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 "services/rpz.h" 65 #include "util/data/msgparse.h" 66 #include "util/data/msgencode.h" 67 #include "util/data/dname.h" 68 #include "util/fptr_wlist.h" 69 #include "util/proxy_protocol.h" 70 #include "util/tube.h" 71 #include "util/edns.h" 72 #include "util/timeval_func.h" 73 #include "iterator/iter_fwd.h" 74 #include "iterator/iter_hints.h" 75 #include "iterator/iter_utils.h" 76 #include "validator/autotrust.h" 77 #include "validator/val_anchor.h" 78 #include "respip/respip.h" 79 #include "libunbound/context.h" 80 #include "libunbound/libworker.h" 81 #include "sldns/sbuffer.h" 82 #include "sldns/wire2str.h" 83 #include "util/shm_side/shm_main.h" 84 #include "dnscrypt/dnscrypt.h" 85 #include "dnstap/dtstream.h" 86 87 #ifdef HAVE_SYS_TYPES_H 88 # include <sys/types.h> 89 #endif 90 #ifdef HAVE_NETDB_H 91 #include <netdb.h> 92 #endif 93 #include <signal.h> 94 #ifdef UB_ON_WINDOWS 95 #include "winrc/win_svc.h" 96 #endif 97 98 /** Size of an UDP datagram */ 99 #define NORMAL_UDP_SIZE 512 /* bytes */ 100 /** ratelimit for error responses */ 101 #define ERROR_RATELIMIT 100 /* qps */ 102 103 /** 104 * seconds to add to prefetch leeway. This is a TTL that expires old rrsets 105 * earlier than they should in order to put the new update into the cache. 106 * This additional value is to make sure that if not all TTLs are equal in 107 * the message to be updated(and replaced), that rrsets with up to this much 108 * extra TTL are also replaced. This means that the resulting new message 109 * will have (most likely) this TTL at least, avoiding very small 'split 110 * second' TTLs due to operators choosing relative primes for TTLs (or so). 111 * Also has to be at least one to break ties (and overwrite cached entry). 112 */ 113 #define PREFETCH_EXPIRY_ADD 60 114 115 /** Report on memory usage by this thread and global */ 116 static void 117 worker_mem_report(struct worker* ATTR_UNUSED(worker), 118 struct serviced_query* ATTR_UNUSED(cur_serv)) 119 { 120 #ifdef UNBOUND_ALLOC_STATS 121 /* measure memory leakage */ 122 extern size_t unbound_mem_alloc, unbound_mem_freed; 123 /* debug func in validator module */ 124 size_t total, front, back, mesh, msg, rrset, infra, ac, superac; 125 size_t me, iter, val, anch; 126 int i; 127 #ifdef CLIENT_SUBNET 128 size_t subnet = 0; 129 #endif /* CLIENT_SUBNET */ 130 if(verbosity < VERB_ALGO) 131 return; 132 front = listen_get_mem(worker->front); 133 back = outnet_get_mem(worker->back); 134 msg = slabhash_get_mem(worker->env.msg_cache); 135 rrset = slabhash_get_mem(&worker->env.rrset_cache->table); 136 infra = infra_get_mem(worker->env.infra_cache); 137 mesh = mesh_get_mem(worker->env.mesh); 138 ac = alloc_get_mem(worker->alloc); 139 superac = alloc_get_mem(&worker->daemon->superalloc); 140 anch = anchors_get_mem(worker->env.anchors); 141 iter = 0; 142 val = 0; 143 for(i=0; i<worker->env.mesh->mods.num; i++) { 144 fptr_ok(fptr_whitelist_mod_get_mem(worker->env.mesh-> 145 mods.mod[i]->get_mem)); 146 if(strcmp(worker->env.mesh->mods.mod[i]->name, "validator")==0) 147 val += (*worker->env.mesh->mods.mod[i]->get_mem) 148 (&worker->env, i); 149 #ifdef CLIENT_SUBNET 150 else if(strcmp(worker->env.mesh->mods.mod[i]->name, 151 "subnetcache")==0) 152 subnet += (*worker->env.mesh->mods.mod[i]->get_mem) 153 (&worker->env, i); 154 #endif /* CLIENT_SUBNET */ 155 else iter += (*worker->env.mesh->mods.mod[i]->get_mem) 156 (&worker->env, i); 157 } 158 me = sizeof(*worker) + sizeof(*worker->base) + sizeof(*worker->comsig) 159 + comm_point_get_mem(worker->cmd_com) 160 + sizeof(worker->rndstate) 161 + regional_get_mem(worker->scratchpad) 162 + sizeof(*worker->env.scratch_buffer) 163 + sldns_buffer_capacity(worker->env.scratch_buffer); 164 if(worker->daemon->env->fwds) 165 log_info("forwards=%u", (unsigned)forwards_get_mem(worker->env.fwds)); 166 if(worker->daemon->env->hints) 167 log_info("hints=%u", (unsigned)hints_get_mem(worker->env.hints)); 168 if(worker->thread_num == 0) 169 me += acl_list_get_mem(worker->daemon->acl); 170 if(cur_serv) { 171 me += serviced_get_mem(cur_serv); 172 } 173 total = front+back+mesh+msg+rrset+infra+iter+val+ac+superac+me; 174 #ifdef CLIENT_SUBNET 175 total += subnet; 176 log_info("Memory conditions: %u front=%u back=%u mesh=%u msg=%u " 177 "rrset=%u infra=%u iter=%u val=%u subnet=%u anchors=%u " 178 "alloccache=%u globalalloccache=%u me=%u", 179 (unsigned)total, (unsigned)front, (unsigned)back, 180 (unsigned)mesh, (unsigned)msg, (unsigned)rrset, (unsigned)infra, 181 (unsigned)iter, (unsigned)val, 182 (unsigned)subnet, (unsigned)anch, (unsigned)ac, 183 (unsigned)superac, (unsigned)me); 184 #else /* no CLIENT_SUBNET */ 185 log_info("Memory conditions: %u front=%u back=%u mesh=%u msg=%u " 186 "rrset=%u infra=%u iter=%u val=%u anchors=%u " 187 "alloccache=%u globalalloccache=%u me=%u", 188 (unsigned)total, (unsigned)front, (unsigned)back, 189 (unsigned)mesh, (unsigned)msg, (unsigned)rrset, 190 (unsigned)infra, (unsigned)iter, (unsigned)val, (unsigned)anch, 191 (unsigned)ac, (unsigned)superac, (unsigned)me); 192 #endif /* CLIENT_SUBNET */ 193 log_info("Total heap memory estimate: %u total-alloc: %u " 194 "total-free: %u", (unsigned)total, 195 (unsigned)unbound_mem_alloc, (unsigned)unbound_mem_freed); 196 #else /* no UNBOUND_ALLOC_STATS */ 197 size_t val = 0; 198 #ifdef CLIENT_SUBNET 199 size_t subnet = 0; 200 #endif /* CLIENT_SUBNET */ 201 int i; 202 if(verbosity < VERB_QUERY) 203 return; 204 for(i=0; i<worker->env.mesh->mods.num; i++) { 205 fptr_ok(fptr_whitelist_mod_get_mem(worker->env.mesh-> 206 mods.mod[i]->get_mem)); 207 if(strcmp(worker->env.mesh->mods.mod[i]->name, "validator")==0) 208 val += (*worker->env.mesh->mods.mod[i]->get_mem) 209 (&worker->env, i); 210 #ifdef CLIENT_SUBNET 211 else if(strcmp(worker->env.mesh->mods.mod[i]->name, 212 "subnetcache")==0) 213 subnet += (*worker->env.mesh->mods.mod[i]->get_mem) 214 (&worker->env, i); 215 #endif /* CLIENT_SUBNET */ 216 } 217 #ifdef CLIENT_SUBNET 218 verbose(VERB_QUERY, "cache memory msg=%u rrset=%u infra=%u val=%u " 219 "subnet=%u", 220 (unsigned)slabhash_get_mem(worker->env.msg_cache), 221 (unsigned)slabhash_get_mem(&worker->env.rrset_cache->table), 222 (unsigned)infra_get_mem(worker->env.infra_cache), 223 (unsigned)val, (unsigned)subnet); 224 #else /* no CLIENT_SUBNET */ 225 verbose(VERB_QUERY, "cache memory msg=%u rrset=%u infra=%u val=%u", 226 (unsigned)slabhash_get_mem(worker->env.msg_cache), 227 (unsigned)slabhash_get_mem(&worker->env.rrset_cache->table), 228 (unsigned)infra_get_mem(worker->env.infra_cache), 229 (unsigned)val); 230 #endif /* CLIENT_SUBNET */ 231 #endif /* UNBOUND_ALLOC_STATS */ 232 } 233 234 void 235 worker_send_cmd(struct worker* worker, enum worker_commands cmd) 236 { 237 uint32_t c = (uint32_t)htonl(cmd); 238 if(!tube_write_msg(worker->cmd, (uint8_t*)&c, sizeof(c), 0)) { 239 log_err("worker send cmd %d failed", (int)cmd); 240 } 241 } 242 243 int 244 worker_handle_service_reply(struct comm_point* c, void* arg, int error, 245 struct comm_reply* reply_info) 246 { 247 struct outbound_entry* e = (struct outbound_entry*)arg; 248 struct worker* worker = e->qstate->env->worker; 249 struct serviced_query *sq = e->qsent; 250 251 verbose(VERB_ALGO, "worker svcd callback for qstate %p", e->qstate); 252 if(error != 0) { 253 mesh_report_reply(worker->env.mesh, e, reply_info, error); 254 worker_mem_report(worker, sq); 255 return 0; 256 } 257 /* sanity check. */ 258 if(!LDNS_QR_WIRE(sldns_buffer_begin(c->buffer)) 259 || LDNS_OPCODE_WIRE(sldns_buffer_begin(c->buffer)) != 260 LDNS_PACKET_QUERY 261 || LDNS_QDCOUNT(sldns_buffer_begin(c->buffer)) > 1) { 262 /* error becomes timeout for the module as if this reply 263 * never arrived. */ 264 verbose(VERB_ALGO, "worker: bad reply handled as timeout"); 265 mesh_report_reply(worker->env.mesh, e, reply_info, 266 NETEVENT_TIMEOUT); 267 worker_mem_report(worker, sq); 268 return 0; 269 } 270 mesh_report_reply(worker->env.mesh, e, reply_info, NETEVENT_NOERROR); 271 worker_mem_report(worker, sq); 272 return 0; 273 } 274 275 /** ratelimit error replies 276 * @param worker: the worker struct with ratelimit counter 277 * @param err: error code that would be wanted. 278 * @return value of err if okay, or -1 if it should be discarded instead. 279 */ 280 static int 281 worker_err_ratelimit(struct worker* worker, int err) 282 { 283 if(worker->err_limit_time == *worker->env.now) { 284 /* see if limit is exceeded for this second */ 285 if(worker->err_limit_count++ > ERROR_RATELIMIT) 286 return -1; 287 } else { 288 /* new second, new limits */ 289 worker->err_limit_time = *worker->env.now; 290 worker->err_limit_count = 1; 291 } 292 return err; 293 } 294 295 /** 296 * Structure holding the result of the worker_check_request function. 297 * Based on configuration it could be called up to four times; ideally should 298 * be called once. 299 */ 300 struct check_request_result { 301 int checked; 302 int value; 303 }; 304 /** check request sanity. 305 * @param pkt: the wire packet to examine for sanity. 306 * @param worker: parameters for checking. 307 * @param out: struct to update with the result. 308 */ 309 static void 310 worker_check_request(sldns_buffer* pkt, struct worker* worker, 311 struct check_request_result* out) 312 { 313 if(out->checked) return; 314 out->checked = 1; 315 if(sldns_buffer_limit(pkt) < LDNS_HEADER_SIZE) { 316 verbose(VERB_QUERY, "request too short, discarded"); 317 out->value = -1; 318 return; 319 } 320 if(sldns_buffer_limit(pkt) > NORMAL_UDP_SIZE && 321 worker->daemon->cfg->harden_large_queries) { 322 verbose(VERB_QUERY, "request too large, discarded"); 323 out->value = -1; 324 return; 325 } 326 if(LDNS_QR_WIRE(sldns_buffer_begin(pkt))) { 327 verbose(VERB_QUERY, "request has QR bit on, discarded"); 328 out->value = -1; 329 return; 330 } 331 if(LDNS_TC_WIRE(sldns_buffer_begin(pkt))) { 332 LDNS_TC_CLR(sldns_buffer_begin(pkt)); 333 verbose(VERB_QUERY, "request bad, has TC bit on"); 334 out->value = worker_err_ratelimit(worker, LDNS_RCODE_FORMERR); 335 return; 336 } 337 if(LDNS_OPCODE_WIRE(sldns_buffer_begin(pkt)) != LDNS_PACKET_QUERY && 338 LDNS_OPCODE_WIRE(sldns_buffer_begin(pkt)) != LDNS_PACKET_NOTIFY) { 339 verbose(VERB_QUERY, "request unknown opcode %d", 340 LDNS_OPCODE_WIRE(sldns_buffer_begin(pkt))); 341 out->value = worker_err_ratelimit(worker, LDNS_RCODE_NOTIMPL); 342 return; 343 } 344 if(LDNS_QDCOUNT(sldns_buffer_begin(pkt)) != 1) { 345 verbose(VERB_QUERY, "request wrong nr qd=%d", 346 LDNS_QDCOUNT(sldns_buffer_begin(pkt))); 347 out->value = worker_err_ratelimit(worker, LDNS_RCODE_FORMERR); 348 return; 349 } 350 if(LDNS_ANCOUNT(sldns_buffer_begin(pkt)) != 0 && 351 (LDNS_ANCOUNT(sldns_buffer_begin(pkt)) != 1 || 352 LDNS_OPCODE_WIRE(sldns_buffer_begin(pkt)) != LDNS_PACKET_NOTIFY)) { 353 verbose(VERB_QUERY, "request wrong nr an=%d", 354 LDNS_ANCOUNT(sldns_buffer_begin(pkt))); 355 out->value = worker_err_ratelimit(worker, LDNS_RCODE_FORMERR); 356 return; 357 } 358 if(LDNS_NSCOUNT(sldns_buffer_begin(pkt)) != 0) { 359 verbose(VERB_QUERY, "request wrong nr ns=%d", 360 LDNS_NSCOUNT(sldns_buffer_begin(pkt))); 