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 LIMITED 25 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 26 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE 27 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 28 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 29 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 30 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 31 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 32 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 33 * 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 <ldns/wire2host.h> 47 #include "services/mesh.h" 48 #include "services/outbound_list.h" 49 #include "services/cache/dns.h" 50 #include "util/log.h" 51 #include "util/net_help.h" 52 #include "util/module.h" 53 #include "util/regional.h" 54 #include "util/data/msgencode.h" 55 #include "util/timehist.h" 56 #include "util/fptr_wlist.h" 57 #include "util/alloc.h" 58 #include "util/config_file.h" 59 60 /** subtract timers and the values do not overflow or become negative */ 61 static void 62 timeval_subtract(struct timeval* d, const struct timeval* end, const struct timeval* start) 63 { 64 #ifndef S_SPLINT_S 65 time_t end_usec = end->tv_usec; 66 d->tv_sec = end->tv_sec - start->tv_sec; 67 if(end_usec < start->tv_usec) { 68 end_usec += 1000000; 69 d->tv_sec--; 70 } 71 d->tv_usec = end_usec - start->tv_usec; 72 #endif 73 } 74 75 /** add timers and the values do not overflow or become negative */ 76 static void 77 timeval_add(struct timeval* d, const struct timeval* add) 78 { 79 #ifndef S_SPLINT_S 80 d->tv_sec += add->tv_sec; 81 d->tv_usec += add->tv_usec; 82 if(d->tv_usec > 1000000 ) { 83 d->tv_usec -= 1000000; 84 d->tv_sec++; 85 } 86 #endif 87 } 88 89 /** divide sum of timers to get average */ 90 static void 91 timeval_divide(struct timeval* avg, const struct timeval* sum, size_t d) 92 { 93 #ifndef S_SPLINT_S 94 size_t leftover; 95 if(d == 0) { 96 avg->tv_sec = 0; 97 avg->tv_usec = 0; 98 return; 99 } 100 avg->tv_sec = sum->tv_sec / d; 101 avg->tv_usec = sum->tv_usec / d; 102 /* handle fraction from seconds divide */ 103 leftover = sum->tv_sec - avg->tv_sec*d; 104 avg->tv_usec += (leftover*1000000)/d; 105 #endif 106 } 107 108 /** histogram compare of time values */ 109 static int 110 timeval_smaller(const struct timeval* x, const struct timeval* y) 111 { 112 #ifndef S_SPLINT_S 113 if(x->tv_sec < y->tv_sec) 114 return 1; 115 else if(x->tv_sec == y->tv_sec) { 116 if(x->tv_usec <= y->tv_usec) 117 return 1; 118 else return 0; 119 } 120 else return 0; 121 #endif 122 } 123 124 int 125 mesh_state_compare(const void* ap, const void* bp) 126 { 127 struct mesh_state* a = (struct mesh_state*)ap; 128 struct mesh_state* b = (struct mesh_state*)bp; 129 130 if(a->s.is_priming && !b->s.is_priming) 131 return -1; 132 if(!a->s.is_priming && b->s.is_priming) 133 return 1; 134 135 if((a->s.query_flags&BIT_RD) && !(b->s.query_flags&BIT_RD)) 136 return -1; 137 if(!(a->s.query_flags&BIT_RD) && (b->s.query_flags&BIT_RD)) 138 return 1; 139 140 if((a->s.query_flags&BIT_CD) && !(b->s.query_flags&BIT_CD)) 141 return -1; 142 if(!(a->s.query_flags&BIT_CD) && (b->s.query_flags&BIT_CD)) 143 return 1; 144 145 return query_info_compare(&a->s.qinfo, &b->s.qinfo); 146 } 147 148 int 149 mesh_state_ref_compare(const void* ap, const void* bp) 150 { 151 struct mesh_state_ref* a = (struct mesh_state_ref*)ap; 152 struct mesh_state_ref* b = (struct mesh_state_ref*)bp; 153 return mesh_state_compare(a->s, b->s); 154 } 155 156 struct mesh_area* 157 mesh_create(struct module_stack* stack, struct module_env* env) 158 { 159 struct mesh_area* mesh = calloc(1, sizeof(struct mesh_area)); 160 if(!mesh) { 161 log_err("mesh area alloc: out of memory"); 162 return NULL; 163 } 164 mesh->histogram = timehist_setup(); 165 mesh->qbuf_bak = ldns_buffer_new(env->cfg->msg_buffer_size); 166 if(!mesh->histogram || !mesh->qbuf_bak) { 167 free(mesh); 168 log_err("mesh area alloc: out of memory"); 169 return NULL; 170 } 171 mesh->mods = *stack; 172 mesh->env = env; 173 rbtree_init(&mesh->run, &mesh_state_compare); 174 rbtree_init(&mesh->all, &mesh_state_compare); 175 mesh->num_reply_addrs = 0; 176 mesh->num_reply_states = 0; 177 mesh->num_detached_states = 0; 178 mesh->num_forever_states = 0; 179 mesh->stats_jostled = 0; 180 mesh->stats_dropped = 0; 181 mesh->max_reply_states = env->cfg->num_queries_per_thread; 182 mesh->max_forever_states = (mesh->max_reply_states+1)/2; 183 #ifndef S_SPLINT_S 184 mesh->jostle_max.tv_sec = (time_t)(env->cfg->jostle_time / 1000); 