1 /* 2 * daemon/worker.c - worker that handles a pending list of requests. 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 implements the worker that handles callbacks on events, for 40 * pending requests. 41 */ 42 #include "config.h" 43 #include "util/log.h" 44 #include "util/net_help.h" 45 #include "util/random.h" 46 #include "daemon/worker.h" 47 #include "daemon/daemon.h" 48 #include "daemon/remote.h" 49 #include "daemon/acl_list.h" 50 #include "util/netevent.h" 51 #include "util/config_file.h" 52 #include "util/module.h" 53 #include "util/regional.h" 54 #include "util/storage/slabhash.h" 55 #include "services/listen_dnsport.h" 56 #include "services/outside_network.h" 57 #include "services/outbound_list.h" 58 #include "services/cache/rrset.h" 59 #include "services/cache/infra.h" 60 #include "services/cache/dns.h" 61 #include "services/authzone.h" 62 #include "services/mesh.h" 63 #include "services/localzone.h" 64 #include "util/data/msgparse.h" 65 #include "util/data/msgencode.h" 66 #include "util/data/dname.h" 67 #include "util/fptr_wlist.h" 68 #include "util/tube.h" 69 #include "util/edns.h" 70 #include "iterator/iter_fwd.h" 71 #include "iterator/iter_hints.h" 72 #include "validator/autotrust.h" 73 #include "validator/val_anchor.h" 74 #include "respip/respip.h" 75 #include "libunbound/context.h" 76 #include "libunbound/libworker.h" 77 #include "sldns/sbuffer.h" 78 #include "sldns/wire2str.h" 79 #include "util/shm_side/shm_main.h" 80 #include "dnscrypt/dnscrypt.h" 81 82 #ifdef HAVE_SYS_TYPES_H 83 # include <sys/types.h> 84 #endif 85 #ifdef HAVE_NETDB_H 86 #include <netdb.h> 87 #endif 88 #include <signal.h> 89 #ifdef UB_ON_WINDOWS 90 #include "winrc/win_svc.h" 91 #endif 92 93 /** Size of an UDP datagram */ 94 #define NORMAL_UDP_SIZE 512 /* bytes */ 95 /** ratelimit for error responses */ 96 #define ERROR_RATELIMIT 100 /* qps */ 97 98 /** 99 * seconds to add to prefetch leeway. This is a TTL that expires old rrsets 100 * earlier than they should in order to put the new update into the cache. 101 * This additional value is to make sure that if not all TTLs are equal in 102 * the message to be updated(and replaced), that rrsets with up to this much 103 * extra TTL are also replaced. This means that the resulting new message 104 * will have (most likely) this TTL at least, avoiding very small 'split 105 * second' TTLs due to operators choosing relative primes for TTLs (or so). 106 * Also has to be at least one to break ties (and overwrite cached entry). 107 */ 108 #define PREFETCH_EXPIRY_ADD 60 109 110 /** Report on memory usage by this thread and global */ 111 static void 112 worker_mem_report(struct worker* ATTR_UNUSED(worker), 113 struct serviced_query* ATTR_UNUSED(cur_serv)) 114 { 115 #ifdef UNBOUND_ALLOC_STATS 116 /* measure memory leakage */ 117 extern size_t unbound_mem_alloc, unbound_mem_freed; 118 /* debug func in validator module */ 119 size_t total, front, back, mesh, msg, rrset, infra, ac, superac; 120 size_t me, iter, val, anch; 121 int i; 122 #ifdef CLIENT_SUBNET 123 size_t subnet = 0; 124 #endif /* CLIENT_SUBNET */ 125 if(verbosity < VERB_ALGO) 126 return; 127 front = listen_get_mem(worker->front); 128 back = outnet_get_mem(worker->back); 129 msg = slabhash_get_mem(worker->env.msg_cache); 130 rrset = slabhash_get_mem(&worker->env.rrset_cache->table); 131 infra = infra_get_mem(worker->env.infra_cache); 132 mesh = mesh_get_mem(worker->env.mesh); 133 ac = alloc_get_mem(&worker->alloc); 134 superac = alloc_get_mem(&worker->daemon->superalloc); 135 anch = anchors_get_mem(worker->env.anchors); 136 iter = 0; 137 val = 0; 138 for(i=0; i<worker->env.mesh->mods.num; i++) { 139 fptr_ok(fptr_whitelist_mod_get_mem(worker->env.mesh-> 140 mods.mod[i]->get_mem)); 141 if(strcmp(worker->env.mesh->mods.mod[i]->name, "validator")==0) 142 val += (*worker->env.mesh->mods.mod[i]->get_mem) 143 (&worker->env, i); 144 #ifdef CLIENT_SUBNET 145 else if(strcmp(worker->env.mesh->mods.mod[i]->name, 146 "subnet")==0) 147 subnet += (*worker->env.mesh->mods.mod[i]->get_mem) 148 (&worker->env, i); 149 #endif /* CLIENT_SUBNET */ 150 else iter += (*worker->env.mesh->mods.mod[i]->get_mem) 151 (&worker->env, i); 152 } 153 me = sizeof(*worker) + sizeof(*worker->base) + sizeof(*worker->comsig) 154 + comm_point_get_mem(worker->cmd_com) 155 + sizeof(worker->rndstate) 156 + regional_get_mem(worker->scratchpad) 157 + sizeof(*worker->env.scratch_buffer) 158 + sldns_buffer_capacity(worker->env.scratch_buffer) 159 + forwards_get_mem(worker->env.fwds) 160 + hints_get_mem(worker->env.hints); 161 if(worker->thread_num == 0) 162 me += acl_list_get_mem(worker->daemon->acl); 163 if(cur_serv) { 164 me += serviced_get_mem(cur_serv); 165 } 166 total = front+back+mesh+msg+rrset+infra+iter+val+ac+superac+me; 167 #ifdef CLIENT_SUBNET 168 total += subnet; 169 log_info("Memory conditions: %u front=%u back=%u mesh=%u msg=%u " 170 "rrset=%u infra=%u iter=%u val=%u subnet=%u anchors=%u " 171 "alloccache=%u globalalloccache=%u me=%u", 172 (unsigned)total, (unsigned)front, (unsigned)back, 173 (unsigned)mesh, (unsigned)msg, (unsigned)rrset, (unsigned)infra, 174 (unsigned)iter, (unsigned)val, 175 (unsigned)subnet, (unsigned)anch, (unsigned)ac, 176 (unsigned)superac, (unsigned)me); 177 #else /* no CLIENT_SUBNET */ 178 log_info("Memory conditions: %u front=%u back=%u mesh=%u msg=%u " 179 "rrset=%u infra=%u iter=%u val=%u anchors=%u " 180 "alloccache=%u globalalloccache=%u me=%u", 181 (unsigned)total, (unsigned)front, (unsigned)back, 182 (unsigned)mesh, (unsigned)msg, (unsigned)rrset, 183 (unsigned)infra, (unsigned)iter, (unsigned)val, (unsigned)anch, 184 (unsigned)ac, (unsigned)superac, (unsigned)me); 185 #endif /* CLIENT_SUBNET */ 186 log_info("Total heap memory estimate: %u total-alloc: %u " 187 "total-free: %u", (unsigned)total, 188 (unsigned)unbound_mem_alloc, (unsigned)unbound_mem_freed); 189 #else /* no UNBOUND_ALLOC_STATS */ 190 size_t val = 0; 191 #ifdef CLIENT_SUBNET 192 size_t subnet = 0; 193 #endif /* CLIENT_SUBNET */ 194 int i; 195 if(verbosity < VERB_QUERY) 196 return; 197 for(i=0; i<worker->env.mesh->mods.num; i++) { 198 fptr_ok(fptr_whitelist_mod_get_mem(worker->env.mesh-> 199 mods.mod[i]->get_mem)); 200 if(strcmp(worker->env.mesh->mods.mod[i]->name, "validator")==0) 201 val += (*worker->env.mesh->mods.mod[i]->get_mem) 202 (&worker->env, i); 203 #ifdef CLIENT_SUBNET 204 else if(strcmp(worker->env.mesh->mods.mod[i]->name, 205 "subnet")==0) 206 subnet += (*worker->env.mesh->mods.mod[i]->get_mem) 207 (&worker->env, i); 208 #endif /* CLIENT_SUBNET */ 209 } 210 #ifdef CLIENT_SUBNET 211 verbose(VERB_QUERY, "cache memory msg=%u rrset=%u infra=%u val=%u " 212 "subnet=%u", 213 (unsigned)slabhash_get_mem(worker->env.msg_cache), 214 (unsigned)slabhash_get_mem(&worker->env.rrset_cache->table), 215 (unsigned)infra_get_mem(worker->env.infra_cache), 216 (unsigned)val, (unsigned)subnet); 217 #else /* no CLIENT_SUBNET */ 218 verbose(VERB_QUERY, "cache memory msg=%u rrset=%u infra=%u val=%u", 219 (unsigned)slabhash_get_mem(worker->env.msg_cache), 220 (unsigned)slabhash_get_mem(&worker->env.rrset_cache->table), 221 (unsigned)infra_get_mem(worker->env.infra_cache), 222 (unsigned)val); 223 #endif /* CLIENT_SUBNET */ 224 #endif /* UNBOUND_ALLOC_STATS */ 225 } 226 227 void 228 worker_send_cmd(struct worker* worker, enum worker_commands cmd) 229 { 230 uint32_t c = (uint32_t)htonl(cmd); 231 if(!tube_write_msg(worker->cmd, (uint8_t*)&c, sizeof(c), 0)) { 232 log_err("worker send cmd %d failed", (int)cmd); 233 } 234 } 235 236 int 237 worker_handle_reply(struct comm_point* c, void* arg, int error, 238 struct comm_reply* reply_info) 239 { 240 struct module_qstate* q = (struct module_qstate*)arg; 241 struct worker* worker = q->env->worker; 242 struct outbound_entry e; 243 e.qstate = q; 244 e.qsent = NULL; 245 246 if(error != 0) { 247 mesh_report_reply(worker->env.mesh, &e, reply_info, error); 248 worker_mem_report(worker, NULL); 249 return 0; 250 } 251 /* sanity check. */ 252 if(!LDNS_QR_WIRE(sldns_buffer_begin(c->buffer)) 253 || LDNS_OPCODE_WIRE(sldns_buffer_begin(c->buffer)) != 254 LDNS_PACKET_QUERY 255 || LDNS_QDCOUNT(sldns_buffer_begin(c->buffer)) > 1) { 256 /* error becomes timeout for the module as if this reply 257 * never arrived. */ 258 mesh_report_reply(worker->env.mesh, &e, reply_info, 259 NETEVENT_TIMEOUT); 260 worker_mem_report(worker, NULL); 261 return 0; 262 } 263 mesh_report_reply(worker->env.mesh, &e, reply_info, NETEVENT_NOERROR); 264 worker_mem_report(worker, NULL); 265 return 0; 266 } 267 268 int 269 worker_handle_service_reply(struct comm_point* c, void* arg, int error, 270 struct comm_reply* reply_info) 271 { 272 struct outbound_entry* e = (struct outbound_entry*)arg; 273 struct worker* worker = e->qstate->env->worker; 274 struct serviced_query *sq = e->qsent; 275 276 verbose(VERB_ALGO, "worker svcd callback for qstate %p", e->qstate); 277 if(error != 0) { 278 mesh_report_reply(worker->env.mesh, e, reply_info, error); 279 worker_mem_report(worker, sq); 280 return 0; 281 } 282 /* sanity check. */ 283 if(!LDNS_QR_WIRE(sldns_buffer_begin(c->buffer)) 284 || LDNS_OPCODE_WIRE(sldns_buffer_begin(c->buffer)) != 285 LDNS_PACKET_QUERY 286 || LDNS_QDCOUNT(sldns_buffer_begin(c->buffer)) > 1) { 287 /* error becomes timeout for the module as if this reply 288 * never arrived. */ 289 verbose(VERB_ALGO, "worker: bad reply handled as timeout"); 290 mesh_report_reply(worker->env.mesh, e, reply_info, 291 NETEVENT_TIMEOUT); 292 worker_mem_report(worker, sq); 293 return 0; 294 } 295 mesh_report_reply(worker->env.mesh, e, reply_info, NETEVENT_NOERROR); 296 worker_mem_report(worker, sq); 297 return 0; 298 } 299 300 /** ratelimit error replies 301 * @param worker: the worker struct with ratelimit counter 302 * @param err: error code that would be wanted. 