contrib/unbound/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
  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/mesh.h"
  62 #include "services/localzone.h"
  63 #include "util/data/msgparse.h"
  64 #include "util/data/msgencode.h"
  65 #include "util/data/dname.h"
  66 #include "util/fptr_wlist.h"
  67 #include "util/tube.h"
  68 #include "iterator/iter_fwd.h"
  69 #include "iterator/iter_hints.h"
  70 #include "validator/autotrust.h"
  71 #include "validator/val_anchor.h"
  72 #include "libunbound/context.h"
  73 #include "libunbound/libworker.h"
  74 #include "sldns/sbuffer.h"
  75
  76 #ifdef HAVE_SYS_TYPES_H
  77 #  include <sys/types.h>
  78 #endif
  79 #ifdef HAVE_NETDB_H
  80 #include <netdb.h>
  81 #endif
  82 #include <signal.h>
  83 #ifdef UB_ON_WINDOWS
  84 #include "winrc/win_svc.h"
  85 #endif
  86
  87 /** Size of an UDP datagram */
  88 #define NORMAL_UDP_SIZE 512 /* bytes */
  89 /** ratelimit for error responses */
  90 #define ERROR_RATELIMIT 100 /* qps */
  91
  92 /**
  93  * seconds to add to prefetch leeway.  This is a TTL that expires old rrsets
  94  * earlier than they should in order to put the new update into the cache.
  95  * This additional value is to make sure that if not all TTLs are equal in
  96  * the message to be updated(and replaced), that rrsets with up to this much
  97  * extra TTL are also replaced.  This means that the resulting new message
  98  * will have (most likely) this TTL at least, avoiding very small 'split
  99  * second' TTLs due to operators choosing relative primes for TTLs (or so).
 100  * Also has to be at least one to break ties (and overwrite cached entry).
 101  */
 102 #define PREFETCH_EXPIRY_ADD 60
 103
 104 #ifdef UNBOUND_ALLOC_STATS
 105 /** measure memory leakage */
 106 static void
 107 debug_memleak(size_t accounted, size_t heap,
 108         size_t total_alloc, size_t total_free)
 109 {
 110         static int init = 0;
 111         static size_t base_heap, base_accounted, base_alloc, base_free;
 112         size_t base_af, cur_af, grow_af, grow_acc;
 113         if(!init) {
 114                 init = 1;
 115                 base_heap = heap;
 116                 base_accounted = accounted;
 117                 base_alloc = total_alloc;
 118                 base_free = total_free;
 119         }
 120         base_af = base_alloc - base_free;
 121         cur_af = total_alloc - total_free;
 122         grow_af = cur_af - base_af;
 123         grow_acc = accounted - base_accounted;
 124         log_info("Leakage: %d leaked. growth: %u use, %u acc, %u heap",
 125                 (int)(grow_af - grow_acc), (unsigned)grow_af,
 126                 (unsigned)grow_acc, (unsigned)(heap - base_heap));
 127 }
 128
 129 /** give debug heap size indication */
 130 static void
 131 debug_total_mem(size_t calctotal)
 132 {
 133 #ifdef HAVE_SBRK
 134         extern void* unbound_start_brk;
 135         extern size_t unbound_mem_alloc, unbound_mem_freed;
 136         void* cur = sbrk(0);
 137         int total = cur-unbound_start_brk;
 138         log_info("Total heap memory estimate: %u  total-alloc: %u  "
 139                 "total-free: %u", (unsigned)total,
 140                 (unsigned)unbound_mem_alloc, (unsigned)unbound_mem_freed);
 141         debug_memleak(calctotal, (size_t)total,
 142                 unbound_mem_alloc, unbound_mem_freed);
 143 #else
 144         (void)calctotal;
 145 #endif /* HAVE_SBRK */
 146 }
 147 #endif /* UNBOUND_ALLOC_STATS */
 148
 149 /** Report on memory usage by this thread and global */
 150 static void
 151 worker_mem_report(struct worker* ATTR_UNUSED(worker),
 152         struct serviced_query* ATTR_UNUSED(cur_serv))
 153 {
 154 #ifdef UNBOUND_ALLOC_STATS
 155         /* debug func in validator module */
 156         size_t total, front, back, mesh, msg, rrset, infra, ac, superac;
 157         size_t me, iter, val, anch;
 158         int i;
 159         if(verbosity < VERB_ALGO)
 160                 return;
 161         front = listen_get_mem(worker->front);
 162         back = outnet_get_mem(worker->back);
 163         msg = slabhash_get_mem(worker->env.msg_cache);
 164         rrset = slabhash_get_mem(&worker->env.rrset_cache->table);
 165         infra = infra_get_mem(worker->env.infra_cache);
 166         mesh = mesh_get_mem(worker->env.mesh);
 167         ac = alloc_get_mem(&worker->alloc);
 168         superac = alloc_get_mem(&worker->daemon->superalloc);
 169         anch = anchors_get_mem(worker->env.anchors);
 170         iter = 0;
 171         val = 0;
 172         for(i=0; i<worker->env.mesh->mods.num; i++) {
 173                 fptr_ok(fptr_whitelist_mod_get_mem(worker->env.mesh->
 174                         mods.mod[i]->get_mem));
 175                 if(strcmp(worker->env.mesh->mods.mod[i]->name, "validator")==0)
 176                         val += (*worker->env.mesh->mods.mod[i]->get_mem)
 177                                 (&worker->env, i);
 178                 else    iter += (*worker->env.mesh->mods.mod[i]->get_mem)
 179                                 (&worker->env, i);
 180         }
 181         me = sizeof(*worker) + sizeof(*worker->base) + sizeof(*worker->comsig)
 182                 + comm_point_get_mem(worker->cmd_com)
 183                 + sizeof(worker->rndstate)
 184                 + regional_get_mem(worker->scratchpad)
 185                 + sizeof(*worker->env.scratch_buffer)
 186                 + sldns_buffer_capacity(worker->env.scratch_buffer)
 187                 + forwards_get_mem(worker->env.fwds)
 188                 + hints_get_mem(worker->env.hints);
 189         if(worker->thread_num == 0)
 190                 me += acl_list_get_mem(worker->daemon->acl);
 191         if(cur_serv) {
 192                 me += serviced_get_mem(cur_serv);
 193         }
 194         total = front+back+mesh+msg+rrset+infra+iter+val+ac+superac+me;
 195         log_info("Memory conditions: %u front=%u back=%u mesh=%u msg=%u "
 196                 "rrset=%u infra=%u iter=%u val=%u anchors=%u "
 197                 "alloccache=%u globalalloccache=%u me=%u",
 198                 (unsigned)total, (unsigned)front, (unsigned)back,
 199                 (unsigned)mesh, (unsigned)msg, (unsigned)rrset,
 200                 (unsigned)infra, (unsigned)iter, (unsigned)val, (unsigned)anch,
 201                 (unsigned)ac, (unsigned)superac, (unsigned)me);
 202         debug_total_mem(total);
 203 #else /* no UNBOUND_ALLOC_STATS */
 204         size_t val = 0;
 205         int i;
 206         if(verbosity < VERB_QUERY)
 207                 return;
 208         for(i=0; i<worker->env.mesh->mods.num; i++) {
 209                 fptr_ok(fptr_whitelist_mod_get_mem(worker->env.mesh->
 210                         mods.mod[i]->get_mem));
 211                 if(strcmp(worker->env.mesh->mods.mod[i]->name, "validator")==0)
 212                         val += (*worker->env.mesh->mods.mod[i]->get_mem)
 213                                 (&worker->env, i);
 214         }
 215         verbose(VERB_QUERY, "cache memory msg=%u rrset=%u infra=%u val=%u",
 216                 (unsigned)slabhash_get_mem(worker->env.msg_cache),
 217                 (unsigned)slabhash_get_mem(&worker->env.rrset_cache->table),
 218                 (unsigned)infra_get_mem(worker->env.infra_cache),
 219                 (unsigned)val);
 220 #endif /* UNBOUND_ALLOC_STATS */
 221 }
 222
 223 void
 224 worker_send_cmd(struct worker* worker, enum worker_commands cmd)
 225 {
 226         uint32_t c = (uint32_t)htonl(cmd);