361 out->value = worker_err_ratelimit(worker, LDNS_RCODE_FORMERR); 362 return; 363 } 364 if(LDNS_ARCOUNT(sldns_buffer_begin(pkt)) > 1) { 365 verbose(VERB_QUERY, "request wrong nr ar=%d", 366 LDNS_ARCOUNT(sldns_buffer_begin(pkt))); 367 out->value = worker_err_ratelimit(worker, LDNS_RCODE_FORMERR); 368 return; 369 } 370 out->value = 0; 371 return; 372 } 373 374 void 375 worker_handle_control_cmd(struct tube* ATTR_UNUSED(tube), uint8_t* msg, 376 size_t len, int error, void* arg) 377 { 378 struct worker* worker = (struct worker*)arg; 379 enum worker_commands cmd; 380 if(error != NETEVENT_NOERROR) { 381 free(msg); 382 if(error == NETEVENT_CLOSED) 383 comm_base_exit(worker->base); 384 else log_info("control event: %d", error); 385 return; 386 } 387 if(len != sizeof(uint32_t)) { 388 fatal_exit("bad control msg length %d", (int)len); 389 } 390 cmd = sldns_read_uint32(msg); 391 free(msg); 392 switch(cmd) { 393 case worker_cmd_quit: 394 verbose(VERB_ALGO, "got control cmd quit"); 395 comm_base_exit(worker->base); 396 break; 397 case worker_cmd_stats: 398 verbose(VERB_ALGO, "got control cmd stats"); 399 server_stats_reply(worker, 1); 400 break; 401 case worker_cmd_stats_noreset: 402 verbose(VERB_ALGO, "got control cmd stats_noreset"); 403 server_stats_reply(worker, 0); 404 break; 405 case worker_cmd_remote: 406 verbose(VERB_ALGO, "got control cmd remote"); 407 daemon_remote_exec(worker); 408 break; 409 default: 410 log_err("bad command %d", (int)cmd); 411 break; 412 } 413 } 414 415 /** check if a delegation is secure */ 416 static enum sec_status 417 check_delegation_secure(struct reply_info *rep) 418 { 419 /* return smallest security status */ 420 size_t i; 421 enum sec_status sec = sec_status_secure; 422 enum sec_status s; 423 size_t num = rep->an_numrrsets + rep->ns_numrrsets; 424 /* check if answer and authority are OK */ 425 for(i=0; i<num; i++) { 426 s = ((struct packed_rrset_data*)rep->rrsets[i]->entry.data) 427 ->security; 428 if(s < sec) 429 sec = s; 430 } 431 /* in additional, only unchecked triggers revalidation */ 432 for(i=num; i<rep->rrset_count; i++) { 433 s = ((struct packed_rrset_data*)rep->rrsets[i]->entry.data) 434 ->security; 435 if(s == sec_status_unchecked) 436 return s; 437 } 438 return sec; 439 } 440 441 /** remove nonsecure from a delegation referral additional section */ 442 static void 443 deleg_remove_nonsecure_additional(struct reply_info* rep) 444 { 445 /* we can simply edit it, since we are working in the scratch region */ 446 size_t i; 447 enum sec_status s; 448 449 for(i = rep->an_numrrsets+rep->ns_numrrsets; i<rep->rrset_count; i++) { 450 s = ((struct packed_rrset_data*)rep->rrsets[i]->entry.data) 451 ->security; 452 if(s != sec_status_secure) { 453 memmove(rep->rrsets+i, rep->rrsets+i+1, 454 sizeof(struct ub_packed_rrset_key*)* 455 (rep->rrset_count - i - 1)); 456 rep->ar_numrrsets--; 457 rep->rrset_count--; 458 i--; 459 } 460 } 461 } 462 463 /** answer nonrecursive query from the cache */ 464 static int 465 answer_norec_from_cache(struct worker* worker, struct query_info* qinfo, 466 uint16_t id, uint16_t flags, struct comm_reply* repinfo, 467 struct edns_data* edns) 468 { 469 /* for a nonrecursive query return either: 470 * o an error (servfail; we try to avoid this) 471 * o a delegation (closest we have; this routine tries that) 472 * o the answer (checked by answer_from_cache) 473 * 474 * So, grab a delegation from the rrset cache. 475 * Then check if it needs validation, if so, this routine fails, 476 * so that iterator can prime and validator can verify rrsets. 477 */ 478 uint16_t udpsize = edns->udp_size; 479 int secure = 0; 480 time_t timenow = *worker->env.now; 481 int has_cd_bit = (flags&BIT_CD); 482 int must_validate = (!has_cd_bit || worker->env.cfg->ignore_cd) 483 && worker->env.need_to_validate; 484 struct dns_msg *msg = NULL; 485 struct delegpt *dp; 486 487 dp = dns_cache_find_delegation(&worker->env, qinfo->qname, 488 qinfo->qname_len, qinfo->qtype, qinfo->qclass, 489 worker->scratchpad, &msg, timenow, 0, NULL, 0); 490 if(!dp) { /* no delegation, need to reprime */ 491 return 0; 492 } 493 /* In case we have a local alias, copy it into the delegation message. 494 * Shallow copy should be fine, as we'll be done with msg in this 495 * function. */ 496 msg->qinfo.local_alias = qinfo->local_alias; 497 if(must_validate) { 498 switch(check_delegation_secure(msg->rep)) { 499 case sec_status_unchecked: 500 /* some rrsets have not been verified yet, go and 501 * let validator do that */ 502 return 0; 503 case sec_status_bogus: 504 case sec_status_secure_sentinel_fail: 505 /* some rrsets are bogus, reply servfail */ 506 edns->edns_version = EDNS_ADVERTISED_VERSION; 507 edns->udp_size = EDNS_ADVERTISED_SIZE; 508 edns->ext_rcode = 0; 509 edns->bits &= EDNS_DO; 510 if(!inplace_cb_reply_servfail_call(&worker->env, qinfo, NULL, 511 msg->rep, LDNS_RCODE_SERVFAIL, edns, repinfo, worker->scratchpad, 512 worker->env.now_tv)) 513 return 0; 514 /* Attach the cached EDE (RFC8914) */ 515 if(worker->env.cfg->ede && 516 msg->rep->reason_bogus != LDNS_EDE_NONE) { 517 edns_opt_list_append_ede(&edns->opt_list_out, 518 worker->scratchpad, msg->rep->reason_bogus, 519 msg->rep->reason_bogus_str); 520 } 521 error_encode(repinfo->c->buffer, LDNS_RCODE_SERVFAIL, 522 &msg->qinfo, id, flags, edns); 523 if(worker->stats.extended) { 524 worker->stats.ans_bogus++; 525 worker->stats.ans_rcode[LDNS_RCODE_SERVFAIL]++; 526 } 527 return 1; 528 case sec_status_secure: 529 /* all rrsets are secure */ 530 /* remove non-secure rrsets from the add. section*/ 531 if(worker->env.cfg->val_clean_additional) 532 deleg_remove_nonsecure_additional(msg->rep); 533 secure = 1; 534 break; 535 case sec_status_indeterminate: 536 case sec_status_insecure: 537 default: 538 /* not secure */ 539 secure = 0; 540 break; 541 } 542 } 543 /* return this delegation from the cache */ 544 edns->edns_version = EDNS_ADVERTISED_VERSION; 545 edns->udp_size = EDNS_ADVERTISED_SIZE; 546 edns->ext_rcode = 0; 547 edns->bits &= EDNS_DO; 548 if(worker->env.cfg->disable_edns_do && (edns->bits & EDNS_DO)) 549 edns->edns_present = 0; 550 if(!inplace_cb_reply_cache_call(&worker->env, qinfo, NULL, msg->rep, 551 (int)(flags&LDNS_RCODE_MASK), edns, repinfo, worker->scratchpad, 552 worker->env.now_tv)) 553 return 0; 554 msg->rep->flags |= BIT_QR|BIT_RA; 555 /* Attach the cached EDE (RFC8914) if CD bit is set and the answer is 556 * bogus. */ 557 if(worker->env.cfg->ede && has_cd_bit && 558 (check_delegation_secure(msg->rep) == sec_status_bogus || 559 check_delegation_secure(msg->rep) == sec_status_secure_sentinel_fail) && 560 msg->rep->reason_bogus != LDNS_EDE_NONE) { 561 edns_opt_list_append_ede(&edns->opt_list_out, 562 worker->scratchpad, msg->rep->reason_bogus, 563 msg->rep->reason_bogus_str); 564 } 565 if(!reply_info_answer_encode(&msg->qinfo, msg->rep, id, flags, 566 repinfo->c->buffer, 0, 1, worker->scratchpad, 567 udpsize, edns, (int)(edns->bits & EDNS_DO), secure)) { 568 if(!inplace_cb_reply_servfail_call(&worker->env, qinfo, NULL, NULL, 569 LDNS_RCODE_SERVFAIL, edns, repinfo, worker->scratchpad, 570 worker->env.now_tv)) 571 edns->opt_list_inplace_cb_out = NULL; 572 error_encode(repinfo->c->buffer, LDNS_RCODE_SERVFAIL, 573 &msg->qinfo, id, flags, edns); 574 } 575 if(worker->stats.extended) { 576 if(secure) worker->stats.ans_secure++; 577 server_stats_insrcode(&worker->stats, repinfo->c->buffer); 578 } 579 return 1; 580 } 581 582 /** Apply, if applicable, a response IP action to a cached answer. 583 * If the answer is rewritten as a result of an action, '*encode_repp' will 584 * point to the reply info containing the modified answer. '*encode_repp' will 585 * be intact otherwise. 586 * It returns 1 on success, 0 otherwise. */ 587 static int 588 apply_respip_action(struct worker* worker, const struct query_info* qinfo, 589 struct respip_client_info* cinfo, struct reply_info* rep, 590 struct sockaddr_storage* addr, socklen_t addrlen, 591 struct ub_packed_rrset_key** alias_rrset, 592 struct reply_info** encode_repp, struct auth_zones* az) 593 { 594 struct respip_action_info actinfo = {0, 0, 0, 0, NULL, 0, NULL}; 595 actinfo.action = respip_none; 596 597 if(qinfo->qtype != LDNS_RR_TYPE_A && 598 qinfo->qtype != LDNS_RR_TYPE_AAAA && 599 qinfo->qtype != LDNS_RR_TYPE_ANY) 600 return 1; 601 602 if(!respip_rewrite_reply(qinfo, cinfo, rep, encode_repp, &actinfo, 603 alias_rrset, 0, worker->scratchpad, az, NULL)) 604 return 0; 605 606 /* xxx_deny actions mean dropping the reply, unless the original reply 607 * was redirected to response-ip data. */ 608 if(actinfo.action == respip_always_deny || 609 ((actinfo.action == respip_deny || 610 actinfo.action == respip_inform_deny) && 611 *encode_repp == rep)) 612 *encode_repp = NULL; 613 614 /* If address info is returned, it means the action should be an 615 * 'inform' variant and the information should be logged. */ 616 if(actinfo.addrinfo) { 617 respip_inform_print(&actinfo, qinfo->qname, 618 qinfo->qtype, qinfo->qclass, qinfo->local_alias, 619 addr, addrlen); 620 621 if(worker->stats.extended && actinfo.rpz_used) { 622 if(actinfo.rpz_disabled) 623 worker->stats.rpz_action[RPZ_DISABLED_ACTION]++; 624 if(actinfo.rpz_cname_override) 625 worker->stats.rpz_action[RPZ_CNAME_OVERRIDE_ACTION]++; 626 else 627 worker->stats.rpz_action[ 628 respip_action_to_rpz_action(actinfo.action)]++; 629 } 630 } 631 632 return 1; 633 } 634 635 /** answer query from the cache. 636 * Normally, the answer message will be built in repinfo->c->buffer; if the 637 * answer is supposed to be suppressed or the answer is supposed to be an 638 * incomplete CNAME chain, the buffer is explicitly cleared to signal the 639 * caller as such. In the latter case *partial_rep will point to the incomplete 640 * reply, and this function is (possibly) supposed to be called again with that 641 * *partial_rep value to complete the chain. In addition, if the query should 642 * be completely dropped, '*need_drop' will be set to 1. */ 643 static int 644 answer_from_cache(struct worker* worker, struct query_info* qinfo, 645 struct respip_client_info* cinfo, int* need_drop, int* is_expired_answer, 646 int* is_secure_answer, struct ub_packed_rrset_key** alias_rrset, 647 struct reply_info** partial_repp, 648 struct reply_info* rep, uint16_t id, uint16_t flags, 649 struct comm_reply* repinfo, struct edns_data* edns) 650 { 651 time_t timenow = *worker->env.now; 652 uint16_t udpsize = edns->udp_size; 653 struct reply_info* encode_rep = rep; 654 struct reply_info* partial_rep = *partial_repp; 655 int has_cd_bit = (flags&BIT_CD); 656 int must_validate = (!has_cd_bit || worker->env.cfg->ignore_cd) 657 && worker->env.need_to_validate; 658 *partial_repp = NULL; /* avoid accidental further pass */ 659 660 /* Check TTL */ 661 if(rep->ttl < timenow) { 662 /* Check if we need to serve expired now */ 663 if(worker->env.cfg->serve_expired && 664 /* if serve-expired-client-timeout is set, serve 665 * an expired record without attempting recursion 666 * if the serve_expired_norec_ttl is set for the record 667 * as we know that recursion is currently failing. */ 668 (!worker->env.cfg->serve_expired_client_timeout || 669 timenow < rep->serve_expired_norec_ttl) 670 #ifdef USE_CACHEDB 671 && !