303 * @return value of err if okay, or -1 if it should be discarded instead. 304 */ 305 static int 306 worker_err_ratelimit(struct worker* worker, int err) 307 { 308 if(worker->err_limit_time == *worker->env.now) { 309 /* see if limit is exceeded for this second */ 310 if(worker->err_limit_count++ > ERROR_RATELIMIT) 311 return -1; 312 } else { 313 /* new second, new limits */ 314 worker->err_limit_time = *worker->env.now; 315 worker->err_limit_count = 1; 316 } 317 return err; 318 } 319 320 /** check request sanity. 321 * @param pkt: the wire packet to examine for sanity. 322 * @param worker: parameters for checking. 323 * @return error code, 0 OK, or -1 discard. 324 */ 325 static int 326 worker_check_request(sldns_buffer* pkt, struct worker* worker) 327 { 328 if(sldns_buffer_limit(pkt) < LDNS_HEADER_SIZE) { 329 verbose(VERB_QUERY, "request too short, discarded"); 330 return -1; 331 } 332 if(sldns_buffer_limit(pkt) > NORMAL_UDP_SIZE && 333 worker->daemon->cfg->harden_large_queries) { 334 verbose(VERB_QUERY, "request too large, discarded"); 335 return -1; 336 } 337 if(LDNS_QR_WIRE(sldns_buffer_begin(pkt))) { 338 verbose(VERB_QUERY, "request has QR bit on, discarded"); 339 return -1; 340 } 341 if(LDNS_TC_WIRE(sldns_buffer_begin(pkt))) { 342 LDNS_TC_CLR(sldns_buffer_begin(pkt)); 343 verbose(VERB_QUERY, "request bad, has TC bit on"); 344 return worker_err_ratelimit(worker, LDNS_RCODE_FORMERR); 345 } 346 if(LDNS_OPCODE_WIRE(sldns_buffer_begin(pkt)) != LDNS_PACKET_QUERY && 347 LDNS_OPCODE_WIRE(sldns_buffer_begin(pkt)) != LDNS_PACKET_NOTIFY) { 348 verbose(VERB_QUERY, "request unknown opcode %d", 349 LDNS_OPCODE_WIRE(sldns_buffer_begin(pkt))); 350 return worker_err_ratelimit(worker, LDNS_RCODE_NOTIMPL); 351 } 352 if(LDNS_QDCOUNT(sldns_buffer_begin(pkt)) != 1) { 353 verbose(VERB_QUERY, "request wrong nr qd=%d", 354 LDNS_QDCOUNT(sldns_buffer_begin(pkt))); 355 return worker_err_ratelimit(worker, LDNS_RCODE_FORMERR); 356 } 357 if(LDNS_ANCOUNT(sldns_buffer_begin(pkt)) != 0 && 358 (LDNS_ANCOUNT(sldns_buffer_begin(pkt)) != 1 || 359 LDNS_OPCODE_WIRE(sldns_buffer_begin(pkt)) != LDNS_PACKET_NOTIFY)) { 360 verbose(VERB_QUERY, "request wrong nr an=%d", 361 LDNS_ANCOUNT(sldns_buffer_begin(pkt))); 362 return worker_err_ratelimit(worker, LDNS_RCODE_FORMERR); 363 } 364 if(LDNS_NSCOUNT(sldns_buffer_begin(pkt)) != 0) { 365 verbose(VERB_QUERY, "request wrong nr ns=%d", 366 LDNS_NSCOUNT(sldns_buffer_begin(pkt))); 367 return worker_err_ratelimit(worker, LDNS_RCODE_FORMERR); 368 } 369 if(LDNS_ARCOUNT(sldns_buffer_begin(pkt)) > 1) { 370 verbose(VERB_QUERY, "request wrong nr ar=%d", 371 LDNS_ARCOUNT(sldns_buffer_begin(pkt))); 372 return worker_err_ratelimit(worker, LDNS_RCODE_FORMERR); 373 } 374 return 0; 375 } 376 377 void 378 worker_handle_control_cmd(struct tube* ATTR_UNUSED(tube), uint8_t* msg, 379 size_t len, int error, void* arg) 380 { 381 struct worker* worker = (struct worker*)arg; 382 enum worker_commands cmd; 383 if(error != NETEVENT_NOERROR) { 384 free(msg); 385 if(error == NETEVENT_CLOSED) 386 comm_base_exit(worker->base); 387 else log_info("control event: %d", error); 388 return; 389 } 390 if(len != sizeof(uint32_t)) { 391 fatal_exit("bad control msg length %d", (int)len); 392 } 393 cmd = sldns_read_uint32(msg); 394 free(msg); 395 switch(cmd) { 396 case worker_cmd_quit: 397 verbose(VERB_ALGO, "got control cmd quit"); 398 comm_base_exit(worker->base); 399 break; 400 case worker_cmd_stats: 401 verbose(VERB_ALGO, "got control cmd stats"); 402 server_stats_reply(worker, 1); 403 break; 404 case worker_cmd_stats_noreset: 405 verbose(VERB_ALGO, "got control cmd stats_noreset"); 406 server_stats_reply(worker, 0); 407 break; 408 case worker_cmd_remote: 409 verbose(VERB_ALGO, "got control cmd remote"); 410 daemon_remote_exec(worker); 411 break; 412 default: 413 log_err("bad command %d", (int)cmd); 414 break; 415 } 416 } 417 418 /** check if a delegation is secure */ 419 static enum sec_status 420 check_delegation_secure(struct reply_info *rep) 421 { 422 /* return smallest security status */ 423 size_t i; 424 enum sec_status sec = sec_status_secure; 425 enum sec_status s; 426 size_t num = rep->an_numrrsets + rep->ns_numrrsets; 427 /* check if answer and authority are OK */ 428 for(i=0; i<num; i++) { 429 s = ((struct packed_rrset_data*)rep->rrsets[i]->entry.data) 430 ->security; 431 if(s < sec) 432 sec = s; 433 } 434 /* in additional, only unchecked triggers revalidation */ 435 for(i=num; i<rep->rrset_count; i++) { 436 s = ((struct packed_rrset_data*)rep->rrsets[i]->entry.data) 437 ->security; 438 if(s == sec_status_unchecked) 439 return s; 440 } 441 return sec; 442 } 443 444 /** remove nonsecure from a delegation referral additional section */ 445 static void 446 deleg_remove_nonsecure_additional(struct reply_info* rep) 447 { 448 /* we can simply edit it, since we are working in the scratch region */ 449 size_t i; 450 enum sec_status s; 451 452 for(i = rep->an_numrrsets+rep->ns_numrrsets; i<rep->rrset_count; i++) { 453 s = ((struct packed_rrset_data*)rep->rrsets[i]->entry.data) 454 ->security; 455 if(s != sec_status_secure) { 456 memmove(rep->rrsets+i, rep->rrsets+i+1, 457 sizeof(struct ub_packed_rrset_key*)* 458 (rep->rrset_count - i - 1)); 459 rep->ar_numrrsets--; 460 rep->rrset_count--; 461 i--; 462 } 463 } 464 } 465 466 /** answer nonrecursive query from the cache */ 467 static int 468 answer_norec_from_cache(struct worker* worker, struct query_info* qinfo, 469 uint16_t id, uint16_t flags, struct comm_reply* repinfo, 470 struct edns_data* edns) 471 { 472 /* for a nonrecursive query return either: 473 * o an error (servfail; we try to avoid this) 474 * o a delegation (closest we have; this routine tries that) 475 * o the answer (checked by answer_from_cache) 476 * 477 * So, grab a delegation from the rrset cache. 478 * Then check if it needs validation, if so, this routine fails, 479 * so that iterator can prime and validator can verify rrsets. 480 */ 481 struct edns_data edns_bak; 482 uint16_t udpsize = edns->udp_size; 483 int secure = 0; 484 time_t timenow = *worker->env.now; 485 int must_validate = (!(flags&BIT_CD) || worker->env.cfg->ignore_cd) 486 && worker->env.need_to_validate; 487 struct dns_msg *msg = NULL; 488 struct delegpt *dp; 489 490 dp = dns_cache_find_delegation(&worker->env, qinfo->qname, 491 qinfo->qname_len, qinfo->qtype, qinfo->qclass, 492 worker->scratchpad, &msg, timenow); 493 if(!dp) { /* no delegation, need to reprime */ 494 return 0; 495 } 496 /* In case we have a local alias, copy it into the delegation message. 497 * Shallow copy should be fine, as we'll be done with msg in this 498 * function. */ 499 msg->qinfo.local_alias = qinfo->local_alias; 500 if(must_validate) { 501 switch(check_delegation_secure(msg->rep)) { 502 case sec_status_unchecked: 503 /* some rrsets have not been verified yet, go and 504 * let validator do that */ 505 return 0; 506 case sec_status_bogus: 507 case sec_status_secure_sentinel_fail: 508 /* some rrsets are bogus, reply servfail */ 509 edns->edns_version = EDNS_ADVERTISED_VERSION; 510 edns->udp_size = EDNS_ADVERTISED_SIZE; 511 edns->ext_rcode = 0; 512 edns->bits &= EDNS_DO; 513 if(!inplace_cb_reply_servfail_call(&worker->env, qinfo, NULL, 514 msg->rep, LDNS_RCODE_SERVFAIL, edns, repinfo, worker->scratchpad)) 515 return 0; 516 error_encode(repinfo->c->buffer, LDNS_RCODE_SERVFAIL, 517 &msg->qinfo, id, flags, edns); 518 if(worker->stats.extended) { 519 worker->stats.ans_bogus++; 520 worker->stats.ans_rcode[LDNS_RCODE_SERVFAIL]++; 521 } 522 return 1; 523 case sec_status_secure: 524 /* all rrsets are secure */ 525 /* remove non-secure rrsets from the add. section*/ 526 if(worker->env.cfg->val_clean_additional) 527 deleg_remove_nonsecure_additional(msg->rep); 528 secure = 1; 529 break; 530 case sec_status_indeterminate: 531 case sec_status_insecure: 532 default: 533 /* not secure */ 534 secure = 0; 535 break; 536 } 537 } 538 /* return this delegation from the cache */ 539 edns_bak = *edns; 540 edns->edns_version = EDNS_ADVERTISED_VERSION; 541 edns->udp_size = EDNS_ADVERTISED_SIZE; 542 edns->ext_rcode = 0; 543 edns->bits &= EDNS_DO; 544 if(!inplace_cb_reply_cache_call(&worker->env, qinfo, NULL, msg->rep, 545 (int)(flags&LDNS_RCODE_MASK), edns, repinfo, worker->scratchpad)) 546 return 0; 547 msg->rep->flags |= BIT_QR|BIT_RA; 548 if(!apply_edns_options(edns, &edns_bak, worker->env.cfg, 549 repinfo->c, worker->scratchpad) || 550 !reply_info_answer_encode(&msg->qinfo, msg->rep, id, flags, 551 repinfo->c->buffer, 0, 1, worker->scratchpad, 552 udpsize, edns, (int)(edns->bits & EDNS_DO), secure)) { 553 if(!inplace_cb_reply_servfail_call(&worker->env, qinfo, NULL, NULL, 554 LDNS_RCODE_SERVFAIL, edns, repinfo, worker->scratchpad)) 555 edns->opt_list = NULL; 556 error_encode(repinfo->c->buffer, LDNS_RCODE_SERVFAIL, 557 &msg->qinfo, id, flags, edns); 558 } 559 if(worker->stats.extended) { 560 if(secure) worker->stats.ans_secure++; 561 server_stats_insrcode(&worker->stats, repinfo->c->buffer); 562 } 563 return 1; 564 } 565 566 /** Apply, if applicable, a response IP action to a cached answer. 