 227         if(!tube_write_msg(worker->cmd, (uint8_t*)&c, sizeof(c), 0)) {
 228                 log_err("worker send cmd %d failed", (int)cmd);
 229         }
 230 }
 231
 232 int
 233 worker_handle_reply(struct comm_point* c, void* arg, int error,
 234         struct comm_reply* reply_info)
 235 {
 236         struct module_qstate* q = (struct module_qstate*)arg;
 237         struct worker* worker = q->env->worker;
 238         struct outbound_entry e;
 239         e.qstate = q;
 240         e.qsent = NULL;
 241
 242         if(error != 0) {
 243                 mesh_report_reply(worker->env.mesh, &e, reply_info, error);
 244                 worker_mem_report(worker, NULL);
 245                 return 0;
 246         }
 247         /* sanity check. */
 248         if(!LDNS_QR_WIRE(sldns_buffer_begin(c->buffer))
 249                 || LDNS_OPCODE_WIRE(sldns_buffer_begin(c->buffer)) !=
 250                         LDNS_PACKET_QUERY
 251                 || LDNS_QDCOUNT(sldns_buffer_begin(c->buffer)) > 1) {
 252                 /* error becomes timeout for the module as if this reply
 253                  * never arrived. */
 254                 mesh_report_reply(worker->env.mesh, &e, reply_info,
 255                         NETEVENT_TIMEOUT);
 256                 worker_mem_report(worker, NULL);
 257                 return 0;
 258         }
 259         mesh_report_reply(worker->env.mesh, &e, reply_info, NETEVENT_NOERROR);
 260         worker_mem_report(worker, NULL);
 261         return 0;
 262 }
 263
 264 int
 265 worker_handle_service_reply(struct comm_point* c, void* arg, int error,
 266         struct comm_reply* reply_info)
 267 {
 268         struct outbound_entry* e = (struct outbound_entry*)arg;
 269         struct worker* worker = e->qstate->env->worker;
 270         struct serviced_query *sq = e->qsent;
 271
 272         verbose(VERB_ALGO, "worker svcd callback for qstate %p", e->qstate);
 273         if(error != 0) {
 274                 mesh_report_reply(worker->env.mesh, e, reply_info, error);
 275                 worker_mem_report(worker, sq);
 276                 return 0;
 277         }
 278         /* sanity check. */
 279         if(!LDNS_QR_WIRE(sldns_buffer_begin(c->buffer))
 280                 || LDNS_OPCODE_WIRE(sldns_buffer_begin(c->buffer)) !=
 281                         LDNS_PACKET_QUERY
 282                 || LDNS_QDCOUNT(sldns_buffer_begin(c->buffer)) > 1) {
 283                 /* error becomes timeout for the module as if this reply
 284                  * never arrived. */
 285                 verbose(VERB_ALGO, "worker: bad reply handled as timeout");
 286                 mesh_report_reply(worker->env.mesh, e, reply_info,
 287                         NETEVENT_TIMEOUT);
 288                 worker_mem_report(worker, sq);
 289                 return 0;
 290         }
 291         mesh_report_reply(worker->env.mesh, e, reply_info, NETEVENT_NOERROR);
 292         worker_mem_report(worker, sq);
 293         return 0;
 294 }
 295
 296 /** ratelimit error replies
 297  * @param worker: the worker struct with ratelimit counter
 298  * @param err: error code that would be wanted.
 299  * @return value of err if okay, or -1 if it should be discarded instead.
 300  */
 301 static int
 302 worker_err_ratelimit(struct worker* worker, int err)
 303 {
 304         if(worker->err_limit_time == *worker->env.now) {
 305                 /* see if limit is exceeded for this second */
 306                 if(worker->err_limit_count++ > ERROR_RATELIMIT)
 307                         return -1;
 308         } else {
 309                 /* new second, new limits */
 310                 worker->err_limit_time = *worker->env.now;
 311                 worker->err_limit_count = 1;
 312         }
 313         return err;
 314 }
 315
 316 /** check request sanity.
 317  * @param pkt: the wire packet to examine for sanity.
 318  * @param worker: parameters for checking.
 319  * @return error code, 0 OK, or -1 discard.
 320 */
 321 static int
 322 worker_check_request(sldns_buffer* pkt, struct worker* worker)
 323 {
 324         if(sldns_buffer_limit(pkt) < LDNS_HEADER_SIZE) {
 325                 verbose(VERB_QUERY, "request too short, discarded");
 326                 return -1;
 327         }
 328         if(sldns_buffer_limit(pkt) > NORMAL_UDP_SIZE &&
 329                 worker->daemon->cfg->harden_large_queries) {
 330                 verbose(VERB_QUERY, "request too large, discarded");
 331                 return -1;
 332         }
 333         if(LDNS_QR_WIRE(sldns_buffer_begin(pkt))) {
 334                 verbose(VERB_QUERY, "request has QR bit on, discarded");
 335                 return -1;
 336         }
 337         if(LDNS_TC_WIRE(sldns_buffer_begin(pkt))) {
 338                 LDNS_TC_CLR(sldns_buffer_begin(pkt));
 339                 verbose(VERB_QUERY, "request bad, has TC bit on");
 340                 return worker_err_ratelimit(worker, LDNS_RCODE_FORMERR);
 341         }
 342         if(LDNS_OPCODE_WIRE(sldns_buffer_begin(pkt)) != LDNS_PACKET_QUERY) {
 343                 verbose(VERB_QUERY, "request unknown opcode %d",
 344                         LDNS_OPCODE_WIRE(sldns_buffer_begin(pkt)));
 345                 return worker_err_ratelimit(worker, LDNS_RCODE_NOTIMPL);
 346         }
 347         if(LDNS_QDCOUNT(sldns_buffer_begin(pkt)) != 1) {
 348                 verbose(VERB_QUERY, "request wrong nr qd=%d",
 349                         LDNS_QDCOUNT(sldns_buffer_begin(pkt)));
 350                 return worker_err_ratelimit(worker, LDNS_RCODE_FORMERR);
 351         }
 352         if(LDNS_ANCOUNT(sldns_buffer_begin(pkt)) != 0) {
 353                 verbose(VERB_QUERY, "request wrong nr an=%d",
 354                         LDNS_ANCOUNT(sldns_buffer_begin(pkt)));
 355                 return worker_err_ratelimit(worker, LDNS_RCODE_FORMERR);
 356         }
 357         if(LDNS_NSCOUNT(sldns_buffer_begin(pkt)) != 0) {
 358                 verbose(VERB_QUERY, "request wrong nr ns=%d",
 359                         LDNS_NSCOUNT(sldns_buffer_begin(pkt)));
 360                 return worker_err_ratelimit(worker, LDNS_RCODE_FORMERR);
 361         }
 362         if(LDNS_ARCOUNT(sldns_buffer_begin(pkt)) > 1) {
 363                 verbose(VERB_QUERY, "request wrong nr ar=%d",
 364                         LDNS_ARCOUNT(sldns_buffer_begin(pkt)));
 365                 return worker_err_ratelimit(worker, LDNS_RCODE_FORMERR);
 366         }
 367         return 0;
 368 }
 369
 370 void
 371 worker_handle_control_cmd(struct tube* ATTR_UNUSED(tube), uint8_t* msg,
 372         size_t len, int error, void* arg)
 373 {
 374         struct worker* worker = (struct worker*)arg;
 375         enum worker_commands cmd;
 376         if(error != NETEVENT_NOERROR) {
 377                 free(msg);
 378                 if(error == NETEVENT_CLOSED)
 379                         comm_base_exit(worker->base);
 380                 else    log_info("control event: %d", error);
 381                 return;
 382         }
 383         if(len != sizeof(uint32_t)) {
 384                 fatal_exit("bad control msg length %d", (int)len);
 385         }
 386         cmd = sldns_read_uint32(msg);
 387         free(msg);
 388         switch(cmd) {
 389         case worker_cmd_quit:
 390                 verbose(VERB_ALGO, "got control cmd quit");
 391                 comm_base_exit(worker->base);
 392                 break;
 393         case worker_cmd_stats:
 394                 verbose(VERB_ALGO, "got control cmd stats");
 395                 server_stats_reply(worker, 1);
