2 * Copyright (C) 2004-2016 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.
20 #include <isc/formatcheck.h>
21 #include <isc/mutex.h>
23 #include <isc/platform.h>
24 #include <isc/print.h>
25 #include <isc/queue.h>
26 #include <isc/random.h>
27 #include <isc/stats.h>
28 #include <isc/stdio.h>
29 #include <isc/string.h>
31 #include <isc/timer.h>
35 #include <dns/dispatch.h>
36 #include <dns/events.h>
37 #include <dns/message.h>
39 #include <dns/rcode.h>
40 #include <dns/rdata.h>
41 #include <dns/rdataclass.h>
42 #include <dns/rdatalist.h>
43 #include <dns/rdataset.h>
44 #include <dns/resolver.h>
45 #include <dns/stats.h>
50 #include <named/interfacemgr.h>
51 #include <named/log.h>
52 #include <named/notify.h>
54 #include <named/server.h>
55 #include <named/update.h>
66 * All client state changes, other than that from idle to listening, occur
67 * as a result of events. This guarantees serialization and avoids the
70 * If a routine is ever created that allows someone other than the client's
71 * task to change the client, then the client will have to be locked.
74 #define NS_CLIENT_TRACE
75 #ifdef NS_CLIENT_TRACE
76 #define CTRACE(m) ns_client_log(client, \
77 NS_LOGCATEGORY_CLIENT, \
78 NS_LOGMODULE_CLIENT, \
81 #define MTRACE(m) isc_log_write(ns_g_lctx, \
82 NS_LOGCATEGORY_GENERAL, \
83 NS_LOGMODULE_CLIENT, \
85 "clientmgr @%p: %s", manager, (m))
87 #define CTRACE(m) ((void)(m))
88 #define MTRACE(m) ((void)(m))
91 #define TCP_CLIENT(c) (((c)->attributes & NS_CLIENTATTR_TCP) != 0)
93 #define TCP_BUFFER_SIZE (65535 + 2)
94 #define SEND_BUFFER_SIZE 4096
95 #define RECV_BUFFER_SIZE 4096
97 #ifdef ISC_PLATFORM_USETHREADS
100 * Number of 'mctx pools' for clients. (Should this be configurable?)
101 * When enabling threads, we use a pool of memory contexts shared by
102 * client objects, since concurrent access to a shared context would cause
103 * heavy contentions. The above constant is expected to be enough for
104 * completely avoiding contentions among threads for an authoritative-only
110 * If named with built without thread, simply share manager's context. Using
111 * 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) {
332 if (TCP_CLIENT(client))
333 sock = client->tcpsocket;
335 sock = client->udpsocket;
336 isc_socket_cancel(sock, 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;
832 isc_netaddr_t netaddr;
834 unsigned int sockflags = ISC_SOCKFLAG_IMMEDIATE;
836 if (TCP_CLIENT(client)) {
837 sock = client->tcpsocket;
840 sock = 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(sock, &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;
956 if (preferred_glue == 0) {
957 if (isc_sockaddr_pf(&client->peeraddr) == AF_INET)
958 preferred_glue = DNS_MESSAGERENDER_PREFER_A;
960 preferred_glue = DNS_MESSAGERENDER_PREFER_AAAA;
963 #ifdef ALLOW_FILTER_AAAA_ON_V4
965 * filter-aaaa-on-v4 yes or break-dnssec option to suppress
967 * We already know that request came via IPv4,
968 * that we have both AAAA and A records,
969 * and that we either have no signatures that the client wants
970 * or we are supposed to break DNSSEC.
972 * Override preferred glue if necessary.
974 if ((client->attributes & NS_CLIENTATTR_FILTER_AAAA) != 0) {
975 render_opts |= DNS_MESSAGERENDER_FILTER_AAAA;
976 if (preferred_glue == DNS_MESSAGERENDER_PREFER_AAAA)
977 preferred_glue = DNS_MESSAGERENDER_PREFER_A;
982 * XXXRTH The following doesn't deal with TCP buffer resizing.
984 result = client_allocsendbuf(client, &buffer, &tcpbuffer, 0,
986 if (result != ISC_R_SUCCESS)
989 result = dns_compress_init(&cctx, -1, client->mctx);
990 if (result != ISC_R_SUCCESS)
992 if (client->peeraddr_valid && client->view != NULL) {
993 isc_netaddr_t netaddr;
994 dns_name_t *name = NULL;
996 isc_netaddr_fromsockaddr(&netaddr, &client->peeraddr);
997 if (client->message->tsigkey != NULL)
998 name = &client->message->tsigkey->name;
999 if (client->view->nocasecompress == NULL ||
1000 !allowed(&netaddr, name, client->view->nocasecompress))
1002 dns_compress_setsensitive(&cctx, ISC_TRUE);
1005 cleanup_cctx = ISC_TRUE;
1007 result = dns_message_renderbegin(client->message, &cctx, &buffer);
1008 if (result != ISC_R_SUCCESS)
1011 if (client->opt != NULL) {
1012 result = dns_message_setopt(client->message, client->opt);
1013 opt_included = ISC_TRUE;
1015 if (result != ISC_R_SUCCESS)
1018 result = dns_message_rendersection(client->message,
1019 DNS_SECTION_QUESTION, 0);
1020 if (result == ISC_R_NOSPACE) {
1021 client->message->flags |= DNS_MESSAGEFLAG_TC;
1024 if (result != ISC_R_SUCCESS)
1028 * Stop after the question if TC was set for rate limiting.
1030 if ((client->message->flags & DNS_MESSAGEFLAG_TC) != 0)
1032 #endif /* USE_RRL */
1033 result = dns_message_rendersection(client->message,
1035 DNS_MESSAGERENDER_PARTIAL |
1037 if (result == ISC_R_NOSPACE) {
1038 client->message->flags |= DNS_MESSAGEFLAG_TC;
1041 if (result != ISC_R_SUCCESS)
1043 result = dns_message_rendersection(client->message,
1044 DNS_SECTION_AUTHORITY,
1045 DNS_MESSAGERENDER_PARTIAL |
1047 if (result == ISC_R_NOSPACE) {
1048 client->message->flags |= DNS_MESSAGEFLAG_TC;
1051 if (result != ISC_R_SUCCESS)
1053 result = dns_message_rendersection(client->message,
1054 DNS_SECTION_ADDITIONAL,
1055 preferred_glue | render_opts);
1056 if (result != ISC_R_SUCCESS && result != ISC_R_NOSPACE)
1059 result = dns_message_renderend(client->message);
1061 if (result != ISC_R_SUCCESS)
1065 dns_compress_invalidate(&cctx);
1066 cleanup_cctx = ISC_FALSE;
1069 if (TCP_CLIENT(client)) {
1070 isc_buffer_usedregion(&buffer, &r);
1071 isc_buffer_putuint16(&tcpbuffer, (isc_uint16_t) r.length);
1072 isc_buffer_add(&tcpbuffer, r.length);
1073 result = client_sendpkg(client, &tcpbuffer);
1075 result = client_sendpkg(client, &buffer);
1077 /* update statistics (XXXJT: is it okay to access message->xxxkey?) */
1078 isc_stats_increment(ns_g_server->nsstats, dns_nsstatscounter_response);
1080 isc_stats_increment(ns_g_server->nsstats,
1081 dns_nsstatscounter_edns0out);
1083 if (client->message->tsigkey != NULL) {
1084 isc_stats_increment(ns_g_server->nsstats,
1085 dns_nsstatscounter_tsigout);
1087 if (client->message->sig0key != NULL) {
1088 isc_stats_increment(ns_g_server->nsstats,
1089 dns_nsstatscounter_sig0out);
1091 if ((client->message->flags & DNS_MESSAGEFLAG_TC) != 0)
1092 isc_stats_increment(ns_g_server->nsstats,
1093 dns_nsstatscounter_truncatedresp);
1095 if (result == ISC_R_SUCCESS)
1099 if (client->tcpbuf != NULL) {
1100 isc_mem_put(client->mctx, client->tcpbuf, TCP_BUFFER_SIZE);
1101 client->tcpbuf = NULL;
1105 dns_compress_invalidate(&cctx);
1107 ns_client_next(client, result);
1110 #if NS_CLIENT_DROPPORT
1111 #define DROPPORT_NO 0
1112 #define DROPPORT_REQUEST 1
1113 #define DROPPORT_RESPONSE 2
1115 * ns_client_dropport determines if certain requests / responses
1116 * should be dropped based on the port number.
