services/mesh.c

   1 /*
   2  * services/mesh.c - deal with mesh of query states and handle events for that.
   3  *
   4  * Copyright (c) 2007, NLnet Labs. All rights reserved.
   5  *
   6  * This software is open source.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  *
  12  * Redistributions of source code must retain the above copyright notice,
  13  * this list of conditions and the following disclaimer.
  14  *
  15  * Redistributions in binary form must reproduce the above copyright notice,
  16  * this list of conditions and the following disclaimer in the documentation
  17  * and/or other materials provided with the distribution.
  18  *
  19  * Neither the name of the NLNET LABS nor the names of its contributors may
  20  * be used to endorse or promote products derived from this software without
  21  * specific prior written permission.
  22  *
  23  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  24  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  25  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  26  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  27  * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  28  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
  29  * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
  30  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  31  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  32  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  33  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  34  */
  35
  36 /**
  37  * \file
  38  *
  39  * This file contains functions to assist in dealing with a mesh of
  40  * query states. This mesh is supposed to be thread-specific.
  41  * It consists of query states (per qname, qtype, qclass) and connections
  42  * between query states and the super and subquery states, and replies to
  43  * send back to clients.
  44  */
  45 #include "config.h"
  46 #include "services/mesh.h"
  47 #include "services/outbound_list.h"
  48 #include "services/cache/dns.h"
  49 #include "util/log.h"
  50 #include "util/net_help.h"
  51 #include "util/module.h"
  52 #include "util/regional.h"
  53 #include "util/data/msgencode.h"
  54 #include "util/timehist.h"
  55 #include "util/fptr_wlist.h"
  56 #include "util/alloc.h"
  57 #include "util/config_file.h"
  58 #include "util/edns.h"
  59 #include "sldns/sbuffer.h"
  60 #include "sldns/wire2str.h"
  61 #include "services/localzone.h"
  62 #include "util/data/dname.h"
  63 #include "respip/respip.h"
  64 #include "services/listen_dnsport.h"
  65
  66 /** subtract timers and the values do not overflow or become negative */
  67 static void
  68 timeval_subtract(struct timeval* d, const struct timeval* end, const struct timeval* start)
  69 {
  70 #ifndef S_SPLINT_S
  71         time_t end_usec = end->tv_usec;
  72         d->tv_sec = end->tv_sec - start->tv_sec;
  73         if(end_usec < start->tv_usec) {
  74                 end_usec += 1000000;
  75                 d->tv_sec--;
  76         }
  77         d->tv_usec = end_usec - start->tv_usec;
  78 #endif
  79 }
  80
  81 /** add timers and the values do not overflow or become negative */
  82 static void
  83 timeval_add(struct timeval* d, const struct timeval* add)
  84 {
  85 #ifndef S_SPLINT_S
  86         d->tv_sec += add->tv_sec;
  87         d->tv_usec += add->tv_usec;
  88         if(d->tv_usec > 1000000 ) {
  89                 d->tv_usec -= 1000000;
  90                 d->tv_sec++;
  91         }
  92 #endif
  93 }
  94
  95 /** divide sum of timers to get average */
  96 static void
  97 timeval_divide(struct timeval* avg, const struct timeval* sum, size_t d)
  98 {
  99 #ifndef S_SPLINT_S
 100         size_t leftover;
 101         if(d == 0) {
 102                 avg->tv_sec = 0;
 103                 avg->tv_usec = 0;
 104                 return;
 105         }
 106         avg->tv_sec = sum->tv_sec / d;
 107         avg->tv_usec = sum->tv_usec / d;
 108         /* handle fraction from seconds divide */
 109         leftover = sum->tv_sec - avg->tv_sec*d;
 110         avg->tv_usec += (leftover*1000000)/d;
 111 #endif
 112 }
 113
 114 /** histogram compare of time values */
 115 static int
 116 timeval_smaller(const struct timeval* x, const struct timeval* y)
 117 {
 118 #ifndef S_SPLINT_S
 119         if(x->tv_sec < y->tv_sec)
 120                 return 1;
 121         else if(x->tv_sec == y->tv_sec) {
 122                 if(x->tv_usec <= y->tv_usec)
 123                         return 1;
 124                 else    return 0;
 125         }
 126         else    return 0;
 127 #endif
 128 }
 129
 130 /*
 131  * Compare two response-ip client info entries for the purpose of mesh state
 132  * compare.  It returns 0 if ci_a and ci_b are considered equal; otherwise
 133  * 1 or -1 (they mean 'ci_a is larger/smaller than ci_b', respectively, but
 134  * in practice it should be only used to mean they are different).
 135  * We cannot share the mesh state for two queries if different response-ip
 136  * actions can apply in the end, even if those queries are otherwise identical.
 137  * For this purpose we compare tag lists and tag action lists; they should be
 138  * identical to share the same state.
 139  * For tag data, we don't look into the data content, as it can be
 140  * expensive; unless tag data are not defined for both or they point to the
 141  * exact same data in memory (i.e., they come from the same ACL entry), we
 142  * consider these data different.
 143  * Likewise, if the client info is associated with views, we don't look into
 144  * the views.  They are considered different unless they are exactly the same
 145  * even if the views only differ in the names.
 146  */
 147 static int
 148 client_info_compare(const struct respip_client_info* ci_a,
 149         const struct respip_client_info* ci_b)
 150 {
 151         int cmp;
 152
 153         if(!ci_a && !ci_b)
 154                 return 0;
 155         if(ci_a && !ci_b)
 156                 return -1;
 157         if(!ci_a && ci_b)
 158                 return 1;
 159         if(ci_a->taglen != ci_b->taglen)
 160                 return (ci_a->taglen < ci_b->taglen) ? -1 : 1;
 161         cmp = memcmp(ci_a->taglist, ci_b->taglist, ci_a->taglen);
 162         if(cmp != 0)
 163                 return cmp;
 164         if(ci_a->tag_actions_size != ci_b->tag_actions_size)
 165                 return (ci_a->tag_actions_size < ci_b->tag_actions_size) ?
 166                         -1 : 1;
 167         cmp = memcmp(ci_a->tag_actions, ci_b->tag_actions,
 168                 ci_a->tag_actions_size);
 169         if(cmp != 0)
 170                 return cmp;
 171         if(ci_a->tag_datas != ci_b->tag_datas)
 172                 return ci_a->tag_datas < ci_b->tag_datas ? -1 : 1;
 173         if(ci_a->view != ci_b->view)
 174                 return ci_a->view < ci_b->view ? -1 : 1;
 175         /* For the unbound daemon these should be non-NULL and identical,
 176          * but we check that just in case. */
 177         if(ci_a->respip_set != ci_b->respip_set)
 178                 return ci_a->respip_set < ci_b->respip_set ? -1 : 1;
 179         return 0;
 180 }
 181
 182 int
 183 mesh_state_compare(const void* ap, const void* bp)
 184 {
 185         struct mesh_state* a = (struct mesh_state*)ap;
 186         struct mesh_state* b = (struct mesh_state*)bp;
 187         int cmp;
 188
 189         if(a->unique < b->unique)
 190                 return -1;
 191         if(a->unique > b->unique)
 192                 return 1;
 193
 194         if(a->s.is_priming && !b->s.is_priming)
 195                 return -1;
 196         if(!a->s.is_priming && b->s.is_priming)
 197                 return 1;
 198
 199         if(a->s.is_valrec && !b->s.is_valrec)
 200                 return -1;
 201         if(!a->s.is_valrec && b->s.is_valrec)
 202                 return 1;
 203
 204         if((a->s.query_flags&BIT_RD) && !(b->s.query_flags&BIT_RD))
 205                 return -1;
 206         if(!(a->s.query_flags&BIT_RD) && (b->s.query_flags&BIT_RD))
 207                 return 1;
 208
 209         if((a->s.query_flags&BIT_CD) && !(b->s.query_flags&BIT_CD))
 210                 return -1;
 211         if(!(a->s.query_flags&BIT_CD) && (b->s.query_flags&BIT_CD))
 212                 return 1;
 213
 214         cmp = query_info_compare(&a->s.qinfo, &b->s.qinfo);
 215         if(cmp != 0)
 216                 return cmp;
 217         return client_info_compare(a->s.client_info, b->s.client_info);
 218 }
 219
 220 int
 221 mesh_state_ref_compare(const void* ap, const void* bp)
 222 {
 223         struct mesh_state_ref* a = (struct mesh_state_ref*)ap;
 224         struct mesh_state_ref* b = (struct mesh_state_ref*)bp;
 225         return mesh_state_compare(a->s, b->s);
 226 }
 227
 228 struct mesh_area*
 229 mesh_create(struct module_stack* stack, struct module_env* env)
 230 {
 231         struct mesh_area* mesh = calloc(1, sizeof(struct mesh_area));
 232         if(!mesh) {
 233                 log_err("mesh area alloc: out of memory");
 234                 return NULL;
 235         }
 236         mesh->histogram = timehist_setup();
 237         mesh->qbuf_bak = sldns_buffer_new(env->cfg->msg_buffer_size);
 238         if(!mesh->histogram || !mesh->qbuf_bak) {
 239                 free(mesh);
 240                 log_err("mesh area alloc: out of memory");
 241                 return NULL;
 242         }
 243         mesh->mods = *stack;
 244         mesh->env = env;
 245         rbtree_init(&mesh->run, &mesh_state_compare);
 246         rbtree_init(&mesh->all, &mesh_state_compare);
 247         mesh->num_reply_addrs = 0;
 248         mesh->num_reply_states = 0;
 249         mesh->num_detached_states = 0;
 250         mesh->num_forever_states = 0;
 251         mesh->stats_jostled = 0;
 252         mesh->stats_dropped = 0;
 253         mesh->max_reply_states = env->cfg->num_queries_per_thread;
 254         mesh->max_forever_states = (mesh->max_reply_states+1)/2;
 255 #ifndef S_SPLINT_S
 256         mesh->jostle_max.tv_sec = (time_t)(env->cfg->jostle_time / 1000);
 257         mesh->jostle_max.tv_usec = (time_t)((env->cfg->jostle_time % 1000)
 258                 *1000);
 259 #endif
 260         return mesh;
 261 }
 262
 263 /** help mesh delete delete mesh states */
 264 static void
 265 mesh_delete_helper(rbnode_type* n)
 266 {
 267         struct mesh_state* mstate = (struct mesh_state*)n->key;
 268         /* perform a full delete, not only 'cleanup' routine,
 269          * because other callbacks expect a clean state in the mesh.