 396                 break;
 397         case worker_cmd_stats_noreset:
 398                 verbose(VERB_ALGO, "got control cmd stats_noreset");
 399                 server_stats_reply(worker, 0);
 400                 break;
 401         case worker_cmd_remote:
 402                 verbose(VERB_ALGO, "got control cmd remote");
 403                 daemon_remote_exec(worker);
 404                 break;
 405         default:
 406                 log_err("bad command %d", (int)cmd);
 407                 break;
 408         }
 409 }
 410
 411 /** check if a delegation is secure */
 412 static enum sec_status
 413 check_delegation_secure(struct reply_info *rep)
 414 {
 415         /* return smallest security status */
 416         size_t i;
 417         enum sec_status sec = sec_status_secure;
 418         enum sec_status s;
 419         size_t num = rep->an_numrrsets + rep->ns_numrrsets;
 420         /* check if answer and authority are OK */
 421         for(i=0; i<num; i++) {
 422                 s = ((struct packed_rrset_data*)rep->rrsets[i]->entry.data)
 423                         ->security;
 424                 if(s < sec)
 425                         sec = s;
 426         }
 427         /* in additional, only unchecked triggers revalidation */
 428         for(i=num; i<rep->rrset_count; i++) {
 429                 s = ((struct packed_rrset_data*)rep->rrsets[i]->entry.data)
 430                         ->security;
 431                 if(s == sec_status_unchecked)
 432                         return s;
 433         }
 434         return sec;
 435 }
 436
 437 /** remove nonsecure from a delegation referral additional section */
 438 static void
 439 deleg_remove_nonsecure_additional(struct reply_info* rep)
 440 {
 441         /* we can simply edit it, since we are working in the scratch region */
 442         size_t i;
 443         enum sec_status s;
 444
 445         for(i = rep->an_numrrsets+rep->ns_numrrsets; i<rep->rrset_count; i++) {
 446                 s = ((struct packed_rrset_data*)rep->rrsets[i]->entry.data)
 447                         ->security;
 448                 if(s != sec_status_secure) {
 449                         memmove(rep->rrsets+i, rep->rrsets+i+1,
 450                                 sizeof(struct ub_packed_rrset_key*)*
 451                                 (rep->rrset_count - i - 1));
 452                         rep->ar_numrrsets--;
 453                         rep->rrset_count--;
 454                         i--;
 455                 }
 456         }
 457 }
 458
 459 /** answer nonrecursive query from the cache */
 460 static int
 461 answer_norec_from_cache(struct worker* worker, struct query_info* qinfo,
 462         uint16_t id, uint16_t flags, struct comm_reply* repinfo,
 463         struct edns_data* edns)
 464 {
 465         /* for a nonrecursive query return either:
 466          *      o an error (servfail; we try to avoid this)
 467          *      o a delegation (closest we have; this routine tries that)
 468          *      o the answer (checked by answer_from_cache)
 469          *
 470          * So, grab a delegation from the rrset cache.
 471          * Then check if it needs validation, if so, this routine fails,
 472          * so that iterator can prime and validator can verify rrsets.
 473          */
 474         uint16_t udpsize = edns->udp_size;
 475         int secure = 0;
 476         time_t timenow = *worker->env.now;
 477         int must_validate = (!(flags&BIT_CD) || worker->env.cfg->ignore_cd)
 478                 && worker->env.need_to_validate;
 479         struct dns_msg *msg = NULL;
 480         struct delegpt *dp;
 481
 482         dp = dns_cache_find_delegation(&worker->env, qinfo->qname,
 483                 qinfo->qname_len, qinfo->qtype, qinfo->qclass,
 484                 worker->scratchpad, &msg, timenow);
 485         if(!dp) { /* no delegation, need to reprime */
 486                 regional_free_all(worker->scratchpad);
 487                 return 0;
 488         }
 489         if(must_validate) {
 490                 switch(check_delegation_secure(msg->rep)) {
 491                 case sec_status_unchecked:
 492                         /* some rrsets have not been verified yet, go and
 493                          * let validator do that */
 494                         regional_free_all(worker->scratchpad);
 495                         return 0;
 496                 case sec_status_bogus:
 497                         /* some rrsets are bogus, reply servfail */
 498                         edns->edns_version = EDNS_ADVERTISED_VERSION;
 499                         edns->udp_size = EDNS_ADVERTISED_SIZE;
 500                         edns->ext_rcode = 0;
 501                         edns->bits &= EDNS_DO;
 502                         error_encode(repinfo->c->buffer, LDNS_RCODE_SERVFAIL,
 503                                 &msg->qinfo, id, flags, edns);
 504                         regional_free_all(worker->scratchpad);
 505                         if(worker->stats.extended) {
 506                                 worker->stats.ans_bogus++;
 507                                 worker->stats.ans_rcode[LDNS_RCODE_SERVFAIL]++;
 508                         }
 509                         return 1;
 510                 case sec_status_secure:
 511                         /* all rrsets are secure */
 512                         /* remove non-secure rrsets from the add. section*/
 513                         if(worker->env.cfg->val_clean_additional)
 514                                 deleg_remove_nonsecure_additional(msg->rep);
 515                         secure = 1;
 516                         break;
 517                 case sec_status_indeterminate:
 518                 case sec_status_insecure:
 519                 default:
 520                         /* not secure */
 521                         secure = 0;
 522                         break;
 523                 }
 524         }
 525         /* return this delegation from the cache */
 526         edns->edns_version = EDNS_ADVERTISED_VERSION;
 527         edns->udp_size = EDNS_ADVERTISED_SIZE;
 528         edns->ext_rcode = 0;
 529         edns->bits &= EDNS_DO;
 530         msg->rep->flags |= BIT_QR|BIT_RA;
 531         if(!reply_info_answer_encode(&msg->qinfo, msg->rep, id, flags,
 532                 repinfo->c->buffer, 0, 1, worker->scratchpad,
 533                 udpsize, edns, (int)(edns->bits & EDNS_DO), secure)) {
 534                 error_encode(repinfo->c->buffer, LDNS_RCODE_SERVFAIL,
 535                         &msg->qinfo, id, flags, edns);
 536         }
 537         regional_free_all(worker->scratchpad);
 538         if(worker->stats.extended) {
 539                 if(secure) worker->stats.ans_secure++;
 540                 server_stats_insrcode(&worker->stats, repinfo->c->buffer);
 541         }
 542         return 1;
 543 }
 544
 545 /** answer query from the cache */
 546 static int
 547 answer_from_cache(struct worker* worker, struct query_info* qinfo,
 548         struct reply_info* rep, uint16_t id, uint16_t flags,
 549         struct comm_reply* repinfo, struct edns_data* edns)
 550 {
 551         time_t timenow = *worker->env.now;
 552         uint16_t udpsize = edns->udp_size;
 553         int secure;
 554         int must_validate = (!(flags&BIT_CD) || worker->env.cfg->ignore_cd)
 555                 && worker->env.need_to_validate;
 556         /* see if it is possible */
 557         if(rep->ttl < timenow) {
 558                 /* the rrsets may have been updated in the meantime.