185 mesh->jostle_max.tv_usec = (time_t)((env->cfg->jostle_time % 1000) 186 *1000); 187 #endif 188 return mesh; 189 } 190 191 /** help mesh delete delete mesh states */ 192 static void 193 mesh_delete_helper(rbnode_t* n) 194 { 195 struct mesh_state* mstate = (struct mesh_state*)n->key; 196 /* perform a full delete, not only 'cleanup' routine, 197 * because other callbacks expect a clean state in the mesh. 198 * For 're-entrant' calls */ 199 mesh_state_delete(&mstate->s); 200 /* but because these delete the items from the tree, postorder 201 * traversal and rbtree rebalancing do not work together */ 202 } 203 204 void 205 mesh_delete(struct mesh_area* mesh) 206 { 207 if(!mesh) 208 return; 209 /* free all query states */ 210 while(mesh->all.count) 211 mesh_delete_helper(mesh->all.root); 212 timehist_delete(mesh->histogram); 213 ldns_buffer_free(mesh->qbuf_bak); 214 free(mesh); 215 } 216 217 void 218 mesh_delete_all(struct mesh_area* mesh) 219 { 220 /* free all query states */ 221 while(mesh->all.count) 222 mesh_delete_helper(mesh->all.root); 223 mesh->stats_dropped += mesh->num_reply_addrs; 224 /* clear mesh area references */ 225 rbtree_init(&mesh->run, &mesh_state_compare); 226 rbtree_init(&mesh->all, &mesh_state_compare); 227 mesh->num_reply_addrs = 0; 228 mesh->num_reply_states = 0; 229 mesh->num_detached_states = 0; 230 mesh->num_forever_states = 0; 231 mesh->forever_first = NULL; 232 mesh->forever_last = NULL; 233 mesh->jostle_first = NULL; 234 mesh->jostle_last = NULL; 235 } 236 237 int mesh_make_new_space(struct mesh_area* mesh, ldns_buffer* qbuf) 238 { 239 struct mesh_state* m = mesh->jostle_first; 240 /* free space is available */ 241 if(mesh->num_reply_states < mesh->max_reply_states) 242 return 1; 243 /* try to kick out a jostle-list item */ 244 if(m && m->reply_list && m->list_select == mesh_jostle_list) { 245 /* how old is it? */ 246 struct timeval age; 247 timeval_subtract(&age, mesh->env->now_tv, 248 &m->reply_list->start_time); 249 if(timeval_smaller(&mesh->jostle_max, &age)) { 250 /* its a goner */ 251 log_nametypeclass(VERB_ALGO, "query jostled out to " 252 "make space for a new one", 253 m->s.qinfo.qname, m->s.qinfo.qtype, 254 m->s.qinfo.qclass); 255 /* backup the query */ 256 if(qbuf) ldns_buffer_copy(mesh->qbuf_bak, qbuf); 257 /* notify supers */ 258 if(m->super_set.count > 0) { 259 verbose(VERB_ALGO, "notify supers of failure"); 260 m->s.return_msg = NULL; 261 m->s.return_rcode = LDNS_RCODE_SERVFAIL; 262 mesh_walk_supers(mesh, m); 263 } 264 mesh->stats_jostled ++; 265 mesh_state_delete(&m->s); 266 /* restore the query - note that the qinfo ptr to 267 * the querybuffer is then correct again. */ 268 if(qbuf) ldns_buffer_copy(qbuf, mesh->qbuf_bak); 269 return 1; 270 } 271 } 272 /* no space for new item */ 273 return 0; 274 } 275 276 void mesh_new_client(struct mesh_area* mesh, struct query_info* qinfo, 277 uint16_t qflags, struct edns_data* edns, struct comm_reply* rep, 278 uint16_t qid) 279 { 280 /* do not use CD flag from user for mesh state, we want the CD-query 281 * to receive validation anyway, to protect out cache contents and 282 * avoid bad-data in this cache that a downstream validator cannot 283 * remove from this cache */ 284 struct mesh_state* s = mesh_area_find(mesh, qinfo, qflags&BIT_RD, 0); 285 int was_detached = 0; 286 int was_noreply = 0; 287 int added = 0; 288 /* does this create a new reply state? */ 289 if(!s || s->list_select == mesh_no_list) { 290 if(!mesh_make_new_space(mesh, rep->c->buffer)) { 291 verbose(VERB_ALGO, "Too many queries. dropping " 292 "incoming query."); 293 comm_point_drop_reply(rep); 294 mesh->stats_dropped ++; 295 return; 296 } 297 /* for this new reply state, the reply address is free, 298 * so the limit of reply addresses does not stop reply states*/ 299 } else { 300 /* protect our memory usage from storing reply addresses */ 301 if(mesh->num_reply_addrs > mesh->max_reply_states*16) { 302 verbose(VERB_ALGO, "Too many requests queued. " 303 "dropping incoming query."); 304 mesh->stats_dropped++; 305 comm_point_drop_reply(rep); 306 return; 307 } 308 } 309 /* see if it already exists, if not, create one */ 310 if(!s) { 311 #ifdef UNBOUND_DEBUG 312 struct rbnode_t* n; 313 #endif 314 s = mesh_state_create(mesh->env, qinfo, qflags&BIT_RD, 0); 315 if(!s) { 316 log_err("mesh_state_create: out of memory; SERVFAIL"); 317 error_encode(rep->c->buffer, LDNS_RCODE_SERVFAIL, 318 qinfo, qid, qflags, edns); 319 comm_point_send_reply(rep); 320 return; 321 } 322 #ifdef UNBOUND_DEBUG 323 n = 324 #endif 325 