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 65 /** subtract timers and the values do not overflow or become negative */ 66 static void 67 timeval_subtract(struct timeval* d, const struct timeval* end, const struct timeval* start) 68 { 69 #ifndef S_SPLINT_S 70 time_t end_usec = end->tv_usec; 71 d->tv_sec = end->tv_sec - start->tv_sec; 72 if(end_usec < start->tv_usec) { 73 end_usec += 1000000; 74 d->tv_sec--; 75 } 76 d->tv_usec = end_usec - start->tv_usec; 77 #endif 78 } 79 80 /** add timers and the values do not overflow or become negative */ 81 static void 82 timeval_add(struct timeval* d, const struct timeval* add) 83 { 84 #ifndef S_SPLINT_S 85 d->tv_sec += add->tv_sec; 86 d->tv_usec += add->tv_usec; 87 if(d->tv_usec > 1000000 ) { 88 d->tv_usec -= 1000000; 89 d->tv_sec++; 90 } 91 #endif 92 } 93 94 /** divide sum of timers to get average */ 95 static void 96 timeval_divide(struct timeval* avg, const struct timeval* sum, size_t d) 97 { 98 #ifndef S_SPLINT_S 99 size_t leftover; 100 if(d == 0) { 101 avg->tv_sec = 0; 102 avg->tv_usec = 0; 103 return; 104 } 105 avg->tv_sec = sum->tv_sec / d; 106 avg->tv_usec = sum->tv_usec / d; 107 /* handle fraction from seconds divide */ 108 leftover = sum->tv_sec - avg->tv_sec*d; 109 avg->tv_usec += (leftover*1000000)/d; 110 #endif 111 } 112 113 /** histogram compare of time values */ 114 static int 115 timeval_smaller(const struct timeval* x, const struct timeval* y) 116 { 117 #ifndef S_SPLINT_S 118 if(x->tv_sec < y->tv_sec) 119 return 1; 120 else if(x->tv_sec == y->tv_sec) { 121 if(x->tv_usec <= y->tv_usec) 122 return 1; 123 else return 0; 124 } 125 else return 0; 126 #endif 127 } 128 129 /* 130 * Compare two response-ip client info entries for the purpose of mesh state 131 * compare. It returns 0 if ci_a and ci_b are considered equal; otherwise 132 * 1 or -1 (they mean 'ci_a is larger/smaller than ci_b', respectively, but 133 * in practice it should be only used to mean they are different). 134 * We cannot share the mesh state for two queries if different response-ip 135 * actions can apply in the end, even if those queries are otherwise identical. 136 * For this purpose we compare tag lists and tag action lists; they should be 137 * identical to share the same state. 138 * For tag data, we don't look into the data content, as it can be 139 * expensive; unless tag data are not defined for both or they point to the 140 * exact same data in memory (i.e., they come from the same ACL entry), we 141 * consider these data different. 142 * Likewise, if the client info is associated with views, we don't look into 143 * the views. They are considered different unless they are exactly the same 144 * even if the views only differ in the names. 145 */ 146 static int 147 client_info_compare(const struct respip_client_info* ci_a, 148 const struct respip_client_info* ci_b) 149 { 150 int cmp; 151 152 if(!ci_a && !ci_b) 153 return 0; 154 if(ci_a && !ci_b) 155 return -1; 156 if(!ci_a && ci_b) 157 return 1; 158 if(ci_a->taglen != ci_b->taglen) 159 return (ci_a->taglen < ci_b->taglen) ? -1 : 1; 160 cmp = memcmp(ci_a->taglist, ci_b->taglist, ci_a->taglen); 161 if(cmp != 0) 162 return cmp; 163 if(ci_a->tag_actions_size != ci_b->tag_actions_size) 164 return (ci_a->tag_actions_size < ci_b->tag_actions_size) ? 165 -1 : 1; 166 cmp = memcmp(ci_a->tag_actions, ci_b->tag_actions, 167 ci_a->tag_actions_size); 168 if(cmp != 0) 169 return cmp; 170 if(ci_a->tag_datas != ci_b->tag_datas) 171 return ci_a->tag_datas < ci_b->tag_datas ? -1 : 1; 172 if(ci_a->view != ci_b->view) 173 return ci_a->view < ci_b->view ? -1 : 1; 174 /* For the unbound daemon these should be non-NULL and identical, 175 * but we check that just in case. */ 176 if(ci_a->respip_set != ci_b->respip_set) 177 return ci_a->respip_set < ci_b->respip_set ? -1 : 1; 178 return 0; 179 } 180 181 int 182 mesh_state_compare(const void* ap, const void* bp) 183 { 184 struct mesh_state* a = (struct mesh_state*)ap; 185 struct mesh_state* b = (struct mesh_state*)bp; 186 int cmp; 187 188 if(a->unique < b->unique) 189 return -1; 190 if(a->unique > b->unique) 191 return 1; 192 193 if(a->s.is_priming && !b->s.is_priming) 194 return -1; 195 if(!a->s.is_priming && b->s.is_priming) 196 return 1; 197 198 if(a->s.is_valrec && !b->s.is_valrec) 199 return -1; 200 if(!a->s.is_valrec && b->s.is_valrec) 201 return 1; 202 203 if((a->s.query_flags&BIT_RD) && !(b->s.query_flags&BIT_RD)) 204 return -1; 205 if(!(a->s.query_flags&BIT_RD) && (b->s.query_flags&BIT_RD)) 206 return 1; 207 208 if((a->s.query_flags&BIT_CD) && !(b->s.query_flags&BIT_CD)) 209 return -1; 210 if(!(a->s.query_flags&BIT_CD) && (b->s.query_flags&BIT_CD)) 211 return 1; 212 213 cmp = query_info_compare(&a->s.qinfo, &b->s.qinfo); 214 if(cmp != 0) 215 return cmp; 216 return client_info_compare(a->s.client_info, b->s.client_info); 217 } 218 219 int 220 mesh_state_ref_compare(const void* ap, const void* bp) 221 { 222 struct mesh_state_ref* a = (struct mesh_state_ref*)ap; 223 struct mesh_state_ref* b = (struct mesh_state_ref*)bp; 224 return mesh_state_compare(a->s, b->s); 225 } 226 227 struct mesh_area* 228 mesh_create(struct module_stack* stack, struct module_env* env) 229 { 230 struct mesh_area* mesh = calloc(1, sizeof(struct mesh_area)); 231 if(!mesh) { 232 log_err("mesh area alloc: out of memory"); 233 return NULL; 234 } 235 mesh->histogram = timehist_setup(); 236 mesh->qbuf_bak = sldns_buffer_new(env->cfg->msg_buffer_size); 237 if(!mesh->histogram || !mesh->qbuf_bak) { 238 free(mesh); 239 log_err("mesh area alloc: out of memory"); 240 return NULL; 241 } 242 mesh->mods = *stack; 243 mesh->env = env; 244 rbtree_init(&mesh->run, &mesh_state_compare); 245 rbtree_init(&mesh->all, &mesh_state_compare); 246 mesh->num_reply_addrs = 0; 247 mesh->num_reply_states = 0; 248 mesh->num_detached_states = 0; 249 mesh->num_forever_states = 0; 250 mesh->stats_jostled = 0; 251 mesh->stats_dropped = 0; 252 mesh->max_reply_states = env->cfg->num_queries_per_thread; 253 mesh->max_forever_states = (mesh->max_reply_states+1)/2; 254 #ifndef S_SPLINT_S 255 mesh->jostle_max.tv_sec = (time_t)(env->cfg->jostle_time / 1000); 256 mesh->jostle_max.tv_usec = (time_t)((env->cfg->jostle_time % 1000) 257 *1000); 258 #endif 259 return mesh; 260 } 261 262 /** help mesh delete delete mesh states */ 263 static void 264 mesh_delete_helper(rbnode_type* n) 265 { 