daemon/worker.c

   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 LIMITED
  25  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
  26  * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE
  27  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  28  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  29  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
  30  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
  31  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
  32  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  33  * POSSIBILITY OF SUCH DAMAGE.
  34  */
  35
  36 /**
  37  * \file
  38  *
  39  * This file implements the worker that handles callbacks on events, for
  40  * pending requests.
  41  */
  42 #include "config.h"
  43 #include <ldns/wire2host.h>
  44 #include "util/log.h"
  45 #include "util/net_help.h"
  46 #include "util/random.h"
  47 #include "daemon/worker.h"
  48 #include "daemon/daemon.h"
  49 #include "daemon/remote.h"
  50 #include "daemon/acl_list.h"
  51 #include "util/netevent.h"
  52 #include "util/config_file.h"
  53 #include "util/module.h"
  54 #include "util/regional.h"
  55 #include "util/storage/slabhash.h"
  56 #include "services/listen_dnsport.h"
  57 #include "services/outside_network.h"
  58 #include "services/outbound_list.h"
  59 #include "services/cache/rrset.h"
  60 #include "services/cache/infra.h"
  61 #include "services/cache/dns.h"
  62 #include "services/mesh.h"
  63 #include "services/localzone.h"
  64 #include "util/data/msgparse.h"
  65 #include "util/data/msgencode.h"
  66 #include "util/data/dname.h"
  67 #include "util/fptr_wlist.h"
  68 #include "util/tube.h"
  69 #include "iterator/iter_fwd.h"
  70 #include "iterator/iter_hints.h"
  71 #include "validator/autotrust.h"
  72 #include "validator/val_anchor.h"
  73
  74 #ifdef HAVE_SYS_TYPES_H
  75 #  include <sys/types.h>
  76 #endif
  77 #ifdef HAVE_NETDB_H
  78 #include <netdb.h>
  79 #endif
  80 #include <signal.h>
  81 #ifdef UB_ON_WINDOWS
  82 #include "winrc/win_svc.h"
  83 #endif
  84
  85 /** Size of an UDP datagram */
  86 #define NORMAL_UDP_SIZE 512 /* bytes */
  87
  88 /**
  89  * seconds to add to prefetch leeway.  This is a TTL that expires old rrsets
  90  * earlier than they should in order to put the new update into the cache.
  91  * This additional value is to make sure that if not all TTLs are equal in
  92  * the message to be updated(and replaced), that rrsets with up to this much
  93  * extra TTL are also replaced.  This means that the resulting new message
  94  * will have (most likely) this TTL at least, avoiding very small 'split
  95  * second' TTLs due to operators choosing relative primes for TTLs (or so).
  96  * Also has to be at least one to break ties (and overwrite cached entry).
  97  */
  98 #define PREFETCH_EXPIRY_ADD 60
  99
 100 #ifdef UNBOUND_ALLOC_STATS
 101 /** measure memory leakage */
 102 static void
 103 debug_memleak(size_t accounted, size_t heap,
 104         size_t total_alloc, size_t total_free)
 105 {
 106         static int init = 0;
 107         static size_t base_heap, base_accounted, base_alloc, base_free;
 108         size_t base_af, cur_af, grow_af, grow_acc;
 109         if(!init) {
 110                 init = 1;
 111                 base_heap = heap;
 112                 base_accounted = accounted;
 113                 base_alloc = total_alloc;
 114                 base_free = total_free;
 115         }
 116         base_af = base_alloc - base_free;
 117         cur_af = total_alloc - total_free;
 118         grow_af = cur_af - base_af;
 119         grow_acc = accounted - base_accounted;
 120         log_info("Leakage: %d leaked. growth: %u use, %u acc, %u heap",
 121                 (int)(grow_af - grow_acc), (unsigned)grow_af,
 122                 (unsigned)grow_acc, (unsigned)(heap - base_heap));
 123 }
 124
 125 /** give debug heap size indication */
 126 static void
 127 debug_total_mem(size_t calctotal)
 128 {
 129 #ifdef HAVE_SBRK
 130         extern void* unbound_start_brk;
 131         extern size_t unbound_mem_alloc, unbound_mem_freed;
 132         void* cur = sbrk(0);
 133         int total = cur-unbound_start_brk;
 134         log_info("Total heap memory estimate: %u  total-alloc: %u  "
 135                 "total-free: %u", (unsigned)total,
 136                 (unsigned)unbound_mem_alloc, (unsigned)unbound_mem_freed);
 137         debug_memleak(calctotal, (size_t)total,
 138                 unbound_mem_alloc, unbound_mem_freed);
 139 #else
 140         (void)calctotal;
 141 #endif /* HAVE_SBRK */
 142 }
 143 #endif /* UNBOUND_ALLOC_STATS */
 144
 145 /** Report on memory usage by this thread and global */
 146 static void
 147 worker_mem_report(struct worker* ATTR_UNUSED(worker),
 148         struct serviced_query* ATTR_UNUSED(cur_serv))
 149 {
 150 #ifdef UNBOUND_ALLOC_STATS
 151         /* debug func in validator module */
 152         size_t total, front, back, mesh, msg, rrset, infra, ac, superac;
 153         size_t me, iter, val, anch;
 154         int i;
 155         if(verbosity < VERB_ALGO)
 156                 return;
 157         front = listen_get_mem(worker->front);
 158         back = outnet_get_mem(worker->back);
 159         msg = slabhash_get_mem(worker->env.msg_cache);
 160         rrset = slabhash_get_mem(&worker->env.rrset_cache->table);
 161         infra = infra_get_mem(worker->env.infra_cache);
 162         mesh = mesh_get_mem(worker->env.mesh);
 163         ac = alloc_get_mem(&worker->alloc);
 164         superac = alloc_get_mem(&worker->daemon->superalloc);
 165         anch = anchors_get_mem(worker->env.anchors);
 166         iter = 0;
 167         val = 0;
 168         for(i=0; i<worker->env.mesh->mods.num; i++) {
 169                 fptr_ok(fptr_whitelist_mod_get_mem(worker->env.mesh->
 170                         mods.mod[i]->get_mem));
 171                 if(strcmp(worker->env.mesh->mods.mod[i]->name, "validator")==0)
 172                         val += (*worker->env.mesh->mods.mod[i]->get_mem)
 173                                 (&worker->env, i);
 174                 else    iter += (*worker->env.mesh->mods.mod[i]->get_mem)
 175                                 (&worker->env, i);
 176         }
 177         me = sizeof(*worker) + sizeof(*worker->base) + sizeof(*worker->comsig)
 178                 + comm_point_get_mem(worker->cmd_com)
 179                 + sizeof(worker->rndstate)
 180                 + regional_get_mem(worker->scratchpad)
 181                 + sizeof(*worker->env.scratch_buffer)
 182                 + ldns_buffer_capacity(worker->env.scratch_buffer)
 183                 + forwards_get_mem(worker->env.fwds)
 184                 + hints_get_mem(worker->env.hints);
 185         if(worker->thread_num == 0)
 186                 me += acl_list_get_mem(worker->daemon->acl);
 187         if(cur_serv) {
 188                 me += serviced_get_mem(cur_serv);
 189         }
 190         total = front+back+mesh+msg+rrset+infra+iter+val+ac+superac+me;
 191         log_info("Memory conditions: %u front=%u back=%u mesh=%u msg=%u "
 192                 "rrset=%u infra=%u iter=%u val=%u anchors=%u "
 193                 "alloccache=%u globalalloccache=%u me=%u",
 194                 (unsigned)total, (unsigned)front, (unsigned)back,
 195                 (unsigned)mesh, (unsigned)msg, (unsigned)rrset,
 196                 (unsigned)infra, (unsigned)iter, (unsigned)val, (unsigned)anch,