 559                  * we will refetch the message format from the
 560                  * authoritative server
 561                  */
 562                 return 0;
 563         }
 564         if(!rrset_array_lock(rep->ref, rep->rrset_count, timenow))
 565                 return 0;
 566         /* locked and ids and ttls are OK. */
 567         /* check CNAME chain (if any) */
 568         if(rep->an_numrrsets > 0 && (rep->rrsets[0]->rk.type ==
 569                 htons(LDNS_RR_TYPE_CNAME) || rep->rrsets[0]->rk.type ==
 570                 htons(LDNS_RR_TYPE_DNAME))) {
 571                 if(!reply_check_cname_chain(qinfo, rep)) {
 572                         /* cname chain invalid, redo iterator steps */
 573                         verbose(VERB_ALGO, "Cache reply: cname chain broken");
 574                 bail_out:
 575                         rrset_array_unlock_touch(worker->env.rrset_cache,
 576                                 worker->scratchpad, rep->ref, rep->rrset_count);
 577                         regional_free_all(worker->scratchpad);
 578                         return 0;
 579                 }
 580         }
 581         /* check security status of the cached answer */
 582         if( rep->security == sec_status_bogus && must_validate) {
 583                 /* BAD cached */
 584                 edns->edns_version = EDNS_ADVERTISED_VERSION;
 585                 edns->udp_size = EDNS_ADVERTISED_SIZE;
 586                 edns->ext_rcode = 0;
 587                 edns->bits &= EDNS_DO;
 588                 error_encode(repinfo->c->buffer, LDNS_RCODE_SERVFAIL,
 589                         qinfo, id, flags, edns);
 590                 rrset_array_unlock_touch(worker->env.rrset_cache,
 591                         worker->scratchpad, rep->ref, rep->rrset_count);
 592                 regional_free_all(worker->scratchpad);
 593                 if(worker->stats.extended) {
 594                         worker->stats.ans_bogus ++;
 595                         worker->stats.ans_rcode[LDNS_RCODE_SERVFAIL] ++;
 596                 }
 597                 return 1;
 598         } else if( rep->security == sec_status_unchecked && must_validate) {
 599                 verbose(VERB_ALGO, "Cache reply: unchecked entry needs "
 600                         "validation");
 601                 goto bail_out; /* need to validate cache entry first */
 602         } else if(rep->security == sec_status_secure) {
 603                 if(reply_all_rrsets_secure(rep))
 604                         secure = 1;
 605                 else    {
 606                         if(must_validate) {
 607                                 verbose(VERB_ALGO, "Cache reply: secure entry"
 608                                         " changed status");
 609                                 goto bail_out; /* rrset changed, re-verify */
 610                         }
 611                         secure = 0;
 612                 }
 613         } else  secure = 0;
 614
 615         edns->edns_version = EDNS_ADVERTISED_VERSION;
 616         edns->udp_size = EDNS_ADVERTISED_SIZE;
 617         edns->ext_rcode = 0;
 618         edns->bits &= EDNS_DO;
 619         if(!reply_info_answer_encode(qinfo, rep, id, flags,
 620                 repinfo->c->buffer, timenow, 1, worker->scratchpad,
 621                 udpsize, edns, (int)(edns->bits & EDNS_DO), secure)) {
 622                 error_encode(repinfo->c->buffer, LDNS_RCODE_SERVFAIL,
 623                         qinfo, id, flags, edns);
 624         }
 625         /* cannot send the reply right now, because blocking network syscall
 626          * is bad while holding locks. */
 627         rrset_array_unlock_touch(worker->env.rrset_cache, worker->scratchpad,
 628                 rep->ref, rep->rrset_count);
 629         regional_free_all(worker->scratchpad);
 630         if(worker->stats.extended) {
 631                 if(secure) worker->stats.ans_secure++;
 632                 server_stats_insrcode(&worker->stats, repinfo->c->buffer);
 633         }
 634         /* go and return this buffer to the client */
 635         return 1;
 636 }
 637
 638 /** Reply to client and perform prefetch to keep cache up to date */
 639 static void
 640 reply_and_prefetch(struct worker* worker, struct query_info* qinfo,
 641         uint16_t flags, struct comm_reply* repinfo, time_t leeway)
 642 {
 643         /* first send answer to client to keep its latency
 644          * as small as a cachereply */
 645         comm_point_send_reply(repinfo);
 646         server_stats_prefetch(&worker->stats, worker);
 647
 648         /* create the prefetch in the mesh as a normal lookup without
 649          * client addrs waiting, which has the cache blacklisted (to bypass
 650          * the cache and go to the network for the data). */
 651         /* this (potentially) runs the mesh for the new query */
 652         mesh_new_prefetch(worker->env.mesh, qinfo, flags, leeway +
 653                 PREFETCH_EXPIRY_ADD);
 654 }
 655
 656 /**
 657  * Fill CH class answer into buffer. Keeps query.
 658  * @param pkt: buffer
 659  * @param str: string to put into text record (<255).
 660  * @param edns: edns reply information.
 661  */
 662 static void
 663 chaos_replystr(sldns_buffer* pkt, const char* str, struct edns_data* edns)
 664 {
 665         size_t len = strlen(str);
 666         unsigned int rd = LDNS_RD_WIRE(sldns_buffer_begin(pkt));
 667         unsigned int cd = LDNS_CD_WIRE(sldns_buffer_begin(pkt));
 668         if(len>255) len=255; /* cap size of TXT record */
 669         sldns_buffer_clear(pkt);
 670         sldns_buffer_skip(pkt, (ssize_t)sizeof(uint16_t)); /* skip id */
 671         sldns_buffer_write_u16(pkt, (uint16_t)(BIT_QR|BIT_RA));
 672         if(rd) LDNS_RD_SET(sldns_buffer_begin(pkt));
 673         if(cd) LDNS_CD_SET(sldns_buffer_begin(pkt));
 674         sldns_buffer_write_u16(pkt, 1); /* qdcount */
 675         sldns_buffer_write_u16(pkt, 1); /* ancount */
 676         sldns_buffer_write_u16(pkt, 0); /* nscount */
 677         sldns_buffer_write_u16(pkt, 0); /* arcount */
 678         (void)query_dname_len(pkt); /* skip qname */
 679         sldns_buffer_skip(pkt, (ssize_t)sizeof(uint16_t)); /* skip qtype */
 680         sldns_buffer_skip(pkt, (ssize_t)sizeof(uint16_t)); /* skip qclass */
 681         sldns_buffer_write_u16(pkt, 0xc00c); /* compr ptr to query */
 682         sldns_buffer_write_u16(pkt, LDNS_RR_TYPE_TXT);
 683         sldns_buffer_write_u16(pkt, LDNS_RR_CLASS_CH);
 684         sldns_buffer_write_u32(pkt, 0); /* TTL */
 685         sldns_buffer_write_u16(pkt, sizeof(uint8_t) + len);
 686         sldns_buffer_write_u8(pkt, len);
 687         sldns_buffer_write(pkt, str, len);
 688         sldns_buffer_flip(pkt);
 689         edns->edns_version = EDNS_ADVERTISED_VERSION;
 690         edns->udp_size = EDNS_ADVERTISED_SIZE;
 691         edns->bits &= EDNS_DO;
 692         attach_edns_record(pkt, edns);
 693 }
 694
 695 /**
 696  * Answer CH class queries.
 697  * @param w: worker
 698  * @param qinfo: query info. Pointer into packet buffer.
 699  * @param edns: edns info from query.
 700  * @param pkt: packet buffer.
 701  * @return: true if a reply is to be sent.