567 * If the answer is rewritten as a result of an action, '*encode_repp' will 568 * point to the reply info containing the modified answer. '*encode_repp' will 569 * be intact otherwise. 570 * It returns 1 on success, 0 otherwise. */ 571 static int 572 apply_respip_action(struct worker* worker, const struct query_info* qinfo, 573 struct respip_client_info* cinfo, struct reply_info* rep, 574 struct comm_reply* repinfo, struct ub_packed_rrset_key** alias_rrset, 575 struct reply_info** encode_repp) 576 { 577 struct respip_action_info actinfo = {respip_none, NULL}; 578 579 if(qinfo->qtype != LDNS_RR_TYPE_A && 580 qinfo->qtype != LDNS_RR_TYPE_AAAA && 581 qinfo->qtype != LDNS_RR_TYPE_ANY) 582 return 1; 583 584 if(!respip_rewrite_reply(qinfo, cinfo, rep, encode_repp, &actinfo, 585 alias_rrset, 0, worker->scratchpad)) 586 return 0; 587 588 /* xxx_deny actions mean dropping the reply, unless the original reply 589 * was redirected to response-ip data. */ 590 if((actinfo.action == respip_deny || 591 actinfo.action == respip_inform_deny) && 592 *encode_repp == rep) 593 *encode_repp = NULL; 594 595 /* If address info is returned, it means the action should be an 596 * 'inform' variant and the information should be logged. */ 597 if(actinfo.addrinfo) { 598 respip_inform_print(actinfo.addrinfo, qinfo->qname, 599 qinfo->qtype, qinfo->qclass, qinfo->local_alias, 600 repinfo); 601 } 602 603 return 1; 604 } 605 606 /** answer query from the cache. 607 * Normally, the answer message will be built in repinfo->c->buffer; if the 608 * answer is supposed to be suppressed or the answer is supposed to be an 609 * incomplete CNAME chain, the buffer is explicitly cleared to signal the 610 * caller as such. In the latter case *partial_rep will point to the incomplete 611 * reply, and this function is (possibly) supposed to be called again with that 612 * *partial_rep value to complete the chain. In addition, if the query should 613 * be completely dropped, '*need_drop' will be set to 1. */ 614 static int 615 answer_from_cache(struct worker* worker, struct query_info* qinfo, 616 struct respip_client_info* cinfo, int* need_drop, 617 struct ub_packed_rrset_key** alias_rrset, 618 struct reply_info** partial_repp, 619 struct reply_info* rep, uint16_t id, uint16_t flags, 620 struct comm_reply* repinfo, struct edns_data* edns) 621 { 622 struct edns_data edns_bak; 623 time_t timenow = *worker->env.now; 624 uint16_t udpsize = edns->udp_size; 625 struct reply_info* encode_rep = rep; 626 struct reply_info* partial_rep = *partial_repp; 627 int secure; 628 int must_validate = (!(flags&BIT_CD) || worker->env.cfg->ignore_cd) 629 && worker->env.need_to_validate; 630 *partial_repp = NULL; /* avoid accidental further pass */ 631 if(worker->env.cfg->serve_expired) { 632 if(worker->env.cfg->serve_expired_ttl && 633 rep->serve_expired_ttl < timenow) 634 return 0; 635 if(!rrset_array_lock(rep->ref, rep->rrset_count, 0)) 636 return 0; 637 /* below, rrsets with ttl before timenow become TTL 0 in 638 * the response */ 639 /* This response was served with zero TTL */ 640 if (timenow >= rep->ttl) { 641 worker->stats.zero_ttl_responses++; 642 } 643 } else { 644 /* see if it is possible */ 645 if(rep->ttl < timenow) { 646 /* the rrsets may have been updated in the meantime. 647 * we will refetch the message format from the 648 * authoritative server 649 */ 650 return 0; 651 } 652 if(!rrset_array_lock(rep->ref, rep->rrset_count, timenow)) 653 return 0; 654 /* locked and ids and ttls are OK. */ 655 } 656 /* check CNAME chain (if any) */ 657 if(rep->an_numrrsets > 0 && (rep->rrsets[0]->rk.type == 658 htons(LDNS_RR_TYPE_CNAME) || rep->rrsets[0]->rk.type == 659 htons(LDNS_RR_TYPE_DNAME))) { 660 if(!reply_check_cname_chain(qinfo, rep)) { 661 /* cname chain invalid, redo iterator steps */ 662 verbose(VERB_ALGO, "Cache reply: cname chain broken"); 663 bail_out: 664 rrset_array_unlock_touch(worker->env.rrset_cache, 665 worker->scratchpad, rep->ref, rep->rrset_count); 666 return 0; 667 } 668 } 669 /* check security status of the cached answer */ 670 if(must_validate && (rep->security == sec_status_bogus || 671 rep->security == sec_status_secure_sentinel_fail)) { 672 /* BAD cached */ 673 edns->edns_version = EDNS_ADVERTISED_VERSION; 674 edns->udp_size = EDNS_ADVERTISED_SIZE; 675 edns->ext_rcode = 0; 676 edns->bits &= EDNS_DO; 677 if(!inplace_cb_reply_servfail_call(&worker->env, qinfo, NULL, rep, 678 LDNS_RCODE_SERVFAIL, edns, repinfo, worker->scratchpad)) 679 goto bail_out; 680 error_encode(repinfo->c->buffer, LDNS_RCODE_SERVFAIL, 681 qinfo, id, flags, edns); 682 rrset_array_unlock_touch(worker->env.rrset_cache, 683 worker->scratchpad, rep->ref, rep->rrset_count); 684 if(worker->stats.extended) { 685 worker->stats.ans_bogus ++; 686 worker->stats.ans_rcode[LDNS_RCODE_SERVFAIL] ++; 687 } 688 return 1; 689 } else if( rep->security == sec_status_unchecked && must_validate) { 690 verbose(VERB_ALGO, "Cache reply: unchecked entry needs " 691 "validation"); 692 goto bail_out; /* need to validate cache entry first */ 693 } else if(rep->security == sec_status_secure) { 694 if(reply_all_rrsets_secure(rep)) 695 secure = 1; 696 else { 697 if(must_validate) { 698 verbose(VERB_ALGO, "Cache reply: secure entry" 699 " changed status"); 700 goto bail_out; /* rrset changed, re-verify */ 701 } 702 secure = 0; 703 } 704 } else secure = 0; 705 706 edns_bak = *edns; 707 edns->edns_version = EDNS_ADVERTISED_VERSION; 708 edns->udp_size = EDNS_ADVERTISED_SIZE; 709 edns->ext_rcode = 0; 710 edns->bits &= EDNS_DO; 711 if(!inplace_cb_reply_cache_call(&worker->env, qinfo, NULL, rep, 712 (int)(flags&LDNS_RCODE_MASK), edns, repinfo, worker->scratchpad)) 713 goto bail_out; 714 *alias_rrset = NULL; /* avoid confusion if caller set it to non-NULL */ 715 if(worker->daemon->use_response_ip && !partial_rep && 716 !apply_respip_action(worker, qinfo, cinfo, rep, repinfo, alias_rrset, 717 &encode_rep)) { 718 goto bail_out; 719 } else if(partial_rep && 720 !respip_merge_cname(partial_rep, qinfo, rep, cinfo, 721 must_validate, &encode_rep, worker->scratchpad)) { 722 goto bail_out; 723 } 724 if(encode_rep != rep) 725 secure = 0; /* if rewritten, it can't be considered "secure" */ 726 if(!encode_rep || *alias_rrset) { 727 sldns_buffer_clear(repinfo->c->buffer); 728 sldns_buffer_flip(repinfo->c->buffer); 729 if(!encode_rep) 730 *need_drop = 1; 731 else { 732 /* If a partial CNAME chain is found, we first need to 733 * make a copy of the reply in the scratchpad so we 734 * can release the locks and lookup the cache again. */ 735 *partial_repp = reply_info_copy(encode_rep, NULL, 736 worker->scratchpad); 737 if(!*partial_repp) 738 goto bail_out; 739 } 740 } else if(!apply_edns_options(edns, &edns_bak, worker->env.cfg, 741 repinfo->c, worker->scratchpad) || 742 !reply_info_answer_encode(qinfo, encode_rep, id, flags, 743 repinfo->c->buffer, timenow, 1, worker->scratchpad, 744 udpsize, edns, (int)(edns->bits & EDNS_DO), secure)) { 745 if(!inplace_cb_reply_servfail_call(&worker->env, qinfo, NULL, NULL, 746 LDNS_RCODE_SERVFAIL, edns, repinfo, worker->scratchpad)) 747 edns->opt_list = NULL; 748 error_encode(repinfo->c->buffer, LDNS_RCODE_SERVFAIL, 749 qinfo, id, flags, edns); 750 } 751 /* cannot send the reply right now, because blocking network syscall 752 * is bad while holding locks. */ 753 rrset_array_unlock_touch(worker->env.rrset_cache, worker->scratchpad, 754 rep->ref, rep->rrset_count); 755 if(worker->stats.extended) { 756 if(secure) worker->stats.ans_secure++; 757 server_stats_insrcode(&worker->stats, repinfo->c->buffer); 758 } 759 /* go and return this buffer to the client */ 760 return 1; 761 } 762 763 /** Reply to client and perform prefetch to keep cache up to date. 764 * If the buffer for the reply is empty, it indicates that only prefetch is 765 * necessary and the reply should be suppressed (because it's dropped or 766 * being deferred). */ 767 static void 768 reply_and_prefetch(struct worker* worker, struct query_info* qinfo, 769 uint16_t flags, struct comm_reply* repinfo, time_t leeway) 770 { 771 /* first send answer to client to keep its latency 772 * as small as a cachereply */ 773 if(sldns_buffer_limit(repinfo->c->buffer) != 0) 774 comm_point_send_reply(repinfo); 775 server_stats_prefetch(&worker->stats, worker); 776 777 /* create the prefetch in the mesh as a normal lookup without 778 * client addrs waiting, which has the cache blacklisted (to bypass 779 * the cache and go to the network for the data). */ 780 /* this (potentially) runs the mesh for the new query */ 781 mesh_new_prefetch(worker->env.mesh, qinfo, flags, leeway + 782 PREFETCH_EXPIRY_ADD); 783 } 784 785 /** 786 * Fill CH class answer into buffer. Keeps query. 