 197                 (unsigned)ac, (unsigned)superac, (unsigned)me);
 198         debug_total_mem(total);
 199 #else /* no UNBOUND_ALLOC_STATS */
 200         size_t val = 0;
 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         }
 211         verbose(VERB_QUERY, "cache memory msg=%u rrset=%u infra=%u val=%u",
 212                 (unsigned)slabhash_get_mem(worker->env.msg_cache),
 213                 (unsigned)slabhash_get_mem(&worker->env.rrset_cache->table),
 214                 (unsigned)infra_get_mem(worker->env.infra_cache),
 215                 (unsigned)val);
 216 #endif /* UNBOUND_ALLOC_STATS */
 217 }
 218
 219 void
 220 worker_send_cmd(struct worker* worker, enum worker_commands cmd)
 221 {
 222         uint32_t c = (uint32_t)htonl(cmd);
 223         if(!tube_write_msg(worker->cmd, (uint8_t*)&c, sizeof(c), 0)) {
 224                 log_err("worker send cmd %d failed", (int)cmd);
 225         }
 226 }
 227
 228 int
 229 worker_handle_reply(struct comm_point* c, void* arg, int error,
 230         struct comm_reply* reply_info)
 231 {
 232         struct module_qstate* q = (struct module_qstate*)arg;
 233         struct worker* worker = q->env->worker;
 234         struct outbound_entry e;
 235         e.qstate = q;
 236         e.qsent = NULL;
 237
 238         if(error != 0) {
 239                 mesh_report_reply(worker->env.mesh, &e, reply_info, error);
 240                 worker_mem_report(worker, NULL);
 241                 return 0;
 242         }
 243         /* sanity check. */
 244         if(!LDNS_QR_WIRE(ldns_buffer_begin(c->buffer))
 245                 || LDNS_OPCODE_WIRE(ldns_buffer_begin(c->buffer)) !=
 246                         LDNS_PACKET_QUERY
 247                 || LDNS_QDCOUNT(ldns_buffer_begin(c->buffer)) > 1) {
 248                 /* error becomes timeout for the module as if this reply
 249                  * never arrived. */
 250                 mesh_report_reply(worker->env.mesh, &e, reply_info,
 251                         NETEVENT_TIMEOUT);
 252                 worker_mem_report(worker, NULL);
 253                 return 0;
 254         }
 255         mesh_report_reply(worker->env.mesh, &e, reply_info, NETEVENT_NOERROR);
 256         worker_mem_report(worker, NULL);
 257         return 0;
 258 }
 259
 260 int
 261 worker_handle_service_reply(struct comm_point* c, void* arg, int error,
 262         struct comm_reply* reply_info)
 263 {
 264         struct outbound_entry* e = (struct outbound_entry*)arg;
 265         struct worker* worker = e->qstate->env->worker;
 266         struct serviced_query *sq = e->qsent;
 267
 268         verbose(VERB_ALGO, "worker svcd callback for qstate %p", e->qstate);
 269         if(error != 0) {
 270                 mesh_report_reply(worker->env.mesh, e, reply_info, error);
 271                 worker_mem_report(worker, sq);
 272                 return 0;
 273         }
 274         /* sanity check. */
 275         if(!LDNS_QR_WIRE(ldns_buffer_begin(c->buffer))
 276                 || LDNS_OPCODE_WIRE(ldns_buffer_begin(c->buffer)) !=
 277                         LDNS_PACKET_QUERY
 278                 || LDNS_QDCOUNT(ldns_buffer_begin(c->buffer)) > 1) {
 279                 /* error becomes timeout for the module as if this reply
 280                  * never arrived. */
 281                 verbose(VERB_ALGO, "worker: bad reply handled as timeout");
 282                 mesh_report_reply(worker->env.mesh, e, reply_info,
 283                         NETEVENT_TIMEOUT);
 284                 worker_mem_report(worker, sq);
 285                 return 0;
 286         }
 287         mesh_report_reply(worker->env.mesh, e, reply_info, NETEVENT_NOERROR);
 288         worker_mem_report(worker, sq);
 289         return 0;
 290 }
 291
 292 /** check request sanity.
 293  * @param pkt: the wire packet to examine for sanity.
 294  * @param worker: parameters for checking.
 295  * @return error code, 0 OK, or -1 discard.
 296 */
 297 static int
 298 worker_check_request(ldns_buffer* pkt, struct worker* worker)
 299 {
 300         if(ldns_buffer_limit(pkt) < LDNS_HEADER_SIZE) {
 301                 verbose(VERB_QUERY, "request too short, discarded");
 302                 return -1;
 303         }
 304         if(ldns_buffer_limit(pkt) > NORMAL_UDP_SIZE &&
 305                 worker->daemon->cfg->harden_large_queries) {
 306                 verbose(VERB_QUERY, "request too large, discarded");
 307                 return -1;
 308         }
 309         if(LDNS_QR_WIRE(ldns_buffer_begin(pkt))) {
 310                 verbose(VERB_QUERY, "request has QR bit on, discarded");
 311                 return -1;
 312         }
 313         if(LDNS_TC_WIRE(ldns_buffer_begin(pkt))) {
 314                 LDNS_TC_CLR(ldns_buffer_begin(pkt));
 315                 verbose(VERB_QUERY, "request bad, has TC bit on");
 316                 return LDNS_RCODE_FORMERR;
 317         }
 318         if(LDNS_OPCODE_WIRE(ldns_buffer_begin(pkt)) != LDNS_PACKET_QUERY) {
 319                 verbose(VERB_QUERY, "request unknown opcode %d",
 320                         LDNS_OPCODE_WIRE(ldns_buffer_begin(pkt)));
 321                 return LDNS_RCODE_NOTIMPL;
 322         }
 323         if(LDNS_QDCOUNT(ldns_buffer_begin(pkt)) != 1) {
 324                 verbose(VERB_QUERY, "request wrong nr qd=%d",
 325                         LDNS_QDCOUNT(ldns_buffer_begin(pkt)));
 326                 return LDNS_RCODE_FORMERR;
 327         }
 328         if(LDNS_ANCOUNT(ldns_buffer_begin(pkt)) != 0) {
 329                 verbose(VERB_QUERY, "request wrong nr an=%d",
 330                         LDNS_ANCOUNT(ldns_buffer_begin(pkt)));
 331                 return LDNS_RCODE_FORMERR;
 332         }
 333         if(LDNS_NSCOUNT(ldns_buffer_begin(pkt)) != 0) {
 334                 verbose(VERB_QUERY, "request wrong nr ns=%d",
 335                         LDNS_NSCOUNT(ldns_buffer_begin(pkt)));
 336                 return LDNS_RCODE_FORMERR;
 337         }
 338         if(LDNS_ARCOUNT(ldns_buffer_begin(pkt)) > 1) {
 339                 verbose(VERB_QUERY, "request wrong nr ar=%d",
 340                         LDNS_ARCOUNT(ldns_buffer_begin(pkt)));
 341                 return LDNS_RCODE_FORMERR;
 342         }
 343         return 0;
 344 }
 345
 346 void
 347 worker_handle_control_cmd(struct tube* ATTR_UNUSED(tube), uint8_t* msg,
 348         size_t len, int error, void* arg)
 349 {
 350         struct worker* worker = (struct worker*)arg;
 351         enum worker_commands cmd;
 352         if(error != NETEVENT_NOERROR) {
 353                 free(msg);
 354                 if(error == NETEVENT_CLOSED)
 355                         comm_base_exit(worker->base);
 356                 else    log_info("control event: %d", error);
 357                 return;
 358         }
 359         if(len != sizeof(uint32_t)) {
 360                 fatal_exit("bad control msg length %d", (int)len);
 361         }
 362         cmd = ldns_read_uint32(msg);
 363         free(msg);
 364         switch(cmd) {
 365         case worker_cmd_quit:
 366                 verbose(VERB_ALGO, "got control cmd quit");
 367                 comm_base_exit(worker->base);
 368                 break;
 369         case worker_cmd_stats:
 370                 verbose(VERB_ALGO, "got control cmd stats");
 371                 server_stats_reply(worker, 1);
 372                 break;
 373         case worker_cmd_stats_noreset:
