]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/blob - contrib/unbound/services/mesh.c
openssh: cherry-pick OpenSSL 1.1.1 compatibility
[FreeBSD/FreeBSD.git] / contrib / unbound / services / mesh.c
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                 while((cb = mstate->cb_list)!=NULL) {
740                         mstate->cb_list = cb->next;
741                         fptr_ok(fptr_whitelist_mesh_cb(cb->cb));
742                         (*cb->cb)(cb->cb_arg, LDNS_RCODE_SERVFAIL, NULL,
743                                 sec_status_unchecked, NULL);
744                         mesh->num_reply_addrs--;
745                 }
746         }
747
748         /* de-init modules */
749         for(i=0; i<mesh->mods.num; i++) {
750                 fptr_ok(fptr_whitelist_mod_clear(mesh->mods.mod[i]->clear));
751                 (*mesh->mods.mod[i]->clear)(&mstate->s, i);
752                 mstate->s.minfo[i] = NULL;
753                 mstate->s.ext_state[i] = module_finished;
754         }
755         alloc_reg_release(mstate->s.env->alloc, mstate->s.region);
756 }
757
758 void 
759 mesh_state_delete(struct module_qstate* qstate)
760 {
761         struct mesh_area* mesh;
762         struct mesh_state_ref* super, ref;
763         struct mesh_state* mstate;
764         if(!qstate)
765                 return;
766         mstate = qstate->mesh_info;
767         mesh = mstate->s.env->mesh;
768         mesh_detach_subs(&mstate->s);
769         if(mstate->list_select == mesh_forever_list) {
770                 mesh->num_forever_states --;
771                 mesh_list_remove(mstate, &mesh->forever_first, 
772                         &mesh->forever_last);
773         } else if(mstate->list_select == mesh_jostle_list) {
774                 mesh_list_remove(mstate, &mesh->jostle_first, 
775                         &mesh->jostle_last);
776         }
777         if(!mstate->reply_list && !mstate->cb_list
778                 && mstate->super_set.count == 0) {
779                 log_assert(mesh->num_detached_states > 0);
780                 mesh->num_detached_states--;
781         }
782         if(mstate->reply_list || mstate->cb_list) {
783                 log_assert(mesh->num_reply_states > 0);
784                 mesh->num_reply_states--;
785         }
786         ref.node.key = &ref;
787         ref.s = mstate;
788         RBTREE_FOR(super, struct mesh_state_ref*, &mstate->super_set) {
789                 (void)rbtree_delete(&super->s->sub_set, &ref);
790         }
791         (void)rbtree_delete(&mesh->run, mstate);
792         (void)rbtree_delete(&mesh->all, mstate);
793         mesh_state_cleanup(mstate);
794 }
795
796 /** helper recursive rbtree find routine */
797 static int
798 find_in_subsub(struct mesh_state* m, struct mesh_state* tofind, size_t *c)
799 {
800         struct mesh_state_ref* r;
801         if((*c)++ > MESH_MAX_SUBSUB)
802                 return 1;
803         RBTREE_FOR(r, struct mesh_state_ref*, &m->sub_set) {
804                 if(r->s == tofind || find_in_subsub(r->s, tofind, c))
805                         return 1;
806         }
807         return 0;
808 }
809
810 /** find cycle for already looked up mesh_state */
811 static int 
812 mesh_detect_cycle_found(struct module_qstate* qstate, struct mesh_state* dep_m)
813 {
814         struct mesh_state* cyc_m = qstate->mesh_info;
815         size_t counter = 0;
816         if(!dep_m)
817                 return 0;
818         if(dep_m == cyc_m || find_in_subsub(dep_m, cyc_m, &counter)) {
819                 if(counter > MESH_MAX_SUBSUB)
820                         return 2;
821                 return 1;
822         }
823         return 0;
824 }
825
826 void mesh_detach_subs(struct module_qstate* qstate)
827 {
828         struct mesh_area* mesh = qstate->env->mesh;
829         struct mesh_state_ref* ref, lookup;
830 #ifdef UNBOUND_DEBUG
831         struct rbnode_type* n;
832 #endif
833         lookup.node.key = &lookup;
834         lookup.s = qstate->mesh_info;
835         RBTREE_FOR(ref, struct mesh_state_ref*, &qstate->mesh_info->sub_set) {
836 #ifdef UNBOUND_DEBUG
837                 n =
838 #else
839                 (void)
840 #endif
841                 rbtree_delete(&ref->s->super_set, &lookup);
842                 log_assert(n != NULL); /* must have been present */
843                 if(!ref->s->reply_list && !ref->s->cb_list
844                         && ref->s->super_set.count == 0) {
845                         mesh->num_detached_states++;
846                         log_assert(mesh->num_detached_states + 
847                                 mesh->num_reply_states <= mesh->all.count);
