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 "services/cache/rrset.h"
  50 #include "util/log.h"
  51 #include "util/net_help.h"
  52 #include "util/module.h"
  53 #include "util/regional.h"
  54 #include "util/data/msgencode.h"
  55 #include "util/timehist.h"
  56 #include "util/fptr_wlist.h"
  57 #include "util/alloc.h"
  58 #include "util/config_file.h"
  59 #include "util/edns.h"
  60 #include "sldns/sbuffer.h"
  61 #include "sldns/wire2str.h"
  62 #include "services/localzone.h"
  63 #include "util/data/dname.h"
  64 #include "respip/respip.h"
  65 #include "services/listen_dnsport.h"
  66
  67 #ifdef CLIENT_SUBNET
  68 #include "edns-subnet/subnetmod.h"
  69 #include "edns-subnet/edns-subnet.h"
  70 #endif
  71
  72 /** subtract timers and the values do not overflow or become negative */
  73 static void
  74 timeval_subtract(struct timeval* d, const struct timeval* end, const struct timeval* start)
  75 {
  76 #ifndef S_SPLINT_S
  77         time_t end_usec = end->tv_usec;
  78         d->tv_sec = end->tv_sec - start->tv_sec;
  79         if(end_usec < start->tv_usec) {
  80                 end_usec += 1000000;
  81                 d->tv_sec--;
  82         }
  83         d->tv_usec = end_usec - start->tv_usec;
  84 #endif
  85 }
  86
  87 /** add timers and the values do not overflow or become negative */
  88 static void
  89 timeval_add(struct timeval* d, const struct timeval* add)
  90 {
  91 #ifndef S_SPLINT_S
  92         d->tv_sec += add->tv_sec;
  93         d->tv_usec += add->tv_usec;
  94         if(d->tv_usec >= 1000000 ) {
  95                 d->tv_usec -= 1000000;
  96                 d->tv_sec++;
  97         }
  98 #endif
  99 }
 100
 101 /** divide sum of timers to get average */
 102 static void
 103 timeval_divide(struct timeval* avg, const struct timeval* sum, size_t d)
 104 {
 105 #ifndef S_SPLINT_S
 106         size_t leftover;
 107         if(d <= 0) {
 108                 avg->tv_sec = 0;
 109                 avg->tv_usec = 0;
 110                 return;
 111         }
 112         avg->tv_sec = sum->tv_sec / d;
 113         avg->tv_usec = sum->tv_usec / d;
 114         /* handle fraction from seconds divide */
 115         leftover = sum->tv_sec - avg->tv_sec*d;
 116         if(leftover <= 0)
 117                 leftover = 0;
 118         avg->tv_usec += (((long long)leftover)*((long long)1000000))/d;
 119         if(avg->tv_sec < 0)
 120                 avg->tv_sec = 0;
 121         if(avg->tv_usec < 0)
 122                 avg->tv_usec = 0;
 123 #endif
 124 }
 125
 126 /** histogram compare of time values */
 127 static int
 128 timeval_smaller(const struct timeval* x, const struct timeval* y)
 129 {
 130 #ifndef S_SPLINT_S
 131         if(x->tv_sec < y->tv_sec)
 132                 return 1;
 133         else if(x->tv_sec == y->tv_sec) {
 134                 if(x->tv_usec <= y->tv_usec)
 135                         return 1;
 136                 else    return 0;
 137         }
 138         else    return 0;
 139 #endif
 140 }
 141
 142 /**
 143  * Compare two response-ip client info entries for the purpose of mesh state
 144  * compare.  It returns 0 if ci_a and ci_b are considered equal; otherwise
 145  * 1 or -1 (they mean 'ci_a is larger/smaller than ci_b', respectively, but
 146  * in practice it should be only used to mean they are different).
 147  * We cannot share the mesh state for two queries if different response-ip
 148  * actions can apply in the end, even if those queries are otherwise identical.
 149  * For this purpose we compare tag lists and tag action lists; they should be
 150  * identical to share the same state.
 151  * For tag data, we don't look into the data content, as it can be
 152  * expensive; unless tag data are not defined for both or they point to the
 153  * exact same data in memory (i.e., they come from the same ACL entry), we
 154  * consider these data different.
 155  * Likewise, if the client info is associated with views, we don't look into
 156  * the views.  They are considered different unless they are exactly the same
 157  * even if the views only differ in the names.
 158  */
 159 static int
 160 client_info_compare(const struct respip_client_info* ci_a,
 161         const struct respip_client_info* ci_b)
 162 {
 163         int cmp;
 164
 165         if(!ci_a && !ci_b)
 166                 return 0;
 167         if(ci_a && !ci_b)
 168                 return -1;
 169         if(!ci_a && ci_b)
 170                 return 1;
 171         if(ci_a->taglen != ci_b->taglen)
 172                 return (ci_a->taglen < ci_b->taglen) ? -1 : 1;
 173         if(ci_a->taglist && !ci_b->taglist)
 174                 return -1;
 175         if(!ci_a->taglist && ci_b->taglist)
 176                 return 1;
 177         if(ci_a->taglist && ci_b->taglist) {
 178                 cmp = memcmp(ci_a->taglist, ci_b->taglist, ci_a->taglen);
 179                 if(cmp != 0)
 180                         return cmp;
 181         }
 182         if(ci_a->tag_actions_size != ci_b->tag_actions_size)
 183                 return (ci_a->tag_actions_size < ci_b->tag_actions_size) ?
 184                         -1 : 1;
 185         if(ci_a->tag_actions && !ci_b->tag_actions)
 186                 return -1;
 187         if(!ci_a->tag_actions && ci_b->tag_actions)
 188                 return 1;
 189         if(ci_a->tag_actions && ci_b->tag_actions) {
 190                 cmp = memcmp(ci_a->tag_actions, ci_b->tag_actions,
 191                         ci_a->tag_actions_size);
 192                 if(cmp != 0)
 193                         return cmp;
 194         }
 195         if(ci_a->tag_datas != ci_b->tag_datas)
 196                 return ci_a->tag_datas < ci_b->tag_datas ? -1 : 1;
 197         if(ci_a->view != ci_b->view)
 198                 return ci_a->view < ci_b->view ? -1 : 1;
 199         /* For the unbound daemon these should be non-NULL and identical,
 200          * but we check that just in case. */
 201         if(ci_a->respip_set != ci_b->respip_set)
 202                 return ci_a->respip_set < ci_b->respip_set ? -1 : 1;
 203         return 0;
 204 }
 205
 206 int
 207 mesh_state_compare(const void* ap, const void* bp)
 208 {
 209         struct mesh_state* a = (struct mesh_state*)ap;
 210         struct mesh_state* b = (struct mesh_state*)bp;
 211         int cmp;
 212
 213         if(a->unique < b->unique)
 214                 return -1;
 215         if(a->unique > b->unique)
 216                 return 1;
 217
 218         if(a->s.is_priming && !b->s.is_priming)
 219                 return -1;
 220         if(!a->s.is_priming && b->s.is_priming)
 221                 return 1;
 222
 223         if(a->s.is_valrec && !b->s.is_valrec)
 224                 return -1;
 225         if(!a->s.is_valrec && b->s.is_valrec)
 226                 return 1;
 227
 228         if((a->s.query_flags&BIT_RD) && !(b->s.query_flags&BIT_RD))
 229                 return -1;
 230         if(!(a->s.query_flags&BIT_RD) && (b->s.query_flags&BIT_RD))
 231                 return 1;
 232
 233         if((a->s.query_flags&BIT_CD) && !(b->s.query_flags&BIT_CD))
 234                 return -1;
 235         if(!(a->s.query_flags&BIT_CD) && (b->s.query_flags&BIT_CD))
 236                 return 1;
 237
 238         cmp = query_info_compare(&a->s.qinfo, &b->s.qinfo);
 239         if(cmp != 0)
 240                 return cmp;
 241         return client_info_compare(a->s.client_info, b->s.client_info);
 242 }
 243
 244 int
 245 mesh_state_ref_compare(const void* ap, const void* bp)
 246 {
 247         struct mesh_state_ref* a = (struct mesh_state_ref*)ap;
 248         struct mesh_state_ref* b = (struct mesh_state_ref*)bp;
 249         return mesh_state_compare(a->s, b->s);
 250 }
 251
 252 struct mesh_area*
 253 mesh_create(struct module_stack* stack, struct module_env* env)
 254 {
 255         struct mesh_area* mesh = calloc(1, sizeof(struct mesh_area));
 256         if(!mesh) {
 257                 log_err("mesh area alloc: out of memory");
 258                 return NULL;
 259         }
 260         mesh->histogram = timehist_setup();
 261         mesh->qbuf_bak = sldns_buffer_new(env->cfg->msg_buffer_size);
 262         if(!mesh->histogram || !mesh->qbuf_bak) {
 263                 free(mesh);
 264                 log_err("mesh area alloc: out of memory");
 265                 return NULL;
 266         }
 267         mesh->mods = *stack;
 268         mesh->env = env;
 269         rbtree_init(&mesh->run, &mesh_state_compare);
 270         rbtree_init(&mesh->all, &mesh_state_compare);
 271         mesh->num_reply_addrs = 0;
 272         mesh->num_reply_states = 0;
 273         mesh->num_detached_states = 0;
 274         mesh->num_forever_states = 0;
 275         mesh->stats_jostled = 0;
 276         mesh->stats_dropped = 0;
 277         mesh->ans_expired = 0;
 278         mesh->max_reply_states = env->cfg->num_queries_per_thread;
 279         mesh->max_forever_states = (mesh->max_reply_states+1)/2;
 280 #ifndef S_SPLINT_S
 281         mesh->jostle_max.tv_sec = (time_t)(env->cfg->jostle_time / 1000);
 282         mesh->jostle_max.tv_usec = (time_t)((env->cfg->jostle_time % 1000)
 283                 *1000);
 284 #endif
 285         return mesh;
 286 }
 287
 288 /** help mesh delete delete mesh states */
 289 static void
 290 mesh_delete_helper(rbnode_type* n)
 291 {
 292         struct mesh_state* mstate = (struct mesh_state*)n->key;
 293         /* perform a full delete, not only 'cleanup' routine,
 294          * because other callbacks expect a clean state in the mesh.