(worker->env.cachedb_enabled && 672 worker->env.cfg->cachedb_check_when_serve_expired) 673 #endif 674 ) { 675 if(!reply_info_can_answer_expired(rep, timenow)) 676 return 0; 677 if(!rrset_array_lock(rep->ref, rep->rrset_count, 0)) 678 return 0; 679 *is_expired_answer = 1; 680 } else { 681 /* the rrsets may have been updated in the meantime. 682 * we will refetch the message format from the 683 * authoritative server 684 */ 685 return 0; 686 } 687 } else { 688 if(!rrset_array_lock(rep->ref, rep->rrset_count, timenow)) 689 return 0; 690 } 691 /* locked and ids and ttls are OK. */ 692 693 /* check CNAME chain (if any) */ 694 if(rep->an_numrrsets > 0 && (rep->rrsets[0]->rk.type == 695 htons(LDNS_RR_TYPE_CNAME) || rep->rrsets[0]->rk.type == 696 htons(LDNS_RR_TYPE_DNAME))) { 697 if(!reply_check_cname_chain(qinfo, rep)) { 698 /* cname chain invalid, redo iterator steps */ 699 verbose(VERB_ALGO, "Cache reply: cname chain broken"); 700 goto bail_out; 701 } 702 } 703 /* check security status of the cached answer */ 704 if(must_validate && (rep->security == sec_status_bogus || 705 rep->security == sec_status_secure_sentinel_fail)) { 706 /* BAD cached */ 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(worker->env.cfg->disable_edns_do && (edns->bits & EDNS_DO)) 712 edns->edns_present = 0; 713 if(!inplace_cb_reply_servfail_call(&worker->env, qinfo, NULL, rep, 714 LDNS_RCODE_SERVFAIL, edns, repinfo, worker->scratchpad, 715 worker->env.now_tv)) 716 goto bail_out; 717 /* Attach the cached EDE (RFC8914) */ 718 if(worker->env.cfg->ede && rep->reason_bogus != LDNS_EDE_NONE) { 719 edns_opt_list_append_ede(&edns->opt_list_out, 720 worker->scratchpad, rep->reason_bogus, 721 rep->reason_bogus_str); 722 } 723 error_encode(repinfo->c->buffer, LDNS_RCODE_SERVFAIL, 724 qinfo, id, flags, edns); 725 rrset_array_unlock_touch(worker->env.rrset_cache, 726 worker->scratchpad, rep->ref, rep->rrset_count); 727 if(worker->stats.extended) { 728 worker->stats.ans_bogus ++; 729 worker->stats.ans_rcode[LDNS_RCODE_SERVFAIL] ++; 730 } 731 return 1; 732 } else if(rep->security == sec_status_unchecked && must_validate) { 733 verbose(VERB_ALGO, "Cache reply: unchecked entry needs " 734 "validation"); 735 goto bail_out; /* need to validate cache entry first */ 736 } else if(rep->security == sec_status_secure) { 737 if(reply_all_rrsets_secure(rep)) { 738 *is_secure_answer = 1; 739 } else { 740 if(must_validate) { 741 verbose(VERB_ALGO, "Cache reply: secure entry" 742 " changed status"); 743 goto bail_out; /* rrset changed, re-verify */ 744 } 745 *is_secure_answer = 0; 746 } 747 } else *is_secure_answer = 0; 748 749 edns->edns_version = EDNS_ADVERTISED_VERSION; 750 edns->udp_size = EDNS_ADVERTISED_SIZE; 751 edns->ext_rcode = 0; 752 edns->bits &= EDNS_DO; 753 if(worker->env.cfg->disable_edns_do && (edns->bits & EDNS_DO)) 754 edns->edns_present = 0; 755 *alias_rrset = NULL; /* avoid confusion if caller set it to non-NULL */ 756 if((worker->daemon->use_response_ip || worker->daemon->use_rpz) && 757 !partial_rep && !apply_respip_action(worker, qinfo, cinfo, rep, 758 &repinfo->client_addr, repinfo->client_addrlen, alias_rrset, 759 &encode_rep, worker->env.auth_zones)) { 760 goto bail_out; 761 } else if(partial_rep && 762 !respip_merge_cname(partial_rep, qinfo, rep, cinfo, 763 must_validate, &encode_rep, worker->scratchpad, 764 worker->env.auth_zones)) { 765 goto bail_out; 766 } 767 if(encode_rep != rep) { 768 /* if rewritten, it can't be considered "secure" */ 769 *is_secure_answer = 0; 770 } 771 if(!encode_rep || *alias_rrset) { 772 if(!encode_rep) 773 *need_drop = 1; 774 else { 775 /* If a partial CNAME chain is found, we first need to 776 * make a copy of the reply in the scratchpad so we 777 * can release the locks and lookup the cache again. */ 778 *partial_repp = reply_info_copy(encode_rep, NULL, 779 worker->scratchpad); 780 if(!*partial_repp) 781 goto bail_out; 782 } 783 } else { 784 if(*is_expired_answer == 1 && 785 worker->env.cfg->ede_serve_expired && worker->env.cfg->ede) { 786 EDNS_OPT_LIST_APPEND_EDE(&edns->opt_list_out, 787 worker->scratchpad, LDNS_EDE_STALE_ANSWER, ""); 788 } 789 /* Attach the cached EDE (RFC8914) if CD bit is set and the 790 * answer is bogus. */ 791 if(*is_secure_answer == 0 && 792 worker->env.cfg->ede && has_cd_bit && 793 encode_rep->reason_bogus != LDNS_EDE_NONE) { 794 edns_opt_list_append_ede(&edns->opt_list_out, 795 worker->scratchpad, encode_rep->reason_bogus, 796 encode_rep->reason_bogus_str); 797 } 798 if(!inplace_cb_reply_cache_call(&worker->env, qinfo, NULL, encode_rep, 799 (int)(flags&LDNS_RCODE_MASK), edns, repinfo, worker->scratchpad, 800 worker->env.now_tv)) 801 goto bail_out; 802 if(!reply_info_answer_encode(qinfo, encode_rep, id, flags, 803 repinfo->c->buffer, timenow, 1, worker->scratchpad, 804 udpsize, edns, (int)(edns->bits & EDNS_DO), 805 *is_secure_answer)) { 806 if(!inplace_cb_reply_servfail_call(&worker->env, qinfo, 807 NULL, NULL, LDNS_RCODE_SERVFAIL, edns, repinfo, 808 worker->scratchpad, worker->env.now_tv)) 809 edns->opt_list_inplace_cb_out = NULL; 810 error_encode(repinfo->c->buffer, LDNS_RCODE_SERVFAIL, 811 qinfo, id, flags, edns); 812 } 813 } 814 /* cannot send the reply right now, because blocking network syscall 815 * is bad while holding locks. */ 816 rrset_array_unlock_touch(worker->env.rrset_cache, worker->scratchpad, 817 rep->ref, rep->rrset_count); 818 /* go and return this buffer to the client */ 819 return 1; 820 821 bail_out: 822 rrset_array_unlock_touch(worker->env.rrset_cache, 823 worker->scratchpad, rep->ref, rep->rrset_count); 824 return 0; 825 } 826 827 /** Reply to client and perform prefetch to keep cache up to date. */ 828 static void 829 reply_and_prefetch(struct worker* worker, struct query_info* qinfo, 830 uint16_t flags, struct comm_reply* repinfo, time_t leeway, int noreply, 831 int rpz_passthru, struct edns_option* opt_list) 832 { 833 (void)opt_list; 834 /* first send answer to client to keep its latency 835 * as small as a cachereply */ 836 if(!noreply) { 837 if(repinfo->c->tcp_req_info) { 838 sldns_buffer_copy( 839 repinfo->c->tcp_req_info->spool_buffer, 840 repinfo->c->buffer); 841 } 842 comm_point_send_reply(repinfo); 843 } 844 server_stats_prefetch(&worker->stats, worker); 845 #ifdef CLIENT_SUBNET 846 /* Check if the subnet module is enabled. In that case pass over the 847 * comm_reply information for ECS generation later. The mesh states are 848 * unique when subnet is enabled. */ 849 if(modstack_find(&worker->env.mesh->mods, "subnetcache") != -1 850 && worker->env.unique_mesh) { 851 mesh_new_prefetch(worker->env.mesh, qinfo, flags, leeway + 852 PREFETCH_EXPIRY_ADD, rpz_passthru, 853 &repinfo->client_addr, opt_list); 854 return; 855 } 856 #endif 857 /* create the prefetch in the mesh as a normal lookup without 858 * client addrs waiting, which has the cache blacklisted (to bypass 859 * the cache and go to the network for the data). */ 860 /* this (potentially) runs the mesh for the new query */ 861 mesh_new_prefetch(worker->env.mesh, qinfo, flags, leeway + 862 PREFETCH_EXPIRY_ADD, rpz_passthru, NULL, NULL); 863 } 864 865 /** 866 * Fill CH class answer into buffer. Keeps query. 867 * @param pkt: buffer 868 * @param str: string to put into text record (<255). 869 * array of strings, every string becomes a text record. 870 * @param num: number of strings in array. 871 * @param edns: edns reply information. 872 * @param worker: worker with scratch region. 873 * @param repinfo: reply information for a communication point. 874 */ 875 static void 876 chaos_replystr(sldns_buffer* pkt, char** str, int num, struct edns_data* edns, 877 struct worker* worker, struct comm_reply* repinfo) 878 { 879 int i; 880 unsigned int rd = LDNS_RD_WIRE(sldns_buffer_begin(pkt)); 881 unsigned int cd = LDNS_CD_WIRE(sldns_buffer_begin(pkt)); 882 size_t udpsize = edns->udp_size; 883 edns->edns_version = EDNS_ADVERTISED_VERSION; 884 edns->udp_size = EDNS_ADVERTISED_SIZE; 885 edns->bits &= EDNS_DO; 886 if(!inplace_cb_reply_local_call(&worker->env, NULL, NULL, NULL, 887 LDNS_RCODE_NOERROR, edns, repinfo, worker->scratchpad, 888 worker->env.now_tv)) 889 edns->opt_list_inplace_cb_out = NULL; 890 sldns_buffer_clear(pkt); 891 sldns_buffer_skip(pkt, (ssize_t)sizeof(uint16_t)); /* skip id */ 892 sldns_buffer_write_u16(pkt, (uint16_t)(BIT_QR|BIT_RA)); 893 if(rd) LDNS_RD_SET(sldns_buffer_begin(pkt)); 894 if(cd) LDNS_CD_SET(sldns_buffer_begin(pkt)); 895 sldns_buffer_write_u16(pkt, 1); /* qdcount */ 896 sldns_buffer_write_u16(pkt, (uint16_t)num); /* ancount */ 897 sldns_buffer_write_u16(pkt, 0); /* nscount */ 898 sldns_buffer_write_u16(pkt, 0); /* arcount */ 899 (void)query_dname_len(pkt); /* skip qname */ 900 sldns_buffer_skip(pkt, (ssize_t)sizeof(uint16_t)); /* skip qtype */ 901 sldns_buffer_skip(pkt, (ssize_t)sizeof(uint16_t)); /* skip qclass */ 902 for(i=0; i<num; i++) { 903 size_t len = strlen(str[i]); 904 if(len>255) len=255; /* cap size of TXT record */ 905 if(sldns_buffer_position(pkt)+2+2+2+4+2+1+len+ 906 calc_edns_field_size(edns) > udpsize) { 907 sldns_buffer_write_u16_at(pkt, 6, i); /* ANCOUNT */ 908 LDNS_TC_SET(sldns_buffer_begin(pkt)); 909 break; 910 } 911 sldns_buffer_write_u16(pkt, 0xc00c); /* compr ptr to query */ 912 sldns_buffer_write_u16(pkt, LDNS_RR_TYPE_TXT); 913 sldns_buffer_write_u16(pkt, LDNS_RR_CLASS_CH); 914 sldns_buffer_write_u32(pkt, 0); /* TTL */ 915 sldns_buffer_write_u16(pkt, sizeof(uint8_t) + len); 916 sldns_buffer_write_u8(pkt, len); 917 sldns_buffer_write(pkt, str[i], len); 918 } 919 sldns_buffer_flip(pkt); 920 if(sldns_buffer_capacity(pkt) >= 921 sldns_buffer_limit(pkt)+calc_edns_field_size(edns)) 922 attach_edns_record(pkt, edns); 923 } 924 925 /** Reply with one string */ 926 static void 927 chaos_replyonestr(sldns_buffer* pkt, const char* str, struct edns_data* edns, 928 struct worker* worker, struct comm_reply* repinfo) 929 { 930 chaos_replystr(pkt, (char**)&str, 1, edns, worker, repinfo); 931 } 932 933 /** 934 * Create CH class trustanchor answer. 935 * @param pkt: buffer 936 * @param edns: edns reply information. 937 * @param w: worker with scratch region. 938 * @param repinfo: reply information for a communication point. 