xref: /openbsd-src/usr.sbin/unbound/services/mesh.c (revision aa997e528a848ca5596493c2a801bdd6fb26ae61)
1 /*
2  * services/mesh.c - deal with mesh of query states and handle events for that.
3  *
4  * Copyright (c) 2007, NLnet Labs. All rights reserved.
5  *
6  * This software is open source.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  *
12  * Redistributions of source code must retain the above copyright notice,
13  * this list of conditions and the following disclaimer.
14  *
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.
18  *
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.
22  *
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.
34  */
35 
36 /**
37  * \file
38  *
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.
44  */
45 #include "config.h"
46 #include "services/mesh.h"
47 #include "services/outbound_list.h"
48 #include "services/cache/dns.h"
49 #include "util/log.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"
59 #include "sldns/wire2str.h"
60 #include "services/localzone.h"
61 #include "util/data/dname.h"
62 #include "respip/respip.h"
63 
64 /** subtract timers and the values do not overflow or become negative */
65 static void
66 timeval_subtract(struct timeval* d, const struct timeval* end, const struct timeval* start)
67 {
68 #ifndef S_SPLINT_S
69 	time_t end_usec = end->tv_usec;
70 	d->tv_sec = end->tv_sec - start->tv_sec;
71 	if(end_usec < start->tv_usec) {
72 		end_usec += 1000000;
73 		d->tv_sec--;
74 	}
75 	d->tv_usec = end_usec - start->tv_usec;
76 #endif
77 }
78 
79 /** add timers and the values do not overflow or become negative */
80 static void
81 timeval_add(struct timeval* d, const struct timeval* add)
82 {
83 #ifndef S_SPLINT_S
84 	d->tv_sec += add->tv_sec;
85 	d->tv_usec += add->tv_usec;
86 	if(d->tv_usec > 1000000 ) {
87 		d->tv_usec -= 1000000;
88 		d->tv_sec++;
89 	}
90 #endif
91 }
92 
93 /** divide sum of timers to get average */
94 static void
95 timeval_divide(struct timeval* avg, const struct timeval* sum, size_t d)
96 {
97 #ifndef S_SPLINT_S
98 	size_t leftover;
99 	if(d == 0) {
100 		avg->tv_sec = 0;
101 		avg->tv_usec = 0;
102 		return;
103 	}
104 	avg->tv_sec = sum->tv_sec / d;
105 	avg->tv_usec = sum->tv_usec / d;
106 	/* handle fraction from seconds divide */
107 	leftover = sum->tv_sec - avg->tv_sec*d;
108 	avg->tv_usec += (leftover*1000000)/d;
109 #endif
110 }
111 
112 /** histogram compare of time values */
113 static int
114 timeval_smaller(const struct timeval* x, const struct timeval* y)
115 {
116 #ifndef S_SPLINT_S
117 	if(x->tv_sec < y->tv_sec)
118 		return 1;
119 	else if(x->tv_sec == y->tv_sec) {
120 		if(x->tv_usec <= y->tv_usec)
121 			return 1;
122 		else	return 0;
123 	}
124 	else	return 0;
125 #endif
126 }
127 
128 /*
129  * Compare two response-ip client info entries for the purpose of mesh state
130  * compare.  It returns 0 if ci_a and ci_b are considered equal; otherwise
131  * 1 or -1 (they mean 'ci_a is larger/smaller than ci_b', respectively, but
132  * in practice it should be only used to mean they are different).
133  * We cannot share the mesh state for two queries if different response-ip
134  * actions can apply in the end, even if those queries are otherwise identical.
135  * For this purpose we compare tag lists and tag action lists; they should be
136  * identical to share the same state.
137  * For tag data, we don't look into the data content, as it can be
138  * expensive; unless tag data are not defined for both or they point to the
139  * exact same data in memory (i.e., they come from the same ACL entry), we
140  * consider these data different.
141  * Likewise, if the client info is associated with views, we don't look into
142  * the views.  They are considered different unless they are exactly the same
143  * even if the views only differ in the names.
144  */
145 static int
146 client_info_compare(const struct respip_client_info* ci_a,
147 	const struct respip_client_info* ci_b)
148 {
149 	int cmp;
150 
151 	if(!ci_a && !ci_b)
152 		return 0;
153 	if(ci_a && !ci_b)
154 		return -1;
155 	if(!ci_a && ci_b)
156 		return 1;
157 	if(ci_a->taglen != ci_b->taglen)
158 		return (ci_a->taglen < ci_b->taglen) ? -1 : 1;
159 	cmp = memcmp(ci_a->taglist, ci_b->taglist, ci_a->taglen);
160 	if(cmp != 0)
161 		return cmp;
162 	if(ci_a->tag_actions_size != ci_b->tag_actions_size)
163 		return (ci_a->tag_actions_size < ci_b->tag_actions_size) ?
164 			-1 : 1;
165 	cmp = memcmp(ci_a->tag_actions, ci_b->tag_actions,
166 		ci_a->tag_actions_size);
167 	if(cmp != 0)
168 		return cmp;
169 	if(ci_a->tag_datas != ci_b->tag_datas)
170 		return ci_a->tag_datas < ci_b->tag_datas ? -1 : 1;
171 	if(ci_a->view != ci_b->view)
172 		return ci_a->view < ci_b->view ? -1 : 1;
173 	/* For the unbound daemon these should be non-NULL and identical,
174 	 * but we check that just in case. */
175 	if(ci_a->respip_set != ci_b->respip_set)
176 		return ci_a->respip_set < ci_b->respip_set ? -1 : 1;
177 	return 0;
178 }
179 
180 int
181 mesh_state_compare(const void* ap, const void* bp)
182 {
183 	struct mesh_state* a = (struct mesh_state*)ap;
184 	struct mesh_state* b = (struct mesh_state*)bp;
185 	int cmp;
186 
187 	if(a->unique < b->unique)
188 		return -1;
189 	if(a->unique > b->unique)
190 		return 1;
191 
192 	if(a->s.is_priming && !b->s.is_priming)
193 		return -1;
194 	if(!a->s.is_priming && b->s.is_priming)
195 		return 1;
196 
197 	if(a->s.is_valrec && !b->s.is_valrec)
198 		return -1;
199 	if(!a->s.is_valrec && b->s.is_valrec)
200 		return 1;
201 
202 	if((a->s.query_flags&BIT_RD) && !(b->s.query_flags&BIT_RD))
203 		return -1;
204 	if(!(a->s.query_flags&BIT_RD) && (b->s.query_flags&BIT_RD))
205 		return 1;
206 
207 	if((a->s.query_flags&BIT_CD) && !(b->s.query_flags&BIT_CD))
208 		return -1;
209 	if(!(a->s.query_flags&BIT_CD) && (b->s.query_flags&BIT_CD))
210 		return 1;
211 
212 	cmp = query_info_compare(&a->s.qinfo, &b->s.qinfo);
213 	if(cmp != 0)
214 		return cmp;
215 	return client_info_compare(a->s.client_info, b->s.client_info);
216 }
217 
218 int
219 mesh_state_ref_compare(const void* ap, const void* bp)
220 {
221 	struct mesh_state_ref* a = (struct mesh_state_ref*)ap;
222 	struct mesh_state_ref* b = (struct mesh_state_ref*)bp;
223 	return mesh_state_compare(a->s, b->s);
224 }
225 
226 struct mesh_area*
227 mesh_create(struct module_stack* stack, struct module_env* env)
228 {
229 	struct mesh_area* mesh = calloc(1, sizeof(struct mesh_area));
230 	if(!mesh) {
231 		log_err("mesh area alloc: out of memory");
232 		return NULL;
233 	}
234 	mesh->histogram = timehist_setup();
235 	mesh->qbuf_bak = sldns_buffer_new(env->cfg->msg_buffer_size);
236 	if(!mesh->histogram || !mesh->qbuf_bak) {
237 		free(mesh);
238 		log_err("mesh area alloc: out of memory");
239 		return NULL;
240 	}
241 	mesh->mods = *stack;
242 	mesh->env = env;
243 	rbtree_init(&mesh->run, &mesh_state_compare);
244 	rbtree_init(&mesh->all, &mesh_state_compare);
245 	mesh->num_reply_addrs = 0;
246 	mesh->num_reply_states = 0;
247 	mesh->num_detached_states = 0;
248 	mesh->num_forever_states = 0;
249 	mesh->stats_jostled = 0;
250 	mesh->stats_dropped = 0;
251 	mesh->max_reply_states = env->cfg->num_queries_per_thread;
252 	mesh->max_forever_states = (mesh->max_reply_states+1)/2;
253 #ifndef S_SPLINT_S
254 	mesh->jostle_max.tv_sec = (time_t)(env->cfg->jostle_time / 1000);
255 	mesh->jostle_max.tv_usec = (time_t)((env->cfg->jostle_time % 1000)
256 		*1000);
257 #endif
258 	return mesh;
259 }
260 
261 /** help mesh delete delete mesh states */
262 static void
263 mesh_delete_helper(rbnode_type* n)
264 {
265 	struct mesh_state* mstate = (struct mesh_state*)n->key;
266 	/* perform a full delete, not only 'cleanup' routine,
267 	 * because other callbacks expect a clean state in the mesh.
