2 * Copyright (C) 2004-2015 Internet Systems Consortium, Inc. ("ISC")
3 * Copyright (C) 1999-2003 Internet Software Consortium.
5 * Permission to use, copy, modify, and/or distribute this software for any
6 * purpose with or without fee is hereby granted, provided that the above
7 * copyright notice and this permission notice appear in all copies.
9 * THE SOFTWARE IS PROVIDED "AS IS" AND ISC DISCLAIMS ALL WARRANTIES WITH
10 * REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
11 * AND FITNESS. IN NO EVENT SHALL ISC BE LIABLE FOR ANY SPECIAL, DIRECT,
12 * INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
13 * LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
14 * OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
15 * PERFORMANCE OF THIS SOFTWARE.
22 #include <isc/formatcheck.h>
23 #include <isc/mutex.h>
25 #include <isc/platform.h>
26 #include <isc/print.h>
27 #include <isc/queue.h>
28 #include <isc/stats.h>
29 #include <isc/stdio.h>
30 #include <isc/string.h>
32 #include <isc/timer.h>
36 #include <dns/dispatch.h>
37 #include <dns/events.h>
38 #include <dns/message.h>
40 #include <dns/rcode.h>
41 #include <dns/rdata.h>
42 #include <dns/rdataclass.h>
43 #include <dns/rdatalist.h>
44 #include <dns/rdataset.h>
45 #include <dns/resolver.h>
46 #include <dns/stats.h>
51 #include <named/interfacemgr.h>
52 #include <named/log.h>
53 #include <named/notify.h>
55 #include <named/server.h>
56 #include <named/update.h>
67 * All client state changes, other than that from idle to listening, occur
68 * as a result of events. This guarantees serialization and avoids the
71 * If a routine is ever created that allows someone other than the client's
72 * task to change the client, then the client will have to be locked.
75 #define NS_CLIENT_TRACE
76 #ifdef NS_CLIENT_TRACE
77 #define CTRACE(m) ns_client_log(client, \
78 NS_LOGCATEGORY_CLIENT, \
79 NS_LOGMODULE_CLIENT, \
82 #define MTRACE(m) isc_log_write(ns_g_lctx, \
83 NS_LOGCATEGORY_GENERAL, \
84 NS_LOGMODULE_CLIENT, \
86 "clientmgr @%p: %s", manager, (m))
88 #define CTRACE(m) ((void)(m))
89 #define MTRACE(m) ((void)(m))
92 #define TCP_CLIENT(c) (((c)->attributes & NS_CLIENTATTR_TCP) != 0)
94 #define TCP_BUFFER_SIZE (65535 + 2)
95 #define SEND_BUFFER_SIZE 4096
96 #define RECV_BUFFER_SIZE 4096
98 #ifdef ISC_PLATFORM_USETHREADS
101 * Number of 'mctx pools' for clients. (Should this be configurable?)
102 * When enabling threads, we use a pool of memory contexts shared by
103 * client objects, since concurrent access to a shared context would cause
104 * heavy contentions. The above constant is expected to be enough for
105 * completely avoiding contentions among threads for an authoritative-only
111 * If named with built without thread, simply share manager's context. Using
112 * a separate context in this case would simply waste memory.
116 /*% nameserver client manager structure */
117 struct ns_clientmgr {
121 /* The queue object has its own locks */
122 client_queue_t inactive; /*%< To be recycled */
125 isc_taskmgr_t * taskmgr;
126 isc_timermgr_t * timermgr;
128 /* Lock covers manager state. */
130 isc_boolean_t exiting;
132 /* Lock covers the clients list */
133 isc_mutex_t listlock;
134 client_list_t clients; /*%< All active clients */
136 /* Lock covers the recursing list */
138 client_list_t recursing; /*%< Recursing clients */
141 /*%< mctx pool for clients. */
142 unsigned int nextmctx;
143 isc_mem_t * mctxpool[NMCTXS];
147 #define MANAGER_MAGIC ISC_MAGIC('N', 'S', 'C', 'm')
148 #define VALID_MANAGER(m) ISC_MAGIC_VALID(m, MANAGER_MAGIC)
151 * Client object states. Ordering is significant: higher-numbered
152 * states are generally "more active", meaning that the client can
153 * have more dynamically allocated data, outstanding events, etc.
154 * In the list below, any such properties listed for state N
155 * also apply to any state > N.
157 * To force the client into a less active state, set client->newstate
158 * to that state and call exit_check(). This will cause any
159 * activities defined for higher-numbered states to be aborted.
162 #define NS_CLIENTSTATE_FREED 0
164 * The client object no longer exists.
167 #define NS_CLIENTSTATE_INACTIVE 1
169 * The client object exists and has a task and timer.
170 * Its "query" struct and sendbuf are initialized.
171 * It is on the client manager's list of inactive clients.
172 * It has a message and OPT, both in the reset state.
175 #define NS_CLIENTSTATE_READY 2
177 * The client object is either a TCP or a UDP one, and
178 * it is associated with a network interface. It is on the
179 * client manager's list of active clients.
181 * If it is a TCP client object, it has a TCP listener socket
182 * and an outstanding TCP listen request.
184 * If it is a UDP client object, it has a UDP listener socket
185 * and an outstanding UDP receive request.
188 #define NS_CLIENTSTATE_READING 3
190 * The client object is a TCP client object that has received
191 * a connection. It has a tcpsocket, tcpmsg, TCP quota, and an
192 * outstanding TCP read request. This state is not used for
193 * UDP client objects.
196 #define NS_CLIENTSTATE_WORKING 4
198 * The client object has received a request and is working
199 * on it. It has a view, and it may have any of a non-reset OPT,
200 * recursion quota, and an outstanding write request.
203 #define NS_CLIENTSTATE_RECURSING 5
205 * The client object is recursing. It will be on the 'recursing'
209 #define NS_CLIENTSTATE_MAX 9
211 * Sentinel value used to indicate "no state". When client->newstate
212 * has this value, we are not attempting to exit the current state.
213 * Must be greater than any valid state.
217 * Enable ns_client_dropport() by default.
219 #ifndef NS_CLIENT_DROPPORT
220 #define NS_CLIENT_DROPPORT 1
223 unsigned int ns_client_requests;
225 static void client_read(ns_client_t *client);
226 static void client_accept(ns_client_t *client);
227 static void client_udprecv(ns_client_t *client);
228 static void clientmgr_destroy(ns_clientmgr_t *manager);
229 static isc_boolean_t exit_check(ns_client_t *client);
230 static void ns_client_endrequest(ns_client_t *client);
231 static void client_start(isc_task_t *task, isc_event_t *event);
232 static void client_request(isc_task_t *task, isc_event_t *event);
233 static void ns_client_dumpmessage(ns_client_t *client, const char *reason);
234 static isc_result_t get_client(ns_clientmgr_t *manager, ns_interface_t *ifp,
235 dns_dispatch_t *disp, isc_boolean_t tcp);
236 static inline isc_boolean_t
237 allowed(isc_netaddr_t *addr, dns_name_t *signer, dns_acl_t *acl);
240 ns_client_recursing(ns_client_t *client) {
241 REQUIRE(NS_CLIENT_VALID(client));
242 REQUIRE(client->state == NS_CLIENTSTATE_WORKING);
244 LOCK(&client->manager->reclock);
245 client->newstate = client->state = NS_CLIENTSTATE_RECURSING;
246 ISC_LIST_APPEND(client->manager->recursing, client, rlink);
247 UNLOCK(&client->manager->reclock);
251 ns_client_killoldestquery(ns_client_t *client) {
253 REQUIRE(NS_CLIENT_VALID(client));
255 LOCK(&client->manager->reclock);
256 oldest = ISC_LIST_HEAD(client->manager->recursing);
257 if (oldest != NULL) {
258 ISC_LIST_UNLINK(client->manager->recursing, oldest, rlink);
259 UNLOCK(&client->manager->reclock);
260 ns_query_cancel(oldest);
262 UNLOCK(&client->manager->reclock);
266 ns_client_settimeout(ns_client_t *client, unsigned int seconds) {
268 isc_interval_t interval;
270 isc_interval_set(&interval, seconds, 0);
271 result = isc_timer_reset(client->timer, isc_timertype_once, NULL,
272 &interval, ISC_FALSE);
273 client->timerset = ISC_TRUE;
274 if (result != ISC_R_SUCCESS) {
275 ns_client_log(client, NS_LOGCATEGORY_CLIENT,
276 NS_LOGMODULE_CLIENT, ISC_LOG_ERROR,
277 "setting timeout: %s",
278 isc_result_totext(result));
279 /* Continue anyway. */
284 * Check for a deactivation or shutdown request and take appropriate
285 * action. Returns ISC_TRUE if either is in progress; in this case
286 * the caller must no longer use the client object as it may have been
290 exit_check(ns_client_t *client) {
291 isc_boolean_t destroy_manager = ISC_FALSE;
292 ns_clientmgr_t *manager = NULL;
294 REQUIRE(NS_CLIENT_VALID(client));
295 manager = client->manager;
297 if (client->state <= client->newstate)
298 return (ISC_FALSE); /* Business as usual. */
300 INSIST(client->newstate < NS_CLIENTSTATE_RECURSING);
303 * We need to detach from the view early when shutting down
304 * the server to break the following vicious circle:
306 * - The resolver will not shut down until the view refcount is zero
307 * - The view refcount does not go to zero until all clients detach
308 * - The client does not detach from the view until references is zero
309 * - references does not go to zero until the resolver has shut down
311 * Keep the view attached until any outstanding updates complete.
313 if (client->nupdates == 0 &&
314 client->newstate == NS_CLIENTSTATE_FREED && client->view != NULL)
315 dns_view_detach(&client->view);
317 if (client->state == NS_CLIENTSTATE_WORKING ||
318 client->state == NS_CLIENTSTATE_RECURSING)
320 INSIST(client->newstate <= NS_CLIENTSTATE_READING);
322 * Let the update processing complete.
324 if (client->nupdates > 0)
328 * We are trying to abort request processing.
330 if (client->nsends > 0) {
331 isc_socket_t *socket;
332 if (TCP_CLIENT(client))
333 socket = client->tcpsocket;
335 socket = client->udpsocket;
336 isc_socket_cancel(socket, client->task,
337 ISC_SOCKCANCEL_SEND);
340 if (! (client->nsends == 0 && client->nrecvs == 0 &&
341 client->references == 0))
344 * Still waiting for I/O cancel completion.
345 * or lingering references.
351 * I/O cancel is complete. Burn down all state
352 * related to the current request. Ensure that
353 * the client is no longer on the recursing list.
