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/tube.h" 70 #include "util/edns.h" 71 #include "iterator/iter_fwd.h" 72 #include "iterator/iter_hints.h" 73 #include "validator/autotrust.h" 74 #include "validator/val_anchor.h" 75 #include "respip/respip.h" 76 #include "libunbound/context.h" 77 #include "libunbound/libworker.h" 78 #include "sldns/sbuffer.h" 79 #include "sldns/wire2str.h" 80 #include "util/shm_side/shm_main.h" 81 #include "dnscrypt/dnscrypt.h" 82 #include "dnstap/dtstream.h" 83 84 #ifdef HAVE_SYS_TYPES_H 85 # include <sys/types.h> 86 #endif 87 #ifdef HAVE_NETDB_H 88 #include <netdb.h> 89 #endif 90 #include <signal.h> 91 #ifdef UB_ON_WINDOWS 92 #include "winrc/win_svc.h" 93 #endif 94 95 /** Size of an UDP datagram */ 96 #define NORMAL_UDP_SIZE 512 /* bytes */ 97 /** ratelimit for error responses */ 98 #define ERROR_RATELIMIT 100 /* qps */ 99 100 /** 101 * seconds to add to prefetch leeway. This is a TTL that expires old rrsets 102 * earlier than they should in order to put the new update into the cache. 103 * This additional value is to make sure that if not all TTLs are equal in 104 * the message to be updated(and replaced), that rrsets with up to this much 105 * extra TTL are also replaced. This means that the resulting new message 106 * will have (most likely) this TTL at least, avoiding very small 'split 107 * second' TTLs due to operators choosing relative primes for TTLs (or so). 108 * Also has to be at least one to break ties (and overwrite cached entry). 109 */ 110 #define PREFETCH_EXPIRY_ADD 60 111 112 /** Report on memory usage by this thread and global */ 113 static void 114 worker_mem_report(struct worker* ATTR_UNUSED(worker), 115 struct serviced_query* ATTR_UNUSED(cur_serv)) 116 { 117 #ifdef UNBOUND_ALLOC_STATS 118 /* measure memory leakage */ 119 extern size_t unbound_mem_alloc, unbound_mem_freed; 120 /* debug func in validator module */ 121 size_t total, front, back, mesh, msg, rrset, infra, ac, superac; 122 size_t me, iter, val, anch; 123 int i; 124 #ifdef CLIENT_SUBNET 125 size_t subnet = 0; 126 #endif /* CLIENT_SUBNET */ 127 if(verbosity < VERB_ALGO) 128 return; 129 front = listen_get_mem(worker->front); 130 back = outnet_get_mem(worker->back); 131 msg = slabhash_get_mem(worker->env.msg_cache); 132 rrset = slabhash_get_mem(&worker->env.rrset_cache->table); 133 infra = infra_get_mem(worker->env.infra_cache); 134 mesh = mesh_get_mem(worker->env.mesh); 135 ac = alloc_get_mem(&worker->alloc); 136 superac = alloc_get_mem(&worker->daemon->superalloc); 137 anch = anchors_get_mem(worker->env.anchors); 138 iter = 0; 139 val = 0; 140 for(i=0; i<worker->env.mesh->mods.num; i++) { 141 fptr_ok(fptr_whitelist_mod_get_mem(worker->env.mesh-> 142 mods.mod[i]->get_mem)); 143 if(strcmp(worker->env.mesh->mods.mod[i]->name, "validator")==0) 144 val += (*worker->env.mesh->mods.mod[i]->get_mem) 145 (&worker->env, i); 146 #ifdef CLIENT_SUBNET 147 else if(strcmp(worker->env.mesh->mods.mod[i]->name, 148 "subnet")==0) 149 subnet += (*worker->env.mesh->mods.mod[i]->get_mem) 150 (&worker->env, i); 151 #endif /* CLIENT_SUBNET */ 152 else iter += (*worker->env.mesh->mods.mod[i]->get_mem) 153 (&worker->env, i); 154 } 155 me = sizeof(*worker) + sizeof(*worker->base) + sizeof(*worker->comsig) 156 + comm_point_get_mem(worker->cmd_com) 157 + sizeof(worker->rndstate) 158 + regional_get_mem(worker->scratchpad) 159 + sizeof(*worker->env.scratch_buffer) 160 + sldns_buffer_capacity(worker->env.scratch_buffer) 161 + forwards_get_mem(worker->env.fwds) 162 + hints_get_mem(worker->env.hints); 163 if(worker->thread_num == 0) 164 me += acl_list_get_mem(worker->daemon->acl); 165 if(cur_serv) { 166 me += serviced_get_mem(cur_serv); 167 } 168 total = front+back+mesh+msg+rrset+infra+iter+val+ac+superac+me; 169 #ifdef CLIENT_SUBNET 170 total += subnet; 171 log_info("Memory conditions: %u front=%u back=%u mesh=%u msg=%u " 172 "rrset=%u infra=%u iter=%u val=%u subnet=%u anchors=%u " 173 "alloccache=%u globalalloccache=%u me=%u", 174 (unsigned)total, (unsigned)front, (unsigned)back, 175 (unsigned)mesh, (unsigned)msg, (unsigned)rrset, (unsigned)infra, 176 (unsigned)iter, (unsigned)val, 177 (unsigned)subnet, (unsigned)anch, (unsigned)ac, 178 (unsigned)superac, (unsigned)me); 179 #else /* no CLIENT_SUBNET */ 180 log_info("Memory conditions: %u front=%u back=%u mesh=%u msg=%u " 181 "rrset=%u infra=%u iter=%u val=%u anchors=%u " 182 "alloccache=%u globalalloccache=%u me=%u", 183 (unsigned)total, (unsigned)front, (unsigned)back, 184 (unsigned)mesh, (unsigned)msg, (unsigned)rrset, 185 (unsigned)infra, (unsigned)iter, (unsigned)val, (unsigned)anch, 186 (unsigned)ac, (unsigned)superac, (unsigned)me); 187 #endif /* CLIENT_SUBNET */ 188 log_info("Total heap memory estimate: %u total-alloc: %u " 189 "total-free: %u", (unsigned)total, 190 (unsigned)unbound_mem_alloc, (unsigned)unbound_mem_freed); 191 #else /* no UNBOUND_ALLOC_STATS */ 192 size_t val = 0; 193 #ifdef CLIENT_SUBNET 194 size_t subnet = 0; 195 #endif /* CLIENT_SUBNET */ 196 int i; 197 if(verbosity < VERB_QUERY) 198 return; 199 for(i=0; i<worker->env.mesh->mods.num; i++) { 200 fptr_ok(fptr_whitelist_mod_get_mem(worker->env.mesh-> 201 mods.mod[i]->get_mem)); 202 if(strcmp(worker->env.mesh->mods.mod[i]->name, "validator")==0) 203 val += (*worker->env.mesh->mods.mod[i]->get_mem) 204 (&worker->env, i); 205 #ifdef CLIENT_SUBNET 206 else if(strcmp(worker->env.mesh->mods.mod[i]->name, 207 "subnet")==0) 208 subnet += (*worker->env.mesh->mods.mod[i]->get_mem) 209 (&worker->env, i); 210 #endif /* CLIENT_SUBNET */ 211 } 212 #ifdef CLIENT_SUBNET 213 verbose(VERB_QUERY, "cache memory msg=%u rrset=%u infra=%u val=%u " 214 "subnet=%u", 215 (unsigned)slabhash_get_mem(worker->env.msg_cache), 216 (unsigned)slabhash_get_mem(&worker->env.rrset_cache->table), 217 (unsigned)infra_get_mem(worker->env.infra_cache), 218 (unsigned)val, (unsigned)subnet); 219 #else /* no CLIENT_SUBNET */ 220 verbose(VERB_QUERY, "cache memory msg=%u rrset=%u infra=%u val=%u", 221 (unsigned)slabhash_get_mem(worker->env.msg_cache), 222 (unsigned)slabhash_get_mem(&worker->env.rrset_cache->table), 223 (unsigned)infra_get_mem(worker->env.infra_cache), 224 (unsigned)val); 225 #endif /* CLIENT_SUBNET */ 226 #endif /* UNBOUND_ALLOC_STATS */ 227 } 228 229 void 230 worker_send_cmd(struct worker* worker, enum worker_commands cmd) 231 { 232 uint32_t c = (uint32_t)htonl(cmd); 233 if(!tube_write_msg(worker->cmd, (uint8_t*)&c, sizeof(c), 0)) { 234 log_err("worker send cmd %d failed", (int)cmd); 235 } 236 } 237 238 int 239 worker_handle_reply(struct comm_point* c, void* arg, int error, 240 struct comm_reply* reply_info) 241 { 242 struct module_qstate* q = (struct module_qstate*)arg; 243 struct worker* worker = q->env->worker; 244 struct outbound_entry e; 245 e.qstate = q; 246 e.qsent = NULL; 247 248 if(error != 0) { 249 mesh_report_reply(worker->env.mesh, &e, reply_info, error); 250 worker_mem_report(worker, NULL); 251 return 0; 252 } 253 /* sanity check. */ 254 if(!LDNS_QR_WIRE(sldns_buffer_begin(c->buffer)) 255 || LDNS_OPCODE_WIRE(sldns_buffer_begin(c->buffer)) != 256 LDNS_PACKET_QUERY 257 || LDNS_QDCOUNT(sldns_buffer_begin(c->buffer)) > 1) { 258 /* error becomes timeout for the module as if this reply 259 * never arrived. */ 260 mesh_report_reply(worker->env.mesh, &e, reply_info, 261 NETEVENT_TIMEOUT); 262 worker_mem_report(worker, NULL); 263 return 0; 264 } 265 mesh_report_reply(worker->env.mesh, &e, reply_info, NETEVENT_NOERROR); 266 worker_mem_report(worker, NULL); 267 return 0; 268 } 269 270 int 271 worker_handle_service_reply(struct comm_point* c, void* arg, int error, 272 struct comm_reply* reply_info) 273 { 274 struct outbound_entry* e = (struct outbound_entry*)arg; 275 struct worker* worker = e->qstate->env->worker; 276 struct serviced_query *sq = e->qsent; 277 278 verbose(VERB_ALGO, "worker svcd callback for qstate %p", e->qstate); 279 if(error != 0) { 280 mesh_report_reply(worker->env.mesh, e, reply_info, error); 281 worker_mem_report(worker, sq); 282 return 0; 283 } 284 /* sanity check. */ 285 if(!LDNS_QR_WIRE(sldns_buffer_begin(c->buffer)) 286 || LDNS_OPCODE_WIRE(sldns_buffer_begin(c->buffer)) != 287 LDNS_PACKET_QUERY 288 || LDNS_QDCOUNT(sldns_buffer_begin(c->buffer)) > 1) { 289 /* error becomes timeout for the module as if this reply 290 * never arrived. */ 291 verbose(VERB_ALGO, "worker: bad reply handled as timeout"); 292 mesh_report_reply(worker->env.mesh, e, reply_info, 293 NETEVENT_TIMEOUT); 294 worker_mem_report(worker, sq); 295 return 0; 296 } 297 mesh_report_reply(worker->env.mesh, e, reply_info, NETEVENT_NOERROR); 298 worker_mem_report(worker, sq); 299 return 0; 300 } 301 302 /** ratelimit error replies 303 * @param worker: the worker struct with ratelimit counter 304 * @param err: error code that would be wanted. 305 * @return value of err if okay, or -1 if it should be discarded instead. 306 */ 307 static int 308 worker_err_ratelimit(struct worker* worker, int err) 309 { 310 if(worker->err_limit_time == *worker->env.now) { 311 /* see if limit is exceeded for this second */ 312 if(worker->err_limit_count++ > ERROR_RATELIMIT) 313 return -1; 314 } else { 315 /* new second, new limits */ 316 worker->err_limit_time = *worker->env.now; 317 worker->err_limit_count = 1; 318 } 319 return err; 320 } 321 322 /** check request sanity. 