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         /* add serve expired timer if required and not already there */
 555         if(timeout && !mesh_serve_expired_init(s, timeout)) {
 556                 log_err("mesh_new_client: out of memory initializing serve expired");
 557                 goto servfail_mem;
 558         }
 559         /* update statistics */
 560         if(was_detached) {
 561                 log_assert(mesh->num_detached_states > 0);
 562                 mesh->num_detached_states--;
 563         }
 564         if(was_noreply) {
 565                 mesh->num_reply_states ++;
 566         }
 567         mesh->num_reply_addrs++;
 568         if(s->list_select == mesh_no_list) {
 569                 /* move to either the forever or the jostle_list */
 570                 if(mesh->num_forever_states < mesh->max_forever_states) {
 571                         mesh->num_forever_states ++;
 572                         mesh_list_insert(s, &mesh->forever_first,
 573                                 &mesh->forever_last);
 574                         s->list_select = mesh_forever_list;
 575                 } else {
 576                         mesh_list_insert(s, &mesh->jostle_first,
 577                                 &mesh->jostle_last);
 578                         s->list_select = mesh_jostle_list;
 579                 }
 580         }
 581         if(added)
 582                 mesh_run(mesh, s, module_event_new, NULL);
 583         return;
 584
 585 servfail_mem:
 586         if(!inplace_cb_reply_servfail_call(mesh->env, qinfo, &s->s,
 587                 NULL, LDNS_RCODE_SERVFAIL, edns, rep, mesh->env->scratch))
 588                         edns->opt_list = NULL;
 589         error_encode(r_buffer, LDNS_RCODE_SERVFAIL,
 590                 qinfo, qid, qflags, edns);
 591         comm_point_send_reply(rep);
 592         if(added)
 593                 mesh_state_delete(&s->s);
 594         return;
 595 }
 596
 597 int
 598 mesh_new_callback(struct mesh_area* mesh, struct query_info* qinfo,
 599         uint16_t qflags, struct edns_data* edns, sldns_buffer* buf,
 600         uint16_t qid, mesh_cb_func_type cb, void* cb_arg)
 601 {
 602         struct mesh_state* s = NULL;
 603         int unique = unique_mesh_state(edns->opt_list, mesh->env);
 604         int timeout = mesh->env->cfg->serve_expired?
 605                 mesh->env->cfg->serve_expired_client_timeout:0;
 606         int was_detached = 0;
 607         int was_noreply = 0;
 608         int added = 0;
 609         if(!unique)
 610                 s = mesh_area_find(mesh, NULL, qinfo, qflags&(BIT_RD|BIT_CD), 0, 0);
 611
 612         /* there are no limits on the number of callbacks */
 613
 614         /* see if it already exists, if not, create one */
 615         if(!s) {
 616 #ifdef UNBOUND_DEBUG
 617                 struct rbnode_type* n;
 618 #endif
 619                 s = mesh_state_create(mesh->env, qinfo, NULL,
 620                         qflags&(BIT_RD|BIT_CD), 0, 0);
 621                 if(!s) {
 622                         return 0;
 623                 }
 624                 if(unique)
 625                         mesh_state_make_unique(s);
 626                 if(edns->opt_list) {
 627                         s->s.edns_opts_front_in = edns_opt_copy_region(edns->opt_list,
 628                                 s->s.region);
 629                         if(!s->s.edns_opts_front_in) {
 630                                 return 0;
 631                         }
 632                 }
 633 #ifdef UNBOUND_DEBUG
 634                 n =
 635 #else
 636                 (void)
 637 #endif
 638                 rbtree_insert(&mesh->all, &s->node);
 639                 log_assert(n != NULL);
 640                 /* set detached (it is now) */
 641                 mesh->num_detached_states++;
 642                 added = 1;
 643         }
 644         if(!s->reply_list && !s->cb_list) {
 645                 was_noreply = 1;
 646                 if(s->super_set.count == 0) {
 647                         was_detached = 1;
 648                 }
 649         }
 650         /* add reply to s */
 651         if(!mesh_state_add_cb(s, edns, buf, cb, cb_arg, qid, qflags)) {
 652                 if(added)
 653                         mesh_state_delete(&s->s);
 654                 return 0;
 655         }
 656         /* add serve expired timer if not already there */
 657         if(timeout && !mesh_serve_expired_init(s, timeout)) {
 658                 return 0;
 659         }
 660         /* update statistics */
 661         if(was_detached) {
 662                 log_assert(mesh->num_detached_states > 0);
 663                 mesh->num_detached_states--;
 664         }
 665         if(was_noreply) {
 666                 mesh->num_reply_states ++;
 667         }
 668         mesh->num_reply_addrs++;
 669         if(added)
 670                 mesh_run(mesh, s, module_event_new, NULL);
 671         return 1;
 672 }
 673
 674 /* Internal backend routine of mesh_new_prefetch().  It takes one additional
 675  * parameter, 'run', which controls whether to run the prefetch state
 676  * immediately.  When this function is called internally 'run' could be
 677  * 0 (false), in which case the new state is only made runnable so it
 678  * will not be run recursively on top of the current state. */
 679 static void mesh_schedule_prefetch(struct mesh_area* mesh,
 680         struct query_info* qinfo, uint16_t qflags, time_t leeway, int run)
 681 {
 682         struct mesh_state* s = mesh_area_find(mesh, NULL, qinfo,
 683                 qflags&(BIT_RD|BIT_CD), 0, 0);
 684 #ifdef UNBOUND_DEBUG
 685         struct rbnode_type* n;
 686 #endif
 687         /* already exists, and for a different purpose perhaps.
 688          * if mesh_no_list, keep it that way. */
 689         if(s) {
 690                 /* make it ignore the cache from now on */
 691                 if(!s->s.blacklist)
 692                         sock_list_insert(&s->s.blacklist, NULL, 0, s->s.region);
 693                 if(s->s.prefetch_leeway < leeway)
 694                         s->s.prefetch_leeway = leeway;
 695                 return;
 696         }
 697         if(!mesh_make_new_space(mesh, NULL)) {
 698                 verbose(VERB_ALGO, "Too many queries. dropped prefetch.");
 699                 mesh->stats_dropped ++;
 700                 return;
 701         }
 702
 703         s = mesh_state_create(mesh->env, qinfo, NULL,
 704                 qflags&(BIT_RD|BIT_CD), 0, 0);
 705         if(!s) {
 706                 log_err("prefetch mesh_state_create: out of memory");
 707                 return;
 708         }
 709 #ifdef UNBOUND_DEBUG
 710         n =
 711 #else
 712         (void)
 713 #endif
 714         rbtree_insert(&mesh->all, &s->node);
 715         log_assert(n != NULL);
 716         /* set detached (it is now) */
 717         mesh->num_detached_states++;
 718         /* make it ignore the cache */
 719         sock_list_insert(&s->s.blacklist, NULL, 0, s->s.region);
 720         s->s.prefetch_leeway = leeway;
 721
 722         if(s->list_select == mesh_no_list) {
 723                 /* move to either the forever or the jostle_list */
 724                 if(mesh->num_forever_states < mesh->max_forever_states) {
 725                         mesh->num_forever_states ++;
 726                         mesh_list_insert(s, &mesh->forever_first,
 727                                 &mesh->forever_last);
 728                         s->list_select = mesh_forever_list;
 729                 } else {
 730                         mesh_list_insert(s, &mesh->jostle_first,
 731                                 &mesh->jostle_last);
 732                         s->list_select = mesh_jostle_list;
 733                 }
 734         }
 735
 736         if(!run) {
 737 #ifdef UNBOUND_DEBUG
 738                 n =
 739 #else
 740                 (void)
 741 #endif
 742                 rbtree_insert(&mesh->run, &s->run_node);
 743                 log_assert(n != NULL);
 744                 return;
 745         }
 746
 747         mesh_run(mesh, s, module_event_new, NULL);
 748 }
 749
 750 void mesh_new_prefetch(struct mesh_area* mesh, struct query_info* qinfo,
 751         uint16_t qflags, time_t leeway)
 752 {
 753         mesh_schedule_prefetch(mesh, qinfo, qflags, leeway, 1);
 754 }
 755
 756 void mesh_report_reply(struct mesh_area* mesh, struct outbound_entry* e,
 757         struct comm_reply* reply, int what)
 758 {
 759         enum module_ev event = module_event_reply;
 760         e->qstate->reply = reply;
 761         if(what != NETEVENT_NOERROR) {
 762                 event = module_event_noreply;
 763                 if(what == NETEVENT_CAPSFAIL)
 764                         event = module_event_capsfail;