1119 * \li 0: Don't drop.
1120 * \li 1: Drop request.
1121 * \li 2: Drop (error) response.
1124 ns_client_dropport(in_port_t port) {
1127 case 13: /* daytime */
1128 case 19: /* chargen */
1130 return (DROPPORT_REQUEST);
1131 case 464: /* kpasswd */
1132 return (DROPPORT_RESPONSE);
1134 return (DROPPORT_NO);
1139 ns_client_error(ns_client_t *client, isc_result_t result) {
1141 dns_message_t *message;
1143 REQUIRE(NS_CLIENT_VALID(client));
1147 message = client->message;
1148 rcode = dns_result_torcode(result);
1150 #if NS_CLIENT_DROPPORT
1152 * Don't send FORMERR to ports on the drop port list.
1154 if (rcode == dns_rcode_formerr &&
1155 ns_client_dropport(isc_sockaddr_getport(&client->peeraddr)) !=
1160 isc_buffer_init(&b, buf, sizeof(buf) - 1);
1161 if (dns_rcode_totext(rcode, &b) != ISC_R_SUCCESS)
1162 isc_buffer_putstr(&b, "UNKNOWN RCODE");
1163 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
1164 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(10),
1165 "dropped error (%.*s) response: suspicious port",
1166 (int)isc_buffer_usedlength(&b), buf);
1167 ns_client_next(client, ISC_R_SUCCESS);
1174 * Try to rate limit error responses.
1176 if (client->view != NULL && client->view->rrl != NULL) {
1177 isc_boolean_t wouldlog;
1178 char log_buf[DNS_RRL_LOG_BUF_LEN];
1179 dns_rrl_result_t rrl_result;
1182 INSIST(rcode != dns_rcode_noerror &&
1183 rcode != dns_rcode_nxdomain);
1184 if (ns_g_server->log_queries)
1185 loglevel = DNS_RRL_LOG_DROP;
1187 loglevel = ISC_LOG_DEBUG(1);
1188 wouldlog = isc_log_wouldlog(ns_g_lctx, loglevel);
1189 rrl_result = dns_rrl(client->view, &client->peeraddr,
1191 dns_rdataclass_in, dns_rdatatype_none,
1192 NULL, result, client->now,
1193 wouldlog, log_buf, sizeof(log_buf));
1194 if (rrl_result != DNS_RRL_RESULT_OK) {
1196 * Log dropped errors in the query category
1197 * so that they are not lost in silence.
1198 * Starts of rate-limited bursts are logged in
1199 * NS_LOGCATEGORY_RRL.
1202 ns_client_log(client,
1203 NS_LOGCATEGORY_QUERY_ERRORS,
1204 NS_LOGMODULE_CLIENT,
1209 * Some error responses cannot be 'slipped',
1210 * so don't try to slip any error responses.
1212 if (!client->view->rrl->log_only) {
1213 isc_stats_increment(ns_g_server->nsstats,
1214 dns_nsstatscounter_ratedropped);
1215 isc_stats_increment(ns_g_server->nsstats,
1216 dns_nsstatscounter_dropped);
1217 ns_client_next(client, DNS_R_DROP);
1222 #endif /* USE_RRL */
1225 * Message may be an in-progress reply that we had trouble
1226 * with, in which case QR will be set. We need to clear QR before
1227 * calling dns_message_reply() to avoid triggering an assertion.
1229 message->flags &= ~DNS_MESSAGEFLAG_QR;
1231 * AA and AD shouldn't be set.
1233 message->flags &= ~(DNS_MESSAGEFLAG_AA | DNS_MESSAGEFLAG_AD);
1234 result = dns_message_reply(message, ISC_TRUE);
1235 if (result != ISC_R_SUCCESS) {
1237 * It could be that we've got a query with a good header,
1238 * but a bad question section, so we try again with
1239 * want_question_section set to ISC_FALSE.
1241 result = dns_message_reply(message, ISC_FALSE);
1242 if (result != ISC_R_SUCCESS) {
1243 ns_client_next(client, result);
1247 message->rcode = rcode;
1250 * FORMERR loop avoidance: If we sent a FORMERR message
1251 * with the same ID to the same client less than two
1252 * seconds ago, assume that we are in an infinite error
1253 * packet dialog with a server for some protocol whose
1254 * error responses look enough like DNS queries to
1255 * elicit a FORMERR response. Drop a packet to break
1258 if (rcode == dns_rcode_formerr) {
1259 if (isc_sockaddr_equal(&client->peeraddr,
1260 &client->formerrcache.addr) &&
1261 message->id == client->formerrcache.id &&
1262 client->requesttime - client->formerrcache.time < 2) {
1264 ns_client_log(client, NS_LOGCATEGORY_CLIENT,
1265 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(1),
1266 "possible error packet loop, "
1268 ns_client_next(client, result);
1271 client->formerrcache.addr = client->peeraddr;
1272 client->formerrcache.time = client->requesttime;
1273 client->formerrcache.id = message->id;
1275 ns_client_send(client);
1278 static inline isc_result_t
1279 client_addopt(ns_client_t *client) {
1280 char nsid[BUFSIZ], *nsidp;
1281 isc_result_t result;
1283 dns_resolver_t *resolver;
1284 isc_uint16_t udpsize;
1285 dns_ednsopt_t ednsopts[DNS_EDNSOPTIONS];
1289 REQUIRE(client->opt == NULL); /* XXXRTH free old. */
1291 view = client->view;
1292 resolver = (view != NULL) ? view->resolver : NULL;
1293 if (resolver != NULL)
1294 udpsize = dns_resolver_getudpsize(resolver);
1296 udpsize = ns_g_udpsize;
1298 flags = client->extflags & DNS_MESSAGEEXTFLAG_REPLYPRESERVE;
1300 /* Set EDNS options if applicable */
1301 if ((client->attributes & NS_CLIENTATTR_WANTNSID) != 0 &&
1302 (ns_g_server->server_id != NULL ||
1303 ns_g_server->server_usehostname)) {
1304 if (ns_g_server->server_usehostname) {
1305 result = ns_os_gethostname(nsid, sizeof(nsid));
1306 if (result != ISC_R_SUCCESS) {
1311 nsidp = ns_g_server->server_id;
1313 INSIST(count < DNS_EDNSOPTIONS);
1314 ednsopts[count].code = DNS_OPT_NSID;
1315 ednsopts[count].length = strlen(nsidp);
1316 ednsopts[count].value = (unsigned char *)nsidp;
1320 result = dns_message_buildopt(client->message, &client->opt, 0,
1321 udpsize, flags, ednsopts, count);
1325 static inline isc_boolean_t
1326 allowed(isc_netaddr_t *addr, dns_name_t *signer, dns_acl_t *acl) {
1328 isc_result_t result;
1332 result = dns_acl_match(addr, signer, acl, &ns_g_server->aclenv,
1334 if (result == ISC_R_SUCCESS && match > 0)
1340 * Callback to see if a non-recursive query coming from 'srcaddr' to
1341 * 'destaddr', with optional key 'mykey' for class 'rdclass' would be
1342 * delivered to 'myview'.
1344 * We run this unlocked as both the view list and the interface list
1345 * are updated when the appropriate task has exclusivity.
1348 ns_client_isself(dns_view_t *myview, dns_tsigkey_t *mykey,
1349 isc_sockaddr_t *srcaddr, isc_sockaddr_t *dstaddr,
1350 dns_rdataclass_t rdclass, void *arg)
1353 dns_tsigkey_t *key = NULL;
1354 dns_name_t *tsig = NULL;
1355 isc_netaddr_t netsrc;
1356 isc_netaddr_t netdst;
1361 * ns_g_server->interfacemgr is task exclusive locked.