323 * @param pkt: the wire packet to examine for sanity. 324 * @param worker: parameters for checking. 325 * @return error code, 0 OK, or -1 discard. 326 */ 327 static int 328 worker_check_request(sldns_buffer* pkt, struct worker* worker) 329 { 330 if(sldns_buffer_limit(pkt) < LDNS_HEADER_SIZE) { 331 verbose(VERB_QUERY, "request too short, discarded"); 332 return -1; 333 } 334 if(sldns_buffer_limit(pkt) > NORMAL_UDP_SIZE && 335 worker->daemon->cfg->harden_large_queries) { 336 verbose(VERB_QUERY, "request too large, discarded"); 337 return -1; 338 } 339 if(LDNS_QR_WIRE(sldns_buffer_begin(pkt))) { 340 verbose(VERB_QUERY, "request has QR bit on, discarded"); 341 return -1; 342 } 343 if(LDNS_TC_WIRE(sldns_buffer_begin(pkt))) { 344 LDNS_TC_CLR(sldns_buffer_begin(pkt)); 345 verbose(VERB_QUERY, "request bad, has TC bit on"); 346 return worker_err_ratelimit(worker, LDNS_RCODE_FORMERR); 347 } 348 if(LDNS_OPCODE_WIRE(sldns_buffer_begin(pkt)) != LDNS_PACKET_QUERY && 349 LDNS_OPCODE_WIRE(sldns_buffer_begin(pkt)) != LDNS_PACKET_NOTIFY) { 350 verbose(VERB_QUERY, "request unknown opcode %d", 351 LDNS_OPCODE_WIRE(sldns_buffer_begin(pkt))); 352 return worker_err_ratelimit(worker, LDNS_RCODE_NOTIMPL); 353 } 354 if(LDNS_QDCOUNT(sldns_buffer_begin(pkt)) != 1) { 355 verbose(VERB_QUERY, "request wrong nr qd=%d", 356 LDNS_QDCOUNT(sldns_buffer_begin(pkt))); 357 return worker_err_ratelimit(worker, LDNS_RCODE_FORMERR); 358 } 359 if(LDNS_ANCOUNT(sldns_buffer_begin(pkt)) != 0 && 360 (LDNS_ANCOUNT(sldns_buffer_begin(pkt)) != 1 || 361 LDNS_OPCODE_WIRE(sldns_buffer_begin(pkt)) != LDNS_PACKET_NOTIFY)) { 362 verbose(VERB_QUERY, "request wrong nr an=%d", 363 LDNS_ANCOUNT(sldns_buffer_begin(pkt))); 364 return worker_err_ratelimit(worker, LDNS_RCODE_FORMERR); 365 } 366 if(LDNS_NSCOUNT(sldns_buffer_begin(pkt)) != 0) { 367 verbose(VERB_QUERY, "request wrong nr ns=%d", 368 LDNS_NSCOUNT(sldns_buffer_begin(pkt))); 369 return worker_err_ratelimit(worker, LDNS_RCODE_FORMERR); 370 } 371 if(LDNS_ARCOUNT(sldns_buffer_begin(pkt)) > 1) { 372 verbose(VERB_QUERY, "request wrong nr ar=%d", 373 LDNS_ARCOUNT(sldns_buffer_begin(pkt))); 374 return worker_err_ratelimit(worker, LDNS_RCODE_FORMERR); 375 } 376 return 0; 377 } 378 379 void 380 worker_handle_control_cmd(struct tube* ATTR_UNUSED(tube), uint8_t* msg, 381 size_t len, int error, void* arg) 382 { 383 struct worker* worker = (struct worker*)arg; 384 enum worker_commands cmd; 385 if(error != NETEVENT_NOERROR) { 386 free(msg); 387 if(error == NETEVENT_CLOSED) 388 comm_base_exit(worker->base); 389 else log_info("control event: %d", error); 390 return; 391 } 392 if(len != sizeof(uint32_t)) { 393 fatal_exit("bad control msg length %d", (int)len); 394 } 395 cmd = sldns_read_uint32(msg); 396 free(msg); 397 switch(cmd) { 398 case worker_cmd_quit: 399 verbose(VERB_ALGO, "got control cmd quit"); 400 comm_base_exit(worker->base); 401 break; 402 case worker_cmd_stats: 403 verbose(VERB_ALGO, "got control cmd stats"); 404 server_stats_reply(worker, 1); 405 break; 406 case worker_cmd_stats_noreset: 407 verbose(VERB_ALGO, "got control cmd stats_noreset"); 408 server_stats_reply(worker, 0); 409 break; 410 case worker_cmd_remote: 411 verbose(VERB_ALGO, "got control cmd remote"); 412 daemon_remote_exec(worker); 413 break; 414 default: 415 log_err("bad command %d", (int)cmd); 416 break; 417 } 418 } 419 420 /** check if a delegation is secure */ 421 static enum sec_status 422 check_delegation_secure(struct reply_info *rep) 423 { 424 /* return smallest security status */ 425 size_t i; 426 enum sec_status sec = sec_status_secure; 427 enum sec_status s; 428 size_t num = rep->an_numrrsets + rep->ns_numrrsets; 429 /* check if answer and authority are OK */ 430 for(i=0; i<num; i++) { 431 s = ((struct packed_rrset_data*)rep->rrsets[i]->entry.data) 432 ->security; 433 if(s < sec) 434 sec = s; 435 } 436 /* in additional, only unchecked triggers revalidation */ 437 for(i=num; i<rep->rrset_count; i++) { 438 s = ((struct packed_rrset_data*)rep->rrsets[i]->entry.data) 439 ->security; 440 if(s == sec_status_unchecked) 441 return s; 442 } 443 return sec; 444 } 445 446 /** remove nonsecure from a delegation referral additional section */ 447 static void 448 deleg_remove_nonsecure_additional(struct reply_info* rep) 449 { 450 /* we can simply edit it, since we are working in the scratch region */ 451 size_t i; 452 enum sec_status s; 453 454 for(i = rep->an_numrrsets+rep->ns_numrrsets; i<rep->rrset_count; i++) { 455 s = ((struct packed_rrset_data*)rep->rrsets[i]->entry.data) 456 ->security; 457 if(s != sec_status_secure) { 458 memmove(rep->rrsets+i, rep->rrsets+i+1, 459 sizeof(struct ub_packed_rrset_key*)* 460 (rep->rrset_count - i - 1)); 461 rep->ar_numrrsets--; 462 rep->rrset_count--; 463 i--; 464 } 465 } 466 } 467 468 /** answer nonrecursive query from the cache */ 469 static int 470 answer_norec_from_cache(struct worker* worker, struct query_info* qinfo, 471 uint16_t id, uint16_t flags, struct comm_reply* repinfo, 472 struct edns_data* edns) 473 { 474 /* for a nonrecursive query return either: 475 * o an error (servfail; we try to avoid this) 476 * o a delegation (closest we have; this routine tries that) 477 * o the answer (checked by answer_from_cache) 478 * 479 * So, grab a delegation from the rrset cache. 480 * Then check if it needs validation, if so, this routine fails, 481 * so that iterator can prime and validator can verify rrsets. 482 */ 483 struct edns_data edns_bak; 484 uint16_t udpsize = edns->udp_size; 485 int secure = 0; 486 time_t timenow = *worker->env.now; 487 int must_validate = (!(flags&BIT_CD) || worker->env.cfg->ignore_cd) 488 && worker->env.need_to_validate; 489 struct dns_msg *msg = NULL; 490 struct delegpt *dp; 491 492 dp = dns_cache_find_delegation(&worker->env, qinfo->qname, 493 qinfo->qname_len, qinfo->qtype, qinfo->qclass, 494 worker->scratchpad, &msg, timenow); 495 if(!dp) { /* no delegation, need to reprime */ 496 return 0; 497 } 498 /* In case we have a local alias, copy it into the delegation message. 499 * Shallow copy should be fine, as we'll be done with msg in this 500 * function. */ 501 msg->qinfo.local_alias = qinfo->local_alias; 502 if(must_validate) { 503 switch(check_delegation_secure(msg->rep)) { 504 case sec_status_unchecked: 505 /* some rrsets have not been verified yet, go and 506 * let validator do that */ 507 return 0; 508 case sec_status_bogus: 509 case sec_status_secure_sentinel_fail: 510 /* some rrsets are bogus, reply servfail */ 511 edns->edns_version = EDNS_ADVERTISED_VERSION; 512 edns->udp_size = EDNS_ADVERTISED_SIZE; 513 edns->ext_rcode = 0; 514 edns->bits &= EDNS_DO; 515 if(!inplace_cb_reply_servfail_call(&worker->env, qinfo, NULL, 516 msg->rep, LDNS_RCODE_SERVFAIL, edns, repinfo, worker->scratchpad)) 517 return 0; 518 error_encode(repinfo->c->buffer, LDNS_RCODE_SERVFAIL, 519 &msg->qinfo, id, flags, edns); 520 if(worker->stats.extended) { 521 worker->stats.ans_bogus++; 522 worker->stats.ans_rcode[LDNS_RCODE_SERVFAIL]++; 523 } 524 return 1; 525 case sec_status_secure: 526 /* all rrsets are secure */ 527 /* remove non-secure rrsets from the add. section*/ 528 if(worker->env.cfg->val_clean_additional) 529 deleg_remove_nonsecure_additional(msg->rep); 530 secure = 1; 531 break; 532 case sec_status_indeterminate: 533 case sec_status_insecure: 534 default: 535 /* not secure */ 536 secure = 0; 537 break; 538 } 539 } 540 /* return this delegation from the cache */ 541 edns_bak = *edns; 542 edns->edns_version = EDNS_ADVERTISED_VERSION; 543 edns->udp_size = EDNS_ADVERTISED_SIZE; 544 edns->ext_rcode = 0; 545 edns->bits &= EDNS_DO; 546 if(!inplace_cb_reply_cache_call(&worker->env, qinfo, NULL, msg->rep, 547 (int)(flags&LDNS_RCODE_MASK), edns, repinfo, worker->scratchpad)) 548 return 0; 549 msg->rep->flags |= BIT_QR|BIT_RA; 550 if(!apply_edns_options(edns, &edns_bak, worker->env.cfg, 551 repinfo->c, worker->scratchpad) || 552 !reply_info_answer_encode(&msg->qinfo, msg->rep, id, flags, 553 repinfo->c->buffer, 0, 1, worker->scratchpad, 554 udpsize, edns, (int)(edns->bits & EDNS_DO), secure)) { 555 if(!inplace_cb_reply_servfail_call(&worker->env, qinfo, NULL, NULL, 556 LDNS_RCODE_SERVFAIL, edns, repinfo, worker->scratchpad)) 557 edns->opt_list = NULL; 558 error_encode(repinfo->c->buffer, LDNS_RCODE_SERVFAIL, 559 &msg->qinfo, id, flags, edns); 560 } 561 if(worker->stats.extended) { 562 if(secure) worker->stats.ans_secure++; 563 server_stats_insrcode(&worker->stats, repinfo->c->buffer); 564 } 565 return 1; 566 } 567 568 /** Apply, if applicable, a response IP action to a cached answer. 569 * If the answer is rewritten as a result of an action, '*encode_repp' will 570 * point to the reply info containing the modified answer. '*encode_repp' will 571 * be intact otherwise. 