 765         }
 766         mesh_run(mesh, e->qstate->mesh_info, event, e);
 767 }
 768
 769 struct mesh_state*
 770 mesh_state_create(struct module_env* env, struct query_info* qinfo,
 771         struct respip_client_info* cinfo, uint16_t qflags, int prime,
 772         int valrec)
 773 {
 774         struct regional* region = alloc_reg_obtain(env->alloc);
 775         struct mesh_state* mstate;
 776         int i;
 777         if(!region)
 778                 return NULL;
 779         mstate = (struct mesh_state*)regional_alloc(region,
 780                 sizeof(struct mesh_state));
 781         if(!mstate) {
 782                 alloc_reg_release(env->alloc, region);
 783                 return NULL;
 784         }
 785         memset(mstate, 0, sizeof(*mstate));
 786         mstate->node = *RBTREE_NULL;
 787         mstate->run_node = *RBTREE_NULL;
 788         mstate->node.key = mstate;
 789         mstate->run_node.key = mstate;
 790         mstate->reply_list = NULL;
 791         mstate->list_select = mesh_no_list;
 792         mstate->replies_sent = 0;
 793         rbtree_init(&mstate->super_set, &mesh_state_ref_compare);
 794         rbtree_init(&mstate->sub_set, &mesh_state_ref_compare);
 795         mstate->num_activated = 0;
 796         mstate->unique = NULL;
 797         /* init module qstate */
 798         mstate->s.qinfo.qtype = qinfo->qtype;
 799         mstate->s.qinfo.qclass = qinfo->qclass;
 800         mstate->s.qinfo.local_alias = NULL;
 801         mstate->s.qinfo.qname_len = qinfo->qname_len;
 802         mstate->s.qinfo.qname = regional_alloc_init(region, qinfo->qname,
 803                 qinfo->qname_len);
 804         if(!mstate->s.qinfo.qname) {
 805                 alloc_reg_release(env->alloc, region);
 806                 return NULL;
 807         }
 808         if(cinfo) {
 809                 mstate->s.client_info = regional_alloc_init(region, cinfo,
 810                         sizeof(*cinfo));
 811                 if(!mstate->s.client_info) {
 812                         alloc_reg_release(env->alloc, region);
 813                         return NULL;
 814                 }
 815         }
 816         /* remove all weird bits from qflags */
 817         mstate->s.query_flags = (qflags & (BIT_RD|BIT_CD));
 818         mstate->s.is_priming = prime;
 819         mstate->s.is_valrec = valrec;
 820         mstate->s.reply = NULL;
 821         mstate->s.region = region;
 822         mstate->s.curmod = 0;
 823         mstate->s.return_msg = 0;
 824         mstate->s.return_rcode = LDNS_RCODE_NOERROR;
 825         mstate->s.env = env;
 826         mstate->s.mesh_info = mstate;
 827         mstate->s.prefetch_leeway = 0;
 828         mstate->s.serve_expired_data = NULL;
 829         mstate->s.no_cache_lookup = 0;
 830         mstate->s.no_cache_store = 0;
 831         mstate->s.need_refetch = 0;
 832         mstate->s.was_ratelimited = 0;
 833
 834         /* init modules */
 835         for(i=0; i<env->mesh->mods.num; i++) {
 836                 mstate->s.minfo[i] = NULL;
 837                 mstate->s.ext_state[i] = module_state_initial;
 838         }
 839         /* init edns option lists */
 840         mstate->s.edns_opts_front_in = NULL;
 841         mstate->s.edns_opts_back_out = NULL;
 842         mstate->s.edns_opts_back_in = NULL;
 843         mstate->s.edns_opts_front_out = NULL;
 844
 845         return mstate;
 846 }
 847
 848 int
 849 mesh_state_is_unique(struct mesh_state* mstate)
 850 {
 851         return mstate->unique != NULL;
 852 }
 853
 854 void
 855 mesh_state_make_unique(struct mesh_state* mstate)
 856 {
 857         mstate->unique = mstate;
 858 }
 859
 860 void
 861 mesh_state_cleanup(struct mesh_state* mstate)
 862 {
 863         struct mesh_area* mesh;
 864         int i;
 865         if(!mstate)
 866                 return;
 867         mesh = mstate->s.env->mesh;
 868         /* Stop and delete the serve expired timer */
 869         if(mstate->s.serve_expired_data && mstate->s.serve_expired_data->timer) {
 870                 comm_timer_delete(mstate->s.serve_expired_data->timer);
 871                 mstate->s.serve_expired_data->timer = NULL;
 872         }
 873         /* drop unsent replies */
 874         if(!mstate->replies_sent) {
 875                 struct mesh_reply* rep = mstate->reply_list;
 876                 struct mesh_cb* cb;
 877                 /* in tcp_req_info, the mstates linked are removed, but
 878                  * the reply_list is now NULL, so the remove-from-empty-list
 879                  * takes no time and also it does not do the mesh accounting */
 880                 mstate->reply_list = NULL;
 881                 for(; rep; rep=rep->next) {
 882                         comm_point_drop_reply(&rep->query_reply);
 883                         log_assert(mesh->num_reply_addrs > 0);
 884                         mesh->num_reply_addrs--;
 885                 }
 886                 while((cb = mstate->cb_list)!=NULL) {
 887                         mstate->cb_list = cb->next;
 888                         fptr_ok(fptr_whitelist_mesh_cb(cb->cb));
 889                         (*cb->cb)(cb->cb_arg, LDNS_RCODE_SERVFAIL, NULL,
 890                                 sec_status_unchecked, NULL, 0);
 891                         log_assert(mesh->num_reply_addrs > 0);
 892                         mesh->num_reply_addrs--;
 893                 }
 894         }
 895
 896         /* de-init modules */
 897         for(i=0; i<mesh->mods.num; i++) {
 898                 fptr_ok(fptr_whitelist_mod_clear(mesh->mods.mod[i]->clear));
 899                 (*mesh->mods.mod[i]->clear)(&mstate->s, i);
 900                 mstate->s.minfo[i] = NULL;
 901                 mstate->s.ext_state[i] = module_finished;
 902         }
 903         alloc_reg_release(mstate->s.env->alloc, mstate->s.region);
 904 }
 905
 906 void
 907 mesh_state_delete(struct module_qstate* qstate)
 908 {
 909         struct mesh_area* mesh;
 910         struct mesh_state_ref* super, ref;
 911         struct mesh_state* mstate;
 912         if(!qstate)
 913                 return;
 914         mstate = qstate->mesh_info;
 915         mesh = mstate->s.env->mesh;
 916         mesh_detach_subs(&mstate->s);
 917         if(mstate->list_select == mesh_forever_list) {
 918                 mesh->num_forever_states --;
 919                 mesh_list_remove(mstate, &mesh->forever_first,
 920                         &mesh->forever_last);
 921         } else if(mstate->list_select == mesh_jostle_list) {
 922                 mesh_list_remove(mstate, &mesh->jostle_first,
 923                         &mesh->jostle_last);
 924         }
 925         if(!mstate->reply_list && !mstate->cb_list
 926                 && mstate->super_set.count == 0) {
 927                 log_assert(mesh->num_detached_states > 0);
 928                 mesh->num_detached_states--;
 929         }
 930         if(mstate->reply_list || mstate->cb_list) {
 931                 log_assert(mesh->num_reply_states > 0);
 932                 mesh->num_reply_states--;
 933         }
 934         ref.node.key = &ref;
 935         ref.s = mstate;
 936         RBTREE_FOR(super, struct mesh_state_ref*, &mstate->super_set) {
 937                 (void)rbtree_delete(&super->s->sub_set, &ref);
 938         }
 939         (void)rbtree_delete(&mesh->run, mstate);
 940         (void)rbtree_delete(&mesh->all, mstate);
 941         mesh_state_cleanup(mstate);
 942 }
 943
 944 /** helper recursive rbtree find routine */
 945 static int
 946 find_in_subsub(struct mesh_state* m, struct mesh_state* tofind, size_t *c)
 947 {
 948         struct mesh_state_ref* r;
 949         if((*c)++ > MESH_MAX_SUBSUB)
 950                 return 1;
 951         RBTREE_FOR(r, struct mesh_state_ref*, &m->sub_set) {
 952                 if(r->s == tofind || find_in_subsub(r->s, tofind, c))
 953                         return 1;
 954         }
 955         return 0;
 956 }
 957
 958 /** find cycle for already looked up mesh_state */
 959 static int
 960 mesh_detect_cycle_found(struct module_qstate* qstate, struct mesh_state* dep_m)
 961 {
 962         struct mesh_state* cyc_m = qstate->mesh_info;
 963         size_t counter = 0;
 964         if(!dep_m)
 965                 return 0;
 966         if(dep_m == cyc_m || find_in_subsub(dep_m, cyc_m, &counter)) {
 967                 if(counter > MESH_MAX_SUBSUB)
 968                         return 2;
 969                 return 1;
 970         }
 971         return 0;
 972 }
 973
 974 void mesh_detach_subs(struct module_qstate* qstate)
 975 {
 976         struct mesh_area* mesh = qstate->env->mesh;
 977         struct mesh_state_ref* ref, lookup;
 978 #ifdef UNBOUND_DEBUG
 979         struct rbnode_type* n;
 980 #endif
 981         lookup.node.key = &lookup;
