2 * services/mesh.c - deal with mesh of query states and handle events for that.
4 * Copyright (c) 2007, NLnet Labs. All rights reserved.
6 * This software is open source.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
12 * Redistributions of source code must retain the above copyright notice,
13 * this list of conditions and the following disclaimer.
15 * Redistributions in binary form must reproduce the above copyright notice,
16 * this list of conditions and the following disclaimer in the documentation
17 * and/or other materials provided with the distribution.
19 * Neither the name of the NLNET LABS nor the names of its contributors may
20 * be used to endorse or promote products derived from this software without
21 * specific prior written permission.
23 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
26 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
27 * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
28 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
29 * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
30 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
31 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
32 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
39 * This file contains functions to assist in dealing with a mesh of
40 * query states. This mesh is supposed to be thread-specific.
41 * It consists of query states (per qname, qtype, qclass) and connections
42 * between query states and the super and subquery states, and replies to
43 * send back to clients.
46 #include "services/mesh.h"
47 #include "services/outbound_list.h"
48 #include "services/cache/dns.h"
50 #include "util/net_help.h"
51 #include "util/module.h"
52 #include "util/regional.h"
53 #include "util/data/msgencode.h"
54 #include "util/timehist.h"
55 #include "util/fptr_wlist.h"
56 #include "util/alloc.h"
57 #include "util/config_file.h"
58 #include "sldns/sbuffer.h"
60 /** subtract timers and the values do not overflow or become negative */
62 timeval_subtract(struct timeval* d, const struct timeval* end, const struct timeval* start)
65 time_t end_usec = end->tv_usec;
66 d->tv_sec = end->tv_sec - start->tv_sec;
67 if(end_usec < start->tv_usec) {
71 d->tv_usec = end_usec - start->tv_usec;
75 /** add timers and the values do not overflow or become negative */
77 timeval_add(struct timeval* d, const struct timeval* add)
80 d->tv_sec += add->tv_sec;
81 d->tv_usec += add->tv_usec;
82 if(d->tv_usec > 1000000 ) {
83 d->tv_usec -= 1000000;
89 /** divide sum of timers to get average */
91 timeval_divide(struct timeval* avg, const struct timeval* sum, size_t d)
100 avg->tv_sec = sum->tv_sec / d;
101 avg->tv_usec = sum->tv_usec / d;
102 /* handle fraction from seconds divide */
103 leftover = sum->tv_sec - avg->tv_sec*d;
104 avg->tv_usec += (leftover*1000000)/d;
108 /** histogram compare of time values */
110 timeval_smaller(const struct timeval* x, const struct timeval* y)
113 if(x->tv_sec < y->tv_sec)
115 else if(x->tv_sec == y->tv_sec) {
116 if(x->tv_usec <= y->tv_usec)
125 mesh_state_compare(const void* ap, const void* bp)
127 struct mesh_state* a = (struct mesh_state*)ap;
128 struct mesh_state* b = (struct mesh_state*)bp;
130 if(a->s.is_priming && !b->s.is_priming)
132 if(!a->s.is_priming && b->s.is_priming)
135 if(a->s.is_valrec && !b->s.is_valrec)
137 if(!a->s.is_valrec && b->s.is_valrec)
140 if((a->s.query_flags&BIT_RD) && !(b->s.query_flags&BIT_RD))
142 if(!(a->s.query_flags&BIT_RD) && (b->s.query_flags&BIT_RD))
145 if((a->s.query_flags&BIT_CD) && !(b->s.query_flags&BIT_CD))
147 if(!(a->s.query_flags&BIT_CD) && (b->s.query_flags&BIT_CD))
150 return query_info_compare(&a->s.qinfo, &b->s.qinfo);
154 mesh_state_ref_compare(const void* ap, const void* bp)
156 struct mesh_state_ref* a = (struct mesh_state_ref*)ap;
157 struct mesh_state_ref* b = (struct mesh_state_ref*)bp;
158 return mesh_state_compare(a->s, b->s);
162 mesh_create(struct module_stack* stack, struct module_env* env)
164 struct mesh_area* mesh = calloc(1, sizeof(struct mesh_area));
166 log_err("mesh area alloc: out of memory");
169 mesh->histogram = timehist_setup();
170 mesh->qbuf_bak = sldns_buffer_new(env->cfg->msg_buffer_size);
171 if(!mesh->histogram || !mesh->qbuf_bak) {
173 log_err("mesh area alloc: out of memory");
178 rbtree_init(&mesh->run, &mesh_state_compare);
179 rbtree_init(&mesh->all, &mesh_state_compare);
180 mesh->num_reply_addrs = 0;
181 mesh->num_reply_states = 0;
182 mesh->num_detached_states = 0;
183 mesh->num_forever_states = 0;
184 mesh->stats_jostled = 0;
185 mesh->stats_dropped = 0;
186 mesh->max_reply_states = env->cfg->num_queries_per_thread;
187 mesh->max_forever_states = (mesh->max_reply_states+1)/2;
189 mesh->jostle_max.tv_sec = (time_t)(env->cfg->jostle_time / 1000);
190 mesh->jostle_max.tv_usec = (time_t)((env->cfg->jostle_time % 1000)
196 /** help mesh delete delete mesh states */
198 mesh_delete_helper(rbnode_t* n)
200 struct mesh_state* mstate = (struct mesh_state*)n->key;
201 /* perform a full delete, not only 'cleanup' routine,
202 * because other callbacks expect a clean state in the mesh.