266 struct mesh_state* mstate = (struct mesh_state*)n->key; 267 /* perform a full delete, not only 'cleanup' routine, 268 * because other callbacks expect a clean state in the mesh. 269 * For 're-entrant' calls */ 270 mesh_state_delete(&mstate->s); 271 /* but because these delete the items from the tree, postorder 272 * traversal and rbtree rebalancing do not work together */ 273 } 274 275 void 276 mesh_delete(struct mesh_area* mesh) 277 { 278 if(!mesh) 279 return; 280 /* free all query states */ 281 while(mesh->all.count) 282 mesh_delete_helper(mesh->all.root); 283 timehist_delete(mesh->histogram); 284 sldns_buffer_free(mesh->qbuf_bak); 285 free(mesh); 286 } 287 288 void 289 mesh_delete_all(struct mesh_area* mesh) 290 { 291 /* free all query states */ 292 while(mesh->all.count) 293 mesh_delete_helper(mesh->all.root); 294 mesh->stats_dropped += mesh->num_reply_addrs; 295 /* clear mesh area references */ 296 rbtree_init(&mesh->run, &mesh_state_compare); 297 rbtree_init(&mesh->all, &mesh_state_compare); 298 mesh->num_reply_addrs = 0; 299 mesh->num_reply_states = 0; 300 mesh->num_detached_states = 0; 301 mesh->num_forever_states = 0; 302 mesh->forever_first = NULL; 303 mesh->forever_last = NULL; 304 mesh->jostle_first = NULL; 305 mesh->jostle_last = NULL; 306 } 307 308 int mesh_make_new_space(struct mesh_area* mesh, sldns_buffer* qbuf) 309 { 310 struct mesh_state* m = mesh->jostle_first; 311 /* free space is available */ 312 if(mesh->num_reply_states < mesh->max_reply_states) 313 return 1; 314 /* try to kick out a jostle-list item */ 315 if(m && m->reply_list && m->list_select == mesh_jostle_list) { 316 /* how old is it? */ 317 struct timeval age; 318 timeval_subtract(&age, mesh->env->now_tv, 319 &m->reply_list->start_time); 320 if(timeval_smaller(&mesh->jostle_max, &age)) { 321 /* its a goner */ 322 log_nametypeclass(VERB_ALGO, "query jostled out to " 323 "make space for a new one", 324 m->s.qinfo.qname, m->s.qinfo.qtype, 325 m->s.qinfo.qclass); 326 /* backup the query */ 327 if(qbuf) sldns_buffer_copy(mesh->qbuf_bak, qbuf); 328 /* notify supers */ 329 if(m->super_set.count > 0) { 330 verbose(VERB_ALGO, "notify supers of failure"); 331 m->s.return_msg = NULL; 332 m->s.return_rcode = LDNS_RCODE_SERVFAIL; 333 mesh_walk_supers(mesh, m); 334 } 335 mesh->stats_jostled ++; 336 mesh_state_delete(&m->s); 337 /* restore the query - note that the qinfo ptr to 338 * the querybuffer is then correct again. */ 339 if(qbuf) sldns_buffer_copy(qbuf, mesh->qbuf_bak); 340 return 1; 341 } 342 } 343 /* no space for new item */ 344 return 0; 345 } 346 347 void mesh_new_client(struct mesh_area* mesh, struct query_info* qinfo, 348 struct respip_client_info* cinfo, uint16_t qflags, 349 struct edns_data* edns, struct comm_reply* rep, uint16_t qid) 350 { 351 struct mesh_state* s = NULL; 352 int unique = unique_mesh_state(edns->opt_list, mesh->env); 353 int was_detached = 0; 354 int was_noreply = 0; 355 int added = 0; 356 if(!unique) 357 s = mesh_area_find(mesh, cinfo, qinfo, qflags&(BIT_RD|BIT_CD), 0, 0); 358 /* does this create a new reply state? */ 359 if(!s || s->list_select == mesh_no_list) { 360 if(!mesh_make_new_space(mesh, rep->c->buffer)) { 361 verbose(VERB_ALGO, "Too many queries. dropping " 362 "incoming query."); 363 comm_point_drop_reply(rep); 364 mesh->stats_dropped ++; 365 return; 366 } 367 /* for this new reply state, the reply address is free, 368 * so the limit of reply addresses does not stop reply states*/ 369 } else { 370 /* protect our memory usage from storing reply addresses */ 371 if(mesh->num_reply_addrs > mesh->max_reply_states*16) { 372 verbose(VERB_ALGO, "Too many requests queued. " 373 "dropping incoming query."); 374 mesh->stats_dropped++; 375 comm_point_drop_reply(rep); 376 return; 377 } 378 } 379 /* see if it already exists, if not, create one */ 380 if(!s) { 381 #ifdef UNBOUND_DEBUG 382 struct rbnode_type* n; 383 #endif 384 s = mesh_state_create(mesh->env, qinfo, cinfo, 385 qflags&(BIT_RD|BIT_CD), 0, 0); 386 if(!s) { 387 log_err("mesh_state_create: out of memory; SERVFAIL"); 388 if(!inplace_cb_reply_servfail_call(mesh->env, qinfo, NULL, NULL, 389 LDNS_RCODE_SERVFAIL, edns, rep, mesh->env->scratch)) 390 edns->opt_list = NULL; 391 error_encode(rep->c->buffer, LDNS_RCODE_SERVFAIL, 392 qinfo, qid, qflags, edns); 393 comm_point_send_reply(rep); 394 return; 395 } 396 if(unique) 397 mesh_state_make_unique(s); 398 /* copy the edns options we got from the front */ 399 if(edns->opt_list) { 400 s->s.edns_opts_front_in = edns_opt_copy_region(edns->opt_list, 401 s->s.region); 402 if(!s->s.edns_opts_front_in) { 403 log_err("mesh_state_create: out of memory; SERVFAIL"); 404 if(!inplace_cb_reply_servfail_call(mesh->env, qinfo, NULL, 405 NULL, LDNS_RCODE_SERVFAIL, edns, rep, mesh->env->scratch)) 406 edns->opt_list = NULL; 407 error_encode(rep->c->buffer, LDNS_RCODE_SERVFAIL, 408 qinfo, qid, qflags, edns); 409 comm_point_send_reply(rep); 410 return; 411 } 412 } 413 414 #ifdef UNBOUND_DEBUG 415 n = 416 #else 417 (void) 418 #endif 419 rbtree_insert(&mesh->all, &s->node); 420 log_assert(n != NULL); 421 /* set detached (it is now) */ 422 mesh->num_detached_states++; 423 added = 1; 424 } 425 if(!s->reply_list && !s->cb_list && s->super_set.count == 0) 426 was_detached = 1; 427 if(!s->reply_list && !s->cb_list) 428 was_noreply = 1; 429 /* add reply to s */ 430 if(!mesh_state_add_reply(s, edns, rep, qid, qflags, qinfo)) { 431 log_err("mesh_new_client: out of memory; SERVFAIL"); 432 if(!inplace_cb_reply_servfail_call(mesh->env, qinfo, &s->s, 433 NULL, LDNS_RCODE_SERVFAIL, edns, rep, mesh->env->scratch)) 434 edns->opt_list = NULL; 435 error_encode(rep->c->buffer, LDNS_RCODE_SERVFAIL, 436 qinfo, qid, qflags, edns); 437 comm_point_send_reply(rep); 438 if(added) 439 mesh_state_delete(&s->s); 440 return; 441 } 442 /* update statistics */ 443 if(was_detached) { 444 log_assert(mesh->num_detached_states > 0); 445 mesh->num_detached_states--; 446 } 447 if(was_noreply) { 448 mesh->num_reply_states ++; 449 } 450 mesh->num_reply_addrs++; 451 if(s->list_select == mesh_no_list) { 452 /* move to either the forever or the jostle_list */ 453 if(mesh->num_forever_states < mesh->max_forever_states) { 454 mesh->num_forever_states ++; 455 mesh_list_insert(s, &mesh->forever_first, 456 &mesh->forever_last); 457 s->list_select = mesh_forever_list; 458 } else { 459 