 295          * For 're-entrant' calls */
 296         mesh_state_delete(&mstate->s);
 297         /* but because these delete the items from the tree, postorder
 298          * traversal and rbtree rebalancing do not work together */
 299 }
 300
 301 void
 302 mesh_delete(struct mesh_area* mesh)
 303 {
 304         if(!mesh)
 305                 return;
 306         /* free all query states */
 307         while(mesh->all.count)
 308                 mesh_delete_helper(mesh->all.root);
 309         timehist_delete(mesh->histogram);
 310         sldns_buffer_free(mesh->qbuf_bak);
 311         free(mesh);
 312 }
 313
 314 void
 315 mesh_delete_all(struct mesh_area* mesh)
 316 {
 317         /* free all query states */
 318         while(mesh->all.count)
 319                 mesh_delete_helper(mesh->all.root);
 320         mesh->stats_dropped += mesh->num_reply_addrs;
 321         /* clear mesh area references */
 322         rbtree_init(&mesh->run, &mesh_state_compare);
 323         rbtree_init(&mesh->all, &mesh_state_compare);
 324         mesh->num_reply_addrs = 0;
 325         mesh->num_reply_states = 0;
 326         mesh->num_detached_states = 0;
 327         mesh->num_forever_states = 0;
 328         mesh->forever_first = NULL;
 329         mesh->forever_last = NULL;
 330         mesh->jostle_first = NULL;
 331         mesh->jostle_last = NULL;
 332 }
 333
 334 int mesh_make_new_space(struct mesh_area* mesh, sldns_buffer* qbuf)
 335 {
 336         struct mesh_state* m = mesh->jostle_first;
 337         /* free space is available */
 338         if(mesh->num_reply_states < mesh->max_reply_states)
 339                 return 1;
 340         /* try to kick out a jostle-list item */
 341         if(m && m->reply_list && m->list_select == mesh_jostle_list) {
 342                 /* how old is it? */
 343                 struct timeval age;
 344                 timeval_subtract(&age, mesh->env->now_tv,
 345                         &m->reply_list->start_time);
 346                 if(timeval_smaller(&mesh->jostle_max, &age)) {
 347                         /* its a goner */
 348                         log_nametypeclass(VERB_ALGO, "query jostled out to "
 349                                 "make space for a new one",
 350                                 m->s.qinfo.qname, m->s.qinfo.qtype,
 351                                 m->s.qinfo.qclass);
 352                         /* backup the query */
 353                         if(qbuf) sldns_buffer_copy(mesh->qbuf_bak, qbuf);
 354                         /* notify supers */
 355                         if(m->super_set.count > 0) {
 356                                 verbose(VERB_ALGO, "notify supers of failure");
 357                                 m->s.return_msg = NULL;
 358                                 m->s.return_rcode = LDNS_RCODE_SERVFAIL;
 359                                 mesh_walk_supers(mesh, m);
 360                         }
 361                         mesh->stats_jostled ++;
 362                         mesh_state_delete(&m->s);
 363                         /* restore the query - note that the qinfo ptr to
 364                          * the querybuffer is then correct again. */
 365                         if(qbuf) sldns_buffer_copy(qbuf, mesh->qbuf_bak);
 366                         return 1;
 367                 }
 368         }
 369         /* no space for new item */
 370         return 0;
 371 }
 372
 373 struct dns_msg*
 374 mesh_serve_expired_lookup(struct module_qstate* qstate,
 375         struct query_info* lookup_qinfo)
 376 {
 377         hashvalue_type h;
 378         struct lruhash_entry* e;
 379         struct dns_msg* msg;
 380         struct reply_info* data;
 381         struct msgreply_entry* key;
 382         time_t timenow = *qstate->env->now;
 383         int must_validate = (!(qstate->query_flags&BIT_CD)
 384                 || qstate->env->cfg->ignore_cd) && qstate->env->need_to_validate;
 385         /* Lookup cache */
 386         h = query_info_hash(lookup_qinfo, qstate->query_flags);
 387         e = slabhash_lookup(qstate->env->msg_cache, h, lookup_qinfo, 0);
 388         if(!e) return NULL;
 389
 390         key = (struct msgreply_entry*)e->key;
 391         data = (struct reply_info*)e->data;
 392         msg = tomsg(qstate->env, &key->key, data, qstate->region, timenow,
 393                 qstate->env->cfg->serve_expired, qstate->env->scratch);
 394         if(!msg)
 395                 goto bail_out;
 396
 397         /* Check CNAME chain (if any)
 398          * This is part of tomsg above; no need to check now. */
 399
 400         /* Check security status of the cached answer.
 401          * tomsg above has a subset of these checks, so we are leaving
 402          * these as is.
 403          * In case of bogus or revalidation we don't care to reply here. */
 404         if(must_validate && (msg->rep->security == sec_status_bogus ||
 405                 msg->rep->security == sec_status_secure_sentinel_fail)) {
 406                 verbose(VERB_ALGO, "Serve expired: bogus answer found in cache");
 407                 goto bail_out;
 408         } else if(msg->rep->security == sec_status_unchecked && must_validate) {
 409                 verbose(VERB_ALGO, "Serve expired: unchecked entry needs "
 410                         "validation");
 411                 goto bail_out; /* need to validate cache entry first */
 412         } else if(msg->rep->security == sec_status_secure &&
 413                 !reply_all_rrsets_secure(msg->rep) && must_validate) {
 414                         verbose(VERB_ALGO, "Serve expired: secure entry"
 415                                 " changed status");
 416                         goto bail_out; /* rrset changed, re-verify */
 417         }
 418
 419         lock_rw_unlock(&e->lock);
 420         return msg;
 421
 422 bail_out:
 423         lock_rw_unlock(&e->lock);
 424         return NULL;
 425 }
 426
 427
 428 /** Init the serve expired data structure */
 429 static int
 430 mesh_serve_expired_init(struct mesh_state* mstate, int timeout)
 431 {
 432         struct timeval t;
 433
 434         /* Create serve_expired_data if not there yet */
 435         if(!mstate->s.serve_expired_data) {
 436                 mstate->s.serve_expired_data = (struct serve_expired_data*)
 437                         regional_alloc_zero(
 438                                 mstate->s.region, sizeof(struct serve_expired_data));
 439                 if(!mstate->s.serve_expired_data)
 440                         return 0;
 441         }
 442
 443         /* Don't overwrite the function if already set */
 444         mstate->s.serve_expired_data->get_cached_answer =
 445                 mstate->s.serve_expired_data->get_cached_answer?
 446                 mstate->s.serve_expired_data->get_cached_answer:
 447                 &mesh_serve_expired_lookup;
 448
 449         /* In case this timer already popped, start it again */
 450         if(!mstate->s.serve_expired_data->timer) {
 451                 mstate->s.serve_expired_data->timer = comm_timer_create(
 452                         mstate->s.env->worker_base, mesh_serve_expired_callback, mstate);
 453                 if(!mstate->s.serve_expired_data->timer)
 454                         return 0;
 455 #ifndef S_SPLINT_S
 456                 t.tv_sec = timeout/1000;
 457                 t.tv_usec = (timeout%1000)*1000;
 458 #endif
 459                 comm_timer_set(mstate->s.serve_expired_data->timer, &t);
 460         }
 461         return 1;
 462 }
 463
 464 void mesh_new_client(struct mesh_area* mesh, struct query_info* qinfo,
 465         struct respip_client_info* cinfo, uint16_t qflags,
 466         struct edns_data* edns, struct comm_reply* rep, uint16_t qid,
 467         int rpz_passthru)
 468 {
 469         struct mesh_state* s = NULL;
 470         int unique = unique_mesh_state(edns->opt_list_in, mesh->env);
 471         int was_detached = 0;
 472         int was_noreply = 0;
 473         int added = 0;
 474         int timeout = mesh->env->cfg->serve_expired?
 475                 mesh->env->cfg->serve_expired_client_timeout:0;
 476         struct sldns_buffer* r_buffer = rep->c->buffer;
 477         if(rep->c->tcp_req_info) {
 478                 r_buffer = rep->c->tcp_req_info->spool_buffer;
 479         }
 480         if(!unique)
 481                 s = mesh_area_find(mesh, cinfo, qinfo, qflags&(BIT_RD|BIT_CD), 0, 0);
 482         /* does this create a new reply state? */
 483         if(!s || s->list_select == mesh_no_list) {
 484                 if(!mesh_make_new_space(mesh, rep->c->buffer)) {
 485                         verbose(VERB_ALGO, "Too many queries. dropping "
 486                                 "incoming query.");
 487                         comm_point_drop_reply(rep);
 488                         mesh->stats_dropped++;
 489                         return;
 490                 }
 491                 /* for this new reply state, the reply address is free,
 492                  * so the limit of reply addresses does not stop reply states*/
 493         } else {
 494                 /* protect our memory usage from storing reply addresses */
 495                 if(mesh->num_reply_addrs > mesh->max_reply_states*16) {
 496                         verbose(VERB_ALGO, "Too many requests queued. "
 497                                 "dropping incoming query.");
 498                         comm_point_drop_reply(rep);
 499                         mesh->stats_dropped++;
 500                         return;
 501                 }
 502         }
 503         /* see if it already exists, if not, create one */
 504         if(!s) {
 505 #ifdef UNBOUND_DEBUG
 506                 struct rbnode_type* n;
 507 #endif
 508                 s = mesh_state_create(mesh->env, qinfo, cinfo,
 509                         qflags&(BIT_RD|BIT_CD), 0, 0);
 510                 if(!s) {
 511                         log_err("mesh_state_create: out of memory; SERVFAIL");
 512                         if(!inplace_cb_reply_servfail_call(mesh->env, qinfo, NULL, NULL,
 513                                 LDNS_RCODE_SERVFAIL, edns, rep, mesh->env->scratch, mesh->env->now_tv))
 514                                         edns->opt_list_inplace_cb_out = NULL;
 515                         error_encode(r_buffer, LDNS_RCODE_SERVFAIL,
 516                                 qinfo, qid, qflags, edns);
 517                         comm_point_send_reply(rep);
 518                         return;
 519                 }
 520                 if(unique)
 521                         mesh_state_make_unique(s);
 522                 s->s.rpz_passthru = rpz_passthru;
 523                 /* copy the edns options we got from the front */
 524                 if(edns->opt_list_in) {
 525                         s->s.edns_opts_front_in = edns_opt_copy_region(edns->opt_list_in,
 526                                 s->s.region);
 527                         if(!s->s.edns_opts_front_in) {
 528                                 log_err("mesh_state_create: out of memory; SERVFAIL");
 529                                 if(!inplace_cb_reply_servfail_call(mesh->env, qinfo, NULL,
 530                                         NULL, LDNS_RCODE_SERVFAIL, edns, rep, mesh->env->scratch, mesh->env->now_tv))
 531                                                 edns->opt_list_inplace_cb_out = NULL;
 532                                 error_encode(r_buffer, LDNS_RCODE_SERVFAIL,
 533                                         qinfo, qid, qflags, edns);
 534                                 comm_point_send_reply(rep);
 535                                 return;
 536                         }
 537                 }
 538
 539 #ifdef UNBOUND_DEBUG
 540                 n =
 541 #else
 542                 (void)
 543 #endif
 544                 rbtree_insert(&mesh->all, &s->node);
 545                 log_assert(n != NULL);
 546                 /* set detached (it is now) */
 547                 mesh->num_detached_states++;
 548                 added = 1;
 549         }
 550         if(!s->reply_list && !s->cb_list) {
 551                 was_noreply = 1;
 552                 if(s->super_set.count == 0) {
 553                         was_detached = 1;
 554                 }
 555         }
 556         /* add reply to s */
 557         if(!mesh_state_add_reply(s, edns, rep, qid, qflags, qinfo)) {
 558                 log_err("mesh_new_client: out of memory; SERVFAIL");
 559                 goto servfail_mem;
 560         }
 561         if(rep->c->tcp_req_info) {
 562                 if(!tcp_req_info_add_meshstate(rep->c->tcp_req_info, mesh, s)) {
 563                         log_err("mesh_new_client: out of memory add tcpreqinfo");
 564                         goto servfail_mem;
 565                 }
 566         }
 567         if(rep->c->use_h2) {
 568                 http2_stream_add_meshstate(rep->c->h2_stream, mesh, s);
 569         }
 570         /* add serve expired timer if required and not already there */
 571         if(timeout && !mesh_serve_expired_init(s, timeout)) {
 572                 log_err("mesh_new_client: out of memory initializing serve expired");
 573                 goto servfail_mem;
 574         }
 575         /* update statistics */
 576         if(was_detached) {
 577                 log_assert(mesh->num_detached_states > 0);
 578                 mesh->num_detached_states--;
 579         }
 580         if(was_noreply) {
 581                 mesh->num_reply_states ++;
 582         }
 583         mesh->num_reply_addrs++;
 584         if(s->list_select == mesh_no_list) {
 585                 /* move to either the forever or the jostle_list */
 586                 if(mesh->num_forever_states < mesh->max_forever_states) {
 587                         mesh->num_forever_states ++;
 588                         mesh_list_insert(s, &mesh->forever_first,
 589                                 &mesh->forever_last);
 590                         s->list_select = mesh_forever_list;
 591                 } else {
 592                         mesh_list_insert(s, &mesh->jostle_first,
 593                                 &mesh->jostle_last);
 594                         s->list_select = mesh_jostle_list;
 595                 }
 596         }
 597         if(added)
 598                 mesh_run(mesh, s, module_event_new, NULL);
 599         return;
 600
 601 servfail_mem:
 602         if(!inplace_cb_reply_servfail_call(mesh->env, qinfo, &s->s,
 603                 NULL, LDNS_RCODE_SERVFAIL, edns, rep, mesh->env->scratch, mesh->env->now_tv))
 604                         edns->opt_list_inplace_cb_out = NULL;
 605         error_encode(r_buffer, LDNS_RCODE_SERVFAIL,
 606                 qinfo, qid, qflags, edns);
 607         comm_point_send_reply(rep);
 608         if(added)
 609                 mesh_state_delete(&s->s);
 610         return;
 611 }
 612
 613 int
 614 mesh_new_callback(struct mesh_area* mesh, struct query_info* qinfo,
 615         uint16_t qflags, struct edns_data* edns, sldns_buffer* buf,
 616         uint16_t qid, mesh_cb_func_type cb, void* cb_arg, int rpz_passthru)
 617 {
 618         struct mesh_state* s = NULL;
 619         int unique = unique_mesh_state(edns->opt_list_in, mesh->env);
 620         int timeout = mesh->env->cfg->serve_expired?