 270          * For 're-entrant' calls */
 271         mesh_state_delete(&mstate->s);
 272         /* but because these delete the items from the tree, postorder
 273          * traversal and rbtree rebalancing do not work together */
 274 }
 275
 276 void
 277 mesh_delete(struct mesh_area* mesh)
 278 {
 279         if(!mesh)
 280                 return;
 281         /* free all query states */
 282         while(mesh->all.count)
 283                 mesh_delete_helper(mesh->all.root);
 284         timehist_delete(mesh->histogram);
 285         sldns_buffer_free(mesh->qbuf_bak);
 286         free(mesh);
 287 }
 288
 289 void
 290 mesh_delete_all(struct mesh_area* mesh)
 291 {
 292         /* free all query states */
 293         while(mesh->all.count)
 294                 mesh_delete_helper(mesh->all.root);
 295         mesh->stats_dropped += mesh->num_reply_addrs;
 296         /* clear mesh area references */
 297         rbtree_init(&mesh->run, &mesh_state_compare);
 298         rbtree_init(&mesh->all, &mesh_state_compare);
 299         mesh->num_reply_addrs = 0;
 300         mesh->num_reply_states = 0;
 301         mesh->num_detached_states = 0;
 302         mesh->num_forever_states = 0;
 303         mesh->forever_first = NULL;
 304         mesh->forever_last = NULL;
 305         mesh->jostle_first = NULL;
 306         mesh->jostle_last = NULL;
 307 }
 308
 309 int mesh_make_new_space(struct mesh_area* mesh, sldns_buffer* qbuf)
 310 {
 311         struct mesh_state* m = mesh->jostle_first;
 312         /* free space is available */
 313         if(mesh->num_reply_states < mesh->max_reply_states)
 314                 return 1;
 315         /* try to kick out a jostle-list item */
 316         if(m && m->reply_list && m->list_select == mesh_jostle_list) {
 317                 /* how old is it? */
 318                 struct timeval age;
 319                 timeval_subtract(&age, mesh->env->now_tv,
 320                         &m->reply_list->start_time);
 321                 if(timeval_smaller(&mesh->jostle_max, &age)) {
 322                         /* its a goner */
 323                         log_nametypeclass(VERB_ALGO, "query jostled out to "
 324                                 "make space for a new one",
 325                                 m->s.qinfo.qname, m->s.qinfo.qtype,
 326                                 m->s.qinfo.qclass);
 327                         /* backup the query */
 328                         if(qbuf) sldns_buffer_copy(mesh->qbuf_bak, qbuf);
 329                         /* notify supers */
 330                         if(m->super_set.count > 0) {
 331                                 verbose(VERB_ALGO, "notify supers of failure");
 332                                 m->s.return_msg = NULL;
 333                                 m->s.return_rcode = LDNS_RCODE_SERVFAIL;
 334                                 mesh_walk_supers(mesh, m);
 335                         }
 336                         mesh->stats_jostled ++;
 337                         mesh_state_delete(&m->s);
 338                         /* restore the query - note that the qinfo ptr to
 339                          * the querybuffer is then correct again. */
 340                         if(qbuf) sldns_buffer_copy(qbuf, mesh->qbuf_bak);
 341                         return 1;
 342                 }
 343         }
 344         /* no space for new item */
 345         return 0;
 346 }
 347
 348 void mesh_new_client(struct mesh_area* mesh, struct query_info* qinfo,
 349         struct respip_client_info* cinfo, uint16_t qflags,
 350         struct edns_data* edns, struct comm_reply* rep, uint16_t qid)
 351 {
 352         struct mesh_state* s = NULL;
 353         int unique = unique_mesh_state(edns->opt_list, mesh->env);
 354         int was_detached = 0;
 355         int was_noreply = 0;
 356         int added = 0;
 357         if(!unique)
 358                 s = mesh_area_find(mesh, cinfo, qinfo, qflags&(BIT_RD|BIT_CD), 0, 0);
 359         /* does this create a new reply state? */
 360         if(!s || s->list_select == mesh_no_list) {
 361                 if(!mesh_make_new_space(mesh, rep->c->buffer)) {
 362                         verbose(VERB_ALGO, "Too many queries. dropping "
 363                                 "incoming query.");
 364                         comm_point_drop_reply(rep);
 365                         mesh->stats_dropped ++;
 366                         return;
 367                 }
 368                 /* for this new reply state, the reply address is free,
 369                  * so the limit of reply addresses does not stop reply states*/
 370         } else {
 371                 /* protect our memory usage from storing reply addresses */
 372                 if(mesh->num_reply_addrs > mesh->max_reply_states*16) {
 373                         verbose(VERB_ALGO, "Too many requests queued. "
 374                                 "dropping incoming query.");
 375                         mesh->stats_dropped++;
 376                         comm_point_drop_reply(rep);
 377                         return;
 378                 }
 379         }
 380         /* see if it already exists, if not, create one */
 381         if(!s) {
 382 #ifdef UNBOUND_DEBUG
 383                 struct rbnode_type* n;
 384 #endif
 385                 s = mesh_state_create(mesh->env, qinfo, cinfo,
 386                         qflags&(BIT_RD|BIT_CD), 0, 0);
 387                 if(!s) {
 388                         log_err("mesh_state_create: out of memory; SERVFAIL");
 389                         if(!inplace_cb_reply_servfail_call(mesh->env, qinfo, NULL, NULL,
 390                                 LDNS_RCODE_SERVFAIL, edns, rep, mesh->env->scratch))
 391                                         edns->opt_list = NULL;
 392                         error_encode(rep->c->buffer, LDNS_RCODE_SERVFAIL,
 393                                 qinfo, qid, qflags, edns);
 394                         comm_point_send_reply(rep);
 395                         return;
 396                 }
 397                 if(unique)
 398                         mesh_state_make_unique(s);
 399                 /* copy the edns options we got from the front */
 400                 if(edns->opt_list) {
 401                         s->s.edns_opts_front_in = edns_opt_copy_region(edns->opt_list,
 402                                 s->s.region);
 403                         if(!s->s.edns_opts_front_in) {
 404                                 log_err("mesh_state_create: out of memory; SERVFAIL");
 405                                 if(!inplace_cb_reply_servfail_call(mesh->env, qinfo, NULL,
 406                                         NULL, LDNS_RCODE_SERVFAIL, edns, rep, mesh->env->scratch))
 407                                                 edns->opt_list = NULL;
 408                                 error_encode(rep->c->buffer, LDNS_RCODE_SERVFAIL,
 409                                         qinfo, qid, qflags, edns);
 410                                 comm_point_send_reply(rep);
 411                                 return;
 412                         }
 413                 }
 414
 415 #ifdef UNBOUND_DEBUG
 416                 n =
 417 #else
 418                 (void)
 419 #endif
 420                 rbtree_insert(&mesh->all, &s->node);
 421                 log_assert(n != NULL);
 422                 /* set detached (it is now) */
 423                 mesh->num_detached_states++;
 424                 added = 1;
 425         }
 426         if(!s->reply_list && !s->cb_list && s->super_set.count == 0)
 427                 was_detached = 1;
 428         if(!s->reply_list && !s->cb_list)
 429                 was_noreply = 1;
 430         /* add reply to s */
 431         if(!mesh_state_add_reply(s, edns, rep, qid, qflags, qinfo)) {
 432                         log_err("mesh_new_client: out of memory; SERVFAIL");
 433                 servfail_mem:
 434                         if(!inplace_cb_reply_servfail_call(mesh->env, qinfo, &s->s,
 435                                 NULL, LDNS_RCODE_SERVFAIL, edns, rep, mesh->env->scratch))
 436                                         edns->opt_list = NULL;
 437                         error_encode(rep->c->buffer, LDNS_RCODE_SERVFAIL,
 438                                 qinfo, qid, qflags, edns);
 439                         comm_point_send_reply(rep);
 440                         if(added)
 441                                 mesh_state_delete(&s->s);
 442                         return;
 443         }
 444         if(rep->c->tcp_req_info) {
 445                 if(!tcp_req_info_add_meshstate(rep->c->tcp_req_info, mesh, s)) {
 446                         log_err("mesh_new_client: out of memory add tcpreqinfo");
 447                         goto servfail_mem;
 448                 }
 449         }
 450         /* update statistics */
