2 * Copyright (C) 2004, 2006 Internet Systems Consortium, Inc. ("ISC")
3 * Copyright (C) 1999-2003 Internet Software Consortium.
5 * Permission to use, copy, modify, and 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.
18 /* $Id: dispatch.c,v 1.101.2.6.2.13 2006/07/19 00:44:04 marka Exp $ */
24 #include <isc/entropy.h>
27 #include <isc/mutex.h>
28 #include <isc/print.h>
29 #include <isc/string.h>
34 #include <dns/dispatch.h>
35 #include <dns/events.h>
37 #include <dns/message.h>
38 #include <dns/portlist.h>
39 #include <dns/tcpmsg.h>
40 #include <dns/types.h>
42 typedef ISC_LIST(dns_dispentry_t) dns_displist_t;
44 typedef struct dns_qid {
46 unsigned int qid_nbuckets; /* hash table size */
47 unsigned int qid_increment; /* id increment on collision */
49 isc_lfsr_t qid_lfsr1; /* state generator info */
50 isc_lfsr_t qid_lfsr2; /* state generator info */
51 dns_displist_t *qid_table; /* the table itself */
54 struct dns_dispatchmgr {
59 dns_portlist_t *portlist;
61 /* Locked by "lock". */
64 ISC_LIST(dns_dispatch_t) list;
66 /* locked by buffer lock */
68 isc_mutex_t buffer_lock;
69 unsigned int buffers; /* allocated buffers */
70 unsigned int buffersize; /* size of each buffer */
71 unsigned int maxbuffers; /* max buffers */
73 /* Locked internally. */
74 isc_mutex_t pool_lock;
75 isc_mempool_t *epool; /* memory pool for events */
76 isc_mempool_t *rpool; /* memory pool for replies */
77 isc_mempool_t *dpool; /* dispatch allocations */
78 isc_mempool_t *bpool; /* memory pool for buffers */
80 isc_entropy_t *entropy; /* entropy source */
83 #define MGR_SHUTTINGDOWN 0x00000001U
84 #define MGR_IS_SHUTTINGDOWN(l) (((l)->state & MGR_SHUTTINGDOWN) != 0)
86 #define IS_PRIVATE(d) (((d)->attributes & DNS_DISPATCHATTR_PRIVATE) != 0)
88 struct dns_dispentry {
95 isc_taskaction_t action;
97 isc_boolean_t item_out;
98 ISC_LIST(dns_dispatchevent_t) items;
99 ISC_LINK(dns_dispentry_t) link;
102 #define INVALID_BUCKET (0xffffdead)
104 struct dns_dispatch {
106 unsigned int magic; /* magic */
107 dns_dispatchmgr_t *mgr; /* dispatch manager */
108 isc_task_t *task; /* internal task */
109 isc_socket_t *socket; /* isc socket attached to */
110 isc_sockaddr_t local; /* local address */
111 unsigned int maxrequests; /* max requests */
112 isc_event_t *ctlevent;
114 /* Locked by mgr->lock. */
115 ISC_LINK(dns_dispatch_t) link;
117 /* Locked by "lock". */
118 isc_mutex_t lock; /* locks all below */
119 isc_sockettype_t socktype;
120 unsigned int attributes;
121 unsigned int refcount; /* number of users */
122 dns_dispatchevent_t *failsafe_ev; /* failsafe cancel event */
123 unsigned int shutting_down : 1,
127 recv_pending : 1; /* is a recv() pending? */
128 isc_result_t shutdown_why;
129 unsigned int requests; /* how many requests we have */
130 unsigned int tcpbuffers; /* allocated buffers */
131 dns_tcpmsg_t tcpmsg; /* for tcp streams */
135 #define QID_MAGIC ISC_MAGIC('Q', 'i', 'd', ' ')
136 #define VALID_QID(e) ISC_MAGIC_VALID((e), QID_MAGIC)
138 #define RESPONSE_MAGIC ISC_MAGIC('D', 'r', 's', 'p')
139 #define VALID_RESPONSE(e) ISC_MAGIC_VALID((e), RESPONSE_MAGIC)
141 #define DISPATCH_MAGIC ISC_MAGIC('D', 'i', 's', 'p')
142 #define VALID_DISPATCH(e) ISC_MAGIC_VALID((e), DISPATCH_MAGIC)
144 #define DNS_DISPATCHMGR_MAGIC ISC_MAGIC('D', 'M', 'g', 'r')
145 #define VALID_DISPATCHMGR(e) ISC_MAGIC_VALID((e), DNS_DISPATCHMGR_MAGIC)
147 #define DNS_QID(disp) ((disp)->socktype == isc_sockettype_tcp) ? \
148 (disp)->qid : (disp)->mgr->qid
152 static dns_dispentry_t *bucket_search(dns_qid_t *, isc_sockaddr_t *,
153 dns_messageid_t, unsigned int);
154 static isc_boolean_t destroy_disp_ok(dns_dispatch_t *);
155 static void destroy_disp(isc_task_t *task, isc_event_t *event);
156 static void udp_recv(isc_task_t *, isc_event_t *);
157 static void tcp_recv(isc_task_t *, isc_event_t *);
158 static void startrecv(dns_dispatch_t *);
159 static dns_messageid_t dns_randomid(dns_qid_t *);
160 static isc_uint32_t dns_hash(dns_qid_t *, isc_sockaddr_t *, dns_messageid_t);
161 static void free_buffer(dns_dispatch_t *disp, void *buf, unsigned int len);
162 static void *allocate_udp_buffer(dns_dispatch_t *disp);
163 static inline void free_event(dns_dispatch_t *disp, dns_dispatchevent_t *ev);
164 static inline dns_dispatchevent_t *allocate_event(dns_dispatch_t *disp);
165 static void do_cancel(dns_dispatch_t *disp);
166 static dns_dispentry_t *linear_first(dns_qid_t *disp);
167 static dns_dispentry_t *linear_next(dns_qid_t *disp,
168 dns_dispentry_t *resp);
169 static void dispatch_free(dns_dispatch_t **dispp);
170 static isc_result_t dispatch_createudp(dns_dispatchmgr_t *mgr,
171 isc_socketmgr_t *sockmgr,
172 isc_taskmgr_t *taskmgr,
173 isc_sockaddr_t *localaddr,
174 unsigned int maxrequests,
175 unsigned int attributes,
176 dns_dispatch_t **dispp);
177 static isc_boolean_t destroy_mgr_ok(dns_dispatchmgr_t *mgr);
178 static void destroy_mgr(dns_dispatchmgr_t **mgrp);
179 static isc_result_t qid_allocate(dns_dispatchmgr_t *mgr, unsigned int buckets,
180 unsigned int increment, dns_qid_t **qidp);
181 static void qid_destroy(isc_mem_t *mctx, dns_qid_t **qidp);
183 #define LVL(x) ISC_LOG_DEBUG(x)
186 mgr_log(dns_dispatchmgr_t *mgr, int level, const char *fmt, ...)
187 ISC_FORMAT_PRINTF(3, 4);
190 mgr_log(dns_dispatchmgr_t *mgr, int level, const char *fmt, ...) {
194 if (! isc_log_wouldlog(dns_lctx, level))
198 vsnprintf(msgbuf, sizeof(msgbuf), fmt, ap);
201 isc_log_write(dns_lctx,
202 DNS_LOGCATEGORY_DISPATCH, DNS_LOGMODULE_DISPATCH,
203 level, "dispatchmgr %p: %s", mgr, msgbuf);
207 dispatch_log(dns_dispatch_t *disp, int level, const char *fmt, ...)
208 ISC_FORMAT_PRINTF(3, 4);
211 dispatch_log(dns_dispatch_t *disp, int level, const char *fmt, ...) {
215 if (! isc_log_wouldlog(dns_lctx, level))
219 vsnprintf(msgbuf, sizeof(msgbuf), fmt, ap);
222 isc_log_write(dns_lctx,
223 DNS_LOGCATEGORY_DISPATCH, DNS_LOGMODULE_DISPATCH,
224 level, "dispatch %p: %s", disp, msgbuf);
228 request_log(dns_dispatch_t *disp, dns_dispentry_t *resp,
229 int level, const char *fmt, ...)