355 * We need to check whether the client is still linked,
356 * because it may already have been removed from the
357 * recursing list by ns_client_killoldestquery()
359 if (client->state == NS_CLIENTSTATE_RECURSING) {
360 LOCK(&manager->reclock);
361 if (ISC_LINK_LINKED(client, rlink))
362 ISC_LIST_UNLINK(manager->recursing,
364 UNLOCK(&manager->reclock);
366 ns_client_endrequest(client);
368 client->state = NS_CLIENTSTATE_READING;
369 INSIST(client->recursionquota == NULL);
371 if (NS_CLIENTSTATE_READING == client->newstate) {
373 client->newstate = NS_CLIENTSTATE_MAX;
374 return (ISC_TRUE); /* We're done. */
378 if (client->state == NS_CLIENTSTATE_READING) {
380 * We are trying to abort the current TCP connection,
383 INSIST(client->recursionquota == NULL);
384 INSIST(client->newstate <= NS_CLIENTSTATE_READY);
385 if (client->nreads > 0)
386 dns_tcpmsg_cancelread(&client->tcpmsg);
387 if (! client->nreads == 0) {
388 /* Still waiting for read cancel completion. */
392 if (client->tcpmsg_valid) {
393 dns_tcpmsg_invalidate(&client->tcpmsg);
394 client->tcpmsg_valid = ISC_FALSE;
396 if (client->tcpsocket != NULL) {
398 isc_socket_detach(&client->tcpsocket);
401 if (client->tcpquota != NULL)
402 isc_quota_detach(&client->tcpquota);
404 if (client->timerset) {
405 (void)isc_timer_reset(client->timer,
406 isc_timertype_inactive,
407 NULL, NULL, ISC_TRUE);
408 client->timerset = ISC_FALSE;
411 client->peeraddr_valid = ISC_FALSE;
413 client->state = NS_CLIENTSTATE_READY;
414 INSIST(client->recursionquota == NULL);
417 * Now the client is ready to accept a new TCP connection
418 * or UDP request, but we may have enough clients doing
419 * that already. Check whether this client needs to remain
420 * active and force it to go inactive if not.
422 * UDP clients go inactive at this point, but TCP clients
423 * may remain active if we have fewer active TCP client
424 * objects than desired due to an earlier quota exhaustion.
426 if (client->mortal && TCP_CLIENT(client) && !ns_g_clienttest) {
427 LOCK(&client->interface->lock);
428 if (client->interface->ntcpcurrent <
429 client->interface->ntcptarget)
430 client->mortal = ISC_FALSE;
431 UNLOCK(&client->interface->lock);
435 * We don't need the client; send it to the inactive
436 * queue for recycling.
438 if (client->mortal) {
439 if (client->newstate > NS_CLIENTSTATE_INACTIVE)
440 client->newstate = NS_CLIENTSTATE_INACTIVE;
443 if (NS_CLIENTSTATE_READY == client->newstate) {
444 if (TCP_CLIENT(client)) {
445 client_accept(client);
447 client_udprecv(client);
448 client->newstate = NS_CLIENTSTATE_MAX;
453 if (client->state == NS_CLIENTSTATE_READY) {
454 INSIST(client->newstate <= NS_CLIENTSTATE_INACTIVE);
457 * We are trying to enter the inactive state.
459 if (client->naccepts > 0)
460 isc_socket_cancel(client->tcplistener, client->task,
461 ISC_SOCKCANCEL_ACCEPT);
463 /* Still waiting for accept cancel completion. */
464 if (! (client->naccepts == 0))
467 /* Accept cancel is complete. */
468 if (client->nrecvs > 0)
469 isc_socket_cancel(client->udpsocket, client->task,
470 ISC_SOCKCANCEL_RECV);
472 /* Still waiting for recv cancel completion. */
473 if (! (client->nrecvs == 0))
476 /* Still waiting for control event to be delivered */
477 if (client->nctls > 0)
480 /* Deactivate the client. */
481 if (client->interface)
482 ns_interface_detach(&client->interface);
484 INSIST(client->naccepts == 0);
485 INSIST(client->recursionquota == NULL);
486 if (client->tcplistener != NULL)
487 isc_socket_detach(&client->tcplistener);
489 if (client->udpsocket != NULL)
490 isc_socket_detach(&client->udpsocket);
492 if (client->dispatch != NULL)
493 dns_dispatch_detach(&client->dispatch);
495 client->attributes = 0;
496 client->mortal = ISC_FALSE;
499 * Put the client on the inactive list. If we are aiming for
500 * the "freed" state, it will be removed from the inactive
501 * list shortly, and we need to keep the manager locked until
502 * that has been done, lest the manager decide to reactivate
503 * the dying client inbetween.
505 client->state = NS_CLIENTSTATE_INACTIVE;
506 INSIST(client->recursionquota == NULL);
508 if (client->state == client->newstate) {
509 client->newstate = NS_CLIENTSTATE_MAX;
510 if (!ns_g_clienttest && manager != NULL &&
512 ISC_QUEUE_PUSH(manager->inactive, client,
514 if (client->needshutdown)
515 isc_task_shutdown(client->task);
520 if (client->state == NS_CLIENTSTATE_INACTIVE) {
521 INSIST(client->newstate == NS_CLIENTSTATE_FREED);
523 * We are trying to free the client.
525 * When "shuttingdown" is true, either the task has received
526 * its shutdown event or no shutdown event has ever been
527 * set up. Thus, we have no outstanding shutdown
528 * event at this point.
530 REQUIRE(client->state == NS_CLIENTSTATE_INACTIVE);
532 INSIST(client->recursionquota == NULL);
533 INSIST(!ISC_QLINK_LINKED(client, ilink));
535 if (manager != NULL) {
536 LOCK(&manager->listlock);
537 ISC_LIST_UNLINK(manager->clients, client, link);
538 LOCK(&manager->lock);
539 if (manager->exiting &&
540 ISC_LIST_EMPTY(manager->clients))
541 destroy_manager = ISC_TRUE;
542 UNLOCK(&manager->lock);
543 UNLOCK(&manager->listlock);
546 ns_query_free(client);
547 isc_mem_put(client->mctx, client->recvbuf, RECV_BUFFER_SIZE);
548 isc_event_free((isc_event_t **)&client->sendevent);
549 isc_event_free((isc_event_t **)&client->recvevent);
550 isc_timer_detach(&client->timer);
552 if (client->tcpbuf != NULL)
553 isc_mem_put(client->mctx, client->tcpbuf,
555 if (client->opt != NULL) {
556 INSIST(dns_rdataset_isassociated(client->opt));
557 dns_rdataset_disassociate(client->opt);
558 dns_message_puttemprdataset(client->message,
562 dns_message_destroy(&client->message);
565 * Detaching the task must be done after unlinking from
566 * the manager's lists because the manager accesses
569 if (client->task != NULL)
570 isc_task_detach(&client->task);
576 * Check that there are no other external references to
577 * the memory context.
579 if (ns_g_clienttest && isc_mem_references(client->mctx) != 1) {
580 isc_mem_stats(client->mctx, stderr);
585 * Destroy the fetchlock mutex that was created in
588 DESTROYLOCK(&client->query.fetchlock);
590 isc_mem_putanddetach(&client->mctx, client, sizeof(*client));
593 if (destroy_manager && manager != NULL)
594 clientmgr_destroy(manager);
600 * The client's task has received the client's control event
601 * as part of the startup process.
604 client_start(isc_task_t *task, isc_event_t *event) {
605 ns_client_t *client = (ns_client_t *) event->ev_arg;
607 INSIST(task == client->task);
611 INSIST(client->nctls == 1);
614 if (exit_check(client))
617 if (TCP_CLIENT(client)) {
618 client_accept(client);
620 client_udprecv(client);
626 * The client's task has received a shutdown event.
629 client_shutdown(isc_task_t *task, isc_event_t *event) {
632 REQUIRE(event != NULL);
633 REQUIRE(event->ev_type == ISC_TASKEVENT_SHUTDOWN);
634 client = event->ev_arg;
635 REQUIRE(NS_CLIENT_VALID(client));
636 REQUIRE(task == client->task);
642 isc_event_free(&event);
644 if (client->shutdown != NULL) {
645 (client->shutdown)(client->shutdown_arg, ISC_R_SHUTTINGDOWN);
646 client->shutdown = NULL;
647 client->shutdown_arg = NULL;
650 if (ISC_QLINK_LINKED(client, ilink))
651 ISC_QUEUE_UNLINK(client->manager->inactive, client, ilink);
653 client->newstate = NS_CLIENTSTATE_FREED;
654 client->needshutdown = ISC_FALSE;
655 (void)exit_check(client);
659 ns_client_endrequest(ns_client_t *client) {
660 INSIST(client->naccepts == 0);
661 INSIST(client->nreads == 0);
662 INSIST(client->nsends == 0);
663 INSIST(client->nrecvs == 0);
664 INSIST(client->nupdates == 0);
665 INSIST(client->state == NS_CLIENTSTATE_WORKING ||
666 client->state == NS_CLIENTSTATE_RECURSING);
668 CTRACE("endrequest");
670 if (client->next != NULL) {
671 (client->next)(client);
675 if (client->view != NULL)
676 dns_view_detach(&client->view);
677 if (client->opt != NULL) {
678 INSIST(dns_rdataset_isassociated(client->opt));
679 dns_rdataset_disassociate(client->opt);
680 dns_message_puttemprdataset(client->message, &client->opt);
683 client->signer = NULL;
684 client->udpsize = 512;
685 client->extflags = 0;
686 client->ednsversion = -1;
687 dns_message_reset(client->message, DNS_MESSAGE_INTENTPARSE);
689 if (client->recursionquota != NULL)
690 isc_quota_detach(&client->recursionquota);
693 * Clear all client attributes that are specific to
694 * the request; that's all except the TCP flag.
696 client->attributes &= NS_CLIENTATTR_TCP;
700 ns_client_next(ns_client_t *client, isc_result_t result) {
703 REQUIRE(NS_CLIENT_VALID(client));
704 REQUIRE(client->state == NS_CLIENTSTATE_WORKING ||
705 client->state == NS_CLIENTSTATE_RECURSING ||
706 client->state == NS_CLIENTSTATE_READING);
710 if (result != ISC_R_SUCCESS)
711 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
712 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(3),
713 "request failed: %s", isc_result_totext(result));
716 * An error processing a TCP request may have left
717 * the connection out of sync. To be safe, we always
718 * sever the connection when result != ISC_R_SUCCESS.