 451         if(was_detached) {
 452                 log_assert(mesh->num_detached_states > 0);
 453                 mesh->num_detached_states--;
 454         }
 455         if(was_noreply) {
 456                 mesh->num_reply_states ++;
 457         }
 458         mesh->num_reply_addrs++;
 459         if(s->list_select == mesh_no_list) {
 460                 /* move to either the forever or the jostle_list */
 461                 if(mesh->num_forever_states < mesh->max_forever_states) {
 462                         mesh->num_forever_states ++;
 463                         mesh_list_insert(s, &mesh->forever_first,
 464                                 &mesh->forever_last);
 465                         s->list_select = mesh_forever_list;
 466                 } else {
 467                         mesh_list_insert(s, &mesh->jostle_first,
 468                                 &mesh->jostle_last);
 469                         s->list_select = mesh_jostle_list;
 470                 }
 471         }
 472         if(added)
 473                 mesh_run(mesh, s, module_event_new, NULL);
 474 }
 475
 476 int
 477 mesh_new_callback(struct mesh_area* mesh, struct query_info* qinfo,
 478         uint16_t qflags, struct edns_data* edns, sldns_buffer* buf,
 479         uint16_t qid, mesh_cb_func_type cb, void* cb_arg)
 480 {
 481         struct mesh_state* s = NULL;
 482         int unique = unique_mesh_state(edns->opt_list, mesh->env);
 483         int was_detached = 0;
 484         int was_noreply = 0;
 485         int added = 0;
 486         if(!unique)
 487                 s = mesh_area_find(mesh, NULL, qinfo, qflags&(BIT_RD|BIT_CD), 0, 0);
 488
 489         /* there are no limits on the number of callbacks */
 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, NULL,
 497                         qflags&(BIT_RD|BIT_CD), 0, 0);
 498                 if(!s) {
 499                         return 0;
 500                 }
 501                 if(unique)
 502                         mesh_state_make_unique(s);
 503                 if(edns->opt_list) {
 504                         s->s.edns_opts_front_in = edns_opt_copy_region(edns->opt_list,
 505                                 s->s.region);
 506                         if(!s->s.edns_opts_front_in) {
 507                                 return 0;
 508                         }
 509                 }
 510 #ifdef UNBOUND_DEBUG
 511                 n =
 512 #else
 513                 (void)
 514 #endif
 515                 rbtree_insert(&mesh->all, &s->node);
 516                 log_assert(n != NULL);
 517                 /* set detached (it is now) */
 518                 mesh->num_detached_states++;
 519                 added = 1;
 520         }
 521         if(!s->reply_list && !s->cb_list && s->super_set.count == 0)
 522                 was_detached = 1;
 523         if(!s->reply_list && !s->cb_list)
 524                 was_noreply = 1;
 525         /* add reply to s */
 526         if(!mesh_state_add_cb(s, edns, buf, cb, cb_arg, qid, qflags)) {
 527                         if(added)
 528                                 mesh_state_delete(&s->s);
 529                         return 0;
 530         }
 531         /* update statistics */
 532         if(was_detached) {
 533                 log_assert(mesh->num_detached_states > 0);
 534                 mesh->num_detached_states--;
 535         }
 536         if(was_noreply) {
 537                 mesh->num_reply_states ++;
 538         }
 539         mesh->num_reply_addrs++;
 540         if(added)
 541                 mesh_run(mesh, s, module_event_new, NULL);
 542         return 1;
 543 }
 544
 545 static void mesh_schedule_prefetch(struct mesh_area* mesh,
 546         struct query_info* qinfo, uint16_t qflags, time_t leeway, int run);
 547
 548 void mesh_new_prefetch(struct mesh_area* mesh, struct query_info* qinfo,
 549         uint16_t qflags, time_t leeway)
 550 {
 551         mesh_schedule_prefetch(mesh, qinfo, qflags, leeway, 1);
 552 }
 553
 554 /* Internal backend routine of mesh_new_prefetch().  It takes one additional
 555  * parameter, 'run', which controls whether to run the prefetch state
 556  * immediately.  When this function is called internally 'run' could be
 557  * 0 (false), in which case the new state is only made runnable so it
 558  * will not be run recursively on top of the current state. */
 559 static void mesh_schedule_prefetch(struct mesh_area* mesh,
 560         struct query_info* qinfo, uint16_t qflags, time_t leeway, int run)
 561 {
 562         struct mesh_state* s = mesh_area_find(mesh, NULL, qinfo,
 563                 qflags&(BIT_RD|BIT_CD), 0, 0);
 564 #ifdef UNBOUND_DEBUG
 565         struct rbnode_type* n;
 566 #endif
 567         /* already exists, and for a different purpose perhaps.
 568          * if mesh_no_list, keep it that way. */
 569         if(s) {
 570                 /* make it ignore the cache from now on */
 571                 if(!s->s.blacklist)
 572                         sock_list_insert(&s->s.blacklist, NULL, 0, s->s.region);
 573                 if(s->s.prefetch_leeway < leeway)
 574                         s->s.prefetch_leeway = leeway;
 575                 return;
 576         }
 577         if(!mesh_make_new_space(mesh, NULL)) {
 578                 verbose(VERB_ALGO, "Too many queries. dropped prefetch.");
 579                 mesh->stats_dropped ++;
 580                 return;
 581         }
 582
 583         s = mesh_state_create(mesh->env, qinfo, NULL,
 584                 qflags&(BIT_RD|BIT_CD), 0, 0);
 585         if(!s) {
 586                 log_err("prefetch mesh_state_create: out of memory");
 587                 return;
 588         }
 589 #ifdef UNBOUND_DEBUG
 590         n =
 591 #else
 592         (void)
 593 #endif
 594         rbtree_insert(&mesh->all, &s->node);
 595         log_assert(n != NULL);
 596         /* set detached (it is now) */
 597         mesh->num_detached_states++;
 598         /* make it ignore the cache */
 599         sock_list_insert(&s->s.blacklist, NULL, 0, s->s.region);
 600         s->s.prefetch_leeway = leeway;
 601
 602         if(s->list_select == mesh_no_list) {
 603                 /* move to either the forever or the jostle_list */
 604                 if(mesh->num_forever_states < mesh->max_forever_states) {
 605                         mesh->num_forever_states ++;
 606                         mesh_list_insert(s, &mesh->forever_first,
 607                                 &mesh->forever_last);
 608                         s->list_select = mesh_forever_list;
 609                 } else {
 610                         mesh_list_insert(s, &mesh->jostle_first,
 611                                 &mesh->jostle_last);
 612                         s->list_select = mesh_jostle_list;
 613                 }
 614         }
 615
 616         if(!run) {
 617 #ifdef UNBOUND_DEBUG
 618                 n =
 619 #else
 620                 (void)
 621 #endif
 622                 rbtree_insert(&mesh->run, &s->run_node);
 623                 log_assert(n != NULL);
 624                 return;
 625         }
 626
 627         mesh_run(mesh, s, module_event_new, NULL);
 628 }
 629
 630 void mesh_report_reply(struct mesh_area* mesh, struct outbound_entry* e,
 631         struct comm_reply* reply, int what)
 632 {
 633         enum module_ev event = module_event_reply;
 634         e->qstate->reply = reply;
 635         if(what != NETEVENT_NOERROR) {
 636                 event = module_event_noreply;
 637                 if(what == NETEVENT_CAPSFAIL)
 638                         event = module_event_capsfail;
 639         }
 640         mesh_run(mesh, e->qstate->mesh_info, event, e);
 641 }
 642
 643 struct mesh_state*
 644 mesh_state_create(struct module_env* env, struct query_info* qinfo,
 645         struct respip_client_info* cinfo, uint16_t qflags, int prime,
 646         int valrec)
 647 {
 648         struct regional* region = alloc_reg_obtain(env->alloc);
 649         struct mesh_state* mstate;
 650         int i;
 651         if(!region)
 652                 return NULL;
 653         mstate = (struct mesh_state*)regional_alloc(region,
 654                 sizeof(struct mesh_state));
 655         if(!mstate) {
 656                 alloc_reg_release(env->alloc, region);
 657                 return NULL;
 658         }
 659         memset(mstate, 0, sizeof(*mstate));
 660         mstate->node = *RBTREE_NULL;
 661         mstate->run_node = *RBTREE_NULL;
 662         mstate->node.key = mstate;
 663         mstate->run_node.key = mstate;
 664         mstate->reply_list = NULL;
 665         mstate->list_select = mesh_no_list;
 666         mstate->replies_sent = 0;
 667         rbtree_init(&mstate->super_set, &mesh_state_ref_compare);
 668         rbtree_init(&mstate->sub_set, &mesh_state_ref_compare);
 669         mstate->num_activated = 0;