230 ISC_FORMAT_PRINTF(4, 5);
233 request_log(dns_dispatch_t *disp, dns_dispentry_t *resp,
234 int level, const char *fmt, ...)
240 if (! isc_log_wouldlog(dns_lctx, level))
244 vsnprintf(msgbuf, sizeof(msgbuf), fmt, ap);
247 if (VALID_RESPONSE(resp)) {
248 isc_sockaddr_format(&resp->host, peerbuf, sizeof(peerbuf));
249 isc_log_write(dns_lctx, DNS_LOGCATEGORY_DISPATCH,
250 DNS_LOGMODULE_DISPATCH, level,
251 "dispatch %p response %p %s: %s", disp, resp,
254 isc_log_write(dns_lctx, DNS_LOGCATEGORY_DISPATCH,
255 DNS_LOGMODULE_DISPATCH, level,
256 "dispatch %p req/resp %p: %s", disp, resp,
262 reseed_lfsr(isc_lfsr_t *lfsr, void *arg)
264 dns_dispatchmgr_t *mgr = arg;
268 REQUIRE(VALID_DISPATCHMGR(mgr));
270 if (mgr->entropy != NULL) {
271 result = isc_entropy_getdata(mgr->entropy, &val, sizeof(val),
273 INSIST(result == ISC_R_SUCCESS);
274 lfsr->count = (val & 0x1f) + 32;
279 lfsr->count = (random() & 0x1f) + 32; /* From 32 to 63 states */
280 lfsr->state = random();
284 * Return an unpredictable message ID.
286 static dns_messageid_t
287 dns_randomid(dns_qid_t *qid) {
290 id = isc_lfsr_generate32(&qid->qid_lfsr1, &qid->qid_lfsr2);
292 return (dns_messageid_t)(id & 0xFFFF);
296 * Return a hash of the destination and message id.
299 dns_hash(dns_qid_t *qid, isc_sockaddr_t *dest, dns_messageid_t id) {
302 ret = isc_sockaddr_hash(dest, ISC_TRUE);
304 ret %= qid->qid_nbuckets;
306 INSIST(ret < qid->qid_nbuckets);
312 * Find the first entry in 'qid'. Returns NULL if there are no entries.
314 static dns_dispentry_t *
315 linear_first(dns_qid_t *qid) {
316 dns_dispentry_t *ret;
321 while (bucket < qid->qid_nbuckets) {
322 ret = ISC_LIST_HEAD(qid->qid_table[bucket]);
332 * Find the next entry after 'resp' in 'qid'. Return NULL if there are
335 static dns_dispentry_t *
336 linear_next(dns_qid_t *qid, dns_dispentry_t *resp) {
337 dns_dispentry_t *ret;
340 ret = ISC_LIST_NEXT(resp, link);
344 bucket = resp->bucket;
346 while (bucket < qid->qid_nbuckets) {
347 ret = ISC_LIST_HEAD(qid->qid_table[bucket]);
357 * The dispatch must be locked.
360 destroy_disp_ok(dns_dispatch_t *disp)
362 if (disp->refcount != 0)
365 if (disp->recv_pending != 0)
368 if (disp->shutting_down == 0)
376 * Called when refcount reaches 0 (and safe to destroy).
378 * The dispatcher must not be locked.
379 * The manager must be locked.
382 destroy_disp(isc_task_t *task, isc_event_t *event) {
383 dns_dispatch_t *disp;
384 dns_dispatchmgr_t *mgr;
385 isc_boolean_t killmgr;
387 INSIST(event->ev_type == DNS_EVENT_DISPATCHCONTROL);
391 disp = event->ev_arg;
395 ISC_LIST_UNLINK(mgr->list, disp, link);
397 dispatch_log(disp, LVL(90),
398 "shutting down; detaching from sock %p, task %p",
399 disp->socket, disp->task);
401 isc_socket_detach(&disp->socket);
402 isc_task_detach(&disp->task);
403 isc_event_free(&event);
405 dispatch_free(&disp);
407 killmgr = destroy_mgr_ok(mgr);
415 * Find an entry for query ID 'id' and socket address 'dest' in 'qid'.
416 * Return NULL if no such entry exists.
418 static dns_dispentry_t *
419 bucket_search(dns_qid_t *qid, isc_sockaddr_t *dest, dns_messageid_t id,
422 dns_dispentry_t *res;
424 REQUIRE(bucket < qid->qid_nbuckets);
426 res = ISC_LIST_HEAD(qid->qid_table[bucket]);
428 while (res != NULL) {
429 if ((res->id == id) && isc_sockaddr_equal(dest, &res->host))
431 res = ISC_LIST_NEXT(res, link);
438 free_buffer(dns_dispatch_t *disp, void *buf, unsigned int len) {
439 INSIST(buf != NULL && len != 0);
442 switch (disp->socktype) {
443 case isc_sockettype_tcp:
444 INSIST(disp->tcpbuffers > 0);
446 isc_mem_put(disp->mgr->mctx, buf, len);
448 case isc_sockettype_udp:
449 LOCK(&disp->mgr->buffer_lock);
450 INSIST(disp->mgr->buffers > 0);
451 INSIST(len == disp->mgr->buffersize);
452 disp->mgr->buffers--;
453 isc_mempool_put(disp->mgr->bpool, buf);
454 UNLOCK(&disp->mgr->buffer_lock);
463 allocate_udp_buffer(dns_dispatch_t *disp) {
466 LOCK(&disp->mgr->buffer_lock);
467 temp = isc_mempool_get(disp->mgr->bpool);
470 disp->mgr->buffers++;
471 UNLOCK(&disp->mgr->buffer_lock);
477 free_event(dns_dispatch_t *disp, dns_dispatchevent_t *ev) {
478 if (disp->failsafe_ev == ev) {
479 INSIST(disp->shutdown_out == 1);
480 disp->shutdown_out = 0;
485 isc_mempool_put(disp->mgr->epool, ev);
488 static inline dns_dispatchevent_t *
489 allocate_event(dns_dispatch_t *disp) {
490 dns_dispatchevent_t *ev;
492 ev = isc_mempool_get(disp->mgr->epool);
495 ISC_EVENT_INIT(ev, sizeof(*ev), 0, NULL, 0,
496 NULL, NULL, NULL, NULL, NULL);
504 * If I/O result == CANCELED or error, free the buffer.
506 * If query, free the buffer, restart.
509 * Allocate event, fill in details.
510 * If cannot allocate, free buffer, restart.
511 * find target. If not found, free buffer, restart.
512 * if event queue is not empty, queue. else, send.
516 udp_recv(isc_task_t *task, isc_event_t *ev_in) {
517 isc_socketevent_t *ev = (isc_socketevent_t *)ev_in;
518 dns_dispatch_t *disp = ev_in->ev_arg;
523 dns_dispentry_t *resp;
524 dns_dispatchevent_t *rev;
526 isc_boolean_t killit;
527 isc_boolean_t queue_response;
528 dns_dispatchmgr_t *mgr;
530 isc_netaddr_t netaddr;
540 dispatch_log(disp, LVL(90),
541 "got packet: requests %d, buffers %d, recvs %d",
542 disp->requests, disp->mgr->buffers, disp->recv_pending);
544 if (ev->ev_type == ISC_SOCKEVENT_RECVDONE) {
546 * Unless the receive event was imported from a listening
547 * interface, in which case the event type is
548 * DNS_EVENT_IMPORTRECVDONE, receive operation must be pending.
550 INSIST(disp->recv_pending != 0);
551 disp->recv_pending = 0;
554 if (disp->shutting_down) {
556 * This dispatcher is shutting down.
558 free_buffer(disp, ev->region.base, ev->region.length);
560 isc_event_free(&ev_in);
563 killit = destroy_disp_ok(disp);
566 isc_task_send(disp->task, &disp->ctlevent);
571 if (ev->result != ISC_R_SUCCESS) {
572 free_buffer(disp, ev->region.base, ev->region.length);
574 if (ev->result != ISC_R_CANCELED)
575 dispatch_log(disp, ISC_LOG_ERROR,
576 "odd socket result in udp_recv(): %s",
577 isc_result_totext(ev->result));
580 isc_event_free(&ev_in);
585 * If this is from a blackholed address, drop it.