mesh_list_insert(s, &mesh->jostle_first, 460 &mesh->jostle_last); 461 s->list_select = mesh_jostle_list; 462 } 463 } 464 if(added) 465 mesh_run(mesh, s, module_event_new, NULL); 466 } 467 468 int 469 mesh_new_callback(struct mesh_area* mesh, struct query_info* qinfo, 470 uint16_t qflags, struct edns_data* edns, sldns_buffer* buf, 471 uint16_t qid, mesh_cb_func_type cb, void* cb_arg) 472 { 473 struct mesh_state* s = NULL; 474 int unique = unique_mesh_state(edns->opt_list, mesh->env); 475 int was_detached = 0; 476 int was_noreply = 0; 477 int added = 0; 478 if(!unique) 479 s = mesh_area_find(mesh, NULL, qinfo, qflags&(BIT_RD|BIT_CD), 0, 0); 480 481 /* there are no limits on the number of callbacks */ 482 483 /* see if it already exists, if not, create one */ 484 if(!s) { 485 #ifdef UNBOUND_DEBUG 486 struct rbnode_type* n; 487 #endif 488 s = mesh_state_create(mesh->env, qinfo, NULL, 489 qflags&(BIT_RD|BIT_CD), 0, 0); 490 if(!s) { 491 return 0; 492 } 493 if(unique) 494 mesh_state_make_unique(s); 495 if(edns->opt_list) { 496 s->s.edns_opts_front_in = edns_opt_copy_region(edns->opt_list, 497 s->s.region); 498 if(!s->s.edns_opts_front_in) { 499 return 0; 500 } 501 } 502 #ifdef UNBOUND_DEBUG 503 n = 504 #else 505 (void) 506 #endif 507 rbtree_insert(&mesh->all, &s->node); 508 log_assert(n != NULL); 509 /* set detached (it is now) */ 510 mesh->num_detached_states++; 511 added = 1; 512 } 513 if(!s->reply_list && !s->cb_list && s->super_set.count == 0) 514 was_detached = 1; 515 if(!s->reply_list && !s->cb_list) 516 was_noreply = 1; 517 /* add reply to s */ 518 if(!mesh_state_add_cb(s, edns, buf, cb, cb_arg, qid, qflags)) { 519 if(added) 520 mesh_state_delete(&s->s); 521 return 0; 522 } 523 /* update statistics */ 524 if(was_detached) { 525 log_assert(mesh->num_detached_states > 0); 526 mesh->num_detached_states--; 527 } 528 if(was_noreply) { 529 mesh->num_reply_states ++; 530 } 531 mesh->num_reply_addrs++; 532 if(added) 533 mesh_run(mesh, s, module_event_new, NULL); 534 return 1; 535 } 536 537 static void mesh_schedule_prefetch(struct mesh_area* mesh, 538 struct query_info* qinfo, uint16_t qflags, time_t leeway, int run); 539 540 void mesh_new_prefetch(struct mesh_area* mesh, struct query_info* qinfo, 541 uint16_t qflags, time_t leeway) 542 { 543 mesh_schedule_prefetch(mesh, qinfo, qflags, leeway, 1); 544 } 545 546 /* Internal backend routine of mesh_new_prefetch(). It takes one additional 547 * parameter, 'run', which controls whether to run the prefetch state 548 * immediately. When this function is called internally 'run' could be 549 * 0 (false), in which case the new state is only made runnable so it 550 * will not be run recursively on top of the current state. */ 551 static void mesh_schedule_prefetch(struct mesh_area* mesh, 552 struct query_info* qinfo, uint16_t qflags, time_t leeway, int run) 553 { 554 struct mesh_state* s = mesh_area_find(mesh, NULL, qinfo, 555 qflags&(BIT_RD|BIT_CD), 0, 0); 556 #ifdef UNBOUND_DEBUG 557 struct rbnode_type* n; 558 #endif 559 /* already exists, and for a different purpose perhaps. 560 * if mesh_no_list, keep it that way. */ 561 if(s) { 562 /* make it ignore the cache from now on */ 563 if(!s->s.blacklist) 564 sock_list_insert(&s->s.blacklist, NULL, 0, s->s.region); 565 if(s->s.prefetch_leeway < leeway) 566 s->s.prefetch_leeway = leeway; 567 return; 568 } 569 if(!mesh_make_new_space(mesh, NULL)) { 570 verbose(VERB_ALGO, "Too many queries. dropped prefetch."); 571 mesh->stats_dropped ++; 572 return; 573 } 574 575 s = mesh_state_create(mesh->env, qinfo, NULL, 576 qflags&(BIT_RD|BIT_CD), 0, 0); 577 if(!s) { 578 log_err("prefetch mesh_state_create: out of memory"); 579 return; 580 } 581 #ifdef UNBOUND_DEBUG 582 n = 583 #else 584 (void) 585 #endif 586 rbtree_insert(&mesh->all, &s->node); 587 log_assert(n != NULL); 588 /* set detached (it is now) */ 589 mesh->num_detached_states++; 590 /* make it ignore the cache */ 591 sock_list_insert(&s->s.blacklist, NULL, 0, s->s.region); 592 s->s.prefetch_leeway = leeway; 593 594 if(s->list_select == mesh_no_list) { 595 /* move to either the forever or the jostle_list */ 596 if(mesh->num_forever_states < mesh->max_forever_states) { 597 mesh->num_forever_states ++; 598 mesh_list_insert(s, &mesh->forever_first, 599 &mesh->forever_last); 600 s->list_select = mesh_forever_list; 601 } else { 602 mesh_list_insert(s, &mesh->jostle_first, 603 &mesh->jostle_last); 604 s->list_select = mesh_jostle_list; 605 } 606 } 607 608 if(!run) { 609 #ifdef UNBOUND_DEBUG 610 n = 611 #else 612 (void) 613 #endif 614 rbtree_insert(&mesh->run, &s->run_node); 615 log_assert(n != NULL); 616 return; 617 } 618 619 mesh_run(mesh, s, module_event_new, NULL); 620 } 621 622 void mesh_report_reply(struct mesh_area* mesh, struct outbound_entry* e, 623 struct comm_reply* reply, int what) 624 { 625 enum module_ev event = module_event_reply; 626 e->qstate->reply = reply; 627 if(what != NETEVENT_NOERROR) { 628 event = module_event_noreply; 629 if(what == NETEVENT_CAPSFAIL) 630 event = module_event_capsfail; 631 } 632 mesh_run(mesh, e->qstate->mesh_info, event, e); 633 } 634 635 struct mesh_state* 636 mesh_state_create(struct module_env* env, struct query_info* qinfo, 637 struct respip_client_info* cinfo, uint16_t qflags, int prime, 638 int valrec) 639 { 640 struct regional* region = alloc_reg_obtain(env->alloc); 641 struct mesh_state* mstate; 642 int i; 643 if(!region) 644 return NULL; 645 mstate = (struct mesh_state*)regional_alloc(region, 646 sizeof(struct mesh_state)); 647 if(!mstate) { 648 alloc_reg_release(env->alloc, region); 649 return NULL; 650 } 651 memset(mstate, 0, sizeof(*mstate)); 652 mstate->node = *RBTREE_NULL; 653 mstate->run_node = *RBTREE_NULL; 654 mstate->node.key = mstate; 655 mstate->run_node.key = mstate; 656 mstate->reply_list = NULL; 657 mstate->list_select = mesh_no_list; 658 mstate->replies_sent = 0; 659 rbtree_init(&mstate->super_set, &mesh_state_ref_compare); 660 rbtree_init(&mstate->sub_set, &mesh_state_ref_compare); 661 mstate->num_activated = 0; 662 mstate->unique = NULL; 663 /* init module qstate */ 664 mstate->s.qinfo.qtype = qinfo->qtype; 665 mstate->s.qinfo.qclass = qinfo->qclass; 666 mstate->s.qinfo.local_alias = NULL; 667 mstate->s.qinfo.qname_len = qinfo->qname_len; 668 mstate->s.qinfo.qname = regional_alloc_init(region, qinfo->qname, 669 qinfo->qname_len); 670 if(!mstate->s.qinfo.qname) { 671 