848                 }
849         }
850         rbtree_init(&qstate->mesh_info->sub_set, &mesh_state_ref_compare);
851 }
852
853 int mesh_add_sub(struct module_qstate* qstate, struct query_info* qinfo,
854         uint16_t qflags, int prime, int valrec, struct module_qstate** newq,
855         struct mesh_state** sub)
856 {
857         /* find it, if not, create it */
858         struct mesh_area* mesh = qstate->env->mesh;
859         *sub = mesh_area_find(mesh, NULL, qinfo, qflags,
860                 prime, valrec);
861         if(mesh_detect_cycle_found(qstate, *sub)) {
862                 verbose(VERB_ALGO, "attach failed, cycle detected");
863                 return 0;
864         }
865         if(!*sub) {
866 #ifdef UNBOUND_DEBUG
867                 struct rbnode_type* n;
868 #endif
869                 /* create a new one */
870                 *sub = mesh_state_create(qstate->env, qinfo, NULL, qflags, prime,
871                         valrec);
872                 if(!*sub) {
873                         log_err("mesh_attach_sub: out of memory");
874                         return 0;
875                 }
876 #ifdef UNBOUND_DEBUG
877                 n =
878 #else
879                 (void)
880 #endif
881                 rbtree_insert(&mesh->all, &(*sub)->node);
882                 log_assert(n != NULL);
883                 /* set detached (it is now) */
884                 mesh->num_detached_states++;
885                 /* set new query state to run */
886 #ifdef UNBOUND_DEBUG
887                 n =
888 #else
889                 (void)
890 #endif
891                 rbtree_insert(&mesh->run, &(*sub)->run_node);
892                 log_assert(n != NULL);
893                 *newq = &(*sub)->s;
894         } else
895                 *newq = NULL;
896         return 1;
897 }
898
899 int mesh_attach_sub(struct module_qstate* qstate, struct query_info* qinfo,
900         uint16_t qflags, int prime, int valrec, struct module_qstate** newq)
901 {
902         struct mesh_area* mesh = qstate->env->mesh;
903         struct mesh_state* sub = NULL;
904         int was_detached;
905         if(!mesh_add_sub(qstate, qinfo, qflags, prime, valrec, newq, &sub))
906                 return 0;
907         was_detached = (sub->super_set.count == 0);
908         if(!mesh_state_attachment(qstate->mesh_info, sub))
909                 return 0;
910         /* if it was a duplicate  attachment, the count was not zero before */
911         if(!sub->reply_list && !sub->cb_list && was_detached && 
912                 sub->super_set.count == 1) {
913                 /* it used to be detached, before this one got added */
914                 log_assert(mesh->num_detached_states > 0);
915                 mesh->num_detached_states--;
916         }
917         /* *newq will be run when inited after the current module stops */
918         return 1;
919 }
920
921 int mesh_state_attachment(struct mesh_state* super, struct mesh_state* sub)
922 {
923 #ifdef UNBOUND_DEBUG
924         struct rbnode_type* n;
925 #endif
926         struct mesh_state_ref* subref; /* points to sub, inserted in super */
927         struct mesh_state_ref* superref; /* points to super, inserted in sub */
928         if( !(subref = regional_alloc(super->s.region,
929                 sizeof(struct mesh_state_ref))) ||
930                 !(superref = regional_alloc(sub->s.region,
931                 sizeof(struct mesh_state_ref))) ) {
932                 log_err("mesh_state_attachment: out of memory");
933                 return 0;
934         }
935         superref->node.key = superref;
936         superref->s = super;
937         subref->node.key = subref;
938         subref->s = sub;
939         if(!rbtree_insert(&sub->super_set, &superref->node)) {
940                 /* this should not happen, iterator and validator do not
941                  * attach subqueries that are identical. */
942                 /* already attached, we are done, nothing todo.
943                  * since superref and subref already allocated in region,
944                  * we cannot free them */
945                 return 1;
946         }
947 #ifdef UNBOUND_DEBUG
948         n =
949 #else
950         (void)
951 #endif
952         rbtree_insert(&super->sub_set, &subref->node);
953         log_assert(n != NULL); /* we checked above if statement, the reverse
954           administration should not fail now, unless they are out of sync */
955         return 1;
956 }
957
958 /**
959  * callback results to mesh cb entry
960  * @param m: mesh state to send it for.
961  * @param rcode: if not 0, error code.
962  * @param rep: reply to send (or NULL if rcode is set).