 374                 verbose(VERB_ALGO, "got control cmd stats_noreset");
 375                 server_stats_reply(worker, 0);
 376                 break;
 377         case worker_cmd_remote:
 378                 verbose(VERB_ALGO, "got control cmd remote");
 379                 daemon_remote_exec(worker);
 380                 break;
 381         default:
 382                 log_err("bad command %d", (int)cmd);
 383                 break;
 384         }
 385 }
 386
 387 /** check if a delegation is secure */
 388 static enum sec_status
 389 check_delegation_secure(struct reply_info *rep)
 390 {
 391         /* return smallest security status */
 392         size_t i;
 393         enum sec_status sec = sec_status_secure;
 394         enum sec_status s;
 395         size_t num = rep->an_numrrsets + rep->ns_numrrsets;
 396         /* check if answer and authority are OK */
 397         for(i=0; i<num; i++) {
 398                 s = ((struct packed_rrset_data*)rep->rrsets[i]->entry.data)
 399                         ->security;
 400                 if(s < sec)
 401                         sec = s;
 402         }
 403         /* in additional, only unchecked triggers revalidation */
 404         for(i=num; i<rep->rrset_count; i++) {
 405                 s = ((struct packed_rrset_data*)rep->rrsets[i]->entry.data)
 406                         ->security;
 407                 if(s == sec_status_unchecked)
 408                         return s;
 409         }
 410         return sec;
 411 }
 412
 413 /** remove nonsecure from a delegation referral additional section */
 414 static void
 415 deleg_remove_nonsecure_additional(struct reply_info* rep)
 416 {
 417         /* we can simply edit it, since we are working in the scratch region */
 418         size_t i;
 419         enum sec_status s;
 420
 421         for(i = rep->an_numrrsets+rep->ns_numrrsets; i<rep->rrset_count; i++) {
 422                 s = ((struct packed_rrset_data*)rep->rrsets[i]->entry.data)
 423                         ->security;
 424                 if(s != sec_status_secure) {
 425                         memmove(rep->rrsets+i, rep->rrsets+i+1,
 426                                 sizeof(struct ub_packed_rrset_key*)*
 427                                 (rep->rrset_count - i - 1));
 428                         rep->ar_numrrsets--;
 429                         rep->rrset_count--;
 430                         i--;
 431                 }
 432         }
 433 }
 434
 435 /** answer nonrecursive query from the cache */
 436 static int
 437 answer_norec_from_cache(struct worker* worker, struct query_info* qinfo,
 438         uint16_t id, uint16_t flags, struct comm_reply* repinfo,
 439         struct edns_data* edns)
 440 {
 441         /* for a nonrecursive query return either:
 442          *      o an error (servfail; we try to avoid this)
 443          *      o a delegation (closest we have; this routine tries that)
 444          *      o the answer (checked by answer_from_cache)
 445          *
 446          * So, grab a delegation from the rrset cache.
 447          * Then check if it needs validation, if so, this routine fails,
 448          * so that iterator can prime and validator can verify rrsets.
 449          */
 450         uint16_t udpsize = edns->udp_size;
 451         int secure = 0;
 452         time_t timenow = *worker->env.now;
 453         int must_validate = (!(flags&BIT_CD) || worker->env.cfg->ignore_cd)
 454                 && worker->env.need_to_validate;
 455         struct dns_msg *msg = NULL;
 456         struct delegpt *dp;
 457
 458         dp = dns_cache_find_delegation(&worker->env, qinfo->qname,
 459                 qinfo->qname_len, qinfo->qtype, qinfo->qclass,
 460                 worker->scratchpad, &msg, timenow);
 461         if(!dp) { /* no delegation, need to reprime */
 462                 regional_free_all(worker->scratchpad);
 463                 return 0;
 464         }
 465         if(must_validate) {
 466                 switch(check_delegation_secure(msg->rep)) {
 467                 case sec_status_unchecked:
 468                         /* some rrsets have not been verified yet, go and
 469                          * let validator do that */
 470                         regional_free_all(worker->scratchpad);
 471                         return 0;
 472                 case sec_status_bogus:
 473                         /* some rrsets are bogus, reply servfail */
 474                         edns->edns_version = EDNS_ADVERTISED_VERSION;
 475                         edns->udp_size = EDNS_ADVERTISED_SIZE;
 476                         edns->ext_rcode = 0;
 477                         edns->bits &= EDNS_DO;
 478                         error_encode(repinfo->c->buffer, LDNS_RCODE_SERVFAIL,
 479                                 &msg->qinfo, id, flags, edns);
 480                         regional_free_all(worker->scratchpad);
 481                         if(worker->stats.extended) {
 482                                 worker->stats.ans_bogus++;
 483                                 worker->stats.ans_rcode[LDNS_RCODE_SERVFAIL]++;
 484                         }
 485                         return 1;
 486                 case sec_status_secure:
 487                         /* all rrsets are secure */
 488                         /* remove non-secure rrsets from the add. section*/
 489                         if(worker->env.cfg->val_clean_additional)
 490                                 deleg_remove_nonsecure_additional(msg->rep);
 491                         secure = 1;
 492                         break;
 493                 case sec_status_indeterminate:
 494                 case sec_status_insecure:
 495                 default:
 496                         /* not secure */
 497                         secure = 0;
 498                         break;
 499                 }
 500         }
 501         /* return this delegation from the cache */
 502         edns->edns_version = EDNS_ADVERTISED_VERSION;
 503         edns->udp_size = EDNS_ADVERTISED_SIZE;
 504         edns->ext_rcode = 0;
 505         edns->bits &= EDNS_DO;
 506         msg->rep->flags |= BIT_QR|BIT_RA;
 507         if(!reply_info_answer_encode(&msg->qinfo, msg->rep, id, flags,
 508                 repinfo->c->buffer, 0, 1, worker->scratchpad,
 509                 udpsize, edns, (int)(edns->bits & EDNS_DO), secure)) {
 510                 error_encode(repinfo->c->buffer, LDNS_RCODE_SERVFAIL,
 511                         &msg->qinfo, id, flags, edns);
 512         }
 513         regional_free_all(worker->scratchpad);
 514         if(worker->stats.extended) {
 515                 if(secure) worker->stats.ans_secure++;
 516                 server_stats_insrcode(&worker->stats, repinfo->c->buffer);
 517         }
 518         return 1;
 519 }
 520
 521 /** answer query from the cache */
 522 static int
 523 answer_from_cache(struct worker* worker, struct query_info* qinfo,
 524         struct reply_info* rep, uint16_t id, uint16_t flags,
 525         struct comm_reply* repinfo, struct edns_data* edns)
 526 {
 527         time_t timenow = *worker->env.now;
 528         uint16_t udpsize = edns->udp_size;
 529         int secure;
 530         int must_validate = (!(flags&BIT_CD) || worker->env.cfg->ignore_cd)
 531                 && worker->env.need_to_validate;
 532         /* see if it is possible */
 533         if(rep->ttl < timenow) {
 534                 /* the rrsets may have been updated in the meantime.