alloc_reg_release(env->alloc, region); 672 return NULL; 673 } 674 if(cinfo) { 675 mstate->s.client_info = regional_alloc_init(region, cinfo, 676 sizeof(*cinfo)); 677 if(!mstate->s.client_info) { 678 alloc_reg_release(env->alloc, region); 679 return NULL; 680 } 681 } 682 /* remove all weird bits from qflags */ 683 mstate->s.query_flags = (qflags & (BIT_RD|BIT_CD)); 684 mstate->s.is_priming = prime; 685 mstate->s.is_valrec = valrec; 686 mstate->s.reply = NULL; 687 mstate->s.region = region; 688 mstate->s.curmod = 0; 689 mstate->s.return_msg = 0; 690 mstate->s.return_rcode = LDNS_RCODE_NOERROR; 691 mstate->s.env = env; 692 mstate->s.mesh_info = mstate; 693 mstate->s.prefetch_leeway = 0; 694 mstate->s.no_cache_lookup = 0; 695 mstate->s.no_cache_store = 0; 696 mstate->s.need_refetch = 0; 697 mstate->s.was_ratelimited = 0; 698 699 /* init modules */ 700 for(i=0; i<env->mesh->mods.num; i++) { 701 mstate->s.minfo[i] = NULL; 702 mstate->s.ext_state[i] = module_state_initial; 703 } 704 /* init edns option lists */ 705 mstate->s.edns_opts_front_in = NULL; 706 mstate->s.edns_opts_back_out = NULL; 707 mstate->s.edns_opts_back_in = NULL; 708 mstate->s.edns_opts_front_out = NULL; 709 710 return mstate; 711 } 712 713 int 714 mesh_state_is_unique(struct mesh_state* mstate) 715 { 716 return mstate->unique != NULL; 717 } 718 719 void 720 mesh_state_make_unique(struct mesh_state* mstate) 721 { 722 mstate->unique = mstate; 723 } 724 725 void 726 mesh_state_cleanup(struct mesh_state* mstate) 727 { 728 struct mesh_area* mesh; 729 int i; 730 if(!mstate) 731 return; 732 mesh = mstate->s.env->mesh; 733 /* drop unsent replies */ 734 if(!mstate->replies_sent) { 735 struct mesh_reply* rep; 736 struct mesh_cb* cb; 737 for(rep=mstate->reply_list; rep; rep=rep->next) { 738 comm_point_drop_reply(&rep->query_reply); 739 mesh->num_reply_addrs--; 740 } 741 while((cb = mstate->cb_list)!=NULL) { 742 mstate->cb_list = cb->next; 743 fptr_ok(fptr_whitelist_mesh_cb(cb->cb)); 744 (*cb->cb)(cb->cb_arg, LDNS_RCODE_SERVFAIL, NULL, 745 sec_status_unchecked, NULL, 0); 746 mesh->num_reply_addrs--; 747 } 748 } 749 750 /* de-init modules */ 751 for(i=0; i<mesh->mods.num; i++) { 752 fptr_ok(fptr_whitelist_mod_clear(mesh->mods.mod[i]->clear)); 753 (*mesh->mods.mod[i]->clear)(&mstate->s, i); 754 mstate->s.minfo[i] = NULL; 755 mstate->s.ext_state[i] = module_finished; 756 } 757 alloc_reg_release(mstate->s.env->alloc, mstate->s.region); 758 } 759 760 void 761 mesh_state_delete(struct module_qstate* qstate) 762 { 763 struct mesh_area* mesh; 764 struct mesh_state_ref* super, ref; 765 struct mesh_state* mstate; 766 if(!qstate) 767 return; 768 mstate = qstate->mesh_info; 769 mesh = mstate->s.env->mesh; 770 mesh_detach_subs(&mstate->s); 771 if(mstate->list_select == mesh_forever_list) { 772 mesh->num_forever_states --; 773 mesh_list_remove(mstate, &mesh->forever_first, 774 &mesh->forever_last); 775 } else if(mstate->list_select == mesh_jostle_list) { 776 mesh_list_remove(mstate, &mesh->jostle_first, 777 &mesh->jostle_last); 778 } 779 if(!mstate->reply_list && !mstate->cb_list 780 && mstate->super_set.count == 0) { 781 log_assert(mesh->num_detached_states > 0); 782 mesh->num_detached_states--; 783 } 784 if(mstate->reply_list || mstate->cb_list) { 785 log_assert(mesh->num_reply_states > 0); 786 mesh->num_reply_states--; 787 } 788 ref.node.key = &ref; 789 ref.s = mstate; 790 RBTREE_FOR(super, struct mesh_state_ref*, &mstate->super_set) { 791 (void)rbtree_delete(&super->s->sub_set, &ref); 792 } 793 (void)rbtree_delete(&mesh->run, mstate); 794 (void)rbtree_delete(&mesh->all, mstate); 795 mesh_state_cleanup(mstate); 796 } 797 798 /** helper recursive rbtree find routine */ 799 static int 800 find_in_subsub(struct mesh_state* m, struct mesh_state* tofind, size_t *c) 801 { 802 struct mesh_state_ref* r; 803 if((*c)++ > MESH_MAX_SUBSUB) 804 return 1; 805 RBTREE_FOR(r, struct mesh_state_ref*, &m->sub_set) { 806 if(r->s == tofind || find_in_subsub(r->s, tofind, c)) 807 return 1; 808 } 809 return 0; 810 } 811 812 /** find cycle for already looked up mesh_state */ 813 static int 814 mesh_detect_cycle_found(struct module_qstate* qstate, struct mesh_state* dep_m) 815 { 816 struct mesh_state* cyc_m = qstate->mesh_info; 817 size_t counter = 0; 818 if(!dep_m) 819 return 0; 820 if(dep_m == cyc_m || find_in_subsub(dep_m, cyc_m, &counter)) { 821 if(counter > MESH_MAX_SUBSUB) 822 return 2; 823 return 1; 824 } 825 return 0; 826 } 827 828 void mesh_detach_subs(struct module_qstate* qstate) 829 { 830 struct mesh_area* mesh = qstate->env->mesh; 831 struct mesh_state_ref* ref, lookup; 832 #ifdef UNBOUND_DEBUG 833 struct rbnode_type* n; 834 #endif 835 lookup.node.key = &lookup; 836 lookup.s = qstate->mesh_info; 837 RBTREE_FOR(ref, struct mesh_state_ref*, &qstate->mesh_info->sub_set) { 838 #ifdef UNBOUND_DEBUG 839 n = 840 #else 841 (void) 842 #endif 843 rbtree_delete(&ref->s->super_set, &lookup); 844 log_assert(n != NULL); /* must have been present */ 845 if(!ref->s->reply_list && !ref->s->cb_list 846 && ref->s->super_set.count == 0) { 847 mesh->num_detached_states++; 848 log_assert(mesh->num_detached_states + 849 mesh->num_reply_states <= mesh->all.count); 850 } 851 } 852 rbtree_init(&qstate->mesh_info->sub_set, &mesh_state_ref_compare); 853 } 854 855 int mesh_add_sub(struct module_qstate* qstate, struct query_info* qinfo, 856 uint16_t qflags, int prime, int valrec, struct module_qstate** newq, 857 struct mesh_state** sub) 858 { 859 /* find it, if not, create it */ 860 struct mesh_area* mesh = qstate->env->mesh; 861 *sub = mesh_area_find(mesh, NULL, qinfo, qflags, 862 prime, valrec); 863 if(mesh_detect_cycle_found(qstate, *sub)) { 864 verbose(VERB_ALGO, "attach failed, cycle detected"); 865 return 0; 866 } 867 if(!*sub) { 868 #ifdef UNBOUND_DEBUG 869 struct rbnode_type* n; 870 #endif 871 /* create a new one */ 872 *sub = mesh_state_create(qstate->env, qinfo, NULL, qflags, prime, 873 valrec); 874 if(!