203 * For 're-entrant' calls */
204 mesh_state_delete(&mstate->s);
205 /* but because these delete the items from the tree, postorder
206 * traversal and rbtree rebalancing do not work together */
210 mesh_delete(struct mesh_area* mesh)
214 /* free all query states */
215 while(mesh->all.count)
216 mesh_delete_helper(mesh->all.root);
217 timehist_delete(mesh->histogram);
218 sldns_buffer_free(mesh->qbuf_bak);
223 mesh_delete_all(struct mesh_area* mesh)
225 /* free all query states */
226 while(mesh->all.count)
227 mesh_delete_helper(mesh->all.root);
228 mesh->stats_dropped += mesh->num_reply_addrs;
229 /* clear mesh area references */
230 rbtree_init(&mesh->run, &mesh_state_compare);
231 rbtree_init(&mesh->all, &mesh_state_compare);
232 mesh->num_reply_addrs = 0;
233 mesh->num_reply_states = 0;
234 mesh->num_detached_states = 0;
235 mesh->num_forever_states = 0;
236 mesh->forever_first = NULL;
237 mesh->forever_last = NULL;
238 mesh->jostle_first = NULL;
239 mesh->jostle_last = NULL;
242 int mesh_make_new_space(struct mesh_area* mesh, sldns_buffer* qbuf)
244 struct mesh_state* m = mesh->jostle_first;
245 /* free space is available */
246 if(mesh->num_reply_states < mesh->max_reply_states)
248 /* try to kick out a jostle-list item */
249 if(m && m->reply_list && m->list_select == mesh_jostle_list) {
252 timeval_subtract(&age, mesh->env->now_tv,
253 &m->reply_list->start_time);
254 if(timeval_smaller(&mesh->jostle_max, &age)) {
256 log_nametypeclass(VERB_ALGO, "query jostled out to "
257 "make space for a new one",
258 m->s.qinfo.qname, m->s.qinfo.qtype,
260 /* backup the query */
261 if(qbuf) sldns_buffer_copy(mesh->qbuf_bak, qbuf);
263 if(m->super_set.count > 0) {
264 verbose(VERB_ALGO, "notify supers of failure");
265 m->s.return_msg = NULL;
266 m->s.return_rcode = LDNS_RCODE_SERVFAIL;
267 mesh_walk_supers(mesh, m);
269 mesh->stats_jostled ++;
270 mesh_state_delete(&m->s);
271 /* restore the query - note that the qinfo ptr to
272 * the querybuffer is then correct again. */
273 if(qbuf) sldns_buffer_copy(qbuf, mesh->qbuf_bak);
277 /* no space for new item */
281 void mesh_new_client(struct mesh_area* mesh, struct query_info* qinfo,
282 uint16_t qflags, struct edns_data* edns, struct comm_reply* rep,
285 struct mesh_state* s = mesh_area_find(mesh, qinfo, qflags&(BIT_RD|BIT_CD), 0, 0);
286 int was_detached = 0;
289 /* does this create a new reply state? */
290 if(!s || s->list_select == mesh_no_list) {
291 if(!mesh_make_new_space(mesh, rep->c->buffer)) {
292 verbose(VERB_ALGO, "Too many queries. dropping "
294 comm_point_drop_reply(rep);
295 mesh->stats_dropped ++;
298 /* for this new reply state, the reply address is free,
299 * so the limit of reply addresses does not stop reply states*/
301 /* protect our memory usage from storing reply addresses */
302 if(mesh->num_reply_addrs > mesh->max_reply_states*16) {
303 verbose(VERB_ALGO, "Too many requests queued. "
304 "dropping incoming query.");
305 mesh->stats_dropped++;
306 comm_point_drop_reply(rep);
310 /* see if it already exists, if not, create one */
315 s = mesh_state_create(mesh->env, qinfo, qflags&(BIT_RD|BIT_CD), 0, 0);
317 log_err("mesh_state_create: out of memory; SERVFAIL");
318 if(!edns_opt_inplace_reply(edns, mesh->env->scratch))
319 edns->opt_list = NULL;
320 error_encode(rep->c->buffer, LDNS_RCODE_SERVFAIL,
321 qinfo, qid, qflags, edns);
322 comm_point_send_reply(rep);
330 rbtree_insert(&mesh->all, &s->node);
331 log_assert(n != NULL);
332 /* set detached (it is now) */
333 mesh->num_detached_states++;
336 if(!s->reply_list && !s->cb_list && s->super_set.count == 0)
338 if(!s->reply_list && !s->cb_list)
341 if(!mesh_state_add_reply(s, edns, rep, qid, qflags, qinfo->qname)) {
342 log_err("mesh_new_client: out of memory; SERVFAIL");
343 if(!edns_opt_inplace_reply(edns, mesh->env->scratch))
344 edns->opt_list = NULL;
345 error_encode(rep->c->buffer, LDNS_RCODE_SERVFAIL,
346 qinfo, qid, qflags, edns);
347 comm_point_send_reply(rep);
349 mesh_state_delete(&s->s);
352 /* update statistics */