268 	 * For 're-entrant' calls */
269 	mesh_state_delete(&mstate->s);
270 	/* but because these delete the items from the tree, postorder
271 	 * traversal and rbtree rebalancing do not work together */
272 }
273 
274 void
275 mesh_delete(struct mesh_area* mesh)
276 {
277 	if(!mesh)
278 		return;
279 	/* free all query states */
280 	while(mesh->all.count)
281 		mesh_delete_helper(mesh->all.root);
282 	timehist_delete(mesh->histogram);
283 	sldns_buffer_free(mesh->qbuf_bak);
284 	free(mesh);
285 }
286 
287 void
288 mesh_delete_all(struct mesh_area* mesh)
289 {
290 	/* free all query states */
291 	while(mesh->all.count)
292 		mesh_delete_helper(mesh->all.root);
293 	mesh->stats_dropped += mesh->num_reply_addrs;
294 	/* clear mesh area references */
295 	rbtree_init(&mesh->run, &mesh_state_compare);
296 	rbtree_init(&mesh->all, &mesh_state_compare);
297 	mesh->num_reply_addrs = 0;
298 	mesh->num_reply_states = 0;
299 	mesh->num_detached_states = 0;
300 	mesh->num_forever_states = 0;
301 	mesh->forever_first = NULL;
302 	mesh->forever_last = NULL;
303 	mesh->jostle_first = NULL;
304 	mesh->jostle_last = NULL;
305 }
306 
307 int mesh_make_new_space(struct mesh_area* mesh, sldns_buffer* qbuf)
308 {
309 	struct mesh_state* m = mesh->jostle_first;
310 	/* free space is available */
311 	if(mesh->num_reply_states < mesh->max_reply_states)
312 		return 1;
313 	/* try to kick out a jostle-list item */
314 	if(m && m->reply_list && m->list_select == mesh_jostle_list) {
315 		/* how old is it? */
316 		struct timeval age;
317 		timeval_subtract(&age, mesh->env->now_tv,
318 			&m->reply_list->start_time);
319 		if(timeval_smaller(&mesh->jostle_max, &age)) {
320 			/* its a goner */
321 			log_nametypeclass(VERB_ALGO, "query jostled out to "
322 				"make space for a new one",
323 				m->s.qinfo.qname, m->s.qinfo.qtype,
324 				m->s.qinfo.qclass);
325 			/* backup the query */
326 			if(qbuf) sldns_buffer_copy(mesh->qbuf_bak, qbuf);
327 			/* notify supers */
328 			if(m->super_set.count > 0) {
329 				verbose(VERB_ALGO, "notify supers of failure");
330 				m->s.return_msg = NULL;
331 				m->s.return_rcode = LDNS_RCODE_SERVFAIL;
332 				mesh_walk_supers(mesh, m);
333 			}
334 			mesh->stats_jostled ++;
335 			mesh_state_delete(&m->s);
336 			/* restore the query - note that the qinfo ptr to
337 			 * the querybuffer is then correct again. */
338 			if(qbuf) sldns_buffer_copy(qbuf, mesh->qbuf_bak);
339 			return 1;
340 		}
341 	}
342 	/* no space for new item */
343 	return 0;
344 }
345 
346 void mesh_new_client(struct mesh_area* mesh, struct query_info* qinfo,
347 	struct respip_client_info* cinfo, uint16_t qflags,
348 	struct edns_data* edns, struct comm_reply* rep, uint16_t qid)
349 {
350 	struct mesh_state* s = NULL;
351 	int unique = unique_mesh_state(edns->opt_list, mesh->env);
352 	int was_detached = 0;
353 	int was_noreply = 0;
354 	int added = 0;
355 	if(!unique)
356 		s = mesh_area_find(mesh, cinfo, qinfo, qflags&(BIT_RD|BIT_CD), 0, 0);
357 	/* does this create a new reply state? */
358 	if(!s || s->list_select == mesh_no_list) {
359 		if(!mesh_make_new_space(mesh, rep->c->buffer)) {
360 			verbose(VERB_ALGO, "Too many queries. dropping "
361 				"incoming query.");
362 			comm_point_drop_reply(rep);
363 			mesh->stats_dropped ++;
364 			return;
365 		}
366 		/* for this new reply state, the reply address is free,
367 		 * so the limit of reply addresses does not stop reply states*/
368 	} else {
369 		/* protect our memory usage from storing reply addresses */
370 		if(mesh->num_reply_addrs > mesh->max_reply_states*16) {
371 			verbose(VERB_ALGO, "Too many requests queued. "
372 				"dropping incoming query.");
373 			mesh->stats_dropped++;
374 			comm_point_drop_reply(rep);
375 			return;
376 		}
377 	}
378 	/* see if it already exists, if not, create one */
379 	if(!s) {
380 #ifdef UNBOUND_DEBUG
381 		struct rbnode_type* n;
382 #endif
383 		s = mesh_state_create(mesh->env, qinfo, cinfo,
384 			qflags&(BIT_RD|BIT_CD), 0, 0);
385 		if(!s) {
386 			log_err("mesh_state_create: out of memory; SERVFAIL");
387 			if(!inplace_cb_reply_servfail_call(mesh->env, qinfo, NULL, NULL,
388 				LDNS_RCODE_SERVFAIL, edns, mesh->env->scratch))
389 					edns->opt_list = NULL;
390 			error_encode(rep->c->buffer, LDNS_RCODE_SERVFAIL,
391 				qinfo, qid, qflags, edns);
392 			comm_point_send_reply(rep);
393 			return;
394 		}
395 		if(unique)
396 			mesh_state_make_unique(s);
397 		/* copy the edns options we got from the front */
398 		if(edns->opt_list) {
399 			s->s.edns_opts_front_in = edns_opt_copy_region(edns->opt_list,
400 				s->s.region);
401 			if(!s->s.edns_opts_front_in) {
402 				log_err("mesh_state_create: out of memory; SERVFAIL");
403 				if(!inplace_cb_reply_servfail_call(mesh->env, qinfo, NULL,
404 					NULL, LDNS_RCODE_SERVFAIL, edns, mesh->env->scratch))
405 						edns->opt_list = NULL;
406 				error_encode(rep->c->buffer, LDNS_RCODE_SERVFAIL,
407 					qinfo, qid, qflags, edns);
408 				comm_point_send_reply(rep);
409 				return;
410 			}
411 		}
412 
413 #ifdef UNBOUND_DEBUG
414 		n =
415 #else
416 		(void)
417 #endif
418 		rbtree_insert(&mesh->all, &s->node);
419 		log_assert(n != NULL);
420 		/* set detached (it is now) */
421 		mesh->num_detached_states++;
422 		added = 1;
423 	}
424 	if(!s->reply_list && !s->cb_list && s->super_set.count == 0)
425 		was_detached = 1;
426 	if(!s->reply_list && !s->cb_list)
427 		was_noreply = 1;
428 	/* add reply to s */
429 	if(!mesh_state_add_reply(s, edns, rep, qid, qflags, qinfo)) {
430 			log_err("mesh_new_client: out of memory; SERVFAIL");
431 			if(!inplace_cb_reply_servfail_call(mesh->env, qinfo, &s->s,
432 				NULL, LDNS_RCODE_SERVFAIL, edns, mesh->env->scratch))
433 					edns->opt_list = NULL;
434 			error_encode(rep->c->buffer, LDNS_RCODE_SERVFAIL,
435 				qinfo, qid, qflags, edns);
436 			comm_point_send_reply(rep);
437 			if(added)
438 				mesh_state_delete(&s->s);
439 			return;
440 	}
441 	/* update statistics */
442 	if(was_detached) {
443 		log_assert(mesh->num_detached_states > 0);
444 		mesh->num_detached_states--;
445 	}
446 	if(was_noreply) {
447 		mesh->num_reply_states ++;
448 	}
449 	mesh->num_reply_addrs++;
450 	if(s->list_select == mesh_no_list) {
451 		/* move to either the forever or the jostle_list */
452 		if(mesh->num_forever_states < mesh->max_forever_states) {
453 			mesh->num_forever_states ++;
454 			mesh_list_insert(s, &mesh->forever_first,
455 				&mesh->forever_last);
456 			s->list_select = mesh_forever_list;