 702  */
 703 static int
 704 answer_chaos(struct worker* w, struct query_info* qinfo,
 705         struct edns_data* edns, sldns_buffer* pkt)
 706 {
 707         struct config_file* cfg = w->env.cfg;
 708         if(qinfo->qtype != LDNS_RR_TYPE_ANY && qinfo->qtype != LDNS_RR_TYPE_TXT)
 709                 return 0;
 710         if(query_dname_compare(qinfo->qname,
 711                 (uint8_t*)"\002id\006server") == 0 ||
 712                 query_dname_compare(qinfo->qname,
 713                 (uint8_t*)"\010hostname\004bind") == 0)
 714         {
 715                 if(cfg->hide_identity)
 716                         return 0;
 717                 if(cfg->identity==NULL || cfg->identity[0]==0) {
 718                         char buf[MAXHOSTNAMELEN+1];
 719                         if (gethostname(buf, MAXHOSTNAMELEN) == 0) {
 720                                 buf[MAXHOSTNAMELEN] = 0;
 721                                 chaos_replystr(pkt, buf, edns);
 722                         } else  {
 723                                 log_err("gethostname: %s", strerror(errno));
 724                                 chaos_replystr(pkt, "no hostname", edns);
 725                         }
 726                 }
 727                 else    chaos_replystr(pkt, cfg->identity, edns);
 728                 return 1;
 729         }
 730         if(query_dname_compare(qinfo->qname,
 731                 (uint8_t*)"\007version\006server") == 0 ||
 732                 query_dname_compare(qinfo->qname,
 733                 (uint8_t*)"\007version\004bind") == 0)
 734         {
 735                 if(cfg->hide_version)
 736                         return 0;
 737                 if(cfg->version==NULL || cfg->version[0]==0)
 738                         chaos_replystr(pkt, PACKAGE_STRING, edns);
 739                 else    chaos_replystr(pkt, cfg->version, edns);
 740                 return 1;
 741         }
 742         return 0;
 743 }
 744
 745 static int
 746 deny_refuse(struct comm_point* c, enum acl_access acl,
 747         enum acl_access deny, enum acl_access refuse,
 748         struct worker* worker, struct comm_reply* repinfo)
 749 {
 750         if(acl == deny) {
 751                 comm_point_drop_reply(repinfo);
 752                 if(worker->stats.extended)
 753                         worker->stats.unwanted_queries++;
 754                 return 0;
 755         } else if(acl == refuse) {
 756                 log_addr(VERB_ALGO, "refused query from",
 757                         &repinfo->addr, repinfo->addrlen);
 758                 log_buf(VERB_ALGO, "refuse", c->buffer);
 759                 if(worker->stats.extended)
 760                         worker->stats.unwanted_queries++;
 761                 if(worker_check_request(c->buffer, worker) == -1) {
 762                         comm_point_drop_reply(repinfo);
 763                         return 0; /* discard this */
 764                 }
 765                 sldns_buffer_set_limit(c->buffer, LDNS_HEADER_SIZE);
 766                 sldns_buffer_write_at(c->buffer, 4,
 767                         (uint8_t*)"\0\0\0\0\0\0\0\0", 8);
 768                 LDNS_QR_SET(sldns_buffer_begin(c->buffer));
 769                 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer),
 770                         LDNS_RCODE_REFUSED);
 771                 return 1;
 772         }
 773
 774         return -1;
 775 }
 776
 777 static int
 778 deny_refuse_all(struct comm_point* c, enum acl_access acl,
 779         struct worker* worker, struct comm_reply* repinfo)
 780 {
 781         return deny_refuse(c, acl, acl_deny, acl_refuse, worker, repinfo);
 782 }
 783
 784 static int
 785 deny_refuse_non_local(struct comm_point* c, enum acl_access acl,
 786         struct worker* worker, struct comm_reply* repinfo)
 787 {
 788         return deny_refuse(c, acl, acl_deny_non_local, acl_refuse_non_local, worker, repinfo);
 789 }
 790
 791 int
 792 worker_handle_request(struct comm_point* c, void* arg, int error,
 793         struct comm_reply* repinfo)
 794 {
 795         struct worker* worker = (struct worker*)arg;
 796         int ret;
 797         hashvalue_t h;
 798         struct lruhash_entry* e;
 799         struct query_info qinfo;
 800         struct edns_data edns;
 801         enum acl_access acl;
 802         int rc = 0;
 803
 804         if(error != NETEVENT_NOERROR) {
 805                 /* some bad tcp query DNS formats give these error calls */
 806                 verbose(VERB_ALGO, "handle request called with err=%d", error);
 807                 return 0;
 808         }
 809 #ifdef USE_DNSTAP
 810         if(worker->dtenv.log_client_query_messages)
 811                 dt_msg_send_client_query(&worker->dtenv, &repinfo->addr, c->type,
 812                         c->buffer);
 813 #endif
 814         acl = acl_list_lookup(worker->daemon->acl, &repinfo->addr,
 815                 repinfo->addrlen);
 816         if((ret=deny_refuse_all(c, acl, worker, repinfo)) != -1)
 817         {
 818                 if(ret == 1)
 819                         goto send_reply;
 820                 return ret;
 821         }
 822         if((ret=worker_check_request(c->buffer, worker)) != 0) {
 823                 verbose(VERB_ALGO, "worker check request: bad query.");
 824                 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen);
 825                 if(ret != -1) {
 826                         LDNS_QR_SET(sldns_buffer_begin(c->buffer));
 827                         LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), ret);
 828                         return 1;
 829                 }
 830                 comm_point_drop_reply(repinfo);
 831                 return 0;
 832         }
 833         worker->stats.num_queries++;
 834         /* see if query is in the cache */
 835         if(!query_info_parse(&qinfo, c->buffer)) {
 836                 verbose(VERB_ALGO, "worker parse request: formerror.");
 837                 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen);
 838                 if(worker_err_ratelimit(worker, LDNS_RCODE_FORMERR) == -1) {
 839                         comm_point_drop_reply(repinfo);
 840                         return 0;
 841                 }
 842                 sldns_buffer_rewind(c->buffer);
 843                 LDNS_QR_SET(sldns_buffer_begin(c->buffer));
 844                 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer),
 845                         LDNS_RCODE_FORMERR);
 846                 server_stats_insrcode(&worker->stats, c->buffer);
 847                 goto send_reply;
 848         }
 849         if(worker->env.cfg->log_queries) {
 850                 char ip[128];
 851                 addr_to_str(&repinfo->addr, repinfo->addrlen, ip, sizeof(ip));
 852                 log_nametypeclass(0, ip, qinfo.qname, qinfo.qtype, qinfo.qclass);
 853         }
 854         if(qinfo.qtype == LDNS_RR_TYPE_AXFR ||
 855                 qinfo.qtype == LDNS_RR_TYPE_IXFR) {
 856                 verbose(VERB_ALGO, "worker request: refused zone transfer.");
 857                 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen);
 858                 sldns_buffer_rewind(c->buffer);
 859                 LDNS_QR_SET(sldns_buffer_begin(c->buffer));
 860                 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer),
 861                         LDNS_RCODE_REFUSED);
 862                 if(worker->stats.extended) {
 863                         worker->stats.qtype[qinfo.qtype]++;
 864                         server_stats_insrcode(&worker->stats, c->buffer);
 865                 }
 866                 goto send_reply;
 867         }
 868         if((ret=parse_edns_from_pkt(c->buffer, &edns)) != 0) {
 869                 struct edns_data reply_edns;
 870                 verbose(VERB_ALGO, "worker parse edns: formerror.");
 871                 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen);
 872                 memset(&reply_edns, 0, sizeof(reply_edns));
 873                 reply_edns.edns_present = 1;
 874                 reply_edns.udp_size = EDNS_ADVERTISED_SIZE;
 875                 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), ret);
 876                 error_encode(c->buffer, ret, &qinfo,
 877                         *(uint16_t*)(void *)sldns_buffer_begin(c->buffer),
 878                         sldns_buffer_read_u16_at(c->buffer, 2), &reply_edns);