720 if (result == ISC_R_SUCCESS && TCP_CLIENT(client))
721 newstate = NS_CLIENTSTATE_READING;
723 newstate = NS_CLIENTSTATE_READY;
725 if (client->newstate > newstate)
726 client->newstate = newstate;
727 (void)exit_check(client);
732 client_senddone(isc_task_t *task, isc_event_t *event) {
734 isc_socketevent_t *sevent = (isc_socketevent_t *) event;
736 REQUIRE(sevent != NULL);
737 REQUIRE(sevent->ev_type == ISC_SOCKEVENT_SENDDONE);
738 client = sevent->ev_arg;
739 REQUIRE(NS_CLIENT_VALID(client));
740 REQUIRE(task == client->task);
741 REQUIRE(sevent == client->sendevent);
747 if (sevent->result != ISC_R_SUCCESS)
748 ns_client_log(client, NS_LOGCATEGORY_CLIENT,
749 NS_LOGMODULE_CLIENT, ISC_LOG_WARNING,
750 "error sending response: %s",
751 isc_result_totext(sevent->result));
753 INSIST(client->nsends > 0);
756 if (client->tcpbuf != NULL) {
757 INSIST(TCP_CLIENT(client));
758 isc_mem_put(client->mctx, client->tcpbuf, TCP_BUFFER_SIZE);
759 client->tcpbuf = NULL;
762 ns_client_next(client, ISC_R_SUCCESS);
766 * We only want to fail with ISC_R_NOSPACE when called from
767 * ns_client_sendraw() and not when called from ns_client_send(),
768 * tcpbuffer is NULL when called from ns_client_sendraw() and
769 * length != 0. tcpbuffer != NULL when called from ns_client_send()
774 client_allocsendbuf(ns_client_t *client, isc_buffer_t *buffer,
775 isc_buffer_t *tcpbuffer, isc_uint32_t length,
776 unsigned char *sendbuf, unsigned char **datap)
779 isc_uint32_t bufsize;
782 INSIST(datap != NULL);
783 INSIST((tcpbuffer == NULL && length != 0) ||
784 (tcpbuffer != NULL && length == 0));
786 if (TCP_CLIENT(client)) {
787 INSIST(client->tcpbuf == NULL);
788 if (length + 2 > TCP_BUFFER_SIZE) {
789 result = ISC_R_NOSPACE;
792 client->tcpbuf = isc_mem_get(client->mctx, TCP_BUFFER_SIZE);
793 if (client->tcpbuf == NULL) {
794 result = ISC_R_NOMEMORY;
797 data = client->tcpbuf;
798 if (tcpbuffer != NULL) {
799 isc_buffer_init(tcpbuffer, data, TCP_BUFFER_SIZE);
800 isc_buffer_init(buffer, data + 2, TCP_BUFFER_SIZE - 2);
802 isc_buffer_init(buffer, data, TCP_BUFFER_SIZE);
803 INSIST(length <= 0xffff);
804 isc_buffer_putuint16(buffer, (isc_uint16_t)length);
808 if (client->udpsize < SEND_BUFFER_SIZE)
809 bufsize = client->udpsize;
811 bufsize = SEND_BUFFER_SIZE;
812 if (length > bufsize) {
813 result = ISC_R_NOSPACE;
816 isc_buffer_init(buffer, data, bufsize);
819 result = ISC_R_SUCCESS;
826 client_sendpkg(ns_client_t *client, isc_buffer_t *buffer) {
827 struct in6_pktinfo *pktinfo;
830 isc_sockaddr_t *address;
831 isc_socket_t *socket;
832 isc_netaddr_t netaddr;
834 unsigned int sockflags = ISC_SOCKFLAG_IMMEDIATE;
836 if (TCP_CLIENT(client)) {
837 socket = client->tcpsocket;
840 socket = client->udpsocket;
841 address = &client->peeraddr;
843 isc_netaddr_fromsockaddr(&netaddr, &client->peeraddr);
844 if (ns_g_server->blackholeacl != NULL &&
845 dns_acl_match(&netaddr, NULL,
846 ns_g_server->blackholeacl,
847 &ns_g_server->aclenv,
848 &match, NULL) == ISC_R_SUCCESS &&
850 return (DNS_R_BLACKHOLED);
851 sockflags |= ISC_SOCKFLAG_NORETRY;
854 if ((client->attributes & NS_CLIENTATTR_PKTINFO) != 0 &&
855 (client->attributes & NS_CLIENTATTR_MULTICAST) == 0)
856 pktinfo = &client->pktinfo;
860 isc_buffer_usedregion(buffer, &r);
864 result = isc_socket_sendto2(socket, &r, client->task,
866 client->sendevent, sockflags);
867 if (result == ISC_R_SUCCESS || result == ISC_R_INPROGRESS) {
869 if (result == ISC_R_SUCCESS)
870 client_senddone(client->task,
871 (isc_event_t *)client->sendevent);
872 result = ISC_R_SUCCESS;
878 ns_client_sendraw(ns_client_t *client, dns_message_t *message) {
884 unsigned char sendbuf[SEND_BUFFER_SIZE];
886 REQUIRE(NS_CLIENT_VALID(client));
890 mr = dns_message_getrawmessage(message);
892 result = ISC_R_UNEXPECTEDEND;
896 result = client_allocsendbuf(client, &buffer, NULL, mr->length,
898 if (result != ISC_R_SUCCESS)
902 * Copy message to buffer and fixup id.
904 isc_buffer_availableregion(&buffer, &r);
905 result = isc_buffer_copyregion(&buffer, mr);
906 if (result != ISC_R_SUCCESS)
908 r.base[0] = (client->message->id >> 8) & 0xff;
909 r.base[1] = client->message->id & 0xff;
911 result = client_sendpkg(client, &buffer);
912 if (result == ISC_R_SUCCESS)
916 if (client->tcpbuf != NULL) {
917 isc_mem_put(client->mctx, client->tcpbuf, TCP_BUFFER_SIZE);
918 client->tcpbuf = NULL;
920 ns_client_next(client, result);
924 ns_client_send(ns_client_t *client) {
928 isc_buffer_t tcpbuffer;
931 isc_boolean_t cleanup_cctx = ISC_FALSE;
932 unsigned char sendbuf[SEND_BUFFER_SIZE];
933 unsigned int render_opts;
934 unsigned int preferred_glue;
935 isc_boolean_t opt_included = ISC_FALSE;
937 REQUIRE(NS_CLIENT_VALID(client));
941 if ((client->attributes & NS_CLIENTATTR_RA) != 0)
942 client->message->flags |= DNS_MESSAGEFLAG_RA;
944 if ((client->attributes & NS_CLIENTATTR_WANTDNSSEC) != 0)
947 render_opts = DNS_MESSAGERENDER_OMITDNSSEC;
950 if (client->view != NULL) {
951 if (client->view->preferred_glue == dns_rdatatype_a)
952 preferred_glue = DNS_MESSAGERENDER_PREFER_A;
953 else if (client->view->preferred_glue == dns_rdatatype_aaaa)
954 preferred_glue = DNS_MESSAGERENDER_PREFER_AAAA;
957 #ifdef ALLOW_FILTER_AAAA_ON_V4
959 * filter-aaaa-on-v4 yes or break-dnssec option to suppress
961 * We already know that request came via IPv4,
962 * that we have both AAAA and A records,
963 * and that we either have no signatures that the client wants
964 * or we are supposed to break DNSSEC.
966 * Override preferred glue if necessary.
968 if ((client->attributes & NS_CLIENTATTR_FILTER_AAAA) != 0) {
969 render_opts |= DNS_MESSAGERENDER_FILTER_AAAA;
970 if (preferred_glue == DNS_MESSAGERENDER_PREFER_AAAA)
971 preferred_glue = DNS_MESSAGERENDER_PREFER_A;
976 * XXXRTH The following doesn't deal with TCP buffer resizing.
978 result = client_allocsendbuf(client, &buffer, &tcpbuffer, 0,
980 if (result != ISC_R_SUCCESS)
983 result = dns_compress_init(&cctx, -1, client->mctx);
984 if (result != ISC_R_SUCCESS)
986 if (client->peeraddr_valid && client->view != NULL) {
987 isc_netaddr_t netaddr;
988 dns_name_t *name = NULL;
990 isc_netaddr_fromsockaddr(&netaddr, &client->peeraddr);
991 if (client->message->tsigkey != NULL)
992 name = &client->message->tsigkey->name;
993 if (client->view->nocasecompress == NULL ||
994 !allowed(&netaddr, name, client->view->nocasecompress))
996 dns_compress_setsensitive(&cctx, ISC_TRUE);
999 cleanup_cctx = ISC_TRUE;
1001 result = dns_message_renderbegin(client->message, &cctx, &buffer);
1002 if (result != ISC_R_SUCCESS)
1005 if (client->opt != NULL) {
1006 result = dns_message_setopt(client->message, client->opt);
1007 opt_included = ISC_TRUE;
1009 if (result != ISC_R_SUCCESS)
1012 result = dns_message_rendersection(client->message,
1013 DNS_SECTION_QUESTION, 0);
1014 if (result == ISC_R_NOSPACE) {
1015 client->message->flags |= DNS_MESSAGEFLAG_TC;
1018 if (result != ISC_R_SUCCESS)
1022 * Stop after the question if TC was set for rate limiting.
1024 if ((client->message->flags & DNS_MESSAGEFLAG_TC) != 0)
1026 #endif /* USE_RRL */
1027 result = dns_message_rendersection(client->message,
1029 DNS_MESSAGERENDER_PARTIAL |
1031 if (result == ISC_R_NOSPACE) {
1032 client->message->flags |= DNS_MESSAGEFLAG_TC;
1035 if (result != ISC_R_SUCCESS)
1037 result = dns_message_rendersection(client->message,
1038 DNS_SECTION_AUTHORITY,
1039 DNS_MESSAGERENDER_PARTIAL |
1041 if (result == ISC_R_NOSPACE) {
1042 client->message->flags |= DNS_MESSAGEFLAG_TC;
1045 if (result != ISC_R_SUCCESS)
1047 result = dns_message_rendersection(client->message,
1048 DNS_SECTION_ADDITIONAL,
1049 preferred_glue | render_opts);
1050 if (result != ISC_R_SUCCESS && result != ISC_R_NOSPACE)
1053 result = dns_message_renderend(client->message);
1055 if (result != ISC_R_SUCCESS)
1059 dns_compress_invalidate(&cctx);
1060 cleanup_cctx = ISC_FALSE;
1063 if (TCP_CLIENT(client)) {
1064 isc_buffer_usedregion(&buffer, &r);
1065 isc_buffer_putuint16(&tcpbuffer, (isc_uint16_t) r.length);
1066 isc_buffer_add(&tcpbuffer, r.length);
1067 result = client_sendpkg(client, &tcpbuffer);
1069 result = client_sendpkg(client, &buffer);
1071 /* update statistics (XXXJT: is it okay to access message->xxxkey?) */
1072 isc_stats_increment(ns_g_server->nsstats, dns_nsstatscounter_response);
1074 isc_stats_increment(ns_g_server->nsstats,
1075 dns_nsstatscounter_edns0out);
1077 if (client->message->tsigkey != NULL) {
1078 isc_stats_increment(ns_g_server->nsstats,
1079 dns_nsstatscounter_tsigout);
1081 if (client->message->sig0key != NULL) {
1082 isc_stats_increment(ns_g_server->nsstats,
1083 dns_nsstatscounter_sig0out);
1085 if ((client->message->flags & DNS_MESSAGEFLAG_TC) != 0)
1086 isc_stats_increment(ns_g_server->nsstats,
1087 dns_nsstatscounter_truncatedresp);
1089 if (result == ISC_R_SUCCESS)
1093 if (client->tcpbuf != NULL) {
1094 isc_mem_put(client->mctx, client->tcpbuf, TCP_BUFFER_SIZE);
1095 client->tcpbuf = NULL;
1099 dns_compress_invalidate(&cctx);
1101 ns_client_next(client, result);
1104 #if NS_CLIENT_DROPPORT
1105 #define DROPPORT_NO 0
1106 #define DROPPORT_REQUEST 1
1107 #define DROPPORT_RESPONSE 2
1109 * ns_client_dropport determines if certain requests / responses
1110 * should be dropped based on the port number.
1113 * \li 0: Don't drop.
1114 * \li 1: Drop request.
1115 * \li 2: Drop (error) response.