*sub) { 875 log_err("mesh_attach_sub: out of memory"); 876 return 0; 877 } 878 #ifdef UNBOUND_DEBUG 879 n = 880 #else 881 (void) 882 #endif 883 rbtree_insert(&mesh->all, &(*sub)->node); 884 log_assert(n != NULL); 885 /* set detached (it is now) */ 886 mesh->num_detached_states++; 887 /* set new query state to run */ 888 #ifdef UNBOUND_DEBUG 889 n = 890 #else 891 (void) 892 #endif 893 rbtree_insert(&mesh->run, &(*sub)->run_node); 894 log_assert(n != NULL); 895 *newq = &(*sub)->s; 896 } else 897 *newq = NULL; 898 return 1; 899 } 900 901 int mesh_attach_sub(struct module_qstate* qstate, struct query_info* qinfo, 902 uint16_t qflags, int prime, int valrec, struct module_qstate** newq) 903 { 904 struct mesh_area* mesh = qstate->env->mesh; 905 struct mesh_state* sub = NULL; 906 int was_detached; 907 if(!mesh_add_sub(qstate, qinfo, qflags, prime, valrec, newq, &sub)) 908 return 0; 909 was_detached = (sub->super_set.count == 0); 910 if(!mesh_state_attachment(qstate->mesh_info, sub)) 911 return 0; 912 /* if it was a duplicate attachment, the count was not zero before */ 913 if(!sub->reply_list && !sub->cb_list && was_detached && 914 sub->super_set.count == 1) { 915 /* it used to be detached, before this one got added */ 916 log_assert(mesh->num_detached_states > 0); 917 mesh->num_detached_states--; 918 } 919 /* *newq will be run when inited after the current module stops */ 920 return 1; 921 } 922 923 int mesh_state_attachment(struct mesh_state* super, struct mesh_state* sub) 924 { 925 #ifdef UNBOUND_DEBUG 926 struct rbnode_type* n; 927 #endif 928 struct mesh_state_ref* subref; /* points to sub, inserted in super */ 929 struct mesh_state_ref* superref; /* points to super, inserted in sub */ 930 if( !(subref = regional_alloc(super->s.region, 931 sizeof(struct mesh_state_ref))) || 932 !(superref = regional_alloc(sub->s.region, 933 sizeof(struct mesh_state_ref))) ) { 934 log_err("mesh_state_attachment: out of memory"); 935 return 0; 936 } 937 superref->node.key = superref; 938 superref->s = super; 939 subref->node.key = subref; 940 subref->s = sub; 941 if(!rbtree_insert(&sub->super_set, &superref->node)) { 942 /* this should not happen, iterator and validator do not 943 * attach subqueries that are identical. */ 944 /* already attached, we are done, nothing todo. 945 * since superref and subref already allocated in region, 946 * we cannot free them */ 947 return 1; 948 } 949 #ifdef UNBOUND_DEBUG 950 n = 951 #else 952 (void) 953 #endif 954 rbtree_insert(&super->sub_set, &subref->node); 955 log_assert(n != NULL); /* we checked above if statement, the reverse 956 administration should not fail now, unless they are out of sync */ 957 return 1; 958 } 959 960 /** 961 * callback results to mesh cb entry 962 * @param m: mesh state to send it for. 963 * @param rcode: if not 0, error code. 964 * @param rep: reply to send (or NULL if rcode is set). 965 * @param r: callback entry 966 */ 967 static void 968 mesh_do_callback(struct mesh_state* m, int rcode, struct reply_info* rep, 969 struct mesh_cb* r) 970 { 971 int secure; 972 char* reason = NULL; 973 int was_ratelimited = m->s.was_ratelimited; 974 /* bogus messages are not made into servfail, sec_status passed 975 * to the callback function */ 976 if(rep && rep->security == sec_status_secure) 977 secure = 1; 978 else secure = 0; 979 if(!rep && rcode == LDNS_RCODE_NOERROR) 980 rcode = LDNS_RCODE_SERVFAIL; 981 if(!rcode && (rep->security == sec_status_bogus || 982 rep->security == sec_status_secure_sentinel_fail)) { 983 if(!(reason = errinf_to_str_bogus(&m->s))) 984 rcode = LDNS_RCODE_SERVFAIL; 985 } 986 /* send the reply */ 987 if(rcode) { 988 if(rcode == LDNS_RCODE_SERVFAIL) { 989 if(!inplace_cb_reply_servfail_call(m->s.env, &m->s.qinfo, &m->s, 990 rep, rcode, &r->edns, NULL, m->s.region)) 991 r->edns.opt_list = NULL; 992 } else { 993 if(!inplace_cb_reply_call(m->s.env, &m->s.qinfo, &m->s, rep, rcode, 994 &r->edns, NULL, m->s.region)) 995 r->edns.opt_list = NULL; 996 } 997 fptr_ok(fptr_whitelist_mesh_cb(r->cb)); 998 (*r->cb)(r->cb_arg, rcode, r->buf, sec_status_unchecked, NULL, 999 was_ratelimited); 1000 } else { 1001 size_t udp_size = r->edns.udp_size; 1002 sldns_buffer_clear(r->buf); 1003 r->edns.edns_version = EDNS_ADVERTISED_VERSION; 1004 r->edns.udp_size = EDNS_ADVERTISED_SIZE; 1005 r->edns.ext_rcode = 0; 1006 r->edns.bits &= EDNS_DO; 1007 1008 if(!inplace_cb_reply_call(m->s.env, &m->s.qinfo, &m->s, rep, 1009 LDNS_RCODE_NOERROR, &r->edns, NULL, m->s.region) || 1010 !reply_info_answer_encode(&m->s.qinfo, rep, r->qid, 1011 r->qflags, r->buf, 0, 1, 1012 m->s.env->scratch, udp_size, &r->edns, 1013 (int)(r->edns.bits & EDNS_DO), secure)) 1014 { 1015 fptr_ok(fptr_whitelist_mesh_cb(r->cb)); 1016 (*r->cb)(r->cb_arg, LDNS_RCODE_SERVFAIL, r->buf, 1017 sec_status_unchecked, NULL, 0); 1018 } else { 1019 fptr_ok(fptr_whitelist_mesh_cb(r->cb)); 1020 (*r->cb)(r->cb_arg, LDNS_RCODE_NOERROR, r->buf, 1021 rep->security, reason, was_ratelimited); 1022 } 1023 } 1024 free(reason); 1025 m->s.env->mesh->num_reply_addrs--; 1026 } 1027 1028 /** 1029 * Send reply to mesh reply entry 1030 * @param m: mesh state to send it for. 1031 * @param rcode: if not 0, error code. 1032 * @param rep: reply to send (or NULL if rcode is set). 1033 * @param r: reply entry 1034 * @param prev: previous reply, already has its answer encoded in buffer. 1035 */ 1036 static void 1037 mesh_send_reply(struct mesh_state* m, int rcode, struct reply_info* rep, 1038 struct mesh_reply* r, struct mesh_reply* prev) 1039 { 1040 struct timeval end_time; 1041 struct timeval duration; 1042 int secure; 1043 /* Copy the client's EDNS for later restore, to make sure the edns 1044 * compare is with the correct edns options. */ 1045 struct edns_data edns_bak = r->edns; 1046 /* examine security status */ 1047 if(m->s.env->need_to_validate && (!(r->qflags&BIT_CD) || 1048 m->s.env->cfg->ignore_cd) && rep && 1049 (rep->security <= sec_status_bogus || 1050 rep->security == sec_status_secure_sentinel_fail)) { 1051 rcode = LDNS_RCODE_SERVFAIL; 1052 if(m->s.env->cfg->stat_extended) 1053 m->s.env->mesh->ans_bogus++; 1054 } 1055 if(rep && rep->security == sec_status_secure) 1056 secure = 1; 1057 else secure = 0; 1058 if(!rep && rcode == LDNS_RCODE_NOERROR) 1059 rcode = LDNS_RCODE_SERVFAIL; 1060 /* send the reply */ 1061 /* We don't reuse the encoded answer if either the previous or current 1062 * response has a local alias. We could compare the alias records 1063 * and still reuse the previous answer if they are the same, but that 1064 * would be complicated and error prone for the relatively minor case. 1065 * So we err on the side of safety. */ 1066 if(prev && prev->qflags == r->qflags && 1067 !prev->local_alias && !r->local_alias && 1068 prev->edns.edns_present == r->edns.edns_present && 1069 