787 * @param pkt: buffer 788 * @param str: string to put into text record (<255). 789 * array of strings, every string becomes a text record. 790 * @param num: number of strings in array. 791 * @param edns: edns reply information. 792 * @param worker: worker with scratch region. 793 * @param repinfo: reply information for a communication point. 794 */ 795 static void 796 chaos_replystr(sldns_buffer* pkt, char** str, int num, struct edns_data* edns, 797 struct worker* worker, struct comm_reply* repinfo) 798 { 799 int i; 800 unsigned int rd = LDNS_RD_WIRE(sldns_buffer_begin(pkt)); 801 unsigned int cd = LDNS_CD_WIRE(sldns_buffer_begin(pkt)); 802 sldns_buffer_clear(pkt); 803 sldns_buffer_skip(pkt, (ssize_t)sizeof(uint16_t)); /* skip id */ 804 sldns_buffer_write_u16(pkt, (uint16_t)(BIT_QR|BIT_RA)); 805 if(rd) LDNS_RD_SET(sldns_buffer_begin(pkt)); 806 if(cd) LDNS_CD_SET(sldns_buffer_begin(pkt)); 807 sldns_buffer_write_u16(pkt, 1); /* qdcount */ 808 sldns_buffer_write_u16(pkt, (uint16_t)num); /* ancount */ 809 sldns_buffer_write_u16(pkt, 0); /* nscount */ 810 sldns_buffer_write_u16(pkt, 0); /* arcount */ 811 (void)query_dname_len(pkt); /* skip qname */ 812 sldns_buffer_skip(pkt, (ssize_t)sizeof(uint16_t)); /* skip qtype */ 813 sldns_buffer_skip(pkt, (ssize_t)sizeof(uint16_t)); /* skip qclass */ 814 for(i=0; i<num; i++) { 815 size_t len = strlen(str[i]); 816 if(len>255) len=255; /* cap size of TXT record */ 817 sldns_buffer_write_u16(pkt, 0xc00c); /* compr ptr to query */ 818 sldns_buffer_write_u16(pkt, LDNS_RR_TYPE_TXT); 819 sldns_buffer_write_u16(pkt, LDNS_RR_CLASS_CH); 820 sldns_buffer_write_u32(pkt, 0); /* TTL */ 821 sldns_buffer_write_u16(pkt, sizeof(uint8_t) + len); 822 sldns_buffer_write_u8(pkt, len); 823 sldns_buffer_write(pkt, str[i], len); 824 } 825 sldns_buffer_flip(pkt); 826 edns->edns_version = EDNS_ADVERTISED_VERSION; 827 edns->udp_size = EDNS_ADVERTISED_SIZE; 828 edns->bits &= EDNS_DO; 829 if(!inplace_cb_reply_local_call(&worker->env, NULL, NULL, NULL, 830 LDNS_RCODE_NOERROR, edns, repinfo, worker->scratchpad)) 831 edns->opt_list = NULL; 832 if(sldns_buffer_capacity(pkt) >= 833 sldns_buffer_limit(pkt)+calc_edns_field_size(edns)) 834 attach_edns_record(pkt, edns); 835 } 836 837 /** Reply with one string */ 838 static void 839 chaos_replyonestr(sldns_buffer* pkt, const char* str, struct edns_data* edns, 840 struct worker* worker, struct comm_reply* repinfo) 841 { 842 chaos_replystr(pkt, (char**)&str, 1, edns, worker, repinfo); 843 } 844 845 /** 846 * Create CH class trustanchor answer. 847 * @param pkt: buffer 848 * @param edns: edns reply information. 849 * @param w: worker with scratch region. 850 * @param repinfo: reply information for a communication point. 851 */ 852 static void 853 chaos_trustanchor(sldns_buffer* pkt, struct edns_data* edns, struct worker* w, 854 struct comm_reply* repinfo) 855 { 856 #define TA_RESPONSE_MAX_TXT 16 /* max number of TXT records */ 857 #define TA_RESPONSE_MAX_TAGS 32 /* max number of tags printed per zone */ 858 char* str_array[TA_RESPONSE_MAX_TXT]; 859 uint16_t tags[TA_RESPONSE_MAX_TAGS]; 860 int num = 0; 861 struct trust_anchor* ta; 862 863 if(!w->env.need_to_validate) { 864 /* no validator module, reply no trustanchors */ 865 chaos_replystr(pkt, NULL, 0, edns, w, repinfo); 866 return; 867 } 868 869 /* fill the string with contents */ 870 lock_basic_lock(&w->env.anchors->lock); 871 RBTREE_FOR(ta, struct trust_anchor*, w->env.anchors->tree) { 872 char* str; 873 size_t i, numtag, str_len = 255; 874 if(num == TA_RESPONSE_MAX_TXT) continue; 875 str = (char*)regional_alloc(w->scratchpad, str_len); 876 if(!str) continue; 877 lock_basic_lock(&ta->lock); 878 numtag = anchor_list_keytags(ta, tags, TA_RESPONSE_MAX_TAGS); 879 if(numtag == 0) { 880 /* empty, insecure point */ 881 lock_basic_unlock(&ta->lock); 882 continue; 883 } 884 str_array[num] = str; 885 num++; 886 887 /* spool name of anchor */ 888 (void)sldns_wire2str_dname_buf(ta->name, ta->namelen, str, str_len); 889 str_len -= strlen(str); str += strlen(str); 890 /* spool tags */ 891 for(i=0; i<numtag; i++) { 892 snprintf(str, str_len, " %u", (unsigned)tags[i]); 893 str_len -= strlen(str); str += strlen(str); 894 } 895 lock_basic_unlock(&ta->lock); 896 } 897 lock_basic_unlock(&w->env.anchors->lock); 898 899 chaos_replystr(pkt, str_array, num, edns, w, repinfo); 900 regional_free_all(w->scratchpad); 901 } 902 903 /** 904 * Answer CH class queries. 905 * @param w: worker 906 * @param qinfo: query info. Pointer into packet buffer. 907 * @param edns: edns info from query. 908 * @param repinfo: reply information for a communication point. 909 * @param pkt: packet buffer. 910 * @return: true if a reply is to be sent. 911 */ 912 static int 913 answer_chaos(struct worker* w, struct query_info* qinfo, 914 struct edns_data* edns, struct comm_reply* repinfo, sldns_buffer* pkt) 915 { 916 struct config_file* cfg = w->env.cfg; 917 if(qinfo->qtype != LDNS_RR_TYPE_ANY && qinfo->qtype != LDNS_RR_TYPE_TXT) 918 return 0; 919 if(query_dname_compare(qinfo->qname, 920 (uint8_t*)"\002id\006server") == 0 || 921 query_dname_compare(qinfo->qname, 922 (uint8_t*)"\010hostname\004bind") == 0) 923 { 924 if(cfg->hide_identity) 925 return 0; 926 if(cfg->identity==NULL || cfg->identity[0]==0) { 927 char buf[MAXHOSTNAMELEN+1]; 928 if (gethostname(buf, MAXHOSTNAMELEN) == 0) { 929 buf[MAXHOSTNAMELEN] = 0; 930 chaos_replyonestr(pkt, buf, edns, w, repinfo); 931 } else { 932 log_err("gethostname: %s", strerror(errno)); 933 chaos_replyonestr(pkt, "no hostname", edns, w, repinfo); 934 } 935 } 936 else chaos_replyonestr(pkt, cfg->identity, edns, w, repinfo); 937 return 1; 938 } 939 if(query_dname_compare(qinfo->qname, 940 (uint8_t*)"\007version\006server") == 0 || 941 query_dname_compare(qinfo->qname, 942 (uint8_t*)"\007version\004bind") == 0) 943 { 944 if(cfg->hide_version) 945 return 0; 946 if(cfg->version==NULL || cfg->version[0]==0) 947 chaos_replyonestr(pkt, PACKAGE_STRING, edns, w, repinfo); 948 else chaos_replyonestr(pkt, cfg->version, edns, w, repinfo); 949 return 1; 950 } 951 if(query_dname_compare(qinfo->qname, 952 (uint8_t*)"\013trustanchor\007unbound") == 0) 953 { 954 if(cfg->hide_trustanchor) 955 return 0; 956 chaos_trustanchor(pkt, edns, w, repinfo); 957 return 1; 958 } 959 960 return 0; 961 } 962 963 /** 964 * Answer notify queries. These are notifies for authoritative zones, 965 * the reply is an ack that the notify has been received. We need to check 966 * access permission here. 967 * @param w: worker 968 * @param qinfo: query info. Pointer into packet buffer. 969 * @param edns: edns info from query. 970 * @param repinfo: reply info with source address. 971 * @param pkt: packet buffer. 972 */ 973 static void 974 answer_notify(struct worker* w, struct query_info* qinfo, 975 struct edns_data* edns, sldns_buffer* pkt, struct comm_reply* repinfo) 976 { 977 int refused = 0; 978 int rcode = LDNS_RCODE_NOERROR; 979 uint32_t serial = 0; 980 int has_serial; 981 if(!w->env.auth_zones) return; 982 has_serial = auth_zone_parse_notify_serial(pkt, &serial); 983 if(auth_zones_notify(w->env.auth_zones, &w->env, qinfo->qname, 984 qinfo->qname_len, qinfo->qclass, &repinfo->addr, 985 repinfo->addrlen, has_serial, serial, &refused)) { 986 rcode = LDNS_RCODE_NOERROR; 987 } else { 988 if(refused) 989 rcode = LDNS_RCODE_REFUSED; 990 else rcode = LDNS_RCODE_SERVFAIL; 991 } 992 993 if(verbosity >= VERB_DETAIL) { 994 char buf[380]; 995 char zname[255+1]; 996 char sr[25]; 997 dname_str(qinfo->qname, zname); 998 sr[0]=0; 999 if(has_serial) 1000 snprintf(sr, sizeof(sr), "serial %u ", 1001 (unsigned)serial); 1002 if(rcode == LDNS_RCODE_REFUSED) 1003 snprintf(buf, sizeof(buf), 1004 "refused NOTIFY %sfor %s from", sr, zname); 1005 else if(rcode == LDNS_RCODE_SERVFAIL) 1006 snprintf(buf, sizeof(buf), 1007 "servfail for NOTIFY %sfor %s from", sr, zname); 1008 else snprintf(buf, sizeof(buf), 1009 "received NOTIFY %sfor %s from", sr, zname); 1010 log_addr(VERB_DETAIL, buf, &repinfo->addr, repinfo->addrlen); 1011 } 1012 edns->edns_version = EDNS_ADVERTISED_VERSION; 1013 edns->udp_size = EDNS_ADVERTISED_SIZE; 1014 edns->ext_rcode = 0; 1015 edns->bits &= EDNS_DO; 1016 edns->opt_list = NULL; 1017 error_encode(pkt, rcode, qinfo, 1018 *(uint16_t*)(void *)sldns_buffer_begin(pkt), 1019 sldns_buffer_read_u16_at(pkt, 2), edns); 1020 LDNS_OPCODE_SET(sldns_buffer_begin(pkt), LDNS_PACKET_NOTIFY); 1021 } 1022 1023 static int 1024 deny_refuse(struct comm_point* c, enum acl_access acl, 1025 enum acl_access deny, enum acl_access refuse, 1026 struct worker* worker, struct comm_reply* repinfo) 1027 { 1028 