963  * @param r: callback entry
964  */
965 static void
966 mesh_do_callback(struct mesh_state* m, int rcode, struct reply_info* rep,
967         struct mesh_cb* r)
968 {
969         int secure;
970         char* reason = NULL;
971         /* bogus messages are not made into servfail, sec_status passed 
972          * to the callback function */
973         if(rep && rep->security == sec_status_secure)
974                 secure = 1;
975         else    secure = 0;
976         if(!rep && rcode == LDNS_RCODE_NOERROR)
977                 rcode = LDNS_RCODE_SERVFAIL;
978         if(!rcode && (rep->security == sec_status_bogus ||
979                 rep->security == sec_status_secure_sentinel_fail)) {
980                 if(!(reason = errinf_to_str(&m->s)))
981                         rcode = LDNS_RCODE_SERVFAIL;
982         }
983         /* send the reply */
984         if(rcode) {
985                 if(rcode == LDNS_RCODE_SERVFAIL) {
986                         if(!inplace_cb_reply_servfail_call(m->s.env, &m->s.qinfo, &m->s,
987                                 rep, rcode, &r->edns, m->s.region))
988                                         r->edns.opt_list = NULL;
989                 } else {
990                         if(!inplace_cb_reply_call(m->s.env, &m->s.qinfo, &m->s, rep, rcode,
991                                 &r->edns, m->s.region))
992                                         r->edns.opt_list = NULL;
993                 }
994                 fptr_ok(fptr_whitelist_mesh_cb(r->cb));
995                 (*r->cb)(r->cb_arg, rcode, r->buf, sec_status_unchecked, NULL);
996         } else {
997                 size_t udp_size = r->edns.udp_size;
998                 sldns_buffer_clear(r->buf);
999                 r->edns.edns_version = EDNS_ADVERTISED_VERSION;
1000                 r->edns.udp_size = EDNS_ADVERTISED_SIZE;
1001                 r->edns.ext_rcode = 0;
1002                 r->edns.bits &= EDNS_DO;
1003
1004                 if(!inplace_cb_reply_call(m->s.env, &m->s.qinfo, &m->s, rep,
1005                         LDNS_RCODE_NOERROR, &r->edns, m->s.region) ||
1006                         !reply_info_answer_encode(&m->s.qinfo, rep, r->qid, 
1007                         r->qflags, r->buf, 0, 1, 
1008                         m->s.env->scratch, udp_size, &r->edns, 
1009                         (int)(r->edns.bits & EDNS_DO), secure)) 
1010                 {
1011                         fptr_ok(fptr_whitelist_mesh_cb(r->cb));
1012                         (*r->cb)(r->cb_arg, LDNS_RCODE_SERVFAIL, r->buf,
1013                                 sec_status_unchecked, NULL);
1014                 } else {
1015                         fptr_ok(fptr_whitelist_mesh_cb(r->cb));
1016                         (*r->cb)(r->cb_arg, LDNS_RCODE_NOERROR, r->buf,
1017                                 rep->security, reason);
1018                 }
1019         }
1020         free(reason);
1021         m->s.env->mesh->num_reply_addrs--;
1022 }
1023
1024 /**
1025  * Send reply to mesh reply entry
1026  * @param m: mesh state to send it for.
1027  * @param rcode: if not 0, error code.
1028  * @param rep: reply to send (or NULL if rcode is set).
1029  * @param r: reply entry
1030  * @param prev: previous reply, already has its answer encoded in buffer.
1031  */
1032 static void
1033 mesh_send_reply(struct mesh_state* m, int rcode, struct reply_info* rep,
1034         struct mesh_reply* r, struct mesh_reply* prev)
1035 {
1036         struct timeval end_time;
1037         struct timeval duration;
1038         int secure;
1039         /* Copy the client's EDNS for later restore, to make sure the edns
1040          * compare is with the correct edns options. */
1041         struct edns_data edns_bak = r->edns;
1042         /* examine security status */
1043         if(m->s.env->need_to_validate && (!(r->qflags&BIT_CD) ||
1044                 m->s.env->cfg->ignore_cd) && rep && 
1045                 (rep->security <= sec_status_bogus ||
1046                 rep->security == sec_status_secure_sentinel_fail)) {
1047                 rcode = LDNS_RCODE_SERVFAIL;
1048                 if(m->s.env->cfg->stat_extended) 
1049                         m->s.env->mesh->ans_bogus++;
1050         }
1051         if(rep && rep->security == sec_status_secure)
1052                 secure = 1;
1053         else    secure = 0;
1054         if(!rep && rcode == LDNS_RCODE_NOERROR)
1055                 rcode = LDNS_RCODE_SERVFAIL;
1056         /* send the reply */
1057         /* We don't reuse the encoded answer if either the previous or current
1058          * response has a local alias.  We could compare the alias records
1059          * and still reuse the previous answer if they are the same, but that
1060          * would be complicated and error prone for the relatively minor case.
