qctx->qtype == dns_rdatatype_any &&
qctx->rdataset->type == dns_rdatatype_ns)
{
qctx->answer_has_ns = true;
}
/*
* Note: if we're in this function, then qctx->type
* is guaranteed to be ANY, but qctx->qtype (i.e. the
* original type requested) might have been RRSIG or
* SIG; we need to check for that.
*/
if (qctx->is_zone && qctx->qtype == dns_rdatatype_any &&
!dns_db_issecure(qctx->db) &&
dns_rdatatype_isdnssec(qctx->rdataset->type))
{
/*
* The zone may be transitioning from insecure
* to secure. Hide DNSSEC records from ANY queries.
*/
dns_rdataset_disassociate(qctx->rdataset);
hidden = true;
} else if (qctx->view->minimal_any && !TCP(qctx->client) &&
!WANTDNSSEC(qctx->client) &&
qctx->qtype == dns_rdatatype_any &&
(dns_rdatatype_issig(qctx->rdataset->type)))
{
CCTRACE(ISC_LOG_DEBUG(5), "query_respond_any: "
"minimal-any skip signature");
dns_rdataset_disassociate(qctx->rdataset);
} else if (qctx->view->minimal_any && !TCP(qctx->client) &&
onetype != 0 && qctx->rdataset->type != onetype &&
qctx->rdataset->covers != onetype)
{
CCTRACE(ISC_LOG_DEBUG(5), "query_respond_any: "
"minimal-any skip rdataset");
dns_rdataset_disassociate(qctx->rdataset);
} else if ((qctx->qtype == dns_rdatatype_any ||
qctx->rdataset->type == qctx->qtype) &&
qctx->rdataset->type != 0)
{
if (NOQNAME(qctx->rdataset) && WANTDNSSEC(qctx->client))
{
qctx->noqname = qctx->rdataset;
} else {
qctx->noqname = NULL;
}
qctx->rpz_st = qctx->client->query.rpz_st;
if (qctx->rpz_st != NULL &&
qctx->rpz_st->m.policy != DNS_RPZ_POLICY_MISS &&
qctx->rpz_st->m.policy != DNS_RPZ_POLICY_PASSTHRU)
{
qctx->rdataset->ttl =
ISC_MIN(qctx->rdataset->ttl,
qctx->rpz_st->m.ttl);
}
if (!qctx->is_zone && RECURSIONOK(qctx->client)) {
dns_name_t *name;
name = (qctx->fname != NULL) ? qctx->fname
: qctx->tname;
query_prefetch(qctx->client, name,
qctx->rdataset);
}
/*
* Remember the first RRtype we find so we
* can skip others with minimal-any.
*/
if (dns_rdatatype_issig(qctx->rdataset->type)) {
onetype = qctx->rdataset->covers;
} else {
onetype = qctx->rdataset->type;
}
query_addrrset(qctx,
(qctx->fname != NULL) ? &qctx->fname
: &qctx->tname,
&qctx->rdataset, NULL, NULL,
DNS_SECTION_ANSWER);
query_addnoqnameproof(qctx);
found = true;
INSIST(qctx->tname != NULL);
/*
* rdataset is non-NULL only in certain
* pathological cases involving DNAMEs.
*/
if (qctx->rdataset != NULL) {
ns_client_putrdataset(qctx->client,
&qctx->rdataset);
}
qctx->rdataset = ns_client_newrdataset(qctx->client);
} else {
/*
* We're not interested in this rdataset.
*/
dns_rdataset_disassociate(qctx->rdataset);
}
}
dns_rdatasetiter_destroy(&rdsiter);
if (found) {
/*
* Call hook if any answers were found.
* Do this before releasing qctx->fname, in case
* the hook function needs it.
*/
CALL_HOOK(NS_QUERY_RESPOND_ANY_FOUND, qctx);
}
if (qctx->fname != NULL) {
dns_message_puttempname(qctx->client->message, &qctx->fname);
}
if (found) {
/*
* At least one matching rdataset was found
*/
query_addauth(qctx);
} else if (dns_rdatatype_issig(qctx->qtype)) {
/*
* No matching rdatasets were found, but we got
* here on a search for RRSIG/SIG, so that's okay.
*/
if (!qctx->is_zone) {
qctx->authoritative = false;
qctx->client->inner.attributes &= ~NS_CLIENTATTR_RA;
query_addauth(qctx);
return ns_query_done(qctx);
}
if (qctx->qtype == dns_rdatatype_rrsig &&
dns_db_issecure(qctx->db))
{
char namebuf[DNS_NAME_FORMATSIZE];
dns_name_format(qctx->client->query.qname, namebuf,
sizeof(namebuf));
ns_client_log(qctx->client, DNS_LOGCATEGORY_DNSSEC,
NS_LOGMODULE_QUERY, ISC_LOG_WARNING,
"missing signature for %s", namebuf);
}
qctx->fname = ns_client_newname(qctx->client, qctx->dbuf, &b);
return query_sign_nodata(qctx);
} else if (!hidden) {
/*
* No matching rdatasets were found and nothing was
* deliberately hidden: something must have gone wrong.
*/
QUERY_ERROR(qctx, DNS_R_SERVFAIL);
}
return ns_query_done(qctx);
cleanup:
return result;
}
/*
* Set the expire time, if requested, when answering from a secondary,
* mirror, or primary zone.
*/
static void
query_getexpire(query_ctx_t *qctx) {
dns_zone_t *raw = NULL, *mayberaw;
CCTRACE(ISC_LOG_DEBUG(3), "query_getexpire");
if (qctx->zone == NULL || !qctx->is_zone ||
qctx->qtype != dns_rdatatype_soa ||
qctx->client->query.restarts != 0 ||
(qctx->client->inner.attributes & NS_CLIENTATTR_WANTEXPIRE) == 0)
{
return;
}
dns_zone_getraw(qctx->zone, &raw);
mayberaw = (raw != NULL) ? raw : qctx->zone;
if (dns_zone_gettype(mayberaw) == dns_zone_secondary ||
dns_zone_gettype(mayberaw) == dns_zone_mirror)
{
isc_time_t expiretime;
uint32_t secs;
dns_zone_getexpiretime(qctx->zone, &expiretime);
secs = isc_time_seconds(&expiretime);
if (secs >= qctx->client->inner.now &&
qctx->result == ISC_R_SUCCESS)
{
qctx->client->inner.attributes |=
NS_CLIENTATTR_HAVEEXPIRE;
qctx->client->inner.expire = secs -
qctx->client->inner.now;
}
} else if (dns_zone_gettype(mayberaw) == dns_zone_primary) {
isc_result_t result;
dns_rdata_t rdata = DNS_RDATA_INIT;
dns_rdata_soa_t soa;
result = dns_rdataset_first(qctx->rdataset);
RUNTIME_CHECK(result == ISC_R_SUCCESS);
dns_rdataset_current(qctx->rdataset, &rdata);
result = dns_rdata_tostruct(&rdata, &soa, NULL);
RUNTIME_CHECK(result == ISC_R_SUCCESS);
qctx->client->inner.expire = soa.expire;
qctx->client->inner.attributes |= NS_CLIENTATTR_HAVEEXPIRE;
}
if (raw != NULL) {
dns_zone_detach(&raw);
}
}
/*
* Set the zone version, if requested, when answering from a secondary,
* mirror, or primary zone.
*/
static void
query_getzoneversion(query_ctx_t *qctx) {
CCTRACE(ISC_LOG_DEBUG(3), __func__);
if (qctx->zone == NULL || !qctx->is_zone ||
qctx->client->query.restarts != 0 ||
(qctx->client->inner.attributes & NS_CLIENTATTR_WANTZONEVERSION) ==
0 ||
(qctx->client->inner.attributes & NS_CLIENTATTR_HAVEZONEVERSION) !=
0)
{
return;
}
switch (dns_zone_gettype(qctx->zone)) {
case dns_zone_mirror:
case dns_zone_primary:
case dns_zone_secondary: {
isc_buffer_t b;
unsigned char buf[128];
isc_buffer_init(&b, buf, sizeof(buf));
isc_result_t result = dns_zone_getzoneversion(qctx->zone, &b);
if (result == ISC_R_SUCCESS) {
size_t len = isc_buffer_usedlength(&b);
/*
* Sanity check zone version from database
* implementations. Minimum length and type 0
* contraints.
*/
if (len < 2 || (buf[1] == 0 && len != 6)) {
return;
}
qctx->client->inner.attributes |=
NS_CLIENTATTR_HAVEZONEVERSION;
INSIST(qctx->client->inner.zoneversion == NULL);
qctx->client->inner.zoneversion =
isc_mem_get(qctx->client->manager->mctx, len);
qctx->client->inner.zoneversionlength = len;
memmove(qctx->client->inner.zoneversion, buf, len);
}
break;
}
default:
break;
}
}
/*%
* Fill the ANSWER section of a positive response.
*/
static isc_result_t
query_addanswer(query_ctx_t *qctx) {
dns_rdataset_t **sigrdatasetp = NULL;
isc_result_t result = ISC_R_UNSET;
CCTRACE(ISC_LOG_DEBUG(3), "query_addanswer");
CALL_HOOK(NS_QUERY_ADDANSWER_BEGIN, qctx);
/*
* On normal lookups, clear any rdatasets that were added on a
* lookup due to stale-answer-client-timeout. Do not clear if we
* are going to refresh the RRset, because the stale contents are
* prioritized.
*/
if (QUERY_STALEOK(&qctx->client->query) &&
!QUERY_STALETIMEOUT(&qctx->client->query) && !qctx->refresh_rrset)
{
CCTRACE(ISC_LOG_DEBUG(3), "query_clear_stale");
/*
* Clear any rdatasets from the client's message that were added
* on a lookup due to a client timeout.
*/
message_clearrdataset(qctx->client->message, true);
/*
* We can clear the attribute to prevent redundant clearing
* in subsequent lookups.
*/
qctx->client->query.attributes &= ~NS_QUERYATTR_STALEOK;
}
if (qctx->dns64) {
result = query_dns64(qctx);
qctx->noqname = NULL;
dns_rdataset_disassociate(qctx->rdataset);
dns_message_puttemprdataset(qctx->client->message,
&qctx->rdataset);
if (result == ISC_R_NOMORE) {
#ifndef dns64_bis_return_excluded_addresses
if (qctx->dns64_exclude) {
if (!qctx->is_zone) {
return ns_query_done(qctx);
}
/*
* Add a fake SOA record.
*/
(void)query_addsoa(qctx, 600,
DNS_SECTION_AUTHORITY);
return ns_query_done(qctx);
}
#endif /* ifndef dns64_bis_return_excluded_addresses */
if (qctx->is_zone) {
return query_nodata(qctx, DNS_R_NXDOMAIN);
} else {
return query_ncache(qctx, DNS_R_NXDOMAIN);
}
} else if (result != ISC_R_SUCCESS) {
qctx->result = result;
return ns_query_done(qctx);
}
} else if (qctx->client->query.dns64_aaaaok != NULL) {
query_filter64(qctx);
ns_client_putrdataset(qctx->client, &qctx->rdataset);
} else {
if (!qctx->is_zone && RECURSIONOK(qctx->client) &&
!QUERY_STALETIMEOUT(&qctx->client->query))
{
query_prefetch(qctx->client, qctx->fname,
qctx->rdataset);
}
if (WANTDNSSEC(qctx->client) && qctx->sigrdataset != NULL) {
sigrdatasetp = &qctx->sigrdataset;
}
query_addrrset(qctx, &qctx->fname, &qctx->rdataset,
sigrdatasetp, qctx->dbuf, DNS_SECTION_ANSWER);
}
return ISC_R_COMPLETE;
cleanup:
return result;
}
/*%
* Build a response for a "normal" query, for a type other than ANY,
* for which we have an answer (either positive or negative).
*/
static isc_result_t
query_respond(query_ctx_t *qctx) {
isc_result_t result = ISC_R_UNSET;
CCTRACE(ISC_LOG_DEBUG(3), "query_respond");
/*
* Check to see if the AAAA RRset has non-excluded addresses
* in it. If not look for a A RRset.
*/
INSIST(qctx->client->query.dns64_aaaaok == NULL);
if (qctx->qtype == dns_rdatatype_aaaa && !qctx->dns64_exclude &&
!ISC_LIST_EMPTY(qctx->view->dns64) &&
qctx->client->message->rdclass == dns_rdataclass_in &&
!dns64_aaaaok(qctx->client, qctx->rdataset, qctx->sigrdataset))
{
/*
* Look to see if there are A records for this name.
*/
qctx->client->query.dns64_ttl = qctx->rdataset->ttl;
SAVE(qctx->client->query.dns64_aaaa, qctx->rdataset);
SAVE(qctx->client->query.dns64_sigaaaa, qctx->sigrdataset);
ns_client_releasename(qctx->client, &qctx->fname);
dns_db_detachnode(qctx->db, &qctx->node);
qctx->type = qctx->qtype = dns_rdatatype_a;
qctx->dns64_exclude = qctx->dns64 = true;
return query_lookup(qctx);
}
/*
* XXX: This hook is meant to be at the top of this function,
* but is postponed until after DNS64 in order to avoid an
* assertion if the hook causes recursion. (When DNS64 also
* becomes a plugin, it will be necessary to find some
* other way to prevent that assertion, since the order in
* which plugins are configured can't be enforced.)
*/
CALL_HOOK(NS_QUERY_RESPOND_BEGIN, qctx);
if (NOQNAME(qctx->rdataset) && WANTDNSSEC(qctx->client)) {
qctx->noqname = qctx->rdataset;
} else {
qctx->noqname = NULL;
}
/*
* Special case NS handling
*/
if (qctx->is_zone && qctx->qtype == dns_rdatatype_ns) {
/*
* We've already got an NS, no need to add one in
* the authority section
*/
if (dns_name_equal(qctx->client->query.qname,
dns_db_origin(qctx->db)))
{
qctx->answer_has_ns = true;
}
/*
* Always add glue for root priming queries, regardless
* of "minimal-responses" setting.
*/
if (dns_name_equal(qctx->client->query.qname, dns_rootname)) {
qctx->client->query.attributes &=
~NS_QUERYATTR_NOADDITIONAL;
dns_db_attach(qctx->db, &qctx->client->query.gluedb);
}
}
/*
* Set expire time
*/
query_getexpire(qctx);
/*
* Set the zone version
*/
query_getzoneversion(qctx);
result = query_addanswer(qctx);
if (result != ISC_R_COMPLETE) {
return result;
}
query_addnoqnameproof(qctx);
/*
* 'qctx->rdataset' will only be non-NULL here if the ANSWER section of
* the message to be sent to the client already contains an RRset with
* the same owner name and the same type as 'qctx->rdataset'. This
* should never happen, with one exception: when chasing DNAME records,
* one of the DNAME records placed in the ANSWER section may turn out
* to be the final answer to the client's query, but we have no way of
* knowing that until now. In such a case, 'qctx->rdataset' will be
* freed later, so we do not need to free it here.
*/
INSIST(qctx->rdataset == NULL || qctx->qtype == dns_rdatatype_dname);
query_addauth(qctx);
return ns_query_done(qctx);
cleanup:
return result;
}
static isc_result_t
query_dns64(query_ctx_t *qctx) {
isc_result_t result;
ns_client_t *client = qctx->client;
dns_name_t *name = NULL, *mname = NULL;
dns_rdataset_t *mrdataset = NULL;
dns_rdataset_t *dns64_rdataset = NULL;
dns_view_t *view = client->inner.view;
const dns_section_t section = DNS_SECTION_ANSWER;
unsigned int flags = 0;
/*%
* To the current response for 'qctx->client', add the answer RRset
* '*rdatasetp' and an optional signature set '*sigrdatasetp', with
* owner name '*namep', to the answer section, unless they are
* already there. Also add any pertinent additional data.
*
* If 'qctx->dbuf' is not NULL, then 'qctx->fname' is the name
* whose data is stored 'qctx->dbuf'. In this case,
* query_addrrset() guarantees that when it returns the name
* will either have been kept or released.
*/
CTRACE(ISC_LOG_DEBUG(3), "query_dns64");
qctx->qtype = qctx->type = dns_rdatatype_aaaa;
name = qctx->fname;
result = dns_message_findname(
client->message, section, name, dns_rdatatype_aaaa,
qctx->rdataset->covers, &mname, &mrdataset);
if (result == ISC_R_SUCCESS) {
/*
* We've already got an RRset of the given name and type.
* There's nothing else to do;
*/
CTRACE(ISC_LOG_DEBUG(3), "query_dns64: dns_message_findname "
"succeeded: done");
if (qctx->dbuf != NULL) {
ns_client_releasename(client, &qctx->fname);
}
return ISC_R_SUCCESS;
} else if (result == DNS_R_NXDOMAIN) {
/*
* The name doesn't exist.