587 isc_netaddr_fromsockaddr(&netaddr, &ev->address);
588 if (disp->mgr->blackhole != NULL &&
589 dns_acl_match(&netaddr, NULL, disp->mgr->blackhole,
590 NULL, &match, NULL) == ISC_R_SUCCESS &&
593 if (isc_log_wouldlog(dns_lctx, LVL(10))) {
594 char netaddrstr[ISC_NETADDR_FORMATSIZE];
595 isc_netaddr_format(&netaddr, netaddrstr,
597 dispatch_log(disp, LVL(10),
598 "blackholed packet from %s",
601 free_buffer(disp, ev->region.base, ev->region.length);
606 * Peek into the buffer to see what we can see.
608 isc_buffer_init(&source, ev->region.base, ev->region.length);
609 isc_buffer_add(&source, ev->n);
610 dres = dns_message_peekheader(&source, &id, &flags);
611 if (dres != ISC_R_SUCCESS) {
612 free_buffer(disp, ev->region.base, ev->region.length);
613 dispatch_log(disp, LVL(10), "got garbage packet");
617 dispatch_log(disp, LVL(92),
618 "got valid DNS message header, /QR %c, id %u",
619 ((flags & DNS_MESSAGEFLAG_QR) ? '1' : '0'), id);
622 * Look at flags. If query, drop it. If response,
623 * look to see where it goes.
625 queue_response = ISC_FALSE;
626 if ((flags & DNS_MESSAGEFLAG_QR) == 0) {
628 free_buffer(disp, ev->region.base, ev->region.length);
633 bucket = dns_hash(qid, &ev->address, id);
635 resp = bucket_search(qid, &ev->address, id, bucket);
636 dispatch_log(disp, LVL(90),
637 "search for response in bucket %d: %s",
638 bucket, (resp == NULL ? "not found" : "found"));
641 free_buffer(disp, ev->region.base, ev->region.length);
646 * Now that we have the original dispatch the query was sent
647 * from check that the address and port the response was
648 * sent to make sense.
650 if (disp != resp->disp) {
655 * Check that the socket types and ports match.
657 if (disp->socktype != resp->disp->socktype ||
658 isc_sockaddr_getport(&disp->local) !=
659 isc_sockaddr_getport(&resp->disp->local)) {
660 free_buffer(disp, ev->region.base, ev->region.length);
665 * If both dispatches are bound to an address then fail as
666 * the addresses can't be equal (enforced by the IP stack).
668 * Note under Linux a packet can be sent out via IPv4 socket
669 * and the response be received via a IPv6 socket.
671 * Requests sent out via IPv6 should always come back in
674 if (isc_sockaddr_pf(&resp->disp->local) == PF_INET6 &&
675 isc_sockaddr_pf(&disp->local) != PF_INET6) {
676 free_buffer(disp, ev->region.base, ev->region.length);
679 isc_sockaddr_anyofpf(&a1, isc_sockaddr_pf(&resp->disp->local));
680 isc_sockaddr_anyofpf(&a2, isc_sockaddr_pf(&disp->local));
681 if (!isc_sockaddr_eqaddr(&a1, &resp->disp->local) &&
682 !isc_sockaddr_eqaddr(&a2, &disp->local)) {
683 free_buffer(disp, ev->region.base, ev->region.length);
688 queue_response = resp->item_out;
689 rev = allocate_event(resp->disp);
691 free_buffer(disp, ev->region.base, ev->region.length);
696 * At this point, rev contains the event we want to fill in, and
697 * resp contains the information on the place to send it to.
698 * Send the event off.
700 isc_buffer_init(&rev->buffer, ev->region.base, ev->region.length);
701 isc_buffer_add(&rev->buffer, ev->n);
702 rev->result = ISC_R_SUCCESS;
704 rev->addr = ev->address;
705 rev->pktinfo = ev->pktinfo;
706 rev->attributes = ev->attributes;
707 if (queue_response) {
708 ISC_LIST_APPEND(resp->items, rev, ev_link);
710 ISC_EVENT_INIT(rev, sizeof(*rev), 0, NULL,
712 resp->action, resp->arg, resp, NULL, NULL);
713 request_log(disp, resp, LVL(90),
714 "[a] Sent event %p buffer %p len %d to task %p",
715 rev, rev->buffer.base, rev->buffer.length,
717 resp->item_out = ISC_TRUE;
718 isc_task_send(resp->task, ISC_EVENT_PTR(&rev));
724 * Restart recv() to get the next packet.
731 isc_event_free(&ev_in);
737 * If I/O result == CANCELED, EOF, or error, notify everyone as the
738 * various queues drain.
743 * Allocate event, fill in details.
744 * If cannot allocate, restart.
745 * find target. If not found, restart.
746 * if event queue is not empty, queue. else, send.
750 tcp_recv(isc_task_t *task, isc_event_t *ev_in) {
751 dns_dispatch_t *disp = ev_in->ev_arg;
752 dns_tcpmsg_t *tcpmsg = &disp->tcpmsg;
756 dns_dispentry_t *resp;
757 dns_dispatchevent_t *rev;
759 isc_boolean_t killit;
760 isc_boolean_t queue_response;
763 char buf[ISC_SOCKADDR_FORMATSIZE];
767 REQUIRE(VALID_DISPATCH(disp));
771 dispatch_log(disp, LVL(90),
772 "got TCP packet: requests %d, buffers %d, recvs %d",
773 disp->requests, disp->tcpbuffers, disp->recv_pending);
777 INSIST(disp->recv_pending != 0);
778 disp->recv_pending = 0;
780 if (disp->refcount == 0) {
782 * This dispatcher is shutting down. Force cancelation.
784 tcpmsg->result = ISC_R_CANCELED;
787 if (tcpmsg->result != ISC_R_SUCCESS) {
788 switch (tcpmsg->result) {
793 dispatch_log(disp, LVL(90), "shutting down on EOF");
797 case ISC_R_CONNECTIONRESET:
798 level = ISC_LOG_INFO;
802 level = ISC_LOG_ERROR;
804 isc_sockaddr_format(&tcpmsg->address, buf, sizeof(buf));
805 dispatch_log(disp, level, "shutting down due to TCP "
806 "receive error: %s: %s", buf,
807 isc_result_totext(tcpmsg->result));
813 * The event is statically allocated in the tcpmsg
814 * structure, and destroy_disp() frees the tcpmsg, so we must
815 * free the event *before* calling destroy_disp().
817 isc_event_free(&ev_in);
819 disp->shutting_down = 1;
820 disp->shutdown_why = tcpmsg->result;
823 * If the recv() was canceled pass the word on.
825 killit = destroy_disp_ok(disp);
828 isc_task_send(disp->task, &disp->ctlevent);
832 dispatch_log(disp, LVL(90), "result %d, length == %d, addr = %p",
834 tcpmsg->buffer.length, tcpmsg->buffer.base);
837 * Peek into the buffer to see what we can see.
839 dres = dns_message_peekheader(&tcpmsg->buffer, &id, &flags);
840 if (dres != ISC_R_SUCCESS) {
841 dispatch_log(disp, LVL(10), "got garbage packet");
845 dispatch_log(disp, LVL(92),
846 "got valid DNS message header, /QR %c, id %u",
847 ((flags & DNS_MESSAGEFLAG_QR) ? '1' : '0'), id);
850 * Allocate an event to send to the query or response client, and
851 * allocate a new buffer for our use.
855 * Look at flags. If query, drop it. If response,
856 * look to see where it goes.
858 queue_response = ISC_FALSE;
859 if ((flags & DNS_MESSAGEFLAG_QR) == 0) {
869 bucket = dns_hash(qid, &tcpmsg->address, id);
871 resp = bucket_search(qid, &tcpmsg->address, id, bucket);
872 dispatch_log(disp, LVL(90),
873 "search for response in bucket %d: %s",
874 bucket, (resp == NULL ? "not found" : "found"));
878 queue_response = resp->item_out;
879 rev = allocate_event(disp);
884 * At this point, rev contains the event we want to fill in, and
885 * resp contains the information on the place to send it to.