1061          * So we err on the side of safety. */
1062         if(prev && prev->qflags == r->qflags && 
1063                 !prev->local_alias && !r->local_alias &&
1064                 prev->edns.edns_present == r->edns.edns_present && 
1065                 prev->edns.bits == r->edns.bits && 
1066                 prev->edns.udp_size == r->edns.udp_size &&
1067                 edns_opt_list_compare(prev->edns.opt_list, r->edns.opt_list)
1068                 == 0) {
1069                 /* if the previous reply is identical to this one, fix ID */
1070                 if(prev->query_reply.c->buffer != r->query_reply.c->buffer)
1071                         sldns_buffer_copy(r->query_reply.c->buffer, 
1072                                 prev->query_reply.c->buffer);
1073                 sldns_buffer_write_at(r->query_reply.c->buffer, 0, 
1074                         &r->qid, sizeof(uint16_t));
1075                 sldns_buffer_write_at(r->query_reply.c->buffer, 12, 
1076                         r->qname, m->s.qinfo.qname_len);
1077                 comm_point_send_reply(&r->query_reply);
1078         } else if(rcode) {
1079                 m->s.qinfo.qname = r->qname;
1080                 m->s.qinfo.local_alias = r->local_alias;
1081                 if(rcode == LDNS_RCODE_SERVFAIL) {
1082                         if(!inplace_cb_reply_servfail_call(m->s.env, &m->s.qinfo, &m->s,
1083                                 rep, rcode, &r->edns, m->s.region))
1084                                         r->edns.opt_list = NULL;
1085                 } else { 
1086                         if(!inplace_cb_reply_call(m->s.env, &m->s.qinfo, &m->s, rep, rcode,
1087                                 &r->edns, m->s.region))
1088                                         r->edns.opt_list = NULL;
1089                 }
1090                 error_encode(r->query_reply.c->buffer, rcode, &m->s.qinfo,
1091                         r->qid, r->qflags, &r->edns);
1092                 comm_point_send_reply(&r->query_reply);
1093         } else {
1094                 size_t udp_size = r->edns.udp_size;
1095                 r->edns.edns_version = EDNS_ADVERTISED_VERSION;
1096                 r->edns.udp_size = EDNS_ADVERTISED_SIZE;
1097                 r->edns.ext_rcode = 0;
1098                 r->edns.bits &= EDNS_DO;
1099                 m->s.qinfo.qname = r->qname;
1100                 m->s.qinfo.local_alias = r->local_alias;
1101                 if(!inplace_cb_reply_call(m->s.env, &m->s.qinfo, &m->s, rep,
1102                         LDNS_RCODE_NOERROR, &r->edns, m->s.region) ||
1103                         !reply_info_answer_encode(&m->s.qinfo, rep, r->qid, 
1104                         r->qflags, r->query_reply.c->buffer, 0, 1, 
1105                         m->s.env->scratch, udp_size, &r->edns, 
1106                         (int)(r->edns.bits & EDNS_DO), secure)) 
1107                 {
1108                         if(!inplace_cb_reply_servfail_call(m->s.env, &m->s.qinfo, &m->s,
1109                         rep, LDNS_RCODE_SERVFAIL, &r->edns, m->s.region))
1110                                 r->edns.opt_list = NULL;
1111                         error_encode(r->query_reply.c->buffer, 
1112                                 LDNS_RCODE_SERVFAIL, &m->s.qinfo, r->qid, 
1113                                 r->qflags, &r->edns);
1114                 }
1115                 r->edns = edns_bak;
1116                 comm_point_send_reply(&r->query_reply);
1117         }
1118         /* account */
1119         m->s.env->mesh->num_reply_addrs--;
1120         end_time = *m->s.env->now_tv;
1121         timeval_subtract(&duration, &end_time, &r->start_time);
1122         verbose(VERB_ALGO, "query took " ARG_LL "d.%6.6d sec",
1123                 (long long)duration.tv_sec, (int)duration.tv_usec);
1124         m->s.env->mesh->replies_sent++;
1125         timeval_add(&m->s.env->mesh->replies_sum_wait, &duration);
1126         timehist_insert(m->s.env->mesh->histogram, &duration);
1127         if(m->s.env->cfg->stat_extended) {
1128                 uint16_t rc = FLAGS_GET_RCODE(sldns_buffer_read_u16_at(r->
1129                         query_reply.c->buffer, 2));
1130                 if(secure) m->s.env->mesh->ans_secure++;
1131                 m->s.env->mesh->ans_rcode[ rc ] ++;
1132                 if(rc == 0 && LDNS_ANCOUNT(sldns_buffer_begin(r->
1133                         query_reply.c->buffer)) == 0)
1134                         m->s.env->mesh->ans_nodata++;
1135         }
1136         /* Log reply sent */
1137         if(m->s.env->cfg->log_replies) {
1138                 log_reply_info(0, &m->s.qinfo, &r->query_reply.addr,
1139                         r->query_reply.addrlen, duration, 0,
1140                         r->query_reply.c->buffer);
1141         }
1142 }
1143
1144 void mesh_query_done(struct mesh_state* mstate)
1145 {
1146         struct mesh_reply* r;
1147         struct mesh_reply* prev = NULL;
1148         struct mesh_cb* c;
1149         struct reply_info* rep = (mstate->s.return_msg?