354 log_assert(mesh->num_detached_states > 0);
355 mesh->num_detached_states--;
358 mesh->num_reply_states ++;
360 mesh->num_reply_addrs++;
361 if(s->list_select == mesh_no_list) {
362 /* move to either the forever or the jostle_list */
363 if(mesh->num_forever_states < mesh->max_forever_states) {
364 mesh->num_forever_states ++;
365 mesh_list_insert(s, &mesh->forever_first,
366 &mesh->forever_last);
367 s->list_select = mesh_forever_list;
369 mesh_list_insert(s, &mesh->jostle_first,
371 s->list_select = mesh_jostle_list;
375 mesh_run(mesh, s, module_event_new, NULL);
379 mesh_new_callback(struct mesh_area* mesh, struct query_info* qinfo,
380 uint16_t qflags, struct edns_data* edns, sldns_buffer* buf,
381 uint16_t qid, mesh_cb_func_t cb, void* cb_arg)
383 struct mesh_state* s = mesh_area_find(mesh, qinfo, qflags&(BIT_RD|BIT_CD), 0, 0);
384 int was_detached = 0;
387 /* there are no limits on the number of callbacks */
389 /* see if it already exists, if not, create one */
394 s = mesh_state_create(mesh->env, qinfo, qflags&(BIT_RD|BIT_CD), 0, 0);
403 rbtree_insert(&mesh->all, &s->node);
404 log_assert(n != NULL);
405 /* set detached (it is now) */
406 mesh->num_detached_states++;
409 if(!s->reply_list && !s->cb_list && s->super_set.count == 0)
411 if(!s->reply_list && !s->cb_list)
414 if(!mesh_state_add_cb(s, edns, buf, cb, cb_arg, qid, qflags)) {
416 mesh_state_delete(&s->s);
419 /* update statistics */
421 log_assert(mesh->num_detached_states > 0);
422 mesh->num_detached_states--;
425 mesh->num_reply_states ++;
427 mesh->num_reply_addrs++;
429 mesh_run(mesh, s, module_event_new, NULL);
433 void mesh_new_prefetch(struct mesh_area* mesh, struct query_info* qinfo,
434 uint16_t qflags, time_t leeway)
436 struct mesh_state* s = mesh_area_find(mesh, qinfo, qflags&(BIT_RD|BIT_CD), 0, 0);
440 /* already exists, and for a different purpose perhaps.
441 * if mesh_no_list, keep it that way. */
443 /* make it ignore the cache from now on */
445 sock_list_insert(&s->s.blacklist, NULL, 0, s->s.region);
446 if(s->s.prefetch_leeway < leeway)
447 s->s.prefetch_leeway = leeway;
450 if(!mesh_make_new_space(mesh, NULL)) {
451 verbose(VERB_ALGO, "Too many queries. dropped prefetch.");
452 mesh->stats_dropped ++;
455 s = mesh_state_create(mesh->env, qinfo, qflags&(BIT_RD|BIT_CD), 0, 0);
457 log_err("prefetch mesh_state_create: out of memory");
465 rbtree_insert(&mesh->all, &s->node);
466 log_assert(n != NULL);
467 /* set detached (it is now) */
468 mesh->num_detached_states++;
469 /* make it ignore the cache */
470 sock_list_insert(&s->s.blacklist, NULL, 0, s->s.region);
471 s->s.prefetch_leeway = leeway;
473 if(s->list_select == mesh_no_list) {
474 /* move to either the forever or the jostle_list */
475 if(mesh->num_forever_states < mesh->max_forever_states) {
476 mesh->num_forever_states ++;
477 mesh_list_insert(s, &mesh->forever_first,
478 &mesh->forever_last);
479 s->list_select = mesh_forever_list;
481 mesh_list_insert(s, &mesh->jostle_first,
483 s->list_select = mesh_jostle_list;
486 mesh_run(mesh, s, module_event_new, NULL);
489 void mesh_report_reply(struct mesh_area* mesh, struct outbound_entry* e,
490 struct comm_reply* reply, int what)
492 enum module_ev event = module_event_reply;
493 e->qstate->reply = reply;
494 if(what != NETEVENT_NOERROR) {
495 event = module_event_noreply;
496 if(what == NETEVENT_CAPSFAIL)
497 event = module_event_capsfail;
499 mesh_run(mesh, e->qstate->mesh_info, event, e);
503 mesh_state_create(struct module_env* env, struct query_info* qinfo,
504 uint16_t qflags, int prime, int valrec)
506 struct regional* region = alloc_reg_obtain(env->alloc);
507 struct mesh_state* mstate;
511 mstate = (struct mesh_state*)regional_alloc(region,
512 sizeof(struct mesh_state));
514 alloc_reg_release(env->alloc, region);
517 memset(mstate, 0, sizeof(*mstate));
518 mstate->node = *RBTREE_NULL;
519 mstate->run_node = *RBTREE_NULL;
520 mstate->node.key = mstate;
521 mstate->run_node.key = mstate;
522 mstate->reply_list = NULL;
523 mstate->list_select = mesh_no_list;