 621                 mesh->env->cfg->serve_expired_client_timeout:0;
 622         int was_detached = 0;
 623         int was_noreply = 0;
 624         int added = 0;
 625         if(!unique)
 626                 s = mesh_area_find(mesh, NULL, qinfo, qflags&(BIT_RD|BIT_CD), 0, 0);
 627
 628         /* there are no limits on the number of callbacks */
 629
 630         /* see if it already exists, if not, create one */
 631         if(!s) {
 632 #ifdef UNBOUND_DEBUG
 633                 struct rbnode_type* n;
 634 #endif
 635                 s = mesh_state_create(mesh->env, qinfo, NULL,
 636                         qflags&(BIT_RD|BIT_CD), 0, 0);
 637                 if(!s) {
 638                         return 0;
 639                 }
 640                 if(unique)
 641                         mesh_state_make_unique(s);
 642                 s->s.rpz_passthru = rpz_passthru;
 643                 if(edns->opt_list_in) {
 644                         s->s.edns_opts_front_in = edns_opt_copy_region(edns->opt_list_in,
 645                                 s->s.region);
 646                         if(!s->s.edns_opts_front_in) {
 647                                 return 0;
 648                         }
 649                 }
 650 #ifdef UNBOUND_DEBUG
 651                 n =
 652 #else
 653                 (void)
 654 #endif
 655                 rbtree_insert(&mesh->all, &s->node);
 656                 log_assert(n != NULL);
 657                 /* set detached (it is now) */
 658                 mesh->num_detached_states++;
 659                 added = 1;
 660         }
 661         if(!s->reply_list && !s->cb_list) {
 662                 was_noreply = 1;
 663                 if(s->super_set.count == 0) {
 664                         was_detached = 1;
 665                 }
 666         }
 667         /* add reply to s */
 668         if(!mesh_state_add_cb(s, edns, buf, cb, cb_arg, qid, qflags)) {
 669                 if(added)
 670                         mesh_state_delete(&s->s);
 671                 return 0;
 672         }
 673         /* add serve expired timer if not already there */
 674         if(timeout && !mesh_serve_expired_init(s, timeout)) {
 675                 return 0;
 676         }
 677         /* update statistics */
 678         if(was_detached) {
 679                 log_assert(mesh->num_detached_states > 0);
 680                 mesh->num_detached_states--;
 681         }
 682         if(was_noreply) {
 683                 mesh->num_reply_states ++;
 684         }
 685         mesh->num_reply_addrs++;
 686         if(added)
 687                 mesh_run(mesh, s, module_event_new, NULL);
 688         return 1;
 689 }
 690
 691 /* Internal backend routine of mesh_new_prefetch().  It takes one additional
 692  * parameter, 'run', which controls whether to run the prefetch state
 693  * immediately.  When this function is called internally 'run' could be
 694  * 0 (false), in which case the new state is only made runnable so it
 695  * will not be run recursively on top of the current state. */
 696 static void mesh_schedule_prefetch(struct mesh_area* mesh,
 697         struct query_info* qinfo, uint16_t qflags, time_t leeway, int run,
 698         int rpz_passthru)
 699 {
 700         struct mesh_state* s = mesh_area_find(mesh, NULL, qinfo,
 701                 qflags&(BIT_RD|BIT_CD), 0, 0);
 702 #ifdef UNBOUND_DEBUG
 703         struct rbnode_type* n;
 704 #endif
 705         /* already exists, and for a different purpose perhaps.
 706          * if mesh_no_list, keep it that way. */
 707         if(s) {
 708                 /* make it ignore the cache from now on */
 709                 if(!s->s.blacklist)
 710                         sock_list_insert(&s->s.blacklist, NULL, 0, s->s.region);
 711                 if(s->s.prefetch_leeway < leeway)
 712                         s->s.prefetch_leeway = leeway;
 713                 return;
 714         }
 715         if(!mesh_make_new_space(mesh, NULL)) {
 716                 verbose(VERB_ALGO, "Too many queries. dropped prefetch.");
 717                 mesh->stats_dropped ++;
 718                 return;
 719         }
 720
 721         s = mesh_state_create(mesh->env, qinfo, NULL,
 722                 qflags&(BIT_RD|BIT_CD), 0, 0);
 723         if(!s) {
 724                 log_err("prefetch mesh_state_create: out of memory");
 725                 return;
 726         }
 727 #ifdef UNBOUND_DEBUG
 728         n =
 729 #else
 730         (void)
 731 #endif
 732         rbtree_insert(&mesh->all, &s->node);
 733         log_assert(n != NULL);
 734         /* set detached (it is now) */
 735         mesh->num_detached_states++;
 736         /* make it ignore the cache */
 737         sock_list_insert(&s->s.blacklist, NULL, 0, s->s.region);
 738         s->s.prefetch_leeway = leeway;
 739
 740         if(s->list_select == mesh_no_list) {
 741                 /* move to either the forever or the jostle_list */
 742                 if(mesh->num_forever_states < mesh->max_forever_states) {
 743                         mesh->num_forever_states ++;
 744                         mesh_list_insert(s, &mesh->forever_first,
 745                                 &mesh->forever_last);
 746                         s->list_select = mesh_forever_list;
 747                 } else {
 748                         mesh_list_insert(s, &mesh->jostle_first,
 749                                 &mesh->jostle_last);
 750                         s->list_select = mesh_jostle_list;
 751                 }
 752         }
 753         s->s.rpz_passthru = rpz_passthru;
 754
 755         if(!run) {
 756 #ifdef UNBOUND_DEBUG
 757                 n =
 758 #else
 759                 (void)
 760 #endif
 761                 rbtree_insert(&mesh->run, &s->run_node);
 762                 log_assert(n != NULL);
 763                 return;
 764         }
 765
 766         mesh_run(mesh, s, module_event_new, NULL);
 767 }
 768
 769 #ifdef CLIENT_SUBNET
 770 /* Same logic as mesh_schedule_prefetch but tailored to the subnet module logic
 771  * like passing along the comm_reply info. This will be faked into an EDNS
 772  * option for processing by the subnet module if the client has not already
 773  * attached its own ECS data. */
 774 static void mesh_schedule_prefetch_subnet(struct mesh_area* mesh,
 775         struct query_info* qinfo, uint16_t qflags, time_t leeway, int run,
 776         int rpz_passthru, struct comm_reply* rep, struct edns_option* edns_list)
 777 {
 778         struct mesh_state* s = NULL;
 779         struct edns_option* opt = NULL;
 780 #ifdef UNBOUND_DEBUG
 781         struct rbnode_type* n;
 782 #endif
 783         if(!mesh_make_new_space(mesh, NULL)) {
 784                 verbose(VERB_ALGO, "Too many queries. dropped prefetch.");
 785                 mesh->stats_dropped ++;
 786                 return;
 787         }
 788
 789         s = mesh_state_create(mesh->env, qinfo, NULL,
 790                 qflags&(BIT_RD|BIT_CD), 0, 0);
 791         if(!s) {
 792                 log_err("prefetch_subnet mesh_state_create: out of memory");
 793                 return;
 794         }
 795         mesh_state_make_unique(s);
 796
 797         opt = edns_opt_list_find(edns_list, mesh->env->cfg->client_subnet_opcode);
 798         if(opt) {
 799                 /* Use the client's ECS data */
 800                 if(!edns_opt_list_append(&s->s.edns_opts_front_in, opt->opt_code,
 801                         opt->opt_len, opt->opt_data, s->s.region)) {
 802                         log_err("prefetch_subnet edns_opt_list_append: out of memory");
 803                         return;
 804                 }
 805         } else {
 806                 /* Fake the ECS data from the client's IP */
 807                 struct ecs_data ecs;
 808                 memset(&ecs, 0, sizeof(ecs));
 809                 subnet_option_from_ss(&rep->addr, &ecs, mesh->env->cfg);
 810                 if(ecs.subnet_validdata == 0) {
 811                         log_err("prefetch_subnet subnet_option_from_ss: invalid data");
 812                         return;
 813                 }
 814                 subnet_ecs_opt_list_append(&ecs, &s->s.edns_opts_front_in, &s->s);
 815                 if(!s->s.edns_opts_front_in) {
 816                         log_err("prefetch_subnet subnet_ecs_opt_list_append: out of memory");
 817                         return;
 818                 }
 819         }
 820 #ifdef UNBOUND_DEBUG
 821         n =
 822 #else
 823         (void)
 824 #endif
 825         rbtree_insert(&mesh->all, &s->node);
 826         log_assert(n != NULL);
 827         /* set detached (it is now) */
 828         mesh->num_detached_states++;
 829         /* make it ignore the cache */
 830         sock_list_insert(&s->s.blacklist, NULL, 0, s->s.region);
 831         s->s.prefetch_leeway = leeway;
 832
 833         if(s->list_select == mesh_no_list) {
 834                 /* move to either the forever or the jostle_list */
 835                 if(mesh->num_forever_states < mesh->max_forever_states) {
 836                         mesh->num_forever_states ++;
 837                         mesh_list_insert(s, &mesh->forever_first,
 838                                 &mesh->forever_last);
 839                         s->list_select = mesh_forever_list;
 840                 } else {
 841                         mesh_list_insert(s, &mesh->jostle_first,
 842                                 &mesh->jostle_last);
 843                         s->list_select = mesh_jostle_list;
 844                 }
 845         }
 846         s->s.rpz_passthru = rpz_passthru;
 847
 848         if(!run) {
 849 #ifdef UNBOUND_DEBUG
 850                 n =
 851 #else
 852                 (void)
 853 #endif
 854                 rbtree_insert(&mesh->run, &s->run_node);
 855                 log_assert(n != NULL);
 856                 return;
 857         }
 858
 859         mesh_run(mesh, s, module_event_new, NULL);
 860 }
 861 #endif /* CLIENT_SUBNET */
 862
 863 void mesh_new_prefetch(struct mesh_area* mesh, struct query_info* qinfo,
 864         uint16_t qflags, time_t leeway, int rpz_passthru,
 865         struct comm_reply* rep, struct edns_option* opt_list)
 866 {
 867         (void)opt_list;
 868         (void)rep;
 869 #ifdef CLIENT_SUBNET
 870         if(rep)
 871                 mesh_schedule_prefetch_subnet(mesh, qinfo, qflags, leeway, 1,
 872                         rpz_passthru, rep, opt_list);
 873         else
 874 #endif
 875                 mesh_schedule_prefetch(mesh, qinfo, qflags, leeway, 1,
 876                         rpz_passthru);
 877 }
 878
 879 void mesh_report_reply(struct mesh_area* mesh, struct outbound_entry* e,
 880         struct comm_reply* reply, int what)
 881 {
 882         enum module_ev event = module_event_reply;
 883         e->qstate->reply = reply;
 884         if(what != NETEVENT_NOERROR) {
 885                 event = module_event_noreply;
 886                 if(what == NETEVENT_CAPSFAIL)
 887                         event = module_event_capsfail;
 888         }
 889         mesh_run(mesh, e->qstate->mesh_info, event, e);
 890 }
 891
 892 struct mesh_state*
 893 mesh_state_create(struct module_env* env, struct query_info* qinfo,
 894         struct respip_client_info* cinfo, uint16_t qflags, int prime,
 895         int valrec)
 896 {
 897         struct regional* region = alloc_reg_obtain(env->alloc);
 898         struct mesh_state* mstate;
 899         int i;
 900         if(!region)
 901                 return NULL;
 902         mstate = (struct mesh_state*)regional_alloc(region,
 903                 sizeof(struct mesh_state));
 904         if(!mstate) {
 905                 alloc_reg_release(env->alloc, region);
 906                 return NULL;
 907         }
 908         memset(mstate, 0, sizeof(*mstate));
 909         mstate->node = *RBTREE_NULL;
 910         mstate->run_node = *RBTREE_NULL;
 911         mstate->node.key = mstate;
 912         mstate->run_node.key = mstate;
 913         mstate->reply_list = NULL;
 914         mstate->list_select = mesh_no_list;
 915         mstate->replies_sent = 0;
 916         rbtree_init(&mstate->super_set, &mesh_state_ref_compare);
 917         rbtree_init(&mstate->sub_set, &mesh_state_ref_compare);
 918         mstate->num_activated = 0;