 982         lookup.s = qstate->mesh_info;
 983         RBTREE_FOR(ref, struct mesh_state_ref*, &qstate->mesh_info->sub_set) {
 984 #ifdef UNBOUND_DEBUG
 985                 n =
 986 #else
 987                 (void)
 988 #endif
 989                 rbtree_delete(&ref->s->super_set, &lookup);
 990                 log_assert(n != NULL); /* must have been present */
 991                 if(!ref->s->reply_list && !ref->s->cb_list
 992                         && ref->s->super_set.count == 0) {
 993                         mesh->num_detached_states++;
 994                         log_assert(mesh->num_detached_states +
 995                                 mesh->num_reply_states <= mesh->all.count);
 996                 }
 997         }
 998         rbtree_init(&qstate->mesh_info->sub_set, &mesh_state_ref_compare);
 999 }
1000
1001 int mesh_add_sub(struct module_qstate* qstate, struct query_info* qinfo,
1002         uint16_t qflags, int prime, int valrec, struct module_qstate** newq,
1003         struct mesh_state** sub)
1004 {
1005         /* find it, if not, create it */
1006         struct mesh_area* mesh = qstate->env->mesh;
1007         *sub = mesh_area_find(mesh, NULL, qinfo, qflags,
1008                 prime, valrec);
1009         if(mesh_detect_cycle_found(qstate, *sub)) {
1010                 verbose(VERB_ALGO, "attach failed, cycle detected");
1011                 return 0;
1012         }
1013         if(!*sub) {
1014 #ifdef UNBOUND_DEBUG
1015                 struct rbnode_type* n;
1016 #endif
1017                 /* create a new one */
1018                 *sub = mesh_state_create(qstate->env, qinfo, NULL, qflags, prime,
1019                         valrec);
1020                 if(!*sub) {
1021                         log_err("mesh_attach_sub: out of memory");
1022                         return 0;
1023                 }
1024 #ifdef UNBOUND_DEBUG
1025                 n =
1026 #else
1027                 (void)
1028 #endif
1029                 rbtree_insert(&mesh->all, &(*sub)->node);
1030                 log_assert(n != NULL);
1031                 /* set detached (it is now) */
1032                 mesh->num_detached_states++;
1033                 /* set new query state to run */
1034 #ifdef UNBOUND_DEBUG
1035                 n =
1036 #else
1037                 (void)
1038 #endif
1039                 rbtree_insert(&mesh->run, &(*sub)->run_node);
1040                 log_assert(n != NULL);
1041                 *newq = &(*sub)->s;
1042         } else
1043                 *newq = NULL;
1044         return 1;
1045 }
1046
1047 int mesh_attach_sub(struct module_qstate* qstate, struct query_info* qinfo,
1048         uint16_t qflags, int prime, int valrec, struct module_qstate** newq)
1049 {
1050         struct mesh_area* mesh = qstate->env->mesh;
1051         struct mesh_state* sub = NULL;
1052         int was_detached;
1053         if(!mesh_add_sub(qstate, qinfo, qflags, prime, valrec, newq, &sub))
1054                 return 0;
1055         was_detached = (sub->super_set.count == 0);
1056         if(!mesh_state_attachment(qstate->mesh_info, sub))
1057                 return 0;
1058         /* if it was a duplicate  attachment, the count was not zero before */
1059         if(!sub->reply_list && !sub->cb_list && was_detached &&
1060                 sub->super_set.count == 1) {
1061                 /* it used to be detached, before this one got added */
1062                 log_assert(mesh->num_detached_states > 0);
1063                 mesh->num_detached_states--;
1064         }
1065         /* *newq will be run when inited after the current module stops */
1066         return 1;
1067 }
1068
1069 int mesh_state_attachment(struct mesh_state* super, struct mesh_state* sub)
1070 {
1071 #ifdef UNBOUND_DEBUG
1072         struct rbnode_type* n;
1073 #endif
1074         struct mesh_state_ref* subref; /* points to sub, inserted in super */
1075         struct mesh_state_ref* superref; /* points to super, inserted in sub */
1076         if( !(subref = regional_alloc(super->s.region,
1077                 sizeof(struct mesh_state_ref))) ||
1078                 !(superref = regional_alloc(sub->s.region,
1079                 sizeof(struct mesh_state_ref))) ) {
1080                 log_err("mesh_state_attachment: out of memory");
1081                 return 0;
1082         }
1083         superref->node.key = superref;
1084         superref->s = super;
1085         subref->node.key = subref;
1086         subref->s = sub;
1087         if(!rbtree_insert(&sub->super_set, &superref->node)) {
1088                 /* this should not happen, iterator and validator do not
1089                  * attach subqueries that are identical. */
1090                 /* already attached, we are done, nothing todo.
1091                  * since superref and subref already allocated in region,
1092                  * we cannot free them */
1093                 return 1;
1094         }
1095 #ifdef UNBOUND_DEBUG
1096         n =
1097 #else
1098         (void)
1099 #endif
1100         rbtree_insert(&super->sub_set, &subref->node);
1101         log_assert(n != NULL); /* we checked above if statement, the reverse
1102           administration should not fail now, unless they are out of sync */
1103         return 1;
1104 }
1105
1106 /**
1107  * callback results to mesh cb entry
1108  * @param m: mesh state to send it for.
1109  * @param rcode: if not 0, error code.
1110  * @param rep: reply to send (or NULL if rcode is set).
1111  * @param r: callback entry
1112  */
1113 static void
1114 mesh_do_callback(struct mesh_state* m, int rcode, struct reply_info* rep,
1115         struct mesh_cb* r)
1116 {
1117         int secure;
1118         char* reason = NULL;
1119         int was_ratelimited = m->s.was_ratelimited;
1120         /* bogus messages are not made into servfail, sec_status passed
1121          * to the callback function */
1122         if(rep && rep->security == sec_status_secure)
1123                 secure = 1;
1124         else    secure = 0;
1125         if(!rep && rcode == LDNS_RCODE_NOERROR)
1126                 rcode = LDNS_RCODE_SERVFAIL;
1127         if(!rcode && (rep->security == sec_status_bogus ||
1128                 rep->security == sec_status_secure_sentinel_fail)) {
1129                 if(!(reason = errinf_to_str_bogus(&m->s)))
1130                         rcode = LDNS_RCODE_SERVFAIL;
1131         }
1132         /* send the reply */
1133         if(rcode) {
1134                 if(rcode == LDNS_RCODE_SERVFAIL) {
1135                         if(!inplace_cb_reply_servfail_call(m->s.env, &m->s.qinfo, &m->s,
1136                                 rep, rcode, &r->edns, NULL, m->s.region))
1137                                         r->edns.opt_list = NULL;
1138                 } else {
1139                         if(!inplace_cb_reply_call(m->s.env, &m->s.qinfo, &m->s, rep, rcode,
1140                                 &r->edns, NULL, m->s.region))
1141                                         r->edns.opt_list = NULL;
1142                 }
1143                 fptr_ok(fptr_whitelist_mesh_cb(r->cb));
1144                 (*r->cb)(r->cb_arg, rcode, r->buf, sec_status_unchecked, NULL,
1145                         was_ratelimited);
1146         } else {
1147                 size_t udp_size = r->edns.udp_size;
1148                 sldns_buffer_clear(r->buf);
1149                 r->edns.edns_version = EDNS_ADVERTISED_VERSION;
1150                 r->edns.udp_size = EDNS_ADVERTISED_SIZE;
1151                 r->edns.ext_rcode = 0;
1152                 r->edns.bits &= EDNS_DO;
1153
1154                 if(!inplace_cb_reply_call(m->s.env, &m->s.qinfo, &m->s, rep,
1155                         LDNS_RCODE_NOERROR, &r->edns, NULL, m->s.region) ||
1156                         !reply_info_answer_encode(&m->s.qinfo, rep, r->qid,
1157                         r->qflags, r->buf, 0, 1,
1158                         m->s.env->scratch, udp_size, &r->edns,
1159                         (int)(r->edns.bits & EDNS_DO), secure))
1160                 {
1161                         fptr_ok(fptr_whitelist_mesh_cb(r->cb));
1162                         (*r->cb)(r->cb_arg, LDNS_RCODE_SERVFAIL, r->buf,
1163                                 sec_status_unchecked, NULL, 0);
1164                 } else {
1165                         fptr_ok(fptr_whitelist_mesh_cb(r->cb));
1166                         (*r->cb)(r->cb_arg, LDNS_RCODE_NOERROR, r->buf,
1167                                 rep->security, reason, was_ratelimited);
1168                 }
1169         }
1170         free(reason);
1171         log_assert(m->s.env->mesh->num_reply_addrs > 0);
1172         m->s.env->mesh->num_reply_addrs--;
1173 }
1174
1175 /**
1176  * Send reply to mesh reply entry
1177  * @param m: mesh state to send it for.