524 mstate->replies_sent = 0;
525 rbtree_init(&mstate->super_set, &mesh_state_ref_compare);
526 rbtree_init(&mstate->sub_set, &mesh_state_ref_compare);
527 mstate->num_activated = 0;
528 /* init module qstate */
529 mstate->s.qinfo.qtype = qinfo->qtype;
530 mstate->s.qinfo.qclass = qinfo->qclass;
531 mstate->s.qinfo.qname_len = qinfo->qname_len;
532 mstate->s.qinfo.qname = regional_alloc_init(region, qinfo->qname,
534 if(!mstate->s.qinfo.qname) {
535 alloc_reg_release(env->alloc, region);
538 /* remove all weird bits from qflags */
539 mstate->s.query_flags = (qflags & (BIT_RD|BIT_CD));
540 mstate->s.is_priming = prime;
541 mstate->s.is_valrec = valrec;
542 mstate->s.reply = NULL;
543 mstate->s.region = region;
544 mstate->s.curmod = 0;
545 mstate->s.return_msg = 0;
546 mstate->s.return_rcode = LDNS_RCODE_NOERROR;
548 mstate->s.mesh_info = mstate;
549 mstate->s.prefetch_leeway = 0;
551 for(i=0; i<env->mesh->mods.num; i++) {
552 mstate->s.minfo[i] = NULL;
553 mstate->s.ext_state[i] = module_state_initial;
559 mesh_state_cleanup(struct mesh_state* mstate)
561 struct mesh_area* mesh;
565 mesh = mstate->s.env->mesh;
566 /* drop unsent replies */
567 if(!mstate->replies_sent) {
568 struct mesh_reply* rep;
570 for(rep=mstate->reply_list; rep; rep=rep->next) {
571 comm_point_drop_reply(&rep->query_reply);
572 mesh->num_reply_addrs--;
574 for(cb=mstate->cb_list; cb; cb=cb->next) {
575 fptr_ok(fptr_whitelist_mesh_cb(cb->cb));
576 (*cb->cb)(cb->cb_arg, LDNS_RCODE_SERVFAIL, NULL,
577 sec_status_unchecked, NULL);
578 mesh->num_reply_addrs--;
582 /* de-init modules */
583 for(i=0; i<mesh->mods.num; i++) {
584 fptr_ok(fptr_whitelist_mod_clear(mesh->mods.mod[i]->clear));
585 (*mesh->mods.mod[i]->clear)(&mstate->s, i);
586 mstate->s.minfo[i] = NULL;
587 mstate->s.ext_state[i] = module_finished;
589 alloc_reg_release(mstate->s.env->alloc, mstate->s.region);
593 mesh_state_delete(struct module_qstate* qstate)
595 struct mesh_area* mesh;
596 struct mesh_state_ref* super, ref;
597 struct mesh_state* mstate;
600 mstate = qstate->mesh_info;
601 mesh = mstate->s.env->mesh;
602 mesh_detach_subs(&mstate->s);
603 if(mstate->list_select == mesh_forever_list) {
604 mesh->num_forever_states --;
605 mesh_list_remove(mstate, &mesh->forever_first,
606 &mesh->forever_last);
607 } else if(mstate->list_select == mesh_jostle_list) {
608 mesh_list_remove(mstate, &mesh->jostle_first,
611 if(!mstate->reply_list && !mstate->cb_list
612 && mstate->super_set.count == 0) {
613 log_assert(mesh->num_detached_states > 0);
614 mesh->num_detached_states--;
616 if(mstate->reply_list || mstate->cb_list) {
617 log_assert(mesh->num_reply_states > 0);
618 mesh->num_reply_states--;
622 RBTREE_FOR(super, struct mesh_state_ref*, &mstate->super_set) {
623 (void)rbtree_delete(&super->s->sub_set, &ref);
625 (void)rbtree_delete(&mesh->run, mstate);
626 (void)rbtree_delete(&mesh->all, mstate);
627 mesh_state_cleanup(mstate);
630 /** helper recursive rbtree find routine */
632 find_in_subsub(struct mesh_state* m, struct mesh_state* tofind, size_t *c)
634 struct mesh_state_ref* r;
635 if((*c)++ > MESH_MAX_SUBSUB)
637 RBTREE_FOR(r, struct mesh_state_ref*, &m->sub_set) {
638 if(r->s == tofind || find_in_subsub(r->s, tofind, c))
644 /** find cycle for already looked up mesh_state */
646 mesh_detect_cycle_found(struct module_qstate* qstate, struct mesh_state* dep_m)
648 struct mesh_state* cyc_m = qstate->mesh_info;
652 if(dep_m == cyc_m || find_in_subsub(dep_m, cyc_m, &counter)) {
653 if(counter > MESH_MAX_SUBSUB)
660 void mesh_detach_subs(struct module_qstate* qstate)
662 struct mesh_area* mesh = qstate->env->mesh;
663 struct mesh_state_ref* ref, lookup;
667 lookup.node.key = &lookup;
668 lookup.s = qstate->mesh_info;
669 RBTREE_FOR(ref, struct mesh_state_ref*, &qstate->mesh_info->sub_set) {
675 rbtree_delete(&ref->s->super_set, &lookup);
676 log_assert(n != NULL); /* must have been present */
677 if(!ref->s->reply_list && !ref->s->cb_list
678 && ref->s->super_set.count == 0) {
679 mesh->num_detached_states++;
680 log_assert(mesh->num_detached_states +