 670         mstate->unique = NULL;
 671         /* init module qstate */
 672         mstate->s.qinfo.qtype = qinfo->qtype;
 673         mstate->s.qinfo.qclass = qinfo->qclass;
 674         mstate->s.qinfo.local_alias = NULL;
 675         mstate->s.qinfo.qname_len = qinfo->qname_len;
 676         mstate->s.qinfo.qname = regional_alloc_init(region, qinfo->qname,
 677                 qinfo->qname_len);
 678         if(!mstate->s.qinfo.qname) {
 679                 alloc_reg_release(env->alloc, region);
 680                 return NULL;
 681         }
 682         if(cinfo) {
 683                 mstate->s.client_info = regional_alloc_init(region, cinfo,
 684                         sizeof(*cinfo));
 685                 if(!mstate->s.client_info) {
 686                         alloc_reg_release(env->alloc, region);
 687                         return NULL;
 688                 }
 689         }
 690         /* remove all weird bits from qflags */
 691         mstate->s.query_flags = (qflags & (BIT_RD|BIT_CD));
 692         mstate->s.is_priming = prime;
 693         mstate->s.is_valrec = valrec;
 694         mstate->s.reply = NULL;
 695         mstate->s.region = region;
 696         mstate->s.curmod = 0;
 697         mstate->s.return_msg = 0;
 698         mstate->s.return_rcode = LDNS_RCODE_NOERROR;
 699         mstate->s.env = env;
 700         mstate->s.mesh_info = mstate;
 701         mstate->s.prefetch_leeway = 0;
 702         mstate->s.no_cache_lookup = 0;
 703         mstate->s.no_cache_store = 0;
 704         mstate->s.need_refetch = 0;
 705         mstate->s.was_ratelimited = 0;
 706
 707         /* init modules */
 708         for(i=0; i<env->mesh->mods.num; i++) {
 709                 mstate->s.minfo[i] = NULL;
 710                 mstate->s.ext_state[i] = module_state_initial;
 711         }
 712         /* init edns option lists */
 713         mstate->s.edns_opts_front_in = NULL;
 714         mstate->s.edns_opts_back_out = NULL;
 715         mstate->s.edns_opts_back_in = NULL;
 716         mstate->s.edns_opts_front_out = NULL;
 717
 718         return mstate;
 719 }
 720
 721 int
 722 mesh_state_is_unique(struct mesh_state* mstate)
 723 {
 724         return mstate->unique != NULL;
 725 }
 726
 727 void
 728 mesh_state_make_unique(struct mesh_state* mstate)
 729 {
 730         mstate->unique = mstate;
 731 }
 732
 733 void
 734 mesh_state_cleanup(struct mesh_state* mstate)
 735 {
 736         struct mesh_area* mesh;
 737         int i;
 738         if(!mstate)
 739                 return;
 740         mesh = mstate->s.env->mesh;
 741         /* drop unsent replies */
 742         if(!mstate->replies_sent) {
 743                 struct mesh_reply* rep = mstate->reply_list;
 744                 struct mesh_cb* cb;
 745                 /* in tcp_req_info, the mstates linked are removed, but
 746                  * the reply_list is now NULL, so the remove-from-empty-list
 747                  * takes no time and also it does not do the mesh accounting */
 748                 mstate->reply_list = NULL;
 749                 for(; rep; rep=rep->next) {
 750                         comm_point_drop_reply(&rep->query_reply);
 751                         mesh->num_reply_addrs--;
 752                 }
 753                 while((cb = mstate->cb_list)!=NULL) {
 754                         mstate->cb_list = cb->next;
 755                         fptr_ok(fptr_whitelist_mesh_cb(cb->cb));
 756                         (*cb->cb)(cb->cb_arg, LDNS_RCODE_SERVFAIL, NULL,
 757                                 sec_status_unchecked, NULL, 0);
 758                         mesh->num_reply_addrs--;
 759                 }
 760         }
 761
 762         /* de-init modules */
 763         for(i=0; i<mesh->mods.num; i++) {
 764                 fptr_ok(fptr_whitelist_mod_clear(mesh->mods.mod[i]->clear));
 765                 (*mesh->mods.mod[i]->clear)(&mstate->s, i);
 766                 mstate->s.minfo[i] = NULL;
 767                 mstate->s.ext_state[i] = module_finished;
 768         }
 769         alloc_reg_release(mstate->s.env->alloc, mstate->s.region);
 770 }
 771
 772 void
 773 mesh_state_delete(struct module_qstate* qstate)
 774 {
 775         struct mesh_area* mesh;
 776         struct mesh_state_ref* super, ref;
 777         struct mesh_state* mstate;
 778         if(!qstate)
 779                 return;
 780         mstate = qstate->mesh_info;
 781         mesh = mstate->s.env->mesh;
 782         mesh_detach_subs(&mstate->s);
 783         if(mstate->list_select == mesh_forever_list) {
 784                 mesh->num_forever_states --;
 785                 mesh_list_remove(mstate, &mesh->forever_first,
 786                         &mesh->forever_last);
 787         } else if(mstate->list_select == mesh_jostle_list) {
 788                 mesh_list_remove(mstate, &mesh->jostle_first,
 789                         &mesh->jostle_last);
 790         }
 791         if(!mstate->reply_list && !mstate->cb_list
 792                 && mstate->super_set.count == 0) {
 793                 log_assert(mesh->num_detached_states > 0);
 794                 mesh->num_detached_states--;
 795         }
 796         if(mstate->reply_list || mstate->cb_list) {
 797                 log_assert(mesh->num_reply_states > 0);
 798                 mesh->num_reply_states--;
 799         }
 800         ref.node.key = &ref;
 801         ref.s = mstate;
 802         RBTREE_FOR(super, struct mesh_state_ref*, &mstate->super_set) {
 803                 (void)rbtree_delete(&super->s->sub_set, &ref);
 804         }
 805         (void)rbtree_delete(&mesh->run, mstate);
 806         (void)rbtree_delete(&mesh->all, mstate);
 807         mesh_state_cleanup(mstate);
 808 }
 809
 810 /** helper recursive rbtree find routine */
 811 static int
 812 find_in_subsub(struct mesh_state* m, struct mesh_state* tofind, size_t *c)
 813 {
 814         struct mesh_state_ref* r;
 815         if((*c)++ > MESH_MAX_SUBSUB)
 816                 return 1;
 817         RBTREE_FOR(r, struct mesh_state_ref*, &m->sub_set) {
 818                 if(r->s == tofind || find_in_subsub(r->s, tofind, c))
 819                         return 1;
 820         }
 821         return 0;
 822 }
 823
 824 /** find cycle for already looked up mesh_state */
 825 static int
 826 mesh_detect_cycle_found(struct module_qstate* qstate, struct mesh_state* dep_m)
 827 {
 828         struct mesh_state* cyc_m = qstate->mesh_info;
 829         size_t counter = 0;
 830         if(!dep_m)
 831                 return 0;
 832         if(dep_m == cyc_m || find_in_subsub(dep_m, cyc_m, &counter)) {
 833                 if(counter > MESH_MAX_SUBSUB)
 834                         return 2;
 835                 return 1;
 836         }
 837         return 0;
 838 }
 839
 840 void mesh_detach_subs(struct module_qstate* qstate)
 841 {
 842         struct mesh_area* mesh = qstate->env->mesh;
 843         struct mesh_state_ref* ref, lookup;
 844 #ifdef UNBOUND_DEBUG
 845         struct rbnode_type* n;
 846 #endif
 847         lookup.node.key = &lookup;
 848         lookup.s = qstate->mesh_info;
 849         RBTREE_FOR(ref, struct mesh_state_ref*, &qstate->mesh_info->sub_set) {
 850 #ifdef UNBOUND_DEBUG
 851                 n =
 852 #else
 853                 (void)
 854 #endif
 855                 rbtree_delete(&ref->s->super_set, &lookup);
 856                 log_assert(n != NULL); /* must have been present */
 857                 if(!ref->s->reply_list && !ref->s->cb_list
 858                         && ref->s->super_set.count == 0) {
 859                         mesh->num_detached_states++;
 860                         log_assert(mesh->num_detached_states +
 861                                 mesh->num_reply_states <= mesh->all.count);
 862                 }
 863         }
 864         rbtree_init(&qstate->mesh_info->sub_set, &mesh_state_ref_compare);
 865 }
 866
 867 int mesh_add_sub(struct module_qstate* qstate, struct query_info* qinfo,
 868         uint16_t qflags, int prime, int valrec, struct module_qstate** newq,
 869         struct mesh_state** sub)
 870 {
 871         /* find it, if not, create it */
 872         struct mesh_area* mesh = qstate->env->mesh;
 873         *sub = mesh_area_find(mesh, NULL, qinfo, qflags,
 874                 prime, valrec);
 875         if(mesh_detect_cycle_found(qstate, *sub)) {
 876                 verbose(VERB_ALGO, "attach failed, cycle detected");
 877                 return 0;
 878         }
 879         if(!*sub) {
 880 #ifdef UNBOUND_DEBUG
 881                 struct rbnode_type* n;
 882 #endif
 883                 /* create a new one */
 884                 *sub = mesh_state_create(qstate->env, qinfo, NULL, qflags, prime,
 885                         valrec);
 886                 if(!*sub) {
 887                         log_err("mesh_attach_sub: out of memory");