572 * It returns 1 on success, 0 otherwise. */ 573 static int 574 apply_respip_action(struct worker* worker, const struct query_info* qinfo, 575 struct respip_client_info* cinfo, struct reply_info* rep, 576 struct comm_reply* repinfo, struct ub_packed_rrset_key** alias_rrset, 577 struct reply_info** encode_repp, struct auth_zones* az) 578 { 579 struct respip_action_info actinfo = {0}; 580 actinfo.action = respip_none; 581 582 if(qinfo->qtype != LDNS_RR_TYPE_A && 583 qinfo->qtype != LDNS_RR_TYPE_AAAA && 584 qinfo->qtype != LDNS_RR_TYPE_ANY) 585 return 1; 586 587 if(!respip_rewrite_reply(qinfo, cinfo, rep, encode_repp, &actinfo, 588 alias_rrset, 0, worker->scratchpad, az)) 589 return 0; 590 591 /* xxx_deny actions mean dropping the reply, unless the original reply 592 * was redirected to response-ip data. */ 593 if((actinfo.action == respip_deny || 594 actinfo.action == respip_inform_deny) && 595 *encode_repp == rep) 596 *encode_repp = NULL; 597 598 /* If address info is returned, it means the action should be an 599 * 'inform' variant and the information should be logged. */ 600 if(actinfo.addrinfo) { 601 respip_inform_print(&actinfo, qinfo->qname, 602 qinfo->qtype, qinfo->qclass, qinfo->local_alias, 603 repinfo); 604 605 if(worker->stats.extended && actinfo.rpz_used) { 606 if(actinfo.rpz_disabled) 607 worker->stats.rpz_action[RPZ_DISABLED_ACTION]++; 608 if(actinfo.rpz_cname_override) 609 worker->stats.rpz_action[RPZ_CNAME_OVERRIDE_ACTION]++; 610 else 611 worker->stats.rpz_action[ 612 respip_action_to_rpz_action(actinfo.action)]++; 613 } 614 } 615 616 return 1; 617 } 618 619 /** answer query from the cache. 620 * Normally, the answer message will be built in repinfo->c->buffer; if the 621 * answer is supposed to be suppressed or the answer is supposed to be an 622 * incomplete CNAME chain, the buffer is explicitly cleared to signal the 623 * caller as such. In the latter case *partial_rep will point to the incomplete 624 * reply, and this function is (possibly) supposed to be called again with that 625 * *partial_rep value to complete the chain. In addition, if the query should 626 * be completely dropped, '*need_drop' will be set to 1. */ 627 static int 628 answer_from_cache(struct worker* worker, struct query_info* qinfo, 629 struct respip_client_info* cinfo, int* need_drop, int* is_expired_answer, 630 int* is_secure_answer, struct ub_packed_rrset_key** alias_rrset, 631 struct reply_info** partial_repp, 632 struct reply_info* rep, uint16_t id, uint16_t flags, 633 struct comm_reply* repinfo, struct edns_data* edns) 634 { 635 struct edns_data edns_bak; 636 time_t timenow = *worker->env.now; 637 uint16_t udpsize = edns->udp_size; 638 struct reply_info* encode_rep = rep; 639 struct reply_info* partial_rep = *partial_repp; 640 int must_validate = (!(flags&BIT_CD) || worker->env.cfg->ignore_cd) 641 && worker->env.need_to_validate; 642 *partial_repp = NULL; /* avoid accidental further pass */ 643 644 /* Check TTL */ 645 if(rep->ttl < timenow) { 646 /* Check if we need to serve expired now */ 647 if(worker->env.cfg->serve_expired && 648 !worker->env.cfg->serve_expired_client_timeout) { 649 if(worker->env.cfg->serve_expired_ttl && 650 rep->serve_expired_ttl < timenow) 651 return 0; 652 if(!rrset_array_lock(rep->ref, rep->rrset_count, 0)) 653 return 0; 654 *is_expired_answer = 1; 655 } else { 656 /* the rrsets may have been updated in the meantime. 657 * we will refetch the message format from the 658 * authoritative server 659 */ 660 return 0; 661 } 662 } else { 663 if(!rrset_array_lock(rep->ref, rep->rrset_count, timenow)) 664 return 0; 665 } 666 /* locked and ids and ttls are OK. */ 667 668 /* check CNAME chain (if any) */ 669 if(rep->an_numrrsets > 0 && (rep->rrsets[0]->rk.type == 670 htons(LDNS_RR_TYPE_CNAME) || rep->rrsets[0]->rk.type == 671 htons(LDNS_RR_TYPE_DNAME))) { 672 if(!reply_check_cname_chain(qinfo, rep)) { 673 /* cname chain invalid, redo iterator steps */ 674 verbose(VERB_ALGO, "Cache reply: cname chain broken"); 675 goto bail_out; 676 } 677 } 678 /* check security status of the cached answer */ 679 if(must_validate && (rep->security == sec_status_bogus || 680 rep->security == sec_status_secure_sentinel_fail)) { 681 /* BAD cached */ 682 edns->edns_version = EDNS_ADVERTISED_VERSION; 683 edns->udp_size = EDNS_ADVERTISED_SIZE; 684 edns->ext_rcode = 0; 685 edns->bits &= EDNS_DO; 686 if(!inplace_cb_reply_servfail_call(&worker->env, qinfo, NULL, rep, 687 LDNS_RCODE_SERVFAIL, edns, repinfo, worker->scratchpad)) 688 goto bail_out; 689 error_encode(repinfo->c->buffer, LDNS_RCODE_SERVFAIL, 690 qinfo, id, flags, edns); 691 rrset_array_unlock_touch(worker->env.rrset_cache, 692 worker->scratchpad, rep->ref, rep->rrset_count); 693 if(worker->stats.extended) { 694 worker->stats.ans_bogus ++; 695 worker->stats.ans_rcode[LDNS_RCODE_SERVFAIL] ++; 696 } 697 return 1; 698 } else if(rep->security == sec_status_unchecked && must_validate) { 699 verbose(VERB_ALGO, "Cache reply: unchecked entry needs " 700 "validation"); 701 goto bail_out; /* need to validate cache entry first */ 702 } else if(rep->security == sec_status_secure) { 703 if(reply_all_rrsets_secure(rep)) { 704 *is_secure_answer = 1; 705 } else { 706 if(must_validate) { 707 verbose(VERB_ALGO, "Cache reply: secure entry" 708 " changed status"); 709 goto bail_out; /* rrset changed, re-verify */ 710 } 711 *is_secure_answer = 0; 712 } 713 } else *is_secure_answer = 0; 714 715 edns_bak = *edns; 716 edns->edns_version = EDNS_ADVERTISED_VERSION; 717 edns->udp_size = EDNS_ADVERTISED_SIZE; 718 edns->ext_rcode = 0; 719 edns->bits &= EDNS_DO; 720 if(!inplace_cb_reply_cache_call(&worker->env, qinfo, NULL, rep, 721 (int)(flags&LDNS_RCODE_MASK), edns, repinfo, worker->scratchpad)) 722 goto bail_out; 723 *alias_rrset = NULL; /* avoid confusion if caller set it to non-NULL */ 724 if((worker->daemon->use_response_ip || worker->daemon->use_rpz) && 725 !partial_rep && !apply_respip_action(worker, qinfo, cinfo, rep, 726 repinfo, alias_rrset, 727 &encode_rep, worker->env.auth_zones)) { 728 goto bail_out; 729 } else if(partial_rep && 730 !respip_merge_cname(partial_rep, qinfo, rep, cinfo, 731 must_validate, &encode_rep, worker->scratchpad, 732 worker->env.auth_zones)) { 733 goto bail_out; 734 } 735 if(encode_rep != rep) { 736 /* if rewritten, it can't be considered "secure" */ 737 *is_secure_answer = 0; 738 } 739 if(!encode_rep || *alias_rrset) { 740 if(!encode_rep) 741 *need_drop = 1; 742 else { 743 /* If a partial CNAME chain is found, we first need to 744 * make a copy of the reply in the scratchpad so we 745 * can release the locks and lookup the cache again. */ 746 *partial_repp = reply_info_copy(encode_rep, NULL, 747 worker->scratchpad); 748 if(!*partial_repp) 749 goto bail_out; 750 } 751 } else if(!apply_edns_options(edns, &edns_bak, worker->env.cfg, 752 repinfo->c, worker->scratchpad) || 753 !reply_info_answer_encode(qinfo, encode_rep, id, flags, 754 repinfo->c->buffer, timenow, 1, worker->scratchpad, 755 udpsize, edns, (int)(edns->bits & EDNS_DO), *is_secure_answer)) { 756 if(!inplace_cb_reply_servfail_call(&worker->env, qinfo, NULL, NULL, 757 LDNS_RCODE_SERVFAIL, edns, repinfo, worker->scratchpad)) 758 edns->opt_list = NULL; 759 error_encode(repinfo->c->buffer, LDNS_RCODE_SERVFAIL, 760 qinfo, id, flags, edns); 761 } 762 /* cannot send the reply right now, because blocking network syscall 763 * is bad while holding locks. */ 764 rrset_array_unlock_touch(worker->env.rrset_cache, worker->scratchpad, 765 rep->ref, rep->rrset_count); 766 /* go and return this buffer to the client */ 767 return 1; 768 769 bail_out: 770 rrset_array_unlock_touch(worker->env.rrset_cache, 771 worker->scratchpad, rep->ref, rep->rrset_count); 772 return 0; 773 } 774 775 /** Reply to client and perform prefetch to keep cache up to date. */ 776 static void 777 reply_and_prefetch(struct worker* worker, struct query_info* qinfo, 778 uint16_t flags, struct comm_reply* repinfo, time_t leeway, int noreply) 779 { 780 /* first send answer to client to keep its latency 781 * as small as a cachereply */ 782 if(!noreply) { 783 if(repinfo->c->tcp_req_info) { 784 sldns_buffer_copy( 785 repinfo->c->tcp_req_info->spool_buffer, 786 repinfo->c->buffer); 787 } 788 comm_point_send_reply(repinfo); 789 } 790 server_stats_prefetch(&worker->stats, worker); 791 792 /* create the prefetch in the mesh as a normal lookup without 793 * client addrs waiting, which has the cache blacklisted (to bypass 794 * the cache and go to the network for the data). */ 795 /* this (potentially) runs the mesh for the new query */ 796 mesh_new_prefetch(worker->env.mesh, qinfo, flags, leeway + 797 PREFETCH_EXPIRY_ADD); 798 } 799 800 /** 801 * Fill CH class answer into buffer. Keeps query. 802 * @param pkt: buffer 803 * @param str: string to put into text record (<255). 804 * array of strings, every string becomes a text record. 805 * @param num: number of strings in array. 806 * @param edns: edns reply information. 807 * @param worker: worker with scratch region. 808 * @param repinfo: reply information for a communication point. 