*/
if (qctx->dbuf != NULL) {
ns_client_keepname(client, name, qctx->dbuf);
}
dns_message_addname(client->message, name, section);
qctx->fname = NULL;
mname = name;
} else {
RUNTIME_CHECK(result == DNS_R_NXRRSET);
if (qctx->dbuf != NULL) {
ns_client_releasename(client, &qctx->fname);
}
}
if (qctx->rdataset->trust != dns_trust_secure) {
client->query.attributes &= ~NS_QUERYATTR_SECURE;
}
/*
* We use the signatures from the A lookup to set DNS_DNS64_DNSSEC
* as this provides a easy way to see if the answer was signed.
*/
if (WANTDNSSEC(qctx->client) && qctx->sigrdataset != NULL &&
dns_rdataset_isassociated(qctx->sigrdataset))
{
flags |= DNS_DNS64_DNSSEC;
}
if (RECURSIONOK(client)) {
flags |= DNS_DNS64_RECURSIVE;
}
result = dns_dns64_apply(client->manager->mctx, view->dns64,
view->dns64cnt, client->message,
client->manager->aclenv,
&client->inner.peeraddr, client->inner.signer,
flags, qctx->rdataset, &dns64_rdataset);
if (result != ISC_R_SUCCESS) {
goto cleanup;
}
dns_rdataset_setownercase(dns64_rdataset, mname);
client->query.attributes |= NS_QUERYATTR_NOADDITIONAL;
if (client->query.dns64_ttl != UINT32_MAX) {
dns64_rdataset->ttl = ISC_MIN(qctx->rdataset->ttl,
client->query.dns64_ttl);
} else {
dns64_rdataset->ttl = ISC_MIN(qctx->rdataset->ttl, 600);
}
query_addtoname(mname, dns64_rdataset);
query_setorder(qctx, mname, dns64_rdataset);
inc_stats(client, ns_statscounter_dns64);
result = ISC_R_SUCCESS;
cleanup:
CTRACE(ISC_LOG_DEBUG(3), "query_dns64: done");
return result;
}
static void
query_filter64(query_ctx_t *qctx) {
ns_client_t *client = qctx->client;
dns_name_t *name = NULL, *mname = NULL;
dns_rdatalist_t *myrdatalist = NULL;
dns_rdataset_t *myrdataset = NULL;
isc_buffer_t *buffer = NULL;
isc_result_t result;
unsigned int i;
const dns_section_t section = DNS_SECTION_ANSWER;
CTRACE(ISC_LOG_DEBUG(3), "query_filter64");
INSIST(client->query.dns64_aaaaok != NULL);
INSIST(client->query.dns64_aaaaoklen ==
dns_rdataset_count(qctx->rdataset));
name = qctx->fname;
result = dns_message_findname(
client->message, section, name, dns_rdatatype_aaaa,
qctx->rdataset->covers, &mname, &myrdataset);
if (result == ISC_R_SUCCESS) {
/*
* We've already got an RRset of the given name and type.
* There's nothing else to do;
*/
CTRACE(ISC_LOG_DEBUG(3), "query_filter64: dns_message_findname "
"succeeded: done");
if (qctx->dbuf != NULL) {
ns_client_releasename(client, &qctx->fname);
}
return;
} else if (result == DNS_R_NXDOMAIN) {
mname = name;
qctx->fname = NULL;
} else {
RUNTIME_CHECK(result == DNS_R_NXRRSET);
if (qctx->dbuf != NULL) {
ns_client_releasename(client, &qctx->fname);
}
qctx->dbuf = NULL;
}
if (qctx->rdataset->trust != dns_trust_secure) {
client->query.attributes &= ~NS_QUERYATTR_SECURE;
}
isc_buffer_allocate(client->manager->mctx, &buffer,
16 * dns_rdataset_count(qctx->rdataset));
dns_message_gettemprdataset(client->message, &myrdataset);
dns_message_gettemprdatalist(client->message, &myrdatalist);
myrdatalist->rdclass = dns_rdataclass_in;
myrdatalist->type = dns_rdatatype_aaaa;
myrdatalist->ttl = qctx->rdataset->ttl;
i = 0;
DNS_RDATASET_FOREACH (qctx->rdataset) {
dns_rdata_t rdata = DNS_RDATA_INIT;
dns_rdata_t *myrdata = NULL;
isc_region_t r;
if (!client->query.dns64_aaaaok[i++]) {
continue;
}
dns_rdataset_current(qctx->rdataset, &rdata);
INSIST(rdata.length == 16);
isc_buffer_putmem(buffer, rdata.data, rdata.length);
isc_buffer_remainingregion(buffer, &r);
isc_buffer_forward(buffer, rdata.length);
dns_message_gettemprdata(client->message, &myrdata);
dns_rdata_fromregion(myrdata, dns_rdataclass_in,
dns_rdatatype_aaaa, &r);
ISC_LIST_APPEND(myrdatalist->rdata, myrdata, link);
}
dns_rdatalist_tordataset(myrdatalist, myrdataset);
dns_rdataset_setownercase(myrdataset, name);
client->query.attributes |= NS_QUERYATTR_NOADDITIONAL;
if (mname == name) {
if (qctx->dbuf != NULL) {
ns_client_keepname(client, name, qctx->dbuf);
}
dns_message_addname(client->message, name, section);
qctx->dbuf = NULL;
}
myrdataset->trust = qctx->rdataset->trust;
query_addtoname(mname, myrdataset);
query_setorder(qctx, mname, myrdataset);
dns_message_takebuffer(client->message, &buffer);
if (buffer != NULL) {
isc_buffer_free(&buffer);
}
if (qctx->dbuf != NULL) {
ns_client_releasename(client, &name);
}
CTRACE(ISC_LOG_DEBUG(3), "query_filter64: done");
}
/*%
* Handle the case of a name not being found in a database lookup.
* Called from query_gotanswer(). Passes off processing to
* query_delegation() for a root referral if appropriate.
*/
static isc_result_t
query_notfound(query_ctx_t *qctx) {
isc_result_t result = ISC_R_UNSET;
CCTRACE(ISC_LOG_DEBUG(3), "query_notfound");
CALL_HOOK(NS_QUERY_NOTFOUND_BEGIN, qctx);
INSIST(!qctx->is_zone);
if (qctx->db != NULL) {
dns_db_detach(&qctx->db);
}
/*
* If the cache doesn't even have the root NS,
* try to get that from the hints DB.
*/
if (qctx->view->hints != NULL) {
dns_clientinfomethods_t cm;
dns_clientinfo_t ci;
dns_clientinfomethods_init(&cm, ns_client_sourceip);
dns_clientinfo_init(&ci, qctx->client, NULL);
dns_db_attach(qctx->view->hints, &qctx->db);
result = dns_db_findext(
qctx->db, dns_rootname, NULL, dns_rdatatype_ns, 0,
qctx->client->inner.now, &qctx->node, qctx->fname, &cm,
&ci, qctx->rdataset, qctx->sigrdataset);
} else {
/* We have no hints. */
result = ISC_R_FAILURE;
}
if (result != ISC_R_SUCCESS) {
/*
* Nonsensical root hints may require cleanup.
*/
qctx_clean(qctx);
/*
* We don't have any root server hints, but
* we may have working forwarders, so try to
* recurse anyway.
*/
if (RECURSIONOK(qctx->client)) {
INSIST(!REDIRECT(qctx->client));
result = ns_query_recurse(qctx->client, qctx->qtype,
qctx->client->query.qname,
NULL, NULL, qctx->resuming);
if (result == ISC_R_SUCCESS) {
CALL_HOOK(NS_QUERY_NOTFOUND_RECURSE, qctx);
qctx->client->query.attributes |=
NS_QUERYATTR_RECURSING;
if (qctx->dns64) {
qctx->client->query.attributes |=
NS_QUERYATTR_DNS64;
}
if (qctx->dns64_exclude) {
qctx->client->query.attributes |=
NS_QUERYATTR_DNS64EXCLUDE;
}
} else if (query_usestale(qctx, result)) {
/*
* If serve-stale is enabled, query_usestale()
* already set up 'qctx' for looking up a
* stale response.
*/
return query_lookup(qctx);
} else {
QUERY_ERROR(qctx, result);
}
return ns_query_done(qctx);
} else {
/* Unable to give root server referral. */
CCTRACE(ISC_LOG_ERROR, "unable to give root server "
"referral");
QUERY_ERROR(qctx, result);
return ns_query_done(qctx);
}
}
return query_delegation(qctx);
cleanup:
return result;
}
/*%
* We have a delegation but recursion is not allowed, so return the delegation
* to the client.
*/
static isc_result_t
query_prepare_delegation_response(query_ctx_t *qctx) {
isc_result_t result = ISC_R_UNSET;
dns_rdataset_t **sigrdatasetp = NULL;
bool detach = false;
CALL_HOOK(NS_QUERY_PREP_DELEGATION_BEGIN, qctx);
/*
* qctx->fname could be released in query_addrrset(), so save a copy of
* it here in case we need it.
*/
dns_fixedname_init(&qctx->dsname);
dns_name_copy(qctx->fname, dns_fixedname_name(&qctx->dsname));
/*
* This is the best answer.
*/
qctx->client->query.isreferral = true;
if (!dns_db_iscache(qctx->db) && qctx->client->query.gluedb == NULL) {
dns_db_attach(qctx->db, &qctx->client->query.gluedb);
detach = true;
}
/*
* We must ensure NOADDITIONAL is off, because the generation of
* additional data is required in delegations.
*/
qctx->client->query.attributes &= ~NS_QUERYATTR_NOADDITIONAL;
if (WANTDNSSEC(qctx->client) && qctx->sigrdataset != NULL) {
sigrdatasetp = &qctx->sigrdataset;
}
query_addrrset(qctx, &qctx->fname, &qctx->rdataset, sigrdatasetp,
qctx->dbuf, DNS_SECTION_AUTHORITY);
if (detach) {
dns_db_detach(&qctx->client->query.gluedb);
}
/*
* Add DS/NSEC(3) record(s) if needed.
*/
query_addds(qctx);
return ns_query_done(qctx);
cleanup:
return result;
}
/*%
* Handle a delegation response from an authoritative lookup. This
* may trigger additional lookups, e.g. from the cache database to
* see if we have a better answer; if that is not allowed, return the
* delegation to the client and call ns_query_done().
*/
static isc_result_t
query_zone_delegation(query_ctx_t *qctx) {
isc_result_t result = ISC_R_UNSET;
CALL_HOOK(NS_QUERY_ZONE_DELEGATION_BEGIN, qctx);
/*
* If the query type is DS, look to see if we are
* authoritative for the child zone
*/
if (!RECURSIONOK(qctx->client) &&
(qctx->options.noexact && qctx->qtype == dns_rdatatype_ds))
{
dns_db_t *tdb = NULL;
dns_zone_t *tzone = NULL;
dns_dbversion_t *tversion = NULL;
dns_getdb_options_t options = { .partial = true };
result = query_getzonedb(qctx->client,
qctx->client->query.qname, qctx->qtype,
options, &tzone, &tdb, &tversion);
if (result != ISC_R_SUCCESS) {
if (tdb != NULL) {
dns_db_detach(&tdb);
}
if (tzone != NULL) {
dns_zone_detach(&tzone);
}
} else {
qctx->options.noexact = false;
ns_client_putrdataset(qctx->client, &qctx->rdataset);
if (qctx->sigrdataset != NULL) {
ns_client_putrdataset(qctx->client,
&qctx->sigrdataset);
}
if (qctx->fname != NULL) {
ns_client_releasename(qctx->client,
&qctx->fname);
}
if (qctx->node != NULL) {
dns_db_detachnode(qctx->db, &qctx->node);
}
if (qctx->db != NULL) {
dns_db_detach(&qctx->db);
}
if (qctx->zone != NULL) {
dns_zone_detach(&qctx->zone);
}
qctx->version = NULL;
RESTORE(qctx->version, tversion);
RESTORE(qctx->db, tdb);
RESTORE(qctx->zone, tzone);
qctx->authoritative = true;
return query_lookup(qctx);
}
}
if (USECACHE(qctx->client) &&
(RECURSIONOK(qctx->client) ||
(qctx->zone != NULL &&
dns_zone_gettype(qctx->zone) == dns_zone_mirror)))
{
/*
* We might have a better answer or delegation in the
* cache. We'll remember the current values of fname,
* rdataset, and sigrdataset. We'll then go looking for
* QNAME in the cache. If we find something better, we'll
* use it instead. If not, then query_lookup() calls
* query_notfound() which calls query_delegation(), and
* we'll restore these values there.
*/
ns_client_keepname(qctx->client, qctx->fname, qctx->dbuf);
SAVE(qctx->zdb, qctx->db);
SAVE(qctx->znode, qctx->node);
SAVE(qctx->zfname, qctx->fname);
SAVE(qctx->zversion, qctx->version);
SAVE(qctx->zrdataset, qctx->rdataset);
SAVE(qctx->zsigrdataset, qctx->sigrdataset);
dns_db_attach(qctx->view->cachedb, &qctx->db);
qctx->is_zone = false;
/*
* Since 'qctx->is_zone' is now false, we should reconsider
* setting the 'stalefirst' option, which is usually set in
* the beginning in ns__query_start().
*/
if (qctx->view->staleanswerclienttimeout == 0 &&
dns_view_staleanswerenabled(qctx->view))
{
qctx->options.stalefirst = true;
}
result = query_lookup(qctx);
/*
* After fetch completes, this option is not expected to be set.
*/
qctx->options.stalefirst = false;
return result;
}
return query_prepare_delegation_response(qctx);
cleanup:
return result;
}
/*%
* Handle delegation responses, including root referrals.
*
* If the delegation was returned from authoritative data,
* call query_zone_delgation(). Otherwise, we can start
* recursion if allowed; or else return the delegation to the
* client and call ns_query_done().
*/
static isc_result_t
query_delegation(query_ctx_t *qctx) {
isc_result_t result = ISC_R_UNSET;
CCTRACE(ISC_LOG_DEBUG(3), "query_delegation");
CALL_HOOK(NS_QUERY_DELEGATION_BEGIN, qctx);
qctx->authoritative = false;
if (qctx->is_zone) {
return query_zone_delegation(qctx);
}
if (qctx->zfname != NULL &&
(!dns_name_issubdomain(qctx->fname, qctx->zfname) ||
(qctx->is_staticstub_zone &&
dns_name_equal(qctx->fname, qctx->zfname))))
{
/*
* In the following cases use "authoritative"
* data instead of the cache delegation:
* 1. We've already got a delegation from
* authoritative data, and it is better
* than what we found in the cache.
* (See the comment above.)
* 2. The query name matches the origin name
* of a static-stub zone. This needs to be
* considered for the case where the NS of
* the static-stub zone and the cached NS
* are different. We still need to contact
* the nameservers configured in the
* static-stub zone.
*/
ns_client_releasename(qctx->client, &qctx->fname);
/*
* We've already done ns_client_keepname() on
* qctx->zfname, so we must set dbuf to NULL to
* prevent query_addrrset() from trying to
* call ns_client_keepname() again.
*/
qctx->dbuf = NULL;
ns_client_putrdataset(qctx->client, &qctx->rdataset);
if (qctx->sigrdataset != NULL) {
ns_client_putrdataset(qctx->client, &qctx->sigrdataset);
}
qctx->version = NULL;
dns_db_detachnode(qctx->db, &qctx->node);
dns_db_detach(&qctx->db);
RESTORE(qctx->db, qctx->zdb);
RESTORE(qctx->node, qctx->znode);
RESTORE(qctx->fname, qctx->zfname);
RESTORE(qctx->version, qctx->zversion);
RESTORE(qctx->rdataset, qctx->zrdataset);
RESTORE(qctx->sigrdataset, qctx->zsigrdataset);
}
result = query_delegation_recurse(qctx);
if (result != ISC_R_COMPLETE) {
return result;
}
return query_prepare_delegation_response(qctx);
cleanup:
return result;
}
/*%
* Handle recursive queries that are triggered as part of the
* delegation process.
*/
static isc_result_t
query_delegation_recurse(query_ctx_t *qctx) {
isc_result_t result = ISC_R_UNSET;
dns_name_t *qname = qctx->client->query.qname;
CCTRACE(ISC_LOG_DEBUG(3), "query_delegation_recurse");
if (!RECURSIONOK(qctx->client)) {
return ISC_R_COMPLETE;
}
CALL_HOOK(NS_QUERY_DELEGATION_RECURSE_BEGIN, qctx);
/*
* We have a delegation and recursion is allowed,
* so we call ns_query_recurse() to follow it.