 879                 server_stats_insrcode(&worker->stats, c->buffer);
 880                 goto send_reply;
 881         }
 882         if(edns.edns_present && edns.edns_version != 0) {
 883                 edns.ext_rcode = (uint8_t)(EDNS_RCODE_BADVERS>>4);
 884                 edns.edns_version = EDNS_ADVERTISED_VERSION;
 885                 edns.udp_size = EDNS_ADVERTISED_SIZE;
 886                 edns.bits &= EDNS_DO;
 887                 verbose(VERB_ALGO, "query with bad edns version.");
 888                 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen);
 889                 error_encode(c->buffer, EDNS_RCODE_BADVERS&0xf, &qinfo,
 890                         *(uint16_t*)(void *)sldns_buffer_begin(c->buffer),
 891                         sldns_buffer_read_u16_at(c->buffer, 2), NULL);
 892                 attach_edns_record(c->buffer, &edns);
 893                 goto send_reply;
 894         }
 895         if(edns.edns_present && edns.udp_size < NORMAL_UDP_SIZE &&
 896                 worker->daemon->cfg->harden_short_bufsize) {
 897                 verbose(VERB_QUERY, "worker request: EDNS bufsize %d ignored",
 898                         (int)edns.udp_size);
 899                 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen);
 900                 edns.udp_size = NORMAL_UDP_SIZE;
 901         }
 902         if(edns.udp_size > worker->daemon->cfg->max_udp_size &&
 903                 c->type == comm_udp) {
 904                 verbose(VERB_QUERY,
 905                         "worker request: max UDP reply size modified"
 906                         " (%d to max-udp-size)", (int)edns.udp_size);
 907                 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen);
 908                 edns.udp_size = worker->daemon->cfg->max_udp_size;
 909         }
 910         if(edns.udp_size < LDNS_HEADER_SIZE) {
 911                 verbose(VERB_ALGO, "worker request: edns is too small.");
 912                 log_addr(VERB_CLIENT, "from", &repinfo->addr, repinfo->addrlen);
 913                 LDNS_QR_SET(sldns_buffer_begin(c->buffer));
 914                 LDNS_TC_SET(sldns_buffer_begin(c->buffer));
 915                 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer),
 916                         LDNS_RCODE_SERVFAIL);
 917                 sldns_buffer_set_position(c->buffer, LDNS_HEADER_SIZE);
 918                 sldns_buffer_write_at(c->buffer, 4,
 919                         (uint8_t*)"\0\0\0\0\0\0\0\0", 8);
 920                 sldns_buffer_flip(c->buffer);
 921                 goto send_reply;
 922         }
 923         if(worker->stats.extended)
 924                 server_stats_insquery(&worker->stats, c, qinfo.qtype,
 925                         qinfo.qclass, &edns, repinfo);
 926         if(c->type != comm_udp)
 927                 edns.udp_size = 65535; /* max size for TCP replies */
 928         if(qinfo.qclass == LDNS_RR_CLASS_CH && answer_chaos(worker, &qinfo,
 929                 &edns, c->buffer)) {
 930                 server_stats_insrcode(&worker->stats, c->buffer);
 931                 goto send_reply;
 932         }
 933         if(local_zones_answer(worker->daemon->local_zones, &qinfo, &edns,
 934                 c->buffer, worker->scratchpad, repinfo)) {
 935                 regional_free_all(worker->scratchpad);
 936                 if(sldns_buffer_limit(c->buffer) == 0) {
 937                         comm_point_drop_reply(repinfo);
 938                         return 0;
 939                 }
 940                 server_stats_insrcode(&worker->stats, c->buffer);
 941                 goto send_reply;
 942         }
 943
 944         /* We've looked in our local zones. If the answer isn't there, we
 945          * might need to bail out based on ACLs now. */
 946         if((ret=deny_refuse_non_local(c, acl, worker, repinfo)) != -1)
 947         {
 948                 if(ret == 1)
 949                         goto send_reply;
 950                 return ret;
 951         }
 952
 953         /* If this request does not have the recursion bit set, verify
 954          * ACLs allow the snooping. */
 955         if(!(LDNS_RD_WIRE(sldns_buffer_begin(c->buffer))) &&
 956                 acl != acl_allow_snoop ) {
 957                 sldns_buffer_set_limit(c->buffer, LDNS_HEADER_SIZE);
 958                 sldns_buffer_write_at(c->buffer, 4,
 959                         (uint8_t*)"\0\0\0\0\0\0\0\0", 8);
 960                 LDNS_QR_SET(sldns_buffer_begin(c->buffer));
 961                 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer),
 962                         LDNS_RCODE_REFUSED);
 963                 sldns_buffer_flip(c->buffer);
 964                 server_stats_insrcode(&worker->stats, c->buffer);
 965                 log_addr(VERB_ALGO, "refused nonrec (cache snoop) query from",
 966                         &repinfo->addr, repinfo->addrlen);
 967                 goto send_reply;
 968         }
 969         h = query_info_hash(&qinfo, sldns_buffer_read_u16_at(c->buffer, 2));
 970         if((e=slabhash_lookup(worker->env.msg_cache, h, &qinfo, 0))) {
 971                 /* answer from cache - we have acquired a readlock on it */
 972                 if(answer_from_cache(worker, &qinfo,
 973                         (struct reply_info*)e->data,
 974                         *(uint16_t*)(void *)sldns_buffer_begin(c->buffer),
 975                         sldns_buffer_read_u16_at(c->buffer, 2), repinfo,
 976                         &edns)) {
 977                         /* prefetch it if the prefetch TTL expired */
 978                         if(worker->env.cfg->prefetch && *worker->env.now >=
 979                                 ((struct reply_info*)e->data)->prefetch_ttl) {
 980                                 time_t leeway = ((struct reply_info*)e->
 981                                         data)->ttl - *worker->env.now;
 982                                 lock_rw_unlock(&e->lock);
 983                                 reply_and_prefetch(worker, &qinfo,
 984                                         sldns_buffer_read_u16_at(c->buffer, 2),
 985                                         repinfo, leeway);
 986                                 rc = 0;
 987                                 goto send_reply_rc;
 988                         }
 989                         lock_rw_unlock(&e->lock);
 990                         goto send_reply;
 991                 }
 992                 verbose(VERB_ALGO, "answer from the cache failed");
 993                 lock_rw_unlock(&e->lock);
 994         }
 995         if(!LDNS_RD_WIRE(sldns_buffer_begin(c->buffer))) {
 996                 if(answer_norec_from_cache(worker, &qinfo,
 997                         *(uint16_t*)(void *)sldns_buffer_begin(c->buffer),
 998                         sldns_buffer_read_u16_at(c->buffer, 2), repinfo,
 999                         &edns)) {
1000                         goto send_reply;
1001                 }
1002                 verbose(VERB_ALGO, "answer norec from cache -- "
1003                         "need to validate or not primed");
1004         }
1005         sldns_buffer_rewind(c->buffer);
1006         server_stats_querymiss(&worker->stats, worker);
1007
1008         if(verbosity >= VERB_CLIENT) {
1009                 if(c->type == comm_udp)
1010                         log_addr(VERB_CLIENT, "udp request from",
1011                                 &repinfo->addr, repinfo->addrlen);
1012                 else    log_addr(VERB_CLIENT, "tcp request from",
1013                                 &repinfo->addr, repinfo->addrlen);
1014         }
1015
1016         /* grab a work request structure for this new request */
1017         mesh_new_client(worker->env.mesh, &qinfo,
1018                 sldns_buffer_read_u16_at(c->buffer, 2),
1019                 &edns, repinfo, *(uint16_t*)(void *)sldns_buffer_begin(c->buffer));
1020         worker_mem_report(worker, NULL);
1021         return 0;
1022
1023 send_reply:
1024         rc = 1;
1025 send_reply_rc:
1026 #ifdef USE_DNSTAP
1027         if(worker->dtenv.log_client_response_messages)
1028                 dt_msg_send_client_response(&worker->dtenv, &repinfo->addr,
1029                         c->type, c->buffer);
1030 #endif
1031         return rc;
1032 }
1033
1034 void
1035 worker_sighandler(int sig, void* arg)
1036 {
1037         /* note that log, print, syscalls here give race conditions.