457 		} else {
458 			mesh_list_insert(s, &mesh->jostle_first,
459 				&mesh->jostle_last);
460 			s->list_select = mesh_jostle_list;
461 		}
462 	}
463 	if(added)
464 		mesh_run(mesh, s, module_event_new, NULL);
465 }
466 
467 int
468 mesh_new_callback(struct mesh_area* mesh, struct query_info* qinfo,
469 	uint16_t qflags, struct edns_data* edns, sldns_buffer* buf,
470 	uint16_t qid, mesh_cb_func_type cb, void* cb_arg)
471 {
472 	struct mesh_state* s = NULL;
473 	int unique = unique_mesh_state(edns->opt_list, mesh->env);
474 	int was_detached = 0;
475 	int was_noreply = 0;
476 	int added = 0;
477 	if(!unique)
478 		s = mesh_area_find(mesh, NULL, qinfo, qflags&(BIT_RD|BIT_CD), 0, 0);
479 
480 	/* there are no limits on the number of callbacks */
481 
482 	/* see if it already exists, if not, create one */
483 	if(!s) {
484 #ifdef UNBOUND_DEBUG
485 		struct rbnode_type* n;
486 #endif
487 		s = mesh_state_create(mesh->env, qinfo, NULL,
488 			qflags&(BIT_RD|BIT_CD), 0, 0);
489 		if(!s) {
490 			return 0;
491 		}
492 		if(unique)
493 			mesh_state_make_unique(s);
494 		if(edns->opt_list) {
495 			s->s.edns_opts_front_in = edns_opt_copy_region(edns->opt_list,
496 				s->s.region);
497 			if(!s->s.edns_opts_front_in) {
498 				return 0;
499 			}
500 		}
501 #ifdef UNBOUND_DEBUG
502 		n =
503 #else
504 		(void)
505 #endif
506 		rbtree_insert(&mesh->all, &s->node);
507 		log_assert(n != NULL);
508 		/* set detached (it is now) */
509 		mesh->num_detached_states++;
510 		added = 1;
511 	}
512 	if(!s->reply_list && !s->cb_list && s->super_set.count == 0)
513 		was_detached = 1;
514 	if(!s->reply_list && !s->cb_list)
515 		was_noreply = 1;
516 	/* add reply to s */
517 	if(!mesh_state_add_cb(s, edns, buf, cb, cb_arg, qid, qflags)) {
518 			if(added)
519 				mesh_state_delete(&s->s);
520 			return 0;
521 	}
522 	/* update statistics */
523 	if(was_detached) {
524 		log_assert(mesh->num_detached_states > 0);
525 		mesh->num_detached_states--;
526 	}
527 	if(was_noreply) {
528 		mesh->num_reply_states ++;
529 	}
530 	mesh->num_reply_addrs++;
531 	if(added)
532 		mesh_run(mesh, s, module_event_new, NULL);
533 	return 1;
534 }
535 
536 static void mesh_schedule_prefetch(struct mesh_area* mesh,
537 	struct query_info* qinfo, uint16_t qflags, time_t leeway, int run);
538 
539 void mesh_new_prefetch(struct mesh_area* mesh, struct query_info* qinfo,
540         uint16_t qflags, time_t leeway)
541 {
542 	mesh_schedule_prefetch(mesh, qinfo, qflags, leeway, 1);
543 }
544 
545 /* Internal backend routine of mesh_new_prefetch().  It takes one additional
546  * parameter, 'run', which controls whether to run the prefetch state
547  * immediately.  When this function is called internally 'run' could be
548  * 0 (false), in which case the new state is only made runnable so it
549  * will not be run recursively on top of the current state. */
550 static void mesh_schedule_prefetch(struct mesh_area* mesh,
551 	struct query_info* qinfo, uint16_t qflags, time_t leeway, int run)
552 {
553 	struct mesh_state* s = mesh_area_find(mesh, NULL, qinfo,
554 		qflags&(BIT_RD|BIT_CD), 0, 0);
555 #ifdef UNBOUND_DEBUG
556 	struct rbnode_type* n;
557 #endif
558 	/* already exists, and for a different purpose perhaps.
559 	 * if mesh_no_list, keep it that way. */
560 	if(s) {
561 		/* make it ignore the cache from now on */
562 		if(!s->s.blacklist)
563 			sock_list_insert(&s->s.blacklist, NULL, 0, s->s.region);
564 		if(s->s.prefetch_leeway < leeway)
565 			s->s.prefetch_leeway = leeway;
566 		return;
567 	}
568 	if(!mesh_make_new_space(mesh, NULL)) {
569 		verbose(VERB_ALGO, "Too many queries. dropped prefetch.");
570 		mesh->stats_dropped ++;
571 		return;
572 	}
573 
574 	s = mesh_state_create(mesh->env, qinfo, NULL,
575 		qflags&(BIT_RD|BIT_CD), 0, 0);
576 	if(!s) {
577 		log_err("prefetch mesh_state_create: out of memory");
578 		return;
579 	}
580 #ifdef UNBOUND_DEBUG
581 	n =
582 #else
583 	(void)
584 #endif
585 	rbtree_insert(&mesh->all, &s->node);
586 	log_assert(n != NULL);
587 	/* set detached (it is now) */
588 	mesh->num_detached_states++;
589 	/* make it ignore the cache */
590 	sock_list_insert(&s->s.blacklist, NULL, 0, s->s.region);
591 	s->s.prefetch_leeway = leeway;
592 
593 	if(s->list_select == mesh_no_list) {
594 		/* move to either the forever or the jostle_list */
595 		if(mesh->num_forever_states < mesh->max_forever_states) {
596 			mesh->num_forever_states ++;
597 			mesh_list_insert(s, &mesh->forever_first,
598 				&mesh->forever_last);
599 			s->list_select = mesh_forever_list;
600 		} else {
601 			mesh_list_insert(s, &mesh->jostle_first,
602 				&mesh->jostle_last);
603 			s->list_select = mesh_jostle_list;
604 		}
605 	}
606 
607 	if(!run) {
608 #ifdef UNBOUND_DEBUG
609 		n =
610 #else
611 		(void)
612 #endif
613 		rbtree_insert(&mesh->run, &s->run_node);
614 		log_assert(n != NULL);
615 		return;
616 	}
617 
618 	mesh_run(mesh, s, module_event_new, NULL);
619 }
620 
621 void mesh_report_reply(struct mesh_area* mesh, struct outbound_entry* e,
622         struct comm_reply* reply, int what)
623 {
624 	enum module_ev event = module_event_reply;
625 	e->qstate->reply = reply;
626 	if(what != NETEVENT_NOERROR) {
627 		event = module_event_noreply;
628 		if(what == NETEVENT_CAPSFAIL)
629 			event = module_event_capsfail;
630 	}
631 	mesh_run(mesh, e->qstate->mesh_info, event, e);
632 }
633 
634 struct mesh_state*
635 mesh_state_create(struct module_env* env, struct query_info* qinfo,
636 	struct respip_client_info* cinfo, uint16_t qflags, int prime,
637 	int valrec)
638 {
639 	struct regional* region = alloc_reg_obtain(env->alloc);
640 	struct mesh_state* mstate;
641 	int i;
642 	if(!region)
643 		return NULL;
644 	mstate = (struct mesh_state*)regional_alloc(region,
645 		sizeof(struct mesh_state));
646 	if(!mstate) {
647 		alloc_reg_release(env->alloc, region);
648 		return NULL;
649 	}
650 	memset(mstate, 0, sizeof(*mstate));
651 	mstate->node = *RBTREE_NULL;
652 	mstate->run_node = *RBTREE_NULL;
653 	mstate->node.key = mstate;
654 	mstate->run_node.key = mstate;
655 	mstate->reply_list = NULL;
656 	mstate->list_select = mesh_no_list;
657 	mstate->replies_sent = 0;
658 	rbtree_init(&mstate->super_set, &mesh_state_ref_compare);
659 	rbtree_init(&mstate->sub_set, &mesh_state_ref_compare);
660 	mstate->num_activated = 0;
661 	mstate->unique = NULL;
662 	/* init module qstate */
663 	mstate->s.qinfo.qtype = qinfo->qtype;
664 	mstate->s.qinfo.qclass = qinfo->qclass;