1118 ns_client_dropport(in_port_t port) {
1121 case 13: /* daytime */
1122 case 19: /* chargen */
1124 return (DROPPORT_REQUEST);
1125 case 464: /* kpasswd */
1126 return (DROPPORT_RESPONSE);
1128 return (DROPPORT_NO);
1133 ns_client_error(ns_client_t *client, isc_result_t result) {
1135 dns_message_t *message;
1137 REQUIRE(NS_CLIENT_VALID(client));
1141 message = client->message;
1142 rcode = dns_result_torcode(result);
1144 #if NS_CLIENT_DROPPORT
1146 * Don't send FORMERR to ports on the drop port list.
1148 if (rcode == dns_rcode_formerr &&
1149 ns_client_dropport(isc_sockaddr_getport(&client->peeraddr)) !=
1154 isc_buffer_init(&b, buf, sizeof(buf) - 1);
1155 if (dns_rcode_totext(rcode, &b) != ISC_R_SUCCESS)
1156 isc_buffer_putstr(&b, "UNKNOWN RCODE");
1157 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
1158 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(10),
1159 "dropped error (%.*s) response: suspicious port",
1160 (int)isc_buffer_usedlength(&b), buf);
1161 ns_client_next(client, ISC_R_SUCCESS);
1168 * Try to rate limit error responses.
1170 if (client->view != NULL && client->view->rrl != NULL) {
1171 isc_boolean_t wouldlog;
1172 char log_buf[DNS_RRL_LOG_BUF_LEN];
1173 dns_rrl_result_t rrl_result;
1175 INSIST(rcode != dns_rcode_noerror &&
1176 rcode != dns_rcode_nxdomain);
1177 wouldlog = isc_log_wouldlog(ns_g_lctx, DNS_RRL_LOG_DROP);
1178 rrl_result = dns_rrl(client->view, &client->peeraddr,
1180 dns_rdataclass_in, dns_rdatatype_none,
1181 NULL, result, client->now,
1182 wouldlog, log_buf, sizeof(log_buf));
1183 if (rrl_result != DNS_RRL_RESULT_OK) {
1185 * Log dropped errors in the query category
1186 * so that they are not lost in silence.
1187 * Starts of rate-limited bursts are logged in
1188 * NS_LOGCATEGORY_RRL.
1191 ns_client_log(client,
1192 NS_LOGCATEGORY_QUERY_EERRORS,
1193 NS_LOGMODULE_CLIENT,
1198 * Some error responses cannot be 'slipped',
1199 * so don't try to slip any error responses.
1201 if (!client->view->rrl->log_only) {
1202 isc_stats_increment(ns_g_server->nsstats,
1203 dns_nsstatscounter_ratedropped);
1204 isc_stats_increment(ns_g_server->nsstats,
1205 dns_nsstatscounter_dropped);
1206 ns_client_next(client, DNS_R_DROP);
1211 #endif /* USE_RRL */
1214 * Message may be an in-progress reply that we had trouble
1215 * with, in which case QR will be set. We need to clear QR before
1216 * calling dns_message_reply() to avoid triggering an assertion.
1218 message->flags &= ~DNS_MESSAGEFLAG_QR;
1220 * AA and AD shouldn't be set.
1222 message->flags &= ~(DNS_MESSAGEFLAG_AA | DNS_MESSAGEFLAG_AD);
1223 result = dns_message_reply(message, ISC_TRUE);
1224 if (result != ISC_R_SUCCESS) {
1226 * It could be that we've got a query with a good header,
1227 * but a bad question section, so we try again with
1228 * want_question_section set to ISC_FALSE.
1230 result = dns_message_reply(message, ISC_FALSE);
1231 if (result != ISC_R_SUCCESS) {
1232 ns_client_next(client, result);
1236 message->rcode = rcode;
1239 * FORMERR loop avoidance: If we sent a FORMERR message
1240 * with the same ID to the same client less than two
1241 * seconds ago, assume that we are in an infinite error
1242 * packet dialog with a server for some protocol whose
1243 * error responses look enough like DNS queries to
1244 * elicit a FORMERR response. Drop a packet to break
1247 if (rcode == dns_rcode_formerr) {
1248 if (isc_sockaddr_equal(&client->peeraddr,
1249 &client->formerrcache.addr) &&
1250 message->id == client->formerrcache.id &&
1251 client->requesttime - client->formerrcache.time < 2) {
1253 ns_client_log(client, NS_LOGCATEGORY_CLIENT,
1254 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(1),
1255 "possible error packet loop, "
1257 ns_client_next(client, result);
1260 client->formerrcache.addr = client->peeraddr;
1261 client->formerrcache.time = client->requesttime;
1262 client->formerrcache.id = message->id;
1264 ns_client_send(client);
1267 static inline isc_result_t
1268 client_addopt(ns_client_t *client) {
1269 char nsid[BUFSIZ], *nsidp;
1270 isc_result_t result;
1272 dns_resolver_t *resolver;
1273 isc_uint16_t udpsize;
1274 dns_ednsopt_t ednsopts[DNS_EDNSOPTIONS];
1278 REQUIRE(client->opt == NULL); /* XXXRTH free old. */
1280 view = client->view;
1281 resolver = (view != NULL) ? view->resolver : NULL;
1282 if (resolver != NULL)
1283 udpsize = dns_resolver_getudpsize(resolver);
1285 udpsize = ns_g_udpsize;
1287 flags = client->extflags & DNS_MESSAGEEXTFLAG_REPLYPRESERVE;
1289 /* Set EDNS options if applicable */
1290 if ((client->attributes & NS_CLIENTATTR_WANTNSID) != 0 &&
1291 (ns_g_server->server_id != NULL ||
1292 ns_g_server->server_usehostname)) {
1293 if (ns_g_server->server_usehostname) {
1294 result = ns_os_gethostname(nsid, sizeof(nsid));
1295 if (result != ISC_R_SUCCESS) {
1300 nsidp = ns_g_server->server_id;
1302 INSIST(count < DNS_EDNSOPTIONS);
1303 ednsopts[count].code = DNS_OPT_NSID;
1304 ednsopts[count].length = strlen(nsidp);
1305 ednsopts[count].value = (unsigned char *)nsidp;
1309 result = dns_message_buildopt(client->message, &client->opt, 0,
1310 udpsize, flags, ednsopts, count);
1314 static inline isc_boolean_t
1315 allowed(isc_netaddr_t *addr, dns_name_t *signer, dns_acl_t *acl) {
1317 isc_result_t result;
1321 result = dns_acl_match(addr, signer, acl, &ns_g_server->aclenv,
1323 if (result == ISC_R_SUCCESS && match > 0)
1329 * Callback to see if a non-recursive query coming from 'srcaddr' to
1330 * 'destaddr', with optional key 'mykey' for class 'rdclass' would be
1331 * delivered to 'myview'.
1333 * We run this unlocked as both the view list and the interface list
1334 * are updated when the appropriate task has exclusivity.
1337 ns_client_isself(dns_view_t *myview, dns_tsigkey_t *mykey,
1338 isc_sockaddr_t *srcaddr, isc_sockaddr_t *dstaddr,
1339 dns_rdataclass_t rdclass, void *arg)
1342 dns_tsigkey_t *key = NULL;
1343 dns_name_t *tsig = NULL;
1344 isc_netaddr_t netsrc;
1345 isc_netaddr_t netdst;
1350 * ns_g_server->interfacemgr is task exclusive locked.
1352 if (ns_g_server->interfacemgr == NULL)
1355 if (!ns_interfacemgr_listeningon(ns_g_server->interfacemgr, dstaddr))
1358 isc_netaddr_fromsockaddr(&netsrc, srcaddr);
1359 isc_netaddr_fromsockaddr(&netdst, dstaddr);
1361 for (view = ISC_LIST_HEAD(ns_g_server->viewlist);
1363 view = ISC_LIST_NEXT(view, link)) {
1365 if (view->matchrecursiveonly)
1368 if (rdclass != view->rdclass)
1371 if (mykey != NULL) {
1372 isc_boolean_t match;
1373 isc_result_t result;
1375 result = dns_view_gettsig(view, &mykey->name, &key);
1376 if (result != ISC_R_SUCCESS)
1378 match = dst_key_compare(mykey->key, key->key);
1379 dns_tsigkey_detach(&key);
1382 tsig = dns_tsigkey_identity(mykey);
1385 if (allowed(&netsrc, tsig, view->matchclients) &&
1386 allowed(&netdst, tsig, view->matchdestinations))
1389 return (ISC_TF(view == myview));
1393 process_opt(ns_client_t *client, dns_rdataset_t *opt) {
1395 isc_buffer_t optbuf;
1396 isc_result_t result;
1397 isc_uint16_t optcode;
1398 isc_uint16_t optlen;
1401 * Set the client's UDP buffer size.
1403 client->udpsize = opt->rdclass;
1406 * If the requested UDP buffer size is less than 512,
1407 * ignore it and use 512.
1409 if (client->udpsize < 512)
1410 client->udpsize = 512;
1413 * Get the flags out of the OPT record.
1415 client->extflags = (isc_uint16_t)(opt->ttl & 0xFFFF);
1418 * Do we understand this version of EDNS?
1420 * XXXRTH need library support for this!
1422 client->ednsversion = (opt->ttl & 0x00FF0000) >> 16;
1423 if (client->ednsversion > 0) {
1424 isc_stats_increment(ns_g_server->nsstats,
1425 dns_nsstatscounter_badednsver);
1426 result = client_addopt(client);
1427 if (result == ISC_R_SUCCESS)
1428 result = DNS_R_BADVERS;
1429 ns_client_error(client, result);
1433 /* Check for NSID request */
1434 result = dns_rdataset_first(opt);
1435 if (result == ISC_R_SUCCESS) {
1436 dns_rdata_init(&rdata);
1437 dns_rdataset_current(opt, &rdata);
1438 isc_buffer_init(&optbuf, rdata.data, rdata.length);
1439 isc_buffer_add(&optbuf, rdata.length);
1440 while (isc_buffer_remaininglength(&optbuf) >= 4) {
1441 optcode = isc_buffer_getuint16(&optbuf);
1442 optlen = isc_buffer_getuint16(&optbuf);
1445 client->attributes |= NS_CLIENTATTR_WANTNSID;
1446 isc_buffer_forward(&optbuf, optlen);
1449 isc_buffer_forward(&optbuf, optlen);
1455 isc_stats_increment(ns_g_server->nsstats, dns_nsstatscounter_edns0in);
1458 * Create an OPT for our reply.
1460 result = client_addopt(client);
1461 if (result != ISC_R_SUCCESS) {
1462 ns_client_error(client, result);
1470 * Handle an incoming request event from the socket (UDP case)
1471 * or tcpmsg (TCP case).