* This phase of the query processing is done;
* we'll resume via fetch_callback() and
* query_resume() when the recursion is complete.
*/
INSIST(!REDIRECT(qctx->client));
if (dns_rdatatype_atparent(qctx->type)) {
/*
* Parent is authoritative for this RDATA type (i.e. DS).
*/
result = ns_query_recurse(qctx->client, qctx->qtype, qname,
NULL, NULL, qctx->resuming);
} else if (qctx->dns64) {
/*
* Look up an A record so we can synthesize DNS64.
*/
result = ns_query_recurse(qctx->client, dns_rdatatype_a, qname,
NULL, NULL, qctx->resuming);
} else {
/*
* Any other recursion.
*/
result = ns_query_recurse(qctx->client, qctx->qtype, qname,
qctx->fname, qctx->rdataset,
qctx->resuming);
}
if (result == ISC_R_SUCCESS) {
qctx->client->query.attributes |= NS_QUERYATTR_RECURSING;
if (qctx->dns64) {
qctx->client->query.attributes |= NS_QUERYATTR_DNS64;
}
if (qctx->dns64_exclude) {
qctx->client->query.attributes |=
NS_QUERYATTR_DNS64EXCLUDE;
}
} else if (query_usestale(qctx, result)) {
/*
* If serve-stale is enabled, query_usestale() already set up
* 'qctx' for looking up a stale response.
*/
return query_lookup(qctx);
} else {
QUERY_ERROR(qctx, result);
}
return ns_query_done(qctx);
cleanup:
return result;
}
/*%
* Add DS/NSEC(3) record(s) if needed.
*/
static void
query_addds(query_ctx_t *qctx) {
ns_client_t *client = qctx->client;
dns_fixedname_t fixed;
dns_name_t *fname = NULL;
dns_name_t *name;
dns_rdataset_t *rdataset = NULL, *sigrdataset = NULL;
isc_buffer_t *dbuf, b;
isc_result_t result;
unsigned int count;
CTRACE(ISC_LOG_DEBUG(3), "query_addds");
/*
* DS not needed.
*/
if (!WANTDNSSEC(client)) {
return;
}
/*
* We'll need some resources...
*/
rdataset = ns_client_newrdataset(client);
sigrdataset = ns_client_newrdataset(client);
/*
* Look for the DS record, which may or may not be present.
*/
result = dns_db_findrdataset(qctx->db, qctx->node, qctx->version,
dns_rdatatype_ds, 0, client->inner.now,
rdataset, sigrdataset);
/*
* If we didn't find it, look for an NSEC.
*/
if (result == ISC_R_NOTFOUND) {
result = dns_db_findrdataset(
qctx->db, qctx->node, qctx->version, dns_rdatatype_nsec,
0, client->inner.now, rdataset, sigrdataset);
}
if (result != ISC_R_SUCCESS && result != ISC_R_NOTFOUND) {
goto addnsec3;
}
if (!dns_rdataset_isassociated(rdataset) ||
!dns_rdataset_isassociated(sigrdataset))
{
goto addnsec3;
}
/*
* Find the delegation in the response message - it is not necessarily
* the first name in the AUTHORITY section when wildcard processing is
* involved.
*/
MSG_SECTION_FOREACH (client->message, DNS_SECTION_AUTHORITY, rname) {
result = dns_message_findtype(rname, dns_rdatatype_ns, 0, NULL);
if (result == ISC_R_SUCCESS) {
/*
* Add the relevant RRset (DS or NSEC) to the
* delegation.
*/
query_addrrset(qctx, &rname, &rdataset, &sigrdataset,
NULL, DNS_SECTION_AUTHORITY);
break;
}
}
goto cleanup;
addnsec3:
if (!dns_db_iszone(qctx->db)) {
goto cleanup;
}
/*
* Add the NSEC3 which proves the DS does not exist.
*/
dbuf = ns_client_getnamebuf(client);
fname = ns_client_newname(client, dbuf, &b);
dns_fixedname_init(&fixed);
if (dns_rdataset_isassociated(rdataset)) {
dns_rdataset_disassociate(rdataset);
}
if (dns_rdataset_isassociated(sigrdataset)) {
dns_rdataset_disassociate(sigrdataset);
}
name = dns_fixedname_name(&qctx->dsname);
query_findclosestnsec3(name, qctx->db, qctx->version, client, rdataset,
sigrdataset, fname, true,
dns_fixedname_name(&fixed));
if (!dns_rdataset_isassociated(rdataset)) {
goto cleanup;
}
query_addrrset(qctx, &fname, &rdataset, &sigrdataset, dbuf,
DNS_SECTION_AUTHORITY);
/*
* Did we find the closest provable encloser instead?
* If so add the nearest to the closest provable encloser.
*/
if (!dns_name_equal(name, dns_fixedname_name(&fixed))) {
count = dns_name_countlabels(dns_fixedname_name(&fixed)) + 1;
dns_name_getlabelsequence(name,
dns_name_countlabels(name) - count,
count, dns_fixedname_name(&fixed));
fixfname(client, &fname, &dbuf, &b);
fixrdataset(client, &rdataset);
fixrdataset(client, &sigrdataset);
if (fname == NULL || rdataset == NULL || sigrdataset == NULL) {
goto cleanup;
}
query_findclosestnsec3(dns_fixedname_name(&fixed), qctx->db,
qctx->version, client, rdataset,
sigrdataset, fname, false, NULL);
if (!dns_rdataset_isassociated(rdataset)) {
goto cleanup;
}
query_addrrset(qctx, &fname, &rdataset, &sigrdataset, dbuf,
DNS_SECTION_AUTHORITY);
}
cleanup:
if (rdataset != NULL) {
ns_client_putrdataset(client, &rdataset);
}
if (sigrdataset != NULL) {
ns_client_putrdataset(client, &sigrdataset);
}
if (fname != NULL) {
ns_client_releasename(client, &fname);
}
}
/*%
* Handle authoritative NOERROR/NODATA responses.
*/
static isc_result_t
query_nodata(query_ctx_t *qctx, isc_result_t res) {
isc_result_t result = res;
CCTRACE(ISC_LOG_DEBUG(3), "query_nodata");
CALL_HOOK(NS_QUERY_NODATA_BEGIN, qctx);
#ifdef dns64_bis_return_excluded_addresses
if (qctx->dns64)
#else /* ifdef dns64_bis_return_excluded_addresses */
if (qctx->dns64 && !qctx->dns64_exclude)
#endif /* ifdef dns64_bis_return_excluded_addresses */
{
isc_buffer_t b;
/*
* Restore the answers from the previous AAAA lookup.
*/
if (qctx->rdataset != NULL) {
ns_client_putrdataset(qctx->client, &qctx->rdataset);
}
if (qctx->sigrdataset != NULL) {
ns_client_putrdataset(qctx->client, &qctx->sigrdataset);
}
RESTORE(qctx->rdataset, qctx->client->query.dns64_aaaa);
RESTORE(qctx->sigrdataset, qctx->client->query.dns64_sigaaaa);
if (qctx->fname == NULL) {
qctx->dbuf = ns_client_getnamebuf(qctx->client);
qctx->fname = ns_client_newname(qctx->client,
qctx->dbuf, &b);
}
dns_name_copy(qctx->client->query.qname, qctx->fname);
qctx->dns64 = false;
#ifdef dns64_bis_return_excluded_addresses
/*
* Resume the diverted processing of the AAAA response?
*/
if (qctx->dns64_exclude) {
return query_prepresponse(qctx);
}
#endif /* ifdef dns64_bis_return_excluded_addresses */
} else if ((result == DNS_R_NXRRSET || result == DNS_R_NCACHENXRRSET) &&
!ISC_LIST_EMPTY(qctx->view->dns64) && !qctx->nxrewrite &&
qctx->client->message->rdclass == dns_rdataclass_in &&
qctx->qtype == dns_rdatatype_aaaa)
{
/*
* Look to see if there are A records for this name.
*/
switch (result) {
case DNS_R_NCACHENXRRSET:
/*
* This is from the negative cache; if the ttl is
* zero, we need to work out whether we have just
* decremented to zero or there was no negative
* cache ttl in the answer.
*/
if (qctx->rdataset->ttl != 0) {
qctx->client->query.dns64_ttl =
qctx->rdataset->ttl;
break;
}
if (dns_rdataset_first(qctx->rdataset) == ISC_R_SUCCESS)
{
qctx->client->query.dns64_ttl = 0;
}
break;
case DNS_R_NXRRSET:
qctx->client->query.dns64_ttl =
dns64_ttl(qctx->db, qctx->version);
break;
default:
UNREACHABLE();
}
SAVE(qctx->client->query.dns64_aaaa, qctx->rdataset);
SAVE(qctx->client->query.dns64_sigaaaa, qctx->sigrdataset);
ns_client_releasename(qctx->client, &qctx->fname);
dns_db_detachnode(qctx->db, &qctx->node);
qctx->type = qctx->qtype = dns_rdatatype_a;
qctx->dns64 = true;
return query_lookup(qctx);
}
if (qctx->is_zone) {
return query_sign_nodata(qctx);
} else {
/*
* We don't call query_addrrset() because we don't need any
* of its extra features (and things would probably break!).
*/
if (dns_rdataset_isassociated(qctx->rdataset)) {
ns_client_keepname(qctx->client, qctx->fname,
qctx->dbuf);
dns_message_addname(qctx->client->message, qctx->fname,
DNS_SECTION_AUTHORITY);
ISC_LIST_APPEND(qctx->fname->list, qctx->rdataset,
link);
qctx->fname = NULL;
qctx->rdataset = NULL;
}
}
return ns_query_done(qctx);
cleanup:
return result;
}
/*%
* Add RRSIGs for NOERROR/NODATA responses when answering authoritatively.
*/
isc_result_t
query_sign_nodata(query_ctx_t *qctx) {
isc_result_t result;
CCTRACE(ISC_LOG_DEBUG(3), "query_sign_nodata");
/*
* Look for a NSEC3 record if we don't have a NSEC record.
*/
if (qctx->redirected) {
return ns_query_done(qctx);
}
if (!dns_rdataset_isassociated(qctx->rdataset) &&
WANTDNSSEC(qctx->client))
{
if (!qctx->fname->attributes.wildcard) {
dns_name_t *found;
dns_name_t *qname;
dns_fixedname_t fixed;
isc_buffer_t b;
found = dns_fixedname_initname(&fixed);
qname = qctx->client->query.qname;
query_findclosestnsec3(qname, qctx->db, qctx->version,
qctx->client, qctx->rdataset,
qctx->sigrdataset, qctx->fname,
true, found);
/*
* Did we find the closest provable encloser
* instead? If so add the nearest to the
* closest provable encloser.
*/
if (dns_rdataset_isassociated(qctx->rdataset) &&
!dns_name_equal(qname, found) &&
(((qctx->client->manager->sctx->options &
NS_SERVER_NONEAREST) == 0) ||
qctx->qtype == dns_rdatatype_ds))
{
unsigned int count;
unsigned int skip;
/*
* Add the closest provable encloser.
*/
query_addrrset(qctx, &qctx->fname,
&qctx->rdataset,
&qctx->sigrdataset, qctx->dbuf,
DNS_SECTION_AUTHORITY);
count = dns_name_countlabels(found) + 1;
skip = dns_name_countlabels(qname) - count;
dns_name_getlabelsequence(qname, skip, count,
found);
/*
* These ensure fname, rdataset, and
* sigrdataset are non-NULL
*/
fixfname(qctx->client, &qctx->fname,
&qctx->dbuf, &b);
fixrdataset(qctx->client, &qctx->rdataset);
fixrdataset(qctx->client, &qctx->sigrdataset);
/*
* 'nearest' doesn't exist so
* 'exist' is set to false.
*/
query_findclosestnsec3(
found, qctx->db, qctx->version,
qctx->client, qctx->rdataset,
qctx->sigrdataset, qctx->fname, false,
NULL);
}
} else {
ns_client_releasename(qctx->client, &qctx->fname);
query_addwildcardproof(qctx, false, true);
}
}
if (dns_rdataset_isassociated(qctx->rdataset)) {
/*
* If we've got a NSEC record, we need to save the
* name now because we're going call query_addsoa()
* below, and it needs to use the name buffer.
*/
ns_client_keepname(qctx->client, qctx->fname, qctx->dbuf);
} else if (qctx->fname != NULL) {
/*
* We're not going to use fname, and need to release
* our hold on the name buffer so query_addsoa()
* may use it.
*/
ns_client_releasename(qctx->client, &qctx->fname);
}
/*
* The RPZ SOA has already been added to the additional section
* if this was an RPZ rewrite, but if it wasn't, add it now.
*/
if (!qctx->nxrewrite) {
result = query_addsoa(qctx, UINT32_MAX, DNS_SECTION_AUTHORITY);
if (result != ISC_R_SUCCESS) {
QUERY_ERROR(qctx, result);
return ns_query_done(qctx);
}
}
/*
* Add NSEC record if we found one.
*/
if (WANTDNSSEC(qctx->client) &&
dns_rdataset_isassociated(qctx->rdataset))
{
query_addnxrrsetnsec(qctx);
}
return ns_query_done(qctx);
}
static void
query_addnxrrsetnsec(query_ctx_t *qctx) {
ns_client_t *client = qctx->client;
dns_rdata_t sigrdata;
dns_rdata_rrsig_t sig;
unsigned int labels;
isc_buffer_t *dbuf, b;
dns_name_t *fname;
isc_result_t result;
INSIST(qctx->fname != NULL);
if (!qctx->fname->attributes.wildcard) {
query_addrrset(qctx, &qctx->fname, &qctx->rdataset,
&qctx->sigrdataset, NULL, DNS_SECTION_AUTHORITY);
return;
}
if (qctx->sigrdataset == NULL ||
!dns_rdataset_isassociated(qctx->sigrdataset))
{
return;
}
if (dns_rdataset_first(qctx->sigrdataset) != ISC_R_SUCCESS) {
return;
}
dns_rdata_init(&sigrdata);
dns_rdataset_current(qctx->sigrdataset, &sigrdata);
result = dns_rdata_tostruct(&sigrdata, &sig, NULL);
RUNTIME_CHECK(result == ISC_R_SUCCESS);
labels = dns_name_countlabels(qctx->fname);
if ((unsigned int)sig.labels + 1 >= labels) {
return;
}
query_addwildcardproof(qctx, true, false);
/*
* We'll need some resources...
*/
dbuf = ns_client_getnamebuf(client);
fname = ns_client_newname(client, dbuf, &b);
dns_name_split(qctx->fname, sig.labels + 1, NULL, fname);
/* This will succeed, since we've stripped labels. */
RUNTIME_CHECK(dns_name_concatenate(dns_wildcardname, fname, fname) ==
ISC_R_SUCCESS);
query_addrrset(qctx, &fname, &qctx->rdataset, &qctx->sigrdataset, dbuf,
DNS_SECTION_AUTHORITY);
}
/*%
* Handle NXDOMAIN and empty wildcard responses.
*/
static isc_result_t
query_nxdomain(query_ctx_t *qctx, isc_result_t result) {
dns_section_t section;
uint32_t ttl;
bool empty_wild = (result == DNS_R_EMPTYWILD);
CCTRACE(ISC_LOG_DEBUG(3), "query_nxdomain");
CALL_HOOK(NS_QUERY_NXDOMAIN_BEGIN, qctx);
INSIST(qctx->is_zone || REDIRECT(qctx->client));
if (!empty_wild) {
result = query_redirect(qctx, result);
if (result != ISC_R_COMPLETE) {
return result;
}
}
if (dns_rdataset_isassociated(qctx->rdataset)) {
/*
* If we've got a NSEC record, we need to save the
* name now because we're going call query_addsoa()
* below, and it needs to use the name buffer.
*/
ns_client_keepname(qctx->client, qctx->fname, qctx->dbuf);
} else if (qctx->fname != NULL) {
/*
* We're not going to use fname, and need to release
* our hold on the name buffer so query_addsoa()
* may use it.
*/
ns_client_releasename(qctx->client, &qctx->fname);
}
/*
* Add SOA to the additional section if generated by a
* RPZ rewrite.
*
* If the query was for a SOA record force the
* ttl to zero so that it is possible for clients to find
* the containing zone of an arbitrary name with a stub
* resolver and not have it cached.