1178  * @param rcode: if not 0, error code.
1179  * @param rep: reply to send (or NULL if rcode is set).
1180  * @param r: reply entry
1181  * @param r_buffer: buffer to use for reply entry.
1182  * @param prev: previous reply, already has its answer encoded in buffer.
1183  * @param prev_buffer: buffer for previous reply.
1184  */
1185 static void
1186 mesh_send_reply(struct mesh_state* m, int rcode, struct reply_info* rep,
1187         struct mesh_reply* r, struct sldns_buffer* r_buffer,
1188         struct mesh_reply* prev, struct sldns_buffer* prev_buffer)
1189 {
1190         struct timeval end_time;
1191         struct timeval duration;
1192         int secure;
1193         /* Copy the client's EDNS for later restore, to make sure the edns
1194          * compare is with the correct edns options. */
1195         struct edns_data edns_bak = r->edns;
1196         /* examine security status */
1197         if(m->s.env->need_to_validate && (!(r->qflags&BIT_CD) ||
1198                 m->s.env->cfg->ignore_cd) && rep &&
1199                 (rep->security <= sec_status_bogus ||
1200                 rep->security == sec_status_secure_sentinel_fail)) {
1201                 rcode = LDNS_RCODE_SERVFAIL;
1202                 if(m->s.env->cfg->stat_extended)
1203                         m->s.env->mesh->ans_bogus++;
1204         }
1205         if(rep && rep->security == sec_status_secure)
1206                 secure = 1;
1207         else    secure = 0;
1208         if(!rep && rcode == LDNS_RCODE_NOERROR)
1209                 rcode = LDNS_RCODE_SERVFAIL;
1210         /* send the reply */
1211         /* We don't reuse the encoded answer if either the previous or current
1212          * response has a local alias.  We could compare the alias records
1213          * and still reuse the previous answer if they are the same, but that
1214          * would be complicated and error prone for the relatively minor case.
1215          * So we err on the side of safety. */
1216         if(prev && prev_buffer && prev->qflags == r->qflags &&
1217                 !prev->local_alias && !r->local_alias &&
1218                 prev->edns.edns_present == r->edns.edns_present &&
1219                 prev->edns.bits == r->edns.bits &&
1220                 prev->edns.udp_size == r->edns.udp_size &&
1221                 edns_opt_list_compare(prev->edns.opt_list, r->edns.opt_list)
1222                 == 0) {
1223                 /* if the previous reply is identical to this one, fix ID */
1224                 if(prev_buffer != r_buffer)
1225                         sldns_buffer_copy(r_buffer, prev_buffer);
1226                 sldns_buffer_write_at(r_buffer, 0, &r->qid, sizeof(uint16_t));
1227                 sldns_buffer_write_at(r_buffer, 12, r->qname,
1228                         m->s.qinfo.qname_len);
1229                 comm_point_send_reply(&r->query_reply);
1230         } else if(rcode) {
1231                 m->s.qinfo.qname = r->qname;
1232                 m->s.qinfo.local_alias = r->local_alias;
1233                 if(rcode == LDNS_RCODE_SERVFAIL) {
1234                         if(!inplace_cb_reply_servfail_call(m->s.env, &m->s.qinfo, &m->s,
1235                                 rep, rcode, &r->edns, NULL, m->s.region))
1236                                         r->edns.opt_list = NULL;
1237                 } else {
1238                         if(!inplace_cb_reply_call(m->s.env, &m->s.qinfo, &m->s, rep, rcode,
1239                                 &r->edns, NULL, m->s.region))
1240                                         r->edns.opt_list = NULL;
1241                 }
1242                 error_encode(r_buffer, rcode, &m->s.qinfo, r->qid,
1243                         r->qflags, &r->edns);
1244                 comm_point_send_reply(&r->query_reply);
1245         } else {
1246                 size_t udp_size = r->edns.udp_size;
1247                 r->edns.edns_version = EDNS_ADVERTISED_VERSION;
1248                 r->edns.udp_size = EDNS_ADVERTISED_SIZE;
1249                 r->edns.ext_rcode = 0;
1250                 r->edns.bits &= EDNS_DO;
1251                 m->s.qinfo.qname = r->qname;
1252                 m->s.qinfo.local_alias = r->local_alias;
1253                 if(!inplace_cb_reply_call(m->s.env, &m->s.qinfo, &m->s, rep,
1254                         LDNS_RCODE_NOERROR, &r->edns, NULL, m->s.region) ||
1255                         !apply_edns_options(&r->edns, &edns_bak,
1256                                 m->s.env->cfg, r->query_reply.c,
1257                                 m->s.region) ||
1258                         !reply_info_answer_encode(&m->s.qinfo, rep, r->qid,
1259                         r->qflags, r_buffer, 0, 1, m->s.env->scratch,
1260                         udp_size, &r->edns, (int)(r->edns.bits & EDNS_DO),
1261                         secure))
1262                 {
1263                         if(!inplace_cb_reply_servfail_call(m->s.env, &m->s.qinfo, &m->s,
1264                         rep, LDNS_RCODE_SERVFAIL, &r->edns, NULL, m->s.region))
1265                                 r->edns.opt_list = NULL;
1266                         error_encode(r_buffer, LDNS_RCODE_SERVFAIL,
1267                                 &m->s.qinfo, r->qid, r->qflags, &r->edns);
1268                 }
1269                 r->edns = edns_bak;
1270                 comm_point_send_reply(&r->query_reply);
1271         }
1272         /* account */
1273         log_assert(m->s.env->mesh->num_reply_addrs > 0);
1274         m->s.env->mesh->num_reply_addrs--;
1275         end_time = *m->s.env->now_tv;
1276         timeval_subtract(&duration, &end_time, &r->start_time);
1277         verbose(VERB_ALGO, "query took " ARG_LL "d.%6.6d sec",
1278                 (long long)duration.tv_sec, (int)duration.tv_usec);
1279         m->s.env->mesh->replies_sent++;
1280         timeval_add(&m->s.env->mesh->replies_sum_wait, &duration);
1281         timehist_insert(m->s.env->mesh->histogram, &duration);
1282         if(m->s.env->cfg->stat_extended) {
1283                 uint16_t rc = FLAGS_GET_RCODE(sldns_buffer_read_u16_at(
1284                         r_buffer, 2));
1285                 if(secure) m->s.env->mesh->ans_secure++;
1286                 m->s.env->mesh->ans_rcode[ rc ] ++;
1287                 if(rc == 0 && LDNS_ANCOUNT(sldns_buffer_begin(r_buffer)) == 0)
1288                         m->s.env->mesh->ans_nodata++;
1289         }
1290         /* Log reply sent */
1291         if(m->s.env->cfg->log_replies) {
1292                 log_reply_info(NO_VERBOSE, &m->s.qinfo, &r->query_reply.addr,
1293                         r->query_reply.addrlen, duration, 0, r_buffer);
1294         }
1295 }
1296
1297 void mesh_query_done(struct mesh_state* mstate)
1298 {
1299         struct mesh_reply* r;
1300         struct mesh_reply* prev = NULL;
1301         struct sldns_buffer* prev_buffer = NULL;
1302         struct mesh_cb* c;
1303         struct reply_info* rep = (mstate->s.return_msg?
