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