1474 client_request(isc_task_t *task, isc_event_t *event) {
1475 ns_client_t *client;
1476 isc_socketevent_t *sevent;
1477 isc_result_t result;
1478 isc_result_t sigresult = ISC_R_SUCCESS;
1479 isc_buffer_t *buffer;
1480 isc_buffer_t tbuffer;
1482 dns_rdataset_t *opt;
1483 dns_name_t *signame;
1484 isc_boolean_t ra; /* Recursion available. */
1485 isc_netaddr_t netaddr;
1489 isc_boolean_t notimp;
1491 REQUIRE(event != NULL);
1492 client = event->ev_arg;
1493 REQUIRE(NS_CLIENT_VALID(client));
1494 REQUIRE(task == client->task);
1496 INSIST(client->recursionquota == NULL);
1498 INSIST(client->state == (TCP_CLIENT(client) ?
1499 NS_CLIENTSTATE_READING :
1500 NS_CLIENTSTATE_READY));
1502 ns_client_requests++;
1504 if (event->ev_type == ISC_SOCKEVENT_RECVDONE) {
1505 INSIST(!TCP_CLIENT(client));
1506 sevent = (isc_socketevent_t *)event;
1507 REQUIRE(sevent == client->recvevent);
1508 isc_buffer_init(&tbuffer, sevent->region.base, sevent->n);
1509 isc_buffer_add(&tbuffer, sevent->n);
1511 result = sevent->result;
1512 if (result == ISC_R_SUCCESS) {
1513 client->peeraddr = sevent->address;
1514 client->peeraddr_valid = ISC_TRUE;
1516 if ((sevent->attributes & ISC_SOCKEVENTATTR_PKTINFO) != 0) {
1517 client->attributes |= NS_CLIENTATTR_PKTINFO;
1518 client->pktinfo = sevent->pktinfo;
1520 if ((sevent->attributes & ISC_SOCKEVENTATTR_MULTICAST) != 0)
1521 client->attributes |= NS_CLIENTATTR_MULTICAST;
1524 INSIST(TCP_CLIENT(client));
1525 REQUIRE(event->ev_type == DNS_EVENT_TCPMSG);
1526 REQUIRE(event->ev_sender == &client->tcpmsg);
1527 buffer = &client->tcpmsg.buffer;
1528 result = client->tcpmsg.result;
1529 INSIST(client->nreads == 1);
1531 * client->peeraddr was set when the connection was accepted.
1536 if (exit_check(client))
1538 client->state = client->newstate = NS_CLIENTSTATE_WORKING;
1540 isc_task_getcurrenttime(task, &client->requesttime);
1541 client->now = client->requesttime;
1543 if (result != ISC_R_SUCCESS) {
1544 if (TCP_CLIENT(client)) {
1545 ns_client_next(client, result);
1547 if (result != ISC_R_CANCELED)
1548 isc_log_write(ns_g_lctx, NS_LOGCATEGORY_CLIENT,
1549 NS_LOGMODULE_CLIENT,
1551 "UDP client handler shutting "
1552 "down due to fatal receive "
1554 isc_result_totext(result));
1555 isc_task_shutdown(client->task);
1560 isc_netaddr_fromsockaddr(&netaddr, &client->peeraddr);
1562 #if NS_CLIENT_DROPPORT
1563 if (ns_client_dropport(isc_sockaddr_getport(&client->peeraddr)) ==
1565 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
1566 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(10),
1567 "dropped request: suspicious port");
1568 ns_client_next(client, ISC_R_SUCCESS);
1573 ns_client_log(client, NS_LOGCATEGORY_CLIENT,
1574 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(3),
1576 TCP_CLIENT(client) ? "TCP" : "UDP");
1579 * Check the blackhole ACL for UDP only, since TCP is done in
1582 if (!TCP_CLIENT(client)) {
1584 if (ns_g_server->blackholeacl != NULL &&
1585 dns_acl_match(&netaddr, NULL, ns_g_server->blackholeacl,
1586 &ns_g_server->aclenv,
1587 &match, NULL) == ISC_R_SUCCESS &&
1590 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
1591 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(10),
1592 "blackholed UDP datagram");
1593 ns_client_next(client, ISC_R_SUCCESS);
1599 * Silently drop multicast requests for the present.
1600 * XXXMPA revisit this as mDNS spec was published.
1602 if ((client->attributes & NS_CLIENTATTR_MULTICAST) != 0) {
1603 ns_client_log(client, NS_LOGCATEGORY_CLIENT,
1604 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(2),
1605 "dropping multicast request");
1606 ns_client_next(client, DNS_R_REFUSED);
1610 result = dns_message_peekheader(buffer, &id, &flags);
1611 if (result != ISC_R_SUCCESS) {
1613 * There isn't enough header to determine whether
1614 * this was a request or a response. Drop it.
1616 ns_client_next(client, result);
1621 * The client object handles requests, not responses.
1622 * If this is a UDP response, forward it to the dispatcher.
1623 * If it's a TCP response, discard it here.
1625 if ((flags & DNS_MESSAGEFLAG_QR) != 0) {
1626 if (TCP_CLIENT(client)) {
1627 CTRACE("unexpected response");
1628 ns_client_next(client, DNS_R_FORMERR);
1631 dns_dispatch_importrecv(client->dispatch, event);
1632 ns_client_next(client, ISC_R_SUCCESS);
1638 * Update some statistics counters. Don't count responses.
1640 if (isc_sockaddr_pf(&client->peeraddr) == PF_INET) {
1641 isc_stats_increment(ns_g_server->nsstats,
1642 dns_nsstatscounter_requestv4);
1644 isc_stats_increment(ns_g_server->nsstats,
1645 dns_nsstatscounter_requestv6);
1647 if (TCP_CLIENT(client))
1648 isc_stats_increment(ns_g_server->nsstats,
1649 dns_nsstatscounter_tcp);
1652 * It's a request. Parse it.
1654 result = dns_message_parse(client->message, buffer, 0);
1655 if (result != ISC_R_SUCCESS) {
1657 * Parsing the request failed. Send a response
1658 * (typically FORMERR or SERVFAIL).
1660 ns_client_error(client, result);
1664 dns_opcodestats_increment(ns_g_server->opcodestats,
1665 client->message->opcode);
1666 switch (client->message->opcode) {
1667 case dns_opcode_query:
1668 case dns_opcode_update:
1669 case dns_opcode_notify:
1672 case dns_opcode_iquery:
1678 client->message->rcode = dns_rcode_noerror;
1680 /* RFC1123 section 6.1.3.2 */
1681 if ((client->attributes & NS_CLIENTATTR_MULTICAST) != 0)
1682 client->message->flags &= ~DNS_MESSAGEFLAG_RD;
1690 opt = dns_message_getopt(client->message);
1693 * Are we dropping all EDNS queries?
1695 if (ns_g_dropedns) {
1696 ns_client_next(client, ISC_R_SUCCESS);
1699 result = process_opt(client, opt);
1700 if (result != ISC_R_SUCCESS)
1704 if (client->message->rdclass == 0) {
1705 ns_client_log(client, NS_LOGCATEGORY_CLIENT,
1706 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(1),
1707 "message class could not be determined");
1708 ns_client_dumpmessage(client,
1709 "message class could not be determined");
1710 ns_client_error(client, notimp ? DNS_R_NOTIMP : DNS_R_FORMERR);
1715 * Determine the destination address. If the receiving interface is
1716 * bound to a specific address, we simply use it regardless of the
1717 * address family. All IPv4 queries should fall into this case.
1718 * Otherwise, if this is a TCP query, get the address from the
1719 * receiving socket (this needs a system call and can be heavy).
1720 * For IPv6 UDP queries, we get this from the pktinfo structure (if
1722 * If all the attempts fail (this can happen due to memory shortage,
1723 * etc), we regard this as an error for safety.
1725 if ((client->interface->flags & NS_INTERFACEFLAG_ANYADDR) == 0)
1726 isc_netaddr_fromsockaddr(&client->destaddr,
1727 &client->interface->addr);
1729 isc_sockaddr_t sockaddr;
1730 result = ISC_R_FAILURE;
1732 if (TCP_CLIENT(client))
1733 result = isc_socket_getsockname(client->tcpsocket,
1735 if (result == ISC_R_SUCCESS)
1736 isc_netaddr_fromsockaddr(&client->destaddr, &sockaddr);
1737 if (result != ISC_R_SUCCESS &&
1738 client->interface->addr.type.sa.sa_family == AF_INET6 &&
1739 (client->attributes & NS_CLIENTATTR_PKTINFO) != 0) {
1741 * XXXJT technically, we should convert the receiving
1742 * interface ID to a proper scope zone ID. However,
1743 * due to the fact there is no standard API for this,
1744 * we only handle link-local addresses and use the
1745 * interface index as link ID. Despite the assumption,
1746 * it should cover most typical cases.
1748 isc_netaddr_fromin6(&client->destaddr,
1749 &client->pktinfo.ipi6_addr);
1750 if (IN6_IS_ADDR_LINKLOCAL(&client->pktinfo.ipi6_addr))
1751 isc_netaddr_setzone(&client->destaddr,
1752 client->pktinfo.ipi6_ifindex);
1753 result = ISC_R_SUCCESS;
1755 if (result != ISC_R_SUCCESS) {
1756 UNEXPECTED_ERROR(__FILE__, __LINE__,
1757 "failed to get request's "
1759 isc_result_totext(result));
1760 ns_client_next(client, ISC_R_SUCCESS);
1766 * Find a view that matches the client's source address.
1768 for (view = ISC_LIST_HEAD(ns_g_server->viewlist);
1770 view = ISC_LIST_NEXT(view, link)) {
1771 if (client->message->rdclass == view->rdclass ||
1772 client->message->rdclass == dns_rdataclass_any)
1774 dns_name_t *tsig = NULL;
1776 sigresult = dns_message_rechecksig(client->message,
1778 if (sigresult == ISC_R_SUCCESS)
1779 tsig = dns_tsigkey_identity(client->message->tsigkey);
1781 if (allowed(&netaddr, tsig, view->matchclients) &&
1782 allowed(&client->destaddr, tsig,
1783 view->matchdestinations) &&
1784 !((client->message->flags & DNS_MESSAGEFLAG_RD)
1785 == 0 && view->matchrecursiveonly))
1787 dns_view_attach(view, &client->view);
1794 char classname[DNS_RDATACLASS_FORMATSIZE];
1797 * Do a dummy TSIG verification attempt so that the
1798 * response will have a TSIG if the query did, as
1799 * required by RFC2845.