*/
section = qctx->nxrewrite ? DNS_SECTION_ADDITIONAL
: DNS_SECTION_AUTHORITY;
ttl = UINT32_MAX;
if (!qctx->nxrewrite && qctx->qtype == dns_rdatatype_soa &&
qctx->zone != NULL && dns_zone_getzeronosoattl(qctx->zone))
{
ttl = 0;
}
if (!qctx->nxrewrite ||
(qctx->rpz_st != NULL && qctx->rpz_st->m.rpz->addsoa))
{
result = query_addsoa(qctx, ttl, section);
if (result != ISC_R_SUCCESS) {
QUERY_ERROR(qctx, result);
return ns_query_done(qctx);
}
}
if (WANTDNSSEC(qctx->client)) {
/*
* Add NSEC record if we found one.
*/
if (dns_rdataset_isassociated(qctx->rdataset)) {
query_addrrset(qctx, &qctx->fname, &qctx->rdataset,
&qctx->sigrdataset, NULL,
DNS_SECTION_AUTHORITY);
}
query_addwildcardproof(qctx, false, false);
}
/*
* Set message rcode.
*/
if (empty_wild) {
qctx->client->message->rcode = dns_rcode_noerror;
} else {
qctx->client->message->rcode = dns_rcode_nxdomain;
}
return ns_query_done(qctx);
cleanup:
return result;
}
/*
* Handle both types of NXDOMAIN redirection, calling redirect()
* (which implements type redirect zones) and redirect2() (which
* implements recursive nxdomain-redirect lookups).
*
* Any result code other than ISC_R_COMPLETE means redirection was
* successful and the result code should be returned up the call stack.
*
* ISC_R_COMPLETE means we reached the end of this function without
* redirecting, so query processing should continue past it.
*/
static isc_result_t
query_redirect(query_ctx_t *qctx, isc_result_t saved_result) {
isc_result_t result;
CCTRACE(ISC_LOG_DEBUG(3), "query_redirect");
result = redirect(qctx->client, qctx->fname, qctx->rdataset,
&qctx->node, &qctx->db, &qctx->version, qctx->type);
switch (result) {
case ISC_R_SUCCESS:
inc_stats(qctx->client, ns_statscounter_nxdomainredirect);
return query_prepresponse(qctx);
case DNS_R_NXRRSET:
qctx->redirected = true;
qctx->is_zone = true;
return query_nodata(qctx, DNS_R_NXRRSET);
case DNS_R_NCACHENXRRSET:
qctx->redirected = true;
qctx->is_zone = false;
return query_ncache(qctx, DNS_R_NCACHENXRRSET);
default:
break;
}
result = redirect2(qctx->client, qctx->fname, qctx->rdataset,
&qctx->node, &qctx->db, &qctx->version, qctx->type,
&qctx->is_zone);
switch (result) {
case ISC_R_SUCCESS:
inc_stats(qctx->client, ns_statscounter_nxdomainredirect);
return query_prepresponse(qctx);
case DNS_R_CONTINUE:
inc_stats(qctx->client,
ns_statscounter_nxdomainredirect_rlookup);
SAVE(qctx->client->query.redirect.db, qctx->db);
SAVE(qctx->client->query.redirect.node, qctx->node);
SAVE(qctx->client->query.redirect.zone, qctx->zone);
qctx->client->query.redirect.qtype = qctx->qtype;
INSIST(qctx->rdataset != NULL);
SAVE(qctx->client->query.redirect.rdataset, qctx->rdataset);
SAVE(qctx->client->query.redirect.sigrdataset,
qctx->sigrdataset);
qctx->client->query.redirect.result = saved_result;
dns_name_copy(qctx->fname, qctx->client->query.redirect.fname);
qctx->client->query.redirect.authoritative =
qctx->authoritative;
qctx->client->query.redirect.is_zone = qctx->is_zone;
return ns_query_done(qctx);
case DNS_R_NXRRSET:
qctx->redirected = true;
qctx->is_zone = true;
return query_nodata(qctx, DNS_R_NXRRSET);
case DNS_R_NCACHENXRRSET:
qctx->redirected = true;
qctx->is_zone = false;
return query_ncache(qctx, DNS_R_NCACHENXRRSET);
default:
break;
}
return ISC_R_COMPLETE;
}
/*%
* Logging function to be passed to dns_nsec_noexistnodata.
*/
static void
log_noexistnodata(void *val, int level, const char *fmt, ...) {
query_ctx_t *qctx = val;
va_list ap;
va_start(ap, fmt);
ns_client_logv(qctx->client, NS_LOGCATEGORY_QUERIES, NS_LOGMODULE_QUERY,
level, fmt, ap);
va_end(ap);
}
static dns_ttl_t
query_synthttl(dns_rdataset_t *soardataset, dns_rdataset_t *sigsoardataset,
dns_rdataset_t *p1rdataset, dns_rdataset_t *sigp1rdataset,
dns_rdataset_t *p2rdataset, dns_rdataset_t *sigp2rdataset) {
dns_rdata_soa_t soa;
dns_rdata_t rdata = DNS_RDATA_INIT;
dns_ttl_t ttl;
isc_result_t result;
REQUIRE(soardataset != NULL);
REQUIRE(sigsoardataset != NULL);
REQUIRE(p1rdataset != NULL);
REQUIRE(sigp1rdataset != NULL);
result = dns_rdataset_first(soardataset);
RUNTIME_CHECK(result == ISC_R_SUCCESS);
dns_rdataset_current(soardataset, &rdata);
result = dns_rdata_tostruct(&rdata, &soa, NULL);
RUNTIME_CHECK(result == ISC_R_SUCCESS);
ttl = ISC_MIN(soa.minimum, soardataset->ttl);
ttl = ISC_MIN(ttl, sigsoardataset->ttl);
ttl = ISC_MIN(ttl, p1rdataset->ttl);
ttl = ISC_MIN(ttl, sigp1rdataset->ttl);
if (p2rdataset != NULL) {
ttl = ISC_MIN(ttl, p2rdataset->ttl);
}
if (sigp2rdataset != NULL) {
ttl = ISC_MIN(ttl, sigp2rdataset->ttl);
}
return ttl;
}
/*
* Synthesize a NODATA response from the SOA and covering NSEC in cache.
*/
static isc_result_t
query_synthnodata(query_ctx_t *qctx, const dns_name_t *signer,
dns_rdataset_t **soardatasetp,
dns_rdataset_t **sigsoardatasetp) {
dns_name_t *name = NULL;
dns_ttl_t ttl;
isc_buffer_t *dbuf, b;
/*
* Determine the correct TTL to use for the SOA and RRSIG
*/
ttl = query_synthttl(*soardatasetp, *sigsoardatasetp, qctx->rdataset,
qctx->sigrdataset, NULL, NULL);
(*soardatasetp)->ttl = (*sigsoardatasetp)->ttl = ttl;
/*
* We want the SOA record to be first, so save the
* NODATA proof's name now or else discard it.
*/
if (WANTDNSSEC(qctx->client)) {
ns_client_keepname(qctx->client, qctx->fname, qctx->dbuf);
} else {
ns_client_releasename(qctx->client, &qctx->fname);
}
dbuf = ns_client_getnamebuf(qctx->client);
name = ns_client_newname(qctx->client, dbuf, &b);
dns_name_copy(signer, name);
/*
* Add SOA record. Omit the RRSIG if DNSSEC was not requested.
*/
if (!WANTDNSSEC(qctx->client)) {
sigsoardatasetp = NULL;
}
query_addrrset(qctx, &name, soardatasetp, sigsoardatasetp, dbuf,
DNS_SECTION_AUTHORITY);
if (WANTDNSSEC(qctx->client)) {
/*
* Add NODATA proof.
*/
query_addrrset(qctx, &qctx->fname, &qctx->rdataset,
&qctx->sigrdataset, NULL, DNS_SECTION_AUTHORITY);
}
inc_stats(qctx->client, ns_statscounter_nodatasynth);
if (name != NULL) {
ns_client_releasename(qctx->client, &name);
}
return ISC_R_SUCCESS;
}
/*
* Synthesize a wildcard answer using the contents of 'rdataset'.
* qctx contains the NODATA proof.
*/
static isc_result_t
query_synthwildcard(query_ctx_t *qctx, dns_rdataset_t *rdataset,
dns_rdataset_t *sigrdataset) {
dns_name_t *name = NULL;
isc_buffer_t *dbuf, b;
dns_rdataset_t *cloneset = NULL, *clonesigset = NULL;
dns_rdataset_t **sigrdatasetp;
CCTRACE(ISC_LOG_DEBUG(3), "query_synthwildcard");
/*
* We want the answer to be first, so save the
* NOQNAME proof's name now or else discard it.
*/
if (WANTDNSSEC(qctx->client)) {
ns_client_keepname(qctx->client, qctx->fname, qctx->dbuf);
} else {
ns_client_releasename(qctx->client, &qctx->fname);
}
dbuf = ns_client_getnamebuf(qctx->client);
name = ns_client_newname(qctx->client, dbuf, &b);
dns_name_copy(qctx->client->query.qname, name);
cloneset = ns_client_newrdataset(qctx->client);
dns_rdataset_clone(rdataset, cloneset);
/*
* Add answer RRset. Omit the RRSIG if DNSSEC was not requested.
*/
if (WANTDNSSEC(qctx->client)) {
clonesigset = ns_client_newrdataset(qctx->client);
dns_rdataset_clone(sigrdataset, clonesigset);
sigrdatasetp = &clonesigset;
} else {
sigrdatasetp = NULL;
}
query_addrrset(qctx, &name, &cloneset, sigrdatasetp, dbuf,
DNS_SECTION_ANSWER);
if (WANTDNSSEC(qctx->client)) {
/*
* Add NOQNAME proof.
*/
query_addrrset(qctx, &qctx->fname, &qctx->rdataset,
&qctx->sigrdataset, NULL, DNS_SECTION_AUTHORITY);
}
inc_stats(qctx->client, ns_statscounter_wildcardsynth);
if (name != NULL) {
ns_client_releasename(qctx->client, &name);
}
if (cloneset != NULL) {
ns_client_putrdataset(qctx->client, &cloneset);
}
if (clonesigset != NULL) {
ns_client_putrdataset(qctx->client, &clonesigset);
}
return ISC_R_SUCCESS;
}
/*
* Add a synthesized CNAME record from the wildard RRset (rdataset)
* and NODATA proof by calling query_synthwildcard then setup to
* follow the CNAME.
*/
static isc_result_t
query_synthcnamewildcard(query_ctx_t *qctx, dns_rdataset_t *rdataset,
dns_rdataset_t *sigrdataset) {
isc_result_t result;
dns_name_t *tname = NULL;
dns_rdata_t rdata = DNS_RDATA_INIT;
dns_rdata_cname_t cname;
result = query_synthwildcard(qctx, rdataset, sigrdataset);
if (result != ISC_R_SUCCESS) {
return result;
}
qctx->client->query.attributes |= NS_QUERYATTR_PARTIALANSWER;
/*
* Reset qname to be the target name of the CNAME and restart
* the query.
*/
dns_message_gettempname(qctx->client->message, &tname);
result = dns_rdataset_first(rdataset);
if (result != ISC_R_SUCCESS) {
dns_message_puttempname(qctx->client->message, &tname);
return result;
}
dns_rdataset_current(rdataset, &rdata);
result = dns_rdata_tostruct(&rdata, &cname, NULL);
RUNTIME_CHECK(result == ISC_R_SUCCESS);
dns_rdata_reset(&rdata);
if (dns_name_equal(qctx->client->query.qname, &cname.cname)) {
dns_message_puttempname(qctx->client->message, &tname);
dns_rdata_freestruct(&cname);
return ISC_R_SUCCESS;
}
dns_name_copy(&cname.cname, tname);
dns_rdata_freestruct(&cname);
ns_client_qnamereplace(qctx->client, tname);
qctx->want_restart = true;
if (!WANTRECURSION(qctx->client)) {
qctx->options.nolog = true;
}
return result;
}
/*
* Synthesize a NXDOMAIN or NODATA response from qctx (which contains the
* NOQNAME proof), nowild + nowildrdataset + signowildrdataset (which
* contains the NOWILDCARD proof or NODATA at wildcard) and
* signer + soardatasetp + sigsoardatasetp which contain the
* SOA record + RRSIG for the negative answer.
*/
static isc_result_t
query_synthnxdomainnodata(query_ctx_t *qctx, bool nodata, dns_name_t *nowild,
dns_rdataset_t *nowildrdataset,
dns_rdataset_t *signowildrdataset, dns_name_t *signer,
dns_rdataset_t **soardatasetp,
dns_rdataset_t **sigsoardatasetp) {
dns_name_t *name = NULL;
dns_ttl_t ttl;
isc_buffer_t *dbuf, b;
dns_rdataset_t *cloneset = NULL, *clonesigset = NULL;
CCTRACE(ISC_LOG_DEBUG(3), "query_synthnxdomain");
/*
* Determine the correct TTL to use for the SOA and RRSIG
*/
ttl = query_synthttl(*soardatasetp, *sigsoardatasetp, qctx->rdataset,
qctx->sigrdataset, nowildrdataset,
signowildrdataset);
(*soardatasetp)->ttl = (*sigsoardatasetp)->ttl = ttl;
/*
* We want the SOA record to be first, so save the
* NOQNAME proof's name now or else discard it.
*/
if (WANTDNSSEC(qctx->client)) {
ns_client_keepname(qctx->client, qctx->fname, qctx->dbuf);
} else {
ns_client_releasename(qctx->client, &qctx->fname);
}
dbuf = ns_client_getnamebuf(qctx->client);
name = ns_client_newname(qctx->client, dbuf, &b);
dns_name_copy(signer, name);
/*
* Add SOA record. Omit the RRSIG if DNSSEC was not requested.
*/
if (!WANTDNSSEC(qctx->client)) {
sigsoardatasetp = NULL;
}
query_addrrset(qctx, &name, soardatasetp, sigsoardatasetp, dbuf,
DNS_SECTION_AUTHORITY);
if (WANTDNSSEC(qctx->client)) {
/*
* Add NOQNAME proof.
*/
query_addrrset(qctx, &qctx->fname, &qctx->rdataset,
&qctx->sigrdataset, NULL, DNS_SECTION_AUTHORITY);
dbuf = ns_client_getnamebuf(qctx->client);
name = ns_client_newname(qctx->client, dbuf, &b);
dns_name_copy(nowild, name);
cloneset = ns_client_newrdataset(qctx->client);
clonesigset = ns_client_newrdataset(qctx->client);
dns_rdataset_clone(nowildrdataset, cloneset);
dns_rdataset_clone(signowildrdataset, clonesigset);
/*
* Add NOWILDCARD proof.
*/
query_addrrset(qctx, &name, &cloneset, &clonesigset, dbuf,
DNS_SECTION_AUTHORITY);
}
if (nodata) {
inc_stats(qctx->client, ns_statscounter_nodatasynth);
} else {
qctx->client->message->rcode = dns_rcode_nxdomain;
inc_stats(qctx->client, ns_statscounter_nxdomainsynth);
}
if (name != NULL) {
ns_client_releasename(qctx->client, &name);
}
if (cloneset != NULL) {
ns_client_putrdataset(qctx->client, &cloneset);
}
if (clonesigset != NULL) {
ns_client_putrdataset(qctx->client, &clonesigset);
}
return ISC_R_SUCCESS;
}
/*
* Check that all signer names in sigrdataset match the expected signer.
*/
static isc_result_t
checksignames(dns_name_t *signer, dns_rdataset_t *sigrdataset) {
DNS_RDATASET_FOREACH (sigrdataset) {
isc_result_t result;
dns_rdata_t rdata = DNS_RDATA_INIT;
dns_rdata_rrsig_t rrsig;
dns_rdataset_current(sigrdataset, &rdata);
result = dns_rdata_tostruct(&rdata, &rrsig, NULL);
RUNTIME_CHECK(result == ISC_R_SUCCESS);
if (dns_name_countlabels(signer) == 0) {
dns_name_copy(&rrsig.signer, signer);
} else if (!dns_name_equal(signer, &rrsig.signer)) {
return ISC_R_FAILURE;
}
}
return ISC_R_SUCCESS;
}
/*%
* Handle covering NSEC responses.
*
* Verify the NSEC record is appropriate for the QNAME; if not,
* redo the initial query without DNS_DBFIND_COVERINGNSEC.
*
* If the covering NSEC proves that the name exists but not the type,
* synthesize a NODATA response.
*
* If the name doesn't exist, compute the wildcard record and check whether
* the wildcard name exists or not. If we can't determine this, redo the
* initial query without DNS_DBFIND_COVERINGNSEC.
*
* If the wildcard name does not exist, compute the SOA name and look that
* up. If the SOA record does not exist, redo the initial query without
* DNS_DBFIND_COVERINGNSEC. If the SOA record exists, synthesize an
* NXDOMAIN response from the found records.
