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