1363 if (ns_g_server->interfacemgr == NULL)
1366 if (!ns_interfacemgr_listeningon(ns_g_server->interfacemgr, dstaddr))
1369 isc_netaddr_fromsockaddr(&netsrc, srcaddr);
1370 isc_netaddr_fromsockaddr(&netdst, dstaddr);
1372 for (view = ISC_LIST_HEAD(ns_g_server->viewlist);
1374 view = ISC_LIST_NEXT(view, link)) {
1376 if (view->matchrecursiveonly)
1379 if (rdclass != view->rdclass)
1382 if (mykey != NULL) {
1383 isc_boolean_t match;
1384 isc_result_t result;
1386 result = dns_view_gettsig(view, &mykey->name, &key);
1387 if (result != ISC_R_SUCCESS)
1389 match = dst_key_compare(mykey->key, key->key);
1390 dns_tsigkey_detach(&key);
1393 tsig = dns_tsigkey_identity(mykey);
1396 if (allowed(&netsrc, tsig, view->matchclients) &&
1397 allowed(&netdst, tsig, view->matchdestinations))
1400 return (ISC_TF(view == myview));
1404 process_opt(ns_client_t *client, dns_rdataset_t *opt) {
1406 isc_buffer_t optbuf;
1407 isc_result_t result;
1408 isc_uint16_t optcode;
1409 isc_uint16_t optlen;
1412 * Set the client's UDP buffer size.
1414 client->udpsize = opt->rdclass;
1417 * If the requested UDP buffer size is less than 512,
1418 * ignore it and use 512.
1420 if (client->udpsize < 512)
1421 client->udpsize = 512;
1424 * Get the flags out of the OPT record.
1426 client->extflags = (isc_uint16_t)(opt->ttl & 0xFFFF);
1429 * Do we understand this version of EDNS?
1431 * XXXRTH need library support for this!
1433 client->ednsversion = (opt->ttl & 0x00FF0000) >> 16;
1434 if (client->ednsversion > 0) {
1435 isc_stats_increment(ns_g_server->nsstats,
1436 dns_nsstatscounter_badednsver);
1437 result = client_addopt(client);
1438 if (result == ISC_R_SUCCESS)
1439 result = DNS_R_BADVERS;
1440 ns_client_error(client, result);
1444 /* Check for NSID request */
1445 result = dns_rdataset_first(opt);
1446 if (result == ISC_R_SUCCESS) {
1447 dns_rdata_init(&rdata);
1448 dns_rdataset_current(opt, &rdata);
1449 isc_buffer_init(&optbuf, rdata.data, rdata.length);
1450 isc_buffer_add(&optbuf, rdata.length);
1451 while (isc_buffer_remaininglength(&optbuf) >= 4) {
1452 optcode = isc_buffer_getuint16(&optbuf);
1453 optlen = isc_buffer_getuint16(&optbuf);
1456 client->attributes |= NS_CLIENTATTR_WANTNSID;
1457 isc_buffer_forward(&optbuf, optlen);
1460 isc_buffer_forward(&optbuf, optlen);
1466 isc_stats_increment(ns_g_server->nsstats, dns_nsstatscounter_edns0in);
1469 * Create an OPT for our reply.
1471 result = client_addopt(client);
1472 if (result != ISC_R_SUCCESS) {
1473 ns_client_error(client, result);
1481 * Handle an incoming request event from the socket (UDP case)
1482 * or tcpmsg (TCP case).
1485 client_request(isc_task_t *task, isc_event_t *event) {
1486 ns_client_t *client;
1487 isc_socketevent_t *sevent;
1488 isc_result_t result;
1489 isc_result_t sigresult = ISC_R_SUCCESS;
1490 isc_buffer_t *buffer;
1491 isc_buffer_t tbuffer;
1493 dns_rdataset_t *opt;
1494 dns_name_t *signame;
1495 isc_boolean_t ra; /* Recursion available. */
1496 isc_netaddr_t netaddr;
1500 isc_boolean_t notimp;
1502 REQUIRE(event != NULL);
1503 client = event->ev_arg;
1504 REQUIRE(NS_CLIENT_VALID(client));
1505 REQUIRE(task == client->task);
1507 INSIST(client->recursionquota == NULL);
1509 INSIST(client->state == (TCP_CLIENT(client) ?
1510 NS_CLIENTSTATE_READING :
1511 NS_CLIENTSTATE_READY));
1513 ns_client_requests++;
1515 if (event->ev_type == ISC_SOCKEVENT_RECVDONE) {
1516 INSIST(!TCP_CLIENT(client));
1517 sevent = (isc_socketevent_t *)event;
1518 REQUIRE(sevent == client->recvevent);
1519 isc_buffer_init(&tbuffer, sevent->region.base, sevent->n);
1520 isc_buffer_add(&tbuffer, sevent->n);
1522 result = sevent->result;
1523 if (result == ISC_R_SUCCESS) {
1524 client->peeraddr = sevent->address;
1525 client->peeraddr_valid = ISC_TRUE;
1527 if ((sevent->attributes & ISC_SOCKEVENTATTR_PKTINFO) != 0) {
1528 client->attributes |= NS_CLIENTATTR_PKTINFO;
1529 client->pktinfo = sevent->pktinfo;
1531 if ((sevent->attributes & ISC_SOCKEVENTATTR_MULTICAST) != 0)
1532 client->attributes |= NS_CLIENTATTR_MULTICAST;
1535 INSIST(TCP_CLIENT(client));
1536 REQUIRE(event->ev_type == DNS_EVENT_TCPMSG);
1537 REQUIRE(event->ev_sender == &client->tcpmsg);
1538 buffer = &client->tcpmsg.buffer;
1539 result = client->tcpmsg.result;
1540 INSIST(client->nreads == 1);
1542 * client->peeraddr was set when the connection was accepted.
1547 if (exit_check(client))
1549 client->state = client->newstate = NS_CLIENTSTATE_WORKING;
1551 isc_task_getcurrenttime(task, &client->requesttime);
1552 client->now = client->requesttime;
1554 if (result != ISC_R_SUCCESS) {
1555 if (TCP_CLIENT(client)) {
1556 ns_client_next(client, result);
1558 if (result != ISC_R_CANCELED)
1559 isc_log_write(ns_g_lctx, NS_LOGCATEGORY_CLIENT,
1560 NS_LOGMODULE_CLIENT,
1562 "UDP client handler shutting "
1563 "down due to fatal receive "
1565 isc_result_totext(result));
1566 isc_task_shutdown(client->task);
1571 isc_netaddr_fromsockaddr(&netaddr, &client->peeraddr);
1573 #if NS_CLIENT_DROPPORT
1574 if (ns_client_dropport(isc_sockaddr_getport(&client->peeraddr)) ==
1576 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
1577 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(10),
1578 "dropped request: suspicious port");
1579 ns_client_next(client, ISC_R_SUCCESS);
1584 ns_client_log(client, NS_LOGCATEGORY_CLIENT,
1585 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(3),
1587 TCP_CLIENT(client) ? "TCP" : "UDP");
1590 * Check the blackhole ACL for UDP only, since TCP is done in
1593 if (!TCP_CLIENT(client)) {
1595 if (ns_g_server->blackholeacl != NULL &&
1596 dns_acl_match(&netaddr, NULL, ns_g_server->blackholeacl,
1597 &ns_g_server->aclenv,
1598 &match, NULL) == ISC_R_SUCCESS &&
1601 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
1602 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(10),
1603 "blackholed UDP datagram");
1604 ns_client_next(client, ISC_R_SUCCESS);
1610 * Silently drop multicast requests for the present.
1611 * XXXMPA revisit this as mDNS spec was published.
1613 if ((client->attributes & NS_CLIENTATTR_MULTICAST) != 0) {
1614 ns_client_log(client, NS_LOGCATEGORY_CLIENT,
1615 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(2),
1616 "dropping multicast request");
1617 ns_client_next(client, DNS_R_REFUSED);
1621 result = dns_message_peekheader(buffer, &id, &flags);
1622 if (result != ISC_R_SUCCESS) {
1624 * There isn't enough header to determine whether
1625 * this was a request or a response. Drop it.
1627 ns_client_next(client, result);
1632 * The client object handles requests, not responses.
1633 * If this is a UDP response, forward it to the dispatcher.
1634 * If it's a TCP response, discard it here.
1636 if ((flags & DNS_MESSAGEFLAG_QR) != 0) {
1637 if (TCP_CLIENT(client)) {
1638 CTRACE("unexpected response");
1639 ns_client_next(client, DNS_R_FORMERR);
1642 dns_dispatch_importrecv(client->dispatch, event);
1643 ns_client_next(client, ISC_R_SUCCESS);
1649 * Update some statistics counters. Don't count responses.