 888                         return 0;
 889                 }
 890 #ifdef UNBOUND_DEBUG
 891                 n =
 892 #else
 893                 (void)
 894 #endif
 895                 rbtree_insert(&mesh->all, &(*sub)->node);
 896                 log_assert(n != NULL);
 897                 /* set detached (it is now) */
 898                 mesh->num_detached_states++;
 899                 /* set new query state to run */
 900 #ifdef UNBOUND_DEBUG
 901                 n =
 902 #else
 903                 (void)
 904 #endif
 905                 rbtree_insert(&mesh->run, &(*sub)->run_node);
 906                 log_assert(n != NULL);
 907                 *newq = &(*sub)->s;
 908         } else
 909                 *newq = NULL;
 910         return 1;
 911 }
 912
 913 int mesh_attach_sub(struct module_qstate* qstate, struct query_info* qinfo,
 914         uint16_t qflags, int prime, int valrec, struct module_qstate** newq)
 915 {
 916         struct mesh_area* mesh = qstate->env->mesh;
 917         struct mesh_state* sub = NULL;
 918         int was_detached;
 919         if(!mesh_add_sub(qstate, qinfo, qflags, prime, valrec, newq, &sub))
 920                 return 0;
 921         was_detached = (sub->super_set.count == 0);
 922         if(!mesh_state_attachment(qstate->mesh_info, sub))
 923                 return 0;
 924         /* if it was a duplicate  attachment, the count was not zero before */
 925         if(!sub->reply_list && !sub->cb_list && was_detached &&
 926                 sub->super_set.count == 1) {
 927                 /* it used to be detached, before this one got added */
 928                 log_assert(mesh->num_detached_states > 0);
 929                 mesh->num_detached_states--;
 930         }
 931         /* *newq will be run when inited after the current module stops */
 932         return 1;
 933 }
 934
 935 int mesh_state_attachment(struct mesh_state* super, struct mesh_state* sub)
 936 {
 937 #ifdef UNBOUND_DEBUG
 938         struct rbnode_type* n;
 939 #endif
 940         struct mesh_state_ref* subref; /* points to sub, inserted in super */
 941         struct mesh_state_ref* superref; /* points to super, inserted in sub */
 942         if( !(subref = regional_alloc(super->s.region,
 943                 sizeof(struct mesh_state_ref))) ||
 944                 !(superref = regional_alloc(sub->s.region,
 945                 sizeof(struct mesh_state_ref))) ) {
 946                 log_err("mesh_state_attachment: out of memory");
 947                 return 0;
 948         }
 949         superref->node.key = superref;
 950         superref->s = super;
 951         subref->node.key = subref;
 952         subref->s = sub;
 953         if(!rbtree_insert(&sub->super_set, &superref->node)) {
 954                 /* this should not happen, iterator and validator do not
 955                  * attach subqueries that are identical. */
 956                 /* already attached, we are done, nothing todo.
 957                  * since superref and subref already allocated in region,
 958                  * we cannot free them */
 959                 return 1;
 960         }
 961 #ifdef UNBOUND_DEBUG
 962         n =
 963 #else
 964         (void)
 965 #endif
 966         rbtree_insert(&super->sub_set, &subref->node);
 967         log_assert(n != NULL); /* we checked above if statement, the reverse
 968           administration should not fail now, unless they are out of sync */
 969         return 1;
 970 }
 971
 972 /**
 973  * callback results to mesh cb entry
 974  * @param m: mesh state to send it for.
 975  * @param rcode: if not 0, error code.
 976  * @param rep: reply to send (or NULL if rcode is set).
 977  * @param r: callback entry
 978  */
 979 static void
 980 mesh_do_callback(struct mesh_state* m, int rcode, struct reply_info* rep,
 981         struct mesh_cb* r)
 982 {
 983         int secure;
 984         char* reason = NULL;
 985         int was_ratelimited = m->s.was_ratelimited;
 986         /* bogus messages are not made into servfail, sec_status passed
 987          * to the callback function */
 988         if(rep && rep->security == sec_status_secure)
 989                 secure = 1;
 990         else    secure = 0;
 991         if(!rep && rcode == LDNS_RCODE_NOERROR)
 992                 rcode = LDNS_RCODE_SERVFAIL;
 993         if(!rcode && (rep->security == sec_status_bogus ||
 994                 rep->security == sec_status_secure_sentinel_fail)) {
 995                 if(!(reason = errinf_to_str_bogus(&m->s)))
 996                         rcode = LDNS_RCODE_SERVFAIL;
 997         }
 998         /* send the reply */
 999         if(rcode) {
1000                 if(rcode == LDNS_RCODE_SERVFAIL) {
1001                         if(!inplace_cb_reply_servfail_call(m->s.env, &m->s.qinfo, &m->s,
1002                                 rep, rcode, &r->edns, NULL, m->s.region))
1003                                         r->edns.opt_list = NULL;
1004                 } else {
1005                         if(!inplace_cb_reply_call(m->s.env, &m->s.qinfo, &m->s, rep, rcode,
1006                                 &r->edns, NULL, m->s.region))
1007                                         r->edns.opt_list = NULL;
1008                 }
1009                 fptr_ok(fptr_whitelist_mesh_cb(r->cb));
1010                 (*r->cb)(r->cb_arg, rcode, r->buf, sec_status_unchecked, NULL,
1011                         was_ratelimited);
1012         } else {
1013                 size_t udp_size = r->edns.udp_size;
1014                 sldns_buffer_clear(r->buf);
1015                 r->edns.edns_version = EDNS_ADVERTISED_VERSION;
1016                 r->edns.udp_size = EDNS_ADVERTISED_SIZE;
1017                 r->edns.ext_rcode = 0;
1018                 r->edns.bits &= EDNS_DO;
1019
1020                 if(!inplace_cb_reply_call(m->s.env, &m->s.qinfo, &m->s, rep,
1021                         LDNS_RCODE_NOERROR, &r->edns, NULL, m->s.region) ||
1022                         !reply_info_answer_encode(&m->s.qinfo, rep, r->qid,
1023                         r->qflags, r->buf, 0, 1,
1024                         m->s.env->scratch, udp_size, &r->edns,
1025                         (int)(r->edns.bits & EDNS_DO), secure))
1026                 {
1027                         fptr_ok(fptr_whitelist_mesh_cb(r->cb));
1028                         (*r->cb)(r->cb_arg, LDNS_RCODE_SERVFAIL, r->buf,
1029                                 sec_status_unchecked, NULL, 0);
1030                 } else {
1031                         fptr_ok(fptr_whitelist_mesh_cb(r->cb));
1032                         (*r->cb)(r->cb_arg, LDNS_RCODE_NOERROR, r->buf,
1033                                 rep->security, reason, was_ratelimited);
1034                 }
1035         }
1036         free(reason);
1037         m->s.env->mesh->num_reply_addrs--;
1038 }
1039
1040 /**
1041  * Send reply to mesh reply entry
1042  * @param m: mesh state to send it for.
1043  * @param rcode: if not 0, error code.
1044  * @param rep: reply to send (or NULL if rcode is set).
1045  * @param r: reply entry
1046  * @param r_buffer: buffer to use for reply entry.
1047  * @param prev: previous reply, already has its answer encoded in buffer.
1048  * @param prev_buffer: buffer for previous reply.
1049  */
1050 static void
1051 mesh_send_reply(struct mesh_state* m, int rcode, struct reply_info* rep,
1052         struct mesh_reply* r, struct sldns_buffer* r_buffer,
1053         struct mesh_reply* prev, struct sldns_buffer* prev_buffer)
1054 {
1055         struct timeval end_time;
1056         struct timeval duration;
1057         int secure;
1058         /* Copy the client's EDNS for later restore, to make sure the edns
1059          * compare is with the correct edns options. */
1060         struct edns_data edns_bak = r->edns;
1061         /* examine security status */
1062         if(m->s.env->need_to_validate && (!(r->qflags&BIT_CD) ||
1063                 m->s.env->cfg->ignore_cd) && rep &&
1064                 (rep->security <= sec_status_bogus ||
1065                 rep->security == sec_status_secure_sentinel_fail)) {
1066                 rcode = LDNS_RCODE_SERVFAIL;
1067                 if(m->s.env->cfg->stat_extended)
1068                         m->s.env->mesh->ans_bogus++;
1069         }
1070         if(rep && rep->security == sec_status_secure)
1071                 secure = 1;
1072         else    secure = 0;
1073         if(!rep && rcode == LDNS_RCODE_NOERROR)
1074                 rcode = LDNS_RCODE_SERVFAIL;
1075         /* send the reply */
1076         /* We don't reuse the encoded answer if either the previous or current
1077          * response has a local alias.  We could compare the alias records
1078          * and still reuse the previous answer if they are the same, but that
1079          * would be complicated and error prone for the relatively minor case.