939 */ 940 static void 941 chaos_trustanchor(sldns_buffer* pkt, struct edns_data* edns, struct worker* w, 942 struct comm_reply* repinfo) 943 { 944 #define TA_RESPONSE_MAX_TXT 16 /* max number of TXT records */ 945 #define TA_RESPONSE_MAX_TAGS 32 /* max number of tags printed per zone */ 946 char* str_array[TA_RESPONSE_MAX_TXT]; 947 uint16_t tags[TA_RESPONSE_MAX_TAGS]; 948 int num = 0; 949 struct trust_anchor* ta; 950 951 if(!w->env.need_to_validate) { 952 /* no validator module, reply no trustanchors */ 953 chaos_replystr(pkt, NULL, 0, edns, w, repinfo); 954 return; 955 } 956 957 /* fill the string with contents */ 958 lock_basic_lock(&w->env.anchors->lock); 959 RBTREE_FOR(ta, struct trust_anchor*, w->env.anchors->tree) { 960 char* str; 961 size_t i, numtag, str_len = 255; 962 if(num == TA_RESPONSE_MAX_TXT) continue; 963 str = (char*)regional_alloc(w->scratchpad, str_len); 964 if(!str) continue; 965 lock_basic_lock(&ta->lock); 966 numtag = anchor_list_keytags(ta, tags, TA_RESPONSE_MAX_TAGS); 967 if(numtag == 0) { 968 /* empty, insecure point */ 969 lock_basic_unlock(&ta->lock); 970 continue; 971 } 972 str_array[num] = str; 973 num++; 974 975 /* spool name of anchor */ 976 (void)sldns_wire2str_dname_buf(ta->name, ta->namelen, str, str_len); 977 str_len -= strlen(str); str += strlen(str); 978 /* spool tags */ 979 for(i=0; i<numtag; i++) { 980 snprintf(str, str_len, " %u", (unsigned)tags[i]); 981 str_len -= strlen(str); str += strlen(str); 982 } 983 lock_basic_unlock(&ta->lock); 984 } 985 lock_basic_unlock(&w->env.anchors->lock); 986 987 chaos_replystr(pkt, str_array, num, edns, w, repinfo); 988 regional_free_all(w->scratchpad); 989 } 990 991 /** 992 * Answer CH class queries. 993 * @param w: worker 994 * @param qinfo: query info. Pointer into packet buffer. 995 * @param edns: edns info from query. 996 * @param repinfo: reply information for a communication point. 997 * @param pkt: packet buffer. 998 * @return: true if a reply is to be sent. 999 */ 1000 static int 1001 answer_chaos(struct worker* w, struct query_info* qinfo, 1002 struct edns_data* edns, struct comm_reply* repinfo, sldns_buffer* pkt) 1003 { 1004 struct config_file* cfg = w->env.cfg; 1005 if(qinfo->qtype != LDNS_RR_TYPE_ANY && qinfo->qtype != LDNS_RR_TYPE_TXT) 1006 return 0; 1007 if(query_dname_compare(qinfo->qname, 1008 (uint8_t*)"\002id\006server") == 0 || 1009 query_dname_compare(qinfo->qname, 1010 (uint8_t*)"\010hostname\004bind") == 0) 1011 { 1012 if(cfg->hide_identity) 1013 return 0; 1014 if(cfg->identity==NULL || cfg->identity[0]==0) { 1015 char buf[MAXHOSTNAMELEN+1]; 1016 if (gethostname(buf, MAXHOSTNAMELEN) == 0) { 1017 buf[MAXHOSTNAMELEN] = 0; 1018 chaos_replyonestr(pkt, buf, edns, w, repinfo); 1019 } else { 1020 log_err("gethostname: %s", strerror(errno)); 1021 chaos_replyonestr(pkt, "no hostname", edns, w, repinfo); 1022 } 1023 } 1024 else chaos_replyonestr(pkt, cfg->identity, edns, w, repinfo); 1025 return 1; 1026 } 1027 if(query_dname_compare(qinfo->qname, 1028 (uint8_t*)"\007version\006server") == 0 || 1029 query_dname_compare(qinfo->qname, 1030 (uint8_t*)"\007version\004bind") == 0) 1031 { 1032 if(cfg->hide_version) 1033 return 0; 1034 if(cfg->version==NULL || cfg->version[0]==0) 1035 chaos_replyonestr(pkt, PACKAGE_STRING, edns, w, repinfo); 1036 else chaos_replyonestr(pkt, cfg->version, edns, w, repinfo); 1037 return 1; 1038 } 1039 if(query_dname_compare(qinfo->qname, 1040 (uint8_t*)"\013trustanchor\007unbound") == 0) 1041 { 1042 if(cfg->hide_trustanchor) 1043 return 0; 1044 chaos_trustanchor(pkt, edns, w, repinfo); 1045 return 1; 1046 } 1047 1048 return 0; 1049 } 1050 1051 /** 1052 * Answer notify queries. These are notifies for authoritative zones, 1053 * the reply is an ack that the notify has been received. We need to check 1054 * access permission here. 1055 * @param w: worker 1056 * @param qinfo: query info. Pointer into packet buffer. 1057 * @param edns: edns info from query. 1058 * @param addr: client address. 1059 * @param addrlen: client address length. 1060 * @param pkt: packet buffer. 1061 */ 1062 static void 1063 answer_notify(struct worker* w, struct query_info* qinfo, 1064 struct edns_data* edns, sldns_buffer* pkt, 1065 struct sockaddr_storage* addr, socklen_t addrlen) 1066 { 1067 int refused = 0; 1068 int rcode = LDNS_RCODE_NOERROR; 1069 uint32_t serial = 0; 1070 int has_serial; 1071 if(!w->env.auth_zones) return; 1072 has_serial = auth_zone_parse_notify_serial(pkt, &serial); 1073 if(auth_zones_notify(w->env.auth_zones, &w->env, qinfo->qname, 1074 qinfo->qname_len, qinfo->qclass, addr, 1075 addrlen, has_serial, serial, &refused)) { 1076 rcode = LDNS_RCODE_NOERROR; 1077 } else { 1078 if(refused) 1079 rcode = LDNS_RCODE_REFUSED; 1080 else rcode = LDNS_RCODE_SERVFAIL; 1081 } 1082 1083 if(verbosity >= VERB_DETAIL) { 1084 char buf[380]; 1085 char zname[255+1]; 1086 char sr[25]; 1087 dname_str(qinfo->qname, zname); 1088 sr[0]=0; 1089 if(has_serial) 1090 snprintf(sr, sizeof(sr), "serial %u ", 1091 (unsigned)serial); 1092 if(rcode == LDNS_RCODE_REFUSED) 1093 snprintf(buf, sizeof(buf), 1094 "refused NOTIFY %sfor %s from", sr, zname); 1095 else if(rcode == LDNS_RCODE_SERVFAIL) 1096 snprintf(buf, sizeof(buf), 1097 "servfail for NOTIFY %sfor %s from", sr, zname); 1098 else snprintf(buf, sizeof(buf), 1099 "received NOTIFY %sfor %s from", sr, zname); 1100 log_addr(VERB_DETAIL, buf, addr, addrlen); 1101 } 1102 edns->edns_version = EDNS_ADVERTISED_VERSION; 1103 edns->udp_size = EDNS_ADVERTISED_SIZE; 1104 edns->ext_rcode = 0; 1105 edns->bits &= EDNS_DO; 1106 error_encode(pkt, rcode, qinfo, 1107 *(uint16_t*)(void *)sldns_buffer_begin(pkt), 1108 sldns_buffer_read_u16_at(pkt, 2), edns); 1109 LDNS_OPCODE_SET(sldns_buffer_begin(pkt), LDNS_PACKET_NOTIFY); 1110 } 1111 1112 static int 1113 deny_refuse(struct comm_point* c, enum acl_access acl, 1114 enum acl_access deny, enum acl_access refuse, 1115 struct worker* worker, struct comm_reply* repinfo, 1116 struct acl_addr* acladdr, int ede, 1117 struct check_request_result* check_result) 1118 { 1119 if(acl == deny) { 1120 if(verbosity >= VERB_ALGO) { 1121 log_acl_action("dropped", &repinfo->client_addr, 1122 repinfo->client_addrlen, acl, acladdr); 1123 log_buf(VERB_ALGO, "dropped", c->buffer); 1124 } 1125 comm_point_drop_reply(repinfo); 1126 if(worker->stats.extended) 1127 worker->stats.unwanted_queries++; 1128 return 0; 1129 } else if(acl == refuse) { 1130 size_t opt_rr_mark; 1131 1132 if(verbosity >= VERB_ALGO) { 1133 log_acl_action("refused", &repinfo->client_addr, 1134 repinfo->client_addrlen, acl, acladdr); 1135 log_buf(VERB_ALGO, "refuse", c->buffer); 1136 } 1137 1138 if(worker->stats.extended) 1139 worker->stats.unwanted_queries++; 1140 worker_check_request(c->buffer, worker, check_result); 1141 if(check_result->value != 0) { 1142 if(check_result->value != -1) { 1143 LDNS_QR_SET(sldns_buffer_begin(c->buffer)); 1144 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), 1145 check_result->value); 1146 return 1; 1147 } 1148 comm_point_drop_reply(repinfo); 1149 return 0; 1150 } 1151 /* worker_check_request() above guarantees that the buffer contains at 1152 * least a header and that qdcount == 1 1153 */ 1154 log_assert(sldns_buffer_limit(c->buffer) >= LDNS_HEADER_SIZE 1155 && LDNS_QDCOUNT(sldns_buffer_begin(c->buffer)) == 1); 1156 1157 sldns_buffer_set_position(c->buffer, LDNS_HEADER_SIZE); /* skip header */ 1158 1159 /* check additional section is present and that we respond with EDEs */ 1160 if(LDNS_ARCOUNT(sldns_buffer_begin(c->buffer)) != 1 1161 || !ede) { 1162 LDNS_QDCOUNT_SET(sldns_buffer_begin(c->buffer), 0); 1163 LDNS_ANCOUNT_SET(sldns_buffer_begin(c->buffer), 0); 1164 LDNS_NSCOUNT_SET(sldns_buffer_begin(c->buffer), 0); 1165 LDNS_ARCOUNT_SET(sldns_buffer_begin(c->buffer), 0); 1166 LDNS_QR_SET(sldns_buffer_begin(c->buffer)); 1167 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), 1168 LDNS_RCODE_REFUSED); 1169 sldns_buffer_set_position(c->buffer, LDNS_HEADER_SIZE); 1170 sldns_buffer_flip(c->buffer); 1171 return 1; 1172 } 1173 1174 if (!query_dname_len(c->buffer)) { 1175 LDNS_QDCOUNT_SET(sldns_buffer_begin(c->buffer), 0); 1176 LDNS_ANCOUNT_SET(sldns_buffer_begin(c->buffer), 0); 1177 LDNS_NSCOUNT_SET(sldns_buffer_begin(c->buffer), 0); 1178 LDNS_ARCOUNT_SET(sldns_buffer_begin(c->buffer), 0); 1179 LDNS_QR_SET(sldns_buffer_begin(c->buffer)); 1180 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), 1181 LDNS_RCODE_FORMERR); 1182 sldns_buffer_set_position(c->buffer, LDNS_HEADER_SIZE); 1183 sldns_buffer_flip(c->buffer); 1184 return 1; 1185 } 1186 /* space available for query type and class? */ 1187 if (sldns_buffer_remaining(c->buffer) < 2 * sizeof(uint16_t)) { 1188 LDNS_QR_SET(sldns_buffer_begin(c->buffer)); 1189 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), 1190 LDNS_RCODE_FORMERR); 1191 LDNS_QDCOUNT_SET(sldns_buffer_begin(c->buffer), 0); 1192 LDNS_ANCOUNT_SET(sldns_buffer_begin(c->buffer), 0); 1193 LDNS_NSCOUNT_SET(sldns_buffer_begin(c->buffer), 0); 1194 LDNS_ARCOUNT_SET(sldns_buffer_begin(c->buffer), 0); 1195 sldns_buffer_set_position(c->buffer, LDNS_HEADER_SIZE); 1196 sldns_buffer_flip(c->buffer); 1197 return 1; 1198 } 1199 LDNS_QR_SET(sldns_buffer_begin(c->buffer)); 1200 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), 1201 LDNS_RCODE_REFUSED); 1202 1203 sldns_buffer_skip(c->buffer, (ssize_t)sizeof(uint16_t)); /* skip qtype */ 1204 1205 sldns_buffer_skip(c->buffer, (ssize_t)sizeof(uint16_t)); /* skip qclass */ 1206 1207 /* The OPT RR to be returned should come directly after 1208 * the query, so mark this spot. 1209 */ 1210 opt_rr_mark = sldns_buffer_position(c->buffer); 1211 1212 /* Skip through the RR records */ 1213 if(LDNS_ANCOUNT(sldns_buffer_begin(c->buffer)) != 0 || 1214 LDNS_NSCOUNT(sldns_buffer_begin(c->buffer)) != 0) { 1215 if(!skip_pkt_rrs(c->buffer, 1216 ((int)LDNS_ANCOUNT(sldns_buffer_begin(c->buffer)))+ 1217 ((int)LDNS_NSCOUNT(sldns_buffer_begin(c->buffer))))) { 1218 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), 1219 LDNS_RCODE_FORMERR); 1220 LDNS_ANCOUNT_SET(sldns_buffer_begin(c->buffer), 0); 1221 LDNS_NSCOUNT_SET(sldns_buffer_begin(c->buffer), 0); 1222 LDNS_ARCOUNT_SET(sldns_buffer_begin(c->buffer), 0); 1223 sldns_buffer_set_position(c->buffer, opt_rr_mark); 1224 sldns_buffer_flip(c->buffer); 1225 return 1; 1226 } 1227 } 1228 /* Do we have a valid OPT RR here? If not return REFUSED (could be a valid TSIG or something so no FORMERR) */ 1229 /* domain name must be the root of length 1. */ 1230 if(sldns_buffer_remaining(c->buffer) < 1 || *sldns_buffer_current(c->buffer) != 0) { 1231 LDNS_ANCOUNT_SET(sldns_buffer_begin(c->buffer), 0); 1232 LDNS_NSCOUNT_SET(sldns_buffer_begin(c->buffer), 0); 1233 LDNS_ARCOUNT_SET(sldns_buffer_begin(c->buffer), 0); 1234 sldns_buffer_set_position(c->buffer, opt_rr_mark); 1235 sldns_buffer_flip(c->buffer); 1236 return 1; 1237 } else { 1238 sldns_buffer_skip(c->buffer, 1); /* skip root label */ 1239 } 1240 if(sldns_buffer_remaining(c->buffer) < 2 || 1241 sldns_buffer_read_u16(c->buffer) != LDNS_RR_TYPE_OPT) { 1242 LDNS_ANCOUNT_SET(sldns_buffer_begin(c->buffer), 0); 1243 LDNS_NSCOUNT_SET(sldns_buffer_begin(c->buffer), 0); 1244 LDNS_ARCOUNT_SET(sldns_buffer_begin(c->buffer), 0); 1245 sldns_buffer_set_position(c->buffer, opt_rr_mark); 1246 sldns_buffer_flip(c->buffer); 1247 return 1; 1248 } 1249 /* Write OPT RR directly after the query, 1250 * so without the (possibly skipped) Answer and NS RRs 1251 */ 1252 LDNS_ANCOUNT_SET(sldns_buffer_begin(c->buffer), 0); 1253 LDNS_NSCOUNT_SET(sldns_buffer_begin(c->buffer), 0); 1254 sldns_buffer_clear(c->buffer); /* reset write limit */ 1255 sldns_buffer_set_position(c->buffer, opt_rr_mark); 1256 1257 /* Check if OPT record can be written 1258 * 17 == root label (1) + RR type (2) + UDP Size (2) 1259 * + Fields (4) + rdata len (2) + EDE Option code (2) 1260 * + EDE Option length (2) + EDE info-code (2) 1261 */ 1262 if (sldns_buffer_available(c->buffer, 17) == 0) { 1263 LDNS_ARCOUNT_SET(sldns_buffer_begin(c->buffer), 0); 1264 sldns_buffer_flip(c->buffer); 1265 return 1; 1266 } 1267 1268 LDNS_ARCOUNT_SET(sldns_buffer_begin(c->buffer), 1); 1269 1270 /* root label */ 1271 sldns_buffer_write_u8(c->buffer, 0); 1272 sldns_buffer_write_u16(c->buffer, LDNS_RR_TYPE_OPT); 1273 sldns_buffer_write_u16(c->buffer, EDNS_ADVERTISED_SIZE); 1274 1275 /* write OPT Record TTL Field */ 1276 sldns_buffer_write_u32(c->buffer, 0); 1277 1278 /* write rdata len: EDE option + length + info-code */ 1279 sldns_buffer_write_u16(c->buffer, 6); 1280 1281 /* write OPTIONS; add EDE option code */ 1282 sldns_buffer_write_u16(c->buffer, LDNS_EDNS_EDE); 1283 1284 /* write single EDE option length (for just 1 info-code) */ 1285 sldns_buffer_write_u16(c->buffer, 2); 1286 1287 /* write single EDE info-code */ 1288 sldns_buffer_write_u16(c->buffer, LDNS_EDE_PROHIBITED); 1289 1290 sldns_buffer_flip(c->buffer); 1291 1292 verbose(VERB_ALGO, "attached EDE code: %d", LDNS_EDE_PROHIBITED); 1293 1294 return 1; 1295 1296 } 1297 1298 return -1; 1299 } 1300 1301 static int 1302 deny_refuse_all(struct comm_point* c, enum acl_access* acl, 1303 struct worker* worker, struct comm_reply* repinfo, 1304 struct acl_addr** acladdr, int ede, int check_proxy, 1305 struct check_request_result* check_result) 1306 { 1307 if(check_proxy) { 1308 *acladdr = acl_addr_lookup(worker->daemon->acl, 1309 &repinfo->remote_addr, repinfo->remote_addrlen); 1310 } else { 1311 *acladdr = acl_addr_lookup(worker->daemon->acl, 1312 &repinfo->client_addr, repinfo->client_addrlen); 1313 } 1314 /* If there is no ACL based on client IP use the interface ACL. */ 1315 if(!