1651 if (isc_sockaddr_pf(&client->peeraddr) == PF_INET) {
1652 isc_stats_increment(ns_g_server->nsstats,
1653 dns_nsstatscounter_requestv4);
1655 isc_stats_increment(ns_g_server->nsstats,
1656 dns_nsstatscounter_requestv6);
1658 if (TCP_CLIENT(client))
1659 isc_stats_increment(ns_g_server->nsstats,
1660 dns_nsstatscounter_requesttcp);
1663 * It's a request. Parse it.
1665 result = dns_message_parse(client->message, buffer, 0);
1666 if (result != ISC_R_SUCCESS) {
1668 * Parsing the request failed. Send a response
1669 * (typically FORMERR or SERVFAIL).
1671 if (result == DNS_R_OPTERR)
1672 (void)client_addopt(client);
1674 ns_client_log(client, NS_LOGCATEGORY_CLIENT,
1675 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(1),
1676 "message parsing failed: %s",
1677 isc_result_totext(result));
1678 ns_client_error(client, result);
1682 dns_opcodestats_increment(ns_g_server->opcodestats,
1683 client->message->opcode);
1684 switch (client->message->opcode) {
1685 case dns_opcode_query:
1686 case dns_opcode_update:
1687 case dns_opcode_notify:
1690 case dns_opcode_iquery:
1696 client->message->rcode = dns_rcode_noerror;
1698 /* RFC1123 section 6.1.3.2 */
1699 if ((client->attributes & NS_CLIENTATTR_MULTICAST) != 0)
1700 client->message->flags &= ~DNS_MESSAGEFLAG_RD;
1708 opt = dns_message_getopt(client->message);
1711 * Are we dropping all EDNS queries?
1713 if (ns_g_dropedns) {
1714 ns_client_next(client, ISC_R_SUCCESS);
1717 result = process_opt(client, opt);
1718 if (result != ISC_R_SUCCESS)
1722 if (client->message->rdclass == 0) {
1723 ns_client_log(client, NS_LOGCATEGORY_CLIENT,
1724 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(1),
1725 "message class could not be determined");
1726 ns_client_dumpmessage(client,
1727 "message class could not be determined");
1728 ns_client_error(client, notimp ? DNS_R_NOTIMP : DNS_R_FORMERR);
1733 * Determine the destination address. If the receiving interface is
1734 * bound to a specific address, we simply use it regardless of the
1735 * address family. All IPv4 queries should fall into this case.
1736 * Otherwise, if this is a TCP query, get the address from the
1737 * receiving socket (this needs a system call and can be heavy).
1738 * For IPv6 UDP queries, we get this from the pktinfo structure (if
1740 * If all the attempts fail (this can happen due to memory shortage,
1741 * etc), we regard this as an error for safety.
1743 if ((client->interface->flags & NS_INTERFACEFLAG_ANYADDR) == 0)
1744 isc_netaddr_fromsockaddr(&client->destaddr,
1745 &client->interface->addr);
1747 isc_sockaddr_t sockaddr;
1748 result = ISC_R_FAILURE;
1750 if (TCP_CLIENT(client))
1751 result = isc_socket_getsockname(client->tcpsocket,
1753 if (result == ISC_R_SUCCESS)
1754 isc_netaddr_fromsockaddr(&client->destaddr, &sockaddr);
1755 if (result != ISC_R_SUCCESS &&
1756 client->interface->addr.type.sa.sa_family == AF_INET6 &&
1757 (client->attributes & NS_CLIENTATTR_PKTINFO) != 0) {
1759 * XXXJT technically, we should convert the receiving
1760 * interface ID to a proper scope zone ID. However,
1761 * due to the fact there is no standard API for this,
1762 * we only handle link-local addresses and use the
1763 * interface index as link ID. Despite the assumption,
1764 * it should cover most typical cases.
1766 isc_netaddr_fromin6(&client->destaddr,
1767 &client->pktinfo.ipi6_addr);
1768 if (IN6_IS_ADDR_LINKLOCAL(&client->pktinfo.ipi6_addr))
1769 isc_netaddr_setzone(&client->destaddr,
1770 client->pktinfo.ipi6_ifindex);
1771 result = ISC_R_SUCCESS;
1773 if (result != ISC_R_SUCCESS) {
1774 UNEXPECTED_ERROR(__FILE__, __LINE__,
1775 "failed to get request's "
1777 isc_result_totext(result));
1778 ns_client_next(client, ISC_R_SUCCESS);
1784 * Find a view that matches the client's source address.
1786 for (view = ISC_LIST_HEAD(ns_g_server->viewlist);
1788 view = ISC_LIST_NEXT(view, link)) {
1789 if (client->message->rdclass == view->rdclass ||
1790 client->message->rdclass == dns_rdataclass_any)
1792 dns_name_t *tsig = NULL;
1794 sigresult = dns_message_rechecksig(client->message,
1796 if (sigresult == ISC_R_SUCCESS)
1797 tsig = dns_tsigkey_identity(client->message->tsigkey);
1799 if (allowed(&netaddr, tsig, view->matchclients) &&
1800 allowed(&client->destaddr, tsig,
1801 view->matchdestinations) &&
1802 !((client->message->flags & DNS_MESSAGEFLAG_RD)
1803 == 0 && view->matchrecursiveonly))
1805 dns_view_attach(view, &client->view);
1812 char classname[DNS_RDATACLASS_FORMATSIZE];
1815 * Do a dummy TSIG verification attempt so that the
1816 * response will have a TSIG if the query did, as
1817 * required by RFC2845.
1822 dns_message_resetsig(client->message);
1824 r = dns_message_getrawmessage(client->message);
1825 isc_buffer_init(&b, r->base, r->length);
1826 isc_buffer_add(&b, r->length);
1827 (void)dns_tsig_verify(&b, client->message, NULL, NULL);
1829 dns_rdataclass_format(client->message->rdclass, classname,
1831 ns_client_log(client, NS_LOGCATEGORY_CLIENT,
1832 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(1),
1833 "no matching view in class '%s'", classname);
1834 ns_client_dumpmessage(client, "no matching view in class");
1835 ns_client_error(client, notimp ? DNS_R_NOTIMP : DNS_R_REFUSED);
1839 ns_client_log(client, NS_LOGCATEGORY_CLIENT,
1840 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(5),
1841 "using view '%s'", view->name);
1844 * Check for a signature. We log bad signatures regardless of
1845 * whether they ultimately cause the request to be rejected or
1846 * not. We do not log the lack of a signature unless we are
1849 client->signer = NULL;
1850 dns_name_init(&client->signername, NULL);
1851 result = dns_message_signer(client->message, &client->signername);
1852 if (result != ISC_R_NOTFOUND) {
1854 if (dns_message_gettsig(client->message, &signame) != NULL) {
1855 isc_stats_increment(ns_g_server->nsstats,
1856 dns_nsstatscounter_tsigin);
1858 isc_stats_increment(ns_g_server->nsstats,
1859 dns_nsstatscounter_sig0in);
1863 if (result == ISC_R_SUCCESS) {
1864 char namebuf[DNS_NAME_FORMATSIZE];
1865 dns_name_format(&client->signername, namebuf, sizeof(namebuf));
1866 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
1867 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(3),
1868 "request has valid signature: %s", namebuf);
1869 client->signer = &client->signername;
1870 } else if (result == ISC_R_NOTFOUND) {
1871 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
1872 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(3),
1873 "request is not signed");
1874 } else if (result == DNS_R_NOIDENTITY) {
1875 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
1876 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(3),
1877 "request is signed by a nonauthoritative key");
1882 isc_result_t tresult;
1884 /* There is a signature, but it is bad. */
1885 isc_stats_increment(ns_g_server->nsstats,
1886 dns_nsstatscounter_invalidsig);
1888 if (dns_message_gettsig(client->message, &signame) != NULL) {
1889 char namebuf[DNS_NAME_FORMATSIZE];
1890 char cnamebuf[DNS_NAME_FORMATSIZE];
1891 dns_name_format(signame, namebuf, sizeof(namebuf));
1892 status = client->message->tsigstatus;
1893 isc_buffer_init(&b, tsigrcode, sizeof(tsigrcode) - 1);
1894 tresult = dns_tsigrcode_totext(status, &b);
1895 INSIST(tresult == ISC_R_SUCCESS);
1896 tsigrcode[isc_buffer_usedlength(&b)] = '\0';
1897 if (client->message->tsigkey->generated) {
1898 dns_name_format(client->message->tsigkey->creator,
1899 cnamebuf, sizeof(cnamebuf));
1900 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
1901 NS_LOGMODULE_CLIENT,
1903 "request has invalid signature: "
1904 "TSIG %s (%s): %s (%s)", namebuf,
1906 isc_result_totext(result),
1909 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
1910 NS_LOGMODULE_CLIENT,
1912 "request has invalid signature: "
1913 "TSIG %s: %s (%s)", namebuf,
1914 isc_result_totext(result),
1918 status = client->message->sig0status;
1919 isc_buffer_init(&b, tsigrcode, sizeof(tsigrcode) - 1);
1920 tresult = dns_tsigrcode_totext(status, &b);
1921 INSIST(tresult == ISC_R_SUCCESS);
1922 tsigrcode[isc_buffer_usedlength(&b)] = '\0';
1923 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
1924 NS_LOGMODULE_CLIENT, ISC_LOG_ERROR,
1925 "request has invalid signature: %s (%s)",
1926 isc_result_totext(result), tsigrcode);
1929 * Accept update messages signed by unknown keys so that
1930 * update forwarding works transparently through slaves
1931 * that don't have all the same keys as the master.