1304                 mstate->s.return_msg->rep:NULL);
1305         /* No need for the serve expired timer anymore; we are going to reply. */
1306         if(mstate->s.serve_expired_data) {
1307                 comm_timer_delete(mstate->s.serve_expired_data->timer);
1308                 mstate->s.serve_expired_data->timer = NULL;
1309         }
1310         if(mstate->s.return_rcode == LDNS_RCODE_SERVFAIL ||
1311                 (rep && FLAGS_GET_RCODE(rep->flags) == LDNS_RCODE_SERVFAIL)) {
1312                 /* we are SERVFAILing; check for expired asnwer here */
1313                 mesh_serve_expired_callback(mstate);
1314                 if((mstate->reply_list || mstate->cb_list)
1315                 && mstate->s.env->cfg->log_servfail
1316                 && !mstate->s.env->cfg->val_log_squelch) {
1317                         char* err = errinf_to_str_servfail(&mstate->s);
1318                         if(err)
1319                                 log_err("%s", err);
1320                         free(err);
1321                 }
1322         }
1323         for(r = mstate->reply_list; r; r = r->next) {
1324                 /* if a response-ip address block has been stored the
1325                  *  information should be logged for each client. */
1326                 if(mstate->s.respip_action_info &&
1327                         mstate->s.respip_action_info->addrinfo) {
1328                         respip_inform_print(mstate->s.respip_action_info,
1329                                 r->qname, mstate->s.qinfo.qtype,
1330                                 mstate->s.qinfo.qclass, r->local_alias,
1331                                 &r->query_reply);
1332                         if(mstate->s.env->cfg->stat_extended &&
1333                                 mstate->s.respip_action_info->rpz_used) {
1334                                 if(mstate->s.respip_action_info->rpz_disabled)
1335                                         mstate->s.env->mesh->rpz_action[RPZ_DISABLED_ACTION]++;
1336                                 if(mstate->s.respip_action_info->rpz_cname_override)
1337                                         mstate->s.env->mesh->rpz_action[RPZ_CNAME_OVERRIDE_ACTION]++;
1338                                 else
1339                                         mstate->s.env->mesh->rpz_action[respip_action_to_rpz_action(
1340                                                 mstate->s.respip_action_info->action)]++;
1341                         }
1342                 }
1343
1344                 /* if this query is determined to be dropped during the
1345                  * mesh processing, this is the point to take that action. */
1346                 if(mstate->s.is_drop) {
1347                         /* briefly set the reply_list to NULL, so that the
1348                          * tcp req info cleanup routine that calls the mesh
1349                          * to deregister the meshstate for it is not done
1350                          * because the list is NULL and also accounting is not
1351                          * done there, but instead we do that here. */
1352                         struct mesh_reply* reply_list = mstate->reply_list;
1353                         mstate->reply_list = NULL;
1354                         comm_point_drop_reply(&r->query_reply);
1355                         mstate->reply_list = reply_list;
1356                 } else {
1357                         struct sldns_buffer* r_buffer = r->query_reply.c->buffer;
1358                         struct mesh_reply* rlist = mstate->reply_list;
1359                         if(r->query_reply.c->tcp_req_info) {
1360                                 r_buffer = r->query_reply.c->tcp_req_info->spool_buffer;
1361                                 prev_buffer = NULL;
1362                         }
1363                         /* briefly set the replylist to null in case the
1364                          * meshsendreply calls tcpreqinfo sendreply that
1365                          * comm_point_drops because of size, and then the
1366                          * null stops the mesh state remove and thus
1367                          * reply_list modification and accounting */
1368                         mstate->reply_list = NULL;
1369                         mesh_send_reply(mstate, mstate->s.return_rcode, rep,
1370                                 r, r_buffer, prev, prev_buffer);
1371                         mstate->reply_list = rlist;
1372                         if(r->query_reply.c->tcp_req_info) {
1373                                 tcp_req_info_remove_mesh_state(r->query_reply.c->tcp_req_info, mstate);
1374                                 r_buffer = NULL;
1375                         }
1376                         prev = r;
1377                         prev_buffer = r_buffer;
1378                 }
1379         }
1380         if(mstate->reply_list) {
1381                 mstate->reply_list = NULL;
1382                 if(!mstate->reply_list && !mstate->cb_list) {
1383                         /* was a reply state, not anymore */
1384                         log_assert(mstate->s.env->mesh->num_reply_states > 0);
1385                         mstate->s.env->mesh->num_reply_states--;
1386                 }
1387                 if(!mstate->reply_list && !mstate->cb_list &&
1388                         mstate->super_set.count == 0)
1389                         mstate->s.env->mesh->num_detached_states++;
1390         }
1391         mstate->replies_sent = 1;
1392         while((c = mstate->cb_list) != NULL) {
1393                 /* take this cb off the list; so that the list can be
1394                  * changed, eg. by adds from the callback routine */
1395                 if(!mstate->reply_list && mstate->cb_list && !c->next) {
1396                         /* was a reply state, not anymore */
1397                         log_assert(mstate->s.env->mesh->num_reply_states > 0);
1398                         mstate->s.env->mesh->num_reply_states--;
1399                 }
1400                 mstate->cb_list = c->next;
1401                 if(!mstate->reply_list && !mstate->cb_list &&
1402                         mstate->super_set.count == 0)
1403                         mstate->s.env->mesh->num_detached_states++;
1404                 mesh_do_callback(mstate, mstate->s.return_rcode, rep, c);
1405         }
1406 }
1407
1408 void mesh_walk_supers(struct mesh_area* mesh, struct mesh_state* mstate)
1409 {
1410         struct mesh_state_ref* ref;
1411         RBTREE_FOR(ref, struct mesh_state_ref*, &mstate->super_set)
1412         {
1413                 /* make super runnable */
1414                 (void)rbtree_insert(&mesh->run, &ref->s->run_node);
1415                 /* callback the function to inform super of result */
1416                 fptr_ok(fptr_whitelist_mod_inform_super(
1417                         mesh->mods.mod[ref->s->s.curmod]->inform_super));
1418                 (*mesh->mods.mod[ref->s->s.curmod]->inform_super)(&mstate->s,
1419                         ref->s->s.curmod, &ref->s->s);
1420                 /* copy state that is always relevant to super */
1421                 copy_state_to_super(&mstate->s, ref->s->s.curmod, &ref->s->s);
1422         }
1423 }
1424
1425 struct mesh_state* mesh_area_find(struct mesh_area* mesh,
1426         struct respip_client_info* cinfo, struct query_info* qinfo,
1427         uint16_t qflags, int prime, int valrec)
1428 {
1429         struct mesh_state key;
1430         struct mesh_state* result;
1431
1432         key.node.key = &key;
1433         key.s.is_priming = prime;
1434         key.s.is_valrec = valrec;
1435         key.s.qinfo = *qinfo;
1436         key.s.query_flags = qflags;
1437         /* We are searching for a similar mesh state when we DO want to
1438          * aggregate the state. Thus unique is set to NULL. (default when we
1439          * desire aggregation).*/
1440         key.unique = NULL;
1441         key.s.client_info = cinfo;
1442
1443         result = (struct mesh_state*)rbtree_search(&mesh->all, &key);
1444         return result;
1445 }
1446
1447 int mesh_state_add_cb(struct mesh_state* s, struct edns_data* edns,
1448         sldns_buffer* buf, mesh_cb_func_type cb, void* cb_arg,
1449         uint16_t qid, uint16_t qflags)
1450 {
1451         struct mesh_cb* r = regional_alloc(s->s.region,
1452                 sizeof(struct mesh_cb));
1453         if(!r)
1454                 return 0;
1455         r->buf = buf;
1456         log_assert(fptr_whitelist_mesh_cb(cb)); /* early failure ifmissing*/
1457         r->cb = cb;
1458         r->cb_arg = cb_arg;
1459         r->edns = *edns;
1460         if(edns->opt_list) {
1461                 r->edns.opt_list = edns_opt_copy_region(edns->opt_list,
1462                         s->s.region);
1463                 if(!r->edns.opt_list)
1464                         return 0;
1465         }
1466         r->qid = qid;
1467         r->qflags = qflags;