(*acladdr) && c->socket) { 1316 *acladdr = c->socket->acl; 1317 } 1318 *acl = acl_get_control(*acladdr); 1319 return deny_refuse(c, *acl, acl_deny, acl_refuse, worker, repinfo, 1320 *acladdr, ede, check_result); 1321 } 1322 1323 static int 1324 deny_refuse_non_local(struct comm_point* c, enum acl_access acl, 1325 struct worker* worker, struct comm_reply* repinfo, 1326 struct acl_addr* acladdr, int ede, 1327 struct check_request_result* check_result) 1328 { 1329 return deny_refuse(c, acl, acl_deny_non_local, acl_refuse_non_local, 1330 worker, repinfo, acladdr, ede, check_result); 1331 } 1332 1333 /* Check if the query is blocked by source IP rate limiting. 1334 * Returns 1 if it passes the check, 0 otherwise. */ 1335 static int 1336 check_ip_ratelimit(struct worker* worker, struct sockaddr_storage* addr, 1337 socklen_t addrlen, int has_cookie, sldns_buffer* pkt) 1338 { 1339 if(!infra_ip_ratelimit_inc(worker->env.infra_cache, addr, addrlen, 1340 *worker->env.now, has_cookie, 1341 worker->env.cfg->ip_ratelimit_backoff, pkt)) { 1342 /* See if we can pass through with slip factor */ 1343 if(!has_cookie && worker->env.cfg->ip_ratelimit_factor != 0 && 1344 ub_random_max(worker->env.rnd, 1345 worker->env.cfg->ip_ratelimit_factor) == 0) { 1346 char addrbuf[128]; 1347 addr_to_str(addr, addrlen, addrbuf, sizeof(addrbuf)); 1348 verbose(VERB_QUERY, "ip_ratelimit allowed through for " 1349 "ip address %s because of slip in " 1350 "ip_ratelimit_factor", addrbuf); 1351 return 1; 1352 } 1353 return 0; 1354 } 1355 return 1; 1356 } 1357 1358 int 1359 worker_handle_request(struct comm_point* c, void* arg, int error, 1360 struct comm_reply* repinfo) 1361 { 1362 struct worker* worker = (struct worker*)arg; 1363 int ret; 1364 hashvalue_type h; 1365 struct lruhash_entry* e; 1366 struct query_info qinfo; 1367 struct edns_data edns; 1368 struct edns_option* original_edns_list = NULL; 1369 enum acl_access acl; 1370 struct acl_addr* acladdr; 1371 int pre_edns_ip_ratelimit = 1; 1372 int rc = 0; 1373 int need_drop = 0; 1374 int is_expired_answer = 0; 1375 int is_secure_answer = 0; 1376 int rpz_passthru = 0; 1377 long long wait_queue_time = 0; 1378 /* We might have to chase a CNAME chain internally, in which case 1379 * we'll have up to two replies and combine them to build a complete 1380 * answer. These variables control this case. */ 1381 struct ub_packed_rrset_key* alias_rrset = NULL; 1382 struct reply_info* partial_rep = NULL; 1383 struct query_info* lookup_qinfo = &qinfo; 1384 struct query_info qinfo_tmp; /* placeholder for lookup_qinfo */ 1385 struct respip_client_info* cinfo = NULL, cinfo_tmp; 1386 struct timeval wait_time; 1387 struct check_request_result check_result = {0,0}; 1388 memset(&qinfo, 0, sizeof(qinfo)); 1389 1390 if((error != NETEVENT_NOERROR && error != NETEVENT_DONE)|| !repinfo) { 1391 /* some bad tcp query DNS formats give these error calls */ 1392 verbose(VERB_ALGO, "handle request called with err=%d", error); 1393 return 0; 1394 } 1395 1396 if (worker->env.cfg->sock_queue_timeout && timeval_isset(&c->recv_tv)) { 1397 timeval_subtract(&wait_time, worker->env.now_tv, &c->recv_tv); 1398 wait_queue_time = wait_time.tv_sec * 1000000 + wait_time.tv_usec; 1399 if (worker->stats.max_query_time_us < wait_queue_time) 1400 worker->stats.max_query_time_us = wait_queue_time; 1401 if(wait_queue_time > 1402 (long long)(worker->env.cfg->sock_queue_timeout * 1000000)) { 1403 /* count and drop queries that were sitting in the socket queue too long */ 1404 worker->stats.num_queries_timed_out++; 1405 return 0; 1406 } 1407 } 1408 1409 #ifdef USE_DNSCRYPT 1410 repinfo->max_udp_size = worker->daemon->cfg->max_udp_size; 1411 if(!dnsc_handle_curved_request(worker->daemon->dnscenv, repinfo)) { 1412 worker->stats.num_query_dnscrypt_crypted_malformed++; 1413 return 0; 1414 } 1415 if(c->dnscrypt && !repinfo->is_dnscrypted) { 1416 char buf[LDNS_MAX_DOMAINLEN+1]; 1417 /* Check if this is unencrypted and asking for certs */ 1418 worker_check_request(c->buffer, worker, &check_result); 1419 if(check_result.value != 0) { 1420 verbose(VERB_ALGO, 1421 "dnscrypt: worker check request: bad query."); 1422 log_addr(VERB_CLIENT,"from",&repinfo->client_addr, 1423 repinfo->client_addrlen); 1424 comm_point_drop_reply(repinfo); 1425 return 0; 1426 } 1427 if(!query_info_parse(&qinfo, c->buffer)) { 1428 verbose(VERB_ALGO, 1429 "dnscrypt: worker parse request: formerror."); 1430 log_addr(VERB_CLIENT, "from", &repinfo->client_addr, 1431 repinfo->client_addrlen); 1432 comm_point_drop_reply(repinfo); 1433 return 0; 1434 } 1435 dname_str(qinfo.qname, buf); 1436 if(!(qinfo.qtype == LDNS_RR_TYPE_TXT && 1437 strcasecmp(buf, 1438 worker->daemon->dnscenv->provider_name) == 0)) { 1439 verbose(VERB_ALGO, 1440 "dnscrypt: not TXT \"%s\". Received: %s \"%s\"", 1441 worker->daemon->dnscenv->provider_name, 1442 sldns_rr_descript(qinfo.qtype)->_name, 1443 buf); 1444 comm_point_drop_reply(repinfo); 1445 worker->stats.num_query_dnscrypt_cleartext++; 1446 return 0; 1447 } 1448 worker->stats.num_query_dnscrypt_cert++; 1449 sldns_buffer_rewind(c->buffer); 1450 } else if(c->dnscrypt && repinfo->is_dnscrypted) { 1451 worker->stats.num_query_dnscrypt_crypted++; 1452 } 1453 #endif 1454 #ifdef USE_DNSTAP 1455 /* 1456 * sending src (client)/dst (local service) addresses over DNSTAP from incoming request handler 1457 */ 1458 if(worker->dtenv.log_client_query_messages) { 1459 log_addr(VERB_ALGO, "request from client", &repinfo->client_addr, repinfo->client_addrlen); 1460 log_addr(VERB_ALGO, "to local addr", (void*)repinfo->c->socket->addr, repinfo->c->socket->addrlen); 1461 dt_msg_send_client_query(&worker->dtenv, &repinfo->client_addr, (void*)repinfo->c->socket->addr, c->type, c->ssl, c->buffer, 1462 ((worker->env.cfg->sock_queue_timeout && timeval_isset(&c->recv_tv))?&c->recv_tv:NULL)); 1463 } 1464 #endif 1465 /* Check deny/refuse ACLs */ 1466 if(repinfo->is_proxied) { 1467 if((ret=deny_refuse_all(c, &acl, worker, repinfo, &acladdr, 1468 worker->env.cfg->ede, 1, &check_result)) != -1) { 1469 if(ret == 1) 1470 goto send_reply; 1471 return ret; 1472 } 1473 } 1474 if((ret=deny_refuse_all(c, &acl, worker, repinfo, &acladdr, 1475 worker->env.cfg->ede, 0, &check_result)) != -1) { 1476 if(ret == 1) 1477 goto send_reply; 1478 return ret; 1479 } 1480 1481 worker_check_request(c->buffer, worker, &check_result); 1482 if(check_result.value != 0) { 1483 verbose(VERB_ALGO, "worker check request: bad query."); 1484 log_addr(VERB_CLIENT,"from",&repinfo->client_addr, repinfo->client_addrlen); 1485 if(check_result.value != -1) { 1486 LDNS_QR_SET(sldns_buffer_begin(c->buffer)); 1487 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), 1488 check_result.value); 1489 return 1; 1490 } 1491 comm_point_drop_reply(repinfo); 1492 return 0; 1493 } 1494 1495 worker->stats.num_queries++; 1496 pre_edns_ip_ratelimit = !worker->env.cfg->do_answer_cookie 1497 || sldns_buffer_limit(c->buffer) < LDNS_HEADER_SIZE 1498 || LDNS_ARCOUNT(sldns_buffer_begin(c->buffer)) == 0; 1499 1500 /* If the IP rate limiting check needs extra EDNS information (e.g., 1501 * DNS Cookies) postpone the check until after EDNS is parsed. */ 1502 if(pre_edns_ip_ratelimit) { 1503 /* NOTE: we always check the repinfo->client_address. 1504 * IP ratelimiting is implicitly disabled for proxies. */ 1505 if(!check_ip_ratelimit(worker, &repinfo->client_addr, 1506 repinfo->client_addrlen, 0, c->buffer)) { 1507 worker->stats.num_queries_ip_ratelimited++; 1508 comm_point_drop_reply(repinfo); 1509 return 0; 1510 } 1511 } 1512 1513 if(!query_info_parse(&qinfo, c->buffer)) { 1514 verbose(VERB_ALGO, "worker parse request: formerror."); 1515 log_addr(VERB_CLIENT, "from", &repinfo->client_addr, 1516 repinfo->client_addrlen); 1517 memset(&qinfo, 0, sizeof(qinfo)); /* zero qinfo.qname */ 1518 if(worker_err_ratelimit(worker, LDNS_RCODE_FORMERR) == -1) { 1519 comm_point_drop_reply(repinfo); 1520 return 0; 1521 } 1522 sldns_buffer_rewind(c->buffer); 1523 LDNS_QR_SET(sldns_buffer_begin(c->buffer)); 1524 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), 1525 LDNS_RCODE_FORMERR); 1526 goto send_reply; 1527 } 1528 if(worker->env.cfg->log_queries) { 1529 char ip[128]; 1530 addr_to_str(&repinfo->client_addr, repinfo->client_addrlen, ip, sizeof(ip)); 1531 log_query_in(ip, qinfo.qname, qinfo.qtype, qinfo.qclass); 1532 } 1533 if(qinfo.qtype == LDNS_RR_TYPE_AXFR || 1534 qinfo.qtype == LDNS_RR_TYPE_IXFR) { 1535 verbose(VERB_ALGO, "worker request: refused zone transfer."); 1536 log_addr(VERB_CLIENT, "from", &repinfo->client_addr, 1537 repinfo->client_addrlen); 1538 sldns_buffer_rewind(c->buffer); 1539 LDNS_QR_SET(sldns_buffer_begin(c->buffer)); 1540 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), 1541 LDNS_RCODE_REFUSED); 1542 if(worker->stats.extended) { 1543 worker->stats.qtype[qinfo.qtype]++; 1544 } 1545 goto send_reply; 1546 } 1547 if(qinfo.qtype == LDNS_RR_TYPE_OPT || 1548 qinfo.qtype == LDNS_RR_TYPE_TSIG || 1549 qinfo.qtype == LDNS_RR_TYPE_TKEY || 1550 qinfo.qtype == LDNS_RR_TYPE_MAILA || 1551 qinfo.qtype == LDNS_RR_TYPE_MAILB || 1552 (qinfo.qtype >= 128 && qinfo.qtype <= 248)) { 1553 verbose(VERB_ALGO, "worker request: formerror for meta-type."); 1554 log_addr(VERB_CLIENT, "from", &repinfo->client_addr, 1555 repinfo->client_addrlen); 1556 if(worker_err_ratelimit(worker, LDNS_RCODE_FORMERR) == -1) { 1557 comm_point_drop_reply(repinfo); 1558 return 0; 1559 } 1560 sldns_buffer_rewind(c->buffer); 1561 LDNS_QR_SET(sldns_buffer_begin(c->buffer)); 