if(acl == deny) { 1029 comm_point_drop_reply(repinfo); 1030 if(worker->stats.extended) 1031 worker->stats.unwanted_queries++; 1032 return 0; 1033 } else if(acl == refuse) { 1034 log_addr(VERB_ALGO, "refused query from", 1035 &repinfo->addr, repinfo->addrlen); 1036 log_buf(VERB_ALGO, "refuse", c->buffer); 1037 if(worker->stats.extended) 1038 worker->stats.unwanted_queries++; 1039 if(worker_check_request(c->buffer, worker) == -1) { 1040 comm_point_drop_reply(repinfo); 1041 return 0; /* discard this */ 1042 } 1043 sldns_buffer_set_limit(c->buffer, LDNS_HEADER_SIZE); 1044 sldns_buffer_write_at(c->buffer, 4, 1045 (uint8_t*)"\0\0\0\0\0\0\0\0", 8); 1046 LDNS_QR_SET(sldns_buffer_begin(c->buffer)); 1047 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), 1048 LDNS_RCODE_REFUSED); 1049 sldns_buffer_set_position(c->buffer, LDNS_HEADER_SIZE); 1050 sldns_buffer_flip(c->buffer); 1051 return 1; 1052 } 1053 1054 return -1; 1055 } 1056 1057 static int 1058 deny_refuse_all(struct comm_point* c, enum acl_access acl, 1059 struct worker* worker, struct comm_reply* repinfo) 1060 { 1061 return deny_refuse(c, acl, acl_deny, acl_refuse, worker, repinfo); 1062 } 1063 1064 static int 1065 deny_refuse_non_local(struct comm_point* c, enum acl_access acl, 1066 struct worker* worker, struct comm_reply* repinfo) 1067 { 1068 return deny_refuse(c, acl, acl_deny_non_local, acl_refuse_non_local, worker, repinfo); 1069 } 1070 1071 int 1072 worker_handle_request(struct comm_point* c, void* arg, int error, 1073 struct comm_reply* repinfo) 1074 { 1075 struct worker* worker = (struct worker*)arg; 1076 int ret; 1077 hashvalue_type h; 1078 struct lruhash_entry* e; 1079 struct query_info qinfo; 1080 struct edns_data edns; 1081 enum acl_access acl; 1082 struct acl_addr* acladdr; 1083 int rc = 0; 1084 int need_drop = 0; 1085 /* We might have to chase a CNAME chain internally, in which case 1086 * we'll have up to two replies and combine them to build a complete 1087 * answer. These variables control this case. */ 1088 struct ub_packed_rrset_key* alias_rrset = NULL; 1089 struct reply_info* partial_rep = NULL; 1090 struct query_info* lookup_qinfo = &qinfo; 1091 struct query_info qinfo_tmp; /* placeholdoer for lookup_qinfo */ 1092 struct respip_client_info* cinfo = NULL, cinfo_tmp; 1093 memset(&qinfo, 0, sizeof(qinfo)); 1094 1095 if(error != NETEVENT_NOERROR) { 1096 /* some bad tcp query DNS formats give these error calls */ 1097 verbose(VERB_ALGO, "handle request called with err=%d", error); 1098 return 0; 1099 } 1100 #ifdef USE_DNSCRYPT 1101 repinfo->max_udp_size = worker->daemon->cfg->max_udp_size; 1102 if(!dnsc_handle_curved_request(worker->daemon->dnscenv, repinfo)) { 1103 worker->stats.num_query_dnscrypt_crypted_malformed++; 1104 return 0; 1105 } 1106 if(c->dnscrypt && !repinfo->is_dnscrypted) { 1107 char buf[LDNS_MAX_DOMAINLEN+1]; 1108 /* Check if this is unencrypted and asking for certs */ 1109 if(worker_check_request(c->buffer, worker) != 0) { 1110 verbose(VERB_ALGO, 1111 "dnscrypt: worker check request: bad query."); 1112 log_addr(VERB_CLIENT,"from",&repinfo->addr, 1113 repinfo->addrlen); 1114 comm_point_drop_reply(repinfo); 1115 return 0; 1116 } 1117 if(!query_info_parse(&qinfo, c->buffer)) { 1118 verbose(VERB_ALGO, 1119 "dnscrypt: worker parse request: formerror."); 1120 log_addr(VERB_CLIENT, "from", &repinfo->addr, 1121 repinfo->addrlen); 1122 comm_point_drop_reply(repinfo); 1123 return 0; 1124 } 1125 dname_str(qinfo.qname, buf); 1126 if(!(qinfo.qtype == LDNS_RR_TYPE_TXT && 1127 strcasecmp(buf, 1128 worker->daemon->dnscenv->provider_name) == 0)) { 1129 verbose(VERB_ALGO, 1130 "dnscrypt: not TXT \"%s\". Received: %s \"%s\"", 1131 worker->daemon->dnscenv->provider_name, 1132 sldns_rr_descript(qinfo.qtype)->_name, 1133 buf); 1134 comm_point_drop_reply(repinfo); 1135 worker->stats.num_query_dnscrypt_cleartext++; 1136 return 0; 1137 } 1138 worker->stats.num_query_dnscrypt_cert++; 1139 sldns_buffer_rewind(c->buffer); 1140 } else if(c->dnscrypt && repinfo->is_dnscrypted) { 1141 worker->stats.num_query_dnscrypt_crypted++; 1142 } 1143 #endif 1144 #ifdef USE_DNSTAP 1145 if(worker->dtenv.log_client_query_messages) 1146 dt_msg_send_client_query(&worker->dtenv, &repinfo->addr, c->type, 1147 c->buffer); 1148 #endif 1149 acladdr = acl_addr_lookup(worker->daemon->acl, &repinfo->addr, 1150 repinfo->addrlen); 1151 acl = acl_get_control(acladdr); 1152 if((ret=deny_refuse_all(c, acl, worker, repinfo)) != -1) 1153 { 1154 if(ret == 1) 1155 goto send_reply; 1156 return ret; 1157 } 1158 if((ret=worker_check_request(c->buffer, worker)) != 0) { 1159 verbose(VERB_ALGO, "worker check request: bad query."); 1160 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen); 1161 if(ret != -1) { 1162 LDNS_QR_SET(sldns_buffer_begin(c->buffer)); 1163 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), ret); 1164 return 1; 1165 } 1166 comm_point_drop_reply(repinfo); 1167 return 0; 1168 } 1169 1170 worker->stats.num_queries++; 1171 1172 /* check if this query should be dropped based on source ip rate limiting */ 1173 if(!infra_ip_ratelimit_inc(worker->env.infra_cache, repinfo, 1174 *worker->env.now)) { 1175 /* See if we are passed through with slip factor */ 1176 if(worker->env.cfg->ip_ratelimit_factor != 0 && 1177 ub_random_max(worker->env.rnd, 1178 worker->env.cfg->ip_ratelimit_factor) == 1) { 1179 1180 char addrbuf[128]; 1181 addr_to_str(&repinfo->addr, repinfo->addrlen, 1182 addrbuf, sizeof(addrbuf)); 1183 verbose(VERB_QUERY, "ip_ratelimit allowed through for ip address %s because of slip in ip_ratelimit_factor", 1184 addrbuf); 1185 } else { 1186 worker->stats.num_queries_ip_ratelimited++; 1187 comm_point_drop_reply(repinfo); 1188 return 0; 1189 } 1190 } 1191 1192 /* see if query is in the cache */ 1193 if(!query_info_parse(&qinfo, c->buffer)) { 1194 verbose(VERB_ALGO, "worker parse request: formerror."); 1195 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen); 1196 memset(&qinfo, 0, sizeof(qinfo)); /* zero qinfo.qname */ 1197 if(worker_err_ratelimit(worker, LDNS_RCODE_FORMERR) == -1) { 1198 comm_point_drop_reply(repinfo); 1199 return 0; 1200 } 1201 sldns_buffer_rewind(c->buffer); 1202 LDNS_QR_SET(sldns_buffer_begin(c->buffer)); 1203 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), 1204 LDNS_RCODE_FORMERR); 1205 server_stats_insrcode(&worker->stats, c->buffer); 1206 goto send_reply; 1207 } 1208 if(worker->env.cfg->log_queries) { 1209 char ip[128]; 1210 addr_to_str(&repinfo->addr, repinfo->addrlen, ip, sizeof(ip)); 1211 log_nametypeclass(0, ip, qinfo.qname, qinfo.qtype, qinfo.qclass); 1212 } 1213 if(qinfo.qtype == LDNS_RR_TYPE_AXFR || 1214 qinfo.qtype == LDNS_RR_TYPE_IXFR) { 1215 verbose(VERB_ALGO, "worker request: refused zone transfer."); 1216 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen); 1217 sldns_buffer_rewind(c->buffer); 1218 LDNS_QR_SET(sldns_buffer_begin(c->buffer)); 1219 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), 1220 LDNS_RCODE_REFUSED); 1221 if(worker->stats.extended) { 1222 worker->stats.qtype[qinfo.qtype]++; 1223 server_stats_insrcode(&worker->stats, c->buffer); 1224 } 1225 goto send_reply; 1226 } 1227 if(qinfo.qtype == LDNS_RR_TYPE_OPT || 1228 qinfo.qtype == LDNS_RR_TYPE_TSIG || 1229 qinfo.qtype == LDNS_RR_TYPE_TKEY || 1230 qinfo.qtype == LDNS_RR_TYPE_MAILA || 1231 qinfo.qtype == LDNS_RR_TYPE_MAILB || 1232 (qinfo.qtype >= 128 && qinfo.qtype <= 248)) { 1233 verbose(VERB_ALGO, "worker request: formerror for meta-type."); 1234 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen); 1235 if(worker_err_ratelimit(worker, LDNS_RCODE_FORMERR) == -1) { 1236 comm_point_drop_reply(repinfo); 1237 return 0; 1238 } 1239 sldns_buffer_rewind(c->buffer); 1240 LDNS_QR_SET(sldns_buffer_begin(c->buffer)); 1241 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), 1242 LDNS_RCODE_FORMERR); 1243 if(worker->stats.extended) { 1244 worker->stats.qtype[qinfo.qtype]++; 1245 server_stats_insrcode(&worker->stats, c->buffer); 1246 } 1247 goto send_reply; 1248 } 1249 if((ret=parse_edns_from_pkt(c->buffer, &edns, worker->scratchpad)) != 0) { 1250 struct edns_data reply_edns; 1251 verbose(VERB_ALGO, "worker parse edns: formerror."); 1252 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen); 1253 memset(&reply_edns, 0, sizeof(reply_edns)); 1254 reply_edns.edns_present = 1; 1255 reply_edns.udp_size = EDNS_ADVERTISED_SIZE; 1256 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), ret); 1257 error_encode(c->buffer, ret, &qinfo, 1258 *(uint16_t*)(void *)sldns_buffer_begin(c->buffer), 1259 sldns_buffer_read_u16_at(c->buffer, 2), &reply_edns); 1260 regional_free_all(worker->scratchpad); 1261 server_stats_insrcode(&worker->stats, c->buffer); 1262 goto send_reply; 1263 } 1264 if(edns.edns_present) { 1265 struct edns_option* edns_opt; 1266 if(edns.edns_version != 0) { 1267 edns.ext_rcode = (uint8_t)(EDNS_RCODE_BADVERS>>4); 1268 edns.edns_version = EDNS_ADVERTISED_VERSION; 1269 edns.udp_size = EDNS_ADVERTISED_SIZE; 1270 edns.bits &= EDNS_DO; 1271 edns.opt_list = NULL; 1272 verbose(VERB_ALGO, "query