1933 if (!(client->message->tsigstatus == dns_tsigerror_badkey &&
1934 client->message->opcode == dns_opcode_update)) {
1935 ns_client_error(client, sigresult);
1941 * Decide whether recursive service is available to this client.
1942 * We do this here rather than in the query code so that we can
1943 * set the RA bit correctly on all kinds of responses, not just
1944 * responses to ordinary queries. Note if you can't query the
1945 * cache there is no point in setting RA.
1948 if (client->view->resolver != NULL &&
1949 client->view->recursion == ISC_TRUE &&
1950 ns_client_checkaclsilent(client, NULL,
1951 client->view->recursionacl,
1952 ISC_TRUE) == ISC_R_SUCCESS &&
1953 ns_client_checkaclsilent(client, NULL,
1954 client->view->cacheacl,
1955 ISC_TRUE) == ISC_R_SUCCESS &&
1956 ns_client_checkaclsilent(client, &client->destaddr,
1957 client->view->recursiononacl,
1958 ISC_TRUE) == ISC_R_SUCCESS &&
1959 ns_client_checkaclsilent(client, &client->destaddr,
1960 client->view->cacheonacl,
1961 ISC_TRUE) == ISC_R_SUCCESS)
1965 client->attributes |= NS_CLIENTATTR_RA;
1967 ns_client_log(client, DNS_LOGCATEGORY_SECURITY, NS_LOGMODULE_CLIENT,
1968 ISC_LOG_DEBUG(3), ra ? "recursion available" :
1969 "recursion not available");
1972 * Adjust maximum UDP response size for this client.
1974 if (client->udpsize > 512) {
1975 dns_peer_t *peer = NULL;
1976 isc_uint16_t udpsize = view->maxudp;
1977 (void) dns_peerlist_peerbyaddr(view->peers, &netaddr, &peer);
1979 dns_peer_getmaxudp(peer, &udpsize);
1980 if (client->udpsize > udpsize)
1981 client->udpsize = udpsize;
1985 * Dispatch the request.
1987 switch (client->message->opcode) {
1988 case dns_opcode_query:
1990 ns_query_start(client);
1992 case dns_opcode_update:
1994 ns_client_settimeout(client, 60);
1995 ns_update_start(client, sigresult);
1997 case dns_opcode_notify:
1999 ns_client_settimeout(client, 60);
2000 ns_notify_start(client);
2002 case dns_opcode_iquery:
2004 ns_client_error(client, DNS_R_NOTIMP);
2007 CTRACE("unknown opcode");
2008 ns_client_error(client, DNS_R_NOTIMP);
2016 client_timeout(isc_task_t *task, isc_event_t *event) {
2017 ns_client_t *client;
2019 REQUIRE(event != NULL);
2020 REQUIRE(event->ev_type == ISC_TIMEREVENT_LIFE ||
2021 event->ev_type == ISC_TIMEREVENT_IDLE);
2022 client = event->ev_arg;
2023 REQUIRE(NS_CLIENT_VALID(client));
2024 REQUIRE(task == client->task);
2025 REQUIRE(client->timer != NULL);
2031 isc_event_free(&event);
2033 if (client->shutdown != NULL) {
2034 (client->shutdown)(client->shutdown_arg, ISC_R_TIMEDOUT);
2035 client->shutdown = NULL;
2036 client->shutdown_arg = NULL;
2039 if (client->newstate > NS_CLIENTSTATE_READY)
2040 client->newstate = NS_CLIENTSTATE_READY;
2041 (void)exit_check(client);
2045 get_clientmctx(ns_clientmgr_t *manager, isc_mem_t **mctxp) {
2046 isc_mem_t *clientmctx;
2047 isc_result_t result;
2049 unsigned int nextmctx;
2052 MTRACE("clientmctx");
2055 * Caller must be holding the manager lock.
2057 if (ns_g_clienttest) {
2058 result = isc_mem_create(0, 0, mctxp);
2059 if (result == ISC_R_SUCCESS)
2060 isc_mem_setname(*mctxp, "client", NULL);
2064 nextmctx = manager->nextmctx++;
2065 if (manager->nextmctx == NMCTXS)
2066 manager->nextmctx = 0;
2068 INSIST(nextmctx < NMCTXS);
2070 clientmctx = manager->mctxpool[nextmctx];
2071 if (clientmctx == NULL) {
2072 result = isc_mem_create(0, 0, &clientmctx);
2073 if (result != ISC_R_SUCCESS)
2075 isc_mem_setname(clientmctx, "client", NULL);
2077 manager->mctxpool[nextmctx] = clientmctx;
2080 clientmctx = manager->mctx;
2083 isc_mem_attach(clientmctx, mctxp);
2085 return (ISC_R_SUCCESS);
2089 client_create(ns_clientmgr_t *manager, ns_client_t **clientp) {
2090 ns_client_t *client;
2091 isc_result_t result;
2092 isc_mem_t *mctx = NULL;
2095 * Caller must be holding the manager lock.
2097 * Note: creating a client does not add the client to the
2098 * manager's client list or set the client's manager pointer.
2099 * The caller is responsible for that.