1150                 mstate->s.return_msg->rep:NULL);
1151         for(r = mstate->reply_list; r; r = r->next) {
1152                 /* if a response-ip address block has been stored the
1153                  *  information should be logged for each client. */
1154                 if(mstate->s.respip_action_info &&
1155                         mstate->s.respip_action_info->addrinfo) {
1156                         respip_inform_print(mstate->s.respip_action_info->addrinfo,
1157                                 r->qname, mstate->s.qinfo.qtype,
1158                                 mstate->s.qinfo.qclass, r->local_alias,
1159                                 &r->query_reply);
1160                 }
1161
1162                 /* if this query is determined to be dropped during the
1163                  * mesh processing, this is the point to take that action. */
1164                 if(mstate->s.is_drop)
1165                         comm_point_drop_reply(&r->query_reply);
1166                 else {
1167                         mesh_send_reply(mstate, mstate->s.return_rcode, rep,
1168                                 r, prev);
1169                         prev = r;
1170                 }
1171         }
1172         mstate->replies_sent = 1;
1173         while((c = mstate->cb_list) != NULL) {
1174                 /* take this cb off the list; so that the list can be
1175                  * changed, eg. by adds from the callback routine */
1176                 if(!mstate->reply_list && mstate->cb_list && !c->next) {
1177                         /* was a reply state, not anymore */
1178                         mstate->s.env->mesh->num_reply_states--;
1179                 }
1180                 mstate->cb_list = c->next;
1181                 if(!mstate->reply_list && !mstate->cb_list &&
1182                         mstate->super_set.count == 0)
1183                         mstate->s.env->mesh->num_detached_states++;
1184                 mesh_do_callback(mstate, mstate->s.return_rcode, rep, c);
1185         }
1186 }
1187
1188 void mesh_walk_supers(struct mesh_area* mesh, struct mesh_state* mstate)
1189 {
1190         struct mesh_state_ref* ref;
1191         RBTREE_FOR(ref, struct mesh_state_ref*, &mstate->super_set)
1192         {
1193                 /* make super runnable */
1194                 (void)rbtree_insert(&mesh->run, &ref->s->run_node);
1195                 /* callback the function to inform super of result */
1196                 fptr_ok(fptr_whitelist_mod_inform_super(
1197                         mesh->mods.mod[ref->s->s.curmod]->inform_super));
1198                 (*mesh->mods.mod[ref->s->s.curmod]->inform_super)(&mstate->s, 
1199                         ref->s->s.curmod, &ref->s->s);
1200         }
1201 }
1202
1203 struct mesh_state* mesh_area_find(struct mesh_area* mesh,
1204         struct respip_client_info* cinfo, struct query_info* qinfo,
1205         uint16_t qflags, int prime, int valrec)
1206 {
1207         struct mesh_state key;
1208         struct mesh_state* result;
1209
1210         key.node.key = &key;
1211         key.s.is_priming = prime;
1212         key.s.is_valrec = valrec;
1213         key.s.qinfo = *qinfo;
1214         key.s.query_flags = qflags;
1215         /* We are searching for a similar mesh state when we DO want to
1216          * aggregate the state. Thus unique is set to NULL. (default when we
1217          * desire aggregation).*/
1218         key.unique = NULL;
1219         key.s.client_info = cinfo;
1220         
1221         result = (struct mesh_state*)rbtree_search(&mesh->all, &key);
1222         return result;
1223 }
1224
1225 int mesh_state_add_cb(struct mesh_state* s, struct edns_data* edns,
1226         sldns_buffer* buf, mesh_cb_func_type cb, void* cb_arg,
1227         uint16_t qid, uint16_t qflags)
1228 {
1229         struct mesh_cb* r = regional_alloc(s->s.region, 
1230                 sizeof(struct mesh_cb));
1231         if(!r)
1232                 return 0;
1233         r->buf = buf;
1234         log_assert(fptr_whitelist_mesh_cb(cb)); /* early failure ifmissing*/
1235         r->cb = cb;
1236         r->cb_arg = cb_arg;
1237         r->edns = *edns;
1238         if(edns->opt_list) {
1239                 r->edns.opt_list = edns_opt_copy_region(edns->opt_list,
1240                         s->s.region);
1241                 if(!r->edns.opt_list)
1242                         return 0;
1243         }
1244         r->qid = qid;
1245         r->qflags = qflags;
1246         r->next = s->cb_list;
1247         s->cb_list = r;
1248         return 1;
1249
1250 }
1251
1252 int mesh_state_add_reply(struct mesh_state* s, struct edns_data* edns,
1253         struct comm_reply* rep, uint16_t qid, uint16_t qflags,
1254         const struct query_info* qinfo)
1255 {
1256         struct mesh_reply* r = regional_alloc(s->s.region, 
1257                 sizeof(struct mesh_reply));
1258         if(!r)
1259                 return 0;
1260         r->query_reply = *rep;
1261         r->edns = *edns;
1262         if(edns->opt_list) {
1263                 r->edns.opt_list = edns_opt_copy_region(edns->opt_list,