886 * Send the event off.
888 dns_tcpmsg_keepbuffer(tcpmsg, &rev->buffer);
890 rev->result = ISC_R_SUCCESS;
892 rev->addr = tcpmsg->address;
893 if (queue_response) {
894 ISC_LIST_APPEND(resp->items, rev, ev_link);
896 ISC_EVENT_INIT(rev, sizeof(*rev), 0, NULL, DNS_EVENT_DISPATCH,
897 resp->action, resp->arg, resp, NULL, NULL);
898 request_log(disp, resp, LVL(90),
899 "[b] Sent event %p buffer %p len %d to task %p",
900 rev, rev->buffer.base, rev->buffer.length,
902 resp->item_out = ISC_TRUE;
903 isc_task_send(resp->task, ISC_EVENT_PTR(&rev));
909 * Restart recv() to get the next packet.
916 isc_event_free(&ev_in);
920 * disp must be locked.
923 startrecv(dns_dispatch_t *disp) {
927 if (disp->shutting_down == 1)
930 if ((disp->attributes & DNS_DISPATCHATTR_NOLISTEN) != 0)
933 if (disp->recv_pending != 0)
936 if (disp->mgr->buffers >= disp->mgr->maxbuffers)
939 switch (disp->socktype) {
941 * UDP reads are always maximal.
943 case isc_sockettype_udp:
944 region.length = disp->mgr->buffersize;
945 region.base = allocate_udp_buffer(disp);
946 if (region.base == NULL)
948 res = isc_socket_recv(disp->socket, ®ion, 1,
949 disp->task, udp_recv, disp);
950 if (res != ISC_R_SUCCESS) {
951 free_buffer(disp, region.base, region.length);
952 disp->shutdown_why = res;
953 disp->shutting_down = 1;
957 INSIST(disp->recv_pending == 0);
958 disp->recv_pending = 1;
961 case isc_sockettype_tcp:
962 res = dns_tcpmsg_readmessage(&disp->tcpmsg, disp->task,
964 if (res != ISC_R_SUCCESS) {
965 disp->shutdown_why = res;
966 disp->shutting_down = 1;
970 INSIST(disp->recv_pending == 0);
971 disp->recv_pending = 1;
977 * Mgr must be locked when calling this function.
980 destroy_mgr_ok(dns_dispatchmgr_t *mgr) {
981 mgr_log(mgr, LVL(90),
982 "destroy_mgr_ok: shuttingdown=%d, listnonempty=%d, "
983 "epool=%d, rpool=%d, dpool=%d",
984 MGR_IS_SHUTTINGDOWN(mgr), !ISC_LIST_EMPTY(mgr->list),
985 isc_mempool_getallocated(mgr->epool),
986 isc_mempool_getallocated(mgr->rpool),
987 isc_mempool_getallocated(mgr->dpool));
988 if (!MGR_IS_SHUTTINGDOWN(mgr))
990 if (!ISC_LIST_EMPTY(mgr->list))
992 if (isc_mempool_getallocated(mgr->epool) != 0)
994 if (isc_mempool_getallocated(mgr->rpool) != 0)
996 if (isc_mempool_getallocated(mgr->dpool) != 0)
1003 * Mgr must be unlocked when calling this function.
1006 destroy_mgr(dns_dispatchmgr_t **mgrp) {
1008 dns_dispatchmgr_t *mgr;
1017 DESTROYLOCK(&mgr->lock);
1020 isc_mempool_destroy(&mgr->epool);
1021 isc_mempool_destroy(&mgr->rpool);
1022 isc_mempool_destroy(&mgr->dpool);
1023 isc_mempool_destroy(&mgr->bpool);
1025 DESTROYLOCK(&mgr->pool_lock);
1027 if (mgr->entropy != NULL)
1028 isc_entropy_detach(&mgr->entropy);
1029 if (mgr->qid != NULL)
1030 qid_destroy(mctx, &mgr->qid);
1032 DESTROYLOCK(&mgr->buffer_lock);
1034 if (mgr->blackhole != NULL)
1035 dns_acl_detach(&mgr->blackhole);
1037 if (mgr->portlist != NULL)
1038 dns_portlist_detach(&mgr->portlist);
1040 isc_mem_put(mctx, mgr, sizeof(dns_dispatchmgr_t));
1041 isc_mem_detach(&mctx);
1045 create_socket(isc_socketmgr_t *mgr, isc_sockaddr_t *local,
1046 isc_socket_t **sockp)
1049 isc_result_t result;
1052 result = isc_socket_create(mgr, isc_sockaddr_pf(local),
1053 isc_sockettype_udp, &sock);
1054 if (result != ISC_R_SUCCESS)
1057 #ifndef ISC_ALLOW_MAPPED
1058 isc_socket_ipv6only(sock, ISC_TRUE);
1060 result = isc_socket_bind(sock, local);
1061 if (result != ISC_R_SUCCESS) {
1062 isc_socket_detach(&sock);
1067 return (ISC_R_SUCCESS);
1075 dns_dispatchmgr_create(isc_mem_t *mctx, isc_entropy_t *entropy,
1076 dns_dispatchmgr_t **mgrp)
1078 dns_dispatchmgr_t *mgr;
1079 isc_result_t result;
1081 REQUIRE(mctx != NULL);
1082 REQUIRE(mgrp != NULL && *mgrp == NULL);
1084 mgr = isc_mem_get(mctx, sizeof(dns_dispatchmgr_t));
1086 return (ISC_R_NOMEMORY);
1089 isc_mem_attach(mctx, &mgr->mctx);
1091 mgr->blackhole = NULL;
1092 mgr->portlist = NULL;
1094 result = isc_mutex_init(&mgr->lock);
1095 if (result != ISC_R_SUCCESS)
1098 result = isc_mutex_init(&mgr->buffer_lock);
1099 if (result != ISC_R_SUCCESS)
1102 result = isc_mutex_init(&mgr->pool_lock);
1103 if (result != ISC_R_SUCCESS)
1104 goto kill_buffer_lock;
1107 if (isc_mempool_create(mgr->mctx, sizeof(dns_dispatchevent_t),
1108 &mgr->epool) != ISC_R_SUCCESS) {
1109 result = ISC_R_NOMEMORY;
1110 goto kill_pool_lock;
1114 if (isc_mempool_create(mgr->mctx, sizeof(dns_dispentry_t),
1115 &mgr->rpool) != ISC_R_SUCCESS) {
1116 result = ISC_R_NOMEMORY;
1121 if (isc_mempool_create(mgr->mctx, sizeof(dns_dispatch_t),
1122 &mgr->dpool) != ISC_R_SUCCESS) {
1123 result = ISC_R_NOMEMORY;
1127 isc_mempool_setname(mgr->epool, "dispmgr_epool");
1128 isc_mempool_setfreemax(mgr->epool, 1024);
1129 isc_mempool_associatelock(mgr->epool, &mgr->pool_lock);
1131 isc_mempool_setname(mgr->rpool, "dispmgr_rpool");
1132 isc_mempool_setfreemax(mgr->rpool, 1024);
1133 isc_mempool_associatelock(mgr->rpool, &mgr->pool_lock);
1135 isc_mempool_setname(mgr->dpool, "dispmgr_dpool");
1136 isc_mempool_setfreemax(mgr->dpool, 1024);
1137 isc_mempool_associatelock(mgr->dpool, &mgr->pool_lock);
1140 mgr->buffersize = 0;
1141 mgr->maxbuffers = 0;
1143 mgr->entropy = NULL;
1146 ISC_LIST_INIT(mgr->list);
1147 mgr->magic = DNS_DISPATCHMGR_MAGIC;
1149 if (entropy != NULL)
1150 isc_entropy_attach(entropy, &mgr->entropy);
1153 return (ISC_R_SUCCESS);
1156 isc_mempool_destroy(&mgr->rpool);
1158 isc_mempool_destroy(&mgr->epool);
1160 DESTROYLOCK(&mgr->pool_lock);
1162 DESTROYLOCK(&mgr->buffer_lock);
1164 DESTROYLOCK(&mgr->lock);
1166 isc_mem_put(mctx, mgr, sizeof(dns_dispatchmgr_t));
1167 isc_mem_detach(&mctx);
1173 dns_dispatchmgr_setblackhole(dns_dispatchmgr_t *mgr, dns_acl_t *blackhole) {
1174 REQUIRE(VALID_DISPATCHMGR(mgr));
1175 if (mgr->blackhole != NULL)
1176 dns_acl_detach(&mgr->blackhole);
1177 dns_acl_attach(blackhole, &mgr->blackhole);
1181 dns_dispatchmgr_getblackhole(dns_dispatchmgr_t *mgr) {
1182 REQUIRE(VALID_DISPATCHMGR(mgr));
1183 return (mgr->blackhole);
1187 dns_dispatchmgr_setblackportlist(dns_dispatchmgr_t *mgr,
1188 dns_portlist_t *portlist)
1190 REQUIRE(VALID_DISPATCHMGR(mgr));
1191 if (mgr->portlist != NULL)
1192 dns_portlist_detach(&mgr->portlist);
1193 if (portlist != NULL)
1194 dns_portlist_attach(portlist, &mgr->portlist);
1198 dns_dispatchmgr_getblackportlist(dns_dispatchmgr_t *mgr) {
1199 REQUIRE(VALID_DISPATCHMGR(mgr));
1200 return (mgr->portlist);
1204 dns_dispatchmgr_setudp(dns_dispatchmgr_t *mgr,
1205 unsigned int buffersize, unsigned int maxbuffers,
1206 unsigned int buckets, unsigned int increment)
1208 isc_result_t result;
1210 REQUIRE(VALID_DISPATCHMGR(mgr));
1211 REQUIRE(buffersize >= 512 && buffersize < (64 * 1024));
1212 REQUIRE(maxbuffers > 0);
1213 REQUIRE(buckets < 2097169); /* next prime > 65536 * 32 */
1214 REQUIRE(increment > buckets);
1217 * Keep some number of items around. This should be a config
1218 * option. For now, keep 8, but later keep at least two even
1219 * if the caller wants less. This allows us to ensure certain
1220 * things, like an event can be "freed" and the next allocation
1221 * will always succeed.