2102 REQUIRE(clientp != NULL && *clientp == NULL);
2104 result = get_clientmctx(manager, &mctx);
2105 if (result != ISC_R_SUCCESS)
2108 client = isc_mem_get(mctx, sizeof(*client));
2109 if (client == NULL) {
2110 isc_mem_detach(&mctx);
2111 return (ISC_R_NOMEMORY);
2113 client->mctx = mctx;
2115 client->task = NULL;
2116 result = isc_task_create(manager->taskmgr, 0, &client->task);
2117 if (result != ISC_R_SUCCESS)
2118 goto cleanup_client;
2119 isc_task_setname(client->task, "client", client);
2121 client->timer = NULL;
2122 result = isc_timer_create(manager->timermgr, isc_timertype_inactive,
2123 NULL, NULL, client->task, client_timeout,
2124 client, &client->timer);
2125 if (result != ISC_R_SUCCESS)
2127 client->timerset = ISC_FALSE;
2129 client->message = NULL;
2130 result = dns_message_create(client->mctx, DNS_MESSAGE_INTENTPARSE,
2132 if (result != ISC_R_SUCCESS)
2135 /* XXXRTH Hardwired constants */
2137 client->sendevent = (isc_socketevent_t *)
2138 isc_event_allocate(client->mctx, client,
2139 ISC_SOCKEVENT_SENDDONE,
2140 client_senddone, client,
2141 sizeof(isc_socketevent_t));
2142 if (client->sendevent == NULL) {
2143 result = ISC_R_NOMEMORY;
2144 goto cleanup_message;
2147 client->recvbuf = isc_mem_get(client->mctx, RECV_BUFFER_SIZE);
2148 if (client->recvbuf == NULL) {
2149 result = ISC_R_NOMEMORY;
2150 goto cleanup_sendevent;
2153 client->recvevent = (isc_socketevent_t *)
2154 isc_event_allocate(client->mctx, client,
2155 ISC_SOCKEVENT_RECVDONE,
2156 client_request, client,
2157 sizeof(isc_socketevent_t));
2158 if (client->recvevent == NULL) {
2159 result = ISC_R_NOMEMORY;
2160 goto cleanup_recvbuf;
2163 client->magic = NS_CLIENT_MAGIC;
2164 client->manager = NULL;
2165 client->state = NS_CLIENTSTATE_INACTIVE;
2166 client->newstate = NS_CLIENTSTATE_MAX;
2167 client->naccepts = 0;
2171 client->nupdates = 0;
2173 client->references = 0;
2174 client->attributes = 0;
2175 client->view = NULL;
2176 client->dispatch = NULL;
2177 client->udpsocket = NULL;
2178 client->tcplistener = NULL;
2179 client->tcpsocket = NULL;
2180 client->tcpmsg_valid = ISC_FALSE;
2181 client->tcpbuf = NULL;
2183 client->udpsize = 512;
2184 client->extflags = 0;
2185 client->ednsversion = -1;
2186 client->next = NULL;
2187 client->shutdown = NULL;
2188 client->shutdown_arg = NULL;
2189 client->signer = NULL;
2190 dns_name_init(&client->signername, NULL);
2191 client->mortal = ISC_FALSE;
2192 client->tcpquota = NULL;
2193 client->recursionquota = NULL;
2194 client->interface = NULL;
2195 client->peeraddr_valid = ISC_FALSE;
2196 #ifdef ALLOW_FILTER_AAAA_ON_V4
2197 client->filter_aaaa = dns_v4_aaaa_ok;
2199 client->needshutdown = ns_g_clienttest;
2201 ISC_EVENT_INIT(&client->ctlevent, sizeof(client->ctlevent), 0, NULL,
2202 NS_EVENT_CLIENTCONTROL, client_start, client, client,
2205 * Initialize FORMERR cache to sentinel value that will not match
2206 * any actual FORMERR response.
2208 isc_sockaddr_any(&client->formerrcache.addr);
2209 client->formerrcache.time = 0;
2210 client->formerrcache.id = 0;
2211 ISC_LINK_INIT(client, link);
2212 ISC_LINK_INIT(client, rlink);
2213 ISC_QLINK_INIT(client, ilink);
2216 * We call the init routines for the various kinds of client here,
2217 * after we have created an otherwise valid client, because some
2218 * of them call routines that REQUIRE(NS_CLIENT_VALID(client)).
2220 result = ns_query_init(client);
2221 if (result != ISC_R_SUCCESS)
2222 goto cleanup_recvevent;
2224 result = isc_task_onshutdown(client->task, client_shutdown, client);
2225 if (result != ISC_R_SUCCESS)
2232 return (ISC_R_SUCCESS);
2235 ns_query_free(client);
2238 isc_event_free((isc_event_t **)&client->recvevent);
2241 isc_mem_put(client->mctx, client->recvbuf, RECV_BUFFER_SIZE);
2244 isc_event_free((isc_event_t **)&client->sendevent);
2249 dns_message_destroy(&client->message);
2252 isc_timer_detach(&client->timer);
2255 isc_task_detach(&client->task);
2258 isc_mem_putanddetach(&client->mctx, client, sizeof(*client));
2264 client_read(ns_client_t *client) {
2265 isc_result_t result;
2269 result = dns_tcpmsg_readmessage(&client->tcpmsg, client->task,
2270 client_request, client);
2271 if (result != ISC_R_SUCCESS)
2275 * Set a timeout to limit the amount of time we will wait
2276 * for a request on this TCP connection.
2278 ns_client_settimeout(client, 30);
2280 client->state = client->newstate = NS_CLIENTSTATE_READING;
2281 INSIST(client->nreads == 0);
2282 INSIST(client->recursionquota == NULL);
2287 ns_client_next(client, result);
2291 client_newconn(isc_task_t *task, isc_event_t *event) {
2292 ns_client_t *client = event->ev_arg;
2293 isc_socket_newconnev_t *nevent = (isc_socket_newconnev_t *)event;
2294 isc_result_t result;
2296 REQUIRE(event->ev_type == ISC_SOCKEVENT_NEWCONN);
2297 REQUIRE(NS_CLIENT_VALID(client));
2298 REQUIRE(client->task == task);
2302 INSIST(client->state == NS_CLIENTSTATE_READY);
2304 INSIST(client->naccepts == 1);
2307 LOCK(&client->interface->lock);
2308 INSIST(client->interface->ntcpcurrent > 0);
2309 client->interface->ntcpcurrent--;
2310 UNLOCK(&client->interface->lock);
2313 * We must take ownership of the new socket before the exit
2314 * check to make sure it gets destroyed if we decide to exit.
2316 if (nevent->result == ISC_R_SUCCESS) {
2317 client->tcpsocket = nevent->newsocket;
2318 isc_socket_setname(client->tcpsocket, "client-tcp", NULL);
2319 client->state = NS_CLIENTSTATE_READING;
2320 INSIST(client->recursionquota == NULL);
2322 (void)isc_socket_getpeername(client->tcpsocket,
2324 client->peeraddr_valid = ISC_TRUE;
2325 ns_client_log(client, NS_LOGCATEGORY_CLIENT,
2326 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(3),
2327 "new TCP connection");
2330 * XXXRTH What should we do? We're trying to accept but
2331 * it didn't work. If we just give up, then TCP
2332 * service may eventually stop.
2334 * For now, we just go idle.
2336 * Going idle is probably the right thing if the
2339 ns_client_log(client, NS_LOGCATEGORY_CLIENT,
2340 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(3),
2341 "accept failed: %s",
2342 isc_result_totext(nevent->result));
2345 if (exit_check(client))
2348 if (nevent->result == ISC_R_SUCCESS) {
2350 isc_netaddr_t netaddr;
2352 isc_netaddr_fromsockaddr(&netaddr, &client->peeraddr);
2354 if (ns_g_server->blackholeacl != NULL &&
2355 dns_acl_match(&netaddr, NULL,
2356 ns_g_server->blackholeacl,
2357 &ns_g_server->aclenv,
2358 &match, NULL) == ISC_R_SUCCESS &&
2361 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
2362 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(10),
2363 "blackholed connection attempt");
2364 client->newstate = NS_CLIENTSTATE_READY;
2365 (void)exit_check(client);
2369 INSIST(client->tcpmsg_valid == ISC_FALSE);
2370 dns_tcpmsg_init(client->mctx, client->tcpsocket,
2372 client->tcpmsg_valid = ISC_TRUE;
2375 * Let a new client take our place immediately, before
2376 * we wait for a request packet. If we don't,
2377 * telnetting to port 53 (once per CPU) will
2378 * deny service to legitimate TCP clients.
2380 result = isc_quota_attach(&ns_g_server->tcpquota,
2382 if (result == ISC_R_SUCCESS)
2383 result = ns_client_replace(client);
2384 if (result != ISC_R_SUCCESS) {
2385 ns_client_log(client, NS_LOGCATEGORY_CLIENT,
2386 NS_LOGMODULE_CLIENT, ISC_LOG_WARNING,
2387 "no more TCP clients: %s",
2388 isc_result_totext(result));
2391 client_read(client);
2395 isc_event_free(&event);
2399 client_accept(ns_client_t *client) {
2400 isc_result_t result;
2404 result = isc_socket_accept(client->tcplistener, client->task,
2405 client_newconn, client);
2406 if (result != ISC_R_SUCCESS) {
2407 UNEXPECTED_ERROR(__FILE__, __LINE__,
2408 "isc_socket_accept() failed: %s",
2409 isc_result_totext(result));
2411 * XXXRTH What should we do? We're trying to accept but
2412 * it didn't work. If we just give up, then TCP
2413 * service may eventually stop.
2415 * For now, we just go idle.