rbtree_insert(&mesh->all, &s->node); 326 log_assert(n != NULL); 327 /* set detached (it is now) */ 328 mesh->num_detached_states++; 329 added = 1; 330 } 331 if(!s->reply_list && !s->cb_list && s->super_set.count == 0) 332 was_detached = 1; 333 if(!s->reply_list && !s->cb_list) 334 was_noreply = 1; 335 /* add reply to s */ 336 if(!mesh_state_add_reply(s, edns, rep, qid, qflags, qinfo->qname)) { 337 log_err("mesh_new_client: out of memory; SERVFAIL"); 338 error_encode(rep->c->buffer, LDNS_RCODE_SERVFAIL, 339 qinfo, qid, qflags, edns); 340 comm_point_send_reply(rep); 341 if(added) 342 mesh_state_delete(&s->s); 343 return; 344 } 345 /* update statistics */ 346 if(was_detached) { 347 log_assert(mesh->num_detached_states > 0); 348 mesh->num_detached_states--; 349 } 350 if(was_noreply) { 351 mesh->num_reply_states ++; 352 } 353 mesh->num_reply_addrs++; 354 if(s->list_select == mesh_no_list) { 355 /* move to either the forever or the jostle_list */ 356 if(mesh->num_forever_states < mesh->max_forever_states) { 357 mesh->num_forever_states ++; 358 mesh_list_insert(s, &mesh->forever_first, 359 &mesh->forever_last); 360 s->list_select = mesh_forever_list; 361 } else { 362 mesh_list_insert(s, &mesh->jostle_first, 363 &mesh->jostle_last); 364 s->list_select = mesh_jostle_list; 365 } 366 } 367 if(added) 368 mesh_run(mesh, s, module_event_new, NULL); 369 } 370 371 int 372 mesh_new_callback(struct mesh_area* mesh, struct query_info* qinfo, 373 uint16_t qflags, struct edns_data* edns, ldns_buffer* buf, 374 uint16_t qid, mesh_cb_func_t cb, void* cb_arg) 375 { 376 struct mesh_state* s = mesh_area_find(mesh, qinfo, qflags&BIT_RD, 0); 377 int was_detached = 0; 378 int was_noreply = 0; 379 int added = 0; 380 /* there are no limits on the number of callbacks */ 381 382 /* see if it already exists, if not, create one */ 383 if(!s) { 384 #ifdef UNBOUND_DEBUG 385 struct rbnode_t* n; 386 #endif 387 s = mesh_state_create(mesh->env, qinfo, qflags&BIT_RD, 0); 388 if(!s) { 389 return 0; 390 } 391 #ifdef UNBOUND_DEBUG 392 n = 393 #endif 394 rbtree_insert(&mesh->all, &s->node); 395 log_assert(n != NULL); 396 /* set detached (it is now) */ 397 mesh->num_detached_states++; 398 added = 1; 399 } 400 if(!s->reply_list && !s->cb_list && s->super_set.count == 0) 401 was_detached = 1; 402 if(!s->reply_list && !s->cb_list) 403 was_noreply = 1; 404 /* add reply to s */ 405 if(!mesh_state_add_cb(s, edns, buf, cb, cb_arg, qid, qflags)) { 406 if(added) 407 mesh_state_delete(&s->s); 408 return 0; 409 } 410 /* update statistics */ 411 if(was_detached) { 412 log_assert(mesh->num_detached_states > 0); 413 mesh->num_detached_states--; 414 } 415 if(was_noreply) { 416 mesh->num_reply_states ++; 417 } 418 mesh->num_reply_addrs++; 419 if(added) 420 mesh_run(mesh, s, module_event_new, NULL); 421 return 1; 422 } 423 424 void mesh_new_prefetch(struct mesh_area* mesh, struct query_info* qinfo, 425 uint16_t qflags, uint32_t leeway) 426 { 427 struct mesh_state* s = mesh_area_find(mesh, qinfo, qflags&BIT_RD, 0); 428 #ifdef UNBOUND_DEBUG 429 struct rbnode_t* n; 430 #endif 431 /* already exists, and for a different purpose perhaps. 432 * if mesh_no_list, keep it that way. */ 433 if(s) { 434 /* make it ignore the cache from now on */ 435 if(!s->s.blacklist) 436 sock_list_insert(&s->s.blacklist, NULL, 0, s->s.region); 437 if(s->s.prefetch_leeway < leeway) 438 s->s.prefetch_leeway = leeway; 439 return; 440 } 441 if(!mesh_make_new_space(mesh, NULL)) { 442 verbose(VERB_ALGO, "Too many queries. dropped prefetch."); 443 mesh->stats_dropped ++; 444 return; 445 } 446 s = mesh_state_create(mesh->env, qinfo, qflags&BIT_RD, 0); 447 if(!s) { 448 log_err("prefetch mesh_state_create: out of memory"); 449 return; 450 } 451 #ifdef UNBOUND_DEBUG 452 n = 453 #endif 454 rbtree_insert(&mesh->all, &s->node); 455 log_assert(n != NULL); 456 /* set detached (it is now) */ 457 mesh->num_detached_states++; 458 /* make it ignore the cache */ 459 sock_list_insert(&s->s.blacklist, NULL, 0, s->s.region); 460 s->s.prefetch_leeway = leeway; 461 462 if(s->list_select == mesh_no_list) { 463 /* move to either the forever or the jostle_list */ 464 if(mesh->num_forever_states < mesh->max_forever_states) { 465 mesh->num_forever_states ++; 466 mesh_list_insert(s, &mesh->forever_first, 467 &mesh->forever_last); 468 s->list_select = mesh_forever_list; 469 } else { 470 mesh_list_insert(s, &mesh->jostle_first, 471 &mesh->jostle_last); 472 s->list_select = mesh_jostle_list; 473 } 474 } 475 mesh_run(mesh, s, module_event_new, NULL); 476 } 477 478 void