1223 * Note that if limits are placed on anything here, we use one
1224 * event internally, so the actual limit should be "wanted + 1."
1232 LOCK(&mgr->buffer_lock);
1233 if (mgr->bpool != NULL) {
1234 isc_mempool_setmaxalloc(mgr->bpool, maxbuffers);
1235 mgr->maxbuffers = maxbuffers;
1236 UNLOCK(&mgr->buffer_lock);
1237 return (ISC_R_SUCCESS);
1240 if (isc_mempool_create(mgr->mctx, buffersize,
1241 &mgr->bpool) != ISC_R_SUCCESS) {
1242 return (ISC_R_NOMEMORY);
1245 isc_mempool_setname(mgr->bpool, "dispmgr_bpool");
1246 isc_mempool_setmaxalloc(mgr->bpool, maxbuffers);
1247 isc_mempool_associatelock(mgr->bpool, &mgr->pool_lock);
1249 result = qid_allocate(mgr, buckets, increment, &mgr->qid);
1250 if (result != ISC_R_SUCCESS)
1253 mgr->buffersize = buffersize;
1254 mgr->maxbuffers = maxbuffers;
1255 UNLOCK(&mgr->buffer_lock);
1256 return (ISC_R_SUCCESS);
1259 isc_mempool_destroy(&mgr->bpool);
1260 UNLOCK(&mgr->buffer_lock);
1261 return (ISC_R_NOMEMORY);
1265 dns_dispatchmgr_destroy(dns_dispatchmgr_t **mgrp) {
1266 dns_dispatchmgr_t *mgr;
1267 isc_boolean_t killit;
1269 REQUIRE(mgrp != NULL);
1270 REQUIRE(VALID_DISPATCHMGR(*mgrp));
1276 mgr->state |= MGR_SHUTTINGDOWN;
1278 killit = destroy_mgr_ok(mgr);
1281 mgr_log(mgr, LVL(90), "destroy: killit=%d", killit);
1287 static isc_boolean_t
1288 blacklisted(dns_dispatchmgr_t *mgr, isc_socket_t *sock) {
1289 isc_sockaddr_t sockaddr;
1290 isc_result_t result;
1292 if (mgr->portlist == NULL)
1295 result = isc_socket_getsockname(sock, &sockaddr);
1296 if (result != ISC_R_SUCCESS)
1299 if (mgr->portlist != NULL &&
1300 dns_portlist_match(mgr->portlist, isc_sockaddr_pf(&sockaddr),
1301 isc_sockaddr_getport(&sockaddr)))
1306 #define ATTRMATCH(_a1, _a2, _mask) (((_a1) & (_mask)) == ((_a2) & (_mask)))
1308 static isc_boolean_t
1309 local_addr_match(dns_dispatch_t *disp, isc_sockaddr_t *addr) {
1310 isc_sockaddr_t sockaddr;
1311 isc_result_t result;
1317 * Don't match wildcard ports against newly blacklisted ports.
1319 if (disp->mgr->portlist != NULL &&
1320 isc_sockaddr_getport(addr) == 0 &&
1321 isc_sockaddr_getport(&disp->local) == 0 &&
1322 blacklisted(disp->mgr, disp->socket))
1326 * Check if we match the binding <address,port>.
1327 * Wildcard ports match/fail here.
1329 if (isc_sockaddr_equal(&disp->local, addr))
1331 if (isc_sockaddr_getport(addr) == 0)
1335 * Check if we match a bound wildcard port <address,port>.
1337 if (!isc_sockaddr_eqaddr(&disp->local, addr))
1339 result = isc_socket_getsockname(disp->socket, &sockaddr);
1340 if (result != ISC_R_SUCCESS)
1343 return (isc_sockaddr_equal(&sockaddr, addr));
1347 * Requires mgr be locked.
1349 * No dispatcher can be locked by this thread when calling this function.
1353 * If a matching dispatcher is found, it is locked after this function
1354 * returns, and must be unlocked by the caller.
1357 dispatch_find(dns_dispatchmgr_t *mgr, isc_sockaddr_t *local,
1358 unsigned int attributes, unsigned int mask,
1359 dns_dispatch_t **dispp)
1361 dns_dispatch_t *disp;
1362 isc_result_t result;
1365 * Make certain that we will not match a private dispatch.
1367 attributes &= ~DNS_DISPATCHATTR_PRIVATE;
1368 mask |= DNS_DISPATCHATTR_PRIVATE;
1370 disp = ISC_LIST_HEAD(mgr->list);
1371 while (disp != NULL) {
1373 if ((disp->shutting_down == 0)
1374 && ATTRMATCH(disp->attributes, attributes, mask)
1375 && local_addr_match(disp, local))
1377 UNLOCK(&disp->lock);
1378 disp = ISC_LIST_NEXT(disp, link);
1382 result = ISC_R_NOTFOUND;
1387 result = ISC_R_SUCCESS;
1394 qid_allocate(dns_dispatchmgr_t *mgr, unsigned int buckets,
1395 unsigned int increment, dns_qid_t **qidp)
1400 REQUIRE(VALID_DISPATCHMGR(mgr));
1401 REQUIRE(buckets < 2097169); /* next prime > 65536 * 32 */
1402 REQUIRE(increment > buckets);
1403 REQUIRE(qidp != NULL && *qidp == NULL);
1405 qid = isc_mem_get(mgr->mctx, sizeof(*qid));
1407 return (ISC_R_NOMEMORY);
1409 qid->qid_table = isc_mem_get(mgr->mctx,
1410 buckets * sizeof(dns_displist_t));
1411 if (qid->qid_table == NULL) {
1412 isc_mem_put(mgr->mctx, qid, sizeof(*qid));
1413 return (ISC_R_NOMEMORY);
1416 if (isc_mutex_init(&qid->lock) != ISC_R_SUCCESS) {
1417 UNEXPECTED_ERROR(__FILE__, __LINE__, "isc_mutex_init failed");
1418 isc_mem_put(mgr->mctx, qid->qid_table,
1419 buckets * sizeof(dns_displist_t));
1420 isc_mem_put(mgr->mctx, qid, sizeof(*qid));
1421 return (ISC_R_UNEXPECTED);
1424 for (i = 0; i < buckets; i++)
1425 ISC_LIST_INIT(qid->qid_table[i]);
1427 qid->qid_nbuckets = buckets;
1428 qid->qid_increment = increment;
1429 qid->magic = QID_MAGIC;
1432 * Initialize to a 32-bit LFSR. Both of these are from Applied
1436 * x^32 + x^7 + x^5 + x^3 + x^2 + x + 1
1439 * x^32 + x^7 + x^6 + x^2 + 1
1441 isc_lfsr_init(&qid->qid_lfsr1, 0, 32, 0x80000057U,
1442 0, reseed_lfsr, mgr);
1443 isc_lfsr_init(&qid->qid_lfsr2, 0, 32, 0x80000062U,
1444 0, reseed_lfsr, mgr);
1446 return (ISC_R_SUCCESS);
1450 qid_destroy(isc_mem_t *mctx, dns_qid_t **qidp) {
1453 REQUIRE(qidp != NULL);
1456 REQUIRE(VALID_QID(qid));
1460 isc_mem_put(mctx, qid->qid_table,
1461 qid->qid_nbuckets * sizeof(dns_displist_t));
1462 DESTROYLOCK(&qid->lock);
1463 isc_mem_put(mctx, qid, sizeof(*qid));
1467 * Allocate and set important limits.