 919         mstate->unique = NULL;
 920         /* init module qstate */
 921         mstate->s.qinfo.qtype = qinfo->qtype;
 922         mstate->s.qinfo.qclass = qinfo->qclass;
 923         mstate->s.qinfo.local_alias = NULL;
 924         mstate->s.qinfo.qname_len = qinfo->qname_len;
 925         mstate->s.qinfo.qname = regional_alloc_init(region, qinfo->qname,
 926                 qinfo->qname_len);
 927         if(!mstate->s.qinfo.qname) {
 928                 alloc_reg_release(env->alloc, region);
 929                 return NULL;
 930         }
 931         if(cinfo) {
 932                 mstate->s.client_info = regional_alloc_init(region, cinfo,
 933                         sizeof(*cinfo));
 934                 if(!mstate->s.client_info) {
 935                         alloc_reg_release(env->alloc, region);
 936                         return NULL;
 937                 }
 938         }
 939         /* remove all weird bits from qflags */
 940         mstate->s.query_flags = (qflags & (BIT_RD|BIT_CD));
 941         mstate->s.is_priming = prime;
 942         mstate->s.is_valrec = valrec;
 943         mstate->s.reply = NULL;
 944         mstate->s.region = region;
 945         mstate->s.curmod = 0;
 946         mstate->s.return_msg = 0;
 947         mstate->s.return_rcode = LDNS_RCODE_NOERROR;
 948         mstate->s.env = env;
 949         mstate->s.mesh_info = mstate;
 950         mstate->s.prefetch_leeway = 0;
 951         mstate->s.serve_expired_data = NULL;
 952         mstate->s.no_cache_lookup = 0;
 953         mstate->s.no_cache_store = 0;
 954         mstate->s.need_refetch = 0;
 955         mstate->s.was_ratelimited = 0;
 956
 957         /* init modules */
 958         for(i=0; i<env->mesh->mods.num; i++) {
 959                 mstate->s.minfo[i] = NULL;
 960                 mstate->s.ext_state[i] = module_state_initial;
 961         }
 962         /* init edns option lists */
 963         mstate->s.edns_opts_front_in = NULL;
 964         mstate->s.edns_opts_back_out = NULL;
 965         mstate->s.edns_opts_back_in = NULL;
 966         mstate->s.edns_opts_front_out = NULL;
 967
 968         return mstate;
 969 }
 970
 971 int
 972 mesh_state_is_unique(struct mesh_state* mstate)
 973 {
 974         return mstate->unique != NULL;
 975 }
 976
 977 void
 978 mesh_state_make_unique(struct mesh_state* mstate)
 979 {
 980         mstate->unique = mstate;
 981 }
 982
 983 void
 984 mesh_state_cleanup(struct mesh_state* mstate)
 985 {
 986         struct mesh_area* mesh;
 987         int i;
 988         if(!mstate)
 989                 return;
 990         mesh = mstate->s.env->mesh;
 991         /* Stop and delete the serve expired timer */
 992         if(mstate->s.serve_expired_data && mstate->s.serve_expired_data->timer) {
 993                 comm_timer_delete(mstate->s.serve_expired_data->timer);
 994                 mstate->s.serve_expired_data->timer = NULL;
 995         }
 996         /* drop unsent replies */
 997         if(!mstate->replies_sent) {
 998                 struct mesh_reply* rep = mstate->reply_list;
 999                 struct mesh_cb* cb;
1000                 /* in tcp_req_info, the mstates linked are removed, but
1001                  * the reply_list is now NULL, so the remove-from-empty-list
1002                  * takes no time and also it does not do the mesh accounting */
1003                 mstate->reply_list = NULL;
1004                 for(; rep; rep=rep->next) {
1005                         comm_point_drop_reply(&rep->query_reply);
1006                         log_assert(mesh->num_reply_addrs > 0);
1007                         mesh->num_reply_addrs--;
1008                 }
1009                 while((cb = mstate->cb_list)!=NULL) {
1010                         mstate->cb_list = cb->next;
1011                         fptr_ok(fptr_whitelist_mesh_cb(cb->cb));
1012                         (*cb->cb)(cb->cb_arg, LDNS_RCODE_SERVFAIL, NULL,
1013                                 sec_status_unchecked, NULL, 0);
1014                         log_assert(mesh->num_reply_addrs > 0);
1015                         mesh->num_reply_addrs--;
1016                 }
1017         }
1018
1019         /* de-init modules */
1020         for(i=0; i<mesh->mods.num; i++) {
1021                 fptr_ok(fptr_whitelist_mod_clear(mesh->mods.mod[i]->clear));
1022                 (*mesh->mods.mod[i]->clear)(&mstate->s, i);
1023                 mstate->s.minfo[i] = NULL;
1024                 mstate->s.ext_state[i] = module_finished;
1025         }
1026         alloc_reg_release(mstate->s.env->alloc, mstate->s.region);
1027 }
1028
1029 void
1030 mesh_state_delete(struct module_qstate* qstate)
1031 {
1032         struct mesh_area* mesh;
1033         struct mesh_state_ref* super, ref;
1034         struct mesh_state* mstate;
1035         if(!qstate)
1036                 return;
1037         mstate = qstate->mesh_info;
1038         mesh = mstate->s.env->mesh;
1039         mesh_detach_subs(&mstate->s);
1040         if(mstate->list_select == mesh_forever_list) {
1041                 mesh->num_forever_states --;
1042                 mesh_list_remove(mstate, &mesh->forever_first,
1043                         &mesh->forever_last);
1044         } else if(mstate->list_select == mesh_jostle_list) {
1045                 mesh_list_remove(mstate, &mesh->jostle_first,
1046                         &mesh->jostle_last);
1047         }
1048         if(!mstate->reply_list && !mstate->cb_list
1049                 && mstate->super_set.count == 0) {
1050                 log_assert(mesh->num_detached_states > 0);
1051                 mesh->num_detached_states--;
1052         }
1053         if(mstate->reply_list || mstate->cb_list) {
1054                 log_assert(mesh->num_reply_states > 0);
1055                 mesh->num_reply_states--;
1056         }
1057         ref.node.key = &ref;
1058         ref.s = mstate;
1059         RBTREE_FOR(super, struct mesh_state_ref*, &mstate->super_set) {
1060                 (void)rbtree_delete(&super->s->sub_set, &ref);
1061         }
1062         (void)rbtree_delete(&mesh->run, mstate);
1063         (void)rbtree_delete(&mesh->all, mstate);
1064         mesh_state_cleanup(mstate);
1065 }
1066
1067 /** helper recursive rbtree find routine */
1068 static int
1069 find_in_subsub(struct mesh_state* m, struct mesh_state* tofind, size_t *c)
1070 {
1071         struct mesh_state_ref* r;
1072         if((*c)++ > MESH_MAX_SUBSUB)
1073                 return 1;
1074         RBTREE_FOR(r, struct mesh_state_ref*, &m->sub_set) {
1075                 if(r->s == tofind || find_in_subsub(r->s, tofind, c))
1076                         return 1;
1077         }
1078         return 0;
1079 }
1080
1081 /** find cycle for already looked up mesh_state */
1082 static int
1083 mesh_detect_cycle_found(struct module_qstate* qstate, struct mesh_state* dep_m)
1084 {
1085         struct mesh_state* cyc_m = qstate->mesh_info;
1086         size_t counter = 0;
1087         if(!dep_m)
1088                 return 0;
1089         if(dep_m == cyc_m || find_in_subsub(dep_m, cyc_m, &counter)) {
1090                 if(counter > MESH_MAX_SUBSUB)
1091                         return 2;
1092                 return 1;
1093         }
1094         return 0;
1095 }
1096
1097 void mesh_detach_subs(struct module_qstate* qstate)
1098 {
1099         struct mesh_area* mesh = qstate->env->mesh;
1100         struct mesh_state_ref* ref, lookup;
1101 #ifdef UNBOUND_DEBUG
1102         struct rbnode_type* n;
1103 #endif
1104         lookup.node.key = &lookup;
1105         lookup.s = qstate->mesh_info;
1106         RBTREE_FOR(ref, struct mesh_state_ref*, &qstate->mesh_info->sub_set) {
1107 #ifdef UNBOUND_DEBUG
1108                 n =
1109 #else
1110                 (void)
1111 #endif
1112                 rbtree_delete(&ref->s->super_set, &lookup);
1113                 log_assert(n != NULL); /* must have been present */
1114                 if(!ref->s->reply_list && !ref->s->cb_list
1115                         && ref->s->super_set.count == 0) {
1116                         mesh->num_detached_states++;
1117                         log_assert(mesh->num_detached_states +
1118                                 mesh->num_reply_states <= mesh->all.count);
1119                 }
1120         }
1121         rbtree_init(&qstate->mesh_info->sub_set, &mesh_state_ref_compare);
1122 }
1123
1124 int mesh_add_sub(struct module_qstate* qstate, struct query_info* qinfo,
1125         uint16_t qflags, int prime, int valrec, struct module_qstate** newq,
1126         struct mesh_state** sub)
1127 {
1128         /* find it, if not, create it */
1129         struct mesh_area* mesh = qstate->env->mesh;
1130         *sub = mesh_area_find(mesh, NULL, qinfo, qflags,
1131                 prime, valrec);
1132         if(mesh_detect_cycle_found(qstate, *sub)) {
1133                 verbose(VERB_ALGO, "attach failed, cycle detected");
1134                 return 0;
1135         }
1136         if(!*sub) {
1137 #ifdef UNBOUND_DEBUG
1138                 struct rbnode_type* n;
1139 #endif
1140                 /* create a new one */
1141                 *sub = mesh_state_create(qstate->env, qinfo, NULL, qflags, prime,
1142                         valrec);
1143                 if(!*sub) {
1144                         log_err("mesh_attach_sub: out of memory");
1145                         return 0;
1146                 }
1147 #ifdef UNBOUND_DEBUG
1148                 n =
1149 #else
1150                 (void)
1151 #endif
1152                 rbtree_insert(&mesh->all, &(*sub)->node);
1153                 log_assert(n != NULL);
1154                 /* set detached (it is now) */
1155                 mesh->num_detached_states++;
1156                 /* set new query state to run */
1157 #ifdef UNBOUND_DEBUG
1158                 n =
1159 #else
1160                 (void)
1161 #endif
1162                 rbtree_insert(&mesh->run, &(*sub)->run_node);
1163                 log_assert(n != NULL);
1164                 *newq = &(*sub)->s;
1165         } else
1166                 *newq = NULL;
1167         return 1;
1168 }
1169
1170 int mesh_attach_sub(struct module_qstate* qstate, struct query_info* qinfo,
1171         uint16_t qflags, int prime, int valrec, struct module_qstate** newq)
1172 {
1173         struct mesh_area* mesh = qstate->env->mesh;
1174         struct mesh_state* sub = NULL;
1175         int was_detached;
1176         if(!mesh_add_sub(qstate, qinfo, qflags, prime, valrec, newq, &sub))
1177                 return 0;
1178         was_detached = (sub->super_set.count == 0);
1179         if(!mesh_state_attachment(qstate->mesh_info, sub))
1180                 return 0;
1181         /* if it was a duplicate  attachment, the count was not zero before */
1182         if(!sub->reply_list && !sub->cb_list && was_detached &&
1183                 sub->super_set.count == 1) {
1184                 /* it used to be detached, before this one got added */
1185                 log_assert(mesh->num_detached_states > 0);
1186                 mesh->num_detached_states--;
1187         }
1188         /* *newq will be run when inited after the current module stops */
1189         return 1;
1190 }
1191
1192 int mesh_state_attachment(struct mesh_state* super, struct mesh_state* sub)
1193 {
1194 #ifdef UNBOUND_DEBUG
1195         struct rbnode_type* n;
1196 #endif
1197         struct mesh_state_ref* subref; /* points to sub, inserted in super */
1198         struct mesh_state_ref* superref; /* points to super, inserted in sub */
1199         if( !(subref = regional_alloc(super->s.region,
1200                 sizeof(struct mesh_state_ref))) ||
1201                 !(superref = regional_alloc(sub->s.region,
1202                 sizeof(struct mesh_state_ref))) ) {
1203                 log_err("mesh_state_attachment: out of memory");
1204                 return 0;
1205         }
1206         superref->node.key = superref;
1207         superref->s = super;
1208         subref->node.key = subref;
1209         subref->s = sub;
1210         if(!rbtree_insert(&sub->super_set, &superref->node)) {
1211                 /* this should not happen, iterator and validator do not
1212                  * attach subqueries that are identical. */
1213                 /* already attached, we are done, nothing todo.