1468         r->next = s->cb_list;
1469         s->cb_list = r;
1470         return 1;
1471
1472 }
1473
1474 int mesh_state_add_reply(struct mesh_state* s, struct edns_data* edns,
1475         struct comm_reply* rep, uint16_t qid, uint16_t qflags,
1476         const struct query_info* qinfo)
1477 {
1478         struct mesh_reply* r = regional_alloc(s->s.region,
1479                 sizeof(struct mesh_reply));
1480         if(!r)
1481                 return 0;
1482         r->query_reply = *rep;
1483         r->edns = *edns;
1484         if(edns->opt_list) {
1485                 r->edns.opt_list = edns_opt_copy_region(edns->opt_list,
1486                         s->s.region);
1487                 if(!r->edns.opt_list)
1488                         return 0;
1489         }
1490         r->qid = qid;
1491         r->qflags = qflags;
1492         r->start_time = *s->s.env->now_tv;
1493         r->next = s->reply_list;
1494         r->qname = regional_alloc_init(s->s.region, qinfo->qname,
1495                 s->s.qinfo.qname_len);
1496         if(!r->qname)
1497                 return 0;
1498
1499         /* Data related to local alias stored in 'qinfo' (if any) is ephemeral
1500          * and can be different for different original queries (even if the
1501          * replaced query name is the same).  So we need to make a deep copy
1502          * and store the copy for each reply info. */
1503         if(qinfo->local_alias) {
1504                 struct packed_rrset_data* d;
1505                 struct packed_rrset_data* dsrc;
1506                 r->local_alias = regional_alloc_zero(s->s.region,
1507                         sizeof(*qinfo->local_alias));
1508                 if(!r->local_alias)
1509                         return 0;
1510                 r->local_alias->rrset = regional_alloc_init(s->s.region,
1511                         qinfo->local_alias->rrset,
1512                         sizeof(*qinfo->local_alias->rrset));
1513                 if(!r->local_alias->rrset)
1514                         return 0;
1515                 dsrc = qinfo->local_alias->rrset->entry.data;
1516
1517                 /* In the current implementation, a local alias must be
1518                  * a single CNAME RR (see worker_handle_request()). */
1519                 log_assert(!qinfo->local_alias->next && dsrc->count == 1 &&
1520                         qinfo->local_alias->rrset->rk.type ==
1521                         htons(LDNS_RR_TYPE_CNAME));
1522                 /* we should make a local copy for the owner name of
1523                  * the RRset */
1524                 r->local_alias->rrset->rk.dname_len =
1525                         qinfo->local_alias->rrset->rk.dname_len;
1526                 r->local_alias->rrset->rk.dname = regional_alloc_init(
1527                         s->s.region, qinfo->local_alias->rrset->rk.dname,
1528                         qinfo->local_alias->rrset->rk.dname_len);
1529                 if(!r->local_alias->rrset->rk.dname)
1530                         return 0;
1531
1532                 /* the rrset is not packed, like in the cache, but it is
1533                  * individualy allocated with an allocator from localzone. */
1534                 d = regional_alloc_zero(s->s.region, sizeof(*d));
1535                 if(!d)
1536                         return 0;
1537                 r->local_alias->rrset->entry.data = d;
1538                 if(!rrset_insert_rr(s->s.region, d, dsrc->rr_data[0],
1539                         dsrc->rr_len[0], dsrc->rr_ttl[0], "CNAME local alias"))
1540                         return 0;
1541         } else
1542                 r->local_alias = NULL;
1543
1544         s->reply_list = r;
1545         return 1;
1546 }
1547
1548 /* Extract the query info and flags from 'mstate' into '*qinfop' and '*qflags'.
1549  * Since this is only used for internal refetch of otherwise-expired answer,
1550  * we simply ignore the rare failure mode when memory allocation fails. */
1551 static void
1552 mesh_copy_qinfo(struct mesh_state* mstate, struct query_info** qinfop,
1553         uint16_t* qflags)
1554 {
1555         struct regional* region = mstate->s.env->scratch;
1556         struct query_info* qinfo;
1557
1558         qinfo = regional_alloc_init(region, &mstate->s.qinfo, sizeof(*qinfo));
1559         if(!qinfo)
1560                 return;
1561         qinfo->qname = regional_alloc_init(region, qinfo->qname,
1562                 qinfo->qname_len);
1563         if(!qinfo->qname)
1564                 return;
1565         *qinfop = qinfo;
1566         *qflags = mstate->s.query_flags;
1567 }
1568
1569 /**
1570  * Continue processing the mesh state at another module.
1571  * Handles module to modules transfer of control.
1572  * Handles module finished.
1573  * @param mesh: the mesh area.
1574  * @param mstate: currently active mesh state.
1575  *      Deleted if finished, calls _done and _supers to
1576  *      send replies to clients and inform other mesh states.
1577  *      This in turn may create additional runnable mesh states.
1578  * @param s: state at which the current module exited.
1579  * @param ev: the event sent to the module.
1580  *      returned is the event to send to the next module.
1581  * @return true if continue processing at the new module.
1582  *      false if not continued processing is needed.
1583  */
1584 static int
1585 mesh_continue(struct mesh_area* mesh, struct mesh_state* mstate,
1586         enum module_ext_state s, enum module_ev* ev)
1587 {
1588         mstate->num_activated++;
1589         if(mstate->num_activated > MESH_MAX_ACTIVATION) {
1590                 /* module is looping. Stop it. */
1591                 log_err("internal error: looping module (%s) stopped",
1592                         mesh->mods.mod[mstate->s.curmod]->name);
1593                 log_query_info(NO_VERBOSE, "pass error for qstate",
1594                         &mstate->s.qinfo);
1595                 s = module_error;
1596         }
1597         if(s == module_wait_module || s == module_restart_next) {
1598                 /* start next module */
1599                 mstate->s.curmod++;
1600                 if(mesh->mods.num == mstate->s.curmod) {
1601                         log_err("Cannot pass to next module; at last module");
1602                         log_query_info(VERB_QUERY, "pass error for qstate",
1603                                 &mstate->s.qinfo);
1604                         mstate->s.curmod--;
1605                         return mesh_continue(mesh, mstate, module_error, ev);
1606                 }
1607                 if(s == module_restart_next) {
1608                         int curmod = mstate->s.curmod;
1609                         for(; mstate->s.curmod < mesh->mods.num;
1610                                 mstate->s.curmod++) {
1611                                 fptr_ok(fptr_whitelist_mod_clear(
1612                                         mesh->mods.mod[mstate->s.curmod]->clear));
1613                                 (*mesh->mods.mod[mstate->s.curmod]->clear)
1614                                         (&mstate->s, mstate->s.curmod);
1615                                 mstate->s.minfo[mstate->s.curmod] = NULL;
1616                         }
1617                         mstate->s.curmod = curmod;
1618                 }
1619                 *ev = module_event_pass;
1620                 return 1;
1621         }
1622         if(s == module_wait_subquery && mstate->sub_set.count == 0) {
1623                 log_err("module cannot wait for subquery, subquery list empty");
1624                 log_query_info(VERB_QUERY, "pass error for qstate",
1625                         &mstate->s.qinfo);
1626                 s = module_error;
1627         }
1628         if(s == module_error && mstate->s.return_rcode == LDNS_RCODE_NOERROR) {
1629                 /* error is bad, handle pass back up below */
1630                 mstate->s.return_rcode = LDNS_RCODE_SERVFAIL;
1631         }
1632         if(s == module_error) {
1633                 mesh_query_done(mstate);
1634                 mesh_walk_supers(mesh, mstate);
1635                 mesh_state_delete(&mstate->s);
1636                 return 0;
1637         }
1638         if(s == module_finished) {
1639                 if(mstate->s.curmod == 0) {
1640                         struct query_info* qinfo = NULL;
1641                         uint16_t qflags;
1642
1643                         mesh_query_done(mstate);
1644                         mesh_walk_supers(mesh, mstate);
1645
1646                         /* If the answer to the query needs to be refetched
1647                          * from an external DNS server, we'll need to schedule
1648                          * a prefetch after removing the current state, so
1649                          * we need to make a copy of the query info here. */
1650                         if(mstate->s.need_refetch)