1470 dispatch_allocate(dns_dispatchmgr_t *mgr, unsigned int maxrequests,
1471 dns_dispatch_t **dispp)
1473 dns_dispatch_t *disp;
1476 REQUIRE(VALID_DISPATCHMGR(mgr));
1477 REQUIRE(dispp != NULL && *dispp == NULL);
1480 * Set up the dispatcher, mostly. Don't bother setting some of
1481 * the options that are controlled by tcp vs. udp, etc.
1484 disp = isc_mempool_get(mgr->dpool);
1486 return (ISC_R_NOMEMORY);
1490 disp->maxrequests = maxrequests;
1491 disp->attributes = 0;
1492 ISC_LINK_INIT(disp, link);
1494 disp->recv_pending = 0;
1495 memset(&disp->local, 0, sizeof(disp->local));
1496 disp->shutting_down = 0;
1497 disp->shutdown_out = 0;
1498 disp->connected = 0;
1499 disp->tcpmsg_valid = 0;
1500 disp->shutdown_why = ISC_R_UNEXPECTED;
1502 disp->tcpbuffers = 0;
1505 if (isc_mutex_init(&disp->lock) != ISC_R_SUCCESS) {
1506 res = ISC_R_UNEXPECTED;
1507 UNEXPECTED_ERROR(__FILE__, __LINE__, "isc_mutex_init failed");
1511 disp->failsafe_ev = allocate_event(disp);
1512 if (disp->failsafe_ev == NULL) {
1513 res = ISC_R_NOMEMORY;
1517 disp->magic = DISPATCH_MAGIC;
1520 return (ISC_R_SUCCESS);
1526 DESTROYLOCK(&disp->lock);
1528 isc_mempool_put(mgr->dpool, disp);
1535 * MUST be unlocked, and not used by anthing.
1538 dispatch_free(dns_dispatch_t **dispp)
1540 dns_dispatch_t *disp;
1541 dns_dispatchmgr_t *mgr;
1543 REQUIRE(VALID_DISPATCH(*dispp));
1548 REQUIRE(VALID_DISPATCHMGR(mgr));
1550 if (disp->tcpmsg_valid) {
1551 dns_tcpmsg_invalidate(&disp->tcpmsg);
1552 disp->tcpmsg_valid = 0;
1555 INSIST(disp->tcpbuffers == 0);
1556 INSIST(disp->requests == 0);
1557 INSIST(disp->recv_pending == 0);
1559 isc_mempool_put(mgr->epool, disp->failsafe_ev);
1560 disp->failsafe_ev = NULL;
1562 if (disp->qid != NULL)
1563 qid_destroy(mgr->mctx, &disp->qid);
1565 DESTROYLOCK(&disp->lock);
1567 isc_mempool_put(mgr->dpool, disp);
1571 dns_dispatch_createtcp(dns_dispatchmgr_t *mgr, isc_socket_t *sock,
1572 isc_taskmgr_t *taskmgr, unsigned int buffersize,
1573 unsigned int maxbuffers, unsigned int maxrequests,
1574 unsigned int buckets, unsigned int increment,
1575 unsigned int attributes, dns_dispatch_t **dispp)
1577 isc_result_t result;
1578 dns_dispatch_t *disp;
1583 REQUIRE(VALID_DISPATCHMGR(mgr));
1584 REQUIRE(isc_socket_gettype(sock) == isc_sockettype_tcp);
1585 REQUIRE((attributes & DNS_DISPATCHATTR_TCP) != 0);
1586 REQUIRE((attributes & DNS_DISPATCHATTR_UDP) == 0);
1588 attributes |= DNS_DISPATCHATTR_PRIVATE; /* XXXMLG */
1593 * dispatch_allocate() checks mgr for us.
1594 * qid_allocate() checks buckets and increment for us.
1597 result = dispatch_allocate(mgr, maxrequests, &disp);
1598 if (result != ISC_R_SUCCESS) {
1603 result = qid_allocate(mgr, buckets, increment, &disp->qid);
1604 if (result != ISC_R_SUCCESS)
1605 goto deallocate_dispatch;
1607 disp->socktype = isc_sockettype_tcp;
1608 disp->socket = NULL;
1609 isc_socket_attach(sock, &disp->socket);
1612 result = isc_task_create(taskmgr, 0, &disp->task);
1613 if (result != ISC_R_SUCCESS)
1616 disp->ctlevent = isc_event_allocate(mgr->mctx, disp,
1617 DNS_EVENT_DISPATCHCONTROL,
1619 sizeof(isc_event_t));
1620 if (disp->ctlevent == NULL)
1623 isc_task_setname(disp->task, "tcpdispatch", disp);
1625 dns_tcpmsg_init(mgr->mctx, disp->socket, &disp->tcpmsg);
1626 disp->tcpmsg_valid = 1;
1628 disp->attributes = attributes;
1631 * Append it to the dispatcher list.
1633 ISC_LIST_APPEND(mgr->list, disp, link);
1636 mgr_log(mgr, LVL(90), "created TCP dispatcher %p", disp);
1637 dispatch_log(disp, LVL(90), "created task %p", disp->task);
1641 return (ISC_R_SUCCESS);
1647 isc_task_detach(&disp->task);
1649 isc_socket_detach(&disp->socket);
1650 deallocate_dispatch:
1651 dispatch_free(&disp);
1659 dns_dispatch_getudp(dns_dispatchmgr_t *mgr, isc_socketmgr_t *sockmgr,
1660 isc_taskmgr_t *taskmgr, isc_sockaddr_t *localaddr,
1661 unsigned int buffersize,
1662 unsigned int maxbuffers, unsigned int maxrequests,
1663 unsigned int buckets, unsigned int increment,
1664 unsigned int attributes, unsigned int mask,
1665 dns_dispatch_t **dispp)
1667 isc_result_t result;
1668 dns_dispatch_t *disp;
1670 REQUIRE(VALID_DISPATCHMGR(mgr));
1671 REQUIRE(sockmgr != NULL);
1672 REQUIRE(localaddr != NULL);
1673 REQUIRE(taskmgr != NULL);
1674 REQUIRE(buffersize >= 512 && buffersize < (64 * 1024));
1675 REQUIRE(maxbuffers > 0);
1676 REQUIRE(buckets < 2097169); /* next prime > 65536 * 32 */
1677 REQUIRE(increment > buckets);
1678 REQUIRE(dispp != NULL && *dispp == NULL);
1679 REQUIRE((attributes & DNS_DISPATCHATTR_TCP) == 0);
1681 result = dns_dispatchmgr_setudp(mgr, buffersize, maxbuffers,
1682 buckets, increment);
1683 if (result != ISC_R_SUCCESS)
1689 * First, see if we have a dispatcher that matches.