1214                  * since superref and subref already allocated in region,
1215                  * we cannot free them */
1216                 return 1;
1217         }
1218 #ifdef UNBOUND_DEBUG
1219         n =
1220 #else
1221         (void)
1222 #endif
1223         rbtree_insert(&super->sub_set, &subref->node);
1224         log_assert(n != NULL); /* we checked above if statement, the reverse
1225           administration should not fail now, unless they are out of sync */
1226         return 1;
1227 }
1228
1229 /**
1230  * callback results to mesh cb entry
1231  * @param m: mesh state to send it for.
1232  * @param rcode: if not 0, error code.
1233  * @param rep: reply to send (or NULL if rcode is set).
1234  * @param r: callback entry
1235  * @param start_time: the time to pass to callback functions, it is 0 or
1236  *      a value from one of the packets if the mesh state had packets.
1237  */
1238 static void
1239 mesh_do_callback(struct mesh_state* m, int rcode, struct reply_info* rep,
1240         struct mesh_cb* r, struct timeval* start_time)
1241 {
1242         int secure;
1243         char* reason = NULL;
1244         int was_ratelimited = m->s.was_ratelimited;
1245         /* bogus messages are not made into servfail, sec_status passed
1246          * to the callback function */
1247         if(rep && rep->security == sec_status_secure)
1248                 secure = 1;
1249         else    secure = 0;
1250         if(!rep && rcode == LDNS_RCODE_NOERROR)
1251                 rcode = LDNS_RCODE_SERVFAIL;
1252         if(!rcode && (rep->security == sec_status_bogus ||
1253                 rep->security == sec_status_secure_sentinel_fail)) {
1254                 if(!(reason = errinf_to_str_bogus(&m->s)))
1255                         rcode = LDNS_RCODE_SERVFAIL;
1256         }
1257         /* send the reply */
1258         if(rcode) {
1259                 if(rcode == LDNS_RCODE_SERVFAIL) {
1260                         if(!inplace_cb_reply_servfail_call(m->s.env, &m->s.qinfo, &m->s,
1261                                 rep, rcode, &r->edns, NULL, m->s.region, start_time))
1262                                         r->edns.opt_list_inplace_cb_out = NULL;
1263                 } else {
1264                         if(!inplace_cb_reply_call(m->s.env, &m->s.qinfo, &m->s, rep, rcode,
1265                                 &r->edns, NULL, m->s.region, start_time))
1266                                         r->edns.opt_list_inplace_cb_out = NULL;
1267                 }
1268                 fptr_ok(fptr_whitelist_mesh_cb(r->cb));
1269                 (*r->cb)(r->cb_arg, rcode, r->buf, sec_status_unchecked, NULL,
1270                         was_ratelimited);
1271         } else {
1272                 size_t udp_size = r->edns.udp_size;
1273                 sldns_buffer_clear(r->buf);
1274                 r->edns.edns_version = EDNS_ADVERTISED_VERSION;
1275                 r->edns.udp_size = EDNS_ADVERTISED_SIZE;
1276                 r->edns.ext_rcode = 0;
1277                 r->edns.bits &= EDNS_DO;
1278
1279                 if(!inplace_cb_reply_call(m->s.env, &m->s.qinfo, &m->s, rep,
1280                         LDNS_RCODE_NOERROR, &r->edns, NULL, m->s.region, start_time) ||
1281                         !reply_info_answer_encode(&m->s.qinfo, rep, r->qid,
1282                         r->qflags, r->buf, 0, 1,
1283                         m->s.env->scratch, udp_size, &r->edns,
1284                         (int)(r->edns.bits & EDNS_DO), secure))
1285                 {
1286                         fptr_ok(fptr_whitelist_mesh_cb(r->cb));
1287                         (*r->cb)(r->cb_arg, LDNS_RCODE_SERVFAIL, r->buf,
1288                                 sec_status_unchecked, NULL, 0);
1289                 } else {
1290                         fptr_ok(fptr_whitelist_mesh_cb(r->cb));
1291                         (*r->cb)(r->cb_arg, LDNS_RCODE_NOERROR, r->buf,
1292                                 rep->security, reason, was_ratelimited);
1293                 }
1294         }
1295         free(reason);
1296         log_assert(m->s.env->mesh->num_reply_addrs > 0);
1297         m->s.env->mesh->num_reply_addrs--;
1298 }
1299
1300 static inline int
1301 mesh_is_rpz_respip_tcponly_action(struct mesh_state const* m)
1302 {
1303         struct respip_action_info const* respip_info = m->s.respip_action_info;
1304         return respip_info == NULL
1305                         ? 0
1306                         : (respip_info->rpz_used
1307                         && !respip_info->rpz_disabled
1308                         && respip_info->action == respip_truncate);
1309 }
1310
1311 static inline int
1312 mesh_is_udp(struct mesh_reply const* r) {
1313         return r->query_reply.c->type == comm_udp;
1314 }
1315
1316 /**
1317  * Send reply to mesh reply entry
1318  * @param m: mesh state to send it for.
1319  * @param rcode: if not 0, error code.
1320  * @param rep: reply to send (or NULL if rcode is set).
1321  * @param r: reply entry
1322  * @param r_buffer: buffer to use for reply entry.
1323  * @param prev: previous reply, already has its answer encoded in buffer.
1324  * @param prev_buffer: buffer for previous reply.
1325  */
1326 static void
1327 mesh_send_reply(struct mesh_state* m, int rcode, struct reply_info* rep,
1328         struct mesh_reply* r, struct sldns_buffer* r_buffer,
1329         struct mesh_reply* prev, struct sldns_buffer* prev_buffer)
1330 {
1331         struct timeval end_time;
1332         struct timeval duration;
1333         int secure;
1334         /* briefly set the replylist to null in case the
1335          * meshsendreply calls tcpreqinfo sendreply that
1336          * comm_point_drops because of size, and then the
1337          * null stops the mesh state remove and thus
1338          * reply_list modification and accounting */
1339         struct mesh_reply* rlist = m->reply_list;
1340
1341         /* rpz: apply actions */
1342         rcode = mesh_is_udp(r) && mesh_is_rpz_respip_tcponly_action(m)
1343                         ? (rcode|BIT_TC) : rcode;
1344
1345         /* examine security status */
1346         if(m->s.env->need_to_validate && (!(r->qflags&BIT_CD) ||
1347                 m->s.env->cfg->ignore_cd) && rep &&
1348                 (rep->security <= sec_status_bogus ||
1349                 rep->security == sec_status_secure_sentinel_fail)) {
1350                 rcode = LDNS_RCODE_SERVFAIL;
1351                 if(m->s.env->cfg->stat_extended)
1352                         m->s.env->mesh->ans_bogus++;
1353         }
1354         if(rep && rep->security == sec_status_secure)
1355                 secure = 1;
1356         else    secure = 0;
1357         if(!rep && rcode == LDNS_RCODE_NOERROR)
1358                 rcode = LDNS_RCODE_SERVFAIL;
1359         if(r->query_reply.c->use_h2) {
1360                 r->query_reply.c->h2_stream = r->h2_stream;
1361                 /* Mesh reply won't exist for long anymore. Make it impossible
1362                  * for HTTP/2 stream to refer to mesh state, in case
1363                  * connection gets cleanup before HTTP/2 stream close. */
1364                 r->h2_stream->mesh_state = NULL;
1365         }
1366         /* send the reply */
1367         /* We don't reuse the encoded answer if:
1368          * - either the previous or current response has a local alias.  We could
1369          *   compare the alias records and still reuse the previous answer if they
1370          *   are the same, but that would be complicated and error prone for the
1371          *   relatively minor case. So we err on the side of safety.
1372          * - there are registered callback functions for the given rcode, as these
1373          *   need to be called for each reply. */
1374         if(((rcode != LDNS_RCODE_SERVFAIL &&
1375                         !m->s.env->inplace_cb_lists[inplace_cb_reply]) ||
1376                 (rcode == LDNS_RCODE_SERVFAIL &&
1377                         !m->s.env->inplace_cb_lists[inplace_cb_reply_servfail])) &&
1378                 prev && prev_buffer && prev->qflags == r->qflags &&
1379                 !prev->local_alias && !r->local_alias &&
1380                 prev->edns.edns_present == r->edns.edns_present &&
1381                 prev->edns.bits == r->edns.bits &&
1382                 prev->edns.udp_size == r->edns.udp_size &&
1383                 edns_opt_list_compare(prev->edns.opt_list_out, r->edns.opt_list_out) == 0 &&
1384                 edns_opt_list_compare(prev->edns.opt_list_inplace_cb_out, r->edns.opt_list_inplace_cb_out) == 0
1385                 ) {
1386                 /* if the previous reply is identical to this one, fix ID */
1387                 if(prev_buffer != r_buffer)
1388                         sldns_buffer_copy(r_buffer, prev_buffer);
1389                 sldns_buffer_write_at(r_buffer, 0, &r->qid, sizeof(uint16_t));
1390                 sldns_buffer_write_at(r_buffer, 12, r->qname,
1391                         m->s.qinfo.qname_len);
1392                 m->reply_list = NULL;
1393                 comm_point_send_reply(&r->query_reply);
1394                 m->reply_list = rlist;
1395         } else if(rcode) {
1396                 m->s.qinfo.qname = r->qname;
1397                 m->s.qinfo.local_alias = r->local_alias;
1398                 if(rcode == LDNS_RCODE_SERVFAIL) {
1399                         if(!inplace_cb_reply_servfail_call(m->s.env, &m->s.qinfo, &m->s,
1400                                 rep, rcode, &r->edns, &r->query_reply, m->s.region, &r->start_time))
1401                                         r->edns.opt_list_inplace_cb_out = NULL;
1402                 } else {
1403                         if(!inplace_cb_reply_call(m->s.env, &m->s.qinfo, &m->s, rep, rcode,
1404                                 &r->edns, &r->query_reply, m->s.region, &r->start_time))
1405                                         r->edns.opt_list_inplace_cb_out = NULL;
1406                 }
1407                 /* Send along EDE BOGUS EDNS0 option when answer is bogus */
1408                 if(m->s.env->cfg->ede && rcode == LDNS_RCODE_SERVFAIL &&
1409                         m->s.env->need_to_validate && (!(r->qflags&BIT_CD) ||
1410                         m->s.env->cfg->ignore_cd) && rep &&
1411                         (rep->security <= sec_status_bogus ||
1412                         rep->security == sec_status_secure_sentinel_fail)) {
1413                         char *reason = m->s.env->cfg->val_log_level >= 2
1414                                 ? errinf_to_str_bogus(&m->s) : NULL;
1415
1416                         /* During validation the EDE code can be received via two
1417                          * code paths. One code path fills the reply_info EDE, and
1418                          * the other fills it in the errinf_strlist. These paths
1419                          * intersect at some points, but where is opaque due to
1420                          * the complexity of the validator. At the time of writing
1421                          * we make the choice to prefer the EDE from errinf_strlist
1422                          * but a compelling reason to do otherwise is just as valid
1423                          */
1424                         sldns_ede_code reason_bogus = errinf_to_reason_bogus(&m->s);
1425                         if ((reason_bogus == LDNS_EDE_DNSSEC_BOGUS &&
1426                                 rep->reason_bogus != LDNS_EDE_NONE) ||
1427                                 reason_bogus == LDNS_EDE_NONE) {
1428                                         reason_bogus = rep->reason_bogus;
1429                         }
1430
1431                         if(reason_bogus != LDNS_EDE_NONE) {
1432                                 edns_opt_list_append_ede(&r->edns.opt_list_out,
1433                                         m->s.region, reason_bogus, reason);
1434                         }
1435                         free(reason);
1436                 }
1437                 error_encode(r_buffer, rcode, &m->s.qinfo, r->qid,
1438                         r->qflags, &r->edns);
1439                 m->reply_list = NULL;
1440                 comm_point_send_reply(&r->query_reply);
1441                 m->reply_list = rlist;
1442         } else {
1443                 size_t udp_size = r->edns.udp_size;
1444                 r->edns.edns_version = EDNS_ADVERTISED_VERSION;
1445                 r->edns.udp_size = EDNS_ADVERTISED_SIZE;
1446                 r->edns.ext_rcode = 0;
1447                 r->edns.bits &= EDNS_DO;
1448                 m->s.qinfo.qname = r->qname;
1449                 m->s.qinfo.local_alias = r->local_alias;
1450                 if(!inplace_cb_reply_call(m->s.env, &m->s.qinfo, &m->s, rep,
1451                         LDNS_RCODE_NOERROR, &r->edns, &r->query_reply, m->s.region, &r->start_time) ||
1452                         !reply_info_answer_encode(&m->s.qinfo, rep, r->qid,
1453                         r->qflags, r_buffer, 0, 1, m->s.env->scratch,
1454                         udp_size, &r->edns, (int)(r->edns.bits & EDNS_DO),
1455                         secure))
1456                 {
1457                         if(!inplace_cb_reply_servfail_call(m->s.env, &m->s.qinfo, &m->s,
1458                         rep, LDNS_RCODE_SERVFAIL, &r->edns, &r->query_reply, m->s.region, &r->start_time))
1459                                 r->edns.opt_list_inplace_cb_out = NULL;
1460                         /* internal server error (probably malloc failure) so no
1461                          * EDE (RFC8914) needed */
1462                         error_encode(r_buffer, LDNS_RCODE_SERVFAIL,
1463                                 &m->s.qinfo, r->qid, r->qflags, &r->edns);
1464                 }
1465                 m->reply_list = NULL;
1466                 comm_point_send_reply(&r->query_reply);
1467                 m->reply_list = rlist;
1468         }
1469         /* account */
1470         log_assert(m->s.env->mesh->num_reply_addrs > 0);
1471         m->s.env->mesh->num_reply_addrs--;
1472         end_time = *m->s.env->now_tv;
1473         timeval_subtract(&duration, &end_time, &r->start_time);
1474         verbose(VERB_ALGO, "query took " ARG_LL "d.%6.6d sec",
1475                 (long long)duration.tv_sec, (int)duration.tv_usec);
1476         m->s.env->mesh->replies_sent++;
1477         timeval_add(&m->s.env->mesh->replies_sum_wait, &duration);
1478         timehist_insert(m->s.env->mesh->histogram, &duration);
1479         if(m->s.env->cfg->stat_extended) {
1480                 uint16_t rc = FLAGS_GET_RCODE(sldns_buffer_read_u16_at(
1481                         r_buffer, 2));
1482                 if(secure) m->s.env->mesh->ans_secure++;
1483                 m->s.env->mesh->ans_rcode[ rc ] ++;
1484                 if(rc == 0 && LDNS_ANCOUNT(sldns_buffer_begin(r_buffer)) == 0)
1485                         m->s.env->mesh->ans_nodata++;
1486         }
1487         /* Log reply sent */
1488         if(m->s.env->cfg->log_replies) {
1489                 log_reply_info(NO_VERBOSE, &m->s.qinfo, &r->query_reply.addr,
1490                         r->query_reply.addrlen, duration, 0, r_buffer);
1491         }
1492 }
1493
1494 void mesh_query_done(struct mesh_state* mstate)
1495 {
1496         struct mesh_reply* r;
1497         struct mesh_reply* prev = NULL;
1498         struct sldns_buffer* prev_buffer = NULL;
1499         struct mesh_cb* c;
1500         struct reply_info* rep = (mstate->s.return_msg?