681 mesh->num_reply_states <= mesh->all.count);
684 rbtree_init(&qstate->mesh_info->sub_set, &mesh_state_ref_compare);
687 int mesh_attach_sub(struct module_qstate* qstate, struct query_info* qinfo,
688 uint16_t qflags, int prime, int valrec, struct module_qstate** newq)
690 /* find it, if not, create it */
691 struct mesh_area* mesh = qstate->env->mesh;
692 struct mesh_state* sub = mesh_area_find(mesh, qinfo, qflags, prime,
695 if(mesh_detect_cycle_found(qstate, sub)) {
696 verbose(VERB_ALGO, "attach failed, cycle detected");
703 /* create a new one */
704 sub = mesh_state_create(qstate->env, qinfo, qflags, prime,
707 log_err("mesh_attach_sub: out of memory");
715 rbtree_insert(&mesh->all, &sub->node);
716 log_assert(n != NULL);
717 /* set detached (it is now) */
718 mesh->num_detached_states++;
719 /* set new query state to run */
725 rbtree_insert(&mesh->run, &sub->run_node);
726 log_assert(n != NULL);
730 was_detached = (sub->super_set.count == 0);
731 if(!mesh_state_attachment(qstate->mesh_info, sub))
733 /* if it was a duplicate attachment, the count was not zero before */
734 if(!sub->reply_list && !sub->cb_list && was_detached &&
735 sub->super_set.count == 1) {
736 /* it used to be detached, before this one got added */
737 log_assert(mesh->num_detached_states > 0);
738 mesh->num_detached_states--;
740 /* *newq will be run when inited after the current module stops */
744 int mesh_state_attachment(struct mesh_state* super, struct mesh_state* sub)
749 struct mesh_state_ref* subref; /* points to sub, inserted in super */
750 struct mesh_state_ref* superref; /* points to super, inserted in sub */
751 if( !(subref = regional_alloc(super->s.region,
752 sizeof(struct mesh_state_ref))) ||
753 !(superref = regional_alloc(sub->s.region,
754 sizeof(struct mesh_state_ref))) ) {
755 log_err("mesh_state_attachment: out of memory");
758 superref->node.key = superref;
760 subref->node.key = subref;
762 if(!rbtree_insert(&sub->super_set, &superref->node)) {
763 /* this should not happen, iterator and validator do not
764 * attach subqueries that are identical. */
765 /* already attached, we are done, nothing todo.
766 * since superref and subref already allocated in region,
767 * we cannot free them */
775 rbtree_insert(&super->sub_set, &subref->node);
776 log_assert(n != NULL); /* we checked above if statement, the reverse
777 administration should not fail now, unless they are out of sync */
782 * callback results to mesh cb entry
783 * @param m: mesh state to send it for.
784 * @param rcode: if not 0, error code.
785 * @param rep: reply to send (or NULL if rcode is set).
786 * @param r: callback entry
789 mesh_do_callback(struct mesh_state* m, int rcode, struct reply_info* rep,
794 /* bogus messages are not made into servfail, sec_status passed
795 * to the callback function */
796 if(rep && rep->security == sec_status_secure)
799 if(!rep && rcode == LDNS_RCODE_NOERROR)
800 rcode = LDNS_RCODE_SERVFAIL;
801 if(!rcode && rep->security == sec_status_bogus) {
802 if(!(reason = errinf_to_str(&m->s)))
803 rcode = LDNS_RCODE_SERVFAIL;
807 fptr_ok(fptr_whitelist_mesh_cb(r->cb));
808 (*r->cb)(r->cb_arg, rcode, r->buf, sec_status_unchecked, NULL);
810 size_t udp_size = r->edns.udp_size;
811 sldns_buffer_clear(r->buf);
812 r->edns.edns_version = EDNS_ADVERTISED_VERSION;
813 r->edns.udp_size = EDNS_ADVERTISED_SIZE;
814 r->edns.ext_rcode = 0;
815 r->edns.bits &= EDNS_DO;
816 if(!edns_opt_inplace_reply(&r->edns, m->s.region) ||
817 !reply_info_answer_encode(&m->s.qinfo, rep, r->qid,
818 r->qflags, r->buf, 0, 1,
819 m->s.env->scratch, udp_size, &r->edns,
820 (int)(r->edns.bits & EDNS_DO), secure))
822 fptr_ok(fptr_whitelist_mesh_cb(r->cb));
823 (*r->cb)(r->cb_arg, LDNS_RCODE_SERVFAIL, r->buf,
824 sec_status_unchecked, NULL);
826 fptr_ok(fptr_whitelist_mesh_cb(r->cb));
827 (*r->cb)(r->cb_arg, LDNS_RCODE_NOERROR, r->buf,
828 rep->security, reason);
832 m->s.env->mesh->num_reply_addrs--;
836 * Send reply to mesh reply entry
837 * @param m: mesh state to send it for.