prev->edns.bits == r->edns.bits && 1070 prev->edns.udp_size == r->edns.udp_size && 1071 edns_opt_list_compare(prev->edns.opt_list, r->edns.opt_list) 1072 == 0) { 1073 /* if the previous reply is identical to this one, fix ID */ 1074 if(prev->query_reply.c->buffer != r->query_reply.c->buffer) 1075 sldns_buffer_copy(r->query_reply.c->buffer, 1076 prev->query_reply.c->buffer); 1077 sldns_buffer_write_at(r->query_reply.c->buffer, 0, 1078 &r->qid, sizeof(uint16_t)); 1079 sldns_buffer_write_at(r->query_reply.c->buffer, 12, 1080 r->qname, m->s.qinfo.qname_len); 1081 comm_point_send_reply(&r->query_reply); 1082 } else if(rcode) { 1083 m->s.qinfo.qname = r->qname; 1084 m->s.qinfo.local_alias = r->local_alias; 1085 if(rcode == LDNS_RCODE_SERVFAIL) { 1086 if(!inplace_cb_reply_servfail_call(m->s.env, &m->s.qinfo, &m->s, 1087 rep, rcode, &r->edns, NULL, m->s.region)) 1088 r->edns.opt_list = NULL; 1089 } else { 1090 if(!inplace_cb_reply_call(m->s.env, &m->s.qinfo, &m->s, rep, rcode, 1091 &r->edns, NULL, m->s.region)) 1092 r->edns.opt_list = NULL; 1093 } 1094 error_encode(r->query_reply.c->buffer, rcode, &m->s.qinfo, 1095 r->qid, r->qflags, &r->edns); 1096 comm_point_send_reply(&r->query_reply); 1097 } else { 1098 size_t udp_size = r->edns.udp_size; 1099 r->edns.edns_version = EDNS_ADVERTISED_VERSION; 1100 r->edns.udp_size = EDNS_ADVERTISED_SIZE; 1101 r->edns.ext_rcode = 0; 1102 r->edns.bits &= EDNS_DO; 1103 m->s.qinfo.qname = r->qname; 1104 m->s.qinfo.local_alias = r->local_alias; 1105 if(!inplace_cb_reply_call(m->s.env, &m->s.qinfo, &m->s, rep, 1106 LDNS_RCODE_NOERROR, &r->edns, NULL, m->s.region) || 1107 !apply_edns_options(&r->edns, &edns_bak, 1108 m->s.env->cfg, r->query_reply.c, 1109 m->s.region) || 1110 !reply_info_answer_encode(&m->s.qinfo, rep, r->qid, 1111 r->qflags, r->query_reply.c->buffer, 0, 1, 1112 m->s.env->scratch, udp_size, &r->edns, 1113 (int)(r->edns.bits & EDNS_DO), secure)) 1114 { 1115 if(!inplace_cb_reply_servfail_call(m->s.env, &m->s.qinfo, &m->s, 1116 rep, LDNS_RCODE_SERVFAIL, &r->edns, NULL, m->s.region)) 1117 r->edns.opt_list = NULL; 1118 error_encode(r->query_reply.c->buffer, 1119 LDNS_RCODE_SERVFAIL, &m->s.qinfo, r->qid, 1120 r->qflags, &r->edns); 1121 } 1122 r->edns = edns_bak; 1123 comm_point_send_reply(&r->query_reply); 1124 } 1125 /* account */ 1126 m->s.env->mesh->num_reply_addrs--; 1127 end_time = *m->s.env->now_tv; 1128 timeval_subtract(&duration, &end_time, &r->start_time); 1129 verbose(VERB_ALGO, "query took " ARG_LL "d.%6.6d sec", 1130 (long long)duration.tv_sec, (int)duration.tv_usec); 1131 m->s.env->mesh->replies_sent++; 1132 timeval_add(&m->s.env->mesh->replies_sum_wait, &duration); 1133 timehist_insert(m->s.env->mesh->histogram, &duration); 1134 if(m->s.env->cfg->stat_extended) { 1135 uint16_t rc = FLAGS_GET_RCODE(sldns_buffer_read_u16_at(r-> 1136 query_reply.c->buffer, 2)); 1137 if(secure) m->s.env->mesh->ans_secure++; 1138 m->s.env->mesh->ans_rcode[ rc ] ++; 1139 if(rc == 0 && LDNS_ANCOUNT(sldns_buffer_begin(r-> 1140 query_reply.c->buffer)) == 0) 1141 m->s.env->mesh->ans_nodata++; 1142 } 1143 /* Log reply sent */ 1144 if(m->s.env->cfg->log_replies) { 1145 log_reply_info(0, &m->s.qinfo, &r->query_reply.addr, 1146 r->query_reply.addrlen, duration, 0, 1147 r->query_reply.c->buffer); 1148 } 1149 } 1150 1151 void mesh_query_done(struct mesh_state* mstate) 1152 { 1153 struct mesh_reply* r; 1154 struct mesh_reply* prev = NULL; 1155 struct mesh_cb* c; 1156 struct reply_info* rep = (mstate->s.return_msg? 1157 mstate->s.return_msg->rep:NULL); 1158 if((mstate->s.return_rcode == LDNS_RCODE_SERVFAIL || 1159 (rep && FLAGS_GET_RCODE(rep->flags) == LDNS_RCODE_SERVFAIL)) 1160 && mstate->s.env->cfg->log_servfail 1161 && !mstate->s.env->cfg->val_log_squelch) { 1162 char* err = errinf_to_str_servfail(&mstate->s); 1163 if(err) 1164 log_err("%s", err); 1165 free(err); 1166 } 1167 for(r = mstate->reply_list; r; r = r->next) { 1168 /* if a response-ip address block has been stored the 1169 * information should be logged for each client. */ 1170 if(mstate->s.respip_action_info && 1171 mstate->s.respip_action_info->addrinfo) { 1172 respip_inform_print(mstate->s.respip_action_info->addrinfo, 1173 r->qname, mstate->s.qinfo.qtype, 1174 mstate->s.qinfo.qclass, r->local_alias, 1175 &r->query_reply); 1176 } 1177 1178 /* if this query is determined to be dropped during the 1179 * mesh processing, this is the point to take that action. */ 1180 if(mstate->s.is_drop) 1181 comm_point_drop_reply(&r->query_reply); 1182 else { 1183 mesh_send_reply(mstate, mstate->s.return_rcode, rep, 1184 r, prev); 1185 prev = r; 1186 } 1187 } 1188 mstate->replies_sent = 1; 1189 while((c = mstate->cb_list) != NULL) { 1190 /* take this cb off the list; so that the list can be 1191 * changed, eg. by adds from the callback routine */ 1192 if(!mstate->reply_list && mstate->cb_list && !c->next) { 1193 /* was a reply state, not anymore */ 1194 mstate->s.env->mesh->num_reply_states--; 1195 } 1196 mstate->cb_list = c->next; 1197 if(!mstate->reply_list && !mstate->cb_list && 1198 mstate->super_set.count == 0) 1199 mstate->s.env->mesh->num_detached_states++; 1200 mesh_do_callback(mstate, mstate->s.return_rcode, rep, c); 1201 } 1202 } 1203 1204 void mesh_walk_supers(struct mesh_area* mesh, struct mesh_state* mstate) 1205 { 1206 struct mesh_state_ref* ref; 1207 RBTREE_FOR(ref, struct mesh_state_ref*, &mstate->super_set) 1208 { 1209 /* make super runnable */ 1210 (void)rbtree_insert(&mesh->run, &ref->s->run_node); 1211 /* callback the function to inform super of result */ 1212 fptr_ok(fptr_whitelist_mod_inform_super( 1213 mesh->mods.mod[ref->s->s.curmod]->inform_super)); 1214 (*mesh->mods.mod[ref->s->s.curmod]->inform_super)(&mstate->s, 1215 ref->s->s.curmod, &ref->s->s); 1216 /* copy state that is always relevant to super */ 1217 copy_state_to_super(&mstate->s, ref->s->s.curmod, &ref->s->s); 1218 } 1219 } 1220 1221 struct mesh_state* mesh_area_find(struct mesh_area* mesh, 1222 struct respip_client_info* cinfo, struct query_info* qinfo, 1223 uint16_t qflags, int prime, int valrec) 1224 { 1225 struct mesh_state key; 1226 struct mesh_state* result; 1227 1228 key.node.key = &key; 1229 key.s.is_priming = prime; 1230 key.s.is_valrec = valrec; 1231 key.s.qinfo = *qinfo; 1232 key.s.query_flags = qflags; 1233 /* We are searching for a similar mesh state when we DO want to 1234 * aggregate the state. Thus unique is set to NULL. (default when we 1235 * desire aggregation).