1501                 mstate->s.return_msg->rep:NULL);
1502         struct timeval tv = {0, 0};
1503         /* No need for the serve expired timer anymore; we are going to reply. */
1504         if(mstate->s.serve_expired_data) {
1505                 comm_timer_delete(mstate->s.serve_expired_data->timer);
1506                 mstate->s.serve_expired_data->timer = NULL;
1507         }
1508         if(mstate->s.return_rcode == LDNS_RCODE_SERVFAIL ||
1509                 (rep && FLAGS_GET_RCODE(rep->flags) == LDNS_RCODE_SERVFAIL)) {
1510                 /* we are SERVFAILing; check for expired answer here */
1511                 mesh_serve_expired_callback(mstate);
1512                 if((mstate->reply_list || mstate->cb_list)
1513                 && mstate->s.env->cfg->log_servfail
1514                 && !mstate->s.env->cfg->val_log_squelch) {
1515                         char* err = errinf_to_str_servfail(&mstate->s);
1516                         if(err)
1517                                 log_err("%s", err);
1518                         free(err);
1519                 }
1520         }
1521         for(r = mstate->reply_list; r; r = r->next) {
1522                 tv = r->start_time;
1523
1524                 /* if a response-ip address block has been stored the
1525                  *  information should be logged for each client. */
1526                 if(mstate->s.respip_action_info &&
1527                         mstate->s.respip_action_info->addrinfo) {
1528                         respip_inform_print(mstate->s.respip_action_info,
1529                                 r->qname, mstate->s.qinfo.qtype,
1530                                 mstate->s.qinfo.qclass, r->local_alias,
1531                                 &r->query_reply);
1532                         if(mstate->s.env->cfg->stat_extended &&
1533                                 mstate->s.respip_action_info->rpz_used) {
1534                                 if(mstate->s.respip_action_info->rpz_disabled)
1535                                         mstate->s.env->mesh->rpz_action[RPZ_DISABLED_ACTION]++;
1536                                 if(mstate->s.respip_action_info->rpz_cname_override)
1537                                         mstate->s.env->mesh->rpz_action[RPZ_CNAME_OVERRIDE_ACTION]++;
1538                                 else
1539                                         mstate->s.env->mesh->rpz_action[respip_action_to_rpz_action(
1540                                                 mstate->s.respip_action_info->action)]++;
1541                         }
1542                 }
1543
1544                 /* if this query is determined to be dropped during the
1545                  * mesh processing, this is the point to take that action. */
1546                 if(mstate->s.is_drop) {
1547                         /* briefly set the reply_list to NULL, so that the
1548                          * tcp req info cleanup routine that calls the mesh
1549                          * to deregister the meshstate for it is not done
1550                          * because the list is NULL and also accounting is not
1551                          * done there, but instead we do that here. */
1552                         struct mesh_reply* reply_list = mstate->reply_list;
1553                         mstate->reply_list = NULL;
1554                         comm_point_drop_reply(&r->query_reply);
1555                         mstate->reply_list = reply_list;
1556                 } else {
1557                         struct sldns_buffer* r_buffer = r->query_reply.c->buffer;
1558                         if(r->query_reply.c->tcp_req_info) {
1559                                 r_buffer = r->query_reply.c->tcp_req_info->spool_buffer;
1560                                 prev_buffer = NULL;
1561                         }
1562                         mesh_send_reply(mstate, mstate->s.return_rcode, rep,
1563                                 r, r_buffer, prev, prev_buffer);
1564                         if(r->query_reply.c->tcp_req_info) {
1565                                 tcp_req_info_remove_mesh_state(r->query_reply.c->tcp_req_info, mstate);
1566                                 r_buffer = NULL;
1567                         }
1568                         prev = r;
1569                         prev_buffer = r_buffer;
1570                 }
1571         }
1572         if(mstate->reply_list) {
1573                 mstate->reply_list = NULL;
1574                 if(!mstate->reply_list && !mstate->cb_list) {
1575                         /* was a reply state, not anymore */
1576                         log_assert(mstate->s.env->mesh->num_reply_states > 0);
1577                         mstate->s.env->mesh->num_reply_states--;
1578                 }
1579                 if(!mstate->reply_list && !mstate->cb_list &&
1580                         mstate->super_set.count == 0)
1581                         mstate->s.env->mesh->num_detached_states++;
1582         }
1583         mstate->replies_sent = 1;
1584         while((c = mstate->cb_list) != NULL) {
1585                 /* take this cb off the list; so that the list can be
1586                  * changed, eg. by adds from the callback routine */
1587                 if(!mstate->reply_list && mstate->cb_list && !c->next) {
1588                         /* was a reply state, not anymore */
1589                         log_assert(mstate->s.env->mesh->num_reply_states > 0);
1590                         mstate->s.env->mesh->num_reply_states--;
1591                 }
1592                 mstate->cb_list = c->next;
1593                 if(!mstate->reply_list && !mstate->cb_list &&
1594                         mstate->super_set.count == 0)
1595                         mstate->s.env->mesh->num_detached_states++;
1596                 mesh_do_callback(mstate, mstate->s.return_rcode, rep, c, &tv);
1597         }
1598 }
1599
1600 void mesh_walk_supers(struct mesh_area* mesh, struct mesh_state* mstate)
1601 {
1602         struct mesh_state_ref* ref;
1603         RBTREE_FOR(ref, struct mesh_state_ref*, &mstate->super_set)
1604         {
1605                 /* make super runnable */
1606                 (void)rbtree_insert(&mesh->run, &ref->s->run_node);
1607                 /* callback the function to inform super of result */
1608                 fptr_ok(fptr_whitelist_mod_inform_super(
1609                         mesh->mods.mod[ref->s->s.curmod]->inform_super));
1610                 (*mesh->mods.mod[ref->s->s.curmod]->inform_super)(&mstate->s,
1611                         ref->s->s.curmod, &ref->s->s);
1612                 /* copy state that is always relevant to super */
1613                 copy_state_to_super(&mstate->s, ref->s->s.curmod, &ref->s->s);
1614         }
1615 }
1616
1617 struct mesh_state* mesh_area_find(struct mesh_area* mesh,
1618         struct respip_client_info* cinfo, struct query_info* qinfo,
1619         uint16_t qflags, int prime, int valrec)
1620 {
1621         struct mesh_state key;
1622         struct mesh_state* result;
1623
1624         key.node.key = &key;
1625         key.s.is_priming = prime;
1626         key.s.is_valrec = valrec;
1627         key.s.qinfo = *qinfo;
1628         key.s.query_flags = qflags;
1629         /* We are searching for a similar mesh state when we DO want to
1630          * aggregate the state. Thus unique is set to NULL. (default when we
1631          * desire aggregation).*/
1632         key.unique = NULL;
1633         key.s.client_info = cinfo;
1634
1635         result = (struct mesh_state*)rbtree_search(&mesh->all, &key);
1636         return result;
1637 }
1638
1639 int mesh_state_add_cb(struct mesh_state* s, struct edns_data* edns,
1640         sldns_buffer* buf, mesh_cb_func_type cb, void* cb_arg,
1641         uint16_t qid, uint16_t qflags)
1642 {
1643         struct mesh_cb* r = regional_alloc(s->s.region,
1644                 sizeof(struct mesh_cb));
1645         if(!r)
1646                 return 0;
1647         r->buf = buf;
1648         log_assert(fptr_whitelist_mesh_cb(cb)); /* early failure ifmissing*/
1649         r->cb = cb;
1650         r->cb_arg = cb_arg;
1651         r->edns = *edns;
1652         if(edns->opt_list_in && !(r->edns.opt_list_in =
1653                         edns_opt_copy_region(edns->opt_list_in, s->s.region)))
1654                 return 0;
1655         if(edns->opt_list_out && !(r->edns.opt_list_out =
1656                         edns_opt_copy_region(edns->opt_list_out, s->s.region)))
1657                 return 0;
1658         if(edns->opt_list_inplace_cb_out && !(r->edns.opt_list_inplace_cb_out =
1659                         edns_opt_copy_region(edns->opt_list_inplace_cb_out, s->s.region)))
1660                 return 0;
1661         r->qid = qid;
1662         r->qflags = qflags;
1663         r->next = s->cb_list;
1664         s->cb_list = r;
1665         return 1;
1666
1667 }
1668
1669 int mesh_state_add_reply(struct mesh_state* s, struct edns_data* edns,
1670         struct comm_reply* rep, uint16_t qid, uint16_t qflags,
1671         const struct query_info* qinfo)
1672 {
1673         struct mesh_reply* r = regional_alloc(s->s.region,
1674                 sizeof(struct mesh_reply));
1675         if(!r)
1676                 return 0;
1677         r->query_reply = *rep;
1678         r->edns = *edns;
1679         if(edns->opt_list_in && !(r->edns.opt_list_in =
1680                         edns_opt_copy_region(edns->opt_list_in, s->s.region)))
1681                 return 0;
1682         if(edns->opt_list_out && !(r->edns.opt_list_out =
1683                         edns_opt_copy_region(edns->opt_list_out, s->s.region)))
1684                 return 0;
1685         if(edns->opt_list_inplace_cb_out && !(r->edns.opt_list_inplace_cb_out =
1686                         edns_opt_copy_region(edns->opt_list_inplace_cb_out, s->s.region)))
1687                 return 0;
1688         r->qid = qid;
1689         r->qflags = qflags;
1690         r->start_time = *s->s.env->now_tv;
1691         r->next = s->reply_list;
1692         r->qname = regional_alloc_init(s->s.region, qinfo->qname,
1693                 s->s.qinfo.qname_len);
1694         if(!r->qname)
1695                 return 0;
1696         if(rep->c->use_h2)
1697                 r->h2_stream = rep->c->h2_stream;
1698
1699         /* Data related to local alias stored in 'qinfo' (if any) is ephemeral
1700          * and can be different for different original queries (even if the
1701          * replaced query name is the same).  So we need to make a deep copy
1702          * and store the copy for each reply info. */
1703         if(qinfo->local_alias) {
1704                 struct packed_rrset_data* d;
1705                 struct packed_rrset_data* dsrc;
1706                 r->local_alias = regional_alloc_zero(s->s.region,
1707                         sizeof(*qinfo->local_alias));
1708                 if(!r->local_alias)
1709                         return 0;
1710                 r->local_alias->rrset = regional_alloc_init(s->s.region,
1711                         qinfo->local_alias->rrset,
1712                         sizeof(*qinfo->local_alias->rrset));
1713                 if(!r->local_alias->rrset)
1714                         return 0;
1715                 dsrc = qinfo->local_alias->rrset->entry.data;
1716
1717                 /* In the current implementation, a local alias must be
1718                  * a single CNAME RR (see worker_handle_request()). */
1719                 log_assert(!qinfo->local_alias->next && dsrc->count == 1 &&
1720                         qinfo->local_alias->rrset->rk.type ==
1721                         htons(LDNS_RR_TYPE_CNAME));
1722                 /* we should make a local copy for the owner name of
1723                  * the RRset */
1724                 r->local_alias->rrset->rk.dname_len =
1725                         qinfo->local_alias->rrset->rk.dname_len;
1726                 r->local_alias->rrset->rk.dname = regional_alloc_init(
1727                         s->s.region, qinfo->local_alias->rrset->rk.dname,
1728                         qinfo->local_alias->rrset->rk.dname_len);
1729                 if(!r->local_alias->rrset->rk.dname)
1730                         return 0;
1731
1732                 /* the rrset is not packed, like in the cache, but it is
1733                  * individually allocated with an allocator from localzone. */
1734                 d = regional_alloc_zero(s->s.region, sizeof(*d));
1735                 if(!d)
1736                         return 0;
1737                 r->local_alias->rrset->entry.data = d;
1738                 if(!rrset_insert_rr(s->s.region, d, dsrc->rr_data[0],
1739                         dsrc->rr_len[0], dsrc->rr_ttl[0], "CNAME local alias"))
1740                         return 0;
1741         } else
1742                 r->local_alias = NULL;
1743
1744         s->reply_list = r;
1745         return 1;
1746 }
1747
1748 /* Extract the query info and flags from 'mstate' into '*qinfop' and '*qflags'.