2419 INSIST(client->naccepts == 0);
2421 LOCK(&client->interface->lock);
2422 client->interface->ntcpcurrent++;
2423 UNLOCK(&client->interface->lock);
2427 client_udprecv(ns_client_t *client) {
2428 isc_result_t result;
2433 r.base = client->recvbuf;
2434 r.length = RECV_BUFFER_SIZE;
2435 result = isc_socket_recv2(client->udpsocket, &r, 1,
2436 client->task, client->recvevent, 0);
2437 if (result != ISC_R_SUCCESS) {
2438 UNEXPECTED_ERROR(__FILE__, __LINE__,
2439 "isc_socket_recv2() failed: %s",
2440 isc_result_totext(result));
2442 * This cannot happen in the current implementation, since
2443 * isc_socket_recv2() cannot fail if flags == 0.
2445 * If this does fail, we just go idle.
2449 INSIST(client->nrecvs == 0);
2454 ns_client_attach(ns_client_t *source, ns_client_t **targetp) {
2455 REQUIRE(NS_CLIENT_VALID(source));
2456 REQUIRE(targetp != NULL && *targetp == NULL);
2458 source->references++;
2459 ns_client_log(source, NS_LOGCATEGORY_CLIENT,
2460 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(10),
2461 "ns_client_attach: ref = %d", source->references);
2466 ns_client_detach(ns_client_t **clientp) {
2467 ns_client_t *client = *clientp;
2469 client->references--;
2470 INSIST(client->references >= 0);
2472 ns_client_log(client, NS_LOGCATEGORY_CLIENT,
2473 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(10),
2474 "ns_client_detach: ref = %d", client->references);
2475 (void)exit_check(client);
2479 ns_client_shuttingdown(ns_client_t *client) {
2480 return (ISC_TF(client->newstate == NS_CLIENTSTATE_FREED));
2484 ns_client_replace(ns_client_t *client) {
2485 isc_result_t result;
2489 REQUIRE(client != NULL);
2490 REQUIRE(client->manager != NULL);
2492 result = get_client(client->manager, client->interface,
2493 client->dispatch, TCP_CLIENT(client));
2494 if (result != ISC_R_SUCCESS)
2498 * The responsibility for listening for new requests is hereby
2499 * transferred to the new client. Therefore, the old client
2500 * should refrain from listening for any more requests.
2502 client->mortal = ISC_TRUE;
2504 return (ISC_R_SUCCESS);
2512 clientmgr_destroy(ns_clientmgr_t *manager) {
2517 REQUIRE(ISC_LIST_EMPTY(manager->clients));
2519 MTRACE("clientmgr_destroy");
2522 for (i = 0; i < NMCTXS; i++) {
2523 if (manager->mctxpool[i] != NULL)
2524 isc_mem_detach(&manager->mctxpool[i]);
2528 ISC_QUEUE_DESTROY(manager->inactive);
2529 DESTROYLOCK(&manager->lock);
2530 DESTROYLOCK(&manager->listlock);
2531 DESTROYLOCK(&manager->reclock);
2533 isc_mem_put(manager->mctx, manager, sizeof(*manager));
2537 ns_clientmgr_create(isc_mem_t *mctx, isc_taskmgr_t *taskmgr,
2538 isc_timermgr_t *timermgr, ns_clientmgr_t **managerp)
2540 ns_clientmgr_t *manager;
2541 isc_result_t result;
2546 manager = isc_mem_get(mctx, sizeof(*manager));
2547 if (manager == NULL)
2548 return (ISC_R_NOMEMORY);
2550 result = isc_mutex_init(&manager->lock);
2551 if (result != ISC_R_SUCCESS)
2552 goto cleanup_manager;
2554 result = isc_mutex_init(&manager->listlock);
2555 if (result != ISC_R_SUCCESS)
2558 result = isc_mutex_init(&manager->reclock);
2559 if (result != ISC_R_SUCCESS)
2560 goto cleanup_listlock;
2562 manager->mctx = mctx;
2563 manager->taskmgr = taskmgr;
2564 manager->timermgr = timermgr;
2565 manager->exiting = ISC_FALSE;
2566 ISC_LIST_INIT(manager->clients);
2567 ISC_LIST_INIT(manager->recursing);
2568 ISC_QUEUE_INIT(manager->inactive, ilink);
2570 manager->nextmctx = 0;
2571 for (i = 0; i < NMCTXS; i++)
2572 manager->mctxpool[i] = NULL; /* will be created on-demand */
2574 manager->magic = MANAGER_MAGIC;
2578 *managerp = manager;
2580 return (ISC_R_SUCCESS);
2583 (void) isc_mutex_destroy(&manager->listlock);
2586 (void) isc_mutex_destroy(&manager->lock);
2589 isc_mem_put(manager->mctx, manager, sizeof(*manager));
2595 ns_clientmgr_destroy(ns_clientmgr_t **managerp) {
2596 isc_result_t result;
2597 ns_clientmgr_t *manager;
2598 ns_client_t *client;
2599 isc_boolean_t need_destroy = ISC_FALSE, unlock = ISC_FALSE;
2601 REQUIRE(managerp != NULL);
2602 manager = *managerp;
2603 REQUIRE(VALID_MANAGER(manager));
2608 * Check for success because we may already be task-exclusive
2609 * at this point. Only if we succeed at obtaining an exclusive
2610 * lock now will we need to relinquish it later.
2612 result = isc_task_beginexclusive(ns_g_server->task);
2613 if (result == ISC_R_SUCCESS)
2616 manager->exiting = ISC_TRUE;
2618 for (client = ISC_LIST_HEAD(manager->clients);
2620 client = ISC_LIST_NEXT(client, link))
2621 isc_task_shutdown(client->task);
2623 if (ISC_LIST_EMPTY(manager->clients))
2624 need_destroy = ISC_TRUE;
2627 isc_task_endexclusive(ns_g_server->task);
2630 clientmgr_destroy(manager);
2636 get_client(ns_clientmgr_t *manager, ns_interface_t *ifp,
2637 dns_dispatch_t *disp, isc_boolean_t tcp)
2639 isc_result_t result = ISC_R_SUCCESS;
2641 ns_client_t *client;
2642 MTRACE("get client");
2644 REQUIRE(manager != NULL);
2646 if (manager->exiting)
2647 return (ISC_R_SHUTTINGDOWN);
2650 * Allocate a client. First try to get a recycled one;
2651 * if that fails, make a new one.