 535                  * we will refetch the message format from the
 536                  * authoritative server
 537                  */
 538                 return 0;
 539         }
 540         if(!rrset_array_lock(rep->ref, rep->rrset_count, timenow))
 541                 return 0;
 542         /* locked and ids and ttls are OK. */
 543         /* check CNAME chain (if any) */
 544         if(rep->an_numrrsets > 0 && (rep->rrsets[0]->rk.type ==
 545                 htons(LDNS_RR_TYPE_CNAME) || rep->rrsets[0]->rk.type ==
 546                 htons(LDNS_RR_TYPE_DNAME))) {
 547                 if(!reply_check_cname_chain(rep)) {
 548                         /* cname chain invalid, redo iterator steps */
 549                         verbose(VERB_ALGO, "Cache reply: cname chain broken");
 550                 bail_out:
 551                         rrset_array_unlock_touch(worker->env.rrset_cache,
 552                                 worker->scratchpad, rep->ref, rep->rrset_count);
 553                         regional_free_all(worker->scratchpad);
 554                         return 0;
 555                 }
 556         }
 557         /* check security status of the cached answer */
 558         if( rep->security == sec_status_bogus && must_validate) {
 559                 /* BAD cached */
 560                 edns->edns_version = EDNS_ADVERTISED_VERSION;
 561                 edns->udp_size = EDNS_ADVERTISED_SIZE;
 562                 edns->ext_rcode = 0;
 563                 edns->bits &= EDNS_DO;
 564                 error_encode(repinfo->c->buffer, LDNS_RCODE_SERVFAIL,
 565                         qinfo, id, flags, edns);
 566                 rrset_array_unlock_touch(worker->env.rrset_cache,
 567                         worker->scratchpad, rep->ref, rep->rrset_count);
 568                 regional_free_all(worker->scratchpad);
 569                 if(worker->stats.extended) {
 570                         worker->stats.ans_bogus ++;
 571                         worker->stats.ans_rcode[LDNS_RCODE_SERVFAIL] ++;
 572                 }
 573                 return 1;
 574         } else if( rep->security == sec_status_unchecked && must_validate) {
 575                 verbose(VERB_ALGO, "Cache reply: unchecked entry needs "
 576                         "validation");
 577                 goto bail_out; /* need to validate cache entry first */
 578         } else if(rep->security == sec_status_secure) {
 579                 if(reply_all_rrsets_secure(rep))
 580                         secure = 1;
 581                 else    {
 582                         if(must_validate) {
 583                                 verbose(VERB_ALGO, "Cache reply: secure entry"
 584                                         " changed status");
 585                                 goto bail_out; /* rrset changed, re-verify */
 586                         }
 587                         secure = 0;
 588                 }
 589         } else  secure = 0;
 590
 591         edns->edns_version = EDNS_ADVERTISED_VERSION;
 592         edns->udp_size = EDNS_ADVERTISED_SIZE;
 593         edns->ext_rcode = 0;
 594         edns->bits &= EDNS_DO;
 595         if(!reply_info_answer_encode(qinfo, rep, id, flags,
 596                 repinfo->c->buffer, timenow, 1, worker->scratchpad,
 597                 udpsize, edns, (int)(edns->bits & EDNS_DO), secure)) {
 598                 error_encode(repinfo->c->buffer, LDNS_RCODE_SERVFAIL,
 599                         qinfo, id, flags, edns);
 600         }
 601         /* cannot send the reply right now, because blocking network syscall
 602          * is bad while holding locks. */
 603         rrset_array_unlock_touch(worker->env.rrset_cache, worker->scratchpad,
 604                 rep->ref, rep->rrset_count);
 605         regional_free_all(worker->scratchpad);
 606         if(worker->stats.extended) {
 607                 if(secure) worker->stats.ans_secure++;
 608                 server_stats_insrcode(&worker->stats, repinfo->c->buffer);
 609         }
 610         /* go and return this buffer to the client */
 611         return 1;
 612 }
 613
 614 /** Reply to client and perform prefetch to keep cache up to date */
 615 static void
 616 reply_and_prefetch(struct worker* worker, struct query_info* qinfo,
 617         uint16_t flags, struct comm_reply* repinfo, time_t leeway)
 618 {
 619         /* first send answer to client to keep its latency
 620          * as small as a cachereply */
 621         comm_point_send_reply(repinfo);
 622         server_stats_prefetch(&worker->stats, worker);
 623
 624         /* create the prefetch in the mesh as a normal lookup without
 625          * client addrs waiting, which has the cache blacklisted (to bypass
 626          * the cache and go to the network for the data). */
 627         /* this (potentially) runs the mesh for the new query */
 628         mesh_new_prefetch(worker->env.mesh, qinfo, flags, leeway +
 629                 PREFETCH_EXPIRY_ADD);
 630 }
 631
 632 /**
 633  * Fill CH class answer into buffer. Keeps query.
 634  * @param pkt: buffer
 635  * @param str: string to put into text record (<255).
 636  * @param edns: edns reply information.
 637  */
 638 static void
 639 chaos_replystr(ldns_buffer* pkt, const char* str, struct edns_data* edns)
 640 {
 641         size_t len = strlen(str);
 642         unsigned int rd = LDNS_RD_WIRE(ldns_buffer_begin(pkt));
 643         unsigned int cd = LDNS_CD_WIRE(ldns_buffer_begin(pkt));
 644         if(len>255) len=255; /* cap size of TXT record */
 645         ldns_buffer_clear(pkt);
 646         ldns_buffer_skip(pkt, (ssize_t)sizeof(uint16_t)); /* skip id */
 647         ldns_buffer_write_u16(pkt, (uint16_t)(BIT_QR|BIT_RA));
 648         if(rd) LDNS_RD_SET(ldns_buffer_begin(pkt));
 649         if(cd) LDNS_CD_SET(ldns_buffer_begin(pkt));
 650         ldns_buffer_write_u16(pkt, 1); /* qdcount */
 651         ldns_buffer_write_u16(pkt, 1); /* ancount */
 652         ldns_buffer_write_u16(pkt, 0); /* nscount */
 653         ldns_buffer_write_u16(pkt, 0); /* arcount */
 654         (void)query_dname_len(pkt); /* skip qname */
 655         ldns_buffer_skip(pkt, (ssize_t)sizeof(uint16_t)); /* skip qtype */
 656         ldns_buffer_skip(pkt, (ssize_t)sizeof(uint16_t)); /* skip qclass */
 657         ldns_buffer_write_u16(pkt, 0xc00c); /* compr ptr to query */
 658         ldns_buffer_write_u16(pkt, LDNS_RR_TYPE_TXT);
 659         ldns_buffer_write_u16(pkt, LDNS_RR_CLASS_CH);
 660         ldns_buffer_write_u32(pkt, 0); /* TTL */
 661         ldns_buffer_write_u16(pkt, sizeof(uint8_t) + len);
 662         ldns_buffer_write_u8(pkt, len);
 663         ldns_buffer_write(pkt, str, len);
 664         ldns_buffer_flip(pkt);
 665         edns->edns_version = EDNS_ADVERTISED_VERSION;
 666         edns->udp_size = EDNS_ADVERTISED_SIZE;
 667         edns->bits &= EDNS_DO;
 668         attach_edns_record(pkt, edns);
 669 }
 670
 671 /**
 672  * Answer CH class queries.
 673  * @param w: worker
 674  * @param qinfo: query info. Pointer into packet buffer.
 675  * @param edns: edns info from query.
 676  * @param pkt: packet buffer.
 677  * @return: true if a reply is to be sent.