1749  * Since this is only used for internal refetch of otherwise-expired answer,
1750  * we simply ignore the rare failure mode when memory allocation fails. */
1751 static void
1752 mesh_copy_qinfo(struct mesh_state* mstate, struct query_info** qinfop,
1753         uint16_t* qflags)
1754 {
1755         struct regional* region = mstate->s.env->scratch;
1756         struct query_info* qinfo;
1757
1758         qinfo = regional_alloc_init(region, &mstate->s.qinfo, sizeof(*qinfo));
1759         if(!qinfo)
1760                 return;
1761         qinfo->qname = regional_alloc_init(region, qinfo->qname,
1762                 qinfo->qname_len);
1763         if(!qinfo->qname)
1764                 return;
1765         *qinfop = qinfo;
1766         *qflags = mstate->s.query_flags;
1767 }
1768
1769 /**
1770  * Continue processing the mesh state at another module.
1771  * Handles module to modules transfer of control.
1772  * Handles module finished.
1773  * @param mesh: the mesh area.
1774  * @param mstate: currently active mesh state.
1775  *      Deleted if finished, calls _done and _supers to
1776  *      send replies to clients and inform other mesh states.
1777  *      This in turn may create additional runnable mesh states.
1778  * @param s: state at which the current module exited.
1779  * @param ev: the event sent to the module.
1780  *      returned is the event to send to the next module.
1781  * @return true if continue processing at the new module.
1782  *      false if not continued processing is needed.
1783  */
1784 static int
1785 mesh_continue(struct mesh_area* mesh, struct mesh_state* mstate,
1786         enum module_ext_state s, enum module_ev* ev)
1787 {
1788         mstate->num_activated++;
1789         if(mstate->num_activated > MESH_MAX_ACTIVATION) {
1790                 /* module is looping. Stop it. */
1791                 log_err("internal error: looping module (%s) stopped",
1792                         mesh->mods.mod[mstate->s.curmod]->name);
1793                 log_query_info(NO_VERBOSE, "pass error for qstate",
1794                         &mstate->s.qinfo);
1795                 s = module_error;
1796         }
1797         if(s == module_wait_module || s == module_restart_next) {
1798                 /* start next module */
1799                 mstate->s.curmod++;
1800                 if(mesh->mods.num == mstate->s.curmod) {
1801                         log_err("Cannot pass to next module; at last module");
1802                         log_query_info(VERB_QUERY, "pass error for qstate",
1803                                 &mstate->s.qinfo);
1804                         mstate->s.curmod--;
1805                         return mesh_continue(mesh, mstate, module_error, ev);
1806                 }
1807                 if(s == module_restart_next) {
1808                         int curmod = mstate->s.curmod;
1809                         for(; mstate->s.curmod < mesh->mods.num;
1810                                 mstate->s.curmod++) {
1811                                 fptr_ok(fptr_whitelist_mod_clear(
1812                                         mesh->mods.mod[mstate->s.curmod]->clear));
1813                                 (*mesh->mods.mod[mstate->s.curmod]->clear)
1814                                         (&mstate->s, mstate->s.curmod);
1815                                 mstate->s.minfo[mstate->s.curmod] = NULL;
1816                         }
1817                         mstate->s.curmod = curmod;
1818                 }
1819                 *ev = module_event_pass;
1820                 return 1;
1821         }
1822         if(s == module_wait_subquery && mstate->sub_set.count == 0) {
1823                 log_err("module cannot wait for subquery, subquery list empty");
1824                 log_query_info(VERB_QUERY, "pass error for qstate",
1825                         &mstate->s.qinfo);
1826                 s = module_error;
1827         }
1828         if(s == module_error && mstate->s.return_rcode == LDNS_RCODE_NOERROR) {
1829                 /* error is bad, handle pass back up below */
1830                 mstate->s.return_rcode = LDNS_RCODE_SERVFAIL;
1831         }
1832         if(s == module_error) {
1833                 mesh_query_done(mstate);
1834                 mesh_walk_supers(mesh, mstate);
1835                 mesh_state_delete(&mstate->s);
1836                 return 0;
1837         }
1838         if(s == module_finished) {
1839                 if(mstate->s.curmod == 0) {
1840                         struct query_info* qinfo = NULL;
1841                         uint16_t qflags;
1842                         int rpz_p = 0;
1843
1844                         mesh_query_done(mstate);
1845                         mesh_walk_supers(mesh, mstate);
1846
1847                         /* If the answer to the query needs to be refetched
1848                          * from an external DNS server, we'll need to schedule
1849                          * a prefetch after removing the current state, so
1850                          * we need to make a copy of the query info here. */
1851                         if(mstate->s.need_refetch) {
1852                                 mesh_copy_qinfo(mstate, &qinfo, &qflags);
1853                                 rpz_p = mstate->s.rpz_passthru;
1854                         }
1855
1856                         mesh_state_delete(&mstate->s);
1857                         if(qinfo) {
1858                                 mesh_schedule_prefetch(mesh, qinfo, qflags,
1859                                         0, 1, rpz_p);
1860                         }
1861                         return 0;
1862                 }
1863                 /* pass along the locus of control */
1864                 mstate->s.curmod --;
1865                 *ev = module_event_moddone;
1866                 return 1;
1867         }
1868         return 0;
1869 }
1870
1871 void mesh_run(struct mesh_area* mesh, struct mesh_state* mstate,
1872         enum module_ev ev, struct outbound_entry* e)
1873 {
1874         enum module_ext_state s;
1875         verbose(VERB_ALGO, "mesh_run: start");
1876         while(mstate) {
1877                 /* run the module */
1878                 fptr_ok(fptr_whitelist_mod_operate(
1879                         mesh->mods.mod[mstate->s.curmod]->operate));
1880                 (*mesh->mods.mod[mstate->s.curmod]->operate)
1881                         (&mstate->s, ev, mstate->s.curmod, e);
1882
1883                 /* examine results */
1884                 mstate->s.reply = NULL;
1885                 regional_free_all(mstate->s.env->scratch);
1886                 s = mstate->s.ext_state[mstate->s.curmod];
1887                 verbose(VERB_ALGO, "mesh_run: %s module exit state is %s",
1888                         mesh->mods.mod[mstate->s.curmod]->name, strextstate(s));
1889                 e = NULL;
1890                 if(mesh_continue(mesh, mstate, s, &ev))
1891                         continue;
1892
1893                 /* run more modules */
1894                 ev = module_event_pass;
1895                 if(mesh->run.count > 0) {
1896                         /* pop random element off the runnable tree */
1897                         mstate = (struct mesh_state*)mesh->run.root->key;
1898                         (void)rbtree_delete(&mesh->run, mstate);
1899                 } else mstate = NULL;
1900         }
1901         if(verbosity >= VERB_ALGO) {
1902                 mesh_stats(mesh, "mesh_run: end");
1903                 mesh_log_list(mesh);
1904         }
1905 }
1906
1907 void
1908 mesh_log_list(struct mesh_area* mesh)
1909 {
1910         char buf[30];
1911         struct mesh_state* m;
1912         int num = 0;
1913         RBTREE_FOR(m, struct mesh_state*, &mesh->all) {
1914                 snprintf(buf, sizeof(buf), "%d%s%s%s%s%s%s mod%d %s%s",
1915                         num++, (m->s.is_priming)?"p":"",  /* prime */
1916                         (m->s.is_valrec)?"v":"",  /* prime */
1917                         (m->s.query_flags&BIT_RD)?"RD":"",
1918                         (m->s.query_flags&BIT_CD)?"CD":"",
1919                         (m->super_set.count==0)?"d":"", /* detached */
1920                         (m->sub_set.count!=0)?"c":"",  /* children */
1921                         m->s.curmod, (m->reply_list)?"rep":"", /*hasreply*/
1922                         (m->cb_list)?"cb":"" /* callbacks */
1923                         );
1924                 log_query_info(VERB_ALGO, buf, &m->s.qinfo);
1925         }
1926 }
1927
1928 void
1929 mesh_stats(struct mesh_area* mesh, const char* str)
1930 {
1931         verbose(VERB_DETAIL, "%s %u recursion states (%u with reply, "
1932                 "%u detached), %u waiting replies, %u recursion replies "
1933                 "sent, %d replies dropped, %d states jostled out",
1934                 str, (unsigned)mesh->all.count,
1935                 (unsigned)mesh->num_reply_states,
1936                 (unsigned)mesh->num_detached_states,
1937                 (unsigned)mesh->num_reply_addrs,
1938                 (unsigned)mesh->replies_sent,
1939                 (unsigned)mesh->stats_dropped,
1940                 (unsigned)mesh->stats_jostled);
1941         if(mesh->replies_sent > 0) {
1942                 struct timeval avg;
1943                 timeval_divide(&avg, &mesh->replies_sum_wait,
1944                         mesh->replies_sent);
1945                 log_info("average recursion processing time "
1946                         ARG_LL "d.%6.6d sec",
1947                         (long long)avg.tv_sec, (int)avg.tv_usec);
1948                 log_info("histogram of recursion processing times");
1949                 timehist_log(mesh->histogram, "recursions");
1950         }
1951 }
1952
1953 void
1954 mesh_stats_clear(struct mesh_area* mesh)
1955 {
1956         if(!mesh)
1957                 return;
1958         mesh->replies_sent = 0;
1959         mesh->replies_sum_wait.tv_sec = 0;
1960         mesh->replies_sum_wait.tv_usec = 0;
1961         mesh->stats_jostled = 0;
1962         mesh->stats_dropped = 0;
1963         timehist_clear(mesh->histogram);
1964         mesh->ans_secure = 0;
1965         mesh->ans_bogus = 0;
1966         mesh->ans_expired = 0;
1967         memset(&mesh->ans_rcode[0], 0, sizeof(size_t)*UB_STATS_RCODE_NUM);
1968         memset(&mesh->rpz_action[0], 0, sizeof(size_t)*UB_STATS_RPZ_ACTION_NUM);
1969         mesh->ans_nodata = 0;
1970 }
1971
1972 size_t
1973 mesh_get_mem(struct mesh_area* mesh)
1974 {
1975         struct mesh_state* m;
1976         size_t s = sizeof(*mesh) + sizeof(struct timehist) +
1977                 sizeof(struct th_buck)*mesh->histogram->num +
1978                 sizeof(sldns_buffer) + sldns_buffer_capacity(mesh->qbuf_bak);
1979         RBTREE_FOR(m, struct mesh_state*, &mesh->all) {
1980                 /* all, including m itself allocated in qstate region */
1981                 s += regional_get_mem(m->s.region);
1982         }
1983         return s;
1984 }
1985
1986 int
1987 mesh_detect_cycle(struct module_qstate* qstate, struct query_info* qinfo,
1988         uint16_t flags, int prime, int valrec)
1989 {
1990         struct mesh_area* mesh = qstate->env->mesh;
1991         struct mesh_state* dep_m = NULL;
1992         dep_m = mesh_area_find(mesh, NULL, qinfo, flags, prime, valrec);
1993         return mesh_detect_cycle_found(qstate, dep_m);
1994 }
1995
1996 void mesh_list_insert(struct mesh_state* m, struct mesh_state** fp,
1997         struct mesh_state** lp)
1998 {
1999         /* insert as last element */
2000         m->prev = *lp;
2001         m->next = NULL;
2002         if(*lp)
2003                 (*lp)->next = m;