809 */ 810 static void 811 chaos_replystr(sldns_buffer* pkt, char** str, int num, struct edns_data* edns, 812 struct worker* worker, struct comm_reply* repinfo) 813 { 814 int i; 815 unsigned int rd = LDNS_RD_WIRE(sldns_buffer_begin(pkt)); 816 unsigned int cd = LDNS_CD_WIRE(sldns_buffer_begin(pkt)); 817 sldns_buffer_clear(pkt); 818 sldns_buffer_skip(pkt, (ssize_t)sizeof(uint16_t)); /* skip id */ 819 sldns_buffer_write_u16(pkt, (uint16_t)(BIT_QR|BIT_RA)); 820 if(rd) LDNS_RD_SET(sldns_buffer_begin(pkt)); 821 if(cd) LDNS_CD_SET(sldns_buffer_begin(pkt)); 822 sldns_buffer_write_u16(pkt, 1); /* qdcount */ 823 sldns_buffer_write_u16(pkt, (uint16_t)num); /* ancount */ 824 sldns_buffer_write_u16(pkt, 0); /* nscount */ 825 sldns_buffer_write_u16(pkt, 0); /* arcount */ 826 (void)query_dname_len(pkt); /* skip qname */ 827 sldns_buffer_skip(pkt, (ssize_t)sizeof(uint16_t)); /* skip qtype */ 828 sldns_buffer_skip(pkt, (ssize_t)sizeof(uint16_t)); /* skip qclass */ 829 for(i=0; i<num; i++) { 830 size_t len = strlen(str[i]); 831 if(len>255) len=255; /* cap size of TXT record */ 832 sldns_buffer_write_u16(pkt, 0xc00c); /* compr ptr to query */ 833 sldns_buffer_write_u16(pkt, LDNS_RR_TYPE_TXT); 834 sldns_buffer_write_u16(pkt, LDNS_RR_CLASS_CH); 835 sldns_buffer_write_u32(pkt, 0); /* TTL */ 836 sldns_buffer_write_u16(pkt, sizeof(uint8_t) + len); 837 sldns_buffer_write_u8(pkt, len); 838 sldns_buffer_write(pkt, str[i], len); 839 } 840 sldns_buffer_flip(pkt); 841 edns->edns_version = EDNS_ADVERTISED_VERSION; 842 edns->udp_size = EDNS_ADVERTISED_SIZE; 843 edns->bits &= EDNS_DO; 844 if(!inplace_cb_reply_local_call(&worker->env, NULL, NULL, NULL, 845 LDNS_RCODE_NOERROR, edns, repinfo, worker->scratchpad)) 846 edns->opt_list = NULL; 847 if(sldns_buffer_capacity(pkt) >= 848 sldns_buffer_limit(pkt)+calc_edns_field_size(edns)) 849 attach_edns_record(pkt, edns); 850 } 851 852 /** Reply with one string */ 853 static void 854 chaos_replyonestr(sldns_buffer* pkt, const char* str, struct edns_data* edns, 855 struct worker* worker, struct comm_reply* repinfo) 856 { 857 chaos_replystr(pkt, (char**)&str, 1, edns, worker, repinfo); 858 } 859 860 /** 861 * Create CH class trustanchor answer. 862 * @param pkt: buffer 863 * @param edns: edns reply information. 864 * @param w: worker with scratch region. 865 * @param repinfo: reply information for a communication point. 866 */ 867 static void 868 chaos_trustanchor(sldns_buffer* pkt, struct edns_data* edns, struct worker* w, 869 struct comm_reply* repinfo) 870 { 871 #define TA_RESPONSE_MAX_TXT 16 /* max number of TXT records */ 872 #define TA_RESPONSE_MAX_TAGS 32 /* max number of tags printed per zone */ 873 char* str_array[TA_RESPONSE_MAX_TXT]; 874 uint16_t tags[TA_RESPONSE_MAX_TAGS]; 875 int num = 0; 876 struct trust_anchor* ta; 877 878 if(!w->env.need_to_validate) { 879 /* no validator module, reply no trustanchors */ 880 chaos_replystr(pkt, NULL, 0, edns, w, repinfo); 881 return; 882 } 883 884 /* fill the string with contents */ 885 lock_basic_lock(&w->env.anchors->lock); 886 RBTREE_FOR(ta, struct trust_anchor*, w->env.anchors->tree) { 887 char* str; 888 size_t i, numtag, str_len = 255; 889 if(num == TA_RESPONSE_MAX_TXT) continue; 890 str = (char*)regional_alloc(w->scratchpad, str_len); 891 if(!str) continue; 892 lock_basic_lock(&ta->lock); 893 numtag = anchor_list_keytags(ta, tags, TA_RESPONSE_MAX_TAGS); 894 if(numtag == 0) { 895 /* empty, insecure point */ 896 lock_basic_unlock(&ta->lock); 897 continue; 898 } 899 str_array[num] = str; 900 num++; 901 902 /* spool name of anchor */ 903 (void)sldns_wire2str_dname_buf(ta->name, ta->namelen, str, str_len); 904 str_len -= strlen(str); str += strlen(str); 905 /* spool tags */ 906 for(i=0; i<numtag; i++) { 907 snprintf(str, str_len, " %u", (unsigned)tags[i]); 908 str_len -= strlen(str); str += strlen(str); 909 } 910 lock_basic_unlock(&ta->lock); 911 } 912 lock_basic_unlock(&w->env.anchors->lock); 913 914 chaos_replystr(pkt, str_array, num, edns, w, repinfo); 915 regional_free_all(w->scratchpad); 916 } 917 918 /** 919 * Answer CH class queries. 920 * @param w: worker 921 * @param qinfo: query info. Pointer into packet buffer. 922 * @param edns: edns info from query. 923 * @param repinfo: reply information for a communication point. 924 * @param pkt: packet buffer. 925 * @return: true if a reply is to be sent. 926 */ 927 static int 928 answer_chaos(struct worker* w, struct query_info* qinfo, 929 struct edns_data* edns, struct comm_reply* repinfo, sldns_buffer* pkt) 930 { 931 struct config_file* cfg = w->env.cfg; 932 if(qinfo->qtype != LDNS_RR_TYPE_ANY && qinfo->qtype != LDNS_RR_TYPE_TXT) 933 return 0; 934 if(query_dname_compare(qinfo->qname, 935 (uint8_t*)"\002id\006server") == 0 || 936 query_dname_compare(qinfo->qname, 937 (uint8_t*)"\010hostname\004bind") == 0) 938 { 939 if(cfg->hide_identity) 940 return 0; 941 if(cfg->identity==NULL || cfg->identity[0]==0) { 942 char buf[MAXHOSTNAMELEN+1]; 943 if (gethostname(buf, MAXHOSTNAMELEN) == 0) { 944 buf[MAXHOSTNAMELEN] = 0; 945 chaos_replyonestr(pkt, buf, edns, w, repinfo); 946 } else { 947 log_err("gethostname: %s", strerror(errno)); 948 chaos_replyonestr(pkt, "no hostname", edns, w, repinfo); 949 } 950 } 951 else chaos_replyonestr(pkt, cfg->identity, edns, w, repinfo); 952 return 1; 953 } 954 if(query_dname_compare(qinfo->qname, 955 (uint8_t*)"\007version\006server") == 0 || 956 query_dname_compare(qinfo->qname, 957 (uint8_t*)"\007version\004bind") == 0) 958 { 959 if(cfg->hide_version) 960 return 0; 961 if(cfg->version==NULL || cfg->version[0]==0) 962 chaos_replyonestr(pkt, PACKAGE_STRING, edns, w, repinfo); 963 else chaos_replyonestr(pkt, cfg->version, edns, w, repinfo); 964 return 1; 965 } 966 if(query_dname_compare(qinfo->qname, 967 (uint8_t*)"\013trustanchor\007unbound") == 0) 968 { 969 if(cfg->hide_trustanchor) 970 return 0; 971 chaos_trustanchor(pkt, edns, w, repinfo); 972 return 1; 973 } 974 975 return 0; 976 } 977 978 /** 979 * Answer notify queries. These are notifies for authoritative zones, 980 * the reply is an ack that the notify has been received. We need to check 981 * access permission here. 982 * @param w: worker 983 * @param qinfo: query info. Pointer into packet buffer. 984 * @param edns: edns info from query. 985 * @param repinfo: reply info with source address. 986 * @param pkt: packet buffer. 987 */ 988 static void 989 answer_notify(struct worker* w, struct query_info* qinfo, 990 struct edns_data* edns, sldns_buffer* pkt, struct comm_reply* repinfo) 991 { 992 int refused = 0; 993 int rcode = LDNS_RCODE_NOERROR; 994 uint32_t serial = 0; 995 int has_serial; 996 if(!w->env.auth_zones) return; 997 has_serial = auth_zone_parse_notify_serial(pkt, &serial); 998 if(auth_zones_notify(w->env.auth_zones, &w->env, qinfo->qname, 999 qinfo->qname_len, qinfo->qclass, &repinfo->addr, 1000 repinfo->addrlen, has_serial, serial, &refused)) { 1001 rcode = LDNS_RCODE_NOERROR; 1002 } else { 1003 if(refused) 1004 rcode = LDNS_RCODE_REFUSED; 1005 else rcode = LDNS_RCODE_SERVFAIL; 1006 } 1007 1008 if(verbosity >= VERB_DETAIL) { 1009 char buf[380]; 1010 char zname[255+1]; 1011 char sr[25]; 1012 dname_str(qinfo->qname, zname); 1013 sr[0]=0; 1014 if(has_serial) 1015 snprintf(sr, sizeof(sr), "serial %u ", 1016 (unsigned)serial); 1017 if(rcode == LDNS_RCODE_REFUSED) 1018 snprintf(buf, sizeof(buf), 1019 "refused NOTIFY %sfor %s from", sr, zname); 1020 else if(rcode == LDNS_RCODE_SERVFAIL) 1021 snprintf(buf, sizeof(buf), 1022 "servfail for NOTIFY %sfor %s from", sr, zname); 1023 else snprintf(buf, sizeof(buf), 1024 "received NOTIFY %sfor %s from", sr, zname); 1025 log_addr(VERB_DETAIL, buf, &repinfo->addr, repinfo->addrlen); 1026 } 1027 edns->edns_version = EDNS_ADVERTISED_VERSION; 1028 edns->udp_size = EDNS_ADVERTISED_SIZE; 1029 edns->ext_rcode = 0; 1030 edns->bits &= EDNS_DO; 1031 edns->opt_list = NULL; 1032 error_encode(pkt, rcode, qinfo, 1033 *(uint16_t*)(void *)sldns_buffer_begin(pkt), 1034 sldns_buffer_read_u16_at(pkt, 2), edns); 1035 LDNS_OPCODE_SET(sldns_buffer_begin(pkt), LDNS_PACKET_NOTIFY); 1036 } 1037 1038 static int 1039 deny_refuse(struct comm_point* c, enum acl_access acl, 1040 enum acl_access deny, enum acl_access refuse, 1041 struct worker* worker, struct comm_reply* repinfo) 1042 { 1043 if(acl == deny) { 1044 comm_point_drop_reply(repinfo); 1045 if(worker->stats.extended) 1046 worker->stats.unwanted_queries++; 1047 return 0; 1048 } else if(acl == refuse) { 1049 log_addr(VERB_ALGO, "refused query from", 1050 &repinfo->addr, repinfo->addrlen); 1051 log_buf(VERB_ALGO, "refuse", c->buffer); 1052 if(worker->stats.extended) 1053 worker->stats.unwanted_queries++; 1054 if(worker_check_request(c->buffer, worker) == -1) { 1055 comm_point_drop_reply(repinfo); 1056 return 0; /* discard this */ 1057 } 1058 sldns_buffer_set_limit(c->buffer, LDNS_HEADER_SIZE); 1059 sldns_buffer_write_at(c->buffer, 4, 1060 (uint8_t*)"\0\0\0\0\0\0\0\0", 8); 1061 LDNS_QR_SET(sldns_buffer_begin(c->buffer)); 1062 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), 1063 LDNS_RCODE_REFUSED); 1064 sldns_buffer_set_position(c->buffer, LDNS_HEADER_SIZE); 1065 sldns_buffer_flip(c->buffer); 1066 return 1; 1067 } 1068 1069 return -1; 1070 } 1071 1072 static int 1073 deny_refuse_all(struct comm_point* c, enum acl_access acl, 1074 struct worker* worker, struct comm_reply* repinfo) 1075 { 1076 return deny_refuse(c, acl, acl_deny, acl_refuse, worker, repinfo); 1077 } 1078 1079 static int 1080 deny_refuse_non_local(struct comm_point* c, enum acl_access acl, 1081 struct worker* worker, struct comm_reply* repinfo) 1082 { 1083 return deny_refuse(c, acl, acl_deny_non_local, acl_refuse_non_local, worker, repinfo); 1084 } 1085 1086 int 1087 worker_handle_request(struct comm_point* c, void* arg, int error, 1088 struct comm_reply* repinfo) 1089 { 