665 	mstate->s.qinfo.local_alias = NULL;
666 	mstate->s.qinfo.qname_len = qinfo->qname_len;
667 	mstate->s.qinfo.qname = regional_alloc_init(region, qinfo->qname,
668 		qinfo->qname_len);
669 	if(!mstate->s.qinfo.qname) {
670 		alloc_reg_release(env->alloc, region);
671 		return NULL;
672 	}
673 	if(cinfo) {
674 		mstate->s.client_info = regional_alloc_init(region, cinfo,
675 			sizeof(*cinfo));
676 		if(!mstate->s.client_info) {
677 			alloc_reg_release(env->alloc, region);
678 			return NULL;
679 		}
680 	}
681 	/* remove all weird bits from qflags */
682 	mstate->s.query_flags = (qflags & (BIT_RD|BIT_CD));
683 	mstate->s.is_priming = prime;
684 	mstate->s.is_valrec = valrec;
685 	mstate->s.reply = NULL;
686 	mstate->s.region = region;
687 	mstate->s.curmod = 0;
688 	mstate->s.return_msg = 0;
689 	mstate->s.return_rcode = LDNS_RCODE_NOERROR;
690 	mstate->s.env = env;
691 	mstate->s.mesh_info = mstate;
692 	mstate->s.prefetch_leeway = 0;
693 	mstate->s.no_cache_lookup = 0;
694 	mstate->s.no_cache_store = 0;
695 	mstate->s.need_refetch = 0;
696 
697 	/* init modules */
698 	for(i=0; i<env->mesh->mods.num; i++) {
699 		mstate->s.minfo[i] = NULL;
700 		mstate->s.ext_state[i] = module_state_initial;
701 	}
702 	/* init edns option lists */
703 	mstate->s.edns_opts_front_in = NULL;
704 	mstate->s.edns_opts_back_out = NULL;
705 	mstate->s.edns_opts_back_in = NULL;
706 	mstate->s.edns_opts_front_out = NULL;
707 
708 	return mstate;
709 }
710 
711 int
712 mesh_state_is_unique(struct mesh_state* mstate)
713 {
714 	return mstate->unique != NULL;
715 }
716 
717 void
718 mesh_state_make_unique(struct mesh_state* mstate)
719 {
720 	mstate->unique = mstate;
721 }
722 
723 void
724 mesh_state_cleanup(struct mesh_state* mstate)
725 {
726 	struct mesh_area* mesh;
727 	int i;
728 	if(!mstate)
729 		return;
730 	mesh = mstate->s.env->mesh;
731 	/* drop unsent replies */
732 	if(!mstate->replies_sent) {
733 		struct mesh_reply* rep;
734 		struct mesh_cb* cb;
735 		for(rep=mstate->reply_list; rep; rep=rep->next) {
736 			comm_point_drop_reply(&rep->query_reply);
737 			mesh->num_reply_addrs--;
738 		}
739 		for(cb=mstate->cb_list; cb; cb=cb->next) {
740 			fptr_ok(fptr_whitelist_mesh_cb(cb->cb));
741 			(*cb->cb)(cb->cb_arg, LDNS_RCODE_SERVFAIL, NULL,
742 				sec_status_unchecked, NULL);
743 			mesh->num_reply_addrs--;
744 		}
745 	}
746 
747 	/* de-init modules */
748 	for(i=0; i<mesh->mods.num; i++) {
749 		fptr_ok(fptr_whitelist_mod_clear(mesh->mods.mod[i]->clear));
750 		(*mesh->mods.mod[i]->clear)(&mstate->s, i);
751 		mstate->s.minfo[i] = NULL;
752 		mstate->s.ext_state[i] = module_finished;
753 	}
754 	alloc_reg_release(mstate->s.env->alloc, mstate->s.region);
755 }
756 
757 void
758 mesh_state_delete(struct module_qstate* qstate)
759 {
760 	struct mesh_area* mesh;
761 	struct mesh_state_ref* super, ref;
762 	struct mesh_state* mstate;
763 	if(!qstate)
764 		return;
765 	mstate = qstate->mesh_info;
766 	mesh = mstate->s.env->mesh;
767 	mesh_detach_subs(&mstate->s);
768 	if(mstate->list_select == mesh_forever_list) {
769 		mesh->num_forever_states --;
770 		mesh_list_remove(mstate, &mesh->forever_first,
771 			&mesh->forever_last);
772 	} else if(mstate->list_select == mesh_jostle_list) {
773 		mesh_list_remove(mstate, &mesh->jostle_first,
774 			&mesh->jostle_last);
775 	}
776 	if(!mstate->reply_list && !mstate->cb_list
777 		&& mstate->super_set.count == 0) {
778 		log_assert(mesh->num_detached_states > 0);
779 		mesh->num_detached_states--;
780 	}
781 	if(mstate->reply_list || mstate->cb_list) {
782 		log_assert(mesh->num_reply_states > 0);
783 		mesh->num_reply_states--;
784 	}
785 	ref.node.key = &ref;
786 	ref.s = mstate;
787 	RBTREE_FOR(super, struct mesh_state_ref*, &mstate->super_set) {
788 		(void)rbtree_delete(&super->s->sub_set, &ref);
789 	}
790 	(void)rbtree_delete(&mesh->run, mstate);
791 	(void)rbtree_delete(&mesh->all, mstate);
792 	mesh_state_cleanup(mstate);
793 }
794 
795 /** helper recursive rbtree find routine */
796 static int
797 find_in_subsub(struct mesh_state* m, struct mesh_state* tofind, size_t *c)
798 {
799 	struct mesh_state_ref* r;
800 	if((*c)++ > MESH_MAX_SUBSUB)
801 		return 1;
802 	RBTREE_FOR(r, struct mesh_state_ref*, &m->sub_set) {
803 		if(r->s == tofind || find_in_subsub(r->s, tofind, c))
804 			return 1;
805 	}
806 	return 0;
807 }
808 
809 /** find cycle for already looked up mesh_state */
810 static int
811 mesh_detect_cycle_found(struct module_qstate* qstate, struct mesh_state* dep_m)
812 {
813 	struct mesh_state* cyc_m = qstate->mesh_info;
814 	size_t counter = 0;
815 	if(!dep_m)
816 		return 0;
817 	if(dep_m == cyc_m || find_in_subsub(dep_m, cyc_m, &counter)) {
818 		if(counter > MESH_MAX_SUBSUB)
819 			return 2;
820 		return 1;
821 	}
822 	return 0;
823 }
824 
825 void mesh_detach_subs(struct module_qstate* qstate)
826 {
827 	struct mesh_area* mesh = qstate->env->mesh;
828 	struct mesh_state_ref* ref, lookup;
829 #ifdef UNBOUND_DEBUG
830 	struct rbnode_type* n;
831 #endif
832 	lookup.node.key = &lookup;
833 	lookup.s = qstate->mesh_info;
834 	RBTREE_FOR(ref, struct mesh_state_ref*, &qstate->mesh_info->sub_set) {
835 #ifdef UNBOUND_DEBUG
836 		n =
837 #else
838 		(void)
839 #endif
840 		rbtree_delete(&ref->s->super_set, &lookup);
841 		log_assert(n != NULL); /* must have been present */
842 		if(!ref->s->reply_list && !ref->s->cb_list
843 			&& ref->s->super_set.count == 0) {
844 			mesh->num_detached_states++;
845 			log_assert(mesh->num_detached_states +
846 				mesh->num_reply_states <= mesh->all.count);
847 		}
848 	}
849 	rbtree_init(&qstate->mesh_info->sub_set, &mesh_state_ref_compare);
850 }
851 
852 int mesh_add_sub(struct module_qstate* qstate, struct query_info* qinfo,
853         uint16_t qflags, int prime, int valrec, struct module_qstate** newq,
854 	struct mesh_state** sub)
855 {
856 	/* find it, if not, create it */
857 	struct mesh_area* mesh = qstate->env->mesh;
858 	*sub = mesh_area_find(mesh, NULL, qinfo, qflags,
859 		prime, valrec);
860 	if(mesh_detect_cycle_found(qstate, *sub)) {
861 		verbose(VERB_ALGO, "attach failed, cycle detected");
862 		return 0;
863 	}
864 	if(!*sub) {
865 #ifdef UNBOUND_DEBUG
866 		struct rbnode_type* n;
867 #endif
868 		/* create a new one */
869 		*sub = mesh_state_create(qstate->env, qinfo, NULL, qflags, prime,
870 			valrec);
871 		if(!*sub) {
872 			log_err("mesh_attach_sub: out of memory");
873 			return 0;
874 		}
875 #ifdef UNBOUND_DEBUG
876 		n =
877 #else
878 		(void)
879 #endif