with bad edns version."); 1273 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen); 1274 error_encode(c->buffer, EDNS_RCODE_BADVERS&0xf, &qinfo, 1275 *(uint16_t*)(void *)sldns_buffer_begin(c->buffer), 1276 sldns_buffer_read_u16_at(c->buffer, 2), NULL); 1277 if(sldns_buffer_capacity(c->buffer) >= 1278 sldns_buffer_limit(c->buffer)+calc_edns_field_size(&edns)) 1279 attach_edns_record(c->buffer, &edns); 1280 regional_free_all(worker->scratchpad); 1281 goto send_reply; 1282 } 1283 if(edns.udp_size < NORMAL_UDP_SIZE && 1284 worker->daemon->cfg->harden_short_bufsize) { 1285 verbose(VERB_QUERY, "worker request: EDNS bufsize %d ignored", 1286 (int)edns.udp_size); 1287 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen); 1288 edns.udp_size = NORMAL_UDP_SIZE; 1289 } 1290 if(c->type != comm_udp) { 1291 edns_opt = edns_opt_list_find(edns.opt_list, LDNS_EDNS_KEEPALIVE); 1292 if(edns_opt && edns_opt->opt_len > 0) { 1293 edns.ext_rcode = 0; 1294 edns.edns_version = EDNS_ADVERTISED_VERSION; 1295 edns.udp_size = EDNS_ADVERTISED_SIZE; 1296 edns.bits &= EDNS_DO; 1297 edns.opt_list = NULL; 1298 verbose(VERB_ALGO, "query with bad edns keepalive."); 1299 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen); 1300 error_encode(c->buffer, LDNS_RCODE_FORMERR, &qinfo, 1301 *(uint16_t*)(void *)sldns_buffer_begin(c->buffer), 1302 sldns_buffer_read_u16_at(c->buffer, 2), NULL); 1303 if(sldns_buffer_capacity(c->buffer) >= 1304 sldns_buffer_limit(c->buffer)+calc_edns_field_size(&edns)) 1305 attach_edns_record(c->buffer, &edns); 1306 regional_free_all(worker->scratchpad); 1307 goto send_reply; 1308 } 1309 } 1310 } 1311 if(edns.udp_size > worker->daemon->cfg->max_udp_size && 1312 c->type == comm_udp) { 1313 verbose(VERB_QUERY, 1314 "worker request: max UDP reply size modified" 1315 " (%d to max-udp-size)", (int)edns.udp_size); 1316 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen); 1317 edns.udp_size = worker->daemon->cfg->max_udp_size; 1318 } 1319 if(edns.udp_size < LDNS_HEADER_SIZE) { 1320 verbose(VERB_ALGO, "worker request: edns is too small."); 1321 log_addr(VERB_CLIENT, "from", &repinfo->addr, repinfo->addrlen); 1322 LDNS_QR_SET(sldns_buffer_begin(c->buffer)); 1323 LDNS_TC_SET(sldns_buffer_begin(c->buffer)); 1324 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), 1325 LDNS_RCODE_SERVFAIL); 1326 sldns_buffer_set_position(c->buffer, LDNS_HEADER_SIZE); 1327 sldns_buffer_write_at(c->buffer, 4, 1328 (uint8_t*)"\0\0\0\0\0\0\0\0", 8); 1329 sldns_buffer_flip(c->buffer); 1330 regional_free_all(worker->scratchpad); 1331 goto send_reply; 1332 } 1333 if(worker->stats.extended) 1334 server_stats_insquery(&worker->stats, c, qinfo.qtype, 1335 qinfo.qclass, &edns, repinfo); 1336 if(c->type != comm_udp) 1337 edns.udp_size = 65535; /* max size for TCP replies */ 1338 if(qinfo.qclass == LDNS_RR_CLASS_CH && answer_chaos(worker, &qinfo, 1339 &edns, repinfo, c->buffer)) { 1340 server_stats_insrcode(&worker->stats, c->buffer); 1341 regional_free_all(worker->scratchpad); 1342 goto send_reply; 1343 } 1344 if(LDNS_OPCODE_WIRE(sldns_buffer_begin(c->buffer)) == 1345 LDNS_PACKET_NOTIFY) { 1346 answer_notify(worker, &qinfo, &edns, c->buffer, repinfo); 1347 regional_free_all(worker->scratchpad); 1348 goto send_reply; 1349 } 1350 if(local_zones_answer(worker->daemon->local_zones, &worker->env, &qinfo, 1351 &edns, c->buffer, worker->scratchpad, repinfo, acladdr->taglist, 1352 acladdr->taglen, acladdr->tag_actions, 1353 acladdr->tag_actions_size, acladdr->tag_datas, 1354 acladdr->tag_datas_size, worker->daemon->cfg->tagname, 1355 worker->daemon->cfg->num_tags, acladdr->view)) { 1356 regional_free_all(worker->scratchpad); 1357 if(sldns_buffer_limit(c->buffer) == 0) { 1358 comm_point_drop_reply(repinfo); 1359 return 0; 1360 } 1361 server_stats_insrcode(&worker->stats, c->buffer); 1362 goto send_reply; 1363 } 1364 if(worker->env.auth_zones && 1365 auth_zones_answer(worker->env.auth_zones, &worker->env, 1366 &qinfo, &edns, repinfo, c->buffer, worker->scratchpad)) { 1367 regional_free_all(worker->scratchpad); 1368 if(sldns_buffer_limit(c->buffer) == 0) { 1369 comm_point_drop_reply(repinfo); 1370 return 0; 1371 } 1372 /* set RA for everyone that can have recursion (based on 1373 * access control list) */ 1374 if(LDNS_RD_WIRE(sldns_buffer_begin(c->buffer)) && 1375 acl != acl_deny_non_local && acl != acl_refuse_non_local) 1376 LDNS_RA_SET(sldns_buffer_begin(c->buffer)); 1377 server_stats_insrcode(&worker->stats, c->buffer); 1378 goto send_reply; 1379 } 1380 1381 /* We've looked in our local zones. If the answer isn't there, we 1382 * might need to bail out based on ACLs now. */ 1383 if((ret=deny_refuse_non_local(c, acl, worker, repinfo)) != -1) 1384 { 1385 regional_free_all(worker->scratchpad); 1386 if(ret == 1) 1387 goto send_reply; 1388 return ret; 1389 } 1390 1391 /* If this request does not have the recursion bit set, verify 1392 * ACLs allow the recursion bit to be treated as set. */ 1393 if(!(LDNS_RD_WIRE(sldns_buffer_begin(c->buffer))) && 1394 acl == acl_allow_setrd ) { 1395 LDNS_RD_SET(sldns_buffer_begin(c->buffer)); 1396 } 1397 1398 /* If this request does not have the recursion bit set, verify 1399 * ACLs allow the snooping. */ 1400 if(!(LDNS_RD_WIRE(sldns_buffer_begin(c->buffer))) && 1401 acl != acl_allow_snoop ) { 1402 error_encode(c->buffer, LDNS_RCODE_REFUSED, &qinfo, 1403 *(uint16_t*)(void *)sldns_buffer_begin(c->buffer), 1404 sldns_buffer_read_u16_at(c->buffer, 2), NULL); 1405 regional_free_all(worker->scratchpad); 1406 server_stats_insrcode(&worker->stats, c->buffer); 1407 log_addr(VERB_ALGO, "refused nonrec (cache snoop) query from", 1408 &repinfo->addr, repinfo->addrlen); 1409 goto send_reply; 1410 } 1411 1412 /* If we've found a local alias, replace the qname with the alias 1413 * target before resolving it. */ 1414 if(qinfo.local_alias) { 1415 struct ub_packed_rrset_key* rrset = qinfo.local_alias->rrset; 1416 struct packed_rrset_data* d = rrset->entry.data; 1417 1418 /* Sanity check: our current implementation only supports 1419 * a single CNAME RRset as a local alias. */ 1420 if(qinfo.local_alias->next || 1421 rrset->rk.type != htons(LDNS_RR_TYPE_CNAME) || 1422 d->count != 1) { 1423 log_err("assumption failure: unexpected local alias"); 1424 regional_free_all(worker->scratchpad); 1425 return 0; /* drop it */ 1426 } 1427 qinfo.qname = d->rr_data[0] + 2; 1428 qinfo.qname_len = d->rr_len[0] - 2; 1429 } 1430 1431 /* If we may apply IP-based actions to the answer, build the client 1432 * information. As this can be expensive, skip it if there is 1433 * absolutely no possibility of it. */ 1434 if(worker->daemon->use_response_ip && 1435 (qinfo.qtype == LDNS_RR_TYPE_A || 1436 qinfo.qtype == LDNS_RR_TYPE_AAAA || 1437 qinfo.qtype == LDNS_RR_TYPE_ANY)) { 1438 cinfo_tmp.taglist = acladdr->taglist; 1439 cinfo_tmp.taglen = acladdr->taglen; 1440 cinfo_tmp.tag_actions = acladdr->tag_actions; 1441 cinfo_tmp.tag_actions_size = acladdr->tag_actions_size; 1442 cinfo_tmp.tag_datas = acladdr->tag_datas; 1443 cinfo_tmp.tag_datas_size = acladdr->tag_datas_size; 1444 cinfo_tmp.view = acladdr->view; 1445 cinfo_tmp.respip_set = worker->daemon->respip_set; 1446 cinfo = &cinfo_tmp; 1447 } 1448 1449 lookup_cache: 1450 /* Lookup the cache. In case we chase an intermediate CNAME chain 1451 * this is a two-pass operation, and lookup_qinfo is different for 1452 * each pass. We should still pass the original qinfo to 1453 * answer_from_cache(), however, since it's used to build the reply. */ 1454 if(!edns_bypass_cache_stage(edns.opt_list, &worker->env)) { 1455 h = query_info_hash(lookup_qinfo, sldns_buffer_read_u16_at(c->buffer, 2)); 1456 if((e=slabhash_lookup(worker->env.msg_cache, h, lookup_qinfo, 0))) { 1457 /* answer from cache - we have acquired a readlock on it */ 1458 if(answer_from_cache(worker, &qinfo, 1459 cinfo, &need_drop, &alias_rrset, &partial_rep, 1460 (struct reply_info*)e->data, 1461 *(uint16_t*)(void *)sldns_buffer_begin(c->buffer), 1462 sldns_buffer_read_u16_at(c->buffer, 2), repinfo, 1463 &edns)) { 1464 /* prefetch it if the prefetch TTL expired. 