1562 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), 1563 LDNS_RCODE_FORMERR); 1564 if(worker->stats.extended) { 1565 worker->stats.qtype[qinfo.qtype]++; 1566 } 1567 goto send_reply; 1568 } 1569 if((ret=parse_edns_from_query_pkt( 1570 c->buffer, &edns, worker->env.cfg, c, repinfo, 1571 (worker->env.now ? *worker->env.now : time(NULL)), 1572 worker->scratchpad, 1573 worker->daemon->cookie_secrets)) != 0) { 1574 struct edns_data reply_edns; 1575 verbose(VERB_ALGO, "worker parse edns: formerror."); 1576 log_addr(VERB_CLIENT, "from", &repinfo->client_addr, 1577 repinfo->client_addrlen); 1578 memset(&reply_edns, 0, sizeof(reply_edns)); 1579 reply_edns.edns_present = 1; 1580 error_encode(c->buffer, ret, &qinfo, 1581 *(uint16_t*)(void *)sldns_buffer_begin(c->buffer), 1582 sldns_buffer_read_u16_at(c->buffer, 2), &reply_edns); 1583 regional_free_all(worker->scratchpad); 1584 goto send_reply; 1585 } 1586 if(edns.edns_present) { 1587 if(edns.edns_version != 0) { 1588 edns.opt_list_in = NULL; 1589 edns.opt_list_out = NULL; 1590 edns.opt_list_inplace_cb_out = NULL; 1591 verbose(VERB_ALGO, "query with bad edns version."); 1592 log_addr(VERB_CLIENT, "from", &repinfo->client_addr, 1593 repinfo->client_addrlen); 1594 extended_error_encode(c->buffer, EDNS_RCODE_BADVERS, &qinfo, 1595 *(uint16_t*)(void *)sldns_buffer_begin(c->buffer), 1596 sldns_buffer_read_u16_at(c->buffer, 2), 0, &edns); 1597 regional_free_all(worker->scratchpad); 1598 goto send_reply; 1599 } 1600 if(edns.udp_size < NORMAL_UDP_SIZE && 1601 worker->daemon->cfg->harden_short_bufsize) { 1602 verbose(VERB_QUERY, "worker request: EDNS bufsize %d ignored", 1603 (int)edns.udp_size); 1604 log_addr(VERB_CLIENT, "from", &repinfo->client_addr, 1605 repinfo->client_addrlen); 1606 edns.udp_size = NORMAL_UDP_SIZE; 1607 } 1608 } 1609 1610 /* Get stats for cookies */ 1611 server_stats_downstream_cookie(&worker->stats, &edns); 1612 1613 /* If the IP rate limiting check was postponed, check now. */ 1614 if(!pre_edns_ip_ratelimit) { 1615 /* NOTE: we always check the repinfo->client_address. 1616 * IP ratelimiting is implicitly disabled for proxies. */ 1617 if(!check_ip_ratelimit(worker, &repinfo->client_addr, 1618 repinfo->client_addrlen, edns.cookie_valid, 1619 c->buffer)) { 1620 worker->stats.num_queries_ip_ratelimited++; 1621 comm_point_drop_reply(repinfo); 1622 return 0; 1623 } 1624 } 1625 1626 /* "if, else if" sequence below deals with downstream DNS Cookies */ 1627 if(acl != acl_allow_cookie) 1628 ; /* pass; No cookie downstream processing whatsoever */ 1629 1630 else if(edns.cookie_valid) 1631 ; /* pass; Valid cookie is good! */ 1632 1633 else if(c->type != comm_udp) 1634 ; /* pass; Stateful transport */ 1635 1636 else if(edns.cookie_present) { 1637 /* Cookie present, but not valid: Cookie was bad! */ 1638 extended_error_encode(c->buffer, 1639 LDNS_EXT_RCODE_BADCOOKIE, &qinfo, 1640 *(uint16_t*)(void *) 1641 sldns_buffer_begin(c->buffer), 1642 sldns_buffer_read_u16_at(c->buffer, 2), 1643 0, &edns); 1644 regional_free_all(worker->scratchpad); 1645 goto send_reply; 1646 } else { 1647 /* Cookie required, but no cookie present on UDP */ 1648 verbose(VERB_ALGO, "worker request: " 1649 "need cookie or stateful transport"); 1650 log_addr(VERB_ALGO, "from",&repinfo->remote_addr 1651 , repinfo->remote_addrlen); 1652 EDNS_OPT_LIST_APPEND_EDE(&edns.opt_list_out, 1653 worker->scratchpad, LDNS_EDE_OTHER, 1654 "DNS Cookie needed for UDP replies"); 1655 error_encode(c->buffer, 1656 (LDNS_RCODE_REFUSED|BIT_TC), &qinfo, 1657 *(uint16_t*)(void *) 1658 sldns_buffer_begin(c->buffer), 1659 sldns_buffer_read_u16_at(c->buffer, 2), 1660 &edns); 1661 regional_free_all(worker->scratchpad); 1662 goto send_reply; 1663 } 1664 1665 if(edns.udp_size > worker->daemon->cfg->max_udp_size && 1666 c->type == comm_udp) { 1667 verbose(VERB_QUERY, 1668 "worker request: max UDP reply size modified" 1669 " (%d to max-udp-size)", (int)edns.udp_size); 1670 log_addr(VERB_CLIENT, "from", &repinfo->client_addr, 1671 repinfo->client_addrlen); 1672 edns.udp_size = worker->daemon->cfg->max_udp_size; 1673 } 1674 if(edns.udp_size < LDNS_HEADER_SIZE) { 1675 verbose(VERB_ALGO, "worker request: edns is too small."); 1676 log_addr(VERB_CLIENT, "from", &repinfo->client_addr, 1677 repinfo->client_addrlen); 1678 LDNS_QR_SET(sldns_buffer_begin(c->buffer)); 1679 LDNS_TC_SET(sldns_buffer_begin(c->buffer)); 1680 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), 1681 LDNS_RCODE_SERVFAIL); 1682 sldns_buffer_set_position(c->buffer, LDNS_HEADER_SIZE); 1683 sldns_buffer_write_at(c->buffer, 4, 1684 (uint8_t*)"\0\0\0\0\0\0\0\0", 8); 1685 sldns_buffer_flip(c->buffer); 1686 regional_free_all(worker->scratchpad); 1687 goto send_reply; 1688 } 1689 if(worker->stats.extended) 1690 server_stats_insquery(&worker->stats, c, qinfo.qtype, 1691 qinfo.qclass, &edns, repinfo); 1692 if(c->type != comm_udp) 1693 edns.udp_size = 65535; /* max size for TCP replies */ 1694 if(qinfo.qclass == LDNS_RR_CLASS_CH && answer_chaos(worker, &qinfo, 1695 &edns, repinfo, c->buffer)) { 1696 regional_free_all(worker->scratchpad); 1697 goto send_reply; 1698 } 1699 if(LDNS_OPCODE_WIRE(sldns_buffer_begin(c->buffer)) == 1700 LDNS_PACKET_NOTIFY) { 1701 answer_notify(worker, &qinfo, &edns, c->buffer, 1702 &repinfo->client_addr, repinfo->client_addrlen); 1703 regional_free_all(worker->scratchpad); 1704 goto send_reply; 1705 } 1706 if(local_zones_answer(worker->daemon->local_zones, &worker->env, &qinfo, 1707 &edns, c->buffer, worker->scratchpad, repinfo, acladdr->taglist, 1708 acladdr->taglen, acladdr->tag_actions, 1709 acladdr->tag_actions_size, acladdr->tag_datas, 1710 acladdr->tag_datas_size, worker->daemon->cfg->tagname, 1711 worker->daemon->cfg->num_tags, acladdr->view)) { 1712 regional_free_all(worker->scratchpad); 1713 if(sldns_buffer_limit(c->buffer) == 0) { 1714 comm_point_drop_reply(repinfo); 1715 return 0; 1716 } 1717 goto send_reply; 1718 } 1719 if(worker->env.auth_zones && 1720 rpz_callback_from_worker_request(worker->env.auth_zones, 1721 &worker->env, &qinfo, &edns, c->buffer, worker->scratchpad, 1722 repinfo, acladdr->taglist, acladdr->taglen, &worker->stats, 1723 &rpz_passthru)) { 1724 regional_free_all(worker->scratchpad); 1725 if(sldns_buffer_limit(c->buffer) == 0) { 1726 comm_point_drop_reply(repinfo); 1727 return 0; 1728 } 1729 goto send_reply; 1730 } 1731 if(worker->env.auth_zones && 1732 auth_zones_answer(worker->env.auth_zones, &worker->env, 1733 &qinfo, &edns, repinfo, c->buffer, worker->scratchpad)) { 1734 regional_free_all(worker->scratchpad); 1735 if(sldns_buffer_limit(c->buffer) == 0) { 1736 comm_point_drop_reply(repinfo); 1737 return 0; 1738 } 1739 /* set RA for everyone that can have recursion (based on 1740 * access control list) */ 1741 if(LDNS_RD_WIRE(sldns_buffer_begin(c->buffer)) && 1742 acl != acl_deny_non_local && acl != acl_refuse_non_local) 1743 LDNS_RA_SET(sldns_buffer_begin(c->buffer)); 1744 goto send_reply; 1745 } 1746 1747 /* We've looked in our local zones. If the answer isn't there, we 1748 * might need to bail out based on ACLs now. */ 1749 if((ret=deny_refuse_non_local(c, acl, worker, repinfo, acladdr, 1750 worker->env.cfg->ede, &check_result)) != -1) 1751 { 1752 regional_free_all(worker->scratchpad); 1753 if(ret == 1) 1754 goto send_reply; 1755 return ret; 1756 } 1757 1758 /* If this request does not have the recursion bit set, verify 1759 * ACLs allow the recursion bit to be treated as set. */ 1760 if(!(LDNS_RD_WIRE(sldns_buffer_begin(c->buffer))) && 1761 acl == acl_allow_setrd ) { 1762 LDNS_RD_SET(sldns_buffer_begin(c->buffer)); 1763 } 1764 1765 /* If this request does not have the recursion bit set, verify 1766 * ACLs allow the snooping. */ 1767 if(!(LDNS_RD_WIRE(sldns_buffer_begin(c->buffer))) && 1768 acl != acl_allow_snoop ) { 1769 if(worker->env.cfg->ede) { 1770 EDNS_OPT_LIST_APPEND_EDE(&edns.opt_list_out, 1771 worker->scratchpad, LDNS_EDE_NOT_AUTHORITATIVE, ""); 1772 } 1773 error_encode(c->buffer, LDNS_RCODE_REFUSED, &qinfo, 1774 *(uint16_t*)(void *)sldns_buffer_begin(c->buffer), 1775 sldns_buffer_read_u16_at(c->buffer, 2), &edns); 1776 regional_free_all(worker->scratchpad); 1777 log_addr(VERB_ALGO, "refused nonrec (cache snoop) query from", 1778 &repinfo->client_addr, repinfo->client_addrlen); 1779 1780 goto send_reply; 1781 } 1782 1783 /* If we've found a local alias, replace the qname with the alias 1784 * target before resolving it. */ 1785 if(qinfo.local_alias) { 1786 struct ub_packed_rrset_key* rrset = qinfo.local_alias->rrset; 1787 struct packed_rrset_data* d = rrset->entry.data; 1788 1789 /* Sanity check: our current implementation only supports 1790 * a single CNAME RRset as a local alias. */ 1791 if(qinfo.local_alias->next || 1792 rrset->rk.type != htons(LDNS_RR_TYPE_CNAME) || 1793 d->count != 1) { 1794 log_err("assumption failure: unexpected local alias"); 1795 regional_free_all(worker->scratchpad); 1796 return 0; /* drop it */ 1797 } 1798 qinfo.qname = d->rr_data[0] + 2; 1799 qinfo.qname_len = d->rr_len[0] - 2; 1800 } 1801 1802 /* If we may apply IP-based actions to the answer, build the client 1803 * information. As this can be expensive, skip it if there is 1804 * absolutely no possibility of it. */ 1805 if((worker->daemon->use_response_ip || worker->daemon->use_rpz) && 1806 (qinfo.qtype == LDNS_RR_TYPE_A || 1807 qinfo.qtype == LDNS_RR_TYPE_AAAA || 1808 qinfo.qtype == LDNS_RR_TYPE_ANY)) { 1809 cinfo_tmp.taglist = acladdr->taglist; 1810 cinfo_tmp.taglen = acladdr->taglen; 1811 cinfo_tmp.tag_actions = acladdr->tag_actions; 1812 cinfo_tmp.tag_actions_size = acladdr->tag_actions_size; 1813 cinfo_tmp.tag_datas = acladdr->tag_datas; 1814 cinfo_tmp.tag_datas_size = acladdr->tag_datas_size; 1815 cinfo_tmp.view = acladdr->view; 1816 cinfo_tmp.respip_set = worker->daemon->respip_set; 1817 cinfo = &cinfo_tmp; 1818 } 1819 1820 /* Keep the original edns list around. The pointer could change if there is 1821 * a cached answer (through the inplace callback function there). 1822 * No need to actually copy the contents as they shouldn't change. 1823 * Used while prefetching and subnet is enabled. */ 1824 original_edns_list = edns.opt_list_in; 1825 lookup_cache: 1826 /* Lookup the cache. In case we chase an intermediate CNAME chain 1827 * this is a two-pass operation, and lookup_qinfo is different for 1828 * each pass. We should still pass the original qinfo to 1829 * answer_from_cache(), however, since it's used to build the reply. */ 1830 if(!edns_bypass_cache_stage(edns.opt_list_in, &worker->env)) { 1831 is_expired_answer = 0; 1832 is_secure_answer = 0; 1833 h = query_info_hash(lookup_qinfo, sldns_buffer_read_u16_at(c->buffer, 2)); 1834 if((e=slabhash_lookup(worker->env.msg_cache, h, lookup_qinfo, 0))) { 1835 struct reply_info* rep = (struct reply_info*)e->data; 1836 /* answer from cache - we have acquired a readlock on it */ 1837 if(answer_from_cache(worker, &qinfo, cinfo, &need_drop, 1838 &is_expired_answer, &is_secure_answer, 1839 &alias_rrset, &partial_rep, rep, 1840 *(uint16_t*)(void *)sldns_buffer_begin(c->buffer), 1841 sldns_buffer_read_u16_at(c->buffer, 2), repinfo, 1842 &edns)) { 1843 /* prefetch it if the prefetch TTL expired. 