1090 struct worker* worker = (struct worker*)arg; 1091 int ret; 1092 hashvalue_type h; 1093 struct lruhash_entry* e; 1094 struct query_info qinfo; 1095 struct edns_data edns; 1096 enum acl_access acl; 1097 struct acl_addr* acladdr; 1098 int rc = 0; 1099 int need_drop = 0; 1100 int is_expired_answer = 0; 1101 int is_secure_answer = 0; 1102 /* We might have to chase a CNAME chain internally, in which case 1103 * we'll have up to two replies and combine them to build a complete 1104 * answer. These variables control this case. */ 1105 struct ub_packed_rrset_key* alias_rrset = NULL; 1106 struct reply_info* partial_rep = NULL; 1107 struct query_info* lookup_qinfo = &qinfo; 1108 struct query_info qinfo_tmp; /* placeholder for lookup_qinfo */ 1109 struct respip_client_info* cinfo = NULL, cinfo_tmp; 1110 memset(&qinfo, 0, sizeof(qinfo)); 1111 1112 if(error != NETEVENT_NOERROR || !repinfo) { 1113 /* some bad tcp query DNS formats give these error calls */ 1114 verbose(VERB_ALGO, "handle request called with err=%d", error); 1115 return 0; 1116 } 1117 #ifdef USE_DNSCRYPT 1118 repinfo->max_udp_size = worker->daemon->cfg->max_udp_size; 1119 if(!dnsc_handle_curved_request(worker->daemon->dnscenv, repinfo)) { 1120 worker->stats.num_query_dnscrypt_crypted_malformed++; 1121 return 0; 1122 } 1123 if(c->dnscrypt && !repinfo->is_dnscrypted) { 1124 char buf[LDNS_MAX_DOMAINLEN+1]; 1125 /* Check if this is unencrypted and asking for certs */ 1126 if(worker_check_request(c->buffer, worker) != 0) { 1127 verbose(VERB_ALGO, 1128 "dnscrypt: worker check request: bad query."); 1129 log_addr(VERB_CLIENT,"from",&repinfo->addr, 1130 repinfo->addrlen); 1131 comm_point_drop_reply(repinfo); 1132 return 0; 1133 } 1134 if(!query_info_parse(&qinfo, c->buffer)) { 1135 verbose(VERB_ALGO, 1136 "dnscrypt: worker parse request: formerror."); 1137 log_addr(VERB_CLIENT, "from", &repinfo->addr, 1138 repinfo->addrlen); 1139 comm_point_drop_reply(repinfo); 1140 return 0; 1141 } 1142 dname_str(qinfo.qname, buf); 1143 if(!(qinfo.qtype == LDNS_RR_TYPE_TXT && 1144 strcasecmp(buf, 1145 worker->daemon->dnscenv->provider_name) == 0)) { 1146 verbose(VERB_ALGO, 1147 "dnscrypt: not TXT \"%s\". Received: %s \"%s\"", 1148 worker->daemon->dnscenv->provider_name, 1149 sldns_rr_descript(qinfo.qtype)->_name, 1150 buf); 1151 comm_point_drop_reply(repinfo); 1152 worker->stats.num_query_dnscrypt_cleartext++; 1153 return 0; 1154 } 1155 worker->stats.num_query_dnscrypt_cert++; 1156 sldns_buffer_rewind(c->buffer); 1157 } else if(c->dnscrypt && repinfo->is_dnscrypted) { 1158 worker->stats.num_query_dnscrypt_crypted++; 1159 } 1160 #endif 1161 #ifdef USE_DNSTAP 1162 if(worker->dtenv.log_client_query_messages) 1163 dt_msg_send_client_query(&worker->dtenv, &repinfo->addr, c->type, 1164 c->buffer); 1165 #endif 1166 acladdr = acl_addr_lookup(worker->daemon->acl, &repinfo->addr, 1167 repinfo->addrlen); 1168 acl = acl_get_control(acladdr); 1169 if((ret=deny_refuse_all(c, acl, worker, repinfo)) != -1) 1170 { 1171 if(ret == 1) 1172 goto send_reply; 1173 return ret; 1174 } 1175 if((ret=worker_check_request(c->buffer, worker)) != 0) { 1176 verbose(VERB_ALGO, "worker check request: bad query."); 1177 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen); 1178 if(ret != -1) { 1179 LDNS_QR_SET(sldns_buffer_begin(c->buffer)); 1180 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), ret); 1181 return 1; 1182 } 1183 comm_point_drop_reply(repinfo); 1184 return 0; 1185 } 1186 1187 worker->stats.num_queries++; 1188 1189 /* check if this query should be dropped based on source ip rate limiting */ 1190 if(!infra_ip_ratelimit_inc(worker->env.infra_cache, repinfo, 1191 *worker->env.now, c->buffer)) { 1192 /* See if we are passed through with slip factor */ 1193 if(worker->env.cfg->ip_ratelimit_factor != 0 && 1194 ub_random_max(worker->env.rnd, 1195 worker->env.cfg->ip_ratelimit_factor) == 0) { 1196 1197 char addrbuf[128]; 1198 addr_to_str(&repinfo->addr, repinfo->addrlen, 1199 addrbuf, sizeof(addrbuf)); 1200 verbose(VERB_QUERY, "ip_ratelimit allowed through for ip address %s because of slip in ip_ratelimit_factor", 1201 addrbuf); 1202 } else { 1203 worker->stats.num_queries_ip_ratelimited++; 1204 comm_point_drop_reply(repinfo); 1205 return 0; 1206 } 1207 } 1208 1209 /* see if query is in the cache */ 1210 if(!query_info_parse(&qinfo, c->buffer)) { 1211 verbose(VERB_ALGO, "worker parse request: formerror."); 1212 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen); 1213 memset(&qinfo, 0, sizeof(qinfo)); /* zero qinfo.qname */ 1214 if(worker_err_ratelimit(worker, LDNS_RCODE_FORMERR) == -1) { 1215 comm_point_drop_reply(repinfo); 1216 return 0; 1217 } 1218 sldns_buffer_rewind(c->buffer); 1219 LDNS_QR_SET(sldns_buffer_begin(c->buffer)); 1220 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), 1221 LDNS_RCODE_FORMERR); 1222 server_stats_insrcode(&worker->stats, c->buffer); 1223 goto send_reply; 1224 } 1225 if(worker->env.cfg->log_queries) { 1226 char ip[128]; 1227 addr_to_str(&repinfo->addr, repinfo->addrlen, ip, sizeof(ip)); 1228 log_query_in(ip, qinfo.qname, qinfo.qtype, qinfo.qclass); 1229 } 1230 if(qinfo.qtype == LDNS_RR_TYPE_AXFR || 1231 qinfo.qtype == LDNS_RR_TYPE_IXFR) { 1232 verbose(VERB_ALGO, "worker request: refused zone transfer."); 1233 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen); 1234 sldns_buffer_rewind(c->buffer); 1235 LDNS_QR_SET(sldns_buffer_begin(c->buffer)); 1236 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), 1237 LDNS_RCODE_REFUSED); 1238 if(worker->stats.extended) { 1239 worker->stats.qtype[qinfo.qtype]++; 1240 server_stats_insrcode(&worker->stats, c->buffer); 1241 } 1242 goto send_reply; 1243 } 1244 if(qinfo.qtype == LDNS_RR_TYPE_OPT || 1245 qinfo.qtype == LDNS_RR_TYPE_TSIG || 1246 qinfo.qtype == LDNS_RR_TYPE_TKEY || 1247 qinfo.qtype == LDNS_RR_TYPE_MAILA || 1248 qinfo.qtype == LDNS_RR_TYPE_MAILB || 1249 (qinfo.qtype >= 128 && qinfo.qtype <= 248)) { 1250 verbose(VERB_ALGO, "worker request: formerror for meta-type."); 1251 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen); 1252 if(worker_err_ratelimit(worker, LDNS_RCODE_FORMERR) == -1) { 1253 comm_point_drop_reply(repinfo); 1254 return 0; 1255 } 1256 sldns_buffer_rewind(c->buffer); 1257 LDNS_QR_SET(sldns_buffer_begin(c->buffer)); 1258 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), 1259 LDNS_RCODE_FORMERR); 1260 if(worker->stats.extended) { 1261 worker->stats.qtype[qinfo.qtype]++; 1262 server_stats_insrcode(&worker->stats, c->buffer); 1263 } 1264 goto send_reply; 1265 } 1266 if((ret=parse_edns_from_pkt(c->buffer, &edns, worker->scratchpad)) != 0) { 1267 struct edns_data reply_edns; 1268 verbose(VERB_ALGO, "worker parse edns: formerror."); 1269 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen); 1270 memset(&reply_edns, 0, sizeof(reply_edns)); 1271 reply_edns.edns_present = 1; 1272 reply_edns.udp_size = EDNS_ADVERTISED_SIZE; 1273 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), ret); 1274 error_encode(c->buffer, ret, &qinfo, 1275 *(uint16_t*)(void *)sldns_buffer_begin(c->buffer), 1276 sldns_buffer_read_u16_at(c->buffer, 2), &reply_edns); 1277 regional_free_all(worker->scratchpad); 1278 server_stats_insrcode(&worker->stats, c->buffer); 1279 goto send_reply; 1280 } 1281 if(edns.edns_present) { 1282 struct edns_option* edns_opt; 1283 if(edns.edns_version != 0) { 1284 edns.ext_rcode = (uint8_t)(EDNS_RCODE_BADVERS>>4); 1285 edns.edns_version = EDNS_ADVERTISED_VERSION; 1286 edns.udp_size = EDNS_ADVERTISED_SIZE; 1287 edns.bits &= EDNS_DO; 1288 edns.opt_list = NULL; 1289 verbose(VERB_ALGO, "query with bad edns version."); 1290 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen); 1291 error_encode(c->buffer, EDNS_RCODE_BADVERS&0xf, &qinfo, 1292 *(uint16_t*)(void *)sldns_buffer_begin(c->buffer), 1293 sldns_buffer_read_u16_at(c->buffer, 2), NULL); 1294 if(sldns_buffer_capacity(c->buffer) >= 1295 sldns_buffer_limit(c->buffer)+calc_edns_field_size(&edns)) 1296 attach_edns_record(c->buffer, &edns); 1297 regional_free_all(worker->scratchpad); 1298 goto send_reply; 1299 } 1300 if(edns.udp_size < NORMAL_UDP_SIZE && 1301 worker->daemon->cfg->harden_short_bufsize) { 1302 verbose(VERB_QUERY, "worker request: EDNS bufsize %d ignored", 1303 (int)edns.udp_size); 1304 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen); 1305 edns.udp_size = NORMAL_UDP_SIZE; 1306 } 1307 if(c->type != comm_udp) { 1308 edns_opt = edns_opt_list_find(edns.opt_list, LDNS_EDNS_KEEPALIVE); 1309 if(edns_opt && edns_opt->opt_len > 0) { 1310 edns.ext_rcode = 0; 1311 edns.edns_version = EDNS_ADVERTISED_VERSION; 1312 edns.udp_size = EDNS_ADVERTISED_SIZE; 1313 edns.bits &= EDNS_DO; 1314 edns.opt_list = NULL; 1315 verbose(VERB_ALGO, "query with bad edns keepalive."); 1316 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen); 1317 error_encode(c->buffer, LDNS_RCODE_FORMERR, &qinfo, 1318 *(uint16_t*)(void *)sldns_buffer_begin(c->buffer), 1319 sldns_buffer_read_u16_at(c->buffer, 2), NULL); 1320 if(sldns_buffer_capacity(c->buffer) >= 1321 sldns_buffer_limit(c->buffer)+calc_edns_field_size(&edns)) 1322 attach_edns_record(c->buffer, &edns); 1323 regional_free_all(worker->scratchpad); 1324 goto send_reply; 1325 } 1326 } 1327 } 1328 if(edns.udp_size > worker->daemon->cfg->max_udp_size && 1329 c->type == comm_udp) { 1330 verbose(VERB_QUERY, 1331 "worker request: max UDP reply size modified" 1332 " (%d to max-udp-size)", (int)edns.udp_size); 1333 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen); 1334 edns.udp_size = worker->daemon->cfg->max_udp_size; 1335 } 1336 if(edns.udp_size < LDNS_HEADER_SIZE) { 1337 verbose(VERB_ALGO, "worker request: edns is too small."); 1338 log_addr(VERB_CLIENT, "from", &repinfo->addr, repinfo->addrlen); 1339 LDNS_QR_SET(sldns_buffer_begin(c->buffer)); 1340 LDNS_TC_SET(sldns_buffer_begin(c->buffer)); 1341 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), 1342 LDNS_RCODE_SERVFAIL); 1343 sldns_buffer_set_position(c->buffer, LDNS_HEADER_SIZE); 1344 sldns_buffer_write_at(c->buffer, 4, 1345 (uint8_t*)"\0\0\0\0\0\0\0\0", 8); 1346 sldns_buffer_flip(c->buffer); 1347 regional_free_all(worker->scratchpad); 1348 goto send_reply; 1349 } 1350 if(worker->stats.extended) 1351 server_stats_insquery(&worker->stats, c, qinfo.qtype, 1352 qinfo.qclass, &edns, repinfo); 1353 if(c->type != comm_udp) 1354 edns.udp_size = 65535; /* max size for TCP replies */ 1355 if(qinfo.qclass == LDNS_RR_CLASS_CH && answer_chaos(worker, &qinfo, 1356 &edns, repinfo, c->buffer)) { 1357 server_stats_insrcode(&worker->stats, c->buffer); 1358 regional_free_all(worker->scratchpad); 1359 goto send_reply; 1360 } 1361 if(LDNS_OPCODE_WIRE(sldns_buffer_begin(c->buffer)) == 1362 LDNS_PACKET_NOTIFY) { 1363 answer_notify(worker, &qinfo, &edns, c->buffer, repinfo); 1364 regional_free_all(worker->scratchpad); 1365 goto send_reply; 1366 } 1367 if(local_zones_answer(worker->daemon->local_zones, &worker->env, &qinfo, 1368 &edns, c->buffer, worker->scratchpad, repinfo, acladdr->taglist, 1369 acladdr->taglen, acladdr->tag_actions, 1370 acladdr->tag_actions_size, acladdr->tag_datas, 1371 acladdr->tag_datas_size, worker->daemon->cfg->tagname, 1372 worker->daemon->cfg->num_tags, acladdr->view)) { 1373 regional_free_all(worker->scratchpad); 1374 if(sldns_buffer_limit(c->buffer) == 0) { 1375 comm_point_drop_reply(repinfo); 1376 return 0; 1377 } 1378 server_stats_insrcode(&worker->stats, c->buffer); 1379 goto send_reply; 1380 } 1381 if(worker->env.auth_zones && 1382 rpz_apply_qname_trigger(worker->env.auth_zones, 1383 &worker->env, &qinfo, &edns, c->buffer, worker->scratchpad, 1384 repinfo, acladdr->taglist, acladdr->taglen, &worker->stats)) { 1385 regional_free_all(worker->scratchpad); 1386 if(sldns_buffer_limit(c->buffer) == 0) { 1387 comm_point_drop_reply(repinfo); 1388 return 0; 1389 } 1390 server_stats_insrcode(&worker->stats, c->buffer); 1391 goto send_reply; 1392 } 1393 if(worker->env.auth_zones && 1394 auth_zones_answer(worker->env.auth_zones, &worker->env, 1395 &qinfo, &edns, repinfo, c->buffer, worker->scratchpad)) { 1396 regional_free_all(worker->scratchpad); 1397 if(sldns_buffer_limit(c->buffer) == 0) { 1398 comm_point_drop_reply(repinfo); 1399 return 0; 1400 } 1401 /* set RA for everyone that can have recursion (based on 1402 * access control list) */ 1403 if(LDNS_RD_WIRE(sldns_buffer_begin(c->buffer)) && 1404 acl != acl_deny_non_local && acl != acl_refuse_non_local) 1405 LDNS_RA_SET(sldns_buffer_begin(c->buffer)); 1406 server_stats_insrcode(&worker->stats, c->buffer); 1407 goto send_reply; 1408 } 1409 1410 /* We've looked in our local zones. If the answer isn't there, we 1411 * might need to bail out based on ACLs now. */ 1412 if((ret=deny_refuse_non_local(c, acl, worker, repinfo)) != -1) 1413 { 1414 regional_free_all(worker->scratchpad); 1415 if(ret == 1) 1416 goto send_reply; 1417 return ret; 1418 } 1419 1420 /* If this request does not have the recursion bit set, verify 1421 * ACLs allow the recursion bit to be treated as set. */ 1422 if(!(LDNS_RD_WIRE(sldns_buffer_begin(c->buffer))) && 1423 acl == acl_allow_setrd ) { 1424 LDNS_RD_SET(sldns_buffer_begin(c->buffer)); 1425 } 1426 1427 /* If this request does not have the recursion bit set, verify 1428 * ACLs allow the snooping. */ 1429 if(!(LDNS_RD_WIRE(sldns_buffer_begin(c->buffer))) && 1430 acl != acl_allow_snoop ) { 1431 error_encode(c->buffer, LDNS_RCODE_REFUSED, &qinfo, 1432 *(uint16_t*)(void *)sldns_buffer_begin(c->buffer), 1433 sldns_buffer_read_u16_at(c->buffer, 2), NULL); 1434 regional_free_all(worker->scratchpad); 1435 server_stats_insrcode(&worker->stats, c->buffer); 1436 log_addr(VERB_ALGO, "refused nonrec (cache snoop) query from", 1437 &repinfo->addr, repinfo->addrlen); 1438 goto send_reply; 1439 } 1440 1441 /* If we've found a local alias, replace the qname with the alias 1442 * target before resolving it. */ 1443 if(qinfo.local_alias) { 1444 struct ub_packed_rrset_key* rrset = qinfo.local_alias->rrset; 1445 struct packed_rrset_data* d = rrset->entry.data; 1446 1447 /* Sanity check: our current implementation only supports 1448 * a single CNAME RRset as a local alias. */ 1449 if(qinfo.local_alias->next || 1450 rrset->rk.type != htons(LDNS_RR_TYPE_CNAME) || 1451 d->count != 1) { 1452 log_err("assumption failure: unexpected local alias"); 1453 regional_free_all(worker->scratchpad); 1454 return 0; /* drop it */ 1455 } 1456 qinfo.qname = d->rr_data[0] + 2; 1457 qinfo.qname_len = d->rr_len[0] - 2; 1458 } 1459 1460 /* If we may apply IP-based actions to the answer, build the client 1461 * information. As this can be expensive, skip it if there is 1462 * absolutely no possibility of it. */ 1463 if((worker->daemon->use_response_ip || worker->daemon->use_rpz) && 1464 (qinfo.qtype == LDNS_RR_TYPE_A || 1465 qinfo.qtype == LDNS_RR_TYPE_AAAA || 1466 qinfo.qtype == LDNS_RR_TYPE_ANY)) { 1467 cinfo_tmp.taglist = acladdr->taglist; 1468 cinfo_tmp.taglen = acladdr->taglen; 1469 cinfo_tmp.tag_actions = acladdr->tag_actions; 1470 cinfo_tmp.tag_actions_size = acladdr->tag_actions_size; 1471 cinfo_tmp.tag_datas = acladdr->tag_datas; 1472 cinfo_tmp.tag_datas_size = acladdr->tag_datas_size; 1473 cinfo_tmp.view = acladdr->view; 1474 cinfo_tmp.respip_set = worker->daemon->respip_set; 1475 cinfo = &cinfo_tmp; 1476 } 1477 1478 lookup_cache: 1479 /* Lookup the cache. In case we chase an intermediate CNAME chain 1480 * this is a two-pass operation, and lookup_qinfo is different for 1481 * each pass. We should still pass the original qinfo to 1482 * answer_from_cache(), however, since it's used to build the reply. */ 1483 if(!edns_bypass_cache_stage(edns.opt_list, &worker->env)) { 1484 is_expired_answer = 0; 1485 is_secure_answer = 0; 1486 h = query_info_hash(lookup_qinfo, sldns_buffer_read_u16_at(c->buffer, 2)); 1487 if((e=slabhash_lookup(worker->env.msg_cache, h, lookup_qinfo, 0))) { 1488 /* answer from cache - we have acquired a readlock on it */ 1489 if(answer_from_cache(worker, &qinfo, 1490 cinfo, &need_drop, &is_expired_answer, &is_secure_answer, 1491 &alias_rrset, &partial_rep, (struct reply_info*)e->data, 1492 *(uint16_t*)(void *)sldns_buffer_begin(c->buffer), 1493 sldns_buffer_read_u16_at(c->buffer, 2), repinfo, 1494 &edns)) { 1495 /* prefetch it if the prefetch TTL expired. 1496 * Note that if there is more than one pass 1497 * its qname must be that used for cache 1498 * lookup. */ 1499 if((worker->env.cfg->prefetch && *worker->env.now >= 1500 ((struct reply_info*)e->data)->prefetch_ttl) || 1501 (worker->env.cfg->serve_expired && 1502 *worker->env.now >= ((struct reply_info*)e->data)->ttl)) { 1503 1504 time_t leeway = ((struct reply_info*)e-> 1505 data)->ttl - *worker->env.now; 1506 if(((struct reply_info*)e->data)->ttl 1507 < *worker->env.now) 1508 leeway = 0; 1509 lock_rw_unlock(&e->lock); 1510 reply_and_prefetch(worker, lookup_qinfo, 1511 sldns_buffer_read_u16_at(c->buffer, 2), 1512 repinfo, leeway, 1513 (partial_rep || need_drop)); 1514 if(!partial_rep) { 1515 rc = 0; 1516 regional_free_all(worker->scratchpad); 1517 goto send_reply_rc; 1518 } 1519 } else if(!partial_rep) { 1520 lock_rw_unlock(&e->lock); 1521 regional_free_all(worker->scratchpad); 1522 goto send_reply; 1523 } else { 1524 /* Note that we've already released the 1525 * lock if we're here after prefetch. */ 1526 lock_rw_unlock(&e->lock); 1527 } 1528 /* We've found a partial reply ending with an 1529 * alias. Replace the lookup qinfo for the 1530 * alias target and lookup the cache again to 1531 * (possibly) complete the reply. As we're 1532 * passing the "base" reply, there will be no 1533 * more alias chasing. */ 1534 memset(&qinfo_tmp, 0, sizeof(qinfo_tmp)); 1535 get_cname_target(alias_rrset, &qinfo_tmp.qname, 1536 &qinfo_tmp.qname_len); 1537 if(!qinfo_tmp.qname) { 1538 log_err("unexpected: invalid answer alias"); 1539 regional_free_all(worker->scratchpad); 1540 return 0; /* drop query */ 1541 } 1542 qinfo_tmp.qtype = qinfo.qtype; 1543 qinfo_tmp.qclass = qinfo.qclass; 1544 lookup_qinfo = &qinfo_tmp; 1545 goto lookup_cache; 1546 } 1547 verbose(VERB_ALGO, "answer from the cache failed"); 1548 lock_rw_unlock(&e->lock); 1549 } 1550 if(!LDNS_RD_WIRE(sldns_buffer_begin(c->buffer))) { 1551 if(answer_norec_from_cache(worker, &qinfo, 1552 *(uint16_t*)(void *)sldns_buffer_begin(c->buffer), 1553 sldns_buffer_read_u16_at(c->buffer, 2), repinfo, 1554 &edns)) { 1555 regional_free_all(worker->scratchpad); 1556 goto send_reply; 1557 } 1558 verbose(VERB_ALGO, "answer norec from cache -- " 1559 "need to validate or not