1264                         s->s.region);
1265                 if(!r->edns.opt_list)
1266                         return 0;
1267         }
1268         r->qid = qid;
1269         r->qflags = qflags;
1270         r->start_time = *s->s.env->now_tv;
1271         r->next = s->reply_list;
1272         r->qname = regional_alloc_init(s->s.region, qinfo->qname,
1273                 s->s.qinfo.qname_len);
1274         if(!r->qname)
1275                 return 0;
1276
1277         /* Data related to local alias stored in 'qinfo' (if any) is ephemeral
1278          * and can be different for different original queries (even if the
1279          * replaced query name is the same).  So we need to make a deep copy
1280          * and store the copy for each reply info. */
1281         if(qinfo->local_alias) {
1282                 struct packed_rrset_data* d;
1283                 struct packed_rrset_data* dsrc;
1284                 r->local_alias = regional_alloc_zero(s->s.region,
1285                         sizeof(*qinfo->local_alias));
1286                 if(!r->local_alias)
1287                         return 0;
1288                 r->local_alias->rrset = regional_alloc_init(s->s.region,
1289                         qinfo->local_alias->rrset,
1290                         sizeof(*qinfo->local_alias->rrset));
1291                 if(!r->local_alias->rrset)
1292                         return 0;
1293                 dsrc = qinfo->local_alias->rrset->entry.data;
1294
1295                 /* In the current implementation, a local alias must be
1296                  * a single CNAME RR (see worker_handle_request()). */
1297                 log_assert(!qinfo->local_alias->next && dsrc->count == 1 &&
1298                         qinfo->local_alias->rrset->rk.type ==
1299                         htons(LDNS_RR_TYPE_CNAME));
1300                 /* Technically, we should make a local copy for the owner
1301                  * name of the RRset, but in the case of the first (and
1302                  * currently only) local alias RRset, the owner name should
1303                  * point to the qname of the corresponding query, which should
1304                  * be valid throughout the lifetime of this mesh_reply.  So
1305                  * we can skip copying. */
1306                 log_assert(qinfo->local_alias->rrset->rk.dname ==
1307                         sldns_buffer_at(rep->c->buffer, LDNS_HEADER_SIZE));
1308
1309                 d = regional_alloc_init(s->s.region, dsrc,
1310                         sizeof(struct packed_rrset_data)
1311                         + sizeof(size_t) + sizeof(uint8_t*) + sizeof(time_t));
1312                 if(!d)
1313                         return 0;
1314                 r->local_alias->rrset->entry.data = d;
1315                 d->rr_len = (size_t*)((uint8_t*)d +
1316                         sizeof(struct packed_rrset_data));
1317                 d->rr_data = (uint8_t**)&(d->rr_len[1]);
1318                 d->rr_ttl = (time_t*)&(d->rr_data[1]);
1319                 d->rr_len[0] = dsrc->rr_len[0];
1320                 d->rr_ttl[0] = dsrc->rr_ttl[0];
1321                 d->rr_data[0] = regional_alloc_init(s->s.region,
1322                         dsrc->rr_data[0], d->rr_len[0]);
1323                 if(!d->rr_data[0])
1324                         return 0;
1325         } else
1326                 r->local_alias = NULL;
1327
1328         s->reply_list = r;
1329         return 1;
1330 }
1331
1332 /* Extract the query info and flags from 'mstate' into '*qinfop' and '*qflags'.
1333  * Since this is only used for internal refetch of otherwise-expired answer,
1334  * we simply ignore the rare failure mode when memory allocation fails. */
1335 static void
1336 mesh_copy_qinfo(struct mesh_state* mstate, struct query_info** qinfop,
1337         uint16_t* qflags)
1338 {
1339         struct regional* region = mstate->s.env->scratch;
1340         struct query_info* qinfo;
1341
1342         qinfo = regional_alloc_init(region, &mstate->s.qinfo, sizeof(*qinfo));
1343         if(!qinfo)
1344                 return;
1345         qinfo->qname = regional_alloc_init(region, qinfo->qname,
1346                 qinfo->qname_len);
1347         if(!qinfo->qname)
1348                 return;
1349         *qinfop = qinfo;
1350         *qflags = mstate->s.query_flags;
1351 }
1352
1353 /**
1354  * Continue processing the mesh state at another module.
1355  * Handles module to modules transfer of control.
1356  * Handles module finished.
1357  * @param mesh: the mesh area.
1358  * @param mstate: currently active mesh state.
1359  *      Deleted if finished, calls _done and _supers to 
1360  *      send replies to clients and inform other mesh states.
1361  *      This in turn may create additional runnable mesh states.
1362  * @param s: state at which the current module exited.
1363  * @param ev: the event sent to the module.
1364  *      returned is the event to send to the next module.
1365  * @return true if continue processing at the new module.
1366  *      false if not continued processing is needed.