*
* If the wildcard name does exist, perform a lookup for the requested
* type at the wildcard name.
*/
static isc_result_t
query_coveringnsec(query_ctx_t *qctx) {
dns_db_t *db = NULL;
dns_clientinfo_t ci;
dns_clientinfomethods_t cm;
dns_dbnode_t *node = NULL;
dns_fixedname_t fixed;
dns_fixedname_t fnamespace;
dns_fixedname_t fnowild;
dns_fixedname_t fsigner;
dns_fixedname_t fwild;
dns_name_t *fname = NULL;
dns_name_t *namespace = NULL;
dns_name_t *nowild = NULL;
dns_name_t *signer = NULL;
dns_name_t *wild = NULL;
dns_name_t qname;
dns_rdataset_t *soardataset = NULL, *sigsoardataset = NULL;
dns_rdataset_t rdataset, sigrdataset;
bool done = false;
bool exists = true, data = true;
bool redirected = false;
isc_result_t result = ISC_R_SUCCESS;
unsigned int dboptions = qctx->client->query.dboptions;
unsigned int labels;
CCTRACE(ISC_LOG_DEBUG(3), "query_coveringnsec");
dns_name_init(&qname);
dns_rdataset_init(&rdataset);
dns_rdataset_init(&sigrdataset);
namespace = dns_fixedname_initname(&fnamespace);
/*
* Check that the NSEC record is from the correct namespace.
* For records that belong to the parent zone (i.e. DS),
* remove a label to find the correct namespace.
*/
dns_name_clone(qctx->client->query.qname, &qname);
labels = dns_name_countlabels(&qname);
if (dns_rdatatype_atparent(qctx->qtype) && labels > 1) {
dns_name_getlabelsequence(&qname, 1, labels - 1, &qname);
}
dns_view_sfd_find(qctx->view, &qname, namespace);
if (!dns_name_issubdomain(qctx->fname, namespace)) {
goto cleanup;
}
/*
* If we have no signer name, stop immediately.
*/
if (!dns_rdataset_isassociated(qctx->sigrdataset)) {
goto cleanup;
}
wild = dns_fixedname_initname(&fwild);
fname = dns_fixedname_initname(&fixed);
signer = dns_fixedname_initname(&fsigner);
nowild = dns_fixedname_initname(&fnowild);
dns_clientinfomethods_init(&cm, ns_client_sourceip);
dns_clientinfo_init(&ci, qctx->client, NULL);
/*
* All signer names must be the same to accept.
*/
result = checksignames(signer, qctx->sigrdataset);
if (result != ISC_R_SUCCESS) {
result = ISC_R_SUCCESS;
goto cleanup;
}
/*
* If NSEC or RRSIG are missing from the type map
* reject the NSEC RRset.
*/
if (!dns_nsec_requiredtypespresent(qctx->rdataset)) {
goto cleanup;
}
/*
* Check that we have the correct NOQNAME NSEC record.
*/
result = dns_nsec_noexistnodata(qctx->qtype, qctx->client->query.qname,
qctx->fname, qctx->rdataset, &exists,
&data, wild, log_noexistnodata, qctx);
if (result != ISC_R_SUCCESS || (exists && data)) {
goto cleanup;
}
if (exists) {
if (qctx->type == dns_rdatatype_any) { /* XXX not yet */
goto cleanup;
}
if (!ISC_LIST_EMPTY(qctx->view->dns64) &&
dns_rdatatype_isaddr(qctx->type)) /* XXX not yet */
{
goto cleanup;
}
if (!qctx->resuming && !STALE(qctx->rdataset) &&
qctx->rdataset->ttl == 0 && RECURSIONOK(qctx->client))
{
goto cleanup;
}
soardataset = ns_client_newrdataset(qctx->client);
sigsoardataset = ns_client_newrdataset(qctx->client);
/*
* Look for SOA record to construct NODATA response.
*/
dns_db_attach(qctx->db, &db);
result = dns_db_findext(db, signer, qctx->version,
dns_rdatatype_soa, dboptions,
qctx->client->inner.now, &node, fname,
&cm, &ci, soardataset, sigsoardataset);
if (result != ISC_R_SUCCESS) {
goto cleanup;
}
(void)query_synthnodata(qctx, signer, &soardataset,
&sigsoardataset);
done = true;
goto cleanup;
}
/*
* Look up the no-wildcard proof.
*/
dns_db_attach(qctx->db, &db);
result = dns_db_findext(db, wild, qctx->version, qctx->type,
dboptions | DNS_DBFIND_COVERINGNSEC,
qctx->client->inner.now, &node, nowild, &cm,
&ci, &rdataset, &sigrdataset);
if (rdataset.trust != dns_trust_secure ||
sigrdataset.trust != dns_trust_secure)
{
goto cleanup;
}
/*
* Zero TTL handling of wildcard record.
*
* We don't yet have code to handle synthesis and type ANY or dns64
* processing so we abort the synthesis here if there would be a
* interaction.
*/
switch (result) {
case ISC_R_SUCCESS:
if (qctx->type == dns_rdatatype_any) { /* XXX not yet */
goto cleanup;
}
if (!ISC_LIST_EMPTY(qctx->view->dns64) &&
dns_rdatatype_isaddr(qctx->type)) /* XXX not yet */
{
goto cleanup;
}
FALLTHROUGH;
case DNS_R_CNAME:
if (!qctx->resuming && !STALE(&rdataset) && rdataset.ttl == 0 &&
RECURSIONOK(qctx->client))
{
goto cleanup;
}
default:
break;
}
switch (result) {
case DNS_R_COVERINGNSEC:
/*
* Check that the covering NSEC record is from the right
* namespace.
*/
if (!dns_name_issubdomain(nowild, namespace)) {
goto cleanup;
}
result = dns_nsec_noexistnodata(qctx->qtype, wild, nowild,
&rdataset, &exists, &data, NULL,
log_noexistnodata, qctx);
if (result != ISC_R_SUCCESS || (exists && data)) {
goto cleanup;
}
break;
case ISC_R_SUCCESS: /* wild card match */
(void)query_synthwildcard(qctx, &rdataset, &sigrdataset);
done = true;
goto cleanup;
case DNS_R_CNAME: /* wild card cname */
(void)query_synthcnamewildcard(qctx, &rdataset, &sigrdataset);
done = true;
goto cleanup;
case DNS_R_NCACHENXRRSET: /* wild card nodata */
case DNS_R_NCACHENXDOMAIN: /* direct nxdomain */
default:
goto cleanup;
}
/*
* We now have the proof that we have an NXDOMAIN. Apply
* NXDOMAIN redirection if configured.
*/
result = query_redirect(qctx, DNS_R_COVERINGNSEC);
if (result != ISC_R_COMPLETE) {
redirected = true;
goto cleanup;
}
/*
* Must be signed to accept.
*/
if (!dns_rdataset_isassociated(&sigrdataset)) {
goto cleanup;
}
/*
* Check signer signer names again.
*/
result = checksignames(signer, &sigrdataset);
if (result != ISC_R_SUCCESS) {
result = ISC_R_SUCCESS;
goto cleanup;
}
if (node != NULL) {
dns_db_detachnode(db, &node);
}
soardataset = ns_client_newrdataset(qctx->client);
sigsoardataset = ns_client_newrdataset(qctx->client);
/*
* Look for SOA record to construct NXDOMAIN response.
*/
result = dns_db_findext(db, signer, qctx->version, dns_rdatatype_soa,
dboptions, qctx->client->inner.now, &node,
fname, &cm, &ci, soardataset, sigsoardataset);
if (result != ISC_R_SUCCESS) {
goto cleanup;
}
(void)query_synthnxdomainnodata(qctx, exists, nowild, &rdataset,
&sigrdataset, signer, &soardataset,
&sigsoardataset);
done = true;
cleanup:
if (dns_rdataset_isassociated(&rdataset)) {
dns_rdataset_disassociate(&rdataset);
}
if (dns_rdataset_isassociated(&sigrdataset)) {
dns_rdataset_disassociate(&sigrdataset);
}
if (soardataset != NULL) {
ns_client_putrdataset(qctx->client, &soardataset);
}
if (sigsoardataset != NULL) {
ns_client_putrdataset(qctx->client, &sigsoardataset);
}
if (db != NULL) {
if (node != NULL) {
dns_db_detachnode(db, &node);
}
dns_db_detach(&db);
}
if (redirected) {
return result;
}
if (!done) {
/*
* No covering NSEC was found; proceed with recursion.
*/
qctx->findcoveringnsec = false;
if (qctx->fname != NULL) {
ns_client_releasename(qctx->client, &qctx->fname);
}
if (qctx->node != NULL) {
dns_db_detachnode(qctx->db, &qctx->node);
}
ns_client_putrdataset(qctx->client, &qctx->rdataset);
if (qctx->sigrdataset != NULL) {
ns_client_putrdataset(qctx->client, &qctx->sigrdataset);
}
return query_lookup(qctx);
}
return ns_query_done(qctx);
}
/*%
* Handle negative cache responses, DNS_R_NCACHENXRRSET or
* DNS_R_NCACHENXDOMAIN. (Note: may also be called with result
* set to DNS_R_NXDOMAIN when handling DNS64 lookups.)
*/
static isc_result_t
query_ncache(query_ctx_t *qctx, isc_result_t result) {
INSIST(!qctx->is_zone);
INSIST(result == DNS_R_NCACHENXDOMAIN ||
result == DNS_R_NCACHENXRRSET || result == DNS_R_NXDOMAIN);
CCTRACE(ISC_LOG_DEBUG(3), "query_ncache");
CALL_HOOK(NS_QUERY_NCACHE_BEGIN, qctx);
qctx->authoritative = false;
if (result == DNS_R_NCACHENXDOMAIN) {
/*
* Set message rcode. (This is not done when
* result == DNS_R_NXDOMAIN because that means we're
* being called after a DNS64 lookup and don't want
* to update the rcode now.)
*/
qctx->client->message->rcode = dns_rcode_nxdomain;
/* Look for RFC 1918 leakage from Internet. */
if (qctx->qtype == dns_rdatatype_ptr &&
qctx->client->message->rdclass == dns_rdataclass_in &&
dns_name_countlabels(qctx->fname) == 7)
{
warn_rfc1918(qctx->client, qctx->fname, qctx->rdataset);
}
}
return query_nodata(qctx, result);
cleanup:
return result;
}
/*
* If we have a zero ttl from the cache, refetch.
*/
static isc_result_t
query_zerottl_refetch(query_ctx_t *qctx) {
isc_result_t result;
CCTRACE(ISC_LOG_DEBUG(3), "query_zerottl_refetch");
if (qctx->is_zone || qctx->resuming || STALE(qctx->rdataset) ||
qctx->rdataset->ttl != 0 || !RECURSIONOK(qctx->client))
{
return ISC_R_COMPLETE;
}
qctx_clean(qctx);
INSIST(!REDIRECT(qctx->client));
result = ns_query_recurse(qctx->client, qctx->qtype,
qctx->client->query.qname, NULL, NULL,
qctx->resuming);
if (result == ISC_R_SUCCESS) {
CALL_HOOK(NS_QUERY_ZEROTTL_RECURSE, qctx);
qctx->client->query.attributes |= NS_QUERYATTR_RECURSING;
if (qctx->dns64) {
qctx->client->query.attributes |= NS_QUERYATTR_DNS64;
}
if (qctx->dns64_exclude) {
qctx->client->query.attributes |=
NS_QUERYATTR_DNS64EXCLUDE;
}
} else {
/*
* There was a zero ttl from the cache, don't fallback to
* serve-stale lookup.
*/
QUERY_ERROR(qctx, result);
}
return ns_query_done(qctx);
cleanup:
return result;
}
/*
* Handle CNAME responses.
*/
static isc_result_t
query_cname(query_ctx_t *qctx) {
isc_result_t result = ISC_R_UNSET;
dns_name_t *tname = NULL;
dns_rdataset_t *trdataset = NULL;
dns_rdataset_t **sigrdatasetp = NULL;
dns_rdata_t rdata = DNS_RDATA_INIT;
dns_rdata_cname_t cname;
CCTRACE(ISC_LOG_DEBUG(3), "query_cname");
CALL_HOOK(NS_QUERY_CNAME_BEGIN, qctx);
result = query_zerottl_refetch(qctx);
if (result != ISC_R_COMPLETE) {
goto cleanup;
}
/*
* Keep a copy of the rdataset. We have to do this because
* query_addrrset may clear 'rdataset' (to prevent the
* cleanup code from cleaning it up).
*/
trdataset = qctx->rdataset;
/*
* Add the CNAME to the answer section.
*/
if (WANTDNSSEC(qctx->client) && qctx->sigrdataset != NULL) {
sigrdatasetp = &qctx->sigrdataset;
}
if (WANTDNSSEC(qctx->client) && qctx->fname->attributes.wildcard) {
dns_fixedname_init(&qctx->wildcardname);
dns_name_copy(qctx->fname,
dns_fixedname_name(&qctx->wildcardname));
qctx->need_wildcardproof = true;
}
if (NOQNAME(qctx->rdataset) && WANTDNSSEC(qctx->client)) {
qctx->noqname = qctx->rdataset;
} else {
qctx->noqname = NULL;
}
if (!qctx->is_zone && RECURSIONOK(qctx->client)) {
query_prefetch(qctx->client, qctx->fname, qctx->rdataset);
}
query_addrrset(qctx, &qctx->fname, &qctx->rdataset, sigrdatasetp,
qctx->dbuf, DNS_SECTION_ANSWER);
query_addnoqnameproof(qctx);
/*
* We set the PARTIALANSWER attribute so that if anything goes
* wrong later on, we'll return what we've got so far.
*/
qctx->client->query.attributes |= NS_QUERYATTR_PARTIALANSWER;
/*
* Reset qname to be the target name of the CNAME and restart
* the query.
*/
dns_message_gettempname(qctx->client->message, &tname);
result = dns_rdataset_first(trdataset);
if (result != ISC_R_SUCCESS) {
dns_message_puttempname(qctx->client->message, &tname);
(void)ns_query_done(qctx);
goto cleanup;
}
dns_rdataset_current(trdataset, &rdata);
result = dns_rdata_tostruct(&rdata, &cname, NULL);
RUNTIME_CHECK(result == ISC_R_SUCCESS);
dns_rdata_reset(&rdata);
dns_name_copy(&cname.cname, tname);
dns_rdata_freestruct(&cname);
ns_client_qnamereplace(qctx->client, tname);
qctx->want_restart = true;
if (!WANTRECURSION(qctx->client)) {
qctx->options.nolog = true;
}
query_addauth(qctx);
return ns_query_done(qctx);
cleanup:
return result;
}
/*
* Handle DNAME responses.
*/
static isc_result_t
query_dname(query_ctx_t *qctx) {
dns_name_t *tname, *prefix;
dns_rdata_t rdata = DNS_RDATA_INIT;
dns_rdata_dname_t dname;
dns_fixedname_t fixed;
dns_rdataset_t *trdataset;
dns_rdataset_t **sigrdatasetp = NULL;
dns_namereln_t namereln;
isc_buffer_t b;
int order;
isc_result_t result = ISC_R_UNSET;
unsigned int nlabels;
CCTRACE(ISC_LOG_DEBUG(3), "query_dname");
CALL_HOOK(NS_QUERY_DNAME_BEGIN, qctx);
/*
* Compare the current qname to the found name. We need
* to know how many labels and bits are in common because
* we're going to have to split qname later on.
*/
namereln = dns_name_fullcompare(qctx->client->query.qname, qctx->fname,
&order, &nlabels);
INSIST(namereln == dns_namereln_subdomain);
/*
* Keep a copy of the rdataset. We have to do this because
* query_addrrset may clear 'rdataset' (to prevent the
* cleanup code from cleaning it up).
*/
trdataset = qctx->rdataset;
/*
* Add the DNAME to the answer section.
*/
if (WANTDNSSEC(qctx->client) && qctx->sigrdataset != NULL) {
sigrdatasetp = &qctx->sigrdataset;
}
if (WANTDNSSEC(qctx->client) && qctx->fname->attributes.wildcard) {
dns_fixedname_init(&qctx->wildcardname);
dns_name_copy(qctx->fname,
dns_fixedname_name(&qctx->wildcardname));
qctx->need_wildcardproof = true;
}
if (!qctx->is_zone && RECURSIONOK(qctx->client)) {
query_prefetch(qctx->client, qctx->fname, qctx->rdataset);
}
query_addrrset(qctx, &qctx->fname, &qctx->rdataset, sigrdatasetp,
qctx->dbuf, DNS_SECTION_ANSWER);
/*
* We set the PARTIALANSWER attribute so that if anything goes
* wrong later on, we'll return what we've got so far.