880 		rbtree_insert(&mesh->all, &(*sub)->node);
881 		log_assert(n != NULL);
882 		/* set detached (it is now) */
883 		mesh->num_detached_states++;
884 		/* set new query state to run */
885 #ifdef UNBOUND_DEBUG
886 		n =
887 #else
888 		(void)
889 #endif
890 		rbtree_insert(&mesh->run, &(*sub)->run_node);
891 		log_assert(n != NULL);
892 		*newq = &(*sub)->s;
893 	} else
894 		*newq = NULL;
895 	return 1;
896 }
897 
898 int mesh_attach_sub(struct module_qstate* qstate, struct query_info* qinfo,
899         uint16_t qflags, int prime, int valrec, struct module_qstate** newq)
900 {
901 	struct mesh_area* mesh = qstate->env->mesh;
902 	struct mesh_state* sub = NULL;
903 	int was_detached;
904 	if(!mesh_add_sub(qstate, qinfo, qflags, prime, valrec, newq, &sub))
905 		return 0;
906 	was_detached = (sub->super_set.count == 0);
907 	if(!mesh_state_attachment(qstate->mesh_info, sub))
908 		return 0;
909 	/* if it was a duplicate  attachment, the count was not zero before */
910 	if(!sub->reply_list && !sub->cb_list && was_detached &&
911 		sub->super_set.count == 1) {
912 		/* it used to be detached, before this one got added */
913 		log_assert(mesh->num_detached_states > 0);
914 		mesh->num_detached_states--;
915 	}
916 	/* *newq will be run when inited after the current module stops */
917 	return 1;
918 }
919 
920 int mesh_state_attachment(struct mesh_state* super, struct mesh_state* sub)
921 {
922 #ifdef UNBOUND_DEBUG
923 	struct rbnode_type* n;
924 #endif
925 	struct mesh_state_ref* subref; /* points to sub, inserted in super */
926 	struct mesh_state_ref* superref; /* points to super, inserted in sub */
927 	if( !(subref = regional_alloc(super->s.region,
928 		sizeof(struct mesh_state_ref))) ||
929 		!(superref = regional_alloc(sub->s.region,
930 		sizeof(struct mesh_state_ref))) ) {
931 		log_err("mesh_state_attachment: out of memory");
932 		return 0;
933 	}
934 	superref->node.key = superref;
935 	superref->s = super;
936 	subref->node.key = subref;
937 	subref->s = sub;
938 	if(!rbtree_insert(&sub->super_set, &superref->node)) {
939 		/* this should not happen, iterator and validator do not
940 		 * attach subqueries that are identical. */
941 		/* already attached, we are done, nothing todo.
942 		 * since superref and subref already allocated in region,
943 		 * we cannot free them */
944 		return 1;
945 	}
946 #ifdef UNBOUND_DEBUG
947 	n =
948 #else
949 	(void)
950 #endif
951 	rbtree_insert(&super->sub_set, &subref->node);
952 	log_assert(n != NULL); /* we checked above if statement, the reverse
953 	  administration should not fail now, unless they are out of sync */
954 	return 1;
955 }
956 
957 /**
958  * callback results to mesh cb entry
959  * @param m: mesh state to send it for.
960  * @param rcode: if not 0, error code.
961  * @param rep: reply to send (or NULL if rcode is set).
962  * @param r: callback entry
963  */
964 static void
965 mesh_do_callback(struct mesh_state* m, int rcode, struct reply_info* rep,
966 	struct mesh_cb* r)
967 {
968 	int secure;
969 	char* reason = NULL;
970 	/* bogus messages are not made into servfail, sec_status passed
971 	 * to the callback function */
972 	if(rep && rep->security == sec_status_secure)
973 		secure = 1;
974 	else	secure = 0;
975 	if(!rep && rcode == LDNS_RCODE_NOERROR)
976 		rcode = LDNS_RCODE_SERVFAIL;
977 	if(!rcode && rep->security == sec_status_bogus) {
978 		if(!(reason = errinf_to_str(&m->s)))
979 			rcode = LDNS_RCODE_SERVFAIL;
980 	}
981 	/* send the reply */
982 	if(rcode) {
983 		if(rcode == LDNS_RCODE_SERVFAIL) {
984 			if(!inplace_cb_reply_servfail_call(m->s.env, &m->s.qinfo, &m->s,
985 				rep, rcode, &r->edns, m->s.region))
986 					r->edns.opt_list = NULL;
987 		} else {
988 			if(!inplace_cb_reply_call(m->s.env, &m->s.qinfo, &m->s, rep, rcode,
989 				&r->edns, m->s.region))
990 					r->edns.opt_list = NULL;
991 		}
992 		fptr_ok(fptr_whitelist_mesh_cb(r->cb));
993 		(*r->cb)(r->cb_arg, rcode, r->buf, sec_status_unchecked, NULL);
994 	} else {
995 		size_t udp_size = r->edns.udp_size;
996 		sldns_buffer_clear(r->buf);
997 		r->edns.edns_version = EDNS_ADVERTISED_VERSION;
998 		r->edns.udp_size = EDNS_ADVERTISED_SIZE;
999 		r->edns.ext_rcode = 0;
1000 		r->edns.bits &= EDNS_DO;
1001 
1002 		if(!inplace_cb_reply_call(m->s.env, &m->s.qinfo, &m->s, rep,
1003 			LDNS_RCODE_NOERROR, &r->edns, m->s.region) ||
1004 			!reply_info_answer_encode(&m->s.qinfo, rep, r->qid,
1005 			r->qflags, r->buf, 0, 1,
1006 			m->s.env->scratch, udp_size, &r->edns,
1007 			(int)(r->edns.bits & EDNS_DO), secure))
1008 		{
1009 			fptr_ok(fptr_whitelist_mesh_cb(r->cb));
1010 			(*r->cb)(r->cb_arg, LDNS_RCODE_SERVFAIL, r->buf,
1011 				sec_status_unchecked, NULL);
1012 		} else {
1013 			fptr_ok(fptr_whitelist_mesh_cb(r->cb));
1014 			(*r->cb)(r->cb_arg, LDNS_RCODE_NOERROR, r->buf,
1015 				rep->security, reason);
1016 		}
1017 	}
1018 	free(reason);
1019 	m->s.env->mesh->num_reply_addrs--;
1020 }
1021 
1022 /**
1023  * Send reply to mesh reply entry
1024  * @param m: mesh state to send it for.
1025  * @param rcode: if not 0, error code.
1026  * @param rep: reply to send (or NULL if rcode is set).
1027  * @param r: reply entry
1028  * @param prev: previous reply, already has its answer encoded in buffer.
1029  */
1030 static void
1031 mesh_send_reply(struct mesh_state* m, int rcode, struct reply_info* rep,
1032 	struct mesh_reply* r, struct mesh_reply* prev)
1033 {
1034 	struct timeval end_time;
1035 	struct timeval duration;
1036 	int secure;
1037 	/* Copy the client's EDNS for later restore, to make sure the edns
1038 	 * compare is with the correct edns options. */
1039 	struct edns_data edns_bak = r->edns;
1040 	/* examine security status */
1041 	if(m->s.env->need_to_validate && (!(r->qflags&BIT_CD) ||
1042 		m->s.env->cfg->ignore_cd) && rep &&
1043 		rep->security <= sec_status_bogus) {
1044 		rcode = LDNS_RCODE_SERVFAIL;
1045 		if(m->s.env->cfg->stat_extended)
1046 			m->s.env->mesh->ans_bogus++;
1047 	}
1048 	if(rep && rep->security == sec_status_secure)
1049 		secure = 1;
1050 	else	secure = 0;
1051 	if(!rep && rcode == LDNS_RCODE_NOERROR)
1052 		rcode = LDNS_RCODE_SERVFAIL;
1053 	/* send the reply */
1054 	/* We don't reuse the encoded answer if either the previous or current
1055 	 * response has a local alias.  We could compare the alias records
1056 	 * and still reuse the previous answer if they are the same, but that
1057 	 * would be complicated and error prone for the relatively minor case.