1038          * And cause hangups if the log-lock is held by the application. */
1039         struct worker* worker = (struct worker*)arg;
1040         switch(sig) {
1041 #ifdef SIGHUP
1042                 case SIGHUP:
1043                         comm_base_exit(worker->base);
1044                         break;
1045 #endif
1046                 case SIGINT:
1047                         worker->need_to_exit = 1;
1048                         comm_base_exit(worker->base);
1049                         break;
1050 #ifdef SIGQUIT
1051                 case SIGQUIT:
1052                         worker->need_to_exit = 1;
1053                         comm_base_exit(worker->base);
1054                         break;
1055 #endif
1056                 case SIGTERM:
1057                         worker->need_to_exit = 1;
1058                         comm_base_exit(worker->base);
1059                         break;
1060                 default:
1061                         /* unknown signal, ignored */
1062                         break;
1063         }
1064 }
1065
1066 /** restart statistics timer for worker, if enabled */
1067 static void
1068 worker_restart_timer(struct worker* worker)
1069 {
1070         if(worker->env.cfg->stat_interval > 0) {
1071                 struct timeval tv;
1072 #ifndef S_SPLINT_S
1073                 tv.tv_sec = worker->env.cfg->stat_interval;
1074                 tv.tv_usec = 0;
1075 #endif
1076                 comm_timer_set(worker->stat_timer, &tv);
1077         }
1078 }
1079
1080 void worker_stat_timer_cb(void* arg)
1081 {
1082         struct worker* worker = (struct worker*)arg;
1083         server_stats_log(&worker->stats, worker, worker->thread_num);
1084         mesh_stats(worker->env.mesh, "mesh has");
1085         worker_mem_report(worker, NULL);
1086         if(!worker->daemon->cfg->stat_cumulative) {
1087                 worker_stats_clear(worker);
1088         }
1089         /* start next timer */
1090         worker_restart_timer(worker);
1091 }
1092
1093 void worker_probe_timer_cb(void* arg)
1094 {
1095         struct worker* worker = (struct worker*)arg;
1096         struct timeval tv;
1097 #ifndef S_SPLINT_S
1098         tv.tv_sec = (time_t)autr_probe_timer(&worker->env);
1099         tv.tv_usec = 0;
1100 #endif
1101         if(tv.tv_sec != 0)
1102                 comm_timer_set(worker->env.probe_timer, &tv);
1103 }
1104
1105 struct worker*
1106 worker_create(struct daemon* daemon, int id, int* ports, int n)
1107 {
1108         unsigned int seed;
1109         struct worker* worker = (struct worker*)calloc(1,
1110                 sizeof(struct worker));
1111         if(!worker)
1112                 return NULL;
1113         worker->numports = n;
1114         worker->ports = (int*)memdup(ports, sizeof(int)*n);
1115         if(!worker->ports) {
1116                 free(worker);
1117                 return NULL;
1118         }
1119         worker->daemon = daemon;
1120         worker->thread_num = id;
1121         if(!(worker->cmd = tube_create())) {
1122                 free(worker->ports);
1123                 free(worker);
1124                 return NULL;
1125         }
1126         /* create random state here to avoid locking trouble in RAND_bytes */
1127         seed = (unsigned int)time(NULL) ^ (unsigned int)getpid() ^
1128                 (((unsigned int)worker->thread_num)<<17);
1129                 /* shift thread_num so it does not match out pid bits */
1130         if(!(worker->rndstate = ub_initstate(seed, daemon->rand))) {
1131                 seed = 0;
1132                 log_err("could not init random numbers.");
1133                 tube_delete(worker->cmd);
1134                 free(worker->ports);
1135                 free(worker);
1136                 return NULL;
1137         }
1138         seed = 0;
1139 #ifdef USE_DNSTAP
1140         if(daemon->cfg->dnstap) {
1141                 log_assert(daemon->dtenv != NULL);
1142                 memcpy(&worker->dtenv, daemon->dtenv, sizeof(struct dt_env));
1143                 if(!dt_init(&worker->dtenv))
1144                         fatal_exit("dt_init failed");
1145         }
1146 #endif
1147         return worker;
1148 }
1149
1150 int
1151 worker_init(struct worker* worker, struct config_file *cfg,
1152         struct listen_port* ports, int do_sigs)
1153 {
1154 #ifdef USE_DNSTAP
1155         struct dt_env* dtenv = &worker->dtenv;
1156 #else
1157         void* dtenv = NULL;
1158 #endif
1159         worker->need_to_exit = 0;
1160         worker->base = comm_base_create(do_sigs);
1161         if(!worker->base) {
1162                 log_err("could not create event handling base");
1163                 worker_delete(worker);
1164                 return 0;
1165         }
1166         comm_base_set_slow_accept_handlers(worker->base, &worker_stop_accept,
1167                 &worker_start_accept, worker);
1168         if(do_sigs) {
1169 #ifdef SIGHUP
1170                 ub_thread_sig_unblock(SIGHUP);
1171 #endif
1172                 ub_thread_sig_unblock(SIGINT);
1173 #ifdef SIGQUIT
1174                 ub_thread_sig_unblock(SIGQUIT);
1175 #endif
1176                 ub_thread_sig_unblock(SIGTERM);
1177 #ifndef LIBEVENT_SIGNAL_PROBLEM
1178                 worker->comsig = comm_signal_create(worker->base,
1179                         worker_sighandler, worker);
1180                 if(!worker->comsig
1181 #ifdef SIGHUP
1182                         || !comm_signal_bind(worker->comsig, SIGHUP)
1183 #endif
1184 #ifdef SIGQUIT
1185                         || !comm_signal_bind(worker->comsig, SIGQUIT)
1186 #endif
1187                         || !comm_signal_bind(worker->comsig, SIGTERM)
1188                         || !comm_signal_bind(worker->comsig, SIGINT)) {
1189                         log_err("could not create signal handlers");
1190                         worker_delete(worker);
1191                         return 0;
1192                 }
1193 #endif /* LIBEVENT_SIGNAL_PROBLEM */
1194                 if(!daemon_remote_open_accept(worker->daemon->rc,
1195                         worker->daemon->rc_ports, worker)) {
1196                         worker_delete(worker);
1197                         return 0;
1198                 }
1199 #ifdef UB_ON_WINDOWS
1200                 wsvc_setup_worker(worker);
1201 #endif /* UB_ON_WINDOWS */
1202         } else { /* !do_sigs */
1203                 worker->comsig = NULL;
1204         }
1205         worker->front = listen_create(worker->base, ports,
1206                 cfg->msg_buffer_size, (int)cfg->incoming_num_tcp,
1207                 worker->daemon->listen_sslctx, dtenv, worker_handle_request,
1208                 worker);
1209         if(!worker->front) {
1210                 log_err("could not create listening sockets");
1211                 worker_delete(worker);
1212                 return 0;
1213         }
1214         worker->back = outside_network_create(worker->base,
1215                 cfg->msg_buffer_size, (size_t)cfg->outgoing_num_ports,
1216                 cfg->out_ifs, cfg->num_out_ifs, cfg->do_ip4, cfg->do_ip6,
1217                 cfg->do_tcp?cfg->outgoing_num_tcp:0,
1218                 worker->daemon->env->infra_cache, worker->rndstate,
1219                 cfg->use_caps_bits_for_id, worker->ports, worker->numports,
1220                 cfg->unwanted_threshold, &worker_alloc_cleanup, worker,
1221                 cfg->do_udp, worker->daemon->connect_sslctx, cfg->delay_close,
1222                 dtenv);
1223         if(!worker->back) {
1224                 log_err("could not create outgoing sockets");
1225                 worker_delete(worker);
1226                 return 0;
1227         }
1228         /* start listening to commands */
1229         if(!tube_setup_bg_listen(worker->cmd, worker->base,
1230                 &worker_handle_control_cmd, worker)) {
1231                 log_err("could not create control compt.");
1232                 worker_delete(worker);
1233                 return 0;
1234         }
1235         worker->stat_timer = comm_timer_create(worker->base,
1236                 worker_stat_timer_cb, worker);
1237         if(!worker->stat_timer) {
1238                 log_err("could not create statistics timer");
1239         }
1240
1241         /* we use the msg_buffer_size as a good estimate for what the
1242          * user wants for memory usage sizes */
1243         worker->scratchpad = regional_create_custom(cfg->msg_buffer_size);
1244         if(!worker->scratchpad) {
1245                 log_err("malloc failure");
1246                 worker_delete(worker);
1247                 return 0;
1248         }
1249
1250         server_stats_init(&worker->stats, cfg);
1251         alloc_init(&worker->alloc, &worker->daemon->superalloc,
1252                 worker->thread_num);
1253         alloc_set_id_cleanup(&worker->alloc, &worker_alloc_cleanup, worker);
1254         worker->env = *worker->daemon->env;