 678  */
 679 static int
 680 answer_chaos(struct worker* w, struct query_info* qinfo,
 681         struct edns_data* edns, ldns_buffer* pkt)
 682 {
 683         struct config_file* cfg = w->env.cfg;
 684         if(qinfo->qtype != LDNS_RR_TYPE_ANY && qinfo->qtype != LDNS_RR_TYPE_TXT)
 685                 return 0;
 686         if(query_dname_compare(qinfo->qname,
 687                 (uint8_t*)"\002id\006server") == 0 ||
 688                 query_dname_compare(qinfo->qname,
 689                 (uint8_t*)"\010hostname\004bind") == 0)
 690         {
 691                 if(cfg->hide_identity)
 692                         return 0;
 693                 if(cfg->identity==NULL || cfg->identity[0]==0) {
 694                         char buf[MAXHOSTNAMELEN+1];
 695                         if (gethostname(buf, MAXHOSTNAMELEN) == 0) {
 696                                 buf[MAXHOSTNAMELEN] = 0;
 697                                 chaos_replystr(pkt, buf, edns);
 698                         } else  {
 699                                 log_err("gethostname: %s", strerror(errno));
 700                                 chaos_replystr(pkt, "no hostname", edns);
 701                         }
 702                 }
 703                 else    chaos_replystr(pkt, cfg->identity, edns);
 704                 return 1;
 705         }
 706         if(query_dname_compare(qinfo->qname,
 707                 (uint8_t*)"\007version\006server") == 0 ||
 708                 query_dname_compare(qinfo->qname,
 709                 (uint8_t*)"\007version\004bind") == 0)
 710         {
 711                 if(cfg->hide_version)
 712                         return 0;
 713                 if(cfg->version==NULL || cfg->version[0]==0)
 714                         chaos_replystr(pkt, PACKAGE_STRING, edns);
 715                 else    chaos_replystr(pkt, cfg->version, edns);
 716                 return 1;
 717         }
 718         return 0;
 719 }
 720
 721 int
 722 worker_handle_request(struct comm_point* c, void* arg, int error,
 723         struct comm_reply* repinfo)
 724 {
 725         struct worker* worker = (struct worker*)arg;
 726         int ret;
 727         hashvalue_t h;
 728         struct lruhash_entry* e;
 729         struct query_info qinfo;
 730         struct edns_data edns;
 731         enum acl_access acl;
 732
 733         if(error != NETEVENT_NOERROR) {
 734                 /* some bad tcp query DNS formats give these error calls */
 735                 verbose(VERB_ALGO, "handle request called with err=%d", error);
 736                 return 0;
 737         }
 738         acl = acl_list_lookup(worker->daemon->acl, &repinfo->addr,
 739                 repinfo->addrlen);
 740         if(acl == acl_deny) {
 741                 comm_point_drop_reply(repinfo);
 742                 if(worker->stats.extended)
 743                         worker->stats.unwanted_queries++;
 744                 return 0;
 745         } else if(acl == acl_refuse) {
 746                 log_addr(VERB_ALGO, "refused query from",
 747                         &repinfo->addr, repinfo->addrlen);
 748                 log_buf(VERB_ALGO, "refuse", c->buffer);
 749                 if(worker->stats.extended)
 750                         worker->stats.unwanted_queries++;
 751                 if(worker_check_request(c->buffer, worker) == -1) {
 752                         comm_point_drop_reply(repinfo);
 753                         return 0; /* discard this */
 754                 }
 755                 ldns_buffer_set_limit(c->buffer, LDNS_HEADER_SIZE);
 756                 ldns_buffer_write_at(c->buffer, 4,
 757                         (uint8_t*)"\0\0\0\0\0\0\0\0", 8);
 758                 LDNS_QR_SET(ldns_buffer_begin(c->buffer));
 759                 LDNS_RCODE_SET(ldns_buffer_begin(c->buffer),
 760                         LDNS_RCODE_REFUSED);
 761                 return 1;
 762         }
 763         if((ret=worker_check_request(c->buffer, worker)) != 0) {
 764                 verbose(VERB_ALGO, "worker check request: bad query.");
 765                 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen);
 766                 if(ret != -1) {
 767                         LDNS_QR_SET(ldns_buffer_begin(c->buffer));
 768                         LDNS_RCODE_SET(ldns_buffer_begin(c->buffer), ret);
 769                         return 1;
 770                 }
 771                 comm_point_drop_reply(repinfo);
 772                 return 0;
 773         }
 774         worker->stats.num_queries++;
 775         /* see if query is in the cache */
 776         if(!query_info_parse(&qinfo, c->buffer)) {
 777                 verbose(VERB_ALGO, "worker parse request: formerror.");
 778                 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen);
 779                 ldns_buffer_rewind(c->buffer);
 780                 LDNS_QR_SET(ldns_buffer_begin(c->buffer));
 781                 LDNS_RCODE_SET(ldns_buffer_begin(c->buffer),
 782                         LDNS_RCODE_FORMERR);
 783                 server_stats_insrcode(&worker->stats, c->buffer);
 784                 return 1;
 785         }
 786         if(worker->env.cfg->log_queries) {
 787                 char ip[128];
 788                 addr_to_str(&repinfo->addr, repinfo->addrlen, ip, sizeof(ip));
 789                 log_nametypeclass(0, ip, qinfo.qname, qinfo.qtype, qinfo.qclass);
 790         }
 791         if(qinfo.qtype == LDNS_RR_TYPE_AXFR ||
 792                 qinfo.qtype == LDNS_RR_TYPE_IXFR) {
 793                 verbose(VERB_ALGO, "worker request: refused zone transfer.");
 794                 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen);
 795                 ldns_buffer_rewind(c->buffer);
 796                 LDNS_QR_SET(ldns_buffer_begin(c->buffer));
 797                 LDNS_RCODE_SET(ldns_buffer_begin(c->buffer),
 798                         LDNS_RCODE_REFUSED);
 799                 if(worker->stats.extended) {
 800                         worker->stats.qtype[qinfo.qtype]++;
 801                         server_stats_insrcode(&worker->stats, c->buffer);
 802                 }
 803                 return 1;
 804         }
 805         if((ret=parse_edns_from_pkt(c->buffer, &edns)) != 0) {
 806                 verbose(VERB_ALGO, "worker parse edns: formerror.");
 807                 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen);
 808                 ldns_buffer_rewind(c->buffer);
 809                 LDNS_QR_SET(ldns_buffer_begin(c->buffer));
 810                 LDNS_RCODE_SET(ldns_buffer_begin(c->buffer), ret);
 811                 server_stats_insrcode(&worker->stats, c->buffer);
 812                 return 1;
 813         }
 814         if(edns.edns_present && edns.edns_version != 0) {
 815                 edns.ext_rcode = (uint8_t)(EDNS_RCODE_BADVERS>>4);
 816                 edns.edns_version = EDNS_ADVERTISED_VERSION;
 817                 edns.udp_size = EDNS_ADVERTISED_SIZE;
 818                 edns.bits &= EDNS_DO;
 819                 verbose(VERB_ALGO, "query with bad edns version.");
 820                 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen);
 821                 error_encode(c->buffer, EDNS_RCODE_BADVERS&0xf, &qinfo,
 822                         *(uint16_t*)ldns_buffer_begin(c->buffer),
 823                         ldns_buffer_read_u16_at(c->buffer, 2), NULL);
 824                 attach_edns_record(c->buffer, &edns);
 825                 return 1;
 826         }
 827         if(edns.edns_present && edns.udp_size < NORMAL_UDP_SIZE &&
 828                 worker->daemon->cfg->harden_short_bufsize) {
 829                 verbose(VERB_QUERY, "worker request: EDNS bufsize %d ignored",
 830                         (int)edns.udp_size);
 831                 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen);