*/ 1236 key.unique = NULL; 1237 key.s.client_info = cinfo; 1238 1239 result = (struct mesh_state*)rbtree_search(&mesh->all, &key); 1240 return result; 1241 } 1242 1243 int mesh_state_add_cb(struct mesh_state* s, struct edns_data* edns, 1244 sldns_buffer* buf, mesh_cb_func_type cb, void* cb_arg, 1245 uint16_t qid, uint16_t qflags) 1246 { 1247 struct mesh_cb* r = regional_alloc(s->s.region, 1248 sizeof(struct mesh_cb)); 1249 if(!r) 1250 return 0; 1251 r->buf = buf; 1252 log_assert(fptr_whitelist_mesh_cb(cb)); /* early failure ifmissing*/ 1253 r->cb = cb; 1254 r->cb_arg = cb_arg; 1255 r->edns = *edns; 1256 if(edns->opt_list) { 1257 r->edns.opt_list = edns_opt_copy_region(edns->opt_list, 1258 s->s.region); 1259 if(!r->edns.opt_list) 1260 return 0; 1261 } 1262 r->qid = qid; 1263 r->qflags = qflags; 1264 r->next = s->cb_list; 1265 s->cb_list = r; 1266 return 1; 1267 1268 } 1269 1270 int mesh_state_add_reply(struct mesh_state* s, struct edns_data* edns, 1271 struct comm_reply* rep, uint16_t qid, uint16_t qflags, 1272 const struct query_info* qinfo) 1273 { 1274 struct mesh_reply* r = regional_alloc(s->s.region, 1275 sizeof(struct mesh_reply)); 1276 if(!r) 1277 return 0; 1278 r->query_reply = *rep; 1279 r->edns = *edns; 1280 if(edns->opt_list) { 1281 r->edns.opt_list = edns_opt_copy_region(edns->opt_list, 1282 s->s.region); 1283 if(!r->edns.opt_list) 1284 return 0; 1285 } 1286 r->qid = qid; 1287 r->qflags = qflags; 1288 r->start_time = *s->s.env->now_tv; 1289 r->next = s->reply_list; 1290 r->qname = regional_alloc_init(s->s.region, qinfo->qname, 1291 s->s.qinfo.qname_len); 1292 if(!r->qname) 1293 return 0; 1294 1295 /* Data related to local alias stored in 'qinfo' (if any) is ephemeral 1296 * and can be different for different original queries (even if the 1297 * replaced query name is the same). So we need to make a deep copy 1298 * and store the copy for each reply info. */ 1299 if(qinfo->local_alias) { 1300 struct packed_rrset_data* d; 1301 struct packed_rrset_data* dsrc; 1302 r->local_alias = regional_alloc_zero(s->s.region, 1303 sizeof(*qinfo->local_alias)); 1304 if(!r->local_alias) 1305 return 0; 1306 r->local_alias->rrset = regional_alloc_init(s->s.region, 1307 qinfo->local_alias->rrset, 1308 sizeof(*qinfo->local_alias->rrset)); 1309 if(!r->local_alias->rrset) 1310 return 0; 1311 dsrc = qinfo->local_alias->rrset->entry.data; 1312 1313 /* In the current implementation, a local alias must be 1314 * a single CNAME RR (see worker_handle_request()). */ 1315 log_assert(!qinfo->local_alias->next && dsrc->count == 1 && 1316 qinfo->local_alias->rrset->rk.type == 1317 htons(LDNS_RR_TYPE_CNAME)); 1318 /* Technically, we should make a local copy for the owner 1319 * name of the RRset, but in the case of the first (and 1320 * currently only) local alias RRset, the owner name should 1321 * point to the qname of the corresponding query, which should 1322 * be valid throughout the lifetime of this mesh_reply. So 1323 * we can skip copying. */ 1324 log_assert(qinfo->local_alias->rrset->rk.dname == 1325 sldns_buffer_at(rep->c->buffer, LDNS_HEADER_SIZE)); 1326 1327 d = regional_alloc_init(s->s.region, dsrc, 1328 sizeof(struct packed_rrset_data) 1329 + sizeof(size_t) + sizeof(uint8_t*) + sizeof(time_t)); 1330 if(!d) 1331 return 0; 1332 r->local_alias->rrset->entry.data = d; 1333 d->rr_len = (size_t*)((uint8_t*)d + 1334 sizeof(struct packed_rrset_data)); 1335 d->rr_data = (uint8_t**)&(d->rr_len[1]); 1336 d->rr_ttl = (time_t*)&(d->rr_data[1]); 1337 d->rr_len[0] = dsrc->rr_len[0]; 1338 d->rr_ttl[0] = dsrc->rr_ttl[0]; 1339 d->rr_data[0] = regional_alloc_init(s->s.region, 1340 dsrc->rr_data[0], d->rr_len[0]); 1341 if(!d->rr_data[0]) 1342 return 0; 1343 } else 1344 r->local_alias = NULL; 1345 1346 s->reply_list = r; 1347 return 1; 1348 } 1349 1350 /* Extract the query info and flags from 'mstate' into '*qinfop' and '*qflags'. 1351 * Since this is only used for internal refetch of otherwise-expired answer, 1352 * we simply ignore the rare failure mode when memory allocation fails. */ 1353 static void 1354 mesh_copy_qinfo(struct mesh_state* mstate, struct query_info** qinfop, 1355 uint16_t* qflags) 1356 { 1357 struct regional* region = mstate->s.env->scratch; 1358 struct query_info* qinfo; 1359 1360 qinfo = regional_alloc_init(region, &mstate->s.qinfo, sizeof(*qinfo)); 1361 if(!qinfo) 1362 return; 1363 qinfo->qname = regional_alloc_init(region, qinfo->qname, 1364 qinfo->qname_len); 1365 if(!qinfo->qname) 1366 return; 1367 *qinfop = qinfo; 1368 *qflags = mstate->s.query_flags; 1369 } 1370 1371 /** 1372 * Continue processing the mesh state at another module. 1373 * Handles module to modules transfer of control. 1374 * Handles module finished. 1375 * @param mesh: the mesh area. 1376 * @param mstate: currently active mesh state. 1377 * Deleted if finished, calls _done and _supers to 1378 * send replies to clients and inform other mesh states. 1379 * This in turn may create additional runnable mesh states. 1380 * @param s: state at which the current module exited. 1381 * @param ev: the event sent to the module. 1382 * returned is the event to send to the next module. 1383 * @return true if continue processing at the new module. 1384 * false if not continued processing is needed. 1385 */ 1386 static int 1387 mesh_continue(struct mesh_area* mesh, struct mesh_state* mstate, 1388 enum module_ext_state s, enum module_ev* ev) 1389 { 1390 mstate->num_activated++; 1391 if(mstate->num_activated > MESH_MAX_ACTIVATION) { 1392 /* module is looping. Stop it. */ 1393 log_err("internal error: looping module (%s) stopped", 1394 mesh->mods.mod[mstate->s.curmod]->name); 1395 log_query_info(VERB_QUERY, "pass error for qstate", 1396 &mstate->s.qinfo); 1397 s = module_error; 1398 } 1399 if(s == module_wait_module || s == module_restart_next) { 1400 /* start next module */ 1401 mstate->s.curmod++; 1402 if(mesh->mods.num == mstate->s.curmod) { 1403 log_err("Cannot pass to next module; at last module"); 1404 log_query_info(VERB_QUERY, "pass error for qstate", 1405 &mstate->s.qinfo); 1406 mstate->s.curmod--; 1407 return mesh_continue(mesh, mstate, module_error, ev); 1408 } 1409 if(s == module_restart_next) { 1410 int curmod = mstate->s.curmod; 1411 for(; mstate->s.curmod < mesh->mods.num; 1412 mstate->s.curmod++) { 