1804 dns_message_resetsig(client->message);
1806 r = dns_message_getrawmessage(client->message);
1807 isc_buffer_init(&b, r->base, r->length);
1808 isc_buffer_add(&b, r->length);
1809 (void)dns_tsig_verify(&b, client->message, NULL, NULL);
1811 dns_rdataclass_format(client->message->rdclass, classname,
1813 ns_client_log(client, NS_LOGCATEGORY_CLIENT,
1814 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(1),
1815 "no matching view in class '%s'", classname);
1816 ns_client_dumpmessage(client, "no matching view in class");
1817 ns_client_error(client, notimp ? DNS_R_NOTIMP : DNS_R_REFUSED);
1821 ns_client_log(client, NS_LOGCATEGORY_CLIENT,
1822 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(5),
1823 "using view '%s'", view->name);
1826 * Check for a signature. We log bad signatures regardless of
1827 * whether they ultimately cause the request to be rejected or
1828 * not. We do not log the lack of a signature unless we are
1831 client->signer = NULL;
1832 dns_name_init(&client->signername, NULL);
1833 result = dns_message_signer(client->message, &client->signername);
1834 if (result != ISC_R_NOTFOUND) {
1836 if (dns_message_gettsig(client->message, &signame) != NULL) {
1837 isc_stats_increment(ns_g_server->nsstats,
1838 dns_nsstatscounter_tsigin);
1840 isc_stats_increment(ns_g_server->nsstats,
1841 dns_nsstatscounter_sig0in);
1845 if (result == ISC_R_SUCCESS) {
1846 char namebuf[DNS_NAME_FORMATSIZE];
1847 dns_name_format(&client->signername, namebuf, sizeof(namebuf));
1848 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
1849 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(3),
1850 "request has valid signature: %s", namebuf);
1851 client->signer = &client->signername;
1852 } else if (result == ISC_R_NOTFOUND) {
1853 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
1854 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(3),
1855 "request is not signed");
1856 } else if (result == DNS_R_NOIDENTITY) {
1857 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
1858 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(3),
1859 "request is signed by a nonauthoritative key");
1864 isc_result_t tresult;
1866 /* There is a signature, but it is bad. */
1867 isc_stats_increment(ns_g_server->nsstats,
1868 dns_nsstatscounter_invalidsig);
1870 if (dns_message_gettsig(client->message, &signame) != NULL) {
1871 char namebuf[DNS_NAME_FORMATSIZE];
1872 char cnamebuf[DNS_NAME_FORMATSIZE];
1873 dns_name_format(signame, namebuf, sizeof(namebuf));
1874 status = client->message->tsigstatus;
1875 isc_buffer_init(&b, tsigrcode, sizeof(tsigrcode) - 1);
1876 tresult = dns_tsigrcode_totext(status, &b);
1877 INSIST(tresult == ISC_R_SUCCESS);
1878 tsigrcode[isc_buffer_usedlength(&b)] = '\0';
1879 if (client->message->tsigkey->generated) {
1880 dns_name_format(client->message->tsigkey->creator,
1881 cnamebuf, sizeof(cnamebuf));
1882 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
1883 NS_LOGMODULE_CLIENT,
1885 "request has invalid signature: "
1886 "TSIG %s (%s): %s (%s)", namebuf,
1888 isc_result_totext(result),
1891 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
1892 NS_LOGMODULE_CLIENT,
1894 "request has invalid signature: "
1895 "TSIG %s: %s (%s)", namebuf,
1896 isc_result_totext(result),
1900 status = client->message->sig0status;
1901 isc_buffer_init(&b, tsigrcode, sizeof(tsigrcode) - 1);
1902 tresult = dns_tsigrcode_totext(status, &b);
1903 INSIST(tresult == ISC_R_SUCCESS);
1904 tsigrcode[isc_buffer_usedlength(&b)] = '\0';
1905 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
1906 NS_LOGMODULE_CLIENT, ISC_LOG_ERROR,
1907 "request has invalid signature: %s (%s)",
1908 isc_result_totext(result), tsigrcode);
1911 * Accept update messages signed by unknown keys so that
1912 * update forwarding works transparently through slaves
1913 * that don't have all the same keys as the master.
1915 if (!(client->message->tsigstatus == dns_tsigerror_badkey &&
1916 client->message->opcode == dns_opcode_update)) {
1917 ns_client_error(client, sigresult);
1923 * Decide whether recursive service is available to this client.
1924 * We do this here rather than in the query code so that we can
1925 * set the RA bit correctly on all kinds of responses, not just
1926 * responses to ordinary queries. Note if you can't query the
1927 * cache there is no point in setting RA.
1930 if (client->view->resolver != NULL &&
1931 client->view->recursion == ISC_TRUE &&
1932 ns_client_checkaclsilent(client, NULL,
1933 client->view->recursionacl,
1934 ISC_TRUE) == ISC_R_SUCCESS &&
1935 ns_client_checkaclsilent(client, NULL,
1936 client->view->cacheacl,
1937 ISC_TRUE) == ISC_R_SUCCESS &&
1938 ns_client_checkaclsilent(client, &client->destaddr,
1939 client->view->recursiononacl,
1940 ISC_TRUE) == ISC_R_SUCCESS &&
1941 ns_client_checkaclsilent(client, &client->destaddr,
1942 client->view->cacheonacl,
1943 ISC_TRUE) == ISC_R_SUCCESS)
1947 client->attributes |= NS_CLIENTATTR_RA;
1949 ns_client_log(client, DNS_LOGCATEGORY_SECURITY, NS_LOGMODULE_CLIENT,
1950 ISC_LOG_DEBUG(3), ra ? "recursion available" :
1951 "recursion not available");
1954 * Adjust maximum UDP response size for this client.
1956 if (client->udpsize > 512) {
1957 dns_peer_t *peer = NULL;
1958 isc_uint16_t udpsize = view->maxudp;
1959 (void) dns_peerlist_peerbyaddr(view->peers, &netaddr, &peer);
1961 dns_peer_getmaxudp(peer, &udpsize);
1962 if (client->udpsize > udpsize)
1963 client->udpsize = udpsize;
1967 * Dispatch the request.
1969 switch (client->message->opcode) {
1970 case dns_opcode_query:
1972 ns_query_start(client);
1974 case dns_opcode_update:
1976 ns_client_settimeout(client, 60);
1977 ns_update_start(client, sigresult);
1979 case dns_opcode_notify:
1981 ns_client_settimeout(client, 60);
1982 ns_notify_start(client);
1984 case dns_opcode_iquery:
1986 ns_client_error(client, DNS_R_NOTIMP);
1989 CTRACE("unknown opcode");
1990 ns_client_error(client, DNS_R_NOTIMP);
1998 client_timeout(isc_task_t *task, isc_event_t *event) {
1999 ns_client_t *client;
2001 REQUIRE(event != NULL);
2002 REQUIRE(event->ev_type == ISC_TIMEREVENT_LIFE ||
2003 event->ev_type == ISC_TIMEREVENT_IDLE);
2004 client = event->ev_arg;
2005 REQUIRE(NS_CLIENT_VALID(client));
2006 REQUIRE(task == client->task);
2007 REQUIRE(client->timer != NULL);
2013 isc_event_free(&event);
2015 if (client->shutdown != NULL) {
2016 (client->shutdown)(client->shutdown_arg, ISC_R_TIMEDOUT);
2017 client->shutdown = NULL;
2018 client->shutdown_arg = NULL;
2021 if (client->newstate > NS_CLIENTSTATE_READY)
2022 client->newstate = NS_CLIENTSTATE_READY;
2023 (void)exit_check(client);
2027 get_clientmctx(ns_clientmgr_t *manager, isc_mem_t **mctxp) {
2028 isc_mem_t *clientmctx;
2029 isc_result_t result;
2031 unsigned int nextmctx;
2034 MTRACE("clientmctx");
2037 * Caller must be holding the manager lock.
2039 if (ns_g_clienttest) {
2040 result = isc_mem_create(0, 0, mctxp);
2041 if (result == ISC_R_SUCCESS)
2042 isc_mem_setname(*mctxp, "client", NULL);
2046 nextmctx = manager->nextmctx++;
2047 if (manager->nextmctx == NMCTXS)
2048 manager->nextmctx = 0;
2050 INSIST(nextmctx < NMCTXS);
2052 clientmctx = manager->mctxpool[nextmctx];
2053 if (clientmctx == NULL) {
2054 result = isc_mem_create(0, 0, &clientmctx);
2055 if (result != ISC_R_SUCCESS)
2057 isc_mem_setname(clientmctx, "client", NULL);
2059 manager->mctxpool[nextmctx] = clientmctx;
2062 clientmctx = manager->mctx;
2065 isc_mem_attach(clientmctx, mctxp);
2067 return (ISC_R_SUCCESS);
2071 client_create(ns_clientmgr_t *manager, ns_client_t **clientp) {
2072 ns_client_t *client;
2073 isc_result_t result;
2074 isc_mem_t *mctx = NULL;
2077 * Caller must be holding the manager lock.
2079 * Note: creating a client does not add the client to the
2080 * manager's client list or set the client's manager pointer.
2081 * The caller is responsible for that.
2084 REQUIRE(clientp != NULL && *clientp == NULL);
2086 result = get_clientmctx(manager, &mctx);
2087 if (result != ISC_R_SUCCESS)
2090 client = isc_mem_get(mctx, sizeof(*client));
2091 if (client == NULL) {
2092 isc_mem_detach(&mctx);
2093 return (ISC_R_NOMEMORY);
2095 client->mctx = mctx;
2097 client->task = NULL;
2098 result = isc_task_create(manager->taskmgr, 0, &client->task);
2099 if (result != ISC_R_SUCCESS)
2100 goto cleanup_client;
2101 isc_task_setname(client->task, "client", client);
2103 client->timer = NULL;
2104 result = isc_timer_create(manager->timermgr, isc_timertype_inactive,
2105 NULL, NULL, client->task, client_timeout,
2106 client, &client->timer);
2107 if (result != ISC_R_SUCCESS)
2109 client->timerset = ISC_FALSE;
2111 client->message = NULL;
2112 result = dns_message_create(client->mctx, DNS_MESSAGE_INTENTPARSE,
2114 if (result != ISC_R_SUCCESS)
2117 /* XXXRTH Hardwired constants */
2119 client->sendevent = (isc_socketevent_t *)
2120 isc_event_allocate(client->mctx, client,
2121 ISC_SOCKEVENT_SENDDONE,
2122 client_senddone, client,
2123 sizeof(isc_socketevent_t));
2124 if (client->sendevent == NULL) {
2125 result = ISC_R_NOMEMORY;
2126 goto cleanup_message;
2129 client->recvbuf = isc_mem_get(client->mctx, RECV_BUFFER_SIZE);
2130 if (client->recvbuf == NULL) {
2131 result = ISC_R_NOMEMORY;
2132 goto cleanup_sendevent;
2135 client->recvevent = (isc_socketevent_t *)
2136 isc_event_allocate(client->mctx, client,
2137 ISC_SOCKEVENT_RECVDONE,
2138 client_request, client,
2139 sizeof(isc_socketevent_t));
2140 if (client->recvevent == NULL) {
2141 result = ISC_R_NOMEMORY;
2142 goto cleanup_recvbuf;
2145 client->magic = NS_CLIENT_MAGIC;
2146 client->manager = NULL;
2147 client->state = NS_CLIENTSTATE_INACTIVE;
2148 client->newstate = NS_CLIENTSTATE_MAX;
2149 client->naccepts = 0;
2153 client->nupdates = 0;
2155 client->references = 0;
2156 client->attributes = 0;
2157 client->view = NULL;
2158 client->dispatch = NULL;
2159 client->udpsocket = NULL;
2160 client->tcplistener = NULL;
2161 client->tcpsocket = NULL;
2162 client->tcpmsg_valid = ISC_FALSE;
2163 client->tcpbuf = NULL;
2165 client->udpsize = 512;
2166 client->extflags = 0;
2167 client->ednsversion = -1;
2168 client->next = NULL;
2169 client->shutdown = NULL;
2170 client->shutdown_arg = NULL;
2171 client->signer = NULL;
2172 dns_name_init(&client->signername, NULL);
2173 client->mortal = ISC_FALSE;
2174 client->tcpquota = NULL;
2175 client->recursionquota = NULL;
2176 client->interface = NULL;
2177 client->peeraddr_valid = ISC_FALSE;
2178 #ifdef ALLOW_FILTER_AAAA_ON_V4
2179 client->filter_aaaa = dns_v4_aaaa_ok;
2181 client->needshutdown = ns_g_clienttest;
2183 ISC_EVENT_INIT(&client->ctlevent, sizeof(client->ctlevent), 0, NULL,
2184 NS_EVENT_CLIENTCONTROL, client_start, client, client,
2187 * Initialize FORMERR cache to sentinel value that will not match
2188 * any actual FORMERR response.