 832                 edns.udp_size = NORMAL_UDP_SIZE;
 833         }
 834         if(edns.udp_size > worker->daemon->cfg->max_udp_size &&
 835                 c->type == comm_udp) {
 836                 verbose(VERB_QUERY,
 837                         "worker request: max UDP reply size modified"
 838                         " (%d to max-udp-size)", (int)edns.udp_size);
 839                 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen);
 840                 edns.udp_size = worker->daemon->cfg->max_udp_size;
 841         }
 842         if(edns.udp_size < LDNS_HEADER_SIZE) {
 843                 verbose(VERB_ALGO, "worker request: edns is too small.");
 844                 log_addr(VERB_CLIENT, "from", &repinfo->addr, repinfo->addrlen);
 845                 LDNS_QR_SET(ldns_buffer_begin(c->buffer));
 846                 LDNS_TC_SET(ldns_buffer_begin(c->buffer));
 847                 LDNS_RCODE_SET(ldns_buffer_begin(c->buffer),
 848                         LDNS_RCODE_SERVFAIL);
 849                 ldns_buffer_set_position(c->buffer, LDNS_HEADER_SIZE);
 850                 ldns_buffer_write_at(c->buffer, 4,
 851                         (uint8_t*)"\0\0\0\0\0\0\0\0", 8);
 852                 ldns_buffer_flip(c->buffer);
 853                 return 1;
 854         }
 855         if(worker->stats.extended)
 856                 server_stats_insquery(&worker->stats, c, qinfo.qtype,
 857                         qinfo.qclass, &edns, repinfo);
 858         if(c->type != comm_udp)
 859                 edns.udp_size = 65535; /* max size for TCP replies */
 860         if(qinfo.qclass == LDNS_RR_CLASS_CH && answer_chaos(worker, &qinfo,
 861                 &edns, c->buffer)) {
 862                 server_stats_insrcode(&worker->stats, c->buffer);
 863                 return 1;
 864         }
 865         if(local_zones_answer(worker->daemon->local_zones, &qinfo, &edns,
 866                 c->buffer, worker->scratchpad)) {
 867                 regional_free_all(worker->scratchpad);
 868                 if(ldns_buffer_limit(c->buffer) == 0) {
 869                         comm_point_drop_reply(repinfo);
 870                         return 0;
 871                 }
 872                 server_stats_insrcode(&worker->stats, c->buffer);
 873                 return 1;
 874         }
 875         if(!(LDNS_RD_WIRE(ldns_buffer_begin(c->buffer))) &&
 876                 acl != acl_allow_snoop ) {
 877                 ldns_buffer_set_limit(c->buffer, LDNS_HEADER_SIZE);
 878                 ldns_buffer_write_at(c->buffer, 4,
 879                         (uint8_t*)"\0\0\0\0\0\0\0\0", 8);
 880                 LDNS_QR_SET(ldns_buffer_begin(c->buffer));
 881                 LDNS_RCODE_SET(ldns_buffer_begin(c->buffer),
 882                         LDNS_RCODE_REFUSED);
 883                 ldns_buffer_flip(c->buffer);
 884                 server_stats_insrcode(&worker->stats, c->buffer);
 885                 log_addr(VERB_ALGO, "refused nonrec (cache snoop) query from",
 886                         &repinfo->addr, repinfo->addrlen);
 887                 return 1;
 888         }
 889         h = query_info_hash(&qinfo);
 890         if((e=slabhash_lookup(worker->env.msg_cache, h, &qinfo, 0))) {
 891                 /* answer from cache - we have acquired a readlock on it */
 892                 if(answer_from_cache(worker, &qinfo,
 893                         (struct reply_info*)e->data,
 894                         *(uint16_t*)ldns_buffer_begin(c->buffer),
 895                         ldns_buffer_read_u16_at(c->buffer, 2), repinfo,
 896                         &edns)) {
 897                         /* prefetch it if the prefetch TTL expired */
 898                         if(worker->env.cfg->prefetch && *worker->env.now >=
 899                                 ((struct reply_info*)e->data)->prefetch_ttl) {
 900                                 time_t leeway = ((struct reply_info*)e->
 901                                         data)->ttl - *worker->env.now;
 902                                 lock_rw_unlock(&e->lock);
 903                                 reply_and_prefetch(worker, &qinfo,
 904                                         ldns_buffer_read_u16_at(c->buffer, 2),
 905                                         repinfo, leeway);
 906                                 return 0;
 907                         }
 908                         lock_rw_unlock(&e->lock);
 909                         return 1;
 910                 }
 911                 verbose(VERB_ALGO, "answer from the cache failed");
 912                 lock_rw_unlock(&e->lock);
 913         }
 914         if(!LDNS_RD_WIRE(ldns_buffer_begin(c->buffer))) {
 915                 if(answer_norec_from_cache(worker, &qinfo,
 916                         *(uint16_t*)ldns_buffer_begin(c->buffer),
 917                         ldns_buffer_read_u16_at(c->buffer, 2), repinfo,
 918                         &edns)) {
 919                         return 1;
 920                 }
 921                 verbose(VERB_ALGO, "answer norec from cache -- "
 922                         "need to validate or not primed");
 923         }
 924         ldns_buffer_rewind(c->buffer);
 925         server_stats_querymiss(&worker->stats, worker);
 926
 927         if(verbosity >= VERB_CLIENT) {
 928                 if(c->type == comm_udp)
 929                         log_addr(VERB_CLIENT, "udp request from",
 930                                 &repinfo->addr, repinfo->addrlen);
 931                 else    log_addr(VERB_CLIENT, "tcp request from",
 932                                 &repinfo->addr, repinfo->addrlen);
 933         }
 934
 935         /* grab a work request structure for this new request */
 936         mesh_new_client(worker->env.mesh, &qinfo,
 937                 ldns_buffer_read_u16_at(c->buffer, 2),
 938                 &edns, repinfo, *(uint16_t*)ldns_buffer_begin(c->buffer));
 939         worker_mem_report(worker, NULL);
 940         return 0;
 941 }
 942
 943 void
 944 worker_sighandler(int sig, void* arg)
 945 {
 946         /* note that log, print, syscalls here give race conditions. */
 947         /* we still print DETAIL logs, because this is extensive per message
 948          * logging anyway, and the operator may then have an interest
 949          * in the cause for unbound to exit */
 950         struct worker* worker = (struct worker*)arg;
 951         switch(sig) {
 952 #ifdef SIGHUP
 953                 case SIGHUP:
 954                         verbose(VERB_QUERY, "caught signal SIGHUP");
 955                         comm_base_exit(worker->base);
 956                         break;
 957 #endif
 958                 case SIGINT:
 959                         verbose(VERB_QUERY, "caught signal SIGINT");
 960                         worker->need_to_exit = 1;
 961                         comm_base_exit(worker->base);
 962                         break;
 963 #ifdef SIGQUIT
 964                 case SIGQUIT:
 965                         verbose(VERB_QUERY, "caught signal SIGQUIT");
 966                         worker->need_to_exit = 1;
 967                         comm_base_exit(worker->base);
 968                         break;
 969 #endif
 970                 case SIGTERM:
 971                         verbose(VERB_QUERY, "caught signal SIGTERM");
 972                         worker->need_to_exit = 1;
 973                         comm_base_exit(worker->base);
 974                         break;
 975                 default:
 976                         log_err("unknown signal: %d, ignored", sig);
 977                         break;
 978         }
 979 }
 980
 981 /** restart statistics timer for worker, if enabled */
 982 static void
 983 worker_restart_timer(struct worker* worker)
 984 {
 985         if(worker->env.cfg->stat_interval > 0) {
 986                 struct timeval tv;