1080          * So we err on the side of safety. */
1081         if(prev && prev_buffer && prev->qflags == r->qflags &&
1082                 !prev->local_alias && !r->local_alias &&
1083                 prev->edns.edns_present == r->edns.edns_present &&
1084                 prev->edns.bits == r->edns.bits &&
1085                 prev->edns.udp_size == r->edns.udp_size &&
1086                 edns_opt_list_compare(prev->edns.opt_list, r->edns.opt_list)
1087                 == 0) {
1088                 /* if the previous reply is identical to this one, fix ID */
1089                 if(prev_buffer != r_buffer)
1090                         sldns_buffer_copy(r_buffer, prev_buffer);
1091                 sldns_buffer_write_at(r_buffer, 0, &r->qid, sizeof(uint16_t));
1092                 sldns_buffer_write_at(r_buffer, 12, r->qname,
1093                         m->s.qinfo.qname_len);
1094                 comm_point_send_reply(&r->query_reply);
1095         } else if(rcode) {
1096                 m->s.qinfo.qname = r->qname;
1097                 m->s.qinfo.local_alias = r->local_alias;
1098                 if(rcode == LDNS_RCODE_SERVFAIL) {
1099                         if(!inplace_cb_reply_servfail_call(m->s.env, &m->s.qinfo, &m->s,
1100                                 rep, rcode, &r->edns, NULL, m->s.region))
1101                                         r->edns.opt_list = NULL;
1102                 } else {
1103                         if(!inplace_cb_reply_call(m->s.env, &m->s.qinfo, &m->s, rep, rcode,
1104                                 &r->edns, NULL, m->s.region))
1105                                         r->edns.opt_list = NULL;
1106                 }
1107                 error_encode(r_buffer, rcode, &m->s.qinfo, r->qid,
1108                         r->qflags, &r->edns);
1109                 comm_point_send_reply(&r->query_reply);
1110         } else {
1111                 size_t udp_size = r->edns.udp_size;
1112                 r->edns.edns_version = EDNS_ADVERTISED_VERSION;
1113                 r->edns.udp_size = EDNS_ADVERTISED_SIZE;
1114                 r->edns.ext_rcode = 0;
1115                 r->edns.bits &= EDNS_DO;
1116                 m->s.qinfo.qname = r->qname;
1117                 m->s.qinfo.local_alias = r->local_alias;
1118                 if(!inplace_cb_reply_call(m->s.env, &m->s.qinfo, &m->s, rep,
1119                         LDNS_RCODE_NOERROR, &r->edns, NULL, m->s.region) ||
1120                         !apply_edns_options(&r->edns, &edns_bak,
1121                                 m->s.env->cfg, r->query_reply.c,
1122                                 m->s.region) ||
1123                         !reply_info_answer_encode(&m->s.qinfo, rep, r->qid,
1124                         r->qflags, r_buffer, 0, 1, m->s.env->scratch,
1125                         udp_size, &r->edns, (int)(r->edns.bits & EDNS_DO),
1126                         secure))
1127                 {
1128                         if(!inplace_cb_reply_servfail_call(m->s.env, &m->s.qinfo, &m->s,
1129                         rep, LDNS_RCODE_SERVFAIL, &r->edns, NULL, m->s.region))
1130                                 r->edns.opt_list = NULL;
1131                         error_encode(r_buffer, LDNS_RCODE_SERVFAIL,
1132                                 &m->s.qinfo, r->qid, r->qflags, &r->edns);
1133                 }
1134                 r->edns = edns_bak;
1135                 comm_point_send_reply(&r->query_reply);
1136         }
1137         /* account */
1138         m->s.env->mesh->num_reply_addrs--;
1139         end_time = *m->s.env->now_tv;
1140         timeval_subtract(&duration, &end_time, &r->start_time);
1141         verbose(VERB_ALGO, "query took " ARG_LL "d.%6.6d sec",
1142                 (long long)duration.tv_sec, (int)duration.tv_usec);
1143         m->s.env->mesh->replies_sent++;
1144         timeval_add(&m->s.env->mesh->replies_sum_wait, &duration);
1145         timehist_insert(m->s.env->mesh->histogram, &duration);
1146         if(m->s.env->cfg->stat_extended) {
1147                 uint16_t rc = FLAGS_GET_RCODE(sldns_buffer_read_u16_at(
1148                         r_buffer, 2));
1149                 if(secure) m->s.env->mesh->ans_secure++;
1150                 m->s.env->mesh->ans_rcode[ rc ] ++;
1151                 if(rc == 0 && LDNS_ANCOUNT(sldns_buffer_begin(r_buffer)) == 0)
1152                         m->s.env->mesh->ans_nodata++;
1153         }
1154         /* Log reply sent */
1155         if(m->s.env->cfg->log_replies) {
1156                 log_reply_info(0, &m->s.qinfo, &r->query_reply.addr,
1157                         r->query_reply.addrlen, duration, 0, r_buffer);
1158         }
1159 }
1160
1161 void mesh_query_done(struct mesh_state* mstate)
1162 {
1163         struct mesh_reply* r;
1164         struct mesh_reply* prev = NULL;
1165         struct sldns_buffer* prev_buffer = NULL;
1166         struct mesh_cb* c;
1167         struct reply_info* rep = (mstate->s.return_msg?
1168                 mstate->s.return_msg->rep:NULL);
1169         if((mstate->s.return_rcode == LDNS_RCODE_SERVFAIL ||
1170                 (rep && FLAGS_GET_RCODE(rep->flags) == LDNS_RCODE_SERVFAIL))
1171                 && mstate->s.env->cfg->log_servfail
1172                 && !mstate->s.env->cfg->val_log_squelch) {
1173                 char* err = errinf_to_str_servfail(&mstate->s);
1174                 if(err)
1175                         log_err("%s", err);
1176                 free(err);
1177         }
1178         for(r = mstate->reply_list; r; r = r->next) {
1179                 /* if a response-ip address block has been stored the
1180                  *  information should be logged for each client. */
1181                 if(mstate->s.respip_action_info &&
1182                         mstate->s.respip_action_info->addrinfo) {
1183                         respip_inform_print(mstate->s.respip_action_info->addrinfo,
1184                                 r->qname, mstate->s.qinfo.qtype,
1185                                 mstate->s.qinfo.qclass, r->local_alias,
1186                                 &r->query_reply);
1187                 }
1188
1189                 /* if this query is determined to be dropped during the
1190                  * mesh processing, this is the point to take that action. */
1191                 if(mstate->s.is_drop)
1192                         comm_point_drop_reply(&r->query_reply);
1193                 else {
1194                         struct sldns_buffer* r_buffer = r->query_reply.c->buffer;
1195                         if(r->query_reply.c->tcp_req_info)
1196                                 r_buffer = r->query_reply.c->tcp_req_info->spool_buffer;
1197                         mesh_send_reply(mstate, mstate->s.return_rcode, rep,
1198                                 r, r_buffer, prev, prev_buffer);
1199                         if(r->query_reply.c->tcp_req_info)
1200                                 tcp_req_info_remove_mesh_state(r->query_reply.c->tcp_req_info, mstate);
1201                         prev = r;
1202                         prev_buffer = r_buffer;
1203                 }
1204         }
1205         mstate->replies_sent = 1;
1206         while((c = mstate->cb_list) != NULL) {
1207                 /* take this cb off the list; so that the list can be
1208                  * changed, eg. by adds from the callback routine */
1209                 if(!mstate->reply_list && mstate->cb_list && !c->next) {
1210                         /* was a reply state, not anymore */
1211                         mstate->s.env->mesh->num_reply_states--;
1212                 }
1213                 mstate->cb_list = c->next;
1214                 if(!mstate->reply_list && !mstate->cb_list &&
1215                         mstate->super_set.count == 0)
1216                         mstate->s.env->mesh->num_detached_states++;
1217                 mesh_do_callback(mstate, mstate->s.return_rcode, rep, c);
1218         }
1219 }
1220
1221 void mesh_walk_supers(struct mesh_area* mesh, struct mesh_state* mstate)
1222 {
1223         struct mesh_state_ref* ref;
1224         RBTREE_FOR(ref, struct mesh_state_ref*, &mstate->super_set)