2190 isc_sockaddr_any(&client->formerrcache.addr);
2191 client->formerrcache.time = 0;
2192 client->formerrcache.id = 0;
2193 ISC_LINK_INIT(client, link);
2194 ISC_LINK_INIT(client, rlink);
2195 ISC_QLINK_INIT(client, ilink);
2198 * We call the init routines for the various kinds of client here,
2199 * after we have created an otherwise valid client, because some
2200 * of them call routines that REQUIRE(NS_CLIENT_VALID(client)).
2202 result = ns_query_init(client);
2203 if (result != ISC_R_SUCCESS)
2204 goto cleanup_recvevent;
2206 result = isc_task_onshutdown(client->task, client_shutdown, client);
2207 if (result != ISC_R_SUCCESS)
2214 return (ISC_R_SUCCESS);
2217 ns_query_free(client);
2220 isc_event_free((isc_event_t **)&client->recvevent);
2223 isc_mem_put(client->mctx, client->recvbuf, RECV_BUFFER_SIZE);
2226 isc_event_free((isc_event_t **)&client->sendevent);
2231 dns_message_destroy(&client->message);
2234 isc_timer_detach(&client->timer);
2237 isc_task_detach(&client->task);
2240 isc_mem_putanddetach(&client->mctx, client, sizeof(*client));
2246 client_read(ns_client_t *client) {
2247 isc_result_t result;
2251 result = dns_tcpmsg_readmessage(&client->tcpmsg, client->task,
2252 client_request, client);
2253 if (result != ISC_R_SUCCESS)
2257 * Set a timeout to limit the amount of time we will wait
2258 * for a request on this TCP connection.
2260 ns_client_settimeout(client, 30);
2262 client->state = client->newstate = NS_CLIENTSTATE_READING;
2263 INSIST(client->nreads == 0);
2264 INSIST(client->recursionquota == NULL);
2269 ns_client_next(client, result);
2273 client_newconn(isc_task_t *task, isc_event_t *event) {
2274 ns_client_t *client = event->ev_arg;
2275 isc_socket_newconnev_t *nevent = (isc_socket_newconnev_t *)event;
2276 isc_result_t result;
2278 REQUIRE(event->ev_type == ISC_SOCKEVENT_NEWCONN);
2279 REQUIRE(NS_CLIENT_VALID(client));
2280 REQUIRE(client->task == task);
2284 INSIST(client->state == NS_CLIENTSTATE_READY);
2286 INSIST(client->naccepts == 1);
2289 LOCK(&client->interface->lock);
2290 INSIST(client->interface->ntcpcurrent > 0);
2291 client->interface->ntcpcurrent--;
2292 UNLOCK(&client->interface->lock);
2295 * We must take ownership of the new socket before the exit
2296 * check to make sure it gets destroyed if we decide to exit.
2298 if (nevent->result == ISC_R_SUCCESS) {
2299 client->tcpsocket = nevent->newsocket;
2300 isc_socket_setname(client->tcpsocket, "client-tcp", NULL);
2301 client->state = NS_CLIENTSTATE_READING;
2302 INSIST(client->recursionquota == NULL);
2304 (void)isc_socket_getpeername(client->tcpsocket,
2306 client->peeraddr_valid = ISC_TRUE;
2307 ns_client_log(client, NS_LOGCATEGORY_CLIENT,
2308 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(3),
2309 "new TCP connection");
2312 * XXXRTH What should we do? We're trying to accept but
2313 * it didn't work. If we just give up, then TCP
2314 * service may eventually stop.
2316 * For now, we just go idle.
2318 * Going idle is probably the right thing if the
2321 ns_client_log(client, NS_LOGCATEGORY_CLIENT,
2322 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(3),
2323 "accept failed: %s",
2324 isc_result_totext(nevent->result));
2327 if (exit_check(client))
2330 if (nevent->result == ISC_R_SUCCESS) {
2332 isc_netaddr_t netaddr;
2334 isc_netaddr_fromsockaddr(&netaddr, &client->peeraddr);
2336 if (ns_g_server->blackholeacl != NULL &&
2337 dns_acl_match(&netaddr, NULL,
2338 ns_g_server->blackholeacl,
2339 &ns_g_server->aclenv,
2340 &match, NULL) == ISC_R_SUCCESS &&
2343 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
2344 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(10),
2345 "blackholed connection attempt");
2346 client->newstate = NS_CLIENTSTATE_READY;
2347 (void)exit_check(client);
2351 INSIST(client->tcpmsg_valid == ISC_FALSE);
2352 dns_tcpmsg_init(client->mctx, client->tcpsocket,
2354 client->tcpmsg_valid = ISC_TRUE;
2357 * Let a new client take our place immediately, before
2358 * we wait for a request packet. If we don't,
2359 * telnetting to port 53 (once per CPU) will
2360 * deny service to legitimate TCP clients.
2362 result = isc_quota_attach(&ns_g_server->tcpquota,
2364 if (result == ISC_R_SUCCESS)
2365 result = ns_client_replace(client);
2366 if (result != ISC_R_SUCCESS) {
2367 ns_client_log(client, NS_LOGCATEGORY_CLIENT,
2368 NS_LOGMODULE_CLIENT, ISC_LOG_WARNING,
2369 "no more TCP clients: %s",
2370 isc_result_totext(result));
2373 client_read(client);
2377 isc_event_free(&event);
2381 client_accept(ns_client_t *client) {
2382 isc_result_t result;
2386 result = isc_socket_accept(client->tcplistener, client->task,
2387 client_newconn, client);
2388 if (result != ISC_R_SUCCESS) {
2389 UNEXPECTED_ERROR(__FILE__, __LINE__,
2390 "isc_socket_accept() failed: %s",
2391 isc_result_totext(result));
2393 * XXXRTH What should we do? We're trying to accept but
2394 * it didn't work. If we just give up, then TCP
2395 * service may eventually stop.
2397 * For now, we just go idle.
2401 INSIST(client->naccepts == 0);
2403 LOCK(&client->interface->lock);
2404 client->interface->ntcpcurrent++;
2405 UNLOCK(&client->interface->lock);
2409 client_udprecv(ns_client_t *client) {
2410 isc_result_t result;
2415 r.base = client->recvbuf;
2416 r.length = RECV_BUFFER_SIZE;
2417 result = isc_socket_recv2(client->udpsocket, &r, 1,
2418 client->task, client->recvevent, 0);
2419 if (result != ISC_R_SUCCESS) {
2420 UNEXPECTED_ERROR(__FILE__, __LINE__,
2421 "isc_socket_recv2() failed: %s",
2422 isc_result_totext(result));
2424 * This cannot happen in the current implementation, since
2425 * isc_socket_recv2() cannot fail if flags == 0.
2427 * If this does fail, we just go idle.
2431 INSIST(client->nrecvs == 0);
2436 ns_client_attach(ns_client_t *source, ns_client_t **targetp) {
2437 REQUIRE(NS_CLIENT_VALID(source));
2438 REQUIRE(targetp != NULL && *targetp == NULL);
2440 source->references++;
2441 ns_client_log(source, NS_LOGCATEGORY_CLIENT,
2442 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(10),
2443 "ns_client_attach: ref = %d", source->references);
2448 ns_client_detach(ns_client_t **clientp) {
2449 ns_client_t *client = *clientp;
2451 client->references--;
2452 INSIST(client->references >= 0);
2454 ns_client_log(client, NS_LOGCATEGORY_CLIENT,
2455 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(10),
2456 "ns_client_detach: ref = %d", client->references);
2457 (void)exit_check(client);
2461 ns_client_shuttingdown(ns_client_t *client) {
2462 return (ISC_TF(client->newstate == NS_CLIENTSTATE_FREED));
2466 ns_client_replace(ns_client_t *client) {
2467 isc_result_t result;
2471 REQUIRE(client != NULL);
2472 REQUIRE(client->manager != NULL);
2474 result = get_client(client->manager, client->interface,
2475 client->dispatch, TCP_CLIENT(client));
2476 if (result != ISC_R_SUCCESS)
2480 * The responsibility for listening for new requests is hereby
2481 * transferred to the new client. Therefore, the old client
2482 * should refrain from listening for any more requests.
2484 client->mortal = ISC_TRUE;
2486 return (ISC_R_SUCCESS);
2494 clientmgr_destroy(ns_clientmgr_t *manager) {
2499 REQUIRE(ISC_LIST_EMPTY(manager->clients));
2501 MTRACE("clientmgr_destroy");
2504 for (i = 0; i < NMCTXS; i++) {
2505 if (manager->mctxpool[i] != NULL)
2506 isc_mem_detach(&manager->mctxpool[i]);
2510 ISC_QUEUE_DESTROY(manager->inactive);
2511 DESTROYLOCK(&manager->lock);
2512 DESTROYLOCK(&manager->listlock);
2513 DESTROYLOCK(&manager->reclock);
2515 isc_mem_put(manager->mctx, manager, sizeof(*manager));
2519 ns_clientmgr_create(isc_mem_t *mctx, isc_taskmgr_t *taskmgr,
2520 isc_timermgr_t *timermgr, ns_clientmgr_t **managerp)
2522 ns_clientmgr_t *manager;
2523 isc_result_t result;
2528 manager = isc_mem_get(mctx, sizeof(*manager));
2529 if (manager == NULL)
2530 return (ISC_R_NOMEMORY);
2532 result = isc_mutex_init(&manager->lock);
2533 if (result != ISC_R_SUCCESS)
2534 goto cleanup_manager;
2536 result = isc_mutex_init(&manager->listlock);
2537 if (result != ISC_R_SUCCESS)
2540 result = isc_mutex_init(&manager->reclock);
2541 if (result != ISC_R_SUCCESS)
2542 goto cleanup_listlock;
2544 manager->mctx = mctx;
2545 manager->taskmgr = taskmgr;
2546 manager->timermgr = timermgr;
2547 manager->exiting = ISC_FALSE;
2548 ISC_LIST_INIT(manager->clients);
2549 ISC_LIST_INIT(manager->recursing);
2550 ISC_QUEUE_INIT(manager->inactive, ilink);
2552 manager->nextmctx = 0;
2553 for (i = 0; i < NMCTXS; i++)
2554 manager->mctxpool[i] = NULL; /* will be created on-demand */
2556 manager->magic = MANAGER_MAGIC;
2560 *managerp = manager;
2562 return (ISC_R_SUCCESS);
2565 (void) isc_mutex_destroy(&manager->listlock);
2568 (void) isc_mutex_destroy(&manager->lock);
2571 isc_mem_put(manager->mctx, manager, sizeof(*manager));
2577 ns_clientmgr_destroy(ns_clientmgr_t **managerp) {
2578 isc_result_t result;
2579 ns_clientmgr_t *manager;
2580 ns_client_t *client;
2581 isc_boolean_t need_destroy = ISC_FALSE, unlock = ISC_FALSE;
2583 REQUIRE(managerp != NULL);
2584 manager = *managerp;
2585 REQUIRE(VALID_MANAGER(manager));
2590 * Check for success because we may already be task-exclusive
2591 * at this point. Only if we succeed at obtaining an exclusive
2592 * lock now will we need to relinquish it later.