1692 result = dispatch_find(mgr, localaddr, attributes, mask, &disp);
1693 if (result == ISC_R_SUCCESS) {
1696 if (disp->maxrequests < maxrequests)
1697 disp->maxrequests = maxrequests;
1699 if ((disp->attributes & DNS_DISPATCHATTR_NOLISTEN) == 0 &&
1700 (attributes & DNS_DISPATCHATTR_NOLISTEN) != 0)
1702 disp->attributes |= DNS_DISPATCHATTR_NOLISTEN;
1703 if (disp->recv_pending != 0)
1704 isc_socket_cancel(disp->socket, disp->task,
1705 ISC_SOCKCANCEL_RECV);
1708 UNLOCK(&disp->lock);
1713 return (ISC_R_SUCCESS);
1719 result = dispatch_createudp(mgr, sockmgr, taskmgr, localaddr,
1720 maxrequests, attributes, &disp);
1721 if (result != ISC_R_SUCCESS) {
1728 return (ISC_R_SUCCESS);
1732 * mgr should be locked.
1735 #ifndef DNS_DISPATCH_HELD
1736 #define DNS_DISPATCH_HELD 20U
1740 dispatch_createudp(dns_dispatchmgr_t *mgr, isc_socketmgr_t *sockmgr,
1741 isc_taskmgr_t *taskmgr,
1742 isc_sockaddr_t *localaddr,
1743 unsigned int maxrequests,
1744 unsigned int attributes,
1745 dns_dispatch_t **dispp)
1747 isc_result_t result;
1748 dns_dispatch_t *disp;
1749 isc_socket_t *sock = NULL;
1750 isc_socket_t *held[DNS_DISPATCH_HELD];
1751 unsigned int i = 0, j = 0;
1754 * dispatch_allocate() checks mgr for us.
1757 result = dispatch_allocate(mgr, maxrequests, &disp);
1758 if (result != ISC_R_SUCCESS)
1762 * Try to allocate a socket that is not on the blacklist.
1763 * Hold up to DNS_DISPATCH_HELD sockets to prevent the OS
1764 * from returning the same port to us too quickly.
1766 memset(held, 0, sizeof(held));
1768 result = create_socket(sockmgr, localaddr, &sock);
1769 if (result != ISC_R_SUCCESS)
1770 goto deallocate_dispatch;
1771 if (isc_sockaddr_getport(localaddr) == 0 && blacklisted(mgr, sock)) {
1772 if (held[i] != NULL)
1773 isc_socket_detach(&held[i]);
1776 if (i == DNS_DISPATCH_HELD)
1778 if (j++ == 0xffffU) {
1779 mgr_log(mgr, ISC_LOG_ERROR, "avoid-v%s-udp-ports: "
1780 "unable to allocate a non-blacklisted port",
1781 isc_sockaddr_pf(localaddr) == AF_INET ?
1783 result = ISC_R_FAILURE;
1784 goto deallocate_dispatch;
1789 disp->socktype = isc_sockettype_udp;
1790 disp->socket = sock;
1791 disp->local = *localaddr;
1794 result = isc_task_create(taskmgr, 0, &disp->task);
1795 if (result != ISC_R_SUCCESS)
1798 disp->ctlevent = isc_event_allocate(mgr->mctx, disp,
1799 DNS_EVENT_DISPATCHCONTROL,
1801 sizeof(isc_event_t));
1802 if (disp->ctlevent == NULL)
1805 isc_task_setname(disp->task, "udpdispatch", disp);
1807 attributes &= ~DNS_DISPATCHATTR_TCP;
1808 attributes |= DNS_DISPATCHATTR_UDP;
1809 disp->attributes = attributes;
1812 * Append it to the dispatcher list.
1814 ISC_LIST_APPEND(mgr->list, disp, link);
1816 mgr_log(mgr, LVL(90), "created UDP dispatcher %p", disp);
1817 dispatch_log(disp, LVL(90), "created task %p", disp->task);
1818 dispatch_log(disp, LVL(90), "created socket %p", disp->socket);
1828 isc_task_detach(&disp->task);
1830 isc_socket_detach(&disp->socket);
1831 deallocate_dispatch:
1832 dispatch_free(&disp);
1834 for (i = 0; i < DNS_DISPATCH_HELD; i++)
1835 if (held[i] != NULL)
1836 isc_socket_detach(&held[i]);
1841 dns_dispatch_attach(dns_dispatch_t *disp, dns_dispatch_t **dispp) {
1842 REQUIRE(VALID_DISPATCH(disp));
1843 REQUIRE(dispp != NULL && *dispp == NULL);
1847 UNLOCK(&disp->lock);
1853 * It is important to lock the manager while we are deleting the dispatch,
1854 * since dns_dispatch_getudp will call dispatch_find, which returns to
1855 * the caller a dispatch but does not attach to it until later. _getudp
1856 * locks the manager, however, so locking it here will keep us from attaching
1857 * to a dispatcher that is in the process of going away.
1860 dns_dispatch_detach(dns_dispatch_t **dispp) {
1861 dns_dispatch_t *disp;
1862 isc_boolean_t killit;
1864 REQUIRE(dispp != NULL && VALID_DISPATCH(*dispp));
1871 INSIST(disp->refcount > 0);
1874 if (disp->refcount == 0) {
1875 if (disp->recv_pending > 0)
1876 isc_socket_cancel(disp->socket, disp->task,
1877 ISC_SOCKCANCEL_RECV);
1878 disp->shutting_down = 1;
1881 dispatch_log(disp, LVL(90), "detach: refcount %d", disp->refcount);
1883 killit = destroy_disp_ok(disp);
1884 UNLOCK(&disp->lock);
1886 isc_task_send(disp->task, &disp->ctlevent);
1890 dns_dispatch_addresponse(dns_dispatch_t *disp, isc_sockaddr_t *dest,
1891 isc_task_t *task, isc_taskaction_t action, void *arg,
1892 dns_messageid_t *idp, dns_dispentry_t **resp)
1894 dns_dispentry_t *res;
1895 unsigned int bucket;
1901 REQUIRE(VALID_DISPATCH(disp));
1902 REQUIRE(task != NULL);
1903 REQUIRE(dest != NULL);
1904 REQUIRE(resp != NULL && *resp == NULL);
1905 REQUIRE(idp != NULL);
1909 if (disp->shutting_down == 1) {
1910 UNLOCK(&disp->lock);
1911 return (ISC_R_SHUTTINGDOWN);
1914 if (disp->requests >= disp->maxrequests) {
1915 UNLOCK(&disp->lock);
1916 return (ISC_R_QUOTA);
1920 * Try somewhat hard to find an unique ID.
1922 qid = DNS_QID(disp);
1924 id = dns_randomid(qid);
1925 bucket = dns_hash(qid, dest, id);
1927 for (i = 0; i < 64; i++) {
1928 if (bucket_search(qid, dest, id, bucket) == NULL) {
1932 id += qid->qid_increment;
1934 bucket = dns_hash(qid, dest, id);
1939 UNLOCK(&disp->lock);
1940 return (ISC_R_NOMORE);
1943 res = isc_mempool_get(disp->mgr->rpool);
1946 UNLOCK(&disp->lock);
1947 return (ISC_R_NOMEMORY);
1953 isc_task_attach(task, &res->task);
1956 res->bucket = bucket;
1958 res->action = action;
1960 res->item_out = ISC_FALSE;
1961 ISC_LIST_INIT(res->items);
1962 ISC_LINK_INIT(res, link);
1963 res->magic = RESPONSE_MAGIC;
1964 ISC_LIST_APPEND(qid->qid_table[bucket], res, link);
1967 request_log(disp, res, LVL(90),
1968 "attached to task %p", res->task);
1970 if (((disp->attributes & DNS_DISPATCHATTR_UDP) != 0) ||
1971 ((disp->attributes & DNS_DISPATCHATTR_CONNECTED) != 0))
1974 UNLOCK(&disp->lock);
1979 return (ISC_R_SUCCESS);
1983 dns_dispatch_starttcp(dns_dispatch_t *disp) {
1985 REQUIRE(VALID_DISPATCH(disp));
1987 dispatch_log(disp, LVL(90), "starttcp %p", disp->task);
1990 disp->attributes |= DNS_DISPATCHATTR_CONNECTED;
1992 UNLOCK(&disp->lock);
1996 dns_dispatch_removeresponse(dns_dispentry_t **resp,
1997 dns_dispatchevent_t **sockevent)
1999 dns_dispatchmgr_t *mgr;
2000 dns_dispatch_t *disp;
2001 dns_dispentry_t *res;
2002 dns_dispatchevent_t *ev;
2003 unsigned int bucket;
2004 isc_boolean_t killit;
2006 isc_eventlist_t events;
2009 REQUIRE(resp != NULL);
2010 REQUIRE(VALID_RESPONSE(*resp));
2016 REQUIRE(VALID_DISPATCH(disp));
2018 REQUIRE(VALID_DISPATCHMGR(mgr));
2020 qid = DNS_QID(disp);
2022 if (sockevent != NULL) {
2023 REQUIRE(*sockevent != NULL);
2032 INSIST(disp->requests > 0);
2034 INSIST(disp->refcount > 0);
2037 if (disp->refcount == 0) {
2038 if (disp->recv_pending > 0)
2039 isc_socket_cancel(disp->socket, disp->task,
2040 ISC_SOCKCANCEL_RECV);
2041 disp->shutting_down = 1;
2044 bucket = res->bucket;
2047 ISC_LIST_UNLINK(qid->qid_table[bucket], res, link);
2050 if (ev == NULL && res->item_out) {
2052 * We've posted our event, but the caller hasn't gotten it
2053 * yet. Take it back.