1844 * Note that if there is more than one pass 1845 * its qname must be that used for cache 1846 * lookup. */ 1847 if((worker->env.cfg->prefetch && 1848 *worker->env.now >= rep->prefetch_ttl) || 1849 (worker->env.cfg->serve_expired && 1850 *worker->env.now > rep->ttl)) { 1851 1852 time_t leeway = rep->ttl - *worker->env.now; 1853 if(rep->ttl < *worker->env.now) 1854 leeway = 0; 1855 lock_rw_unlock(&e->lock); 1856 1857 reply_and_prefetch(worker, lookup_qinfo, 1858 sldns_buffer_read_u16_at(c->buffer, 2), 1859 repinfo, leeway, 1860 (partial_rep || need_drop), 1861 rpz_passthru, 1862 original_edns_list); 1863 if(!partial_rep) { 1864 rc = 0; 1865 regional_free_all(worker->scratchpad); 1866 goto send_reply_rc; 1867 } 1868 } else if(!partial_rep) { 1869 lock_rw_unlock(&e->lock); 1870 regional_free_all(worker->scratchpad); 1871 goto send_reply; 1872 } else { 1873 /* Note that we've already released the 1874 * lock if we're here after prefetch. */ 1875 lock_rw_unlock(&e->lock); 1876 } 1877 /* We've found a partial reply ending with an 1878 * alias. Replace the lookup qinfo for the 1879 * alias target and lookup the cache again to 1880 * (possibly) complete the reply. As we're 1881 * passing the "base" reply, there will be no 1882 * more alias chasing. */ 1883 memset(&qinfo_tmp, 0, sizeof(qinfo_tmp)); 1884 get_cname_target(alias_rrset, &qinfo_tmp.qname, 1885 &qinfo_tmp.qname_len); 1886 if(!qinfo_tmp.qname) { 1887 log_err("unexpected: invalid answer alias"); 1888 regional_free_all(worker->scratchpad); 1889 return 0; /* drop query */ 1890 } 1891 qinfo_tmp.qtype = qinfo.qtype; 1892 qinfo_tmp.qclass = qinfo.qclass; 1893 lookup_qinfo = &qinfo_tmp; 1894 goto lookup_cache; 1895 } 1896 verbose(VERB_ALGO, "answer from the cache failed"); 1897 lock_rw_unlock(&e->lock); 1898 } 1899 1900 if(!LDNS_RD_WIRE(sldns_buffer_begin(c->buffer))) { 1901 if(answer_norec_from_cache(worker, &qinfo, 1902 *(uint16_t*)(void *)sldns_buffer_begin(c->buffer), 1903 sldns_buffer_read_u16_at(c->buffer, 2), repinfo, 1904 &edns)) { 1905 regional_free_all(worker->scratchpad); 1906 goto send_reply; 1907 } 1908 verbose(VERB_ALGO, "answer norec from cache -- " 1909 "need to validate or not primed"); 1910 } 1911 } 1912 sldns_buffer_rewind(c->buffer); 1913 server_stats_querymiss(&worker->stats, worker); 1914 1915 if(verbosity >= VERB_CLIENT) { 1916 if(c->type == comm_udp) 1917 log_addr(VERB_CLIENT, "udp request from", 1918 &repinfo->client_addr, repinfo->client_addrlen); 1919 else log_addr(VERB_CLIENT, "tcp request from", 1920 &repinfo->client_addr, repinfo->client_addrlen); 1921 } 1922 1923 /* grab a work request structure for this new request */ 1924 mesh_new_client(worker->env.mesh, &qinfo, cinfo, 1925 sldns_buffer_read_u16_at(c->buffer, 2), 1926 &edns, repinfo, *(uint16_t*)(void *)sldns_buffer_begin(c->buffer), 1927 rpz_passthru); 1928 regional_free_all(worker->scratchpad); 1929 worker_mem_report(worker, NULL); 1930 return 0; 1931 1932 send_reply: 1933 rc = 1; 1934 send_reply_rc: 1935 if(need_drop) { 1936 comm_point_drop_reply(repinfo); 1937 return 0; 1938 } 1939 if(is_expired_answer) { 1940 worker->stats.ans_expired++; 1941 } 1942 server_stats_insrcode(&worker->stats, c->buffer); 1943 if(worker->stats.extended) { 1944 if(is_secure_answer) worker->stats.ans_secure++; 1945 } 1946 #ifdef USE_DNSTAP 1947 /* 1948 * sending src (client)/dst (local service) addresses over DNSTAP from send_reply code label (when we serviced local zone for ex.) 1949 */ 1950 if(worker->dtenv.log_client_response_messages && rc !=0) { 1951 log_addr(VERB_ALGO, "from local addr", (void*)repinfo->c->socket->addr, repinfo->c->socket->addrlen); 1952 log_addr(VERB_ALGO, "response to client", &repinfo->client_addr, repinfo->client_addrlen); 1953 dt_msg_send_client_response(&worker->dtenv, &repinfo->client_addr, (void*)repinfo->c->socket->addr, c->type, c->ssl, c->buffer); 1954 } 1955 #endif 1956 if(worker->env.cfg->log_replies) 1957 { 1958 struct timeval tv; 1959 memset(&tv, 0, sizeof(tv)); 1960 if(qinfo.local_alias && qinfo.local_alias->rrset && 1961 qinfo.local_alias->rrset->rk.dname) { 1962 /* log original qname, before the local alias was 1963 * used to resolve that CNAME to something else */ 1964 qinfo.qname = qinfo.local_alias->rrset->rk.dname; 1965 log_reply_info(NO_VERBOSE, &qinfo, 1966 &repinfo->client_addr, repinfo->client_addrlen, 1967 tv, 1, c->buffer, 1968 (worker->env.cfg->log_destaddr?(void*)repinfo->c->socket->addr:NULL), 1969 c->type); 1970 } else { 1971 log_reply_info(NO_VERBOSE, &qinfo, 1972 &repinfo->client_addr, repinfo->client_addrlen, 1973 tv, 1, c->buffer, 1974 (worker->env.cfg->log_destaddr?(void*)repinfo->c->socket->addr:NULL), 1975 c->type); 1976 } 1977 } 1978 #ifdef USE_DNSCRYPT 1979 if(!dnsc_handle_uncurved_request(repinfo)) { 1980 return 0; 1981 } 1982 #endif 1983 return rc; 1984 } 1985 1986 void 1987 worker_sighandler(int sig, void* arg) 1988 { 1989 /* note that log, print, syscalls here give race conditions. 1990 * And cause hangups if the log-lock is held by the application. */ 1991 struct worker* worker = (struct worker*)arg; 1992 switch(sig) { 1993 #ifdef SIGHUP 1994 case SIGHUP: 1995 comm_base_exit(worker->base); 1996 break; 1997 #endif 1998 #ifdef SIGBREAK 1999 case SIGBREAK: 2000 #endif 2001 case SIGINT: 2002 worker->need_to_exit = 1; 2003 comm_base_exit(worker->base); 2004 break; 2005 #ifdef SIGQUIT 2006 case SIGQUIT: 2007 worker->need_to_exit = 1; 2008 comm_base_exit(worker->base); 2009 break; 2010 #endif 2011 case SIGTERM: 2012 worker->need_to_exit = 1; 2013 comm_base_exit(worker->base); 2014 break; 2015 default: 2016 /* unknown signal, ignored */ 2017 break; 2018 } 2019 } 2020 2021 /** restart statistics timer for worker, if enabled */ 2022 static void 2023 worker_restart_timer(struct worker* worker) 2024 { 2025 if(worker->env.cfg->stat_interval > 0) { 2026 struct timeval tv; 2027 #ifndef S_SPLINT_S 2028 tv.tv_sec = worker->env.cfg->stat_interval; 2029 tv.tv_usec = 0; 2030 #endif 2031 comm_timer_set(worker->stat_timer, &tv); 2032 } 2033 } 2034 2035 void worker_stat_timer_cb(void* arg) 2036 { 2037 struct worker* worker = (struct worker*)arg; 2038 server_stats_log(&worker->stats, worker, worker->thread_num); 2039 mesh_stats(worker->env.mesh, "mesh has"); 2040 worker_mem_report(worker, NULL); 2041 /* SHM is enabled, process data to SHM */ 2042 if (worker->daemon->cfg->shm_enable) { 2043 shm_main_run(worker); 2044 } 2045 if(!worker->daemon->cfg->stat_cumulative) { 2046 worker_stats_clear(worker); 2047 } 2048 /* start next timer */ 2049 worker_restart_timer(worker); 2050 } 2051 2052 void worker_probe_timer_cb(void* arg) 2053 { 2054 struct worker* worker = (struct worker*)arg; 2055 struct timeval tv; 2056 #ifndef S_SPLINT_S 2057 tv.tv_sec = (time_t)autr_probe_timer(&worker->env); 2058 tv.tv_usec = 0; 2059 #endif 2060 if(tv.tv_sec != 0) 2061 comm_timer_set(worker->env.probe_timer, &tv); 2062 } 2063 2064 struct worker* 2065 worker_create(struct daemon* daemon, int id, int* ports, int n) 2066 { 2067 unsigned int seed; 2068 struct worker* worker = (struct worker*)calloc(1, 2069 sizeof(struct worker)); 2070 if(!worker) 2071 return NULL; 2072 worker->numports = n; 2073 worker->ports = (int*)memdup(ports, sizeof(int)*n); 2074 if(!worker->ports) { 2075 free(worker); 2076 return NULL; 2077 } 2078 worker->daemon = daemon; 2079 worker->thread_num = id; 2080 if(!(worker->cmd = tube_create())) { 2081 free(worker->ports); 2082 free(worker); 2083 return NULL; 2084 } 2085 /* create random state here to avoid locking trouble in RAND_bytes */ 2086 if(!(worker->rndstate = ub_initstate(daemon->rand))) { 2087 log_err("could not init random numbers."); 2088 tube_delete(worker->cmd); 2089 free(worker->ports); 2090 free(worker); 2091 return NULL; 2092 } 2093 explicit_bzero(&seed, sizeof(seed)); 2094 return worker; 2095 } 2096 2097 int 2098 worker_init(struct worker* worker, struct config_file *cfg, 2099 struct listen_port* ports, int do_sigs) 2100 { 2101 #ifdef USE_DNSTAP 2102 struct dt_env* dtenv = &worker->dtenv; 2103 #else 2104 void* dtenv = NULL; 2105 #endif 2106 #ifdef HAVE_GETTID 2107 worker->thread_tid = gettid(); 2108 #endif 2109 worker->need_to_exit = 0; 2110 worker->base = comm_base_create(do_sigs); 2111 if(!worker->base) { 2112 log_err("could not create event handling base"); 2113 worker_delete(worker); 2114 return 0; 2115 } 2116 comm_base_set_slow_accept_handlers(worker->base, &worker_stop_accept, 2117 &worker_start_accept, worker); 2118 if(do_sigs) { 2119 #ifdef SIGHUP 2120 ub_thread_sig_unblock(SIGHUP); 2121 #endif 2122 #ifdef SIGBREAK 2123 ub_thread_sig_unblock(SIGBREAK); 2124 #endif 2125 ub_thread_sig_unblock(SIGINT); 2126 #ifdef SIGQUIT 2127 ub_thread_sig_unblock(SIGQUIT); 2128 #endif 2129 ub_thread_sig_unblock(SIGTERM); 2130 #ifndef LIBEVENT_SIGNAL_PROBLEM 2131 worker->comsig = comm_signal_create(worker->base, 2132 worker_sighandler, worker); 2133 if(!worker->comsig 2134 #ifdef SIGHUP 2135 || !comm_signal_bind(worker->comsig, SIGHUP) 2136 #endif 2137 #ifdef SIGQUIT 2138 || !comm_signal_bind(worker->comsig, SIGQUIT) 2139 #endif 2140 || !comm_signal_bind(worker->comsig, SIGTERM) 2141 #ifdef SIGBREAK 2142 || !comm_signal_bind(worker->comsig, SIGBREAK) 2143 #endif 2144 || !comm_signal_bind(worker->comsig, SIGINT)) { 2145 log_err("could not create signal handlers"); 2146 worker_delete(worker); 2147 return 0; 2148 } 2149 #endif /* LIBEVENT_SIGNAL_PROBLEM */ 2150 if(!daemon_remote_open_accept(worker->daemon->rc, 2151 worker->daemon->rc_ports, worker)) { 2152 worker_delete(worker); 2153 return 0; 2154 } 2155 #ifdef UB_ON_WINDOWS 2156 wsvc_setup_worker(worker); 2157 #endif /* UB_ON_WINDOWS */ 2158 } else { /* !do_sigs */ 2159 worker->comsig = NULL; 2160 } 2161 #ifdef USE_DNSTAP 2162 if(cfg->dnstap) { 2163 log_assert(worker->daemon->dtenv != NULL); 2164 memcpy(&worker->dtenv, worker->daemon->dtenv, sizeof(struct dt_env)); 2165 if(!dt_init(&worker->dtenv, worker->base)) 2166 fatal_exit("dt_init failed"); 2167 } 2168 #endif 2169 worker->front = listen_create(worker->base, ports, 2170 cfg->msg_buffer_size, (int)cfg->incoming_num_tcp, 2171 cfg->do_tcp_keepalive 2172 ? cfg->tcp_keepalive_timeout 2173 : cfg->tcp_idle_timeout, 2174 cfg->harden_large_queries, cfg->http_max_streams, 2175 cfg->http_endpoint, cfg->http_notls_downstream, 2176 worker->daemon->tcl, worker->daemon->listen_sslctx, 2177 dtenv, worker->daemon->doq_table, worker->env.rnd, 2178 cfg->ssl_service_key, cfg->ssl_service_pem, cfg, 2179 worker_handle_request, worker); 2180 if(!worker->front) { 2181 log_err("could not create listening sockets"); 2182 worker_delete(worker); 2183 return 0; 2184 } 2185 worker->back = outside_network_create(worker->base, 2186 cfg->msg_buffer_size, (size_t)cfg->outgoing_num_ports, 2187 cfg->out_ifs, cfg->num_out_ifs, cfg->do_ip4, cfg->do_ip6, 2188 cfg->do_tcp?cfg->outgoing_num_tcp:0, cfg->ip_dscp, 2189 worker->daemon->env->infra_cache, worker->rndstate, 2190 cfg->use_caps_bits_for_id, worker->ports, worker->numports, 2191 cfg->unwanted_threshold, cfg->outgoing_tcp_mss, 2192 &worker_alloc_cleanup, worker, 2193 cfg->do_udp || cfg->udp_upstream_without_downstream, 2194 worker->daemon->connect_sslctx, cfg->delay_close, 2195 cfg->tls_use_sni, dtenv, cfg->udp_connect, 2196 cfg->max_reuse_tcp_queries, cfg->tcp_reuse_timeout, 2197 cfg->tcp_auth_query_timeout); 2198 if(!worker->back) { 2199 log_err("could not create outgoing sockets"); 2200 worker_delete(worker); 2201 return 0; 2202 } 2203 iterator_set_ip46_support(&worker->daemon->mods, worker->daemon->env, 2204 worker->back); 2205 /* start listening to commands */ 2206 if(!tube_setup_bg_listen(worker->cmd, worker->base, 2207 &worker_handle_control_cmd, worker)) { 2208 log_err("could not create control compt."); 2209 worker_delete(worker); 2210 return 0; 2211 } 2212 worker->stat_timer = comm_timer_create(worker->base, 2213 worker_stat_timer_cb, worker); 2214 if(!worker->stat_timer) { 2215 log_err("could not create statistics timer"); 2216 } 2217 2218 /* we use the msg_buffer_size as a good estimate for what the 2219 * user wants for memory usage sizes */ 2220 worker->scratchpad = regional_create_custom(cfg->msg_buffer_size); 2221 if(!worker->scratchpad) { 2222 log_err("malloc failure"); 2223 worker_delete(worker); 2224 return 0; 2225 } 2226 2227 server_stats_init(&worker->stats, cfg); 2228 worker->alloc = worker->daemon->worker_allocs[worker->thread_num]; 2229 alloc_set_id_cleanup(worker->alloc, &worker_alloc_cleanup, worker); 2230 worker->env = *worker->daemon->env; 2231 comm_base_timept(worker->base, &worker->env.now, &worker->env.now_tv); 2232 worker->env.worker = worker; 2233 worker->env.worker_base = worker->base; 2234 worker->env.send_query = &worker_send_query; 2235 worker->env.alloc = worker->alloc; 2236 worker->env.outnet = worker->back; 2237 worker->env.rnd = worker->rndstate; 2238 /* If case prefetch is triggered, the corresponding mesh will clear 2239 * the scratchpad for the module env in the middle of request handling. 2240 * It would be prone to a use-after-free kind of bug, so we avoid 2241 * sharing it with worker's own scratchpad at the cost of having 2242 * one more pad per worker. */ 2243 worker->env.scratch = regional_create_custom(cfg->msg_buffer_size); 2244 if(!worker->env.scratch) { 2245 log_err("malloc failure"); 2246 worker_delete(worker); 2247 return 0; 2248 } 2249 worker->env.mesh = mesh_create(&worker->daemon->mods, &worker->env); 2250 if(!worker->env.mesh) { 2251 log_err("malloc failure"); 2252 worker_delete(worker); 2253 return 0; 2254 } 2255 /* Pass on daemon variables that we would need in the mesh area */ 2256 worker->env.mesh->use_response_ip = worker->daemon->use_response_ip; 2257 worker->env.mesh->use_rpz = worker->daemon->use_rpz; 2258 2259 worker->env.detach_subs = &mesh_detach_subs; 2260 worker->env.attach_sub = &mesh_attach_sub; 2261 worker->env.add_sub = &mesh_add_sub; 2262 worker->env.kill_sub = &mesh_state_delete; 2263 worker->env.detect_cycle = &mesh_detect_cycle; 2264 worker->env.scratch_buffer = sldns_buffer_new(cfg->msg_buffer_size); 2265 if(!worker->env.scratch_buffer) { 2266 log_err("malloc failure"); 2267 worker_delete(worker); 2268 return 0; 2269 } 2270 /* one probe timer per process -- if we have 5011 anchors */ 2271 if(autr_get_num_anchors(worker->env.anchors) > 0 2272 #ifndef THREADS_DISABLED 2273 && worker->thread_num == 0 2274 #endif 2275 ) { 2276 struct timeval tv; 2277 tv.tv_sec = 0; 2278 tv.tv_usec = 0; 2279 worker->env.probe_timer = comm_timer_create(worker->base, 2280 worker_probe_timer_cb, worker); 2281 if(!worker->env.probe_timer) { 2282 log_err("could not create 5011-probe timer"); 2283 } else { 2284 /* let timer fire, then it can reset itself */ 2285 comm_timer_set(worker->env.probe_timer, &tv); 2286 } 2287 } 2288 /* zone transfer tasks, setup once per process, if any */ 2289 if(worker->env.auth_zones 2290 #ifndef THREADS_DISABLED 2291 && worker->thread_num == 0 2292 #endif 2293 ) { 2294 auth_xfer_pickup_initial(worker->env.auth_zones, &worker->env); 2295 auth_zones_pickup_zonemd_verify(worker->env.auth_zones, 2296 &worker->env); 2297 } 2298 #ifdef USE_DNSTAP 2299 if(worker->daemon->cfg->dnstap 2300 #ifndef THREADS_DISABLED 2301 && worker->thread_num == 0 2302 #endif 2303 ) { 2304 if(!dt_io_thread_start(dtenv->dtio, comm_base_internal( 2305 worker->base), worker->daemon->num)) { 2306 log_err("could not start dnstap io thread"); 2307 worker_delete(worker); 2308 return 0; 2309 } 2310 } 2311 #endif /* USE_DNSTAP */ 2312 worker_mem_report(worker, NULL); 2313 /* if statistics enabled start timer */ 2314 if(worker->env.cfg->stat_interval > 0) { 2315 verbose(VERB_ALGO, "set statistics interval %d secs", 2316 worker->env.cfg->stat_interval); 2317 worker_restart_timer(worker); 2318 } 2319 pp_init(&sldns_write_uint16, &sldns_write_uint32); 2320 return 1; 2321 } 2322 2323 void 2324 worker_work(struct worker* worker) 2325 { 2326 comm_base_dispatch(worker->base); 2327 } 2328 2329 void 2330 worker_delete(struct worker* worker) 2331 { 2332 if(!worker) 2333 return; 2334 if(worker->env.mesh && verbosity >= VERB_OPS) { 2335 server_stats_log(&worker->stats, worker, worker->thread_num); 2336 mesh_stats(worker->env.mesh, "mesh has"); 2337 worker_mem_report(worker, NULL); 2338 } 2339 outside_network_quit_prepare(worker->back); 2340 mesh_delete(worker->env.mesh); 2341 sldns_buffer_free(worker->env.scratch_buffer); 2342 listen_delete(worker->front); 2343 outside_network_delete(worker->back); 2344 comm_signal_delete(worker->comsig); 2345 tube_delete(worker->cmd); 2346 comm_timer_delete(worker->stat_timer); 2347 comm_timer_delete(worker->env.probe_timer); 2348 free(worker->ports); 2349 if(worker->thread_num == 0) { 2350 #ifdef UB_ON_WINDOWS 2351 wsvc_desetup_worker(worker); 2352 #endif /* UB_ON_WINDOWS */ 2353 } 2354 #ifdef USE_DNSTAP 2355 if(worker->daemon->cfg->dnstap 2356 #ifndef THREADS_DISABLED 2357 && worker->thread_num == 0 2358 #endif 2359 ) { 2360 dt_io_thread_stop(worker->dtenv.dtio); 2361 } 2362 dt_deinit(&worker->dtenv); 2363 #endif /* USE_DNSTAP */ 2364 comm_base_delete(worker->base); 2365 ub_randfree(worker->rndstate); 2366 /* don't touch worker->alloc, as it's maintained in daemon */ 2367 regional_destroy(worker->env.scratch); 2368 regional_destroy(worker->scratchpad); 2369 free(worker); 2370 } 2371 2372 struct outbound_entry* 2373 worker_send_query(struct query_info* qinfo, uint16_t flags, int dnssec, 2374 int want_dnssec, int nocaps, int check_ratelimit, 2375 struct sockaddr_storage* addr, socklen_t addrlen, uint8_t* zone, 2376 size_t zonelen, int tcp_upstream, int ssl_upstream, char* tls_auth_name, 2377 struct module_qstate* q, int* was_ratelimited) 2378 { 2379 struct worker* worker = q->env->worker; 2380 struct outbound_entry* e = (struct outbound_entry*)regional_alloc( 2381 q->region, sizeof(*e)); 2382 if(!e) 2383 return NULL; 2384 e->qstate = q; 2385 e->qsent = outnet_serviced_query(worker->back, qinfo, flags, dnssec, 2386 want_dnssec, nocaps, check_ratelimit, tcp_upstream, 2387 ssl_upstream, tls_auth_name, addr, addrlen, zone, zonelen, q, 2388 worker_handle_service_reply, e, worker->back->udp_buff, q->env, 2389 was_ratelimited); 2390 if(!e->qsent) { 2391 return NULL; 2392 } 2393 return e; 2394 } 2395 2396 void 2397 worker_alloc_cleanup(void* arg) 2398 { 2399 struct worker* worker = (struct worker*)arg; 2400 slabhash_clear(&worker->env.rrset_cache->table); 2401 slabhash_clear(worker->env.msg_cache); 2402 } 2403 2404 void worker_stats_clear(struct worker* worker) 2405 { 2406 server_stats_init(&worker->stats, worker->env.cfg); 2407 mesh_stats_clear(worker->env.mesh); 2408 worker->back->unwanted_replies = 0; 2409 worker->back->num_tcp_outgoing = 0; 2410 worker->back->num_udp_outgoing = 0; 2411 } 2412 2413 void worker_start_accept(void* arg) 2414 { 2415 struct worker* worker = (struct worker*)arg; 2416 listen_start_accept(worker->front); 2417 if(worker->thread_num == 0) 2418 daemon_remote_start_accept(worker->daemon->rc); 2419 } 2420 2421 void worker_stop_accept(void* arg) 2422 { 2423 struct worker* worker = (struct worker*)arg; 2424 listen_stop_accept(worker->front); 2425 if(worker->thread_num == 0) 2426 daemon_remote_stop_accept(worker->daemon->rc); 2427 } 2428 2429 /* --- fake callbacks for fptr_wlist to work --- */ 2430 struct outbound_entry* libworker_send_query( 2431 struct query_info* ATTR_UNUSED(qinfo), 2432 uint16_t ATTR_UNUSED(flags), int ATTR_UNUSED(dnssec), 2433 int ATTR_UNUSED(want_dnssec), int ATTR_UNUSED(nocaps), 2434 int ATTR_UNUSED(check_ratelimit), 2435 struct sockaddr_storage* ATTR_UNUSED(addr), socklen_t ATTR_UNUSED(addrlen), 2436 uint8_t* ATTR_UNUSED(zone), size_t ATTR_UNUSED(zonelen), int ATTR_UNUSED(tcp_upstream), 2437 int ATTR_UNUSED(ssl_upstream), char* ATTR_UNUSED(tls_auth_name), 2438 struct module_qstate* ATTR_UNUSED(q), int* ATTR_UNUSED(was_ratelimited)) 2439 { 2440 log_assert(0); 2441 return 0; 2442 } 2443 2444 int libworker_handle_service_reply(struct comm_point* ATTR_UNUSED(c), 2445 void* ATTR_UNUSED(arg), int ATTR_UNUSED(error), 2446 struct comm_reply* ATTR_UNUSED(reply_info)) 2447 { 2448 log_assert(0); 2449 return 0; 2450 } 2451 2452 void libworker_handle_control_cmd(struct tube* ATTR_UNUSED(tube), 2453 uint8_t* ATTR_UNUSED(buffer), size_t ATTR_UNUSED(len), 2454 int ATTR_UNUSED(error), void* ATTR_UNUSED(arg)) 2455 { 2456 log_assert(0); 2457 } 2458 2459 void libworker_fg_done_cb(void* ATTR_UNUSED(arg), int ATTR_UNUSED(rcode), 2460 sldns_buffer* ATTR_UNUSED(buf), enum sec_status ATTR_UNUSED(s), 2461 char* ATTR_UNUSED(why_bogus), int ATTR_UNUSED(was_ratelimited)) 2462 { 2463 log_assert(0); 2464 } 2465 2466 void libworker_bg_done_cb(void* ATTR_UNUSED(arg), int ATTR_UNUSED(rcode), 2467 sldns_buffer* ATTR_UNUSED(buf), enum sec_status ATTR_UNUSED(s), 2468 char* ATTR_UNUSED(why_bogus), int ATTR_UNUSED(was_ratelimited)) 2469 { 2470 log_assert(0); 2471 } 2472 2473 void libworker_event_done_cb(void* ATTR_UNUSED(arg), int ATTR_UNUSED(rcode), 2474 sldns_buffer* ATTR_UNUSED(buf), enum sec_status ATTR_UNUSED(s), 2475 char* ATTR_UNUSED(why_bogus), int ATTR_UNUSED(was_ratelimited)) 2476 { 2477 log_assert(0); 2478 } 2479 2480 int context_query_cmp(const void* ATTR_UNUSED(a), const void* ATTR_UNUSED(b)) 2481 { 2482 log_assert(0); 2483 return 0; 2484 } 2485 2486 int order_lock_cmp(const void* ATTR_UNUSED(e1), const void* ATTR_UNUSED(e2)) 2487 { 2488 log_assert(0); 2489 return 0; 2490 } 2491 2492 int codeline_cmp(const void* ATTR_UNUSED(a), const void* ATTR_UNUSED(b)) 2493 { 2494 log_assert(0); 2495 return 0; 2496 } 2497 2498 #ifdef USE_DNSTAP 2499 void dtio_tap_callback(int ATTR_UNUSED(fd), short ATTR_UNUSED(ev), 2500 void* ATTR_UNUSED(arg)) 2501 { 2502 log_assert(0); 2503 } 2504 #endif 2505 2506 #ifdef USE_DNSTAP 2507 void dtio_mainfdcallback(int ATTR_UNUSED(fd), short ATTR_UNUSED(ev), 2508 void* ATTR_UNUSED(arg)) 2509 { 2510 log_assert(0); 2511 } 2512 #endif 2513 2514 #ifdef HAVE_NGTCP2 2515 void doq_client_event_cb(int ATTR_UNUSED(fd), short ATTR_UNUSED(ev), 2516 void* ATTR_UNUSED(arg)) 2517 { 2518 log_assert(0); 2519 } 2520 #endif 2521 2522 #ifdef HAVE_NGTCP2 2523 void doq_client_timer_cb(int ATTR_UNUSED(fd), short ATTR_UNUSED(ev), 2524 void* ATTR_UNUSED(arg)) 2525 { 2526 log_assert(0); 2527 } 2528 #endif 2529