mesh_report_reply(struct mesh_area* mesh, struct outbound_entry* e, 479 struct comm_reply* reply, int what) 480 { 481 enum module_ev event = module_event_reply; 482 e->qstate->reply = reply; 483 if(what != NETEVENT_NOERROR) { 484 event = module_event_noreply; 485 if(what == NETEVENT_CAPSFAIL) 486 event = module_event_capsfail; 487 } 488 mesh_run(mesh, e->qstate->mesh_info, event, e); 489 } 490 491 struct mesh_state* 492 mesh_state_create(struct module_env* env, struct query_info* qinfo, 493 uint16_t qflags, int prime) 494 { 495 struct regional* region = alloc_reg_obtain(env->alloc); 496 struct mesh_state* mstate; 497 int i; 498 if(!region) 499 return NULL; 500 mstate = (struct mesh_state*)regional_alloc(region, 501 sizeof(struct mesh_state)); 502 if(!mstate) { 503 alloc_reg_release(env->alloc, region); 504 return NULL; 505 } 506 memset(mstate, 0, sizeof(*mstate)); 507 mstate->node = *RBTREE_NULL; 508 mstate->run_node = *RBTREE_NULL; 509 mstate->node.key = mstate; 510 mstate->run_node.key = mstate; 511 mstate->reply_list = NULL; 512 mstate->list_select = mesh_no_list; 513 mstate->replies_sent = 0; 514 rbtree_init(&mstate->super_set, &mesh_state_ref_compare); 515 rbtree_init(&mstate->sub_set, &mesh_state_ref_compare); 516 mstate->num_activated = 0; 517 /* init module qstate */ 518 mstate->s.qinfo.qtype = qinfo->qtype; 519 mstate->s.qinfo.qclass = qinfo->qclass; 520 mstate->s.qinfo.qname_len = qinfo->qname_len; 521 mstate->s.qinfo.qname = regional_alloc_init(region, qinfo->qname, 522 qinfo->qname_len); 523 if(!mstate->s.qinfo.qname) { 524 alloc_reg_release(env->alloc, region); 525 return NULL; 526 } 527 /* remove all weird bits from qflags */ 528 mstate->s.query_flags = (qflags & (BIT_RD|BIT_CD)); 529 mstate->s.is_priming = prime; 530 mstate->s.reply = NULL; 531 mstate->s.region = region; 532 mstate->s.curmod = 0; 533 mstate->s.return_msg = 0; 534 mstate->s.return_rcode = LDNS_RCODE_NOERROR; 535 mstate->s.env = env; 536 mstate->s.mesh_info = mstate; 537 mstate->s.prefetch_leeway = 0; 538 /* init modules */ 539 for(i=0; i<env->mesh->mods.num; i++) { 540 mstate->s.minfo[i] = NULL; 541 mstate->s.ext_state[i] = module_state_initial; 542 } 543 return mstate; 544 } 545 546 void 547 mesh_state_cleanup(struct mesh_state* mstate) 548 { 549 struct mesh_area* mesh; 550 int i; 551 if(!mstate) 552 return; 553 mesh = mstate->s.env->mesh; 554 /* drop unsent replies */ 555 if(!mstate->replies_sent) { 556 struct mesh_reply* rep; 557 struct mesh_cb* cb; 558 for(rep=mstate->reply_list; rep; rep=rep->next) { 559 comm_point_drop_reply(&rep->query_reply); 560 mesh->num_reply_addrs--; 561 } 562 for(cb=mstate->cb_list; cb; cb=cb->next) { 563 fptr_ok(fptr_whitelist_mesh_cb(cb->cb)); 564 (*cb->cb)(cb->cb_arg, LDNS_RCODE_SERVFAIL, NULL, 565 sec_status_unchecked, NULL); 566 mesh->num_reply_addrs--; 567 } 568 } 569 570 /* de-init modules */ 571 for(i=0; i<mesh->mods.num; i++) { 572 fptr_ok(fptr_whitelist_mod_clear(mesh->mods.mod[i]->clear)); 573 (*mesh->mods.mod[i]->clear)(&mstate->s, i); 574 mstate->s.minfo[i] = NULL; 575 mstate->s.ext_state[i] = module_finished; 576 } 577 alloc_reg_release(mstate->s.env->alloc, mstate->s.region); 578 } 579 580 void 581 mesh_state_delete(struct module_qstate* qstate) 582 { 583 struct mesh_area* mesh; 584 struct mesh_state_ref* super, ref; 585 struct mesh_state* mstate; 586 if(!qstate) 587 return; 588 mstate = qstate->mesh_info; 589 mesh = mstate->s.env->mesh; 590 mesh_detach_subs(&mstate->s); 591 if(mstate->list_select == mesh_forever_list) { 592 mesh->num_forever_states --; 593 mesh_list_remove(mstate, &mesh->forever_first, 594 &mesh->forever_last); 595 } else if(mstate->list_select == mesh_jostle_list) { 596 mesh_list_remove(mstate, &mesh->jostle_first, 597 &mesh->jostle_last); 598 } 599 if(!mstate->reply_list && !mstate->cb_list 600 && mstate->super_set.count == 0) { 601 log_assert(mesh->num_detached_states > 0); 602 mesh->num_detached_states--; 603 } 604 if(mstate->reply_list || mstate->cb_list) { 605 log_assert(mesh->num_reply_states > 0); 606 mesh->num_reply_states--; 607 } 608 ref.node.key = &ref; 609 ref.s = mstate; 610 RBTREE_FOR(super, struct mesh_state_ref*, &mstate->super_set) { 611 (void)rbtree_delete(&super->s->sub_set, &ref); 612 } 613 (void)rbtree_delete(&mesh->run, mstate); 614 (void)rbtree_delete(&mesh->all, mstate); 615 mesh_state_cleanup(mstate); 616 } 617 618 /** helper recursive rbtree find routine */ 619 static int 620 find_in_subsub(struct mesh_state* m, struct mesh_state* tofind, size_t *c) 621 { 622 struct mesh_state_ref* r; 623 