2055 ISC_LIST_INIT(events);
2056 n = isc_task_unsend(res->task, res, DNS_EVENT_DISPATCH,
2059 * We had better have gotten it back.
2062 ev = (dns_dispatchevent_t *)ISC_LIST_HEAD(events);
2066 REQUIRE(res->item_out == ISC_TRUE);
2067 res->item_out = ISC_FALSE;
2068 if (ev->buffer.base != NULL)
2069 free_buffer(disp, ev->buffer.base, ev->buffer.length);
2070 free_event(disp, ev);
2073 request_log(disp, res, LVL(90), "detaching from task %p", res->task);
2074 isc_task_detach(&res->task);
2077 * Free any buffered requests as well
2079 ev = ISC_LIST_HEAD(res->items);
2080 while (ev != NULL) {
2081 ISC_LIST_UNLINK(res->items, ev, ev_link);
2082 if (ev->buffer.base != NULL)
2083 free_buffer(disp, ev->buffer.base, ev->buffer.length);
2084 free_event(disp, ev);
2085 ev = ISC_LIST_HEAD(res->items);
2088 isc_mempool_put(disp->mgr->rpool, res);
2089 if (disp->shutting_down == 1)
2094 killit = destroy_disp_ok(disp);
2095 UNLOCK(&disp->lock);
2097 isc_task_send(disp->task, &disp->ctlevent);
2101 do_cancel(dns_dispatch_t *disp) {
2102 dns_dispatchevent_t *ev;
2103 dns_dispentry_t *resp;
2106 if (disp->shutdown_out == 1)
2109 qid = DNS_QID(disp);
2112 * Search for the first response handler without packets outstanding.
2115 for (resp = linear_first(qid);
2116 resp != NULL && resp->item_out != ISC_FALSE;
2118 resp = linear_next(qid, resp);
2120 * No one to send the cancel event to, so nothing to do.
2126 * Send the shutdown failsafe event to this resp.
2128 ev = disp->failsafe_ev;
2129 ISC_EVENT_INIT(ev, sizeof(*ev), 0, NULL, DNS_EVENT_DISPATCH,
2130 resp->action, resp->arg, resp, NULL, NULL);
2131 ev->result = disp->shutdown_why;
2132 ev->buffer.base = NULL;
2133 ev->buffer.length = 0;
2134 disp->shutdown_out = 1;
2135 request_log(disp, resp, LVL(10),
2136 "cancel: failsafe event %p -> task %p",
2138 resp->item_out = ISC_TRUE;
2139 isc_task_send(resp->task, ISC_EVENT_PTR(&ev));
2145 dns_dispatch_getsocket(dns_dispatch_t *disp) {
2146 REQUIRE(VALID_DISPATCH(disp));
2148 return (disp->socket);
2152 dns_dispatch_getlocaladdress(dns_dispatch_t *disp, isc_sockaddr_t *addrp) {
2154 REQUIRE(VALID_DISPATCH(disp));
2155 REQUIRE(addrp != NULL);
2157 if (disp->socktype == isc_sockettype_udp) {
2158 *addrp = disp->local;
2159 return (ISC_R_SUCCESS);
2161 return (ISC_R_NOTIMPLEMENTED);
2165 dns_dispatch_cancel(dns_dispatch_t *disp) {
2166 REQUIRE(VALID_DISPATCH(disp));
2170 if (disp->shutting_down == 1) {
2171 UNLOCK(&disp->lock);
2175 disp->shutdown_why = ISC_R_CANCELED;
2176 disp->shutting_down = 1;
2179 UNLOCK(&disp->lock);
2185 dns_dispatch_changeattributes(dns_dispatch_t *disp,
2186 unsigned int attributes, unsigned int mask)
2188 REQUIRE(VALID_DISPATCH(disp));
2191 * Should check for valid attributes here!
2196 if ((mask & DNS_DISPATCHATTR_NOLISTEN) != 0) {
2197 if ((disp->attributes & DNS_DISPATCHATTR_NOLISTEN) != 0 &&
2198 (attributes & DNS_DISPATCHATTR_NOLISTEN) == 0) {
2199 disp->attributes &= ~DNS_DISPATCHATTR_NOLISTEN;
2201 } else if ((disp->attributes & DNS_DISPATCHATTR_NOLISTEN)
2203 (attributes & DNS_DISPATCHATTR_NOLISTEN) != 0) {
2204 disp->attributes |= DNS_DISPATCHATTR_NOLISTEN;
2205 if (disp->recv_pending != 0)
2206 isc_socket_cancel(disp->socket, disp->task,
2207 ISC_SOCKCANCEL_RECV);
2211 disp->attributes &= ~mask;
2212 disp->attributes |= (attributes & mask);
2213 UNLOCK(&disp->lock);
2217 dns_dispatch_importrecv(dns_dispatch_t *disp, isc_event_t *event) {
2219 isc_socketevent_t *sevent, *newsevent;
2221 REQUIRE(VALID_DISPATCH(disp));
2222 REQUIRE((disp->attributes & DNS_DISPATCHATTR_NOLISTEN) != 0);
2223 REQUIRE(event != NULL);
2225 sevent = (isc_socketevent_t *)event;
2227 INSIST(sevent->n <= disp->mgr->buffersize);
2228 newsevent = (isc_socketevent_t *)
2229 isc_event_allocate(disp->mgr->mctx, NULL,
2230 DNS_EVENT_IMPORTRECVDONE, udp_recv,
2231 disp, sizeof(isc_socketevent_t));
2232 if (newsevent == NULL)
2235 buf = allocate_udp_buffer(disp);
2237 isc_event_free(ISC_EVENT_PTR(&newsevent));
2240 memcpy(buf, sevent->region.base, sevent->n);
2241 newsevent->region.base = buf;
2242 newsevent->region.length = disp->mgr->buffersize;
2243 newsevent->n = sevent->n;
2244 newsevent->result = sevent->result;
2245 newsevent->address = sevent->address;
2246 newsevent->timestamp = sevent->timestamp;
2247 newsevent->pktinfo = sevent->pktinfo;
2248 newsevent->attributes = sevent->attributes;
2250 isc_task_send(disp->task, ISC_EVENT_PTR(&newsevent));
2255 dns_dispatchmgr_dump(dns_dispatchmgr_t *mgr) {
2256 dns_dispatch_t *disp;
2259 disp = ISC_LIST_HEAD(mgr->list);
2260 while (disp != NULL) {
2261 isc_sockaddr_format(&disp->local, foo, sizeof(foo));
2262 printf("\tdispatch %p, addr %s\n", disp, foo);
2263 disp = ISC_LIST_NEXT(disp, link);