1651                                 mesh_copy_qinfo(mstate, &qinfo, &qflags);
1652
1653                         mesh_state_delete(&mstate->s);
1654                         if(qinfo) {
1655                                 mesh_schedule_prefetch(mesh, qinfo, qflags,
1656                                         0, 1);
1657                         }
1658                         return 0;
1659                 }
1660                 /* pass along the locus of control */
1661                 mstate->s.curmod --;
1662                 *ev = module_event_moddone;
1663                 return 1;
1664         }
1665         return 0;
1666 }
1667
1668 void mesh_run(struct mesh_area* mesh, struct mesh_state* mstate,
1669         enum module_ev ev, struct outbound_entry* e)
1670 {
1671         enum module_ext_state s;
1672         verbose(VERB_ALGO, "mesh_run: start");
1673         while(mstate) {
1674                 /* run the module */
1675                 fptr_ok(fptr_whitelist_mod_operate(
1676                         mesh->mods.mod[mstate->s.curmod]->operate));
1677                 (*mesh->mods.mod[mstate->s.curmod]->operate)
1678                         (&mstate->s, ev, mstate->s.curmod, e);
1679
1680                 /* examine results */
1681                 mstate->s.reply = NULL;
1682                 regional_free_all(mstate->s.env->scratch);
1683                 s = mstate->s.ext_state[mstate->s.curmod];
1684                 verbose(VERB_ALGO, "mesh_run: %s module exit state is %s",
1685                         mesh->mods.mod[mstate->s.curmod]->name, strextstate(s));
1686                 e = NULL;
1687                 if(mesh_continue(mesh, mstate, s, &ev))
1688                         continue;
1689
1690                 /* run more modules */
1691                 ev = module_event_pass;
1692                 if(mesh->run.count > 0) {
1693                         /* pop random element off the runnable tree */
1694                         mstate = (struct mesh_state*)mesh->run.root->key;
1695                         (void)rbtree_delete(&mesh->run, mstate);
1696                 } else mstate = NULL;
1697         }
1698         if(verbosity >= VERB_ALGO) {
1699                 mesh_stats(mesh, "mesh_run: end");
1700                 mesh_log_list(mesh);
1701         }
1702 }
1703
1704 void
1705 mesh_log_list(struct mesh_area* mesh)
1706 {
1707         char buf[30];
1708         struct mesh_state* m;
1709         int num = 0;
1710         RBTREE_FOR(m, struct mesh_state*, &mesh->all) {
1711                 snprintf(buf, sizeof(buf), "%d%s%s%s%s%s%s mod%d %s%s",
1712                         num++, (m->s.is_priming)?"p":"",  /* prime */
1713                         (m->s.is_valrec)?"v":"",  /* prime */
1714                         (m->s.query_flags&BIT_RD)?"RD":"",
1715                         (m->s.query_flags&BIT_CD)?"CD":"",
1716                         (m->super_set.count==0)?"d":"", /* detached */
1717                         (m->sub_set.count!=0)?"c":"",  /* children */
1718                         m->s.curmod, (m->reply_list)?"rep":"", /*hasreply*/
1719                         (m->cb_list)?"cb":"" /* callbacks */
1720                         );
1721                 log_query_info(VERB_ALGO, buf, &m->s.qinfo);
1722         }
1723 }
1724
1725 void
1726 mesh_stats(struct mesh_area* mesh, const char* str)
1727 {
1728         verbose(VERB_DETAIL, "%s %u recursion states (%u with reply, "
1729                 "%u detached), %u waiting replies, %u recursion replies "
1730                 "sent, %d replies dropped, %d states jostled out",
1731                 str, (unsigned)mesh->all.count,
1732                 (unsigned)mesh->num_reply_states,
1733                 (unsigned)mesh->num_detached_states,
1734                 (unsigned)mesh->num_reply_addrs,
1735                 (unsigned)mesh->replies_sent,
1736                 (unsigned)mesh->stats_dropped,
1737                 (unsigned)mesh->stats_jostled);
1738         if(mesh->replies_sent > 0) {
1739                 struct timeval avg;
1740                 timeval_divide(&avg, &mesh->replies_sum_wait,
1741                         mesh->replies_sent);
1742                 log_info("average recursion processing time "
1743                         ARG_LL "d.%6.6d sec",
1744                         (long long)avg.tv_sec, (int)avg.tv_usec);
1745                 log_info("histogram of recursion processing times");
1746                 timehist_log(mesh->histogram, "recursions");
1747         }
1748 }
1749
1750 void
1751 mesh_stats_clear(struct mesh_area* mesh)
1752 {
1753         if(!mesh)
1754                 return;
1755         mesh->replies_sent = 0;
1756         mesh->replies_sum_wait.tv_sec = 0;
1757         mesh->replies_sum_wait.tv_usec = 0;
1758         mesh->stats_jostled = 0;
1759         mesh->stats_dropped = 0;
1760         timehist_clear(mesh->histogram);
1761         mesh->ans_secure = 0;
1762         mesh->ans_bogus = 0;
1763         mesh->ans_expired = 0;
1764         memset(&mesh->ans_rcode[0], 0, sizeof(size_t)*UB_STATS_RCODE_NUM);
1765         memset(&mesh->rpz_action[0], 0, sizeof(size_t)*UB_STATS_RPZ_ACTION_NUM);
1766         mesh->ans_nodata = 0;
1767 }
1768
1769 size_t
1770 mesh_get_mem(struct mesh_area* mesh)
1771 {
1772         struct mesh_state* m;
1773         size_t s = sizeof(*mesh) + sizeof(struct timehist) +
1774                 sizeof(struct th_buck)*mesh->histogram->num +
1775                 sizeof(sldns_buffer) + sldns_buffer_capacity(mesh->qbuf_bak);
1776         RBTREE_FOR(m, struct mesh_state*, &mesh->all) {
1777                 /* all, including m itself allocated in qstate region */
1778                 s += regional_get_mem(m->s.region);
1779         }
1780         return s;
1781 }
1782
1783 int
1784 mesh_detect_cycle(struct module_qstate* qstate, struct query_info* qinfo,
1785         uint16_t flags, int prime, int valrec)
1786 {
1787         struct mesh_area* mesh = qstate->env->mesh;
1788         struct mesh_state* dep_m = NULL;
1789         if(!mesh_state_is_unique(qstate->mesh_info))
1790                 dep_m = mesh_area_find(mesh, NULL, qinfo, flags, prime, valrec);
1791         return mesh_detect_cycle_found(qstate, dep_m);
1792 }
1793
1794 void mesh_list_insert(struct mesh_state* m, struct mesh_state** fp,
1795         struct mesh_state** lp)
1796 {
1797         /* insert as last element */
1798         m->prev = *lp;
1799         m->next = NULL;
1800         if(*lp)
1801                 (*lp)->next = m;
1802         else    *fp = m;
1803         *lp = m;
1804 }
1805
1806 void mesh_list_remove(struct mesh_state* m, struct mesh_state** fp,
1807         struct mesh_state** lp)
1808 {
1809         if(m->next)
1810                 m->next->prev = m->prev;
1811         else    *lp = m->prev;
1812         if(m->prev)
1813                 m->prev->next = m->next;
1814         else    *fp = m->next;
1815 }
1816
1817 void mesh_state_remove_reply(struct mesh_area* mesh, struct mesh_state* m,
1818         struct comm_point* cp)
1819 {
1820         struct mesh_reply* n, *prev = NULL;
1821         n = m->reply_list;
1822         /* when in mesh_cleanup, it sets the reply_list to NULL, so that
1823          * there is no accounting twice */
1824         if(!n) return; /* nothing to remove, also no accounting needed */
1825         while(n) {
1826                 if(n->query_reply.c == cp) {
1827                         /* unlink it */
1828                         if(prev) prev->next = n->next;
1829                         else m->reply_list = n->next;
1830                         /* delete it, but allocated in m region */
1831                         log_assert(mesh->num_reply_addrs > 0);
1832                         mesh->num_reply_addrs--;
1833
1834                         /* prev = prev; */
1835                         n = n->next;
1836                         continue;
1837                 }
1838                 prev = n;
1839                 n = n->next;
1840         }
1841         /* it was not detached (because it had a reply list), could be now */
1842         if(!m->reply_list && !m->cb_list
1843                 && m->super_set.count == 0) {
1844                 mesh->num_detached_states++;
1845         }
1846         /* if not replies any more in mstate, it is no longer a reply_state */
1847         if(!m->reply_list && !m->cb_list) {
1848                 log_assert(mesh->num_reply_states > 0);
1849                 mesh->num_reply_states--;
1850         }
1851 }
1852
1853
1854 static int
1855 apply_respip_action(struct module_qstate* qstate,
1856         const struct query_info* qinfo, struct respip_client_info* cinfo,
1857         struct respip_action_info* actinfo, struct reply_info* rep,
1858         struct ub_packed_rrset_key** alias_rrset,
1859         struct reply_info** encode_repp, struct auth_zones* az)
1860 {
1861         if(qinfo->qtype != LDNS_RR_TYPE_A &&
1862                 qinfo->qtype != LDNS_RR_TYPE_AAAA &&