if((*c)++ > MESH_MAX_SUBSUB) 624 return 1; 625 RBTREE_FOR(r, struct mesh_state_ref*, &m->sub_set) { 626 if(r->s == tofind || find_in_subsub(r->s, tofind, c)) 627 return 1; 628 } 629 return 0; 630 } 631 632 /** find cycle for already looked up mesh_state */ 633 static int 634 mesh_detect_cycle_found(struct module_qstate* qstate, struct mesh_state* dep_m) 635 { 636 struct mesh_state* cyc_m = qstate->mesh_info; 637 size_t counter = 0; 638 if(!dep_m) 639 return 0; 640 if(dep_m == cyc_m || find_in_subsub(dep_m, cyc_m, &counter)) { 641 if(counter > MESH_MAX_SUBSUB) 642 return 2; 643 return 1; 644 } 645 return 0; 646 } 647 648 void mesh_detach_subs(struct module_qstate* qstate) 649 { 650 struct mesh_area* mesh = qstate->env->mesh; 651 struct mesh_state_ref* ref, lookup; 652 #ifdef UNBOUND_DEBUG 653 struct rbnode_t* n; 654 #endif 655 lookup.node.key = &lookup; 656 lookup.s = qstate->mesh_info; 657 RBTREE_FOR(ref, struct mesh_state_ref*, &qstate->mesh_info->sub_set) { 658 #ifdef UNBOUND_DEBUG 659 n = 660 #endif 661 rbtree_delete(&ref->s->super_set, &lookup); 662 log_assert(n != NULL); /* must have been present */ 663 if(!ref->s->reply_list && !ref->s->cb_list 664 && ref->s->super_set.count == 0) { 665 mesh->num_detached_states++; 666 log_assert(mesh->num_detached_states + 667 mesh->num_reply_states <= mesh->all.count); 668 } 669 } 670 rbtree_init(&qstate->mesh_info->sub_set, &mesh_state_ref_compare); 671 } 672 673 int mesh_attach_sub(struct module_qstate* qstate, struct query_info* qinfo, 674 uint16_t qflags, int prime, struct module_qstate** newq) 675 { 676 /* find it, if not, create it */ 677 struct mesh_area* mesh = qstate->env->mesh; 678 struct mesh_state* sub = mesh_area_find(mesh, qinfo, qflags, prime); 679 if(mesh_detect_cycle_found(qstate, sub)) { 680 verbose(VERB_ALGO, "attach failed, cycle detected"); 681 return 0; 682 } 683 if(!sub) { 684 #ifdef UNBOUND_DEBUG 685 struct rbnode_t* n; 686 #endif 687 /* create a new one */ 688 sub = mesh_state_create(qstate->env, qinfo, qflags, prime); 689 if(!sub) { 690 log_err("mesh_attach_sub: out of memory"); 691 return 0; 692 } 693 #ifdef UNBOUND_DEBUG 694 n = 695 #endif 696 rbtree_insert(&mesh->all, &sub->node); 697 log_assert(n != NULL); 698 /* set detached (it is now) */ 699 mesh->num_detached_states++; 700 /* set new query state to run */ 701 #ifdef UNBOUND_DEBUG 702 n = 703 #endif 704 rbtree_insert(&mesh->run, &sub->run_node); 705 log_assert(n != NULL); 706 *newq = &sub->s; 707 } else 708 *newq = NULL; 709 if(!mesh_state_attachment(qstate->mesh_info, sub)) 710 return 0; 711 if(!sub->reply_list && !sub->cb_list && sub->super_set.count == 1) { 712 /* it used to be detached, before this one got added */ 713 log_assert(mesh->num_detached_states > 0); 714 mesh->num_detached_states--; 715 } 716 /* *newq will be run when inited after the current module stops */ 717 return 1; 718 } 719 720 int mesh_state_attachment(struct mesh_state* super, struct mesh_state* sub) 721 { 722 #ifdef UNBOUND_DEBUG 723 struct rbnode_t* n; 724 #endif 725 struct mesh_state_ref* subref; /* points to sub, inserted in super */ 726 struct mesh_state_ref* superref; /* points to super, inserted in sub */ 727 if( !(subref = regional_alloc(super->s.region, 728 sizeof(struct mesh_state_ref))) || 729 !(superref = regional_alloc(sub->s.region, 730 sizeof(struct mesh_state_ref))) ) { 731 log_err("mesh_state_attachment: out of memory"); 732 return 0; 733 } 734 superref->node.key = superref; 735 superref->s = super; 736 subref->node.key = subref; 737 subref->s = sub; 738 #ifdef UNBOUND_DEBUG 739 n = 740 #endif 741 rbtree_insert(&sub->super_set, &superref->node); 742 log_assert(n != NULL); 743 #ifdef UNBOUND_DEBUG 744 n = 745 #endif 746 rbtree_insert(&super->sub_set, &subref->node); 747 log_assert(n != NULL); 748 return 1; 749 } 750 751 /** 752 * callback results to mesh cb entry 753 * @param m: mesh state to send it for. 754 * @param rcode: if not 0, error code. 755 * @param rep: reply to send (or NULL if rcode is set). 756 * @param r: callback entry 757 */ 758 static void 759 mesh_do_callback(struct mesh_state* m, int rcode, struct reply_info* rep, 760 struct mesh_cb* r) 761 { 762 int secure; 763 char* reason = NULL; 764 /* bogus messages are not made into servfail, sec_status passed 765 * to the callback function */ 766 if(rep && rep->security == sec_status_secure) 767 secure = 1; 768 else secure = 0; 769 if(!rep && rcode == LDNS_RCODE_NOERROR) 770 rcode = LDNS_RCODE_SERVFAIL; 771 if(!rcode && rep->security == sec_status_bogus) { 772 if(!