2594 result = isc_task_beginexclusive(ns_g_server->task);
2595 if (result == ISC_R_SUCCESS)
2598 manager->exiting = ISC_TRUE;
2600 for (client = ISC_LIST_HEAD(manager->clients);
2602 client = ISC_LIST_NEXT(client, link))
2603 isc_task_shutdown(client->task);
2605 if (ISC_LIST_EMPTY(manager->clients))
2606 need_destroy = ISC_TRUE;
2609 isc_task_endexclusive(ns_g_server->task);
2612 clientmgr_destroy(manager);
2618 get_client(ns_clientmgr_t *manager, ns_interface_t *ifp,
2619 dns_dispatch_t *disp, isc_boolean_t tcp)
2621 isc_result_t result = ISC_R_SUCCESS;
2623 ns_client_t *client;
2624 MTRACE("get client");
2626 REQUIRE(manager != NULL);
2628 if (manager->exiting)
2629 return (ISC_R_SHUTTINGDOWN);
2632 * Allocate a client. First try to get a recycled one;
2633 * if that fails, make a new one.
2636 if (!ns_g_clienttest)
2637 ISC_QUEUE_POP(manager->inactive, ilink, client);
2642 MTRACE("create new");
2644 LOCK(&manager->lock);
2645 result = client_create(manager, &client);
2646 UNLOCK(&manager->lock);
2647 if (result != ISC_R_SUCCESS)
2650 LOCK(&manager->listlock);
2651 ISC_LIST_APPEND(manager->clients, client, link);
2652 UNLOCK(&manager->listlock);
2655 client->manager = manager;
2656 ns_interface_attach(ifp, &client->interface);
2657 client->state = NS_CLIENTSTATE_READY;
2658 INSIST(client->recursionquota == NULL);
2661 client->attributes |= NS_CLIENTATTR_TCP;
2662 isc_socket_attach(ifp->tcpsocket,
2663 &client->tcplistener);
2667 dns_dispatch_attach(disp, &client->dispatch);
2668 sock = dns_dispatch_getsocket(client->dispatch);
2669 isc_socket_attach(sock, &client->udpsocket);
2672 INSIST(client->nctls == 0);
2674 ev = &client->ctlevent;
2675 isc_task_send(client->task, &ev);
2677 return (ISC_R_SUCCESS);
2681 ns_clientmgr_createclients(ns_clientmgr_t *manager, unsigned int n,
2682 ns_interface_t *ifp, isc_boolean_t tcp)
2684 isc_result_t result = ISC_R_SUCCESS;
2687 REQUIRE(VALID_MANAGER(manager));
2690 MTRACE("createclients");
2692 for (disp = 0; disp < n; disp++) {
2693 result = get_client(manager, ifp, ifp->udpdispatch[disp], tcp);
2694 if (result != ISC_R_SUCCESS)
2702 ns_client_getsockaddr(ns_client_t *client) {
2703 return (&client->peeraddr);
2707 ns_client_checkaclsilent(ns_client_t *client, isc_netaddr_t *netaddr,
2708 dns_acl_t *acl, isc_boolean_t default_allow)
2710 isc_result_t result;
2711 isc_netaddr_t tmpnetaddr;
2721 if (netaddr == NULL) {
2722 isc_netaddr_fromsockaddr(&tmpnetaddr, &client->peeraddr);
2723 netaddr = &tmpnetaddr;
2726 result = dns_acl_match(netaddr, client->signer, acl,
2727 &ns_g_server->aclenv, &match, NULL);
2729 if (result != ISC_R_SUCCESS)
2730 goto deny; /* Internal error, already logged. */
2733 goto deny; /* Negative match or no match. */
2736 return (ISC_R_SUCCESS);
2739 return (DNS_R_REFUSED);
2743 ns_client_checkacl(ns_client_t *client, isc_sockaddr_t *sockaddr,
2744 const char *opname, dns_acl_t *acl,
2745 isc_boolean_t default_allow, int log_level)
2747 isc_result_t result;
2748 isc_netaddr_t netaddr;
2750 if (sockaddr != NULL)
2751 isc_netaddr_fromsockaddr(&netaddr, sockaddr);
2753 result = ns_client_checkaclsilent(client, sockaddr ? &netaddr : NULL,
2754 acl, default_allow);
2756 if (result == ISC_R_SUCCESS)
2757 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
2758 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(3),
2759 "%s approved", opname);
2761 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
2762 NS_LOGMODULE_CLIENT,
2763 log_level, "%s denied", opname);
2768 ns_client_name(ns_client_t *client, char *peerbuf, size_t len) {
2769 if (client->peeraddr_valid)
2770 isc_sockaddr_format(&client->peeraddr, peerbuf,
2773 snprintf(peerbuf, len, "@%p", client);
2777 ns_client_logv(ns_client_t *client, isc_logcategory_t *category,
2778 isc_logmodule_t *module, int level, const char *fmt, va_list ap)
2781 char peerbuf[ISC_SOCKADDR_FORMATSIZE];
2782 char signerbuf[DNS_NAME_FORMATSIZE], qnamebuf[DNS_NAME_FORMATSIZE];
2783 const char *viewname = "";
2784 const char *sep1 = "", *sep2 = "", *sep3 = "", *sep4 = "";
2785 const char *signer = "", *qname = "";
2786 dns_name_t *q = NULL;
2788 vsnprintf(msgbuf, sizeof(msgbuf), fmt, ap);
2790 ns_client_name(client, peerbuf, sizeof(peerbuf));
2792 if (client->signer != NULL) {
2793 dns_name_format(client->signer, signerbuf, sizeof(signerbuf));
2798 q = client->query.origqname != NULL
2799 ? client->query.origqname : client->query.qname;
2801 dns_name_format(q, qnamebuf, sizeof(qnamebuf));
2807 if (client->view != NULL && strcmp(client->view->name, "_bind") != 0 &&
2808 strcmp(client->view->name, "_default") != 0) {
2810 viewname = client->view->name;
2813 isc_log_write(ns_g_lctx, category, module, level,
2814 "client %s%s%s%s%s%s%s%s: %s",
2815 peerbuf, sep1, signer, sep2, qname, sep3,
2816 sep4, viewname, msgbuf);
2820 ns_client_log(ns_client_t *client, isc_logcategory_t *category,
2821 isc_logmodule_t *module, int level, const char *fmt, ...)
2825 if (! isc_log_wouldlog(ns_g_lctx, level))
2829 ns_client_logv(client, category, module, level, fmt, ap);
2834 ns_client_aclmsg(const char *msg, dns_name_t *name, dns_rdatatype_t type,
2835 dns_rdataclass_t rdclass, char *buf, size_t len)
2837 char namebuf[DNS_NAME_FORMATSIZE];
2838 char typebuf[DNS_RDATATYPE_FORMATSIZE];
2839 char classbuf[DNS_RDATACLASS_FORMATSIZE];
2841 dns_name_format(name, namebuf, sizeof(namebuf));
2842 dns_rdatatype_format(type, typebuf, sizeof(typebuf));
2843 dns_rdataclass_format(rdclass, classbuf, sizeof(classbuf));
2844 (void)snprintf(buf, len, "%s '%s/%s/%s'", msg, namebuf, typebuf,
2849 ns_client_dumpmessage(ns_client_t *client, const char *reason) {
2850 isc_buffer_t buffer;
2853 isc_result_t result;
2855 if (!isc_log_wouldlog(ns_g_lctx, ISC_LOG_DEBUG(1)))
2859 * Note that these are multiline debug messages. We want a newline
2860 * to appear in the log after each message.
2864 buf = isc_mem_get(client->mctx, len);
2867 isc_buffer_init(&buffer, buf, len);
2868 result = dns_message_totext(client->message,
2869 &dns_master_style_debug,
2871 if (result == ISC_R_NOSPACE) {
2872 isc_mem_put(client->mctx, buf, len);
2874 } else if (result == ISC_R_SUCCESS)
2875 ns_client_log(client, NS_LOGCATEGORY_UNMATCHED,
2876 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(1),
2878 (int)isc_buffer_usedlength(&buffer),
2880 } while (result == ISC_R_NOSPACE);
2883 isc_mem_put(client->mctx, buf, len);
2887 ns_client_dumprecursing(FILE *f, ns_clientmgr_t *manager) {
2888 ns_client_t *client;
2889 char namebuf[DNS_NAME_FORMATSIZE];
2890 char original[DNS_NAME_FORMATSIZE];
2891 char peerbuf[ISC_SOCKADDR_FORMATSIZE];
2892 char typebuf[DNS_RDATATYPE_FORMATSIZE];
2893 char classbuf[DNS_RDATACLASS_FORMATSIZE];
2896 const char *origfor;
2897 dns_rdataset_t *rdataset;
2899 REQUIRE(VALID_MANAGER(manager));
2901 LOCK(&manager->reclock);
2902 client = ISC_LIST_HEAD(manager->recursing);
2903 while (client != NULL) {
2904 INSIST(client->state == NS_CLIENTSTATE_RECURSING);
2906 ns_client_name(client, peerbuf, sizeof(peerbuf));
2907 if (client->view != NULL &&
2908 strcmp(client->view->name, "_bind") != 0 &&
2909 strcmp(client->view->name, "_default") != 0) {
2910 name = client->view->name;
2917 LOCK(&client->query.fetchlock);
2918 INSIST(client->query.qname != NULL);
2919 dns_name_format(client->query.qname, namebuf, sizeof(namebuf));
2920 if (client->query.qname != client->query.origqname &&
2921 client->query.origqname != NULL) {
2923 dns_name_format(client->query.origqname, original,
2929 rdataset = ISC_LIST_HEAD(client->query.qname->list);
2930 if (rdataset == NULL && client->query.origqname != NULL)
2931 rdataset = ISC_LIST_HEAD(client->query.origqname->list);
2932 if (rdataset != NULL) {
2933 dns_rdatatype_format(rdataset->type, typebuf,
2935 dns_rdataclass_format(rdataset->rdclass, classbuf,
2938 strcpy(typebuf, "-");
2939 strcpy(classbuf, "-");
2941 UNLOCK(&client->query.fetchlock);
2942 fprintf(f, "; client %s%s%s: id %u '%s/%s/%s'%s%s "
2943 "requesttime %d\n", peerbuf, sep, name,
2944 client->message->id, namebuf, typebuf, classbuf,
2945 origfor, original, client->requesttime);
2946 client = ISC_LIST_NEXT(client, rlink);
2948 UNLOCK(&manager->reclock);
2952 ns_client_qnamereplace(ns_client_t *client, dns_name_t *name) {
2953 LOCK(&client->query.fetchlock);
2954 if (client->query.restarts > 0) {
2956 * client->query.qname was dynamically allocated.
2958 dns_message_puttempname(client->message,
2959 &client->query.qname);
2961 client->query.qname = name;
2962 UNLOCK(&client->query.fetchlock);
2966 ns_client_sourceip(dns_clientinfo_t *ci, isc_sockaddr_t **addrp) {
2967 ns_client_t *client = (ns_client_t *) ci->data;
2969 REQUIRE(NS_CLIENT_VALID(client));
2970 REQUIRE(addrp != NULL);
2972 *addrp = &client->peeraddr;
2973 return (ISC_R_SUCCESS);