1058 	 * So we err on the side of safety. */
1059 	if(prev && prev->qflags == r->qflags &&
1060 		!prev->local_alias && !r->local_alias &&
1061 		prev->edns.edns_present == r->edns.edns_present &&
1062 		prev->edns.bits == r->edns.bits &&
1063 		prev->edns.udp_size == r->edns.udp_size &&
1064 		edns_opt_list_compare(prev->edns.opt_list, r->edns.opt_list)
1065 		== 0) {
1066 		/* if the previous reply is identical to this one, fix ID */
1067 		if(prev->query_reply.c->buffer != r->query_reply.c->buffer)
1068 			sldns_buffer_copy(r->query_reply.c->buffer,
1069 				prev->query_reply.c->buffer);
1070 		sldns_buffer_write_at(r->query_reply.c->buffer, 0,
1071 			&r->qid, sizeof(uint16_t));
1072 		sldns_buffer_write_at(r->query_reply.c->buffer, 12,
1073 			r->qname, m->s.qinfo.qname_len);
1074 		comm_point_send_reply(&r->query_reply);
1075 	} else if(rcode) {
1076 		m->s.qinfo.qname = r->qname;
1077 		m->s.qinfo.local_alias = r->local_alias;
1078 		if(rcode == LDNS_RCODE_SERVFAIL) {
1079 			if(!inplace_cb_reply_servfail_call(m->s.env, &m->s.qinfo, &m->s,
1080 				rep, rcode, &r->edns, m->s.region))
1081 					r->edns.opt_list = NULL;
1082 		} else {
1083 			if(!inplace_cb_reply_call(m->s.env, &m->s.qinfo, &m->s, rep, rcode,
1084 				&r->edns, m->s.region))
1085 					r->edns.opt_list = NULL;
1086 		}
1087 		error_encode(r->query_reply.c->buffer, rcode, &m->s.qinfo,
1088 			r->qid, r->qflags, &r->edns);
1089 		comm_point_send_reply(&r->query_reply);
1090 	} else {
1091 		size_t udp_size = r->edns.udp_size;
1092 		r->edns.edns_version = EDNS_ADVERTISED_VERSION;
1093 		r->edns.udp_size = EDNS_ADVERTISED_SIZE;
1094 		r->edns.ext_rcode = 0;
1095 		r->edns.bits &= EDNS_DO;
1096 		m->s.qinfo.qname = r->qname;
1097 		m->s.qinfo.local_alias = r->local_alias;
1098 		if(!inplace_cb_reply_call(m->s.env, &m->s.qinfo, &m->s, rep,
1099 			LDNS_RCODE_NOERROR, &r->edns, m->s.region) ||
1100 			!reply_info_answer_encode(&m->s.qinfo, rep, r->qid,
1101 			r->qflags, r->query_reply.c->buffer, 0, 1,
1102 			m->s.env->scratch, udp_size, &r->edns,
1103 			(int)(r->edns.bits & EDNS_DO), secure))
1104 		{
1105 			if(!inplace_cb_reply_servfail_call(m->s.env, &m->s.qinfo, &m->s,
1106 			rep, LDNS_RCODE_SERVFAIL, &r->edns, m->s.region))
1107 				r->edns.opt_list = NULL;
1108 			error_encode(r->query_reply.c->buffer,
1109 				LDNS_RCODE_SERVFAIL, &m->s.qinfo, r->qid,
1110 				r->qflags, &r->edns);
1111 		}
1112 		r->edns = edns_bak;
1113 		comm_point_send_reply(&r->query_reply);
1114 	}
1115 	/* account */
1116 	m->s.env->mesh->num_reply_addrs--;
1117 	end_time = *m->s.env->now_tv;
1118 	timeval_subtract(&duration, &end_time, &r->start_time);
1119 	verbose(VERB_ALGO, "query took " ARG_LL "d.%6.6d sec",
1120 		(long long)duration.tv_sec, (int)duration.tv_usec);
1121 	m->s.env->mesh->replies_sent++;
1122 	timeval_add(&m->s.env->mesh->replies_sum_wait, &duration);
1123 	timehist_insert(m->s.env->mesh->histogram, &duration);
1124 	if(m->s.env->cfg->stat_extended) {
1125 		uint16_t rc = FLAGS_GET_RCODE(sldns_buffer_read_u16_at(r->
1126 			query_reply.c->buffer, 2));
1127 		if(secure) m->s.env->mesh->ans_secure++;
1128 		m->s.env->mesh->ans_rcode[ rc ] ++;
1129 		if(rc == 0 && LDNS_ANCOUNT(sldns_buffer_begin(r->
1130 			query_reply.c->buffer)) == 0)
1131 			m->s.env->mesh->ans_nodata++;
1132 	}
1133 	/* Log reply sent */
1134 	if(m->s.env->cfg->log_replies) {
1135 		log_reply_info(0, &m->s.qinfo, &r->query_reply.addr,
1136 			r->query_reply.addrlen, duration, 0,
1137 			r->query_reply.c->buffer);
1138 	}
1139 }
1140 
1141 void mesh_query_done(struct mesh_state* mstate)
1142 {
1143 	struct mesh_reply* r;
1144 	struct mesh_reply* prev = NULL;
1145 	struct mesh_cb* c;
1146 	struct reply_info* rep = (mstate->s.return_msg?
1147 		mstate->s.return_msg->rep:NULL);
1148 	for(r = mstate->reply_list; r; r = r->next) {
1149 		/* if a response-ip address block has been stored the
1150 		 *  information should be logged for each client. */
1151 		if(mstate->s.respip_action_info &&
1152 			mstate->s.respip_action_info->addrinfo) {
1153 			respip_inform_print(mstate->s.respip_action_info->addrinfo,
1154 				r->qname, mstate->s.qinfo.qtype,
1155 				mstate->s.qinfo.qclass, r->local_alias,
1156 				&r->query_reply);
1157 		}
1158 
1159 		/* if this query is determined to be dropped during the
1160 		 * mesh processing, this is the point to take that action. */
1161 		if(mstate->s.is_drop)
1162 			comm_point_drop_reply(&r->query_reply);
1163 		else {
1164 			mesh_send_reply(mstate, mstate->s.return_rcode, rep,
1165 				r, prev);
1166 			prev = r;
1167 		}
1168 	}
1169 	mstate->replies_sent = 1;
1170 	for(c = mstate->cb_list; c; c = c->next) {
1171 		mesh_do_callback(mstate, mstate->s.return_rcode, rep, c);
1172 	}
1173 }
1174 
1175 void mesh_walk_supers(struct mesh_area* mesh, struct mesh_state* mstate)
1176 {
1177 	struct mesh_state_ref* ref;
1178 	RBTREE_FOR(ref, struct mesh_state_ref*, &mstate->super_set)
1179 	{
1180 		/* make super runnable */
1181 		(void)rbtree_insert(&mesh->run, &ref->s->run_node);
1182 		/* callback the function to inform super of result */
1183 		fptr_ok(fptr_whitelist_mod_inform_super(
1184 			mesh->mods.mod[ref->s->s.curmod]->inform_super));
1185 		(*mesh->mods.mod[ref->s->s.curmod]->inform_super)(&mstate->s,
1186 			ref->s->s.curmod, &ref->s->s);
1187 	}
1188 }
1189 
1190 struct mesh_state* mesh_area_find(struct mesh_area* mesh,
1191 	struct respip_client_info* cinfo, struct query_info* qinfo,
1192 	uint16_t qflags, int prime, int valrec)
1193 {
1194 	struct mesh_state key;
1195 	struct mesh_state* result;
1196 
1197 	key.node.key = &key;
1198 	key.s.is_priming = prime;
1199 	key.s.is_valrec = valrec;
1200 	key.s.qinfo = *qinfo;