1465 * Note that if there is more than one pass 1466 * its qname must be that used for cache 1467 * lookup. */ 1468 if((worker->env.cfg->prefetch || worker->env.cfg->serve_expired) 1469 && *worker->env.now >= 1470 ((struct reply_info*)e->data)->prefetch_ttl) { 1471 time_t leeway = ((struct reply_info*)e-> 1472 data)->ttl - *worker->env.now; 1473 if(((struct reply_info*)e->data)->ttl 1474 < *worker->env.now) 1475 leeway = 0; 1476 lock_rw_unlock(&e->lock); 1477 reply_and_prefetch(worker, lookup_qinfo, 1478 sldns_buffer_read_u16_at(c->buffer, 2), 1479 repinfo, leeway); 1480 if(!partial_rep) { 1481 rc = 0; 1482 regional_free_all(worker->scratchpad); 1483 goto send_reply_rc; 1484 } 1485 } else if(!partial_rep) { 1486 lock_rw_unlock(&e->lock); 1487 regional_free_all(worker->scratchpad); 1488 goto send_reply; 1489 } else { 1490 /* Note that we've already released the 1491 * lock if we're here after prefetch. */ 1492 lock_rw_unlock(&e->lock); 1493 } 1494 /* We've found a partial reply ending with an 1495 * alias. Replace the lookup qinfo for the 1496 * alias target and lookup the cache again to 1497 * (possibly) complete the reply. As we're 1498 * passing the "base" reply, there will be no 1499 * more alias chasing. */ 1500 memset(&qinfo_tmp, 0, sizeof(qinfo_tmp)); 1501 get_cname_target(alias_rrset, &qinfo_tmp.qname, 1502 &qinfo_tmp.qname_len); 1503 if(!qinfo_tmp.qname) { 1504 log_err("unexpected: invalid answer alias"); 1505 regional_free_all(worker->scratchpad); 1506 return 0; /* drop query */ 1507 } 1508 qinfo_tmp.qtype = qinfo.qtype; 1509 qinfo_tmp.qclass = qinfo.qclass; 1510 lookup_qinfo = &qinfo_tmp; 1511 goto lookup_cache; 1512 } 1513 verbose(VERB_ALGO, "answer from the cache failed"); 1514 lock_rw_unlock(&e->lock); 1515 } 1516 if(!LDNS_RD_WIRE(sldns_buffer_begin(c->buffer))) { 1517 if(answer_norec_from_cache(worker, &qinfo, 1518 *(uint16_t*)(void *)sldns_buffer_begin(c->buffer), 1519 sldns_buffer_read_u16_at(c->buffer, 2), repinfo, 1520 &edns)) { 1521 regional_free_all(worker->scratchpad); 1522 goto send_reply; 1523 } 1524 verbose(VERB_ALGO, "answer norec from cache -- " 1525 "need to validate or not primed"); 1526 } 1527 } 1528 sldns_buffer_rewind(c->buffer); 1529 server_stats_querymiss(&worker->stats, worker); 1530 1531 if(verbosity >= VERB_CLIENT) { 1532 if(c->type == comm_udp) 1533 log_addr(VERB_CLIENT, "udp request from", 1534 &repinfo->addr, repinfo->addrlen); 1535 else log_addr(VERB_CLIENT, "tcp request from", 1536 &repinfo->addr, repinfo->addrlen); 1537 } 1538 1539 /* grab a work request structure for this new request */ 1540 mesh_new_client(worker->env.mesh, &qinfo, cinfo, 1541 sldns_buffer_read_u16_at(c->buffer, 2), 1542 &edns, repinfo, *(uint16_t*)(void *)sldns_buffer_begin(c->buffer)); 1543 regional_free_all(worker->scratchpad); 1544 worker_mem_report(worker, NULL); 1545 return 0; 1546 1547 send_reply: 1548 rc = 1; 1549 send_reply_rc: 1550 if(need_drop) { 1551 comm_point_drop_reply(repinfo); 1552 return 0; 1553 } 1554 #ifdef USE_DNSTAP 1555 if(worker->dtenv.log_client_response_messages) 1556 dt_msg_send_client_response(&worker->dtenv, &repinfo->addr, 1557 c->type, c->buffer); 1558 #endif 1559 if(worker->env.cfg->log_replies) 1560 { 1561 struct timeval tv = {0, 0}; 1562 log_reply_info(0, &qinfo, &repinfo->addr, repinfo->addrlen, 1563 tv, 1, c->buffer); 1564 } 1565 #ifdef USE_DNSCRYPT 1566 if(!dnsc_handle_uncurved_request(repinfo)) { 1567 return 0; 1568 } 1569 #endif 1570 return rc; 1571 } 1572 1573 void 1574 worker_sighandler(int sig, void* arg) 1575 { 1576 /* note that log, print, syscalls here give race conditions. 1577 * And cause hangups if the log-lock is held by the application. */ 1578 struct worker* worker = (struct worker*)arg; 1579 switch(sig) { 1580 #ifdef SIGHUP 1581 case SIGHUP: 1582 comm_base_exit(worker->base); 1583 break; 1584 #endif 1585 case SIGINT: 1586 worker->need_to_exit = 1; 1587 comm_base_exit(worker->base); 1588 break; 1589 #ifdef SIGQUIT 1590 case SIGQUIT: 1591 worker->need_to_exit = 1; 1592 comm_base_exit(worker->base); 1593 break; 1594 #endif 1595 case SIGTERM: 1596 worker->need_to_exit = 1; 1597 comm_base_exit(worker->base); 1598 break; 1599 default: 1600 /* unknown signal, ignored */ 1601 break; 1602 } 1603 } 1604 1605 /** restart statistics timer for worker, if enabled */ 1606 static void 1607 worker_restart_timer(struct worker* worker) 1608 { 1609 if(worker->env.cfg->stat_interval > 0) { 1610 struct timeval tv; 1611 #ifndef S_SPLINT_S 1612 tv.tv_sec = worker->env.cfg->stat_interval; 1613 tv.tv_usec = 0; 1614 #endif 1615 comm_timer_set(worker->stat_timer, &tv); 1616 } 1617 } 1618 1619 void worker_stat_timer_cb(void* arg) 1620 { 1621 struct worker* worker = (struct worker*)arg; 1622 server_stats_log(&worker->stats, worker, worker->thread_num); 1623 mesh_stats(worker->env.mesh, "mesh has"); 1624 worker_mem_report(worker, NULL); 1625 /* SHM is enabled, process data to SHM */ 1626 if (worker->daemon->cfg->shm_enable) { 1627 shm_main_run(worker); 1628 } 1629 if(!worker->daemon->cfg->stat_cumulative) { 1630 worker_stats_clear(worker); 1631 } 1632 /* start next timer */ 1633 worker_restart_timer(worker); 1634 } 1635 1636 void worker_probe_timer_cb(void* arg) 1637 { 1638 struct worker* worker = (struct worker*)arg; 1639 struct timeval tv; 1640 #ifndef S_SPLINT_S 1641 tv.tv_sec = (time_t)autr_probe_timer(&worker->env); 1642 tv.tv_usec = 0; 1643 #endif 1644 if(tv.tv_sec != 0) 1645 comm_timer_set(worker->env.probe_timer, &tv); 1646 } 1647 1648 struct worker* 1649 worker_create(struct daemon* daemon, int id, int* ports, int n) 1650 { 1651 unsigned int seed; 1652 struct worker* worker = (struct worker*)calloc(1, 1653 sizeof(struct worker)); 1654 if(!worker) 1655 return NULL; 1656 worker->numports = n; 1657 worker->ports = (int*)memdup(ports, sizeof(int)*n); 1658 if(!worker->ports) { 1659 free(worker); 1660 return NULL; 1661 } 1662 worker->daemon = daemon; 1663 worker->thread_num = id; 1664 if(!(worker->cmd = tube_create())) { 1665 free(worker->ports); 1666 free(worker); 1667 return NULL; 1668 } 1669 /* create random state here to avoid locking trouble in RAND_bytes */ 1670 seed = (unsigned int)time(NULL) ^ (unsigned int)getpid() ^ 1671 (((unsigned int)worker->thread_num)<<17); 1672 /* shift thread_num so it does not match out pid bits */ 1673 if(!(worker->rndstate = ub_initstate(seed, daemon->rand))) { 1674 explicit_bzero(&seed, sizeof(seed)); 1675 log_err("could not init random numbers."); 1676 tube_delete(worker->cmd); 1677 free(worker->ports); 1678 free(worker); 1679 return NULL; 1680 } 1681 explicit_bzero(&seed, sizeof(seed)); 1682 #ifdef USE_DNSTAP 1683 if(daemon->cfg->dnstap) { 1684 log_assert(daemon->dtenv != NULL); 1685 memcpy(&worker->dtenv, daemon->dtenv, sizeof(struct dt_env)); 1686 if(!dt_init(&worker->dtenv)) 1687 fatal_exit("dt_init failed"); 1688 } 1689 #endif 1690 return worker; 1691 } 1692 1693 int 1694 worker_init(struct worker* worker, struct config_file *cfg, 1695 struct listen_port* ports, int do_sigs) 1696 { 1697 #ifdef USE_DNSTAP 1698 struct dt_env* dtenv = &worker->dtenv; 1699 #else 1700 void* dtenv = NULL; 1701 #endif 1702 worker->need_to_exit = 0; 1703 worker->base = comm_base_create(do_sigs); 1704 if(!worker->base) { 1705 log_err("could not create event handling base"); 1706 worker_delete(worker); 1707 return 0; 1708 } 1709 comm_base_set_slow_accept_handlers(worker->base, &worker_stop_accept, 1710 &worker_start_accept, worker); 1711 if(do_sigs) { 1712 #ifdef SIGHUP 1713 ub_thread_sig_unblock(SIGHUP); 1714 #endif 1715 ub_thread_sig_unblock(SIGINT); 1716 #ifdef SIGQUIT 1717 ub_thread_sig_unblock(SIGQUIT); 1718 #endif 1719 ub_thread_sig_unblock(SIGTERM); 1720 #ifndef LIBEVENT_SIGNAL_PROBLEM 1721 worker->comsig = comm_signal_create(worker->base, 1722 worker_sighandler, worker); 1723 if(!worker->comsig 1724 #ifdef SIGHUP 1725 || !comm_signal_bind(worker->comsig, SIGHUP) 1726 #endif 1727 #ifdef SIGQUIT 1728 || !comm_signal_bind(worker->comsig, SIGQUIT) 1729 #endif 1730 || !comm_signal_bind(worker->comsig, SIGTERM) 1731 || !comm_signal_bind(worker->comsig, SIGINT)) { 1732 log_err("could not create signal handlers"); 1733 worker_delete(worker); 1734 return 0; 1735 } 1736 #endif /* LIBEVENT_SIGNAL_PROBLEM */ 1737 if(!daemon_remote_open_accept(worker->daemon->rc, 1738 worker->daemon->rc_ports, worker)) { 1739 worker_delete(worker); 1740 return 0; 1741 } 1742 #ifdef UB_ON_WINDOWS 1743 wsvc_setup_worker(worker); 1744 #endif /* UB_ON_WINDOWS */ 1745 } else { /* !do_sigs */ 1746 worker->comsig = NULL; 1747 } 1748 worker->front = listen_create(worker->base, ports, 1749 cfg->msg_buffer_size, (int)cfg->incoming_num_tcp, 1750 cfg->do_tcp_keepalive 1751 ? cfg->tcp_keepalive_timeout 1752 : cfg->tcp_idle_timeout, 1753 worker->daemon->tcl, 1754 worker->daemon->listen_sslctx, 1755 dtenv, worker_handle_request, worker); 1756 if(!worker->front) { 1757 log_err("could not create listening sockets"); 1758 worker_delete(worker); 1759 return 0; 1760 } 1761 worker->back = outside_network_create(worker->base, 1762 cfg->msg_buffer_size, (size_t)cfg->outgoing_num_ports, 1763 cfg->out_ifs, cfg->num_out_ifs, cfg->do_ip4, cfg->do_ip6, 1764 cfg->do_tcp?cfg->outgoing_num_tcp:0, 1765 worker->daemon->env->infra_cache, worker->rndstate, 1766 cfg->use_caps_bits_for_id, worker->ports, worker->numports, 1767 cfg->unwanted_threshold, cfg->outgoing_tcp_mss, 1768 &worker_alloc_cleanup, worker, 1769 cfg->do_udp || cfg->udp_upstream_without_downstream, 1770 worker->daemon->connect_sslctx, cfg->delay_close, 1771 dtenv); 1772 if(!worker->back) { 1773 log_err("could not create outgoing sockets"); 1774 worker_delete(worker); 1775 return 0; 1776 } 1777 /* start listening to commands */ 1778 if(!tube_setup_bg_listen(worker->cmd, worker->base, 1779 &worker_handle_control_cmd, worker)) { 1780 log_err("could not create control compt."); 1781 worker_delete(worker); 1782 return 0; 1783 } 1784 worker->stat_timer = comm_timer_create(worker->base, 1785 worker_stat_timer_cb, worker); 1786 if(!worker->stat_timer) { 1787 log_err("could not create statistics timer"); 1788 } 1789 1790 /* we use the msg_buffer_size as a good estimate for what the 1791 * user wants for memory usage sizes */ 1792 worker->scratchpad = regional_create_custom(cfg->msg_buffer_size); 1793 if(!worker->scratchpad) { 1794 log_err("malloc failure"); 1795 worker_delete(worker); 1796 return 0; 1797 } 1798 1799 server_stats_init(&worker->stats, cfg); 1800 alloc_init(&worker->alloc, &worker->daemon->superalloc, 1801 worker->thread_num); 1802 alloc_set_id_cleanup(&worker->alloc, &worker_alloc_cleanup, worker); 1803 worker->env = *worker->daemon->env; 1804 comm_base_timept(worker->base, &worker->env.now, &worker->env.now_tv); 1805 if(worker->thread_num == 0) 1806 log_set_time(worker->env.now); 1807 worker->env.worker = worker; 1808 worker->env.worker_base = worker->base; 1809 worker->env.send_query = &worker_send_query; 1810 worker->env.alloc = &worker->alloc; 1811 worker->env.outnet = worker->back; 1812 worker->env.rnd = worker->rndstate; 1813 /* If case prefetch is triggered, the corresponding mesh will clear 1814 * the scratchpad for the module env in the middle of request handling. 