1367  */
1368 static int
1369 mesh_continue(struct mesh_area* mesh, struct mesh_state* mstate,
1370         enum module_ext_state s, enum module_ev* ev)
1371 {
1372         mstate->num_activated++;
1373         if(mstate->num_activated > MESH_MAX_ACTIVATION) {
1374                 /* module is looping. Stop it. */
1375                 log_err("internal error: looping module (%s) stopped",
1376                         mesh->mods.mod[mstate->s.curmod]->name);
1377                 log_query_info(VERB_QUERY, "pass error for qstate",
1378                         &mstate->s.qinfo);
1379                 s = module_error;
1380         }
1381         if(s == module_wait_module || s == module_restart_next) {
1382                 /* start next module */
1383                 mstate->s.curmod++;
1384                 if(mesh->mods.num == mstate->s.curmod) {
1385                         log_err("Cannot pass to next module; at last module");
1386                         log_query_info(VERB_QUERY, "pass error for qstate",
1387                                 &mstate->s.qinfo);
1388                         mstate->s.curmod--;
1389                         return mesh_continue(mesh, mstate, module_error, ev);
1390                 }
1391                 if(s == module_restart_next) {
1392                         int curmod = mstate->s.curmod;
1393                         for(; mstate->s.curmod < mesh->mods.num; 
1394                                 mstate->s.curmod++) {
1395                                 fptr_ok(fptr_whitelist_mod_clear(
1396                                         mesh->mods.mod[mstate->s.curmod]->clear));
1397                                 (*mesh->mods.mod[mstate->s.curmod]->clear)
1398                                         (&mstate->s, mstate->s.curmod);
1399                                 mstate->s.minfo[mstate->s.curmod] = NULL;
1400                         }
1401                         mstate->s.curmod = curmod;
1402                 }
1403                 *ev = module_event_pass;
1404                 return 1;
1405         }
1406         if(s == module_wait_subquery && mstate->sub_set.count == 0) {
1407                 log_err("module cannot wait for subquery, subquery list empty");
1408                 log_query_info(VERB_QUERY, "pass error for qstate",
1409                         &mstate->s.qinfo);
1410                 s = module_error;
1411         }
1412         if(s == module_error && mstate->s.return_rcode == LDNS_RCODE_NOERROR) {
1413                 /* error is bad, handle pass back up below */
1414                 mstate->s.return_rcode = LDNS_RCODE_SERVFAIL;
1415         }
1416         if(s == module_error) {
1417                 mesh_query_done(mstate);
1418                 mesh_walk_supers(mesh, mstate);
1419                 mesh_state_delete(&mstate->s);
1420                 return 0;
1421         }
1422         if(s == module_finished) {
1423                 if(mstate->s.curmod == 0) {
1424                         struct query_info* qinfo = NULL;
1425                         uint16_t qflags;
1426
1427                         mesh_query_done(mstate);
1428                         mesh_walk_supers(mesh, mstate);
1429
1430                         /* If the answer to the query needs to be refetched
1431                          * from an external DNS server, we'll need to schedule
1432                          * a prefetch after removing the current state, so
1433                          * we need to make a copy of the query info here. */
1434                         if(mstate->s.need_refetch)
1435                                 mesh_copy_qinfo(mstate, &qinfo, &qflags);
1436
1437                         mesh_state_delete(&mstate->s);
1438                         if(qinfo) {
1439                                 mesh_schedule_prefetch(mesh, qinfo, qflags,
1440                                         0, 1);
1441                         }
1442                         return 0;
1443                 }
1444                 /* pass along the locus of control */
1445                 mstate->s.curmod --;
1446                 *ev = module_event_moddone;
1447                 return 1;
1448         }
1449         return 0;
1450 }
1451
1452 void mesh_run(struct mesh_area* mesh, struct mesh_state* mstate,
1453         enum module_ev ev, struct outbound_entry* e)
1454 {
1455         enum module_ext_state s;
1456         verbose(VERB_ALGO, "mesh_run: start");
1457         while(mstate) {
1458                 /* run the module */
1459                 fptr_ok(fptr_whitelist_mod_operate(
1460                         mesh->mods.mod[mstate->s.curmod]->operate));
1461                 (*mesh->mods.mod[mstate->s.curmod]->operate)
1462                         (&mstate->s, ev, mstate->s.curmod, e);
1463
1464                 /* examine results */
1465                 mstate->s.reply = NULL;
1466                 regional_free_all(mstate->s.env->scratch);
1467                 s = mstate->s.ext_state[mstate->s.curmod];
1468                 verbose(VERB_ALGO, "mesh_run: %s module exit state is %s", 
1469                         mesh->mods.mod[mstate->s.curmod]->name, strextstate(s));