838 * @param rcode: if not 0, error code.
839 * @param rep: reply to send (or NULL if rcode is set).
840 * @param r: reply entry
841 * @param prev: previous reply, already has its answer encoded in buffer.
844 mesh_send_reply(struct mesh_state* m, int rcode, struct reply_info* rep,
845 struct mesh_reply* r, struct mesh_reply* prev)
847 struct timeval end_time;
848 struct timeval duration;
850 /* examine security status */
851 if(m->s.env->need_to_validate && (!(r->qflags&BIT_CD) ||
852 m->s.env->cfg->ignore_cd) && rep &&
853 rep->security <= sec_status_bogus) {
854 rcode = LDNS_RCODE_SERVFAIL;
855 if(m->s.env->cfg->stat_extended)
856 m->s.env->mesh->ans_bogus++;
858 if(rep && rep->security == sec_status_secure)
861 if(!rep && rcode == LDNS_RCODE_NOERROR)
862 rcode = LDNS_RCODE_SERVFAIL;
864 if(prev && prev->qflags == r->qflags &&
865 prev->edns.edns_present == r->edns.edns_present &&
866 prev->edns.bits == r->edns.bits &&
867 prev->edns.udp_size == r->edns.udp_size &&
868 edns_opt_list_compare(prev->edns.opt_list, r->edns.opt_list)
870 /* if the previous reply is identical to this one, fix ID */
871 if(prev->query_reply.c->buffer != r->query_reply.c->buffer)
872 sldns_buffer_copy(r->query_reply.c->buffer,
873 prev->query_reply.c->buffer);
874 sldns_buffer_write_at(r->query_reply.c->buffer, 0,
875 &r->qid, sizeof(uint16_t));
876 sldns_buffer_write_at(r->query_reply.c->buffer, 12,
877 r->qname, m->s.qinfo.qname_len);
878 comm_point_send_reply(&r->query_reply);
880 m->s.qinfo.qname = r->qname;
881 error_encode(r->query_reply.c->buffer, rcode, &m->s.qinfo,
882 r->qid, r->qflags, &r->edns);
883 comm_point_send_reply(&r->query_reply);
885 size_t udp_size = r->edns.udp_size;
886 r->edns.edns_version = EDNS_ADVERTISED_VERSION;
887 r->edns.udp_size = EDNS_ADVERTISED_SIZE;
888 r->edns.ext_rcode = 0;
889 r->edns.bits &= EDNS_DO;
890 m->s.qinfo.qname = r->qname;
891 if(!edns_opt_inplace_reply(&r->edns, m->s.region) ||
892 !reply_info_answer_encode(&m->s.qinfo, rep, r->qid,
893 r->qflags, r->query_reply.c->buffer, 0, 1,
894 m->s.env->scratch, udp_size, &r->edns,
895 (int)(r->edns.bits & EDNS_DO), secure))
897 error_encode(r->query_reply.c->buffer,
898 LDNS_RCODE_SERVFAIL, &m->s.qinfo, r->qid,
899 r->qflags, &r->edns);
901 comm_point_send_reply(&r->query_reply);
904 m->s.env->mesh->num_reply_addrs--;
905 end_time = *m->s.env->now_tv;
906 timeval_subtract(&duration, &end_time, &r->start_time);
907 verbose(VERB_ALGO, "query took " ARG_LL "d.%6.6d sec",
908 (long long)duration.tv_sec, (int)duration.tv_usec);
909 m->s.env->mesh->replies_sent++;
910 timeval_add(&m->s.env->mesh->replies_sum_wait, &duration);
911 timehist_insert(m->s.env->mesh->histogram, &duration);
912 if(m->s.env->cfg->stat_extended) {
913 uint16_t rc = FLAGS_GET_RCODE(sldns_buffer_read_u16_at(r->
914 query_reply.c->buffer, 2));
915 if(secure) m->s.env->mesh->ans_secure++;
916 m->s.env->mesh->ans_rcode[ rc ] ++;
917 if(rc == 0 && LDNS_ANCOUNT(sldns_buffer_begin(r->
918 query_reply.c->buffer)) == 0)
919 m->s.env->mesh->ans_nodata++;
923 void mesh_query_done(struct mesh_state* mstate)
925 struct mesh_reply* r;
926 struct mesh_reply* prev = NULL;
928 struct reply_info* rep = (mstate->s.return_msg?