(reason = errinf_to_str(&m->s))) 773 rcode = LDNS_RCODE_SERVFAIL; 774 } 775 /* send the reply */ 776 if(rcode) { 777 fptr_ok(fptr_whitelist_mesh_cb(r->cb)); 778 (*r->cb)(r->cb_arg, rcode, r->buf, sec_status_unchecked, NULL); 779 } else { 780 size_t udp_size = r->edns.udp_size; 781 ldns_buffer_clear(r->buf); 782 r->edns.edns_version = EDNS_ADVERTISED_VERSION; 783 r->edns.udp_size = EDNS_ADVERTISED_SIZE; 784 r->edns.ext_rcode = 0; 785 r->edns.bits &= EDNS_DO; 786 if(!reply_info_answer_encode(&m->s.qinfo, rep, r->qid, 787 r->qflags, r->buf, 0, 1, 788 m->s.env->scratch, udp_size, &r->edns, 789 (int)(r->edns.bits & EDNS_DO), secure)) 790 { 791 fptr_ok(fptr_whitelist_mesh_cb(r->cb)); 792 (*r->cb)(r->cb_arg, LDNS_RCODE_SERVFAIL, r->buf, 793 sec_status_unchecked, NULL); 794 } else { 795 fptr_ok(fptr_whitelist_mesh_cb(r->cb)); 796 (*r->cb)(r->cb_arg, LDNS_RCODE_NOERROR, r->buf, 797 rep->security, reason); 798 } 799 } 800 free(reason); 801 m->s.env->mesh->num_reply_addrs--; 802 } 803 804 /** 805 * Send reply to mesh reply entry 806 * @param m: mesh state to send it for. 807 * @param rcode: if not 0, error code. 808 * @param rep: reply to send (or NULL if rcode is set). 809 * @param r: reply entry 810 * @param prev: previous reply, already has its answer encoded in buffer. 811 */ 812 static void 813 mesh_send_reply(struct mesh_state* m, int rcode, struct reply_info* rep, 814 struct mesh_reply* r, struct mesh_reply* prev) 815 { 816 struct timeval end_time; 817 struct timeval duration; 818 int secure; 819 /* examine security status */ 820 if(m->s.env->need_to_validate && (!(r->qflags&BIT_CD) || 821 m->s.env->cfg->ignore_cd) && rep && 822 rep->security <= sec_status_bogus) { 823 rcode = LDNS_RCODE_SERVFAIL; 824 if(m->s.env->cfg->stat_extended) 825 m->s.env->mesh->ans_bogus++; 826 } 827 if(rep && rep->security == sec_status_secure) 828 secure = 1; 829 else secure = 0; 830 if(!rep && rcode == LDNS_RCODE_NOERROR) 831 rcode = LDNS_RCODE_SERVFAIL; 832 /* send the reply */ 833 if(prev && prev->qflags == r->qflags && 834 prev->edns.edns_present == r->edns.edns_present && 835 prev->edns.bits == r->edns.bits && 836 prev->edns.udp_size == r->edns.udp_size) { 837 /* if the previous reply is identical to this one, fix ID */ 838 if(prev->query_reply.c->buffer != r->query_reply.c->buffer) 839 ldns_buffer_copy(r->query_reply.c->buffer, 840 prev->query_reply.c->buffer); 841 ldns_buffer_write_at(r->query_reply.c->buffer, 0, 842 &r->qid, sizeof(uint16_t)); 843 ldns_buffer_write_at(r->query_reply.c->buffer, 12, 844 r->qname, m->s.qinfo.qname_len); 845 comm_point_send_reply(&r->query_reply); 846 } else if(rcode) { 847 m->s.qinfo.qname = r->qname; 848 error_encode(r->query_reply.c->buffer, rcode, &m->s.qinfo, 849 r->qid, r->qflags, &r->edns); 850 comm_point_send_reply(&r->query_reply); 851 } else { 852 size_t udp_size = r->edns.udp_size; 853 r->edns.edns_version = EDNS_ADVERTISED_VERSION; 854 r->edns.udp_size = EDNS_ADVERTISED_SIZE; 855 r->edns.ext_rcode = 0; 856 r->edns.bits &= EDNS_DO; 857 m->s.qinfo.qname = r->qname; 858 if(!reply_info_answer_encode(&m->s.qinfo, rep, r->qid, 859 r->qflags, r->query_reply.c->buffer, 0, 1, 860 m->s.env->scratch, udp_size, &r->edns, 861 (int)(r->edns.bits & EDNS_DO), secure)) 862 { 863 error_encode(r->query_reply.c->buffer, 864 LDNS_RCODE_SERVFAIL, &m->s.qinfo, r->qid, 865 r->qflags, &r->edns); 866 } 867 comm_point_send_reply(&r->query_reply); 868 } 869 /* account */ 870 m->s.env->mesh->num_reply_addrs--; 871 end_time = *m->s.env->now_tv; 872 timeval_subtract(&duration, &end_time, &r->start_time); 873 verbose(VERB_ALGO, "query took %d.%6.6d sec", 874 (int)duration.tv_sec, (int)duration.tv_usec); 875 m->s.env->mesh->replies_sent++; 876 timeval_add(&m->s.env->mesh->replies_sum_wait, &duration); 877 timehist_insert(m->s.env->mesh->histogram, &duration); 878 if(m->s.env->cfg->stat_extended) { 879 uint16_t rc = FLAGS_GET_RCODE(ldns_buffer_read_u16_at(r-> 880 query_reply.c->buffer, 2)); 881 if(secure) m->s.env->mesh->ans_secure++; 882 m->s.env->mesh->ans_rcode[ rc ] ++; 883 if(rc == 0 && LDNS_ANCOUNT(ldns_buffer_begin(r-> 884 query_reply.c->buffer)) == 0) 885 m->s.env->mesh->ans_nodata++; 886 } 887 } 888 889 void mesh_query_done(struct mesh_state* mstate) 890 { 891 struct mesh_reply* r; 892 struct mesh_reply* prev = NULL; 893 struct mesh_cb* c; 894 struct reply_info* rep = (mstate->s.return_msg? 895 mstate->s.return_msg->rep:NULL); 896 for(r = mstate->reply_list; r; r = r->next) { 897 mesh_send_reply(mstate, mstate->s.return_rcode, rep, r, prev); 898 prev = r; 899 } 900 