 987 #ifndef S_SPLINT_S
 988                 tv.tv_sec = worker->env.cfg->stat_interval;
 989                 tv.tv_usec = 0;
 990 #endif
 991                 comm_timer_set(worker->stat_timer, &tv);
 992         }
 993 }
 994
 995 void worker_stat_timer_cb(void* arg)
 996 {
 997         struct worker* worker = (struct worker*)arg;
 998         server_stats_log(&worker->stats, worker, worker->thread_num);
 999         mesh_stats(worker->env.mesh, "mesh has");
1000         worker_mem_report(worker, NULL);
1001         if(!worker->daemon->cfg->stat_cumulative) {
1002                 worker_stats_clear(worker);
1003         }
1004         /* start next timer */
1005         worker_restart_timer(worker);
1006 }
1007
1008 void worker_probe_timer_cb(void* arg)
1009 {
1010         struct worker* worker = (struct worker*)arg;
1011         struct timeval tv;
1012 #ifndef S_SPLINT_S
1013         tv.tv_sec = (time_t)autr_probe_timer(&worker->env);
1014         tv.tv_usec = 0;
1015 #endif
1016         if(tv.tv_sec != 0)
1017                 comm_timer_set(worker->env.probe_timer, &tv);
1018 }
1019
1020 struct worker*
1021 worker_create(struct daemon* daemon, int id, int* ports, int n)
1022 {
1023         unsigned int seed;
1024         struct worker* worker = (struct worker*)calloc(1,
1025                 sizeof(struct worker));
1026         if(!worker)
1027                 return NULL;
1028         worker->numports = n;
1029         worker->ports = (int*)memdup(ports, sizeof(int)*n);
1030         if(!worker->ports) {
1031                 free(worker);
1032                 return NULL;
1033         }
1034         worker->daemon = daemon;
1035         worker->thread_num = id;
1036         if(!(worker->cmd = tube_create())) {
1037                 free(worker->ports);
1038                 free(worker);
1039                 return NULL;
1040         }
1041         /* create random state here to avoid locking trouble in RAND_bytes */
1042         seed = (unsigned int)time(NULL) ^ (unsigned int)getpid() ^
1043                 (((unsigned int)worker->thread_num)<<17);
1044                 /* shift thread_num so it does not match out pid bits */
1045         if(!(worker->rndstate = ub_initstate(seed, daemon->rand))) {
1046                 seed = 0;
1047                 log_err("could not init random numbers.");
1048                 tube_delete(worker->cmd);
1049                 free(worker->ports);
1050                 free(worker);
1051                 return NULL;
1052         }
1053         seed = 0;
1054         return worker;
1055 }
1056
1057 int
1058 worker_init(struct worker* worker, struct config_file *cfg,
1059         struct listen_port* ports, int do_sigs)
1060 {
1061         worker->need_to_exit = 0;
1062         worker->base = comm_base_create(do_sigs);
1063         if(!worker->base) {
1064                 log_err("could not create event handling base");
1065                 worker_delete(worker);
1066                 return 0;
1067         }
1068         comm_base_set_slow_accept_handlers(worker->base, &worker_stop_accept,
1069                 &worker_start_accept, worker);
1070         if(do_sigs) {
1071 #ifdef SIGHUP
1072                 ub_thread_sig_unblock(SIGHUP);
1073 #endif
1074                 ub_thread_sig_unblock(SIGINT);
1075 #ifdef SIGQUIT
1076                 ub_thread_sig_unblock(SIGQUIT);
1077 #endif
1078                 ub_thread_sig_unblock(SIGTERM);
1079 #ifndef LIBEVENT_SIGNAL_PROBLEM
1080                 worker->comsig = comm_signal_create(worker->base,
1081                         worker_sighandler, worker);
1082                 if(!worker->comsig
1083 #ifdef SIGHUP
1084                         || !comm_signal_bind(worker->comsig, SIGHUP)
1085 #endif
1086 #ifdef SIGQUIT
1087                         || !comm_signal_bind(worker->comsig, SIGQUIT)
1088 #endif
1089                         || !comm_signal_bind(worker->comsig, SIGTERM)
1090                         || !comm_signal_bind(worker->comsig, SIGINT)) {
1091                         log_err("could not create signal handlers");
1092                         worker_delete(worker);
1093                         return 0;
1094                 }
1095 #endif /* LIBEVENT_SIGNAL_PROBLEM */
1096                 if(!daemon_remote_open_accept(worker->daemon->rc,
1097                         worker->daemon->rc_ports, worker)) {
1098                         worker_delete(worker);
1099                         return 0;
1100                 }
1101 #ifdef UB_ON_WINDOWS
1102                 wsvc_setup_worker(worker);
1103 #endif /* UB_ON_WINDOWS */
1104         } else { /* !do_sigs */
1105                 worker->comsig = NULL;
1106         }
1107         worker->front = listen_create(worker->base, ports,
1108                 cfg->msg_buffer_size, (int)cfg->incoming_num_tcp,
1109                 worker->daemon->listen_sslctx, worker_handle_request, worker);
1110         if(!worker->front) {
1111                 log_err("could not create listening sockets");
1112                 worker_delete(worker);
1113                 return 0;
1114         }
1115         worker->back = outside_network_create(worker->base,
1116                 cfg->msg_buffer_size, (size_t)cfg->outgoing_num_ports,
1117                 cfg->out_ifs, cfg->num_out_ifs, cfg->do_ip4, cfg->do_ip6,
1118                 cfg->do_tcp?cfg->outgoing_num_tcp:0,
1119                 worker->daemon->env->infra_cache, worker->rndstate,
1120                 cfg->use_caps_bits_for_id, worker->ports, worker->numports,
1121                 cfg->unwanted_threshold, &worker_alloc_cleanup, worker,
1122                 cfg->do_udp, worker->daemon->connect_sslctx);
1123         if(!worker->back) {
1124                 log_err("could not create outgoing sockets");
1125                 worker_delete(worker);
1126                 return 0;
1127         }
1128         /* start listening to commands */
1129         if(!tube_setup_bg_listen(worker->cmd, worker->base,
1130                 &worker_handle_control_cmd, worker)) {
1131                 log_err("could not create control compt.");
1132                 worker_delete(worker);
1133                 return 0;
1134         }
1135         worker->stat_timer = comm_timer_create(worker->base,
1136                 worker_stat_timer_cb, worker);
1137         if(!worker->stat_timer) {
1138                 log_err("could not create statistics timer");
1139         }
1140
1141         /* we use the msg_buffer_size as a good estimate for what the
1142          * user wants for memory usage sizes */
1143         worker->scratchpad = regional_create_custom(cfg->msg_buffer_size);
1144         if(!worker->scratchpad) {
1145                 log_err("malloc failure");
1146                 worker_delete(worker);
1147                 return 0;
1148         }
1149
1150         server_stats_init(&worker->stats, cfg);
1151         alloc_init(&worker->alloc, &worker->daemon->superalloc,
1152                 worker->thread_num);
1153         alloc_set_id_cleanup(&worker->alloc, &worker_alloc_cleanup, worker);
1154         worker->env = *worker->daemon->env;
1155         comm_base_timept(worker->base, &worker->env.now, &worker->env.now_tv);
1156         if(worker->thread_num == 0)
1157                 log_set_time(worker->env.now);
1158         worker->env.worker = worker;
1159         worker->env.send_query = &worker_send_query;
1160         worker->env.alloc = &worker->alloc;
1161         worker->env.rnd = worker->rndstate;
1162         worker->env.scratch = worker->scratchpad;
1163         worker->env.mesh = mesh_create(&worker->daemon->mods, &worker->env);
1164         worker->env.detach_subs = &mesh_detach_subs;
1165         worker->env.attach_sub = &mesh_attach_sub;
1166         worker->env.kill_sub = &mesh_state_delete;
1167         worker->env.detect_cycle = &mesh_detect_cycle;