1225         {
1226                 /* make super runnable */
1227                 (void)rbtree_insert(&mesh->run, &ref->s->run_node);
1228                 /* callback the function to inform super of result */
1229                 fptr_ok(fptr_whitelist_mod_inform_super(
1230                         mesh->mods.mod[ref->s->s.curmod]->inform_super));
1231                 (*mesh->mods.mod[ref->s->s.curmod]->inform_super)(&mstate->s,
1232                         ref->s->s.curmod, &ref->s->s);
1233                 /* copy state that is always relevant to super */
1234                 copy_state_to_super(&mstate->s, ref->s->s.curmod, &ref->s->s);
1235         }
1236 }
1237
1238 struct mesh_state* mesh_area_find(struct mesh_area* mesh,
1239         struct respip_client_info* cinfo, struct query_info* qinfo,
1240         uint16_t qflags, int prime, int valrec)
1241 {
1242         struct mesh_state key;
1243         struct mesh_state* result;
1244
1245         key.node.key = &key;
1246         key.s.is_priming = prime;
1247         key.s.is_valrec = valrec;
1248         key.s.qinfo = *qinfo;
1249         key.s.query_flags = qflags;
1250         /* We are searching for a similar mesh state when we DO want to
1251          * aggregate the state. Thus unique is set to NULL. (default when we
1252          * desire aggregation).*/
1253         key.unique = NULL;
1254         key.s.client_info = cinfo;
1255
1256         result = (struct mesh_state*)rbtree_search(&mesh->all, &key);
1257         return result;
1258 }
1259
1260 int mesh_state_add_cb(struct mesh_state* s, struct edns_data* edns,
1261         sldns_buffer* buf, mesh_cb_func_type cb, void* cb_arg,
1262         uint16_t qid, uint16_t qflags)
1263 {
1264         struct mesh_cb* r = regional_alloc(s->s.region,
1265                 sizeof(struct mesh_cb));
1266         if(!r)
1267                 return 0;
1268         r->buf = buf;
1269         log_assert(fptr_whitelist_mesh_cb(cb)); /* early failure ifmissing*/
1270         r->cb = cb;
1271         r->cb_arg = cb_arg;
1272         r->edns = *edns;
1273         if(edns->opt_list) {
1274                 r->edns.opt_list = edns_opt_copy_region(edns->opt_list,
1275                         s->s.region);
1276                 if(!r->edns.opt_list)
1277                         return 0;
1278         }
1279         r->qid = qid;
1280         r->qflags = qflags;
1281         r->next = s->cb_list;
1282         s->cb_list = r;
1283         return 1;
1284
1285 }
1286
1287 int mesh_state_add_reply(struct mesh_state* s, struct edns_data* edns,
1288         struct comm_reply* rep, uint16_t qid, uint16_t qflags,
1289         const struct query_info* qinfo)
1290 {
1291         struct mesh_reply* r = regional_alloc(s->s.region,
1292                 sizeof(struct mesh_reply));
1293         if(!r)
1294                 return 0;
1295         r->query_reply = *rep;
1296         r->edns = *edns;
1297         if(edns->opt_list) {
1298                 r->edns.opt_list = edns_opt_copy_region(edns->opt_list,
1299                         s->s.region);
1300                 if(!r->edns.opt_list)
1301                         return 0;
1302         }
1303         r->qid = qid;
1304         r->qflags = qflags;
1305         r->start_time = *s->s.env->now_tv;
1306         r->next = s->reply_list;
1307         r->qname = regional_alloc_init(s->s.region, qinfo->qname,
1308                 s->s.qinfo.qname_len);
1309         if(!r->qname)
1310                 return 0;
1311
1312         /* Data related to local alias stored in 'qinfo' (if any) is ephemeral
1313          * and can be different for different original queries (even if the
1314          * replaced query name is the same).  So we need to make a deep copy
1315          * and store the copy for each reply info. */
1316         if(qinfo->local_alias) {
1317                 struct packed_rrset_data* d;
1318                 struct packed_rrset_data* dsrc;
1319                 r->local_alias = regional_alloc_zero(s->s.region,
1320                         sizeof(*qinfo->local_alias));
1321                 if(!r->local_alias)
1322                         return 0;
1323                 r->local_alias->rrset = regional_alloc_init(s->s.region,
1324                         qinfo->local_alias->rrset,
1325                         sizeof(*qinfo->local_alias->rrset));
1326                 if(!r->local_alias->rrset)
1327                         return 0;
1328                 dsrc = qinfo->local_alias->rrset->entry.data;
1329
1330                 /* In the current implementation, a local alias must be
1331                  * a single CNAME RR (see worker_handle_request()). */
1332                 log_assert(!qinfo->local_alias->next && dsrc->count == 1 &&
1333                         qinfo->local_alias->rrset->rk.type ==
1334                         htons(LDNS_RR_TYPE_CNAME));
1335                 /* Technically, we should make a local copy for the owner
1336                  * name of the RRset, but in the case of the first (and
1337                  * currently only) local alias RRset, the owner name should
1338                  * point to the qname of the corresponding query, which should
1339                  * be valid throughout the lifetime of this mesh_reply.  So
1340                  * we can skip copying. */
1341                 log_assert(qinfo->local_alias->rrset->rk.dname ==
1342                         sldns_buffer_at(rep->c->buffer, LDNS_HEADER_SIZE));
1343
1344                 d = regional_alloc_init(s->s.region, dsrc,
1345                         sizeof(struct packed_rrset_data)
1346                         + sizeof(size_t) + sizeof(uint8_t*) + sizeof(time_t));
1347                 if(!d)
1348                         return 0;
1349                 r->local_alias->rrset->entry.data = d;
1350                 d->rr_len = (size_t*)((uint8_t*)d +
1351                         sizeof(struct packed_rrset_data));
1352                 d->rr_data = (uint8_t**)&(d->rr_len[1]);
1353                 d->rr_ttl = (time_t*)&(d->rr_data[1]);
1354                 d->rr_len[0] = dsrc->rr_len[0];
1355                 d->rr_ttl[0] = dsrc->rr_ttl[0];
1356                 d->rr_data[0] = regional_alloc_init(s->s.region,
1357                         dsrc->rr_data[0], d->rr_len[0]);
1358                 if(!d->rr_data[0])
1359                         return 0;
1360         } else
1361                 r->local_alias = NULL;
1362
1363         s->reply_list = r;
1364         return 1;
1365 }
1366
1367 /* Extract the query info and flags from 'mstate' into '*qinfop' and '*qflags'.
1368  * Since this is only used for internal refetch of otherwise-expired answer,
1369  * we simply ignore the rare failure mode when memory allocation fails. */
1370 static void
1371 mesh_copy_qinfo(struct mesh_state* mstate, struct query_info** qinfop,
1372         uint16_t* qflags)
1373 {
1374         struct regional* region = mstate->s.env->scratch;
1375         struct query_info* qinfo;
1376
1377         qinfo = regional_alloc_init(region, &mstate->s.qinfo, sizeof(*qinfo));
1378         if(!qinfo)
1379                 return;
1380         qinfo->qname = regional_alloc_init(region, qinfo->qname,
1381                 qinfo->qname_len);
1382         if(!qinfo->qname)
1383                 return;
1384         *qinfop = qinfo;
1385         *qflags = mstate->s.query_flags;
1386 }
1387
1388 /**
1389  * Continue processing the mesh state at another module.
1390  * Handles module to modules transfer of control.
1391  * Handles module finished.
1392  * @param mesh: the mesh area.
1393  * @param mstate: currently active mesh state.
1394  *      Deleted if finished, calls _done and _supers to
1395  *      send replies to clients and inform other mesh states.
1396  *      This in turn may create additional runnable mesh states.
1397  * @param s: state at which the current module exited.
1398  * @param ev: the event sent to the module.
1399  *      returned is the event to send to the next module.
1400  * @return true if continue processing at the new module.
1401  *      false if not continued processing is needed.