*/
qctx->client->query.attributes |= NS_QUERYATTR_PARTIALANSWER;
/*
* Get the target name of the DNAME.
*/
tname = NULL;
dns_message_gettempname(qctx->client->message, &tname);
result = dns_rdataset_first(trdataset);
if (result != ISC_R_SUCCESS) {
dns_message_puttempname(qctx->client->message, &tname);
(void)ns_query_done(qctx);
goto cleanup;
}
dns_rdataset_current(trdataset, &rdata);
result = dns_rdata_tostruct(&rdata, &dname, NULL);
RUNTIME_CHECK(result == ISC_R_SUCCESS);
dns_rdata_reset(&rdata);
dns_name_copy(&dname.dname, tname);
dns_rdata_freestruct(&dname);
/*
* Construct the new qname consisting of
* <found name prefix>.<dname target>
*/
prefix = dns_fixedname_initname(&fixed);
dns_name_split(qctx->client->query.qname, nlabels, prefix, NULL);
INSIST(qctx->fname == NULL);
qctx->dbuf = ns_client_getnamebuf(qctx->client);
qctx->fname = ns_client_newname(qctx->client, qctx->dbuf, &b);
result = dns_name_concatenate(prefix, tname, qctx->fname);
dns_message_puttempname(qctx->client->message, &tname);
/*
* RFC2672, section 4.1, subsection 3c says
* we should return YXDOMAIN if the constructed
* name would be too long.
*/
if (result == DNS_R_NAMETOOLONG) {
qctx->client->message->rcode = dns_rcode_yxdomain;
}
if (result != ISC_R_SUCCESS) {
(void)ns_query_done(qctx);
goto cleanup;
}
ns_client_keepname(qctx->client, qctx->fname, qctx->dbuf);
/*
* Synthesize a CNAME consisting of
* <old qname> <dname ttl> CNAME <new qname>
* with <dname trust value>
*
* Synthesize a CNAME so old old clients that don't understand
* DNAME can chain.
*
* We do not try to synthesize a signature because we hope
* that security aware servers will understand DNAME. Also,
* even if we had an online key, making a signature
* on-the-fly is costly, and not really legitimate anyway
* since the synthesized CNAME is NOT in the zone.
*/
query_addcname(qctx, trdataset->trust, trdataset->ttl);
/*
* If the original query was not for a CNAME or ANY then follow the
* CNAME.
*/
if (qctx->qtype != dns_rdatatype_cname &&
qctx->qtype != dns_rdatatype_any)
{
/*
* Switch to the new qname and restart.
*/
ns_client_qnamereplace(qctx->client, qctx->fname);
qctx->fname = NULL;
qctx->want_restart = true;
if (!WANTRECURSION(qctx->client)) {
qctx->options.nolog = true;
}
}
query_addauth(qctx);
return ns_query_done(qctx);
cleanup:
return result;
}
/*%
* Add CNAME to response.
*/
static void
query_addcname(query_ctx_t *qctx, dns_trust_t trust, dns_ttl_t ttl) {
ns_client_t *client = qctx->client;
dns_rdataset_t *rdataset = NULL;
dns_rdatalist_t *rdatalist = NULL;
dns_rdata_t *rdata = NULL;
isc_region_t r;
dns_name_t *aname = NULL;
dns_message_gettempname(client->message, &aname);
dns_name_copy(client->query.qname, aname);
dns_message_gettemprdatalist(client->message, &rdatalist);
dns_message_gettemprdata(client->message, &rdata);
dns_message_gettemprdataset(client->message, &rdataset);
rdatalist->type = dns_rdatatype_cname;
rdatalist->rdclass = client->message->rdclass;
rdatalist->ttl = ttl;
dns_name_toregion(qctx->fname, &r);
rdata->data = r.base;
rdata->length = r.length;
rdata->rdclass = client->message->rdclass;
rdata->type = dns_rdatatype_cname;
ISC_LIST_APPEND(rdatalist->rdata, rdata, link);
dns_rdatalist_tordataset(rdatalist, rdataset);
rdataset->trust = trust;
dns_rdataset_setownercase(rdataset, aname);
query_addrrset(qctx, &aname, &rdataset, NULL, NULL, DNS_SECTION_ANSWER);
if (rdataset != NULL) {
if (dns_rdataset_isassociated(rdataset)) {
dns_rdataset_disassociate(rdataset);
}
dns_message_puttemprdataset(client->message, &rdataset);
}
if (aname != NULL) {
dns_message_puttempname(client->message, &aname);
}
}
/*%
* Prepare to respond: determine whether a wildcard proof is needed,
* then hand off to query_respond() or (for type ANY queries)
* query_respond_any().
*/
static isc_result_t
query_prepresponse(query_ctx_t *qctx) {
isc_result_t result = ISC_R_UNSET;
CCTRACE(ISC_LOG_DEBUG(3), "query_prepresponse");
CALL_HOOK(NS_QUERY_PREP_RESPONSE_BEGIN, qctx);
if (WANTDNSSEC(qctx->client) && qctx->fname->attributes.wildcard) {
dns_fixedname_init(&qctx->wildcardname);
dns_name_copy(qctx->fname,
dns_fixedname_name(&qctx->wildcardname));
qctx->need_wildcardproof = true;
}
if (qctx->type == dns_rdatatype_any) {
return query_respond_any(qctx);
}
result = query_zerottl_refetch(qctx);
if (result != ISC_R_COMPLETE) {
goto cleanup;
}
return query_respond(qctx);
cleanup:
return result;
}
/*%
* Add SOA to the authority section when sending negative responses
* (or to the additional section if sending negative responses triggered
* by RPZ rewriting.)
*/
static isc_result_t
query_addsoa(query_ctx_t *qctx, unsigned int override_ttl,
dns_section_t section) {
ns_client_t *client = qctx->client;
dns_name_t *name = NULL;
dns_dbnode_t *node = NULL;
isc_result_t result, eresult = ISC_R_SUCCESS;
dns_rdataset_t *rdataset = NULL, *sigrdataset = NULL;
dns_rdataset_t **sigrdatasetp = NULL;
dns_clientinfomethods_t cm;
dns_clientinfo_t ci;
CTRACE(ISC_LOG_DEBUG(3), "query_addsoa");
dns_clientinfomethods_init(&cm, ns_client_sourceip);
dns_clientinfo_init(&ci, client, NULL);
/*
* Don't add the SOA record for test which set "-T nosoa".
*/
if (((client->manager->sctx->options & NS_SERVER_NOSOA) != 0) &&
(!WANTDNSSEC(client) || !dns_rdataset_isassociated(qctx->rdataset)))
{
return ISC_R_SUCCESS;
}
/*
* Get resources and make 'name' be the database origin.
*/
dns_message_gettempname(client->message, &name);
/*
* We'll be releasing 'name' before returning, so it's safe to
* use clone instead of copying here.
*/
dns_name_clone(dns_db_origin(qctx->db), name);
rdataset = ns_client_newrdataset(client);
if (WANTDNSSEC(client) && dns_db_issecure(qctx->db)) {
sigrdataset = ns_client_newrdataset(client);
}
/*
* Find the SOA.
*/
result = dns_db_getoriginnode(qctx->db, &node);
if (result == ISC_R_SUCCESS) {
result = dns_db_findrdataset(
qctx->db, node, qctx->version, dns_rdatatype_soa, 0,
client->inner.now, rdataset, sigrdataset);
} else {
dns_fixedname_t foundname;
dns_name_t *fname;
fname = dns_fixedname_initname(&foundname);
result = dns_db_findext(qctx->db, name, qctx->version,
dns_rdatatype_soa,
client->query.dboptions, 0, &node,
fname, &cm, &ci, rdataset, sigrdataset);
}
if (result != ISC_R_SUCCESS) {
/*
* This is bad. We tried to get the SOA RR at the zone top
* and it didn't work!
*/
CTRACE(ISC_LOG_ERROR, "unable to find SOA RR at zone apex");
eresult = DNS_R_SERVFAIL;
} else {
/*
* Extract the SOA MINIMUM.
*/
dns_rdata_soa_t soa;
dns_rdata_t rdata = DNS_RDATA_INIT;
result = dns_rdataset_first(rdataset);
RUNTIME_CHECK(result == ISC_R_SUCCESS);
dns_rdataset_current(rdataset, &rdata);
result = dns_rdata_tostruct(&rdata, &soa, NULL);
RUNTIME_CHECK(result == ISC_R_SUCCESS);
if (override_ttl != UINT32_MAX && override_ttl < rdataset->ttl)
{
rdataset->ttl = override_ttl;
if (sigrdataset != NULL) {
sigrdataset->ttl = override_ttl;
}
}
/*
* Add the SOA and its SIG to the response, with the
* TTLs adjusted per RFC2308 section 3.
*/
if (rdataset->ttl > soa.minimum) {
rdataset->ttl = soa.minimum;
}
if (sigrdataset != NULL && sigrdataset->ttl > soa.minimum) {
sigrdataset->ttl = soa.minimum;
}
if (sigrdataset != NULL) {
sigrdatasetp = &sigrdataset;
} else {
sigrdatasetp = NULL;
}
if (section == DNS_SECTION_ADDITIONAL) {
rdataset->attributes.required = true;
}
query_addrrset(qctx, &name, &rdataset, sigrdatasetp, NULL,
section);
}
ns_client_putrdataset(client, &rdataset);
if (sigrdataset != NULL) {
ns_client_putrdataset(client, &sigrdataset);
}
if (name != NULL) {
ns_client_releasename(client, &name);
}
if (node != NULL) {
dns_db_detachnode(qctx->db, &node);
}
return eresult;
}
/*%
* Add NS to authority section (used when the zone apex is already known).
*/
static isc_result_t
query_addns(query_ctx_t *qctx) {
ns_client_t *client = qctx->client;
isc_result_t result, eresult;
dns_name_t *name = NULL, *fname;
dns_dbnode_t *node = NULL;
dns_fixedname_t foundname;
dns_rdataset_t *rdataset = NULL, *sigrdataset = NULL;
dns_rdataset_t **sigrdatasetp = NULL;
dns_clientinfomethods_t cm;
dns_clientinfo_t ci;
CTRACE(ISC_LOG_DEBUG(3), "query_addns");
/*
* Initialization.
*/
eresult = ISC_R_SUCCESS;
fname = dns_fixedname_initname(&foundname);
dns_clientinfomethods_init(&cm, ns_client_sourceip);
dns_clientinfo_init(&ci, client, NULL);
/*
* Get resources and make 'name' be the database origin.
*/
dns_message_gettempname(client->message, &name);
dns_name_clone(dns_db_origin(qctx->db), name);
rdataset = ns_client_newrdataset(client);
if (WANTDNSSEC(client) && dns_db_issecure(qctx->db)) {
sigrdataset = ns_client_newrdataset(client);
}
/*
* Find the NS rdataset.
*/
result = dns_db_getoriginnode(qctx->db, &node);
if (result == ISC_R_SUCCESS) {
result = dns_db_findrdataset(
qctx->db, node, qctx->version, dns_rdatatype_ns, 0,
client->inner.now, rdataset, sigrdataset);
} else {
CTRACE(ISC_LOG_DEBUG(3), "query_addns: calling dns_db_find");
result = dns_db_findext(qctx->db, name, NULL, dns_rdatatype_ns,
client->query.dboptions, 0, &node,
fname, &cm, &ci, rdataset, sigrdataset);
CTRACE(ISC_LOG_DEBUG(3), "query_addns: dns_db_find complete");
}
if (result != ISC_R_SUCCESS) {
CTRACE(ISC_LOG_ERROR, "query_addns: "
"dns_db_findrdataset or dns_db_find "
"failed");
/*
* This is bad. We tried to get the NS rdataset at the zone
* top and it didn't work!
*/
eresult = DNS_R_SERVFAIL;
} else {
if (sigrdataset != NULL) {
sigrdatasetp = &sigrdataset;
}
query_addrrset(qctx, &name, &rdataset, sigrdatasetp, NULL,
DNS_SECTION_AUTHORITY);
}
CTRACE(ISC_LOG_DEBUG(3), "query_addns: cleanup");
ns_client_putrdataset(client, &rdataset);
if (sigrdataset != NULL) {
ns_client_putrdataset(client, &sigrdataset);
}
if (name != NULL) {
ns_client_releasename(client, &name);
}
if (node != NULL) {
dns_db_detachnode(qctx->db, &node);
}
CTRACE(ISC_LOG_DEBUG(3), "query_addns: done");
return eresult;
}
/*%
* Find the zone cut and add the best NS rrset to the authority section.
*/
static void
query_addbestns(query_ctx_t *qctx) {
ns_client_t *client = qctx->client;
dns_db_t *db = NULL, *zdb = NULL;
dns_dbnode_t *node = NULL;
dns_name_t *fname = NULL, *zfname = NULL;
dns_rdataset_t *rdataset = NULL, *sigrdataset = NULL;
dns_rdataset_t *zrdataset = NULL, *zsigrdataset = NULL;
bool is_zone = false, use_zone = false;
isc_buffer_t *dbuf = NULL;
isc_result_t result;
dns_dbversion_t *version = NULL;
dns_zone_t *zone = NULL;
isc_buffer_t b;
dns_clientinfomethods_t cm;
dns_clientinfo_t ci;
dns_name_t qname;
CTRACE(ISC_LOG_DEBUG(3), "query_addbestns");
dns_clientinfomethods_init(&cm, ns_client_sourceip);
dns_clientinfo_init(&ci, client, NULL);
dns_name_init(&qname);
dns_name_clone(client->query.qname, &qname);
/*
* Find the right database.
*/
do {
result = query_getdb(client, &qname, dns_rdatatype_ns,
(dns_getdb_options_t){ 0 }, &zone, &db,
&version, &is_zone);
if (result != ISC_R_SUCCESS) {
goto cleanup;
}
/*
* If this is a static stub zone look for a parent zone.
*/
if (zone != NULL &&
dns_zone_gettype(zone) == dns_zone_staticstub)
{
unsigned int labels = dns_name_countlabels(&qname);
dns_db_detach(&db);
dns_zone_detach(&zone);
version = NULL;
if (labels != 1) {
dns_name_split(&qname, labels - 1, NULL,
&qname);
continue;
}
if (!USECACHE(client)) {
goto cleanup;
}
dns_db_attach(client->inner.view->cachedb, &db);
is_zone = false;
}
break;
} while (true);
db_find:
/*
* We'll need some resources...
*/
dbuf = ns_client_getnamebuf(client);
fname = ns_client_newname(client, dbuf, &b);
rdataset = ns_client_newrdataset(client);
/*
* Get the RRSIGs if the client requested them or if we may
* need to validate answers from the cache.
*/
if (WANTDNSSEC(client) || !is_zone) {
sigrdataset = ns_client_newrdataset(client);
}
/*
* Now look for the zonecut.
*/
if (is_zone) {
result = dns_db_findext(
db, client->query.qname, version, dns_rdatatype_ns,
client->query.dboptions, client->inner.now, &node,
fname, &cm, &ci, rdataset, sigrdataset);
if (result != DNS_R_DELEGATION) {
goto cleanup;
}
if (USECACHE(client)) {
ns_client_keepname(client, fname, dbuf);
dns_db_detachnode(db, &node);
SAVE(zdb, db);
SAVE(zfname, fname);
SAVE(zrdataset, rdataset);
SAVE(zsigrdataset, sigrdataset);
version = NULL;
dns_db_attach(client->inner.view->cachedb, &db);
is_zone = false;
goto db_find;
}
} else {
result = dns_db_findzonecut(db, client->query.qname,
client->query.dboptions,
client->inner.now, &node, fname,
NULL, rdataset, sigrdataset);
if (result == ISC_R_SUCCESS) {
if (zfname != NULL &&
!dns_name_issubdomain(fname, zfname))
{
/*
* We found a zonecut in the cache, but our
* zone delegation is better.
*/
use_zone = true;
}
} else if (result == ISC_R_NOTFOUND && zfname != NULL) {
/*
* We didn't find anything in the cache, but we
* have a zone delegation, so use it.
*/
use_zone = true;
} else {
goto cleanup;
}
}
if (use_zone) {
ns_client_releasename(client, &fname);
/*
* We've already done ns_client_keepname() on
* zfname, so we must set dbuf to NULL to
* prevent query_addrrset() from trying to
* call ns_client_keepname() again.