2654 if (!ns_g_clienttest)
2655 ISC_QUEUE_POP(manager->inactive, ilink, client);
2660 MTRACE("create new");
2662 LOCK(&manager->lock);
2663 result = client_create(manager, &client);
2664 UNLOCK(&manager->lock);
2665 if (result != ISC_R_SUCCESS)
2668 LOCK(&manager->listlock);
2669 ISC_LIST_APPEND(manager->clients, client, link);
2670 UNLOCK(&manager->listlock);
2673 client->manager = manager;
2674 ns_interface_attach(ifp, &client->interface);
2675 client->state = NS_CLIENTSTATE_READY;
2676 INSIST(client->recursionquota == NULL);
2679 client->attributes |= NS_CLIENTATTR_TCP;
2680 isc_socket_attach(ifp->tcpsocket,
2681 &client->tcplistener);
2685 dns_dispatch_attach(disp, &client->dispatch);
2686 sock = dns_dispatch_getsocket(client->dispatch);
2687 isc_socket_attach(sock, &client->udpsocket);
2690 INSIST(client->nctls == 0);
2692 ev = &client->ctlevent;
2693 isc_task_send(client->task, &ev);
2695 return (ISC_R_SUCCESS);
2699 ns_clientmgr_createclients(ns_clientmgr_t *manager, unsigned int n,
2700 ns_interface_t *ifp, isc_boolean_t tcp)
2702 isc_result_t result = ISC_R_SUCCESS;
2705 REQUIRE(VALID_MANAGER(manager));
2708 MTRACE("createclients");
2710 for (disp = 0; disp < n; disp++) {
2711 result = get_client(manager, ifp, ifp->udpdispatch[disp], tcp);
2712 if (result != ISC_R_SUCCESS)
2720 ns_client_getsockaddr(ns_client_t *client) {
2721 return (&client->peeraddr);
2725 ns_client_checkaclsilent(ns_client_t *client, isc_netaddr_t *netaddr,
2726 dns_acl_t *acl, isc_boolean_t default_allow)
2728 isc_result_t result;
2729 isc_netaddr_t tmpnetaddr;
2739 if (netaddr == NULL) {
2740 isc_netaddr_fromsockaddr(&tmpnetaddr, &client->peeraddr);
2741 netaddr = &tmpnetaddr;
2744 result = dns_acl_match(netaddr, client->signer, acl,
2745 &ns_g_server->aclenv, &match, NULL);
2747 if (result != ISC_R_SUCCESS)
2748 goto deny; /* Internal error, already logged. */
2751 goto deny; /* Negative match or no match. */
2754 return (ISC_R_SUCCESS);
2757 return (DNS_R_REFUSED);
2761 ns_client_checkacl(ns_client_t *client, isc_sockaddr_t *sockaddr,
2762 const char *opname, dns_acl_t *acl,
2763 isc_boolean_t default_allow, int log_level)
2765 isc_result_t result;
2766 isc_netaddr_t netaddr;
2768 if (sockaddr != NULL)
2769 isc_netaddr_fromsockaddr(&netaddr, sockaddr);
2771 result = ns_client_checkaclsilent(client, sockaddr ? &netaddr : NULL,
2772 acl, default_allow);
2774 if (result == ISC_R_SUCCESS)
2775 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
2776 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(3),
2777 "%s approved", opname);
2779 ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
2780 NS_LOGMODULE_CLIENT,
2781 log_level, "%s denied", opname);
2786 ns_client_name(ns_client_t *client, char *peerbuf, size_t len) {
2787 if (client->peeraddr_valid)
2788 isc_sockaddr_format(&client->peeraddr, peerbuf,
2791 snprintf(peerbuf, len, "@%p", client);
2795 ns_client_logv(ns_client_t *client, isc_logcategory_t *category,
2796 isc_logmodule_t *module, int level, const char *fmt, va_list ap)
2799 char peerbuf[ISC_SOCKADDR_FORMATSIZE];
2800 char signerbuf[DNS_NAME_FORMATSIZE], qnamebuf[DNS_NAME_FORMATSIZE];
2801 const char *viewname = "";
2802 const char *sep1 = "", *sep2 = "", *sep3 = "", *sep4 = "";
2803 const char *signer = "", *qname = "";
2804 dns_name_t *q = NULL;
2806 vsnprintf(msgbuf, sizeof(msgbuf), fmt, ap);
2808 ns_client_name(client, peerbuf, sizeof(peerbuf));
2810 if (client->signer != NULL) {
2811 dns_name_format(client->signer, signerbuf, sizeof(signerbuf));
2816 q = client->query.origqname != NULL
2817 ? client->query.origqname : client->query.qname;
2819 dns_name_format(q, qnamebuf, sizeof(qnamebuf));
2825 if (client->view != NULL && strcmp(client->view->name, "_bind") != 0 &&
2826 strcmp(client->view->name, "_default") != 0) {
2828 viewname = client->view->name;
2831 isc_log_write(ns_g_lctx, category, module, level,
2832 "client %s%s%s%s%s%s%s%s: %s",
2833 peerbuf, sep1, signer, sep2, qname, sep3,
2834 sep4, viewname, msgbuf);
2838 ns_client_log(ns_client_t *client, isc_logcategory_t *category,
2839 isc_logmodule_t *module, int level, const char *fmt, ...)
2843 if (! isc_log_wouldlog(ns_g_lctx, level))
2847 ns_client_logv(client, category, module, level, fmt, ap);
2852 ns_client_aclmsg(const char *msg, dns_name_t *name, dns_rdatatype_t type,
2853 dns_rdataclass_t rdclass, char *buf, size_t len)
2855 char namebuf[DNS_NAME_FORMATSIZE];
2856 char typebuf[DNS_RDATATYPE_FORMATSIZE];
2857 char classbuf[DNS_RDATACLASS_FORMATSIZE];
2859 dns_name_format(name, namebuf, sizeof(namebuf));
2860 dns_rdatatype_format(type, typebuf, sizeof(typebuf));
2861 dns_rdataclass_format(rdclass, classbuf, sizeof(classbuf));
2862 (void)snprintf(buf, len, "%s '%s/%s/%s'", msg, namebuf, typebuf,
2867 ns_client_dumpmessage(ns_client_t *client, const char *reason) {
2868 isc_buffer_t buffer;
2871 isc_result_t result;
2873 if (!isc_log_wouldlog(ns_g_lctx, ISC_LOG_DEBUG(1)))
2877 * Note that these are multiline debug messages. We want a newline
2878 * to appear in the log after each message.
2882 buf = isc_mem_get(client->mctx, len);
2885 isc_buffer_init(&buffer, buf, len);
2886 result = dns_message_totext(client->message,
2887 &dns_master_style_debug,
2889 if (result == ISC_R_NOSPACE) {
2890 isc_mem_put(client->mctx, buf, len);
2892 } else if (result == ISC_R_SUCCESS)
2893 ns_client_log(client, NS_LOGCATEGORY_UNMATCHED,
2894 NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(1),
2896 (int)isc_buffer_usedlength(&buffer),
2898 } while (result == ISC_R_NOSPACE);
2901 isc_mem_put(client->mctx, buf, len);
2905 ns_client_dumprecursing(FILE *f, ns_clientmgr_t *manager) {
2906 ns_client_t *client;
2907 char namebuf[DNS_NAME_FORMATSIZE];
2908 char original[DNS_NAME_FORMATSIZE];
2909 char peerbuf[ISC_SOCKADDR_FORMATSIZE];
2910 char typebuf[DNS_RDATATYPE_FORMATSIZE];
2911 char classbuf[DNS_RDATACLASS_FORMATSIZE];
2914 const char *origfor;
2915 dns_rdataset_t *rdataset;
2917 REQUIRE(VALID_MANAGER(manager));
2919 LOCK(&manager->reclock);
2920 client = ISC_LIST_HEAD(manager->recursing);
2921 while (client != NULL) {
2922 INSIST(client->state == NS_CLIENTSTATE_RECURSING);
2924 ns_client_name(client, peerbuf, sizeof(peerbuf));
2925 if (client->view != NULL &&
2926 strcmp(client->view->name, "_bind") != 0 &&
2927 strcmp(client->view->name, "_default") != 0) {
2928 name = client->view->name;
2935 LOCK(&client->query.fetchlock);
2936 INSIST(client->query.qname != NULL);
2937 dns_name_format(client->query.qname, namebuf, sizeof(namebuf));
2938 if (client->query.qname != client->query.origqname &&
2939 client->query.origqname != NULL) {
2941 dns_name_format(client->query.origqname, original,
2947 rdataset = ISC_LIST_HEAD(client->query.qname->list);
2948 if (rdataset == NULL && client->query.origqname != NULL)
2949 rdataset = ISC_LIST_HEAD(client->query.origqname->list);
2950 if (rdataset != NULL) {
2951 dns_rdatatype_format(rdataset->type, typebuf,
2953 dns_rdataclass_format(rdataset->rdclass, classbuf,
2956 strcpy(typebuf, "-");
2957 strcpy(classbuf, "-");
2959 UNLOCK(&client->query.fetchlock);
2960 fprintf(f, "; client %s%s%s: id %u '%s/%s/%s'%s%s "
2961 "requesttime %d\n", peerbuf, sep, name,
2962 client->message->id, namebuf, typebuf, classbuf,
2963 origfor, original, client->requesttime);
2964 client = ISC_LIST_NEXT(client, rlink);
2966 UNLOCK(&manager->reclock);
2970 ns_client_qnamereplace(ns_client_t *client, dns_name_t *name) {
2971 LOCK(&client->query.fetchlock);
2972 if (client->query.restarts > 0) {
2974 * client->query.qname was dynamically allocated.
2976 dns_message_puttempname(client->message,
2977 &client->query.qname);
2979 client->query.qname = name;
2980 UNLOCK(&client->query.fetchlock);
2984 ns_client_sourceip(dns_clientinfo_t *ci, isc_sockaddr_t **addrp) {
2985 ns_client_t *client = (ns_client_t *) ci->data;
2987 REQUIRE(NS_CLIENT_VALID(client));
2988 REQUIRE(addrp != NULL);
2990 *addrp = &client->peeraddr;
2991 return (ISC_R_SUCCESS);