1402  */
1403 static int
1404 mesh_continue(struct mesh_area* mesh, struct mesh_state* mstate,
1405         enum module_ext_state s, enum module_ev* ev)
1406 {
1407         mstate->num_activated++;
1408         if(mstate->num_activated > MESH_MAX_ACTIVATION) {
1409                 /* module is looping. Stop it. */
1410                 log_err("internal error: looping module (%s) stopped",
1411                         mesh->mods.mod[mstate->s.curmod]->name);
1412                 log_query_info(0, "pass error for qstate",
1413                         &mstate->s.qinfo);
1414                 s = module_error;
1415         }
1416         if(s == module_wait_module || s == module_restart_next) {
1417                 /* start next module */
1418                 mstate->s.curmod++;
1419                 if(mesh->mods.num == mstate->s.curmod) {
1420                         log_err("Cannot pass to next module; at last module");
1421                         log_query_info(VERB_QUERY, "pass error for qstate",
1422                                 &mstate->s.qinfo);
1423                         mstate->s.curmod--;
1424                         return mesh_continue(mesh, mstate, module_error, ev);
1425                 }
1426                 if(s == module_restart_next) {
1427                         int curmod = mstate->s.curmod;
1428                         for(; mstate->s.curmod < mesh->mods.num;
1429                                 mstate->s.curmod++) {
1430                                 fptr_ok(fptr_whitelist_mod_clear(
1431                                         mesh->mods.mod[mstate->s.curmod]->clear));
1432                                 (*mesh->mods.mod[mstate->s.curmod]->clear)
1433                                         (&mstate->s, mstate->s.curmod);
1434                                 mstate->s.minfo[mstate->s.curmod] = NULL;
1435                         }
1436                         mstate->s.curmod = curmod;
1437                 }
1438                 *ev = module_event_pass;
1439                 return 1;
1440         }
1441         if(s == module_wait_subquery && mstate->sub_set.count == 0) {
1442                 log_err("module cannot wait for subquery, subquery list empty");
1443                 log_query_info(VERB_QUERY, "pass error for qstate",
1444                         &mstate->s.qinfo);
1445                 s = module_error;
1446         }
1447         if(s == module_error && mstate->s.return_rcode == LDNS_RCODE_NOERROR) {
1448                 /* error is bad, handle pass back up below */
1449                 mstate->s.return_rcode = LDNS_RCODE_SERVFAIL;
1450         }
1451         if(s == module_error) {
1452                 mesh_query_done(mstate);
1453                 mesh_walk_supers(mesh, mstate);
1454                 mesh_state_delete(&mstate->s);
1455                 return 0;
1456         }
1457         if(s == module_finished) {
1458                 if(mstate->s.curmod == 0) {
1459                         struct query_info* qinfo = NULL;
1460                         uint16_t qflags;
1461
1462                         mesh_query_done(mstate);
1463                         mesh_walk_supers(mesh, mstate);
1464
1465                         /* If the answer to the query needs to be refetched
1466                          * from an external DNS server, we'll need to schedule
1467                          * a prefetch after removing the current state, so
1468                          * we need to make a copy of the query info here. */
1469                         if(mstate->s.need_refetch)
1470                                 mesh_copy_qinfo(mstate, &qinfo, &qflags);
1471
1472                         mesh_state_delete(&mstate->s);
1473                         if(qinfo) {
1474                                 mesh_schedule_prefetch(mesh, qinfo, qflags,
1475                                         0, 1);
1476                         }
1477                         return 0;
1478                 }
1479                 /* pass along the locus of control */
1480                 mstate->s.curmod --;
1481                 *ev = module_event_moddone;
1482                 return 1;
1483         }
1484         return 0;
1485 }
1486
1487 void mesh_run(struct mesh_area* mesh, struct mesh_state* mstate,
1488         enum module_ev ev, struct outbound_entry* e)
1489 {
1490         enum module_ext_state s;
1491         verbose(VERB_ALGO, "mesh_run: start");
1492         while(mstate) {
1493                 /* run the module */
1494                 fptr_ok(fptr_whitelist_mod_operate(
1495                         mesh->mods.mod[mstate->s.curmod]->operate));
1496                 (*mesh->mods.mod[mstate->s.curmod]->operate)
1497                         (&mstate->s, ev, mstate->s.curmod, e);
1498
1499                 /* examine results */
1500                 mstate->s.reply = NULL;
1501                 regional_free_all(mstate->s.env->scratch);
1502                 s = mstate->s.ext_state[mstate->s.curmod];
1503                 verbose(VERB_ALGO, "mesh_run: %s module exit state is %s",
1504                         mesh->mods.mod[mstate->s.curmod]->name, strextstate(s));
1505                 e = NULL;
1506                 if(mesh_continue(mesh, mstate, s, &ev))
1507                         continue;
1508
1509                 /* run more modules */
1510                 ev = module_event_pass;
1511                 if(mesh->run.count > 0) {
1512                         /* pop random element off the runnable tree */
1513                         mstate = (struct mesh_state*)mesh->run.root->key;
1514                         (void)rbtree_delete(&mesh->run, mstate);
1515                 } else mstate = NULL;
1516         }
1517         if(verbosity >= VERB_ALGO) {
1518                 mesh_stats(mesh, "mesh_run: end");
1519                 mesh_log_list(mesh);
1520         }
1521 }
1522
1523 void
1524 mesh_log_list(struct mesh_area* mesh)
1525 {
1526         char buf[30];
1527         struct mesh_state* m;
1528         int num = 0;
1529         RBTREE_FOR(m, struct mesh_state*, &mesh->all) {
1530                 snprintf(buf, sizeof(buf), "%d%s%s%s%s%s%s mod%d %s%s",
1531                         num++, (m->s.is_priming)?"p":"",  /* prime */
1532                         (m->s.is_valrec)?"v":"",  /* prime */
1533                         (m->s.query_flags&BIT_RD)?"RD":"",
1534                         (m->s.query_flags&BIT_CD)?"CD":"",
1535                         (m->super_set.count==0)?"d":"", /* detached */
1536                         (m->sub_set.count!=0)?"c":"",  /* children */
1537                         m->s.curmod, (m->reply_list)?"rep":"", /*hasreply*/
1538                         (m->cb_list)?"cb":"" /* callbacks */
1539                         );
1540                 log_query_info(VERB_ALGO, buf, &m->s.qinfo);
1541         }
1542 }
1543
1544 void
1545 mesh_stats(struct mesh_area* mesh, const char* str)
1546 {
1547         verbose(VERB_DETAIL, "%s %u recursion states (%u with reply, "
1548                 "%u detached), %u waiting replies, %u recursion replies "
1549                 "sent, %d replies dropped, %d states jostled out",
1550                 str, (unsigned)mesh->all.count,
1551                 (unsigned)mesh->num_reply_states,
1552                 (unsigned)mesh->num_detached_states,
1553                 (unsigned)mesh->num_reply_addrs,
1554                 (unsigned)mesh->replies_sent,
1555                 (unsigned)mesh->stats_dropped,
1556                 (unsigned)mesh->stats_jostled);
1557         if(mesh->replies_sent > 0) {
1558                 struct timeval avg;
1559                 timeval_divide(&avg, &mesh->replies_sum_wait,
1560                         mesh->replies_sent);
1561                 log_info("average recursion processing time "
1562                         ARG_LL "d.%6.6d sec",
1563                         (long long)avg.tv_sec, (int)avg.tv_usec);
1564                 log_info("histogram of recursion processing times");
1565                 timehist_log(mesh->histogram, "recursions");
1566         }
1567 }
1568
1569 void
1570 mesh_stats_clear(struct mesh_area* mesh)
1571 {
1572         if(!mesh)
1573                 return;
1574         mesh->replies_sent = 0;
1575         mesh->replies_sum_wait.tv_sec = 0;
1576         mesh->replies_sum_wait.tv_usec = 0;
1577         mesh->stats_jostled = 0;
1578         mesh->stats_dropped = 0;
1579         timehist_clear(mesh->histogram);
1580         mesh->ans_secure = 0;
1581         mesh->ans_bogus = 0;
1582         memset(&mesh->ans_rcode[0], 0, sizeof(size_t)*16);
1583         mesh->ans_nodata = 0;
1584 }
1585
1586 size_t
1587 mesh_get_mem(struct mesh_area* mesh)
1588 {
1589         struct mesh_state* m;
1590         size_t s = sizeof(*mesh) + sizeof(struct timehist) +
1591                 sizeof(struct th_buck)*mesh->histogram->num +
1592                 sizeof(sldns_buffer) + sldns_buffer_capacity(mesh->qbuf_bak);
1593         RBTREE_FOR(m, struct mesh_state*, &mesh->all) {
1594                 /* all, including m itself allocated in qstate region */
1595                 s += regional_get_mem(m->s.region);
1596         }
1597         return s;
1598 }
1599
1600 int
1601 mesh_detect_cycle(struct module_qstate* qstate, struct query_info* qinfo,
1602         uint16_t flags, int prime, int valrec)
1603 {
1604         struct mesh_area* mesh = qstate->env->mesh;
1605         struct mesh_state* dep_m = NULL;
1606         if(!mesh_state_is_unique(qstate->mesh_info))
1607                 dep_m = mesh_area_find(mesh, NULL, qinfo, flags, prime, valrec);
1608         return mesh_detect_cycle_found(qstate, dep_m);
1609 }
1610
1611 void mesh_list_insert(struct mesh_state* m, struct mesh_state** fp,
1612         struct mesh_state** lp)
1613 {
1614         /* insert as last element */
1615         m->prev = *lp;
1616         m->next = NULL;
1617         if(*lp)
1618                 (*lp)->next = m;
1619         else    *fp = m;
1620         *lp = m;
1621 }
1622
1623 void mesh_list_remove(struct mesh_state* m, struct mesh_state** fp,
1624         struct mesh_state** lp)
1625 {
1626         if(m->next)
1627                 m->next->prev = m->prev;
1628         else    *lp = m->prev;
1629         if(m->prev)
1630                 m->prev->next = m->next;
1631         else    *fp = m->next;
1632 }
1633
1634 void mesh_state_remove_reply(struct mesh_area* mesh, struct mesh_state* m,
1635         struct comm_point* cp)
1636 {
1637         struct mesh_reply* n, *prev = NULL;
1638         n = m->reply_list;
1639         /* when in mesh_cleanup, it sets the reply_list to NULL, so that
1640          * there is no accounting twice */
1641         if(!n) return; /* nothing to remove, also no accounting needed */
1642         while(n) {
1643                 if(n->query_reply.c == cp) {
1644                         /* unlink it */
1645                         if(prev) prev->next = n->next;
1646                         else m->reply_list = n->next;
1647                         /* delete it, but allocated in m region */
1648                         mesh->num_reply_addrs--;
1649
1650                         /* prev = prev; */
1651                         n = n->next;
1652                         continue;
1653                 }
1654                 prev = n;
1655                 n = n->next;
1656         }
1657         /* it was not detached (because it had a reply list), could be now */
1658         if(!m->reply_list && !m->cb_list
1659                 && m->super_set.count == 0) {
1660                 mesh->num_detached_states++;
1661         }
1662         /* if not replies any more in mstate, it is no longer a reply_state */
1663         if(!m->reply_list && !m->cb_list) {
1664                 log_assert(mesh->num_reply_states > 0);
1665                 mesh->num_reply_states--;
1666         }
1667 }