1413 fptr_ok(fptr_whitelist_mod_clear( 1414 mesh->mods.mod[mstate->s.curmod]->clear)); 1415 (*mesh->mods.mod[mstate->s.curmod]->clear) 1416 (&mstate->s, mstate->s.curmod); 1417 mstate->s.minfo[mstate->s.curmod] = NULL; 1418 } 1419 mstate->s.curmod = curmod; 1420 } 1421 *ev = module_event_pass; 1422 return 1; 1423 } 1424 if(s == module_wait_subquery && mstate->sub_set.count == 0) { 1425 log_err("module cannot wait for subquery, subquery list empty"); 1426 log_query_info(VERB_QUERY, "pass error for qstate", 1427 &mstate->s.qinfo); 1428 s = module_error; 1429 } 1430 if(s == module_error && mstate->s.return_rcode == LDNS_RCODE_NOERROR) { 1431 /* error is bad, handle pass back up below */ 1432 mstate->s.return_rcode = LDNS_RCODE_SERVFAIL; 1433 } 1434 if(s == module_error) { 1435 mesh_query_done(mstate); 1436 mesh_walk_supers(mesh, mstate); 1437 mesh_state_delete(&mstate->s); 1438 return 0; 1439 } 1440 if(s == module_finished) { 1441 if(mstate->s.curmod == 0) { 1442 struct query_info* qinfo = NULL; 1443 uint16_t qflags; 1444 1445 mesh_query_done(mstate); 1446 mesh_walk_supers(mesh, mstate); 1447 1448 /* If the answer to the query needs to be refetched 1449 * from an external DNS server, we'll need to schedule 1450 * a prefetch after removing the current state, so 1451 * we need to make a copy of the query info here. */ 1452 if(mstate->s.need_refetch) 1453 mesh_copy_qinfo(mstate, &qinfo, &qflags); 1454 1455 mesh_state_delete(&mstate->s); 1456 if(qinfo) { 1457 mesh_schedule_prefetch(mesh, qinfo, qflags, 1458 0, 1); 1459 } 1460 return 0; 1461 } 1462 /* pass along the locus of control */ 1463 mstate->s.curmod --; 1464 *ev = module_event_moddone; 1465 return 1; 1466 } 1467 return 0; 1468 } 1469 1470 void mesh_run(struct mesh_area* mesh, struct mesh_state* mstate, 1471 enum module_ev ev, struct outbound_entry* e) 1472 { 1473 enum module_ext_state s; 1474 verbose(VERB_ALGO, "mesh_run: start"); 1475 while(mstate) { 1476 /* run the module */ 1477 fptr_ok(fptr_whitelist_mod_operate( 1478 mesh->mods.mod[mstate->s.curmod]->operate)); 1479 (*mesh->mods.mod[mstate->s.curmod]->operate) 1480 (&mstate->s, ev, mstate->s.curmod, e); 1481 1482 /* examine results */ 1483 mstate->s.reply = NULL; 1484 regional_free_all(mstate->s.env->scratch); 1485 s = mstate->s.ext_state[mstate->s.curmod]; 1486 verbose(VERB_ALGO, "mesh_run: %s module exit state is %s", 1487 mesh->mods.mod[mstate->s.curmod]->name, strextstate(s)); 1488 e = NULL; 1489 if(mesh_continue(mesh, mstate, s, &ev)) 1490 continue; 1491 1492 /* run more modules */ 1493 ev = module_event_pass; 1494 if(mesh->run.count > 0) { 1495 /* pop random element off the runnable tree */ 1496 mstate = (struct mesh_state*)mesh->run.root->key; 1497 (void)rbtree_delete(&mesh->run, mstate); 1498 } else mstate = NULL; 1499 } 1500 if(verbosity >= VERB_ALGO) { 1501 mesh_stats(mesh, "mesh_run: end"); 1502 mesh_log_list(mesh); 1503 } 1504 } 1505 1506 void 1507 mesh_log_list(struct mesh_area* mesh) 1508 { 1509 char buf[30]; 1510 struct mesh_state* m; 1511 int num = 0; 1512 RBTREE_FOR(m, struct mesh_state*, &mesh->all) { 1513 snprintf(buf, sizeof(buf), "%d%s%s%s%s%s%s mod%d %s%s", 1514 num++, (m->s.is_priming)?"p":"", /* prime */ 1515 (m->s.is_valrec)?"v":"", /* prime */ 1516 (m->s.query_flags&BIT_RD)?"RD":"", 1517 (m->s.query_flags&BIT_CD)?"CD":"", 1518 (m->super_set.count==0)?"d":"", /* detached */ 1519 (m->sub_set.count!=0)?"c":"", /* children */ 1520 m->s.curmod, (m->reply_list)?"rep":"", /*hasreply*/ 1521 (m->cb_list)?"cb":"" /* callbacks */ 1522 ); 1523 log_query_info(VERB_ALGO, buf, &m->s.qinfo); 1524 } 1525 } 1526 1527 void 1528 mesh_stats(struct mesh_area* mesh, const char* str) 1529 { 1530 verbose(VERB_DETAIL, "%s %u recursion states (%u with reply, " 1531 "%u detached), %u waiting replies, %u recursion replies " 1532 "sent, %d replies dropped, %d states jostled out", 1533 str, (unsigned)mesh->all.count, 1534 (unsigned)mesh->num_reply_states, 1535 (unsigned)mesh->num_detached_states, 1536 (unsigned)mesh->num_reply_addrs, 1537 (unsigned)mesh->replies_sent, 1538 (unsigned)mesh->stats_dropped, 1539 (unsigned)mesh->stats_jostled); 1540 if(mesh->replies_sent > 0) { 1541 struct timeval avg; 1542 timeval_divide(&avg, &mesh->replies_sum_wait, 1543 mesh->replies_sent); 1544 log_info("average recursion processing time " 1545 ARG_LL "d.%6.6d sec", 1546 (long long)avg.tv_sec, (int)avg.tv_usec); 1547 log_info("histogram of recursion processing times"); 1548 timehist_log(mesh->histogram, "recursions"); 1549 } 1550 } 1551 1552 void 1553 mesh_stats_clear(struct mesh_area* mesh) 1554 { 1555 if(!mesh) 1556 return; 1557 mesh->replies_sent = 0; 1558 mesh->replies_sum_wait.tv_sec = 0; 1559 mesh->replies_sum_wait.tv_usec = 0; 1560 mesh->stats_jostled = 0; 1561 mesh->stats_dropped = 0; 1562 timehist_clear(mesh->histogram); 1563 mesh->ans_secure = 0; 1564 mesh->ans_bogus = 0; 1565 memset(&mesh->ans_rcode[0], 0, sizeof(size_t)*16); 1566 mesh->ans_nodata = 0; 1567 } 1568 1569 size_t 1570 mesh_get_mem(struct mesh_area* mesh) 1571 { 1572 struct mesh_state* m; 1573 size_t s = sizeof(*mesh) + sizeof(struct timehist) + 1574 sizeof(struct th_buck)*mesh->histogram->num + 1575 sizeof(sldns_buffer) + sldns_buffer_capacity(mesh->qbuf_bak); 1576 RBTREE_FOR(m, struct mesh_state*, &mesh->all) { 1577 /* all, including m itself allocated in qstate region */ 1578 s += regional_get_mem(m->s.region); 1579 } 1580 return s; 1581 } 1582 1583 int 1584 mesh_detect_cycle(struct module_qstate* qstate, struct query_info* qinfo, 1585 uint16_t flags, int prime, int valrec) 1586 { 1587 struct mesh_area* mesh = qstate->env->mesh; 1588 struct mesh_state* dep_m = NULL; 1589 if(!mesh_state_is_unique(qstate->mesh_info)) 1590 dep_m = mesh_area_find(mesh, NULL, qinfo, flags, prime, valrec); 1591 return mesh_detect_cycle_found(qstate, dep_m); 1592 } 1593 1594 void mesh_list_insert(struct mesh_state* m, struct mesh_state** fp, 1595 struct mesh_state** lp) 1596 { 1597 /* insert as last element */ 1598 m->prev = *lp; 1599 m->next = NULL; 1600 if(*lp) 1601 (*lp)->next = m; 1602 else *fp = m; 1603 *lp = m; 1604 } 1605 1606 void mesh_list_remove(struct mesh_state* m, struct mesh_state** fp, 1607 struct mesh_state** lp) 1608 { 1609 if(m->next) 1610 m->next->prev = m->prev; 1611 else *lp = m->prev; 1612 if(m->prev) 1613 m->prev->next = m->next; 1614 else *fp = m->next; 1615 } 1616