1201 	key.s.query_flags = qflags;
1202 	/* We are searching for a similar mesh state when we DO want to
1203 	 * aggregate the state. Thus unique is set to NULL. (default when we
1204 	 * desire aggregation).*/
1205 	key.unique = NULL;
1206 	key.s.client_info = cinfo;
1207 
1208 	result = (struct mesh_state*)rbtree_search(&mesh->all, &key);
1209 	return result;
1210 }
1211 
1212 int mesh_state_add_cb(struct mesh_state* s, struct edns_data* edns,
1213         sldns_buffer* buf, mesh_cb_func_type cb, void* cb_arg,
1214 	uint16_t qid, uint16_t qflags)
1215 {
1216 	struct mesh_cb* r = regional_alloc(s->s.region,
1217 		sizeof(struct mesh_cb));
1218 	if(!r)
1219 		return 0;
1220 	r->buf = buf;
1221 	log_assert(fptr_whitelist_mesh_cb(cb)); /* early failure ifmissing*/
1222 	r->cb = cb;
1223 	r->cb_arg = cb_arg;
1224 	r->edns = *edns;
1225 	if(edns->opt_list) {
1226 		r->edns.opt_list = edns_opt_copy_region(edns->opt_list,
1227 			s->s.region);
1228 		if(!r->edns.opt_list)
1229 			return 0;
1230 	}
1231 	r->qid = qid;
1232 	r->qflags = qflags;
1233 	r->next = s->cb_list;
1234 	s->cb_list = r;
1235 	return 1;
1236 
1237 }
1238 
1239 int mesh_state_add_reply(struct mesh_state* s, struct edns_data* edns,
1240         struct comm_reply* rep, uint16_t qid, uint16_t qflags,
1241         const struct query_info* qinfo)
1242 {
1243 	struct mesh_reply* r = regional_alloc(s->s.region,
1244 		sizeof(struct mesh_reply));
1245 	if(!r)
1246 		return 0;
1247 	r->query_reply = *rep;
1248 	r->edns = *edns;
1249 	if(edns->opt_list) {
1250 		r->edns.opt_list = edns_opt_copy_region(edns->opt_list,
1251 			s->s.region);
1252 		if(!r->edns.opt_list)
1253 			return 0;
1254 	}
1255 	r->qid = qid;
1256 	r->qflags = qflags;
1257 	r->start_time = *s->s.env->now_tv;
1258 	r->next = s->reply_list;
1259 	r->qname = regional_alloc_init(s->s.region, qinfo->qname,
1260 		s->s.qinfo.qname_len);
1261 	if(!r->qname)
1262 		return 0;
1263 
1264 	/* Data related to local alias stored in 'qinfo' (if any) is ephemeral
1265 	 * and can be different for different original queries (even if the
1266 	 * replaced query name is the same).  So we need to make a deep copy
1267 	 * and store the copy for each reply info. */
1268 	if(qinfo->local_alias) {
1269 		struct packed_rrset_data* d;
1270 		struct packed_rrset_data* dsrc;
1271 		r->local_alias = regional_alloc_zero(s->s.region,
1272 			sizeof(*qinfo->local_alias));
1273 		if(!r->local_alias)
1274 			return 0;
1275 		r->local_alias->rrset = regional_alloc_init(s->s.region,
1276 			qinfo->local_alias->rrset,
1277 			sizeof(*qinfo->local_alias->rrset));
1278 		if(!r->local_alias->rrset)
1279 			return 0;
1280 		dsrc = qinfo->local_alias->rrset->entry.data;
1281 
1282 		/* In the current implementation, a local alias must be
1283 		 * a single CNAME RR (see worker_handle_request()). */
1284 		log_assert(!qinfo->local_alias->next && dsrc->count == 1 &&
1285 			qinfo->local_alias->rrset->rk.type ==
1286 			htons(LDNS_RR_TYPE_CNAME));
1287 		/* Technically, we should make a local copy for the owner
1288 		 * name of the RRset, but in the case of the first (and
1289 		 * currently only) local alias RRset, the owner name should
1290 		 * point to the qname of the corresponding query, which should
1291 		 * be valid throughout the lifetime of this mesh_reply.  So
1292 		 * we can skip copying. */
1293 		log_assert(qinfo->local_alias->rrset->rk.dname ==
1294 			sldns_buffer_at(rep->c->buffer, LDNS_HEADER_SIZE));
1295 
1296 		d = regional_alloc_init(s->s.region, dsrc,
1297 			sizeof(struct packed_rrset_data)
1298 			+ sizeof(size_t) + sizeof(uint8_t*) + sizeof(time_t));
1299 		if(!d)
1300 			return 0;
1301 		r->local_alias->rrset->entry.data = d;
1302 		d->rr_len = (size_t*)((uint8_t*)d +
1303 			sizeof(struct packed_rrset_data));
1304 		d->rr_data = (uint8_t**)&(d->rr_len[1]);
1305 		d->rr_ttl = (time_t*)&(d->rr_data[1]);
1306 		d->rr_len[0] = dsrc->rr_len[0];
1307 		d->rr_ttl[0] = dsrc->rr_ttl[0];
1308 		d->rr_data[0] = regional_alloc_init(s->s.region,
1309 			dsrc->rr_data[0], d->rr_len[0]);
1310 		if(!d->rr_data[0])
1311 			return 0;
1312 	} else
1313 		r->local_alias = NULL;
1314 
1315 	s->reply_list = r;
1316 	return 1;
1317 }
1318 
1319 /* Extract the query info and flags from 'mstate' into '*qinfop' and '*qflags'.
1320  * Since this is only used for internal refetch of otherwise-expired answer,
1321  * we simply ignore the rare failure mode when memory allocation fails. */
1322 static void
1323 mesh_copy_qinfo(struct mesh_state* mstate, struct query_info** qinfop,
1324 	uint16_t* qflags)
1325 {
1326 	struct regional* region = mstate->s.env->scratch;
1327 	struct query_info* qinfo;
1328 
1329 	qinfo = regional_alloc_init(region, &mstate->s.qinfo, sizeof(*qinfo));
1330 	if(!qinfo)
1331 		return;
1332 	qinfo->qname = regional_alloc_init(region, qinfo->qname,
1333 		qinfo->qname_len);
1334 	if(!qinfo->qname)
1335 		return;
1336 	*qinfop = qinfo;
1337 	*qflags = mstate->s.query_flags;
1338 }
1339 
1340 /**
1341  * Continue processing the mesh state at another module.
1342  * Handles module to modules transfer of control.
1343  * Handles module finished.
1344  * @param mesh: the mesh area.
1345  * @param mstate: currently active mesh state.
1346  * 	Deleted if finished, calls _done and _supers to
1347  * 	send replies to clients and inform other mesh states.
1348  * 	This in turn may create additional runnable mesh states.
1349  * @param s: state at which the current module exited.
1350  * @param ev: the event sent to the module.
1351  * 	returned is the event to send to the next module.
1352  * @return true if continue processing at the new module.
1353  * 	false if not continued processing is needed.