929 mstate->s.return_msg->rep:NULL);
930 for(r = mstate->reply_list; r; r = r->next) {
931 mesh_send_reply(mstate, mstate->s.return_rcode, rep, r, prev);
934 mstate->replies_sent = 1;
935 for(c = mstate->cb_list; c; c = c->next) {
936 mesh_do_callback(mstate, mstate->s.return_rcode, rep, c);
940 void mesh_walk_supers(struct mesh_area* mesh, struct mesh_state* mstate)
942 struct mesh_state_ref* ref;
943 RBTREE_FOR(ref, struct mesh_state_ref*, &mstate->super_set)
945 /* make super runnable */
946 (void)rbtree_insert(&mesh->run, &ref->s->run_node);
947 /* callback the function to inform super of result */
948 fptr_ok(fptr_whitelist_mod_inform_super(
949 mesh->mods.mod[ref->s->s.curmod]->inform_super));
950 (*mesh->mods.mod[ref->s->s.curmod]->inform_super)(&mstate->s,
951 ref->s->s.curmod, &ref->s->s);
955 struct mesh_state* mesh_area_find(struct mesh_area* mesh,
956 struct query_info* qinfo, uint16_t qflags, int prime, int valrec)
958 struct mesh_state key;
959 struct mesh_state* result;
962 key.s.is_priming = prime;
963 key.s.is_valrec = valrec;
964 key.s.qinfo = *qinfo;
965 key.s.query_flags = qflags;
967 result = (struct mesh_state*)rbtree_search(&mesh->all, &key);
971 int mesh_state_add_cb(struct mesh_state* s, struct edns_data* edns,
972 sldns_buffer* buf, mesh_cb_func_t cb, void* cb_arg,
973 uint16_t qid, uint16_t qflags)
975 struct mesh_cb* r = regional_alloc(s->s.region,
976 sizeof(struct mesh_cb));
980 log_assert(fptr_whitelist_mesh_cb(cb)); /* early failure ifmissing*/
985 r->edns.opt_list = edns_opt_copy_region(edns->opt_list,
987 if(!r->edns.opt_list)
992 r->next = s->cb_list;
998 int mesh_state_add_reply(struct mesh_state* s, struct edns_data* edns,
999 struct comm_reply* rep, uint16_t qid, uint16_t qflags, uint8_t* qname)
1001 struct mesh_reply* r = regional_alloc(s->s.region,
1002 sizeof(struct mesh_reply));
1005 r->query_reply = *rep;
1007 if(edns->opt_list) {
1008 r->edns.opt_list = edns_opt_copy_region(edns->opt_list,
1010 if(!r->edns.opt_list)
1015 r->start_time = *s->s.env->now_tv;
1016 r->next = s->reply_list;
1017 r->qname = regional_alloc_init(s->s.region, qname,
1018 s->s.qinfo.qname_len);
1026 * Continue processing the mesh state at another module.
1027 * Handles module to modules tranfer of control.
1028 * Handles module finished.
1029 * @param mesh: the mesh area.
1030 * @param mstate: currently active mesh state.
1031 * Deleted if finished, calls _done and _supers to
1032 * send replies to clients and inform other mesh states.
1033 * This in turn may create additional runnable mesh states.
1034 * @param s: state at which the current module exited.
1035 * @param ev: the event sent to the module.
1036 * returned is the event to send to the next module.
1037 * @return true if continue processing at the new module.
1038 * false if not continued processing is needed.
1041 mesh_continue(struct mesh_area* mesh, struct mesh_state* mstate,
1042 enum module_ext_state s, enum module_ev* ev)
1044 mstate->num_activated++;
1045 if(mstate->num_activated > MESH_MAX_ACTIVATION) {
1046 /* module is looping. Stop it. */
1047 log_err("internal error: looping module stopped");
1048 log_query_info(VERB_QUERY, "pass error for qstate",
1052 if(s == module_wait_module || s == module_restart_next) {
1053 /* start next module */
1055 if(mesh->mods.num == mstate->s.curmod) {
1056 log_err("Cannot pass to next module; at last module");
1057 log_query_info(VERB_QUERY, "pass error for qstate",
1060 return mesh_continue(mesh, mstate, module_error, ev);
1062 if(s == module_restart_next) {
1063 fptr_ok(fptr_whitelist_mod_clear(
1064 mesh->mods.mod[mstate->s.curmod]->clear));
1065 (*mesh->mods.mod[mstate->s.curmod]->clear)
1066 (&mstate->s, mstate->s.curmod);
1067 mstate->s.minfo[mstate->s.curmod] = NULL;
1069 *ev = module_event_pass;
1072 if(s == module_wait_subquery && mstate->sub_set.count == 0) {
1073 log_err("module cannot wait for subquery, subquery list empty");
1074 log_query_info(VERB_QUERY, "pass error for qstate",
1078 if(s == module_error && mstate->s.return_rcode == LDNS_RCODE_NOERROR) {
1079 /* error is bad, handle pass back up below */
1080 mstate->s.return_rcode = LDNS_RCODE_SERVFAIL;
1082 if(s == module_error || s == module_finished) {