2004         else    *fp = m;
2005         *lp = m;
2006 }
2007
2008 void mesh_list_remove(struct mesh_state* m, struct mesh_state** fp,
2009         struct mesh_state** lp)
2010 {
2011         if(m->next)
2012                 m->next->prev = m->prev;
2013         else    *lp = m->prev;
2014         if(m->prev)
2015                 m->prev->next = m->next;
2016         else    *fp = m->next;
2017 }
2018
2019 void mesh_state_remove_reply(struct mesh_area* mesh, struct mesh_state* m,
2020         struct comm_point* cp)
2021 {
2022         struct mesh_reply* n, *prev = NULL;
2023         n = m->reply_list;
2024         /* when in mesh_cleanup, it sets the reply_list to NULL, so that
2025          * there is no accounting twice */
2026         if(!n) return; /* nothing to remove, also no accounting needed */
2027         while(n) {
2028                 if(n->query_reply.c == cp) {
2029                         /* unlink it */
2030                         if(prev) prev->next = n->next;
2031                         else m->reply_list = n->next;
2032                         /* delete it, but allocated in m region */
2033                         log_assert(mesh->num_reply_addrs > 0);
2034                         mesh->num_reply_addrs--;
2035
2036                         /* prev = prev; */
2037                         n = n->next;
2038                         continue;
2039                 }
2040                 prev = n;
2041                 n = n->next;
2042         }
2043         /* it was not detached (because it had a reply list), could be now */
2044         if(!m->reply_list && !m->cb_list
2045                 && m->super_set.count == 0) {
2046                 mesh->num_detached_states++;
2047         }
2048         /* if not replies any more in mstate, it is no longer a reply_state */
2049         if(!m->reply_list && !m->cb_list) {
2050                 log_assert(mesh->num_reply_states > 0);
2051                 mesh->num_reply_states--;
2052         }
2053 }
2054
2055
2056 static int
2057 apply_respip_action(struct module_qstate* qstate,
2058         const struct query_info* qinfo, struct respip_client_info* cinfo,
2059         struct respip_action_info* actinfo, struct reply_info* rep,
2060         struct ub_packed_rrset_key** alias_rrset,
2061         struct reply_info** encode_repp, struct auth_zones* az)
2062 {
2063         if(qinfo->qtype != LDNS_RR_TYPE_A &&
2064                 qinfo->qtype != LDNS_RR_TYPE_AAAA &&
2065                 qinfo->qtype != LDNS_RR_TYPE_ANY)
2066                 return 1;
2067
2068         if(!respip_rewrite_reply(qinfo, cinfo, rep, encode_repp, actinfo,
2069                 alias_rrset, 0, qstate->region, az, NULL))
2070                 return 0;
2071
2072         /* xxx_deny actions mean dropping the reply, unless the original reply
2073          * was redirected to response-ip data. */
2074         if((actinfo->action == respip_deny ||
2075                 actinfo->action == respip_inform_deny) &&
2076                 *encode_repp == rep)
2077                 *encode_repp = NULL;
2078
2079         return 1;
2080 }
2081
2082 void
2083 mesh_serve_expired_callback(void* arg)
2084 {
2085         struct mesh_state* mstate = (struct mesh_state*) arg;
2086         struct module_qstate* qstate = &mstate->s;
2087         struct mesh_reply* r;
2088         struct mesh_area* mesh = qstate->env->mesh;
2089         struct dns_msg* msg;
2090         struct mesh_cb* c;
2091         struct mesh_reply* prev = NULL;
2092         struct sldns_buffer* prev_buffer = NULL;
2093         struct sldns_buffer* r_buffer = NULL;
2094         struct reply_info* partial_rep = NULL;
2095         struct ub_packed_rrset_key* alias_rrset = NULL;
2096         struct reply_info* encode_rep = NULL;
2097         struct respip_action_info actinfo;
2098         struct query_info* lookup_qinfo = &qstate->qinfo;
2099         struct query_info qinfo_tmp;
2100         struct timeval tv = {0, 0};
2101         int must_validate = (!(qstate->query_flags&BIT_CD)
2102                 || qstate->env->cfg->ignore_cd) && qstate->env->need_to_validate;
2103         if(!qstate->serve_expired_data) return;
2104         verbose(VERB_ALGO, "Serve expired: Trying to reply with expired data");
2105         comm_timer_delete(qstate->serve_expired_data->timer);
2106         qstate->serve_expired_data->timer = NULL;
2107         /* If is_drop or no_cache_lookup (modules that handle their own cache e.g.,
2108          * subnetmod) ignore stale data from the main cache. */
2109         if(qstate->no_cache_lookup || qstate->is_drop) {
2110                 verbose(VERB_ALGO,
2111                         "Serve expired: Not allowed to look into cache for stale");
2112                 return;
2113         }
2114         /* The following while is used instead of the `goto lookup_cache`
2115          * like in the worker. */
2116         while(1) {
2117                 fptr_ok(fptr_whitelist_serve_expired_lookup(
2118                         qstate->serve_expired_data->get_cached_answer));
2119                 msg = (*qstate->serve_expired_data->get_cached_answer)(qstate,
2120                         lookup_qinfo);
2121                 if(!msg)
2122                         return;
2123                 /* Reset these in case we pass a second time from here. */
2124                 encode_rep = msg->rep;
2125                 memset(&actinfo, 0, sizeof(actinfo));
2126                 actinfo.action = respip_none;
2127                 alias_rrset = NULL;
2128                 if((mesh->use_response_ip || mesh->use_rpz) &&
2129                         !partial_rep && !apply_respip_action(qstate, &qstate->qinfo,
2130                         qstate->client_info, &actinfo, msg->rep, &alias_rrset, &encode_rep,
2131                         qstate->env->auth_zones)) {
2132                         return;
2133                 } else if(partial_rep &&
2134                         !respip_merge_cname(partial_rep, &qstate->qinfo, msg->rep,
2135                         qstate->client_info, must_validate, &encode_rep, qstate->region,
2136                         qstate->env->auth_zones)) {
2137                         return;
2138                 }
2139                 if(!encode_rep || alias_rrset) {
2140                         if(!encode_rep) {
2141                                 /* Needs drop */
2142                                 return;
2143                         } else {
2144                                 /* A partial CNAME chain is found. */
2145                                 partial_rep = encode_rep;
2146                         }
2147                 }
2148                 /* We've found a partial reply ending with an
2149                 * alias.  Replace the lookup qinfo for the
2150                 * alias target and lookup the cache again to
2151                 * (possibly) complete the reply.  As we're
2152                 * passing the "base" reply, there will be no
2153                 * more alias chasing. */
2154                 if(partial_rep) {
2155                         memset(&qinfo_tmp, 0, sizeof(qinfo_tmp));
2156                         get_cname_target(alias_rrset, &qinfo_tmp.qname,
2157                                 &qinfo_tmp.qname_len);
2158                         if(!qinfo_tmp.qname) {
2159                                 log_err("Serve expired: unexpected: invalid answer alias");
2160                                 return;
2161                         }
2162                         qinfo_tmp.qtype = qstate->qinfo.qtype;
2163                         qinfo_tmp.qclass = qstate->qinfo.qclass;
2164                         lookup_qinfo = &qinfo_tmp;
2165                         continue;
2166                 }
2167                 break;
2168         }
2169
2170         if(verbosity >= VERB_ALGO)
2171                 log_dns_msg("Serve expired lookup", &qstate->qinfo, msg->rep);
2172
2173         for(r = mstate->reply_list; r; r = r->next) {
2174                 tv = r->start_time;
2175
2176                 /* If address info is returned, it means the action should be an
2177                 * 'inform' variant and the information should be logged. */
2178                 if(actinfo.addrinfo) {
2179                         respip_inform_print(&actinfo, r->qname,
2180                                 qstate->qinfo.qtype, qstate->qinfo.qclass,
2181                                 r->local_alias, &r->query_reply);
2182
2183                         if(qstate->env->cfg->stat_extended && actinfo.rpz_used) {
2184                                 if(actinfo.rpz_disabled)
2185                                         qstate->env->mesh->rpz_action[RPZ_DISABLED_ACTION]++;
2186                                 if(actinfo.rpz_cname_override)
2187                                         qstate->env->mesh->rpz_action[RPZ_CNAME_OVERRIDE_ACTION]++;
2188                                 else
2189                                         qstate->env->mesh->rpz_action[
2190                                                 respip_action_to_rpz_action(actinfo.action)]++;
2191                         }
2192                 }
2193
2194                 /* Add EDE Stale Answer (RCF8914). Ignore global ede as this is
2195                  * warning instead of an error */
2196                 if (r->edns.edns_present && qstate->env->cfg->ede_serve_expired &&
2197                         qstate->env->cfg->ede) {
2198                         edns_opt_list_append_ede(&r->edns.opt_list_out,
2199                                 mstate->s.region, LDNS_EDE_STALE_ANSWER, NULL);
2200                 }
2201
2202                 r_buffer = r->query_reply.c->buffer;
2203                 if(r->query_reply.c->tcp_req_info)
2204                         r_buffer = r->query_reply.c->tcp_req_info->spool_buffer;
2205                 mesh_send_reply(mstate, LDNS_RCODE_NOERROR, msg->rep,
2206                         r, r_buffer, prev, prev_buffer);
2207                 if(r->query_reply.c->tcp_req_info)
2208                         tcp_req_info_remove_mesh_state(r->query_reply.c->tcp_req_info, mstate);
2209                 prev = r;
2210                 prev_buffer = r_buffer;
2211
2212                 /* Account for each reply sent. */
2213                 mesh->ans_expired++;
2214
2215         }
2216         if(mstate->reply_list) {
2217                 mstate->reply_list = NULL;
2218                 if(!mstate->reply_list && !mstate->cb_list) {
2219                         log_assert(mesh->num_reply_states > 0);
2220                         mesh->num_reply_states--;
2221                         if(mstate->super_set.count == 0) {
2222                                 mesh->num_detached_states++;
2223                         }
2224                 }
2225         }
2226         while((c = mstate->cb_list) != NULL) {
2227                 /* take this cb off the list; so that the list can be
2228                  * changed, eg. by adds from the callback routine */
2229                 if(!mstate->reply_list && mstate->cb_list && !c->next) {
2230                         /* was a reply state, not anymore */
2231                         log_assert(qstate->env->mesh->num_reply_states > 0);
2232                         qstate->env->mesh->num_reply_states--;
2233                 }
2234                 mstate->cb_list = c->next;
2235                 if(!mstate->reply_list && !mstate->cb_list &&
2236                         mstate->super_set.count == 0)
2237                         qstate->env->mesh->num_detached_states++;
2238                 mesh_do_callback(mstate, LDNS_RCODE_NOERROR, msg->rep, c, &tv);
2239         }
2240 }