mstate->replies_sent = 1; 901 for(c = mstate->cb_list; c; c = c->next) { 902 mesh_do_callback(mstate, mstate->s.return_rcode, rep, c); 903 } 904 } 905 906 void mesh_walk_supers(struct mesh_area* mesh, struct mesh_state* mstate) 907 { 908 struct mesh_state_ref* ref; 909 RBTREE_FOR(ref, struct mesh_state_ref*, &mstate->super_set) 910 { 911 /* make super runnable */ 912 (void)rbtree_insert(&mesh->run, &ref->s->run_node); 913 /* callback the function to inform super of result */ 914 fptr_ok(fptr_whitelist_mod_inform_super( 915 mesh->mods.mod[ref->s->s.curmod]->inform_super)); 916 (*mesh->mods.mod[ref->s->s.curmod]->inform_super)(&mstate->s, 917 ref->s->s.curmod, &ref->s->s); 918 } 919 } 920 921 struct mesh_state* mesh_area_find(struct mesh_area* mesh, 922 struct query_info* qinfo, uint16_t qflags, int prime) 923 { 924 struct mesh_state key; 925 struct mesh_state* result; 926 927 key.node.key = &key; 928 key.s.is_priming = prime; 929 key.s.qinfo = *qinfo; 930 key.s.query_flags = qflags; 931 932 result = (struct mesh_state*)rbtree_search(&mesh->all, &key); 933 return result; 934 } 935 936 int mesh_state_add_cb(struct mesh_state* s, struct edns_data* edns, 937 ldns_buffer* buf, mesh_cb_func_t cb, void* cb_arg, 938 uint16_t qid, uint16_t qflags) 939 { 940 struct mesh_cb* r = regional_alloc(s->s.region, 941 sizeof(struct mesh_cb)); 942 if(!r) 943 return 0; 944 r->buf = buf; 945 log_assert(fptr_whitelist_mesh_cb(cb)); /* early failure ifmissing*/ 946 r->cb = cb; 947 r->cb_arg = cb_arg; 948 r->edns = *edns; 949 r->qid = qid; 950 r->qflags = qflags; 951 r->next = s->cb_list; 952 s->cb_list = r; 953 return 1; 954 955 } 956 957 int mesh_state_add_reply(struct mesh_state* s, struct edns_data* edns, 958 struct comm_reply* rep, uint16_t qid, uint16_t qflags, uint8_t* qname) 959 { 960 struct mesh_reply* r = regional_alloc(s->s.region, 961 sizeof(struct mesh_reply)); 962 if(!r) 963 return 0; 964 r->query_reply = *rep; 965 r->edns = *edns; 966 r->qid = qid; 967 r->qflags = qflags; 968 r->start_time = *s->s.env->now_tv; 969 r->next = s->reply_list; 970 r->qname = regional_alloc_init(s->s.region, qname, 971 s->s.qinfo.qname_len); 972 if(!r->qname) 973 return 0; 974 s->reply_list = r; 975 return 1; 976 977 } 978 979 /** 980 * Continue processing the mesh state at another module. 981 * Handles module to modules tranfer of control. 982 * Handles module finished. 983 * @param mesh: the mesh area. 984 * @param mstate: currently active mesh state. 985 * Deleted if finished, calls _done and _supers to 986 * send replies to clients and inform other mesh states. 987 * This in turn may create additional runnable mesh states. 988 * @param s: state at which the current module exited. 989 * @param ev: the event sent to the module. 990 * returned is the event to send to the next module. 991 * @return true if continue processing at the new module. 992 * false if not continued processing is needed. 993 */ 994 static int 995 mesh_continue(struct mesh_area* mesh, struct mesh_state* mstate, 996 enum module_ext_state s, enum module_ev* ev) 997 { 998 mstate->num_activated++; 999 if(mstate->num_activated > MESH_MAX_ACTIVATION) { 1000 /* module is looping. Stop it. */ 1001 log_err("internal error: looping module stopped"); 1002 log_query_info(VERB_QUERY, "pass error for qstate", 1003 &mstate->s.qinfo); 1004 s = module_error; 1005 } 1006 if(s == module_wait_module || s == module_restart_next) { 1007 /* start next module */ 1008 mstate->s.curmod++; 1009 if(mesh->mods.num == mstate->s.curmod) { 1010 log_err("Cannot pass to next module; at last module"); 1011 log_query_info(VERB_QUERY, "pass error for qstate", 1012 &mstate->s.qinfo); 1013 mstate->s.curmod--; 1014 return mesh_continue(mesh, mstate, module_error, ev); 1015 } 1016 if(s == module_restart_next) { 1017 fptr_ok(fptr_whitelist_mod_clear( 1018 mesh->mods.mod[mstate->s.curmod]->clear)); 1019 (*mesh->mods.mod[mstate->s.curmod]->clear) 1020 (&mstate->s, mstate->s.curmod); 1021 mstate->s.minfo[mstate->s.curmod] = NULL; 1022 } 1023 *ev = module_event_pass; 1024 return 1; 1025 } 1026 if(s == module_error && mstate->s.return_rcode == LDNS_RCODE_NOERROR) { 1027 /* error is bad, handle pass back up below */ 1028 mstate->s.return_rcode = LDNS_RCODE_SERVFAIL; 1029 } 1030 if(s == module_error || s == module_finished) { 1031 if(mstate->s.curmod == 0) { 1032 mesh_query_done(mstate); 1033 mesh_walk_supers(mesh, mstate); 1034 mesh_state_delete(&mstate->s); 1035 return 0; 1036 } 1037 /* pass along the locus of control */ 1038 mstate->s.curmod --; 1039 *ev = module_event_moddone; 1040 return 1; 1041 } 1042 return 0; 1043 } 1044 1045 void mesh_run(struct mesh_area* mesh, struct mesh_state* mstate, 1046 enum module_ev ev, struct outbound_entry* e) 1047 { 1048 enum module_ext_state s; 1049 verbose(VERB_ALGO, "mesh_run: start"); 1050 while(mstate) { 1051 /* run the module */ 1052 fptr_ok(fptr_whitelist_mod_operate( 1053 mesh->mods.mod[mstate->s.curmod]->operate)); 1054 (*mesh->mods.mod[mstate->s.curmod]->operate) 1055 (&mstate->s, ev, mstate->s.curmod, e); 1056 1057 /* examine results */ 1058 mstate->s.reply = NULL; 1059 regional_free_all(mstate->s.env->scratch); 1060 s = mstate->s.ext_state[mstate->s.curmod]; 1061 verbose(VERB_ALGO, "mesh_run: %s module exit state is %s", 1062 mesh->mods.mod[mstate->s.curmod]->name, strextstate(s)); 1063 e = NULL; 1064 if(mesh_continue(mesh, mstate, s, &ev)) 1065 continue; 1066 1067 /* run more modules */ 1068 ev = module_event_pass; 1069 if(mesh->run.count > 0) { 1070 /* pop random element off the runnable tree */ 1071 mstate = (struct mesh_state*)mesh->run.root->key; 1072 (void)rbtree_delete(&mesh->run, mstate); 1073 } else mstate = NULL; 1074 } 1075 if(verbosity >= VERB_ALGO) { 1076 mesh_stats(mesh, "mesh_run: end"); 1077 mesh_log_list(mesh); 1078 } 1079 } 1080 1081 void 1082 mesh_log_list(struct mesh_area* mesh) 1083 { 1084 char buf[30]; 1085 struct mesh_state* m; 1086 int num = 0; 1087 RBTREE_FOR(m, struct mesh_state*, &mesh->all) { 1088 snprintf(buf, sizeof(buf), "%d%s%s%s%s%s mod%d %s%s", 1089 num++, (m->s.is_priming)?"p":"", /* prime */ 1090 (m->s.query_flags&BIT_RD)?"RD":"", 1091 (m->s.query_flags&BIT_CD)?"CD":"", 1092 (m->super_set.count==0)?"d":"", /* detached */ 1093 (m->sub_set.count!=0)?"c":"", /* children */ 1094 m->s.curmod, (m->reply_list)?"rep":"", /*hasreply*/ 1095 (m->cb_list)?"cb":"" /* callbacks */ 1096 ); 1097 log_query_info(VERB_ALGO, buf, &m->s.qinfo); 1098 } 1099 } 1100 1101 void 1102 mesh_stats(struct mesh_area* mesh, const char* str) 1103 { 1104 verbose(VERB_DETAIL, "%s %u recursion states (%u with reply, " 1105 "%u detached), %u waiting replies, %u recursion replies " 1106 "sent, %d replies dropped, %d states jostled out", 1107 str, (unsigned)mesh->all.count, 1108 (unsigned)mesh->num_reply_states, 1109 (unsigned)mesh->num_detached_states, 1110 (unsigned)mesh->num_reply_addrs, 1111 (unsigned)mesh->replies_sent, 1112 (unsigned)mesh->stats_dropped, 1113 (unsigned)mesh->stats_jostled); 1114 if(mesh->replies_sent > 0) { 1115 struct timeval avg; 1116 timeval_divide(&avg, &mesh->replies_sum_wait, 1117 mesh->replies_sent); 1118 log_info("average recursion processing time " 1119 "%d.%6.6d sec", (int)avg.tv_sec, (int)avg.tv_usec); 1120 log_info("histogram of recursion processing times"); 1121 timehist_log(mesh->histogram, "recursions"); 1122 } 1123 } 1124 1125 void 1126 mesh_stats_clear(struct mesh_area* mesh) 1127 { 1128 if(!mesh) 1129 return; 1130 mesh->replies_sent = 0; 1131 mesh->replies_sum_wait.tv_sec = 0; 1132 mesh->replies_sum_wait.tv_usec = 0; 1133 mesh->stats_jostled = 0; 1134 mesh->stats_dropped = 0; 1135 timehist_clear(mesh->histogram); 1136 mesh->ans_secure = 0; 1137 mesh->ans_bogus = 0; 1138 memset(&mesh->ans_rcode[0], 0, sizeof(size_t)*16); 1139 mesh->ans_nodata = 0; 1140 } 1141 1142 size_t 1143 mesh_get_mem(struct mesh_area* mesh) 1144 { 1145 struct mesh_state* m; 1146 size_t s = sizeof(*mesh) + sizeof(struct timehist) + 1147 sizeof(struct th_buck)*mesh->histogram->num + 1148 sizeof(ldns_buffer) + ldns_buffer_capacity(mesh->qbuf_bak); 1149 RBTREE_FOR(m, struct mesh_state*, &mesh->all) { 1150 /* all, including m itself allocated in qstate region */ 1151 s += regional_get_mem(m->s.region); 1152 } 1153 return s; 1154 } 1155 1156 int 1157 mesh_detect_cycle(struct module_qstate* qstate, struct query_info* qinfo, 1158 uint16_t flags, int prime) 1159 { 1160 struct mesh_area* mesh = qstate->env->mesh; 1161 struct mesh_state* dep_m = mesh_area_find(mesh, qinfo, flags, prime); 1162 return mesh_detect_cycle_found(qstate, dep_m); 1163 } 1164 1165 void mesh_list_insert(struct mesh_state* m, struct mesh_state** fp, 1166 struct mesh_state** lp) 1167 { 1168 /* insert as last element */ 1169 m->prev = *lp; 1170 m->next = NULL; 1171 if(*lp) 1172 (*lp)->next = m; 1173 else *fp = m; 1174 *lp = m; 1175 } 1176 1177 void mesh_list_remove(struct mesh_state* m, struct mesh_state** fp, 1178 struct mesh_state** lp) 1179 { 1180 if(m->next) 1181 m->next->prev = m->prev; 1182 else *lp = m->prev; 1183 if(m->prev) 1184 m->prev->next = m->next; 1185 else *fp = m->next; 1186 } 1187