1083 if(mstate->s.curmod == 0) {
1084 mesh_query_done(mstate);
1085 mesh_walk_supers(mesh, mstate);
1086 mesh_state_delete(&mstate->s);
1089 /* pass along the locus of control */
1090 mstate->s.curmod --;
1091 *ev = module_event_moddone;
1097 void mesh_run(struct mesh_area* mesh, struct mesh_state* mstate,
1098 enum module_ev ev, struct outbound_entry* e)
1100 enum module_ext_state s;
1101 verbose(VERB_ALGO, "mesh_run: start");
1103 /* run the module */
1104 fptr_ok(fptr_whitelist_mod_operate(
1105 mesh->mods.mod[mstate->s.curmod]->operate));
1106 (*mesh->mods.mod[mstate->s.curmod]->operate)
1107 (&mstate->s, ev, mstate->s.curmod, e);
1109 /* examine results */
1110 mstate->s.reply = NULL;
1111 regional_free_all(mstate->s.env->scratch);
1112 s = mstate->s.ext_state[mstate->s.curmod];
1113 verbose(VERB_ALGO, "mesh_run: %s module exit state is %s",
1114 mesh->mods.mod[mstate->s.curmod]->name, strextstate(s));
1116 if(mesh_continue(mesh, mstate, s, &ev))
1119 /* run more modules */
1120 ev = module_event_pass;
1121 if(mesh->run.count > 0) {
1122 /* pop random element off the runnable tree */
1123 mstate = (struct mesh_state*)mesh->run.root->key;
1124 (void)rbtree_delete(&mesh->run, mstate);
1125 } else mstate = NULL;
1127 if(verbosity >= VERB_ALGO) {
1128 mesh_stats(mesh, "mesh_run: end");
1129 mesh_log_list(mesh);
1134 mesh_log_list(struct mesh_area* mesh)
1137 struct mesh_state* m;
1139 RBTREE_FOR(m, struct mesh_state*, &mesh->all) {
1140 snprintf(buf, sizeof(buf), "%d%s%s%s%s%s%s mod%d %s%s",
1141 num++, (m->s.is_priming)?"p":"", /* prime */
1142 (m->s.is_valrec)?"v":"", /* prime */
1143 (m->s.query_flags&BIT_RD)?"RD":"",
1144 (m->s.query_flags&BIT_CD)?"CD":"",
1145 (m->super_set.count==0)?"d":"", /* detached */
1146 (m->sub_set.count!=0)?"c":"", /* children */
1147 m->s.curmod, (m->reply_list)?"rep":"", /*hasreply*/
1148 (m->cb_list)?"cb":"" /* callbacks */
1150 log_query_info(VERB_ALGO, buf, &m->s.qinfo);
1155 mesh_stats(struct mesh_area* mesh, const char* str)
1157 verbose(VERB_DETAIL, "%s %u recursion states (%u with reply, "
1158 "%u detached), %u waiting replies, %u recursion replies "
1159 "sent, %d replies dropped, %d states jostled out",
1160 str, (unsigned)mesh->all.count,
1161 (unsigned)mesh->num_reply_states,
1162 (unsigned)mesh->num_detached_states,
1163 (unsigned)mesh->num_reply_addrs,
1164 (unsigned)mesh->replies_sent,
1165 (unsigned)mesh->stats_dropped,
1166 (unsigned)mesh->stats_jostled);
1167 if(mesh->replies_sent > 0) {
1169 timeval_divide(&avg, &mesh->replies_sum_wait,
1170 mesh->replies_sent);
1171 log_info("average recursion processing time "
1172 ARG_LL "d.%6.6d sec",
1173 (long long)avg.tv_sec, (int)avg.tv_usec);
1174 log_info("histogram of recursion processing times");
1175 timehist_log(mesh->histogram, "recursions");
1180 mesh_stats_clear(struct mesh_area* mesh)
1184 mesh->replies_sent = 0;
1185 mesh->replies_sum_wait.tv_sec = 0;
1186 mesh->replies_sum_wait.tv_usec = 0;
1187 mesh->stats_jostled = 0;
1188 mesh->stats_dropped = 0;
1189 timehist_clear(mesh->histogram);
1190 mesh->ans_secure = 0;
1191 mesh->ans_bogus = 0;
1192 memset(&mesh->ans_rcode[0], 0, sizeof(size_t)*16);
1193 mesh->ans_nodata = 0;
1197 mesh_get_mem(struct mesh_area* mesh)
1199 struct mesh_state* m;
1200 size_t s = sizeof(*mesh) + sizeof(struct timehist) +
1201 sizeof(struct th_buck)*mesh->histogram->num +
1202 sizeof(sldns_buffer) + sldns_buffer_capacity(mesh->qbuf_bak);
1203 RBTREE_FOR(m, struct mesh_state*, &mesh->all) {
1204 /* all, including m itself allocated in qstate region */
1205 s += regional_get_mem(m->s.region);
1211 mesh_detect_cycle(struct module_qstate* qstate, struct query_info* qinfo,
1212 uint16_t flags, int prime, int valrec)
1214 struct mesh_area* mesh = qstate->env->mesh;
1215 struct mesh_state* dep_m = mesh_area_find(mesh, qinfo, flags, prime,
1217 return mesh_detect_cycle_found(qstate, dep_m);
1220 void mesh_list_insert(struct mesh_state* m, struct mesh_state** fp,
1221 struct mesh_state** lp)
1223 /* insert as last element */
1232 void mesh_list_remove(struct mesh_state* m, struct mesh_state** fp,
1233 struct mesh_state** lp)
1236 m->next->prev = m->prev;
1239 m->prev->next = m->next;