1470                 e = NULL;
1471                 if(mesh_continue(mesh, mstate, s, &ev))
1472                         continue;
1473
1474                 /* run more modules */
1475                 ev = module_event_pass;
1476                 if(mesh->run.count > 0) {
1477                         /* pop random element off the runnable tree */
1478                         mstate = (struct mesh_state*)mesh->run.root->key;
1479                         (void)rbtree_delete(&mesh->run, mstate);
1480                 } else mstate = NULL;
1481         }
1482         if(verbosity >= VERB_ALGO) {
1483                 mesh_stats(mesh, "mesh_run: end");
1484                 mesh_log_list(mesh);
1485         }
1486 }
1487
1488 void 
1489 mesh_log_list(struct mesh_area* mesh)
1490 {
1491         char buf[30];
1492         struct mesh_state* m;
1493         int num = 0;
1494         RBTREE_FOR(m, struct mesh_state*, &mesh->all) {
1495                 snprintf(buf, sizeof(buf), "%d%s%s%s%s%s%s mod%d %s%s", 
1496                         num++, (m->s.is_priming)?"p":"",  /* prime */
1497                         (m->s.is_valrec)?"v":"",  /* prime */
1498                         (m->s.query_flags&BIT_RD)?"RD":"",
1499                         (m->s.query_flags&BIT_CD)?"CD":"",
1500                         (m->super_set.count==0)?"d":"", /* detached */
1501                         (m->sub_set.count!=0)?"c":"",  /* children */
1502                         m->s.curmod, (m->reply_list)?"rep":"", /*hasreply*/
1503                         (m->cb_list)?"cb":"" /* callbacks */
1504                         ); 
1505                 log_query_info(VERB_ALGO, buf, &m->s.qinfo);
1506         }
1507 }
1508
1509 void 
1510 mesh_stats(struct mesh_area* mesh, const char* str)
1511 {
1512         verbose(VERB_DETAIL, "%s %u recursion states (%u with reply, "
1513                 "%u detached), %u waiting replies, %u recursion replies "
1514                 "sent, %d replies dropped, %d states jostled out", 
1515                 str, (unsigned)mesh->all.count, 
1516                 (unsigned)mesh->num_reply_states,
1517                 (unsigned)mesh->num_detached_states,
1518                 (unsigned)mesh->num_reply_addrs,
1519                 (unsigned)mesh->replies_sent,
1520                 (unsigned)mesh->stats_dropped,
1521                 (unsigned)mesh->stats_jostled);
1522         if(mesh->replies_sent > 0) {
1523                 struct timeval avg;
1524                 timeval_divide(&avg, &mesh->replies_sum_wait, 
1525                         mesh->replies_sent);
1526                 log_info("average recursion processing time "
1527                         ARG_LL "d.%6.6d sec",
1528                         (long long)avg.tv_sec, (int)avg.tv_usec);
1529                 log_info("histogram of recursion processing times");
1530                 timehist_log(mesh->histogram, "recursions");
1531         }
1532 }
1533
1534 void 
1535 mesh_stats_clear(struct mesh_area* mesh)
1536 {
1537         if(!mesh)
1538                 return;
1539         mesh->replies_sent = 0;
1540         mesh->replies_sum_wait.tv_sec = 0;
1541         mesh->replies_sum_wait.tv_usec = 0;
1542         mesh->stats_jostled = 0;
1543         mesh->stats_dropped = 0;
1544         timehist_clear(mesh->histogram);
1545         mesh->ans_secure = 0;
1546         mesh->ans_bogus = 0;
1547         memset(&mesh->ans_rcode[0], 0, sizeof(size_t)*16);
1548         mesh->ans_nodata = 0;
1549 }
1550
1551 size_t 
1552 mesh_get_mem(struct mesh_area* mesh)
1553 {
1554         struct mesh_state* m;
1555         size_t s = sizeof(*mesh) + sizeof(struct timehist) +
1556                 sizeof(struct th_buck)*mesh->histogram->num +
1557                 sizeof(sldns_buffer) + sldns_buffer_capacity(mesh->qbuf_bak);
1558         RBTREE_FOR(m, struct mesh_state*, &mesh->all) {
1559                 /* all, including m itself allocated in qstate region */
1560                 s += regional_get_mem(m->s.region);
1561         }
1562         return s;
1563 }
1564
1565 int 
1566 mesh_detect_cycle(struct module_qstate* qstate, struct query_info* qinfo,
1567         uint16_t flags, int prime, int valrec)
1568 {
1569         struct mesh_area* mesh = qstate->env->mesh;
1570         struct mesh_state* dep_m = NULL;
1571         if(!mesh_state_is_unique(qstate->mesh_info))
1572                 dep_m = mesh_area_find(mesh, NULL, qinfo, flags, prime, valrec);
1573         return mesh_detect_cycle_found(qstate, dep_m);
1574 }
1575
1576 void mesh_list_insert(struct mesh_state* m, struct mesh_state** fp,
1577         struct mesh_state** lp)
1578 {
1579         /* insert as last element */
1580         m->prev = *lp;
1581         m->next = NULL;
1582         if(*lp)
1583                 (*lp)->next = m;
1584         else    *fp = m;
1585         *lp = m;
1586 }
1587
1588 void mesh_list_remove(struct mesh_state* m, struct mesh_state** fp,
1589         struct mesh_state** lp)
1590 {
1591         if(m->next)
1592                 m->next->prev = m->prev;
1593         else    *lp = m->prev;
1594         if(m->prev)
1595                 m->prev->next = m->next;
1596         else    *fp = m->next;
1597 }