1255         comm_base_timept(worker->base, &worker->env.now, &worker->env.now_tv);
1256         if(worker->thread_num == 0)
1257                 log_set_time(worker->env.now);
1258         worker->env.worker = worker;
1259         worker->env.send_query = &worker_send_query;
1260         worker->env.alloc = &worker->alloc;
1261         worker->env.rnd = worker->rndstate;
1262         worker->env.scratch = worker->scratchpad;
1263         worker->env.mesh = mesh_create(&worker->daemon->mods, &worker->env);
1264         worker->env.detach_subs = &mesh_detach_subs;
1265         worker->env.attach_sub = &mesh_attach_sub;
1266         worker->env.kill_sub = &mesh_state_delete;
1267         worker->env.detect_cycle = &mesh_detect_cycle;
1268         worker->env.scratch_buffer = sldns_buffer_new(cfg->msg_buffer_size);
1269         if(!(worker->env.fwds = forwards_create()) ||
1270                 !forwards_apply_cfg(worker->env.fwds, cfg)) {
1271                 log_err("Could not set forward zones");
1272                 worker_delete(worker);
1273                 return 0;
1274         }
1275         if(!(worker->env.hints = hints_create()) ||
1276                 !hints_apply_cfg(worker->env.hints, cfg)) {
1277                 log_err("Could not set root or stub hints");
1278                 worker_delete(worker);
1279                 return 0;
1280         }
1281         /* one probe timer per process -- if we have 5011 anchors */
1282         if(autr_get_num_anchors(worker->env.anchors) > 0
1283 #ifndef THREADS_DISABLED
1284                 && worker->thread_num == 0
1285 #endif
1286                 ) {
1287                 struct timeval tv;
1288                 tv.tv_sec = 0;
1289                 tv.tv_usec = 0;
1290                 worker->env.probe_timer = comm_timer_create(worker->base,
1291                         worker_probe_timer_cb, worker);
1292                 if(!worker->env.probe_timer) {
1293                         log_err("could not create 5011-probe timer");
1294                 } else {
1295                         /* let timer fire, then it can reset itself */
1296                         comm_timer_set(worker->env.probe_timer, &tv);
1297                 }
1298         }
1299         if(!worker->env.mesh || !worker->env.scratch_buffer) {
1300                 worker_delete(worker);
1301                 return 0;
1302         }
1303         worker_mem_report(worker, NULL);
1304         /* if statistics enabled start timer */
1305         if(worker->env.cfg->stat_interval > 0) {
1306                 verbose(VERB_ALGO, "set statistics interval %d secs",
1307                         worker->env.cfg->stat_interval);
1308                 worker_restart_timer(worker);
1309         }
1310         return 1;
1311 }
1312
1313 void
1314 worker_work(struct worker* worker)
1315 {
1316         comm_base_dispatch(worker->base);
1317 }
1318
1319 void
1320 worker_delete(struct worker* worker)
1321 {
1322         if(!worker)
1323                 return;
1324         if(worker->env.mesh && verbosity >= VERB_OPS) {
1325                 server_stats_log(&worker->stats, worker, worker->thread_num);
1326                 mesh_stats(worker->env.mesh, "mesh has");
1327                 worker_mem_report(worker, NULL);
1328         }
1329         outside_network_quit_prepare(worker->back);
1330         mesh_delete(worker->env.mesh);
1331         sldns_buffer_free(worker->env.scratch_buffer);
1332         forwards_delete(worker->env.fwds);
1333         hints_delete(worker->env.hints);
1334         listen_delete(worker->front);
1335         outside_network_delete(worker->back);
1336         comm_signal_delete(worker->comsig);
1337         tube_delete(worker->cmd);
1338         comm_timer_delete(worker->stat_timer);
1339         comm_timer_delete(worker->env.probe_timer);
1340         free(worker->ports);
1341         if(worker->thread_num == 0) {
1342                 log_set_time(NULL);
1343 #ifdef UB_ON_WINDOWS
1344                 wsvc_desetup_worker(worker);
1345 #endif /* UB_ON_WINDOWS */
1346         }
1347         comm_base_delete(worker->base);
1348         ub_randfree(worker->rndstate);
1349         alloc_clear(&worker->alloc);
1350         regional_destroy(worker->scratchpad);
1351         free(worker);
1352 }
1353
1354 struct outbound_entry*
1355 worker_send_query(uint8_t* qname, size_t qnamelen, uint16_t qtype,
1356         uint16_t qclass, uint16_t flags, int dnssec, int want_dnssec,
1357         int nocaps, struct sockaddr_storage* addr, socklen_t addrlen,
1358         uint8_t* zone, size_t zonelen, struct module_qstate* q)
1359 {
1360         struct worker* worker = q->env->worker;
1361         struct outbound_entry* e = (struct outbound_entry*)regional_alloc(
1362                 q->region, sizeof(*e));
1363         if(!e)
1364                 return NULL;
1365         e->qstate = q;
1366         e->qsent = outnet_serviced_query(worker->back, qname,
1367                 qnamelen, qtype, qclass, flags, dnssec, want_dnssec, nocaps,
1368                 q->env->cfg->tcp_upstream, q->env->cfg->ssl_upstream, addr,
1369                 addrlen, zone, zonelen, worker_handle_service_reply, e,
1370                 worker->back->udp_buff);
1371         if(!e->qsent) {
1372                 return NULL;
1373         }
1374         return e;
1375 }
1376
1377 void
1378 worker_alloc_cleanup(void* arg)
1379 {
1380         struct worker* worker = (struct worker*)arg;
1381         slabhash_clear(&worker->env.rrset_cache->table);
1382         slabhash_clear(worker->env.msg_cache);
1383 }
1384
1385 void worker_stats_clear(struct worker* worker)
1386 {
1387         server_stats_init(&worker->stats, worker->env.cfg);
1388         mesh_stats_clear(worker->env.mesh);
1389         worker->back->unwanted_replies = 0;
1390         worker->back->num_tcp_outgoing = 0;
1391 }
1392
1393 void worker_start_accept(void* arg)
1394 {
1395         struct worker* worker = (struct worker*)arg;
1396         listen_start_accept(worker->front);
1397         if(worker->thread_num == 0)
1398                 daemon_remote_start_accept(worker->daemon->rc);
1399 }
1400
1401 void worker_stop_accept(void* arg)
1402 {
1403         struct worker* worker = (struct worker*)arg;
1404         listen_stop_accept(worker->front);
1405         if(worker->thread_num == 0)
1406                 daemon_remote_stop_accept(worker->daemon->rc);
1407 }
1408
1409 /* --- fake callbacks for fptr_wlist to work --- */
1410 struct outbound_entry* libworker_send_query(uint8_t* ATTR_UNUSED(qname),
1411         size_t ATTR_UNUSED(qnamelen), uint16_t ATTR_UNUSED(qtype),
1412         uint16_t ATTR_UNUSED(qclass), uint16_t ATTR_UNUSED(flags),
1413         int ATTR_UNUSED(dnssec), int ATTR_UNUSED(want_dnssec),
1414         int ATTR_UNUSED(nocaps), struct sockaddr_storage* ATTR_UNUSED(addr),
1415         socklen_t ATTR_UNUSED(addrlen), uint8_t* ATTR_UNUSED(zone),
1416         size_t ATTR_UNUSED(zonelen), struct module_qstate* ATTR_UNUSED(q))
1417 {
1418         log_assert(0);
1419         return 0;
1420 }
1421
1422 int libworker_handle_reply(struct comm_point* ATTR_UNUSED(c),
1423         void* ATTR_UNUSED(arg), int ATTR_UNUSED(error),
1424         struct comm_reply* ATTR_UNUSED(reply_info))
1425 {
1426         log_assert(0);
1427         return 0;
1428 }
1429
1430 int libworker_handle_service_reply(struct comm_point* ATTR_UNUSED(c),
1431         void* ATTR_UNUSED(arg), int ATTR_UNUSED(error),
1432         struct comm_reply* ATTR_UNUSED(reply_info))
1433 {
1434         log_assert(0);
1435         return 0;
1436 }
1437
1438 void libworker_handle_control_cmd(struct tube* ATTR_UNUSED(tube),
1439         uint8_t* ATTR_UNUSED(buffer), size_t ATTR_UNUSED(len),
1440         int ATTR_UNUSED(error), void* ATTR_UNUSED(arg))
1441 {
1442         log_assert(0);
1443 }
1444
1445 void libworker_fg_done_cb(void* ATTR_UNUSED(arg), int ATTR_UNUSED(rcode),
1446         sldns_buffer* ATTR_UNUSED(buf), enum sec_status ATTR_UNUSED(s),
1447         char* ATTR_UNUSED(why_bogus))
1448 {
1449         log_assert(0);
1450 }
1451
1452 void libworker_bg_done_cb(void* ATTR_UNUSED(arg), int ATTR_UNUSED(rcode),
1453         sldns_buffer* ATTR_UNUSED(buf), enum sec_status ATTR_UNUSED(s),
1454         char* ATTR_UNUSED(why_bogus))
1455 {
1456         log_assert(0);
1457 }
1458
1459 void libworker_event_done_cb(void* ATTR_UNUSED(arg), int ATTR_UNUSED(rcode),
1460         sldns_buffer* ATTR_UNUSED(buf), enum sec_status ATTR_UNUSED(s),
1461         char* ATTR_UNUSED(why_bogus))
1462 {
1463         log_assert(0);
1464 }
1465
1466 int context_query_cmp(const void* ATTR_UNUSED(a), const void* ATTR_UNUSED(b))
1467 {
1468         log_assert(0);
1469         return 0;
1470 }
1471
1472 int order_lock_cmp(const void* ATTR_UNUSED(e1), const void* ATTR_UNUSED(e2))
1473 {
1474         log_assert(0);
1475         return 0;
1476 }
1477
1478 int codeline_cmp(const void* ATTR_UNUSED(a), const void* ATTR_UNUSED(b))
1479 {
1480         log_assert(0);
1481         return 0;
1482 }
1483