*/
dbuf = NULL;
ns_client_putrdataset(client, &rdataset);
if (sigrdataset != NULL) {
ns_client_putrdataset(client, &sigrdataset);
}
if (node != NULL) {
dns_db_detachnode(db, &node);
}
dns_db_detach(&db);
RESTORE(db, zdb);
RESTORE(fname, zfname);
RESTORE(rdataset, zrdataset);
RESTORE(sigrdataset, zsigrdataset);
}
/*
* Attempt to validate RRsets that are pending or that are glue.
*/
if ((DNS_TRUST_PENDING(rdataset->trust) ||
(sigrdataset != NULL && DNS_TRUST_PENDING(sigrdataset->trust))) &&
!validate(client, db, fname, rdataset, sigrdataset) &&
!PENDINGOK(client->query.dboptions))
{
goto cleanup;
}
if ((DNS_TRUST_GLUE(rdataset->trust) ||
(sigrdataset != NULL && DNS_TRUST_GLUE(sigrdataset->trust))) &&
!validate(client, db, fname, rdataset, sigrdataset) &&
SECURE(client) && WANTDNSSEC(client))
{
goto cleanup;
}
/*
* If the answer is secure only add NS records if they are secure
* when the client may be looking for AD in the response.
*/
if (SECURE(client) && (WANTDNSSEC(client) || WANTAD(client)) &&
((rdataset->trust != dns_trust_secure) ||
(sigrdataset != NULL && sigrdataset->trust != dns_trust_secure)))
{
goto cleanup;
}
/*
* If the client doesn't want DNSSEC we can discard the sigrdataset
* now.
*/
if (!WANTDNSSEC(client)) {
ns_client_putrdataset(client, &sigrdataset);
}
query_addrrset(qctx, &fname, &rdataset, &sigrdataset, dbuf,
DNS_SECTION_AUTHORITY);
cleanup:
if (rdataset != NULL) {
ns_client_putrdataset(client, &rdataset);
}
if (sigrdataset != NULL) {
ns_client_putrdataset(client, &sigrdataset);
}
if (fname != NULL) {
ns_client_releasename(client, &fname);
}
if (node != NULL) {
dns_db_detachnode(db, &node);
}
if (db != NULL) {
dns_db_detach(&db);
}
if (zone != NULL) {
dns_zone_detach(&zone);
}
if (zdb != NULL) {
ns_client_putrdataset(client, &zrdataset);
if (zsigrdataset != NULL) {
ns_client_putrdataset(client, &zsigrdataset);
}
if (zfname != NULL) {
ns_client_releasename(client, &zfname);
}
dns_db_detach(&zdb);
}
}
static void
query_addwildcardproof(query_ctx_t *qctx, bool ispositive, bool nodata) {
ns_client_t *client = qctx->client;
isc_buffer_t *dbuf, b;
dns_name_t *name;
dns_name_t *fname = NULL;
dns_rdataset_t *rdataset = NULL, *sigrdataset = NULL;
dns_fixedname_t wfixed;
dns_name_t *wname;
dns_dbnode_t *node = NULL;
unsigned int options;
unsigned int olabels, nlabels, labels;
isc_result_t result;
dns_rdata_t rdata = DNS_RDATA_INIT;
dns_rdata_nsec_t nsec;
bool have_wname;
int order;
dns_fixedname_t cfixed;
dns_name_t *cname;
dns_clientinfomethods_t cm;
dns_clientinfo_t ci;
CTRACE(ISC_LOG_DEBUG(3), "query_addwildcardproof");
dns_clientinfomethods_init(&cm, ns_client_sourceip);
dns_clientinfo_init(&ci, client, NULL);
/*
* If a name has been specifically flagged as needing
* a wildcard proof then it will have been copied to
* qctx->wildcardname. Otherwise we just use the client
* QNAME.
*/
if (qctx->need_wildcardproof) {
name = dns_fixedname_name(&qctx->wildcardname);
} else {
name = client->query.qname;
}
/*
* Get the NOQNAME proof then if !ispositive
* get the NOWILDCARD proof.
*
* DNS_DBFIND_NOWILD finds the NSEC records that covers the
* name ignoring any wildcard. From the owner and next names
* of this record you can compute which wildcard (if it exists)
* will match by finding the longest common suffix of the
* owner name and next names with the qname and prefixing that
* with the wildcard label.
*
* e.g.
* Given:
* example SOA
* example NSEC b.example
* b.example A
* b.example NSEC a.d.example
* a.d.example A
* a.d.example NSEC g.f.example
* g.f.example A
* g.f.example NSEC z.i.example
* z.i.example A
* z.i.example NSEC example
*
* QNAME:
* a.example -> example NSEC b.example
* owner common example
* next common example
* wild *.example
* d.b.example -> b.example NSEC a.d.example
* owner common b.example
* next common example
* wild *.b.example
* a.f.example -> a.d.example NSEC g.f.example
* owner common example
* next common f.example
* wild *.f.example
* j.example -> z.i.example NSEC example
* owner common example
* next common example
* wild *.example
*/
options = client->query.dboptions | DNS_DBFIND_NOWILD;
wname = dns_fixedname_initname(&wfixed);
again:
have_wname = false;
/*
* We'll need some resources...
*/
dbuf = ns_client_getnamebuf(client);
fname = ns_client_newname(client, dbuf, &b);
rdataset = ns_client_newrdataset(client);
sigrdataset = ns_client_newrdataset(client);
result = dns_db_findext(qctx->db, name, qctx->version,
dns_rdatatype_nsec, options, 0, &node, fname,
&cm, &ci, rdataset, sigrdataset);
if (node != NULL) {
dns_db_detachnode(qctx->db, &node);
}
if (!dns_rdataset_isassociated(rdataset)) {
/*
* No NSEC proof available, return NSEC3 proofs instead.
*/
cname = dns_fixedname_initname(&cfixed);
/*
* Find the closest encloser using a binary search.
* maxlabels: suffix length of NXDOMAIN result
* minlabels: suffix length of non NXDOMAIN result
*/
unsigned int maxlabels = dns_name_countlabels(name);
unsigned int minlabels = dns_name_countlabels(fname);
unsigned int namelabels = maxlabels;
bool search = result == DNS_R_NXDOMAIN;
dns_name_copy(name, cname);
while (search) {
labels = (maxlabels + minlabels) / 2;
dns_name_split(name, labels, NULL, cname);
if (labels == minlabels) {
break;
}
result = dns_db_findext(qctx->db, cname, qctx->version,
dns_rdatatype_nsec, options, 0,
NULL, fname, &cm, &ci, NULL,
NULL);
if (result == DNS_R_NXDOMAIN) {
maxlabels = labels;
} else {
minlabels = labels;
}
}
/*
* Add closest (provable) encloser NSEC3.
*/
query_findclosestnsec3(cname, qctx->db, qctx->version, client,
rdataset, sigrdataset, fname, true,
cname);
if (!dns_rdataset_isassociated(rdataset)) {
goto cleanup;
}
if (!ispositive) {
query_addrrset(qctx, &fname, &rdataset, &sigrdataset,
dbuf, DNS_SECTION_AUTHORITY);
}
/*
* Replace resources which were consumed by query_addrrset.
*/
if (fname == NULL) {
dbuf = ns_client_getnamebuf(client);
fname = ns_client_newname(client, dbuf, &b);
}
if (rdataset == NULL) {
rdataset = ns_client_newrdataset(client);
} else if (dns_rdataset_isassociated(rdataset)) {
dns_rdataset_disassociate(rdataset);
}
if (sigrdataset == NULL) {
sigrdataset = ns_client_newrdataset(client);
} else if (dns_rdataset_isassociated(sigrdataset)) {
dns_rdataset_disassociate(sigrdataset);
}
/*
* Add no qname proof.
*/
labels = dns_name_countlabels(cname) + 1;
if (labels > namelabels) {
char namebuf[DNS_NAME_FORMATSIZE];
dns_name_format(cname, namebuf, sizeof(namebuf));
ns_client_log(qctx->client, DNS_LOGCATEGORY_DNSSEC,
NS_LOGMODULE_QUERY, ISC_LOG_WARNING,
"closest-encloser name too long: %s",
namebuf);
dns_name_copy(name, wname);
} else if (labels == namelabels) {
dns_name_copy(name, wname);
} else {
dns_name_split(name, labels, NULL, wname);
}
query_findclosestnsec3(wname, qctx->db, qctx->version, client,
rdataset, sigrdataset, fname, false,
NULL);
if (!dns_rdataset_isassociated(rdataset)) {
goto cleanup;
}
query_addrrset(qctx, &fname, &rdataset, &sigrdataset, dbuf,
DNS_SECTION_AUTHORITY);
if (ispositive) {
goto cleanup;
}
/*
* Replace resources which were consumed by query_addrrset.
*/
if (fname == NULL) {
dbuf = ns_client_getnamebuf(client);
fname = ns_client_newname(client, dbuf, &b);
}
if (rdataset == NULL) {
rdataset = ns_client_newrdataset(client);
} else if (dns_rdataset_isassociated(rdataset)) {
dns_rdataset_disassociate(rdataset);
}
if (sigrdataset == NULL) {
sigrdataset = ns_client_newrdataset(client);
} else if (dns_rdataset_isassociated(sigrdataset)) {
dns_rdataset_disassociate(sigrdataset);
}
/*
* Add the no wildcard proof.
*/
result = dns_name_concatenate(dns_wildcardname, cname, wname);
if (result != ISC_R_SUCCESS) {
goto cleanup;
}
query_findclosestnsec3(wname, qctx->db, qctx->version, client,
rdataset, sigrdataset, fname, nodata,
NULL);
if (!dns_rdataset_isassociated(rdataset)) {
goto cleanup;
}
query_addrrset(qctx, &fname, &rdataset, &sigrdataset, dbuf,
DNS_SECTION_AUTHORITY);
goto cleanup;
} else if (result == DNS_R_NXDOMAIN) {
if (!ispositive) {
result = dns_rdataset_first(rdataset);
}
if (result == ISC_R_SUCCESS) {
dns_rdataset_current(rdataset, &rdata);
result = dns_rdata_tostruct(&rdata, &nsec, NULL);
RUNTIME_CHECK(result == ISC_R_SUCCESS);
(void)dns_name_fullcompare(name, fname, &order,
&olabels);
(void)dns_name_fullcompare(name, &nsec.next, &order,
&nlabels);
/*
* Check for a pathological condition created when
* serving some malformed signed zones and bail out.
*/
if (dns_name_countlabels(name) == nlabels) {
goto cleanup;
}
if (olabels > nlabels) {
dns_name_split(name, olabels, NULL, wname);
} else {
dns_name_split(name, nlabels, NULL, wname);
}
result = dns_name_concatenate(dns_wildcardname, wname,
wname);
if (result == ISC_R_SUCCESS) {
have_wname = true;
}
dns_rdata_freestruct(&nsec);
}
query_addrrset(qctx, &fname, &rdataset, &sigrdataset, dbuf,
DNS_SECTION_AUTHORITY);
}
if (rdataset != NULL) {
ns_client_putrdataset(client, &rdataset);
}
if (sigrdataset != NULL) {
ns_client_putrdataset(client, &sigrdataset);
}
if (fname != NULL) {
ns_client_releasename(client, &fname);
}
if (have_wname) {
ispositive = true; /* prevent loop */
if (!dns_name_equal(name, wname)) {
name = wname;
goto again;
}
}
cleanup:
if (rdataset != NULL) {
ns_client_putrdataset(client, &rdataset);
}
if (sigrdataset != NULL) {
ns_client_putrdataset(client, &sigrdataset);
}
if (fname != NULL) {
ns_client_releasename(client, &fname);
}
}
/*%
* Add NS records, and NSEC/NSEC3 wildcard proof records if needed,
* to the authority section.
*/
static void
query_addauth(query_ctx_t *qctx) {
CCTRACE(ISC_LOG_DEBUG(3), "query_addauth");
/*
* Add NS records to the authority section (if we haven't already
* added them to the answer section).
*/
if (!qctx->want_restart && !NOAUTHORITY(qctx->client)) {
if (qctx->is_zone) {
if (!qctx->answer_has_ns) {
(void)query_addns(qctx);
}
} else if (!qctx->answer_has_ns &&
qctx->qtype != dns_rdatatype_ns)
{
if (qctx->fname != NULL) {
ns_client_releasename(qctx->client,
&qctx->fname);
}
query_addbestns(qctx);
}
}
/*
* Add NSEC records to the authority section if they're needed for
* DNSSEC wildcard proofs.
*/
if (qctx->need_wildcardproof && dns_db_issecure(qctx->db)) {
query_addwildcardproof(qctx, true, false);
}
}
/*
* When sending a referral, if the answer to the question is
* in the glue, sort it to the start of the additional section.
*/
static void
query_glueanswer(query_ctx_t *qctx) {
const dns_namelist_t *secs = qctx->client->message->sections;
const dns_section_t section = DNS_SECTION_ADDITIONAL;
dns_message_t *msg;
if (!ISC_LIST_EMPTY(secs[DNS_SECTION_ANSWER]) ||
qctx->client->message->rcode != dns_rcode_noerror ||
!dns_rdatatype_isaddr(qctx->qtype))
{
return;
}
msg = qctx->client->message;
MSG_SECTION_FOREACH (msg, section, name) {
if (dns_name_equal(name, qctx->client->query.qname)) {
ISC_LIST_FOREACH (name->list, rdataset, link) {
if (rdataset->type == qctx->qtype) {
ISC_LIST_UNLINK(msg->sections[section],
name, link);
ISC_LIST_PREPEND(msg->sections[section],
name, link);
ISC_LIST_UNLINK(name->list, rdataset,
link);
ISC_LIST_PREPEND(name->list, rdataset,
link);
rdataset->attributes.required = true;
break;
}
}
break;
}
}
}
isc_result_t
ns_query_done(query_ctx_t *qctx) {
isc_result_t result = ISC_R_UNSET;
const dns_namelist_t *secs = qctx->client->message->sections;
bool partial_result_with_servfail = false;
CCTRACE(ISC_LOG_DEBUG(3), "ns_query_done");
CALL_HOOK(NS_QUERY_DONE_BEGIN, qctx);
/*
* General cleanup.
*/
qctx->rpz_st = qctx->client->query.rpz_st;
if (qctx->rpz_st != NULL &&
(qctx->rpz_st->state & DNS_RPZ_RECURSING) == 0)
{
rpz_match_clear(qctx->rpz_st);
qctx->rpz_st->state &= ~DNS_RPZ_DONE_QNAME;
}
qctx_clean(qctx);
qctx_freedata(qctx);
/*
* Clear the AA bit if we're not authoritative.
*/
if (qctx->client->query.restarts == 0 && !qctx->authoritative) {
qctx->client->message->flags &= ~DNS_MESSAGEFLAG_AA;
qctx->client->inner.attributes &= ~NS_CLIENTATTR_WANTRC;
}
/*
* Do we need to restart the query (e.g. for CNAME chaining)?
*/
if (qctx->want_restart) {
if (qctx->client->query.restarts <
qctx->client->inner.view->max_restarts)
{
query_ctx_t *saved_qctx = NULL;
qctx->client->query.restarts++;
saved_qctx = isc_mem_get(qctx->client->manager->mctx,
sizeof(*saved_qctx));
qctx_save(qctx, saved_qctx);
isc_nmhandle_attach(qctx->client->inner.handle,
&qctx->client->inner.restarthandle);
isc_async_run(qctx->client->manager->loop,
async_restart, saved_qctx);
return DNS_R_CONTINUE;
} else {
/*
* This is e.g. a long CNAME chain which we cut short.
*/
qctx->client->query.attributes |=
NS_QUERYATTR_PARTIALANSWER;
qctx->client->message->rcode = dns_rcode_servfail;
qctx->result = DNS_R_SERVFAIL;
/*
* Send the answer back with a SERVFAIL result even
* if recursion was requested.
*/
partial_result_with_servfail = true;
dns_ede_add(&qctx->client->edectx, DNS_EDE_OTHER,
"max. restarts reached");
ns_client_log(qctx->client, NS_LOGCATEGORY_CLIENT,
NS_LOGMODULE_QUERY, ISC_LOG_INFO,
"query iterations limit reached");
}
}
if (qctx->result != ISC_R_SUCCESS &&
(!PARTIALANSWER(qctx->client) ||
(WANTRECURSION(qctx->client) && !partial_result_with_servfail) ||
qctx->result == DNS_R_DROP))
{
if (qctx->result == DNS_R_DUPLICATE ||
qctx->result == DNS_R_DROP)
{
/*
* This was a duplicate query that we are
* recursing on or the result of rate limiting.
* Don't send a response now for a duplicate query,
* because the original will still cause a response.
*/
query_next(qctx->client, qctx->result);
} else {
/*
* If we don't have any answer to give the client,
* or if the client requested recursion and thus wanted
* the complete answer, send an error response.