primed"); 1560 } 1561 } 1562 sldns_buffer_rewind(c->buffer); 1563 server_stats_querymiss(&worker->stats, worker); 1564 1565 if(verbosity >= VERB_CLIENT) { 1566 if(c->type == comm_udp) 1567 log_addr(VERB_CLIENT, "udp request from", 1568 &repinfo->addr, repinfo->addrlen); 1569 else log_addr(VERB_CLIENT, "tcp request from", 1570 &repinfo->addr, repinfo->addrlen); 1571 } 1572 1573 /* grab a work request structure for this new request */ 1574 mesh_new_client(worker->env.mesh, &qinfo, cinfo, 1575 sldns_buffer_read_u16_at(c->buffer, 2), 1576 &edns, repinfo, *(uint16_t*)(void *)sldns_buffer_begin(c->buffer)); 1577 regional_free_all(worker->scratchpad); 1578 worker_mem_report(worker, NULL); 1579 return 0; 1580 1581 send_reply: 1582 rc = 1; 1583 send_reply_rc: 1584 if(need_drop) { 1585 comm_point_drop_reply(repinfo); 1586 return 0; 1587 } 1588 if(is_expired_answer) { 1589 worker->stats.ans_expired++; 1590 } 1591 if(worker->stats.extended) { 1592 if(is_secure_answer) worker->stats.ans_secure++; 1593 server_stats_insrcode(&worker->stats, repinfo->c->buffer); 1594 } 1595 #ifdef USE_DNSTAP 1596 if(worker->dtenv.log_client_response_messages) 1597 dt_msg_send_client_response(&worker->dtenv, &repinfo->addr, 1598 c->type, c->buffer); 1599 #endif 1600 if(worker->env.cfg->log_replies) 1601 { 1602 struct timeval tv; 1603 memset(&tv, 0, sizeof(tv)); 1604 if(qinfo.local_alias && qinfo.local_alias->rrset && 1605 qinfo.local_alias->rrset->rk.dname) { 1606 /* log original qname, before the local alias was 1607 * used to resolve that CNAME to something else */ 1608 qinfo.qname = qinfo.local_alias->rrset->rk.dname; 1609 log_reply_info(NO_VERBOSE, &qinfo, &repinfo->addr, repinfo->addrlen, 1610 tv, 1, c->buffer); 1611 } else { 1612 log_reply_info(NO_VERBOSE, &qinfo, &repinfo->addr, repinfo->addrlen, 1613 tv, 1, c->buffer); 1614 } 1615 } 1616 #ifdef USE_DNSCRYPT 1617 if(!dnsc_handle_uncurved_request(repinfo)) { 1618 return 0; 1619 } 1620 #endif 1621 return rc; 1622 } 1623 1624 void 1625 worker_sighandler(int sig, void* arg) 1626 { 1627 /* note that log, print, syscalls here give race conditions. 1628 * And cause hangups if the log-lock is held by the application. */ 1629 struct worker* worker = (struct worker*)arg; 1630 switch(sig) { 1631 #ifdef SIGHUP 1632 case SIGHUP: 1633 comm_base_exit(worker->base); 1634 break; 1635 #endif 1636 case SIGINT: 1637 worker->need_to_exit = 1; 1638 comm_base_exit(worker->base); 1639 break; 1640 #ifdef SIGQUIT 1641 case SIGQUIT: 1642 worker->need_to_exit = 1; 1643 comm_base_exit(worker->base); 1644 break; 1645 #endif 1646 case SIGTERM: 1647 worker->need_to_exit = 1; 1648 comm_base_exit(worker->base); 1649 break; 1650 default: 1651 /* unknown signal, ignored */ 1652 break; 1653 } 1654 } 1655 1656 /** restart statistics timer for worker, if enabled */ 1657 static void 1658 worker_restart_timer(struct worker* worker) 1659 { 1660 if(worker->env.cfg->stat_interval > 0) { 1661 struct timeval tv; 1662 #ifndef S_SPLINT_S 1663 tv.tv_sec = worker->env.cfg->stat_interval; 1664 tv.tv_usec = 0; 1665 #endif 1666 comm_timer_set(worker->stat_timer, &tv); 1667 } 1668 } 1669 1670 void worker_stat_timer_cb(void* arg) 1671 { 1672 struct worker* worker = (struct worker*)arg; 1673 server_stats_log(&worker->stats, worker, worker->thread_num); 1674 mesh_stats(worker->env.mesh, "mesh has"); 1675 worker_mem_report(worker, NULL); 1676 /* SHM is enabled, process data to SHM */ 1677 if (worker->daemon->cfg->shm_enable) { 1678 shm_main_run(worker); 1679 } 1680 if(!worker->daemon->cfg->stat_cumulative) { 1681 worker_stats_clear(worker); 1682 } 1683 /* start next timer */ 1684 worker_restart_timer(worker); 1685 } 1686 1687 void worker_probe_timer_cb(void* arg) 1688 { 1689 struct worker* worker = (struct worker*)arg; 1690 struct timeval tv; 1691 #ifndef S_SPLINT_S 1692 tv.tv_sec = (time_t)autr_probe_timer(&worker->env); 1693 tv.tv_usec = 0; 1694 #endif 1695 if(tv.tv_sec != 0) 1696 comm_timer_set(worker->env.probe_timer, &tv); 1697 } 1698 1699 struct worker* 1700 worker_create(struct daemon* daemon, int id, int* ports, int n) 1701 { 1702 unsigned int seed; 1703 struct worker* worker = (struct worker*)calloc(1, 1704 sizeof(struct worker)); 1705 if(!worker) 1706 return NULL; 1707 worker->numports = n; 1708 worker->ports = (int*)memdup(ports, sizeof(int)*n); 1709 if(!worker->ports) { 1710 free(worker); 1711 return NULL; 1712 } 1713 worker->daemon = daemon; 1714 worker->thread_num = id; 1715 if(!(worker->cmd = tube_create())) { 1716 free(worker->ports); 1717 free(worker); 1718 return NULL; 1719 } 1720 /* create random state here to avoid locking trouble in RAND_bytes */ 1721 if(!(worker->rndstate = ub_initstate(daemon->rand))) { 1722 log_err("could not init random numbers."); 1723 tube_delete(worker->cmd); 1724 free(worker->ports); 1725 free(worker); 1726 return NULL; 1727 } 1728 explicit_bzero(&seed, sizeof(seed)); 1729 #ifdef USE_DNSTAP 1730 if(daemon->cfg->dnstap) { 1731 log_assert(daemon->dtenv != NULL); 1732 memcpy(&worker->dtenv, daemon->dtenv, sizeof(struct dt_env)); 1733 if(!dt_init(&worker->dtenv)) 1734 fatal_exit("dt_init failed"); 1735 } 1736 #endif 1737 return worker; 1738 } 1739 1740 int 1741 worker_init(struct worker* worker, struct config_file *cfg, 1742 struct listen_port* ports, int do_sigs) 1743 { 1744 #ifdef USE_DNSTAP 1745 struct dt_env* dtenv = &worker->dtenv; 1746 #else 1747 void* dtenv = NULL; 1748 #endif 1749 worker->need_to_exit = 0; 1750 worker->base = comm_base_create(do_sigs); 1751 if(!worker->base) { 1752 log_err("could not create event handling base"); 1753 worker_delete(worker); 1754 return 0; 1755 } 1756 comm_base_set_slow_accept_handlers(worker->base, &worker_stop_accept, 1757 &worker_start_accept, worker); 1758 if(do_sigs) { 1759 #ifdef SIGHUP 1760 ub_thread_sig_unblock(SIGHUP); 1761 #endif 1762 ub_thread_sig_unblock(SIGINT); 1763 #ifdef SIGQUIT 1764 ub_thread_sig_unblock(SIGQUIT); 1765 #endif 1766 ub_thread_sig_unblock(SIGTERM); 1767 #ifndef LIBEVENT_SIGNAL_PROBLEM 1768 worker->comsig = comm_signal_create(worker->base, 1769 worker_sighandler, worker); 1770 if(!worker->comsig 1771 #ifdef SIGHUP 1772 || !comm_signal_bind(worker->comsig, SIGHUP) 1773 #endif 1774 #ifdef SIGQUIT 1775 || !comm_signal_bind(worker->comsig, SIGQUIT) 1776 #endif 1777 || !comm_signal_bind(worker->comsig, SIGTERM) 1778 || !comm_signal_bind(worker->comsig, SIGINT)) { 1779 log_err("could not create signal handlers"); 1780 worker_delete(worker); 1781 return 0; 1782 } 1783 #endif /* LIBEVENT_SIGNAL_PROBLEM */ 1784 if(!daemon_remote_open_accept(worker->daemon->rc, 1785 worker->daemon->rc_ports, worker)) { 1786 worker_delete(worker); 1787 return 0; 1788 } 1789 #ifdef UB_ON_WINDOWS 1790 wsvc_setup_worker(worker); 1791 #endif /* UB_ON_WINDOWS */ 1792 } else { /* !do_sigs */ 1793 worker->comsig = NULL; 1794 } 1795 worker->front = listen_create(worker->base, ports, 1796 cfg->msg_buffer_size, (int)cfg->incoming_num_tcp, 1797 cfg->do_tcp_keepalive 1798 ? cfg->tcp_keepalive_timeout 1799 : cfg->tcp_idle_timeout, 1800 worker->daemon->tcl, 1801 worker->daemon->listen_sslctx, 1802 dtenv, worker_handle_request, worker); 1803 if(!worker->front) { 1804 log_err("could not create listening sockets"); 1805 worker_delete(worker); 1806 return 0; 1807 } 1808 worker->back = outside_network_create(worker->base, 1809 cfg->msg_buffer_size, (size_t)cfg->outgoing_num_ports, 1810 cfg->out_ifs, cfg->num_out_ifs, cfg->do_ip4, cfg->do_ip6, 1811 cfg->do_tcp?cfg->outgoing_num_tcp:0, cfg->ip_dscp, 1812 worker->daemon->env->infra_cache, worker->rndstate, 1813 cfg->use_caps_bits_for_id, worker->ports, worker->numports, 1814 cfg->unwanted_threshold, cfg->outgoing_tcp_mss, 1815 &worker_alloc_cleanup, worker, 1816 cfg->do_udp || cfg->udp_upstream_without_downstream, 1817 worker->daemon->connect_sslctx, cfg->delay_close, 1818 cfg->tls_use_sni, dtenv); 1819 if(!worker->back) { 1820 log_err("could not create outgoing sockets"); 1821 worker_delete(worker); 1822 return 0; 1823 } 1824 /* start listening to commands */ 1825 if(!tube_setup_bg_listen(worker->cmd, worker->base, 1826 &worker_handle_control_cmd, worker)) { 1827 log_err("could not create control compt."); 1828 worker_delete(worker); 1829 return 0; 1830 } 1831 worker->stat_timer = comm_timer_create(worker->base, 1832 worker_stat_timer_cb, worker); 1833 if(!worker->stat_timer) { 1834 log_err("could not create statistics timer"); 1835 } 1836 1837 /* we use the msg_buffer_size as a good estimate for what the 1838 * user wants for memory usage sizes */ 1839 worker->scratchpad = regional_create_custom(cfg->msg_buffer_size); 1840 if(!worker->scratchpad) { 1841 log_err("malloc failure"); 1842 worker_delete(worker); 1843 return 0; 1844 } 1845 1846 server_stats_init(&worker->stats, cfg); 1847 alloc_init(&worker->alloc, &worker->daemon->superalloc, 1848 worker->thread_num); 1849 alloc_set_id_cleanup(&worker->alloc, &worker_alloc_cleanup, worker); 1850 worker->env = *worker->daemon->env; 1851 comm_base_timept(worker->base, &worker->env.now, &worker->env.now_tv); 1852 worker->env.worker = worker; 1853 worker->env.worker_base = worker->base; 1854 worker->env.send_query = &worker_send_query; 1855 worker->env.alloc = &worker->alloc; 1856 worker->env.outnet = worker->back; 1857 worker->env.rnd = worker->rndstate; 1858 /* If case prefetch is triggered, the corresponding mesh will clear 1859 * the scratchpad for the module env in the middle of request handling. 