1354  */
1355 static int
1356 mesh_continue(struct mesh_area* mesh, struct mesh_state* mstate,
1357 	enum module_ext_state s, enum module_ev* ev)
1358 {
1359 	mstate->num_activated++;
1360 	if(mstate->num_activated > MESH_MAX_ACTIVATION) {
1361 		/* module is looping. Stop it. */
1362 		log_err("internal error: looping module (%s) stopped",
1363 			mesh->mods.mod[mstate->s.curmod]->name);
1364 		log_query_info(VERB_QUERY, "pass error for qstate",
1365 			&mstate->s.qinfo);
1366 		s = module_error;
1367 	}
1368 	if(s == module_wait_module || s == module_restart_next) {
1369 		/* start next module */
1370 		mstate->s.curmod++;
1371 		if(mesh->mods.num == mstate->s.curmod) {
1372 			log_err("Cannot pass to next module; at last module");
1373 			log_query_info(VERB_QUERY, "pass error for qstate",
1374 				&mstate->s.qinfo);
1375 			mstate->s.curmod--;
1376 			return mesh_continue(mesh, mstate, module_error, ev);
1377 		}
1378 		if(s == module_restart_next) {
1379 			int curmod = mstate->s.curmod;
1380 			for(; mstate->s.curmod < mesh->mods.num;
1381 				mstate->s.curmod++) {
1382 				fptr_ok(fptr_whitelist_mod_clear(
1383 					mesh->mods.mod[mstate->s.curmod]->clear));
1384 				(*mesh->mods.mod[mstate->s.curmod]->clear)
1385 					(&mstate->s, mstate->s.curmod);
1386 				mstate->s.minfo[mstate->s.curmod] = NULL;
1387 			}
1388 			mstate->s.curmod = curmod;
1389 		}
1390 		*ev = module_event_pass;
1391 		return 1;
1392 	}
1393 	if(s == module_wait_subquery && mstate->sub_set.count == 0) {
1394 		log_err("module cannot wait for subquery, subquery list empty");
1395 		log_query_info(VERB_QUERY, "pass error for qstate",
1396 			&mstate->s.qinfo);
1397 		s = module_error;
1398 	}
1399 	if(s == module_error && mstate->s.return_rcode == LDNS_RCODE_NOERROR) {
1400 		/* error is bad, handle pass back up below */
1401 		mstate->s.return_rcode = LDNS_RCODE_SERVFAIL;
1402 	}
1403 	if(s == module_error) {
1404 		mesh_query_done(mstate);
1405 		mesh_walk_supers(mesh, mstate);
1406 		mesh_state_delete(&mstate->s);
1407 		return 0;
1408 	}
1409 	if(s == module_finished) {
1410 		if(mstate->s.curmod == 0) {
1411 			struct query_info* qinfo = NULL;
1412 			uint16_t qflags;
1413 
1414 			mesh_query_done(mstate);
1415 			mesh_walk_supers(mesh, mstate);
1416 
1417 			/* If the answer to the query needs to be refetched
1418 			 * from an external DNS server, we'll need to schedule
1419 			 * a prefetch after removing the current state, so
1420 			 * we need to make a copy of the query info here. */
1421 			if(mstate->s.need_refetch)
1422 				mesh_copy_qinfo(mstate, &qinfo, &qflags);
1423 
1424 			mesh_state_delete(&mstate->s);
1425 			if(qinfo) {
1426 				mesh_schedule_prefetch(mesh, qinfo, qflags,
1427 					0, 1);
1428 			}
1429 			return 0;
1430 		}
1431 		/* pass along the locus of control */
1432 		mstate->s.curmod --;
1433 		*ev = module_event_moddone;
1434 		return 1;
1435 	}
1436 	return 0;
1437 }
1438 
1439 void mesh_run(struct mesh_area* mesh, struct mesh_state* mstate,
1440 	enum module_ev ev, struct outbound_entry* e)
1441 {
1442 	enum module_ext_state s;
1443 	verbose(VERB_ALGO, "mesh_run: start");
1444 	while(mstate) {
1445 		/* run the module */
1446 		fptr_ok(fptr_whitelist_mod_operate(
1447 			mesh->mods.mod[mstate->s.curmod]->operate));
1448 		(*mesh->mods.mod[mstate->s.curmod]->operate)
1449 			(&mstate->s, ev, mstate->s.curmod, e);
1450 
1451 		/* examine results */
1452 		mstate->s.reply = NULL;
1453 		regional_free_all(mstate->s.env->scratch);
1454 		s = mstate->s.ext_state[mstate->s.curmod];
1455 		verbose(VERB_ALGO, "mesh_run: %s module exit state is %s",
1456 			mesh->mods.mod[mstate->s.curmod]->name, strextstate(s));
1457 		e = NULL;
1458 		if(mesh_continue(mesh, mstate, s, &ev))
1459 			continue;
1460 
1461 		/* run more modules */
1462 		ev = module_event_pass;
1463 		if(mesh->run.count > 0) {
1464 			/* pop random element off the runnable tree */
1465 			mstate = (struct mesh_state*)mesh->run.root->key;
1466 			(void)rbtree_delete(&mesh->run, mstate);
1467 		} else mstate = NULL;
1468 	}
1469 	if(verbosity >= VERB_ALGO) {
1470 		mesh_stats(mesh, "mesh_run: end");
1471 		mesh_log_list(mesh);
1472 	}
1473 }
1474 
1475 void
1476 mesh_log_list(struct mesh_area* mesh)
1477 {
1478 	char buf[30];
1479 	struct mesh_state* m;
1480 	int num = 0;
1481 	RBTREE_FOR(m, struct mesh_state*, &mesh->all) {
1482 		snprintf(buf, sizeof(buf), "%d%s%s%s%s%s%s mod%d %s%s",
1483 			num++, (m->s.is_priming)?"p":"",  /* prime */
1484 			(m->s.is_valrec)?"v":"",  /* prime */
1485 			(m->s.query_flags&BIT_RD)?"RD":"",
1486 			(m->s.query_flags&BIT_CD)?"CD":"",
1487 			(m->super_set.count==0)?"d":"", /* detached */
1488 			(m->sub_set.count!=0)?"c":"",  /* children */
1489 			m->s.curmod, (m->reply_list)?"rep":"", /*hasreply*/
1490 			(m->cb_list)?"cb":"" /* callbacks */
1491 			);
1492 		log_query_info(VERB_ALGO, buf, &m->s.qinfo);
1493 	}
1494 }
1495 
1496 void
1497 mesh_stats(struct mesh_area* mesh, const char* str)
1498 {
1499 	verbose(VERB_DETAIL, "%s %u recursion states (%u with reply, "
1500 		"%u detached), %u waiting replies, %u recursion replies "
1501 		"sent, %d replies dropped, %d states jostled out",
1502 		str, (unsigned)mesh->all.count,
1503 		(unsigned)mesh->num_reply_states,
1504 		(unsigned)mesh->num_detached_states,
1505 		(unsigned)mesh->num_reply_addrs,
1506 		(unsigned)mesh->replies_sent,
1507 		(unsigned)mesh->stats_dropped,
1508 		(unsigned)mesh->stats_jostled);
1509 	if(mesh->replies_sent > 0) {
1510 		struct timeval avg;
1511 		timeval_divide(&avg, &mesh->replies_sum_wait,
1512 			mesh->replies_sent);
1513 		log_info("average recursion processing time "
1514 			ARG_LL "d.%6.6d sec",
1515 			(long long)avg.tv_sec, (int)avg.tv_usec);
1516 		log_info("histogram of recursion processing times");
1517 		timehist_log(mesh->histogram, "recursions");
1518 	}
1519 }
1520 
1521 void
1522 mesh_stats_clear(struct mesh_area* mesh)
1523 {
1524 	if(!mesh)
1525 		return;
1526 	mesh->replies_sent = 0;
1527 	mesh->replies_sum_wait.tv_sec = 0;
1528 	mesh->replies_sum_wait.tv_usec = 0;
1529 	mesh->stats_jostled = 0;
1530 	mesh->stats_dropped = 0;
1531 	timehist_clear(mesh->histogram);
1532 	mesh->ans_secure = 0;
1533 	mesh->ans_bogus = 0;
1534 	memset(&mesh->ans_rcode[0], 0, sizeof(size_t)*16);
1535 	mesh->ans_nodata = 0;
1536 }
1537 
1538 size_t
1539 mesh_get_mem(struct mesh_area* mesh)
1540 {
1541 	struct mesh_state* m;
1542 	size_t s = sizeof(*mesh) + sizeof(struct timehist) +
1543 		sizeof(struct th_buck)*mesh->histogram->num +
1544 		sizeof(sldns_buffer) + sldns_buffer_capacity(mesh->qbuf_bak);
1545 	RBTREE_FOR(m, struct mesh_state*, &mesh->all) {
1546 		/* all, including m itself allocated in qstate region */
1547 		s += regional_get_mem(m->s.region);
1548 	}
1549 	return s;
1550 }
1551 
1552 int
1553 mesh_detect_cycle(struct module_qstate* qstate, struct query_info* qinfo,
1554 	uint16_t flags, int prime, int valrec)
1555 {
1556 	struct mesh_area* mesh = qstate->env->mesh;
1557 	struct mesh_state* dep_m = NULL;
1558 	if(!mesh_state_is_unique(qstate->mesh_info))
1559 		dep_m = mesh_area_find(mesh, NULL, qinfo, flags, prime, valrec);
1560 	return mesh_detect_cycle_found(qstate, dep_m);
1561 }
1562 
1563 void mesh_list_insert(struct mesh_state* m, struct mesh_state** fp,
1564         struct mesh_state** lp)
1565 {
1566 	/* insert as last element */
1567 	m->prev = *lp;
1568 	m->next = NULL;
1569 	if(*lp)
1570 		(*lp)->next = m;
1571 	else	*fp = m;
1572 	*lp = m;
1573 }
1574 
1575 void mesh_list_remove(struct mesh_state* m, struct mesh_state** fp,
1576         struct mesh_state** lp)
1577 {
1578 	if(m->next)
1579 		m->next->prev = m->prev;
1580 	else	*lp = m->prev;
1581 	if(m->prev)
1582 		m->prev->next = m->next;
1583 	else	*fp = m->next;
1584 }
1585