*/
INSIST(qctx->line >= 0);
query_error(qctx->client, qctx->result, qctx->line);
}
qctx->detach_client = true;
return qctx->result;
}
/*
* If we're recursing then just return; the query will
* resume when recursion ends.
*/
if (RECURSING(qctx->client) &&
(!QUERY_STALETIMEOUT(&qctx->client->query) ||
qctx->options.stalefirst))
{
return qctx->result;
}
/*
* We are done. Sort the answer to the front of the
* additional section if necessary, make a final tweak
* to the AA bit if the auth-nxdomain config option
* says so, then render and send the response.
*/
query_glueanswer(qctx);
if (qctx->client->message->rcode == dns_rcode_nxdomain &&
qctx->view->auth_nxdomain)
{
qctx->client->message->flags |= DNS_MESSAGEFLAG_AA;
}
/*
* If the response is somehow unexpected for the client and this
* is a result of recursion, return an error to the caller
* to indicate it may need to be logged.
*/
if (qctx->resuming &&
(ISC_LIST_EMPTY(secs[DNS_SECTION_ANSWER]) ||
qctx->client->message->rcode != dns_rcode_noerror))
{
qctx->result = ISC_R_FAILURE;
}
CALL_HOOK(NS_QUERY_DONE_SEND, qctx);
query_send(qctx->client);
if (qctx->refresh_rrset) {
/*
* If we reached this point then it means that we have found a
* stale RRset entry in cache and BIND is configured to allow
* queries to be answered with stale data if no active RRset
* is available, i.e. "stale-anwer-client-timeout 0". But, we
* still need to refresh the RRset. To prevent adding duplicate
* RRsets, clear the RRsets from the message before doing the
* refresh.
*/
message_clearrdataset(qctx->client->message, false);
query_stale_refresh(qctx->client);
}
qctx->detach_client = true;
return qctx->result;
cleanup:
/*
* We'd only get here if one of the hooks above
* (NS_QUERY_DONE_BEGIN or NS_QUERY_DONE_SEND) returned
* NS_HOOK_RETURN. Some housekeeping may be needed.
*/
qctx_clean(qctx);
qctx_freedata(qctx);
if (!qctx->async) {
qctx->detach_client = true;
query_error(qctx->client, DNS_R_SERVFAIL, __LINE__);
}
return result;
}
static void
log_tat(ns_client_t *client) {
char namebuf[DNS_NAME_FORMATSIZE];
char clientbuf[ISC_NETADDR_FORMATSIZE];
char classbuf[DNS_RDATACLASS_FORMATSIZE];
isc_netaddr_t netaddr;
char *tags = NULL;
size_t taglen = 0;
if (!isc_log_wouldlog(ISC_LOG_INFO)) {
return;
}
if ((client->query.qtype != dns_rdatatype_null ||
!dns_name_istat(client->query.qname)) &&
(client->inner.keytag == NULL ||
client->query.qtype != dns_rdatatype_dnskey))
{
return;
}
isc_netaddr_fromsockaddr(&netaddr, &client->inner.peeraddr);
dns_name_format(client->query.qname, namebuf, sizeof(namebuf));
isc_netaddr_format(&netaddr, clientbuf, sizeof(clientbuf));
dns_rdataclass_format(client->inner.view->rdclass, classbuf,
sizeof(classbuf));
if (client->query.qtype == dns_rdatatype_dnskey) {
uint16_t keytags = client->inner.keytag_len / 2;
size_t len = taglen = sizeof("65000") * keytags + 1;
char *cp = tags = isc_mem_get(client->manager->mctx, taglen);
int i = 0;
INSIST(client->inner.keytag != NULL);
if (tags != NULL) {
while (keytags-- > 0U) {
int n;
uint16_t keytag;
keytag = (client->inner.keytag[i * 2] << 8) |
client->inner.keytag[i * 2 + 1];
n = snprintf(cp, len, " %u", keytag);
if (n > 0 && (size_t)n <= len) {
cp += n;
len -= n;
i++;
} else {
break;
}
}
}
}
isc_log_write(NS_LOGCATEGORY_TAT, NS_LOGMODULE_QUERY, ISC_LOG_INFO,
"trust-anchor-telemetry '%s/%s' from %s%s", namebuf,
classbuf, clientbuf, tags != NULL ? tags : "");
if (tags != NULL) {
isc_mem_put(client->manager->mctx, tags, taglen);
}
}
static void
log_query(ns_client_t *client, unsigned int flags, unsigned int extflags) {
char namebuf[DNS_NAME_FORMATSIZE];
char typebuf[DNS_RDATATYPE_FORMATSIZE];
char classbuf[DNS_RDATACLASS_FORMATSIZE];
char sabuf[ISC_SOCKADDR_FORMATSIZE];
char ecsbuf[NS_CLIENT_ECS_FORMATSIZE] = { 0 };
char flagsbuf[NS_CLIENT_FLAGS_FORMATSIZE] = { 0 };
dns_rdataset_t *rdataset;
int level = ISC_LOG_INFO;
if (!isc_log_wouldlog(level)) {
return;
}
rdataset = ISC_LIST_HEAD(client->query.qname->list);
INSIST(rdataset != NULL);
dns_name_format(client->query.qname, namebuf, sizeof(namebuf));
dns_rdataclass_format(rdataset->rdclass, classbuf, sizeof(classbuf));
dns_rdatatype_format(rdataset->type, typebuf, sizeof(typebuf));
isc_sockaddr_format(&client->inner.destsockaddr, sabuf, sizeof(sabuf));
if (HAVEECS(client)) {
ns_client_log_ecs(client, ecsbuf, sizeof(ecsbuf));
}
ns_client_log_flags(client, flags, extflags, flagsbuf,
sizeof(flagsbuf));
ns_client_log(client, NS_LOGCATEGORY_QUERIES, NS_LOGMODULE_QUERY, level,
"query: %s %s %s %s (%s)%s", namebuf, classbuf, typebuf,
flagsbuf, sabuf, ecsbuf);
}
static void
log_queryerror(ns_client_t *client, isc_result_t result, int line, int level) {
char namebuf[DNS_NAME_FORMATSIZE];
char typebuf[DNS_RDATATYPE_FORMATSIZE];
char classbuf[DNS_RDATACLASS_FORMATSIZE];
const char *namep, *typep, *classp, *sep1, *sep2;
dns_rdataset_t *rdataset;
if (!isc_log_wouldlog(level)) {
return;
}
namep = typep = classp = sep1 = sep2 = "";
/*
* Query errors can happen for various reasons. In some cases we cannot
* even assume the query contains a valid question section, so we should
* expect exceptional cases.
*/
if (client->query.origqname != NULL) {
dns_name_format(client->query.origqname, namebuf,
sizeof(namebuf));
namep = namebuf;
sep1 = " for ";
rdataset = ISC_LIST_HEAD(client->query.origqname->list);
if (rdataset != NULL) {
dns_rdataclass_format(rdataset->rdclass, classbuf,
sizeof(classbuf));
classp = classbuf;
dns_rdatatype_format(rdataset->type, typebuf,
sizeof(typebuf));
typep = typebuf;
sep2 = "/";
}
}
ns_client_log(client, NS_LOGCATEGORY_QUERY_ERRORS, NS_LOGMODULE_QUERY,
level, "query failed (%s)%s%s%s%s%s%s at %s:%d",
isc_result_totext(result), sep1, namep, sep2, classp,
sep2, typep, __FILE__, line);
}
void
ns_query_start(ns_client_t *client, isc_nmhandle_t *handle) {
isc_result_t result;
dns_message_t *message;
dns_rdataset_t *rdataset;
dns_rdatatype_t qtype;
unsigned int saved_extflags;
unsigned int saved_flags;
REQUIRE(NS_CLIENT_VALID(client));
/*
* Attach to the request handle
*/
isc_nmhandle_attach(handle, &client->inner.reqhandle);
message = client->message;
saved_extflags = client->inner.extflags;
saved_flags = client->message->flags;
CTRACE(ISC_LOG_DEBUG(3), "ns_query_start");
/*
* Ensure that appropriate cleanups occur.
*/
client->inner.cleanup = query_cleanup;
if ((message->flags & DNS_MESSAGEFLAG_RD) != 0) {
client->query.attributes |= NS_QUERYATTR_WANTRECURSION;
}
if ((client->inner.extflags & DNS_MESSAGEEXTFLAG_DO) != 0) {
client->inner.attributes |= NS_CLIENTATTR_WANTDNSSEC;
}
switch (client->inner.view->minimalresponses) {
case dns_minimal_no:
break;
case dns_minimal_yes:
client->query.attributes |= (NS_QUERYATTR_NOAUTHORITY |
NS_QUERYATTR_NOADDITIONAL);
break;
case dns_minimal_noauth:
client->query.attributes |= NS_QUERYATTR_NOAUTHORITY;
break;
case dns_minimal_noauthrec:
if ((message->flags & DNS_MESSAGEFLAG_RD) != 0) {
client->query.attributes |= NS_QUERYATTR_NOAUTHORITY;
}
break;
}
if (client->inner.view->cachedb == NULL ||
!client->inner.view->recursion)
{
/*
* We don't have a cache. Turn off cache support and
* recursion.
*/
client->query.attributes &= ~(NS_QUERYATTR_RECURSIONOK |
NS_QUERYATTR_CACHEOK);
client->inner.attributes |= NS_CLIENTATTR_NOSETFC;
} else if ((client->inner.attributes & NS_CLIENTATTR_RA) == 0 ||
(message->flags & DNS_MESSAGEFLAG_RD) == 0)
{
/*
* If the client isn't allowed to recurse (due to
* "recursion no", the allow-recursion ACL, or the
* lack of a resolver in this view), or if it
* doesn't want recursion, turn recursion off.
*/
client->query.attributes &= ~NS_QUERYATTR_RECURSIONOK;
client->inner.attributes |= NS_CLIENTATTR_NOSETFC;
}
/*
* Check for multiple question queries, since edns1 is dead.
*/
if (message->counts[DNS_SECTION_QUESTION] > 1) {
query_error(client, DNS_R_FORMERR, __LINE__);
return;
}
/*
* Get the question name.
*/
if (ISC_LIST_EMPTY(message->sections[DNS_SECTION_QUESTION])) {
query_error(client, ISC_R_NOMORE, __LINE__);
return;
}
client->query.qname =
ISC_LIST_HEAD(message->sections[DNS_SECTION_QUESTION]);
client->query.origqname = client->query.qname;
if (ISC_LIST_NEXT(client->query.qname, link) != NULL) {
/*
* There's more than one QNAME in the question
* section.
*/
query_error(client, DNS_R_FORMERR, __LINE__);
return;
}
if ((client->manager->sctx->options & NS_SERVER_LOGQUERIES) != 0) {
log_query(client, saved_flags, saved_extflags);
}
/*
* Check for meta-queries like IXFR and AXFR.
*/
rdataset = ISC_LIST_HEAD(client->query.qname->list);
INSIST(rdataset != NULL);
client->query.qtype = qtype = rdataset->type;
dns_rdatatypestats_increment(client->manager->sctx->rcvquerystats,
qtype);
log_tat(client);
if (dns_rdatatype_ismeta(qtype)) {
switch (qtype) {
case dns_rdatatype_any:
break; /* Let the query logic handle it. */
case dns_rdatatype_ixfr:
case dns_rdatatype_axfr:
if (isc_nm_is_http_handle(handle)) {
/*
* We cannot use DoH for zone transfers.
* According to RFC 8484 a DoH request contains
* exactly one DNS message (see Section 6:
* Definition of the "application/dns-message"
* Media Type).
*
* This makes DoH unsuitable for zone transfers
* as often (and usually!) these need more than
* one DNS message, especially for larger zones.
* As zone transfers over DoH are not (yet)
* standardised, nor discussed in RFC 8484,
* the best thing we can do is to return "not
* implemented".
*/
query_error(client, DNS_R_NOTIMP, __LINE__);
return;
}
if (isc_nm_socket_type(handle) ==
isc_nm_streamdnssocket)
{
/*
* Currently this code is here for DoT, which
* has more complex requirements for zone
* transfers compared to other stream
* protocols. See RFC 9103 for details.
*/
switch (isc_nm_xfr_checkperm(handle)) {
case ISC_R_SUCCESS:
break;
case ISC_R_DOTALPNERROR:
query_error(client, DNS_R_NOALPN,
__LINE__);
return;
default:
query_error(client, DNS_R_REFUSED,
__LINE__);
return;
}
}
ns_xfr_start(client, rdataset->type);
return;
case dns_rdatatype_maila:
case dns_rdatatype_mailb:
query_error(client, DNS_R_NOTIMP, __LINE__);
return;
case dns_rdatatype_tkey:
result = dns_tkey_processquery(
client->message, client->manager->sctx->tkeyctx,
client->inner.view->dynamickeys);
if (result == ISC_R_SUCCESS) {
query_send(client);
} else {
query_error(client, result, __LINE__);
}
return;
default: /* TSIG, etc. */
query_error(client, DNS_R_FORMERR, __LINE__);
return;
}
}
/*
* Turn on minimal response for (C)DNSKEY and (C)DS queries.
*/
if (dns_rdatatype_iskeymaterial(qtype) || qtype == dns_rdatatype_ds) {
client->query.attributes |= (NS_QUERYATTR_NOAUTHORITY |
NS_QUERYATTR_NOADDITIONAL);
} else if (qtype == dns_rdatatype_ns) {
/*
* Always turn on additional records for NS queries.
*/
client->query.attributes &= ~(NS_QUERYATTR_NOAUTHORITY |
NS_QUERYATTR_NOADDITIONAL);
}
/*
* Maybe turn on minimal responses for ANY queries.
*/
if (qtype == dns_rdatatype_any && client->inner.view->minimal_any &&
!TCP(client))
{
client->query.attributes |= (NS_QUERYATTR_NOAUTHORITY |
NS_QUERYATTR_NOADDITIONAL);
}
/*
* Turn on minimal responses for EDNS/UDP bufsize 512 queries.
*/
if (client->inner.ednsversion >= 0 && client->inner.udpsize <= 512U &&
!TCP(client))
{
client->query.attributes |= (NS_QUERYATTR_NOAUTHORITY |
NS_QUERYATTR_NOADDITIONAL);
}
/*
* If the client has requested that DNSSEC checking be disabled,
* allow lookups to return pending data and instruct the resolver
* to return data before validation has completed.
*
* We don't need to set DNS_DBFIND_PENDINGOK when validation is
* disabled as there will be no pending data.
*/
if ((message->flags & DNS_MESSAGEFLAG_CD) != 0 ||
qtype == dns_rdatatype_rrsig)
{
client->query.dboptions |= DNS_DBFIND_PENDINGOK;
client->query.fetchoptions |= DNS_FETCHOPT_NOVALIDATE;
} else if (!client->inner.view->enablevalidation) {
client->query.fetchoptions |= DNS_FETCHOPT_NOVALIDATE;
}
if (client->inner.view->qminimization) {
client->query.fetchoptions |= DNS_FETCHOPT_QMINIMIZE |
DNS_FETCHOPT_QMIN_SKIP_IP6A;
if (client->inner.view->qmin_strict) {
client->query.fetchoptions |= DNS_FETCHOPT_QMIN_STRICT;
}
}
/*
* Allow glue NS records to be added to the authority section
* if the answer is secure.
*/
if ((message->flags & DNS_MESSAGEFLAG_CD) != 0) {
client->query.attributes &= ~NS_QUERYATTR_SECURE;
}
/*
* Set NS_CLIENTATTR_WANTAD if the client has set AD in the query.
* This allows AD to be returned on queries without DO set.
*/
if ((message->flags & DNS_MESSAGEFLAG_AD) != 0) {
client->inner.attributes |= NS_CLIENTATTR_WANTAD;
}
/*
* This is an ordinary query.
*/
result = dns_message_reply(message, true);
if (result != ISC_R_SUCCESS) {
query_next(client, result);
return;
}
/*
* Assume authoritative response until it is known to be
* otherwise.
*
* If "-T noaa" has been set on the command line don't set
* AA on authoritative answers.
*/
if ((client->manager->sctx->options & NS_SERVER_NOAA) == 0) {
message->flags |= DNS_MESSAGEFLAG_AA;
}
/*
* Set AD. We must clear it if we add non-validated data to a
* response.
*/
if (WANTDNSSEC(client) || WANTAD(client)) {
message->flags |= DNS_MESSAGEFLAG_AD;
}
/*
* Start global outgoing query count.
*/
isc_counter_create(client->manager->mctx,
client->inner.view->max_queries, &client->query.qc);
query_setup(client, qtype);
}