1863                 qinfo->qtype != LDNS_RR_TYPE_ANY)
1864                 return 1;
1865
1866         if(!respip_rewrite_reply(qinfo, cinfo, rep, encode_repp, actinfo,
1867                 alias_rrset, 0, qstate->region, az))
1868                 return 0;
1869
1870         /* xxx_deny actions mean dropping the reply, unless the original reply
1871          * was redirected to response-ip data. */
1872         if((actinfo->action == respip_deny ||
1873                 actinfo->action == respip_inform_deny) &&
1874                 *encode_repp == rep)
1875                 *encode_repp = NULL;
1876
1877         return 1;
1878 }
1879
1880 void
1881 mesh_serve_expired_callback(void* arg)
1882 {
1883         struct mesh_state* mstate = (struct mesh_state*) arg;
1884         struct module_qstate* qstate = &mstate->s;
1885         struct mesh_reply* r, *rlist;
1886         struct mesh_area* mesh = qstate->env->mesh;
1887         struct dns_msg* msg;
1888         struct mesh_cb* c;
1889         struct mesh_reply* prev = NULL;
1890         struct sldns_buffer* prev_buffer = NULL;
1891         struct sldns_buffer* r_buffer = NULL;
1892         struct reply_info* partial_rep = NULL;
1893         struct ub_packed_rrset_key* alias_rrset = NULL;
1894         struct reply_info* encode_rep = NULL;
1895         struct respip_action_info actinfo;
1896         struct query_info* lookup_qinfo = &qstate->qinfo;
1897         struct query_info qinfo_tmp;
1898         int must_validate = (!(qstate->query_flags&BIT_CD)
1899                 || qstate->env->cfg->ignore_cd) && qstate->env->need_to_validate;
1900         if(!qstate->serve_expired_data) return;
1901         verbose(VERB_ALGO, "Serve expired: Trying to reply with expired data");
1902         comm_timer_delete(qstate->serve_expired_data->timer);
1903         qstate->serve_expired_data->timer = NULL;
1904         if(qstate->blacklist || qstate->no_cache_lookup || qstate->is_drop) {
1905                 verbose(VERB_ALGO,
1906                         "Serve expired: Not allowed to look into cache for stale");
1907                 return;
1908         }
1909         /* The following while is used instead of the `goto lookup_cache`
1910          * like in the worker. */
1911         while(1) {
1912                 fptr_ok(fptr_whitelist_serve_expired_lookup(
1913                         qstate->serve_expired_data->get_cached_answer));
1914                 msg = qstate->serve_expired_data->get_cached_answer(qstate,
1915                         lookup_qinfo);
1916                 if(!msg)
1917                         return;
1918                 /* Reset these in case we pass a second time from here. */
1919                 encode_rep = msg->rep;
1920                 memset(&actinfo, 0, sizeof(actinfo));
1921                 actinfo.action = respip_none;
1922                 alias_rrset = NULL;
1923                 if((mesh->use_response_ip || mesh->use_rpz) &&
1924                         !partial_rep && !apply_respip_action(qstate, &qstate->qinfo,
1925                         qstate->client_info, &actinfo, msg->rep, &alias_rrset, &encode_rep,
1926                         qstate->env->auth_zones)) {
1927                         return;
1928                 } else if(partial_rep &&
1929                         !respip_merge_cname(partial_rep, &qstate->qinfo, msg->rep,
1930                         qstate->client_info, must_validate, &encode_rep, qstate->region,
1931                         qstate->env->auth_zones)) {
1932                         return;
1933                 }
1934                 if(!encode_rep || alias_rrset) {
1935                         if(!encode_rep) {
1936                                 /* Needs drop */
1937                                 return;
1938                         } else {
1939                                 /* A partial CNAME chain is found. */
1940                                 partial_rep = encode_rep;
1941                         }
1942                 }
1943                 /* We've found a partial reply ending with an
1944                 * alias.  Replace the lookup qinfo for the
1945                 * alias target and lookup the cache again to
1946                 * (possibly) complete the reply.  As we're
1947                 * passing the "base" reply, there will be no
1948                 * more alias chasing. */
1949                 if(partial_rep) {
1950                         memset(&qinfo_tmp, 0, sizeof(qinfo_tmp));
1951                         get_cname_target(alias_rrset, &qinfo_tmp.qname,
1952                                 &qinfo_tmp.qname_len);
1953                         if(!qinfo_tmp.qname) {
1954                                 log_err("Serve expired: unexpected: invalid answer alias");
1955                                 return;
1956                         }
1957                         qinfo_tmp.qtype = qstate->qinfo.qtype;
1958                         qinfo_tmp.qclass = qstate->qinfo.qclass;
1959                         lookup_qinfo = &qinfo_tmp;
1960                         continue;
1961                 }
1962                 break;
1963         }
1964
1965         if(verbosity >= VERB_ALGO)
1966                 log_dns_msg("Serve expired lookup", &qstate->qinfo, msg->rep);
1967
1968         for(r = mstate->reply_list; r; r = r->next) {
1969                 /* If address info is returned, it means the action should be an
1970                 * 'inform' variant and the information should be logged. */
1971                 if(actinfo.addrinfo) {
1972                         respip_inform_print(&actinfo, r->qname,
1973                                 qstate->qinfo.qtype, qstate->qinfo.qclass,
1974                                 r->local_alias, &r->query_reply);
1975
1976                         if(qstate->env->cfg->stat_extended && actinfo.rpz_used) {
1977                                 if(actinfo.rpz_disabled)
1978                                         qstate->env->mesh->rpz_action[RPZ_DISABLED_ACTION]++;
1979                                 if(actinfo.rpz_cname_override)
1980                                         qstate->env->mesh->rpz_action[RPZ_CNAME_OVERRIDE_ACTION]++;
1981                                 else
1982                                         qstate->env->mesh->rpz_action[
1983                                                 respip_action_to_rpz_action(actinfo.action)]++;
1984                         }
1985                 }
1986
1987                 r_buffer = r->query_reply.c->buffer;
1988                 if(r->query_reply.c->tcp_req_info)
1989                         r_buffer = r->query_reply.c->tcp_req_info->spool_buffer;
1990                 /* briefly set the replylist to null in case the meshsendreply
1991                  * calls tcpreqinfo sendreply that comm_point_drops because
1992                  * of size, and then the null stops the mesh state remove and
1993                  * thus reply_list modification and accounting */
1994                 rlist = mstate->reply_list;
1995                 mstate->reply_list = NULL;
1996                 mesh_send_reply(mstate, LDNS_RCODE_NOERROR, msg->rep,
1997                         r, r_buffer, prev, prev_buffer);
1998                 mstate->reply_list = rlist;
1999                 if(r->query_reply.c->tcp_req_info)
2000                         tcp_req_info_remove_mesh_state(r->query_reply.c->tcp_req_info, mstate);
2001                 prev = r;
2002                 prev_buffer = r_buffer;
2003
2004                 /* Account for each reply sent. */
2005                 mesh->ans_expired++;
2006
2007         }
2008         if(mstate->reply_list) {
2009                 mstate->reply_list = NULL;
2010                 if(!mstate->reply_list && !mstate->cb_list) {
2011                         log_assert(mesh->num_reply_states > 0);
2012                         mesh->num_reply_states--;
2013                         if(mstate->super_set.count == 0) {
2014                                 mesh->num_detached_states++;
2015                         }
2016                 }
2017         }
2018         while((c = mstate->cb_list) != NULL) {
2019                 /* take this cb off the list; so that the list can be
2020                  * changed, eg. by adds from the callback routine */
2021                 if(!mstate->reply_list && mstate->cb_list && !c->next) {
2022                         /* was a reply state, not anymore */
2023                         log_assert(qstate->env->mesh->num_reply_states > 0);
2024                         qstate->env->mesh->num_reply_states--;
2025                 }
2026                 mstate->cb_list = c->next;
2027                 if(!mstate->reply_list && !mstate->cb_list &&
2028                         mstate->super_set.count == 0)
2029                         qstate->env->mesh->num_detached_states++;
2030                 mesh_do_callback(mstate, LDNS_RCODE_NOERROR, msg->rep, c);
2031         }
2032 }