1860 * It would be prone to a use-after-free kind of bug, so we avoid 1861 * sharing it with worker's own scratchpad at the cost of having 1862 * one more pad per worker. */ 1863 worker->env.scratch = regional_create_custom(cfg->msg_buffer_size); 1864 if(!worker->env.scratch) { 1865 log_err("malloc failure"); 1866 worker_delete(worker); 1867 return 0; 1868 } 1869 worker->env.mesh = mesh_create(&worker->daemon->mods, &worker->env); 1870 /* Pass on daemon variables that we would need in the mesh area */ 1871 worker->env.mesh->use_response_ip = worker->daemon->use_response_ip; 1872 worker->env.mesh->use_rpz = worker->daemon->use_rpz; 1873 1874 worker->env.detach_subs = &mesh_detach_subs; 1875 worker->env.attach_sub = &mesh_attach_sub; 1876 worker->env.add_sub = &mesh_add_sub; 1877 worker->env.kill_sub = &mesh_state_delete; 1878 worker->env.detect_cycle = &mesh_detect_cycle; 1879 worker->env.scratch_buffer = sldns_buffer_new(cfg->msg_buffer_size); 1880 if(!(worker->env.fwds = forwards_create()) || 1881 !forwards_apply_cfg(worker->env.fwds, cfg)) { 1882 log_err("Could not set forward zones"); 1883 worker_delete(worker); 1884 return 0; 1885 } 1886 if(!(worker->env.hints = hints_create()) || 1887 !hints_apply_cfg(worker->env.hints, cfg)) { 1888 log_err("Could not set root or stub hints"); 1889 worker_delete(worker); 1890 return 0; 1891 } 1892 /* one probe timer per process -- if we have 5011 anchors */ 1893 if(autr_get_num_anchors(worker->env.anchors) > 0 1894 #ifndef THREADS_DISABLED 1895 && worker->thread_num == 0 1896 #endif 1897 ) { 1898 struct timeval tv; 1899 tv.tv_sec = 0; 1900 tv.tv_usec = 0; 1901 worker->env.probe_timer = comm_timer_create(worker->base, 1902 worker_probe_timer_cb, worker); 1903 if(!worker->env.probe_timer) { 1904 log_err("could not create 5011-probe timer"); 1905 } else { 1906 /* let timer fire, then it can reset itself */ 1907 comm_timer_set(worker->env.probe_timer, &tv); 1908 } 1909 } 1910 /* zone transfer tasks, setup once per process, if any */ 1911 if(worker->env.auth_zones 1912 #ifndef THREADS_DISABLED 1913 && worker->thread_num == 0 1914 #endif 1915 ) { 1916 auth_xfer_pickup_initial(worker->env.auth_zones, &worker->env); 1917 } 1918 #ifdef USE_DNSTAP 1919 if(worker->daemon->cfg->dnstap 1920 #ifndef THREADS_DISABLED 1921 && worker->thread_num == 0 1922 #endif 1923 ) { 1924 if(!dt_io_thread_start(dtenv->dtio, comm_base_internal( 1925 worker->base), worker->daemon->num)) { 1926 log_err("could not start dnstap io thread"); 1927 worker_delete(worker); 1928 return 0; 1929 } 1930 } 1931 #endif /* USE_DNSTAP */ 1932 if(!worker->env.mesh || !worker->env.scratch_buffer) { 1933 worker_delete(worker); 1934 return 0; 1935 } 1936 worker_mem_report(worker, NULL); 1937 /* if statistics enabled start timer */ 1938 if(worker->env.cfg->stat_interval > 0) { 1939 verbose(VERB_ALGO, "set statistics interval %d secs", 1940 worker->env.cfg->stat_interval); 1941 worker_restart_timer(worker); 1942 } 1943 return 1; 1944 } 1945 1946 void 1947 worker_work(struct worker* worker) 1948 { 1949 comm_base_dispatch(worker->base); 1950 } 1951 1952 void 1953 worker_delete(struct worker* worker) 1954 { 1955 if(!worker) 1956 return; 1957 if(worker->env.mesh && verbosity >= VERB_OPS) { 1958 server_stats_log(&worker->stats, worker, worker->thread_num); 1959 mesh_stats(worker->env.mesh, "mesh has"); 1960 worker_mem_report(worker, NULL); 1961 } 1962 outside_network_quit_prepare(worker->back); 1963 mesh_delete(worker->env.mesh); 1964 sldns_buffer_free(worker->env.scratch_buffer); 1965 forwards_delete(worker->env.fwds); 1966 hints_delete(worker->env.hints); 1967 listen_delete(worker->front); 1968 outside_network_delete(worker->back); 1969 comm_signal_delete(worker->comsig); 1970 tube_delete(worker->cmd); 1971 comm_timer_delete(worker->stat_timer); 1972 comm_timer_delete(worker->env.probe_timer); 1973 free(worker->ports); 1974 if(worker->thread_num == 0) { 1975 #ifdef UB_ON_WINDOWS 1976 wsvc_desetup_worker(worker); 1977 #endif /* UB_ON_WINDOWS */ 1978 } 1979 #ifdef USE_DNSTAP 1980 if(worker->daemon->cfg->dnstap 1981 #ifndef THREADS_DISABLED 1982 && worker->thread_num == 0 1983 #endif 1984 ) { 1985 dt_io_thread_stop(worker->dtenv.dtio); 1986 } 1987 dt_deinit(&worker->dtenv); 1988 #endif /* USE_DNSTAP */ 1989 comm_base_delete(worker->base); 1990 ub_randfree(worker->rndstate); 1991 alloc_clear(&worker->alloc); 1992 regional_destroy(worker->env.scratch); 1993 regional_destroy(worker->scratchpad); 1994 free(worker); 1995 } 1996 1997 struct outbound_entry* 1998 worker_send_query(struct query_info* qinfo, uint16_t flags, int dnssec, 1999 int want_dnssec, int nocaps, struct sockaddr_storage* addr, 2000 socklen_t addrlen, uint8_t* zone, size_t zonelen, int ssl_upstream, 2001 char* tls_auth_name, struct module_qstate* q) 2002 { 2003 struct worker* worker = q->env->worker; 2004 struct outbound_entry* e = (struct outbound_entry*)regional_alloc( 2005 q->region, sizeof(*e)); 2006 if(!e) 2007 return NULL; 2008 e->qstate = q; 2009 e->qsent = outnet_serviced_query(worker->back, qinfo, flags, dnssec, 2010 want_dnssec, nocaps, q->env->cfg->tcp_upstream, 2011 ssl_upstream, tls_auth_name, addr, addrlen, zone, zonelen, q, 2012 worker_handle_service_reply, e, worker->back->udp_buff, q->env); 2013 if(!e->qsent) { 2014 return NULL; 2015 } 2016 return e; 2017 } 2018 2019 void 2020 worker_alloc_cleanup(void* arg) 2021 { 2022 struct worker* worker = (struct worker*)arg; 2023 slabhash_clear(&worker->env.rrset_cache->table); 2024 slabhash_clear(worker->env.msg_cache); 2025 } 2026 2027 void worker_stats_clear(struct worker* worker) 2028 { 2029 server_stats_init(&worker->stats, worker->env.cfg); 2030 mesh_stats_clear(worker->env.mesh); 2031 worker->back->unwanted_replies = 0; 2032 worker->back->num_tcp_outgoing = 0; 2033 } 2034 2035 void worker_start_accept(void* arg) 2036 { 2037 struct worker* worker = (struct worker*)arg; 2038 listen_start_accept(worker->front); 2039 if(worker->thread_num == 0) 2040 daemon_remote_start_accept(worker->daemon->rc); 2041 } 2042 2043 void worker_stop_accept(void* arg) 2044 { 2045 struct worker* worker = (struct worker*)arg; 2046 listen_stop_accept(worker->front); 2047 if(worker->thread_num == 0) 2048 daemon_remote_stop_accept(worker->daemon->rc); 2049 } 2050 2051 /* --- fake callbacks for fptr_wlist to work --- */ 2052 struct outbound_entry* libworker_send_query( 2053 struct query_info* ATTR_UNUSED(qinfo), 2054 uint16_t ATTR_UNUSED(flags), int ATTR_UNUSED(dnssec), 2055 int ATTR_UNUSED(want_dnssec), int ATTR_UNUSED(nocaps), 2056 struct sockaddr_storage* ATTR_UNUSED(addr), socklen_t ATTR_UNUSED(addrlen), 2057 uint8_t* ATTR_UNUSED(zone), size_t ATTR_UNUSED(zonelen), 2058 int ATTR_UNUSED(ssl_upstream), char* ATTR_UNUSED(tls_auth_name), 2059 struct module_qstate* ATTR_UNUSED(q)) 2060 { 2061 log_assert(0); 2062 return 0; 2063 } 2064 2065 int libworker_handle_reply(struct comm_point* ATTR_UNUSED(c), 2066 void* ATTR_UNUSED(arg), int ATTR_UNUSED(error), 2067 struct comm_reply* ATTR_UNUSED(reply_info)) 2068 { 2069 log_assert(0); 2070 return 0; 2071 } 2072 2073 int libworker_handle_service_reply(struct comm_point* ATTR_UNUSED(c), 2074 void* ATTR_UNUSED(arg), int ATTR_UNUSED(error), 2075 struct comm_reply* ATTR_UNUSED(reply_info)) 2076 { 2077 log_assert(0); 2078 return 0; 2079 } 2080 2081 void libworker_handle_control_cmd(struct tube* ATTR_UNUSED(tube), 2082 uint8_t* ATTR_UNUSED(buffer), size_t ATTR_UNUSED(len), 2083 int ATTR_UNUSED(error), void* ATTR_UNUSED(arg)) 2084 { 2085 log_assert(0); 2086 } 2087 2088 void libworker_fg_done_cb(void* ATTR_UNUSED(arg), int ATTR_UNUSED(rcode), 2089 sldns_buffer* ATTR_UNUSED(buf), enum sec_status ATTR_UNUSED(s), 2090 char* ATTR_UNUSED(why_bogus), int ATTR_UNUSED(was_ratelimited)) 2091 { 2092 log_assert(0); 2093 } 2094 2095 void libworker_bg_done_cb(void* ATTR_UNUSED(arg), int ATTR_UNUSED(rcode), 2096 sldns_buffer* ATTR_UNUSED(buf), enum sec_status ATTR_UNUSED(s), 2097 char* ATTR_UNUSED(why_bogus), int ATTR_UNUSED(was_ratelimited)) 2098 { 2099 log_assert(0); 2100 } 2101 2102 void libworker_event_done_cb(void* ATTR_UNUSED(arg), int ATTR_UNUSED(rcode), 2103 sldns_buffer* ATTR_UNUSED(buf), enum sec_status ATTR_UNUSED(s), 2104 char* ATTR_UNUSED(why_bogus), int ATTR_UNUSED(was_ratelimited)) 2105 { 2106 log_assert(0); 2107 } 2108 2109 int context_query_cmp(const void* ATTR_UNUSED(a), const void* ATTR_UNUSED(b)) 2110 { 2111 log_assert(0); 2112 return 0; 2113 } 2114 2115 int order_lock_cmp(const void* ATTR_UNUSED(e1), const void* ATTR_UNUSED(e2)) 2116 { 2117 log_assert(0); 2118 return 0; 2119 } 2120 2121 int codeline_cmp(const void* ATTR_UNUSED(a), const void* ATTR_UNUSED(b)) 2122 { 2123 log_assert(0); 2124 return 0; 2125 } 2126 2127 #ifdef USE_DNSTAP 2128 void dtio_tap_callback(int ATTR_UNUSED(fd), short ATTR_UNUSED(ev), 2129 void* ATTR_UNUSED(arg)) 2130 { 2131 log_assert(0); 2132 } 2133 #endif 2134 2135 #ifdef USE_DNSTAP 2136 void dtio_mainfdcallback(int ATTR_UNUSED(fd), short ATTR_UNUSED(ev), 2137 void* ATTR_UNUSED(arg)) 2138 { 2139 log_assert(0); 2140 } 2141 #endif 2142