1815 * It would be prone to a use-after-free kind of bug, so we avoid 1816 * sharing it with worker's own scratchpad at the cost of having 1817 * one more pad per worker. */ 1818 worker->env.scratch = regional_create_custom(cfg->msg_buffer_size); 1819 if(!worker->env.scratch) { 1820 log_err("malloc failure"); 1821 worker_delete(worker); 1822 return 0; 1823 } 1824 worker->env.mesh = mesh_create(&worker->daemon->mods, &worker->env); 1825 worker->env.detach_subs = &mesh_detach_subs; 1826 worker->env.attach_sub = &mesh_attach_sub; 1827 worker->env.add_sub = &mesh_add_sub; 1828 worker->env.kill_sub = &mesh_state_delete; 1829 worker->env.detect_cycle = &mesh_detect_cycle; 1830 worker->env.scratch_buffer = sldns_buffer_new(cfg->msg_buffer_size); 1831 if(!(worker->env.fwds = forwards_create()) || 1832 !forwards_apply_cfg(worker->env.fwds, cfg)) { 1833 log_err("Could not set forward zones"); 1834 worker_delete(worker); 1835 return 0; 1836 } 1837 if(!(worker->env.hints = hints_create()) || 1838 !hints_apply_cfg(worker->env.hints, cfg)) { 1839 log_err("Could not set root or stub hints"); 1840 worker_delete(worker); 1841 return 0; 1842 } 1843 /* one probe timer per process -- if we have 5011 anchors */ 1844 if(autr_get_num_anchors(worker->env.anchors) > 0 1845 #ifndef THREADS_DISABLED 1846 && worker->thread_num == 0 1847 #endif 1848 ) { 1849 struct timeval tv; 1850 tv.tv_sec = 0; 1851 tv.tv_usec = 0; 1852 worker->env.probe_timer = comm_timer_create(worker->base, 1853 worker_probe_timer_cb, worker); 1854 if(!worker->env.probe_timer) { 1855 log_err("could not create 5011-probe timer"); 1856 } else { 1857 /* let timer fire, then it can reset itself */ 1858 comm_timer_set(worker->env.probe_timer, &tv); 1859 } 1860 } 1861 /* zone transfer tasks, setup once per process, if any */ 1862 if(worker->env.auth_zones 1863 #ifndef THREADS_DISABLED 1864 && worker->thread_num == 0 1865 #endif 1866 ) { 1867 auth_xfer_pickup_initial(worker->env.auth_zones, &worker->env); 1868 } 1869 if(!worker->env.mesh || !worker->env.scratch_buffer) { 1870 worker_delete(worker); 1871 return 0; 1872 } 1873 worker_mem_report(worker, NULL); 1874 /* if statistics enabled start timer */ 1875 if(worker->env.cfg->stat_interval > 0) { 1876 verbose(VERB_ALGO, "set statistics interval %d secs", 1877 worker->env.cfg->stat_interval); 1878 worker_restart_timer(worker); 1879 } 1880 return 1; 1881 } 1882 1883 void 1884 worker_work(struct worker* worker) 1885 { 1886 comm_base_dispatch(worker->base); 1887 } 1888 1889 void 1890 worker_delete(struct worker* worker) 1891 { 1892 if(!worker) 1893 return; 1894 if(worker->env.mesh && verbosity >= VERB_OPS) { 1895 server_stats_log(&worker->stats, worker, worker->thread_num); 1896 mesh_stats(worker->env.mesh, "mesh has"); 1897 worker_mem_report(worker, NULL); 1898 } 1899 outside_network_quit_prepare(worker->back); 1900 mesh_delete(worker->env.mesh); 1901 sldns_buffer_free(worker->env.scratch_buffer); 1902 forwards_delete(worker->env.fwds); 1903 hints_delete(worker->env.hints); 1904 listen_delete(worker->front); 1905 outside_network_delete(worker->back); 1906 comm_signal_delete(worker->comsig); 1907 tube_delete(worker->cmd); 1908 comm_timer_delete(worker->stat_timer); 1909 comm_timer_delete(worker->env.probe_timer); 1910 free(worker->ports); 1911 if(worker->thread_num == 0) { 1912 log_set_time(NULL); 1913 #ifdef UB_ON_WINDOWS 1914 wsvc_desetup_worker(worker); 1915 #endif /* UB_ON_WINDOWS */ 1916 } 1917 comm_base_delete(worker->base); 1918 ub_randfree(worker->rndstate); 1919 alloc_clear(&worker->alloc); 1920 regional_destroy(worker->env.scratch); 1921 regional_destroy(worker->scratchpad); 1922 free(worker); 1923 } 1924 1925 struct outbound_entry* 1926 worker_send_query(struct query_info* qinfo, uint16_t flags, int dnssec, 1927 int want_dnssec, int nocaps, struct sockaddr_storage* addr, 1928 socklen_t addrlen, uint8_t* zone, size_t zonelen, int ssl_upstream, 1929 char* tls_auth_name, struct module_qstate* q) 1930 { 1931 struct worker* worker = q->env->worker; 1932 struct outbound_entry* e = (struct outbound_entry*)regional_alloc( 1933 q->region, sizeof(*e)); 1934 if(!e) 1935 return NULL; 1936 e->qstate = q; 1937 e->qsent = outnet_serviced_query(worker->back, qinfo, flags, dnssec, 1938 want_dnssec, nocaps, q->env->cfg->tcp_upstream, 1939 ssl_upstream, tls_auth_name, addr, addrlen, zone, zonelen, q, 1940 worker_handle_service_reply, e, worker->back->udp_buff, q->env); 1941 if(!e->qsent) { 1942 return NULL; 1943 } 1944 return e; 1945 } 1946 1947 void 1948 worker_alloc_cleanup(void* arg) 1949 { 1950 struct worker* worker = (struct worker*)arg; 1951 slabhash_clear(&worker->env.rrset_cache->table); 1952 slabhash_clear(worker->env.msg_cache); 1953 } 1954 1955 void worker_stats_clear(struct worker* worker) 1956 { 1957 server_stats_init(&worker->stats, worker->env.cfg); 1958 mesh_stats_clear(worker->env.mesh); 1959 worker->back->unwanted_replies = 0; 1960 worker->back->num_tcp_outgoing = 0; 1961 } 1962 1963 void worker_start_accept(void* arg) 1964 { 1965 struct worker* worker = (struct worker*)arg; 1966 listen_start_accept(worker->front); 1967 if(worker->thread_num == 0) 1968 daemon_remote_start_accept(worker->daemon->rc); 1969 } 1970 1971 void worker_stop_accept(void* arg) 1972 { 1973 struct worker* worker = (struct worker*)arg; 1974 listen_stop_accept(worker->front); 1975 if(worker->thread_num == 0) 1976 daemon_remote_stop_accept(worker->daemon->rc); 1977 } 1978 1979 /* --- fake callbacks for fptr_wlist to work --- */ 1980 struct outbound_entry* libworker_send_query( 1981 struct query_info* ATTR_UNUSED(qinfo), 1982 uint16_t ATTR_UNUSED(flags), int ATTR_UNUSED(dnssec), 1983 int ATTR_UNUSED(want_dnssec), int ATTR_UNUSED(nocaps), 1984 struct sockaddr_storage* ATTR_UNUSED(addr), socklen_t ATTR_UNUSED(addrlen), 1985 uint8_t* ATTR_UNUSED(zone), size_t ATTR_UNUSED(zonelen), 1986 int ATTR_UNUSED(ssl_upstream), char* ATTR_UNUSED(tls_auth_name), 1987 struct module_qstate* ATTR_UNUSED(q)) 1988 { 1989 log_assert(0); 1990 return 0; 1991 } 1992 1993 int libworker_handle_reply(struct comm_point* ATTR_UNUSED(c), 1994 void* ATTR_UNUSED(arg), int ATTR_UNUSED(error), 1995 struct comm_reply* ATTR_UNUSED(reply_info)) 1996 { 1997 log_assert(0); 1998 return 0; 1999 } 2000 2001 int libworker_handle_service_reply(struct comm_point* ATTR_UNUSED(c), 2002 void* ATTR_UNUSED(arg), int ATTR_UNUSED(error), 2003 struct comm_reply* ATTR_UNUSED(reply_info)) 2004 { 2005 log_assert(0); 2006 return 0; 2007 } 2008 2009 void libworker_handle_control_cmd(struct tube* ATTR_UNUSED(tube), 2010 uint8_t* ATTR_UNUSED(buffer), size_t ATTR_UNUSED(len), 2011 int ATTR_UNUSED(error), void* ATTR_UNUSED(arg)) 2012 { 2013 log_assert(0); 2014 } 2015 2016 void libworker_fg_done_cb(void* ATTR_UNUSED(arg), int ATTR_UNUSED(rcode), 2017 sldns_buffer* ATTR_UNUSED(buf), enum sec_status ATTR_UNUSED(s), 2018 char* ATTR_UNUSED(why_bogus), int ATTR_UNUSED(was_ratelimited)) 2019 { 2020 log_assert(0); 2021 } 2022 2023 void libworker_bg_done_cb(void* ATTR_UNUSED(arg), int ATTR_UNUSED(rcode), 2024 sldns_buffer* ATTR_UNUSED(buf), enum sec_status ATTR_UNUSED(s), 2025 char* ATTR_UNUSED(why_bogus), int ATTR_UNUSED(was_ratelimited)) 2026 { 2027 log_assert(0); 2028 } 2029 2030 void libworker_event_done_cb(void* ATTR_UNUSED(arg), int ATTR_UNUSED(rcode), 2031 sldns_buffer* ATTR_UNUSED(buf), enum sec_status ATTR_UNUSED(s), 2032 char* ATTR_UNUSED(why_bogus), int ATTR_UNUSED(was_ratelimited)) 2033 { 2034 log_assert(0); 2035 } 2036 2037 int context_query_cmp(const void* ATTR_UNUSED(a), const void* ATTR_UNUSED(b)) 2038 { 2039 log_assert(0); 2040 return 0; 2041 } 2042 2043 int order_lock_cmp(const void* ATTR_UNUSED(e1), const void* ATTR_UNUSED(e2)) 2044 { 2045 log_assert(0); 2046 return 0; 2047 } 2048 2049 int codeline_cmp(const void* ATTR_UNUSED(a), const void* ATTR_UNUSED(b)) 2050 { 2051 log_assert(0); 2052 return 0; 2053 } 2054 2055