1168         worker->env.scratch_buffer = ldns_buffer_new(cfg->msg_buffer_size);
1169         if(!(worker->env.fwds = forwards_create()) ||
1170                 !forwards_apply_cfg(worker->env.fwds, cfg)) {
1171                 log_err("Could not set forward zones");
1172                 worker_delete(worker);
1173                 return 0;
1174         }
1175         if(!(worker->env.hints = hints_create()) ||
1176                 !hints_apply_cfg(worker->env.hints, cfg)) {
1177                 log_err("Could not set root or stub hints");
1178                 worker_delete(worker);
1179                 return 0;
1180         }
1181         /* one probe timer per process -- if we have 5011 anchors */
1182         if(autr_get_num_anchors(worker->env.anchors) > 0
1183 #ifndef THREADS_DISABLED
1184                 && worker->thread_num == 0
1185 #endif
1186                 ) {
1187                 struct timeval tv;
1188                 tv.tv_sec = 0;
1189                 tv.tv_usec = 0;
1190                 worker->env.probe_timer = comm_timer_create(worker->base,
1191                         worker_probe_timer_cb, worker);
1192                 if(!worker->env.probe_timer) {
1193                         log_err("could not create 5011-probe timer");
1194                 } else {
1195                         /* let timer fire, then it can reset itself */
1196                         comm_timer_set(worker->env.probe_timer, &tv);
1197                 }
1198         }
1199         if(!worker->env.mesh || !worker->env.scratch_buffer) {
1200                 worker_delete(worker);
1201                 return 0;
1202         }
1203         worker_mem_report(worker, NULL);
1204         /* if statistics enabled start timer */
1205         if(worker->env.cfg->stat_interval > 0) {
1206                 verbose(VERB_ALGO, "set statistics interval %d secs",
1207                         worker->env.cfg->stat_interval);
1208                 worker_restart_timer(worker);
1209         }
1210         return 1;
1211 }
1212
1213 void
1214 worker_work(struct worker* worker)
1215 {
1216         comm_base_dispatch(worker->base);
1217 }
1218
1219 void
1220 worker_delete(struct worker* worker)
1221 {
1222         if(!worker)
1223                 return;
1224         if(worker->env.mesh && verbosity >= VERB_OPS) {
1225                 server_stats_log(&worker->stats, worker, worker->thread_num);
1226                 mesh_stats(worker->env.mesh, "mesh has");
1227                 worker_mem_report(worker, NULL);
1228         }
1229         outside_network_quit_prepare(worker->back);
1230         mesh_delete(worker->env.mesh);
1231         ldns_buffer_free(worker->env.scratch_buffer);
1232         forwards_delete(worker->env.fwds);
1233         hints_delete(worker->env.hints);
1234         listen_delete(worker->front);
1235         outside_network_delete(worker->back);
1236         comm_signal_delete(worker->comsig);
1237         tube_delete(worker->cmd);
1238         comm_timer_delete(worker->stat_timer);
1239         comm_timer_delete(worker->env.probe_timer);
1240         free(worker->ports);
1241         if(worker->thread_num == 0) {
1242                 log_set_time(NULL);
1243 #ifdef UB_ON_WINDOWS
1244                 wsvc_desetup_worker(worker);
1245 #endif /* UB_ON_WINDOWS */
1246         }
1247         comm_base_delete(worker->base);
1248         ub_randfree(worker->rndstate);
1249         alloc_clear(&worker->alloc);
1250         regional_destroy(worker->scratchpad);
1251         free(worker);
1252 }
1253
1254 struct outbound_entry*
1255 worker_send_query(uint8_t* qname, size_t qnamelen, uint16_t qtype,
1256         uint16_t qclass, uint16_t flags, int dnssec, int want_dnssec,
1257         struct sockaddr_storage* addr, socklen_t addrlen, uint8_t* zone,
1258         size_t zonelen, struct module_qstate* q)
1259 {
1260         struct worker* worker = q->env->worker;
1261         struct outbound_entry* e = (struct outbound_entry*)regional_alloc(
1262                 q->region, sizeof(*e));
1263         if(!e)
1264                 return NULL;
1265         e->qstate = q;
1266         e->qsent = outnet_serviced_query(worker->back, qname,
1267                 qnamelen, qtype, qclass, flags, dnssec, want_dnssec,
1268                 q->env->cfg->tcp_upstream, q->env->cfg->ssl_upstream, addr,
1269                 addrlen, zone, zonelen, worker_handle_service_reply, e,
1270                 worker->back->udp_buff);
1271         if(!e->qsent) {
1272                 return NULL;
1273         }
1274         return e;
1275 }
1276
1277 void
1278 worker_alloc_cleanup(void* arg)
1279 {
1280         struct worker* worker = (struct worker*)arg;
1281         slabhash_clear(&worker->env.rrset_cache->table);
1282         slabhash_clear(worker->env.msg_cache);
1283 }
1284
1285 void worker_stats_clear(struct worker* worker)
1286 {
1287         server_stats_init(&worker->stats, worker->env.cfg);
1288         mesh_stats_clear(worker->env.mesh);
1289         worker->back->unwanted_replies = 0;
1290 }
1291
1292 void worker_start_accept(void* arg)
1293 {
1294         struct worker* worker = (struct worker*)arg;
1295         listen_start_accept(worker->front);
1296         if(worker->thread_num == 0)
1297                 daemon_remote_start_accept(worker->daemon->rc);
1298 }
1299
1300 void worker_stop_accept(void* arg)
1301 {
1302         struct worker* worker = (struct worker*)arg;
1303         listen_stop_accept(worker->front);
1304         if(worker->thread_num == 0)
1305                 daemon_remote_stop_accept(worker->daemon->rc);
1306 }
1307
1308 /* --- fake callbacks for fptr_wlist to work --- */
1309 struct outbound_entry* libworker_send_query(uint8_t* ATTR_UNUSED(qname),
1310         size_t ATTR_UNUSED(qnamelen), uint16_t ATTR_UNUSED(qtype),
1311         uint16_t ATTR_UNUSED(qclass), uint16_t ATTR_UNUSED(flags),
1312         int ATTR_UNUSED(dnssec), int ATTR_UNUSED(want_dnssec),
1313         struct sockaddr_storage* ATTR_UNUSED(addr),
1314         socklen_t ATTR_UNUSED(addrlen), struct module_qstate* ATTR_UNUSED(q))
1315 {
1316         log_assert(0);
1317         return 0;
1318 }
1319
1320 int libworker_handle_reply(struct comm_point* ATTR_UNUSED(c),
1321         void* ATTR_UNUSED(arg), int ATTR_UNUSED(error),
1322         struct comm_reply* ATTR_UNUSED(reply_info))
1323 {
1324         log_assert(0);
1325         return 0;
1326 }
1327
1328 int libworker_handle_service_reply(struct comm_point* ATTR_UNUSED(c),
1329         void* ATTR_UNUSED(arg), int ATTR_UNUSED(error),
1330         struct comm_reply* ATTR_UNUSED(reply_info))
1331 {
1332         log_assert(0);
1333         return 0;
1334 }
1335
1336 void libworker_handle_control_cmd(struct tube* ATTR_UNUSED(tube),
1337         uint8_t* ATTR_UNUSED(buffer), size_t ATTR_UNUSED(len),
1338         int ATTR_UNUSED(error), void* ATTR_UNUSED(arg))
1339 {
1340         log_assert(0);
1341 }
1342
1343 void libworker_fg_done_cb(void* ATTR_UNUSED(arg), int ATTR_UNUSED(rcode),
1344         ldns_buffer* ATTR_UNUSED(buf), enum sec_status ATTR_UNUSED(s),
1345         char* ATTR_UNUSED(why_bogus))
1346 {
1347         log_assert(0);
1348 }
1349
1350 void libworker_bg_done_cb(void* ATTR_UNUSED(arg), int ATTR_UNUSED(rcode),
1351         ldns_buffer* ATTR_UNUSED(buf), enum sec_status ATTR_UNUSED(s),
1352         char* ATTR_UNUSED(why_bogus))
1353 {
1354         log_assert(0);
1355 }
1356
1357 int context_query_cmp(const void* ATTR_UNUSED(a), const void* ATTR_UNUSED(b))
1358 {
1359         log_assert(0);
1360         return 0;
1361 }
1362
1363 int order_lock_cmp(const void* ATTR_UNUSED(e1), const void* ATTR_UNUSED(e2))
1364 {
1365         log_assert(0);
1366         return 0;
1367 }
1368
1369 int codeline_cmp(const void* ATTR_UNUSED(a), const void* ATTR_UNUSED(b))
1370 {
1371         log_assert(0);
1372         return 0;
1373 }
1374