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 /** Report on memory usage by this thread and global */
 105 static void
 106 worker_mem_report(struct worker* ATTR_UNUSED(worker),
 107         struct serviced_query* ATTR_UNUSED(cur_serv))
 108 {
 109 #ifdef UNBOUND_ALLOC_STATS
 110         /* measure memory leakage */
 111         extern size_t unbound_mem_alloc, unbound_mem_freed;
 112         /* debug func in validator module */
 113         size_t total, front, back, mesh, msg, rrset, infra, ac, superac;
 114         size_t me, iter, val, anch;
 115         int i;
 116         if(verbosity < VERB_ALGO)
 117                 return;
 118         front = listen_get_mem(worker->front);
 119         back = outnet_get_mem(worker->back);
 120         msg = slabhash_get_mem(worker->env.msg_cache);
 121         rrset = slabhash_get_mem(&worker->env.rrset_cache->table);
 122         infra = infra_get_mem(worker->env.infra_cache);
 123         mesh = mesh_get_mem(worker->env.mesh);
 124         ac = alloc_get_mem(&worker->alloc);
 125         superac = alloc_get_mem(&worker->daemon->superalloc);
 126         anch = anchors_get_mem(worker->env.anchors);
 127         iter = 0;
 128         val = 0;
 129         for(i=0; i<worker->env.mesh->mods.num; i++) {
 130                 fptr_ok(fptr_whitelist_mod_get_mem(worker->env.mesh->
 131                         mods.mod[i]->get_mem));
 132                 if(strcmp(worker->env.mesh->mods.mod[i]->name, "validator")==0)
 133                         val += (*worker->env.mesh->mods.mod[i]->get_mem)
 134                                 (&worker->env, i);
 135                 else    iter += (*worker->env.mesh->mods.mod[i]->get_mem)
 136                                 (&worker->env, i);
 137         }
 138         me = sizeof(*worker) + sizeof(*worker->base) + sizeof(*worker->comsig)
 139                 + comm_point_get_mem(worker->cmd_com)
 140                 + sizeof(worker->rndstate)
 141                 + regional_get_mem(worker->scratchpad)
 142                 + sizeof(*worker->env.scratch_buffer)
 143                 + sldns_buffer_capacity(worker->env.scratch_buffer)
 144                 + forwards_get_mem(worker->env.fwds)
 145                 + hints_get_mem(worker->env.hints);
 146         if(worker->thread_num == 0)
 147                 me += acl_list_get_mem(worker->daemon->acl);
 148         if(cur_serv) {
 149                 me += serviced_get_mem(cur_serv);
 150         }
 151         total = front+back+mesh+msg+rrset+infra+iter+val+ac+superac+me;
 152         log_info("Memory conditions: %u front=%u back=%u mesh=%u msg=%u "
 153                 "rrset=%u infra=%u iter=%u val=%u anchors=%u "
 154                 "alloccache=%u globalalloccache=%u me=%u",
 155                 (unsigned)total, (unsigned)front, (unsigned)back,
 156                 (unsigned)mesh, (unsigned)msg, (unsigned)rrset,
 157                 (unsigned)infra, (unsigned)iter, (unsigned)val, (unsigned)anch,
 158                 (unsigned)ac, (unsigned)superac, (unsigned)me);
 159         log_info("Total heap memory estimate: %u  total-alloc: %u  "
 160                 "total-free: %u", (unsigned)total,
 161                 (unsigned)unbound_mem_alloc, (unsigned)unbound_mem_freed);
 162 #else /* no UNBOUND_ALLOC_STATS */
 163         size_t val = 0;
 164         int i;
 165         if(verbosity < VERB_QUERY)
 166                 return;
 167         for(i=0; i<worker->env.mesh->mods.num; i++) {
 168                 fptr_ok(fptr_whitelist_mod_get_mem(worker->env.mesh->
 169                         mods.mod[i]->get_mem));
 170                 if(strcmp(worker->env.mesh->mods.mod[i]->name, "validator")==0)
 171                         val += (*worker->env.mesh->mods.mod[i]->get_mem)
 172                                 (&worker->env, i);
 173         }
 174         verbose(VERB_QUERY, "cache memory msg=%u rrset=%u infra=%u val=%u",
 175                 (unsigned)slabhash_get_mem(worker->env.msg_cache),
 176                 (unsigned)slabhash_get_mem(&worker->env.rrset_cache->table),
 177                 (unsigned)infra_get_mem(worker->env.infra_cache),
 178                 (unsigned)val);
 179 #endif /* UNBOUND_ALLOC_STATS */
 180 }
 181
 182 void
 183 worker_send_cmd(struct worker* worker, enum worker_commands cmd)
 184 {
 185         uint32_t c = (uint32_t)htonl(cmd);
 186         if(!tube_write_msg(worker->cmd, (uint8_t*)&c, sizeof(c), 0)) {
 187                 log_err("worker send cmd %d failed", (int)cmd);
 188         }
 189 }
 190
 191 int
 192 worker_handle_reply(struct comm_point* c, void* arg, int error,
 193         struct comm_reply* reply_info)
 194 {
 195         struct module_qstate* q = (struct module_qstate*)arg;
 196         struct worker* worker = q->env->worker;
 197         struct outbound_entry e;
 198         e.qstate = q;
 199         e.qsent = NULL;
 200
 201         if(error != 0) {
 202                 mesh_report_reply(worker->env.mesh, &e, reply_info, error);
 203                 worker_mem_report(worker, NULL);
 204                 return 0;
 205         }
 206         /* sanity check. */
 207         if(!LDNS_QR_WIRE(sldns_buffer_begin(c->buffer))
 208                 || LDNS_OPCODE_WIRE(sldns_buffer_begin(c->buffer)) !=
 209                         LDNS_PACKET_QUERY
 210                 || LDNS_QDCOUNT(sldns_buffer_begin(c->buffer)) > 1) {
 211                 /* error becomes timeout for the module as if this reply
 212                  * never arrived. */
 213                 mesh_report_reply(worker->env.mesh, &e, reply_info,
 214                         NETEVENT_TIMEOUT);
 215                 worker_mem_report(worker, NULL);
 216                 return 0;
 217         }
 218         mesh_report_reply(worker->env.mesh, &e, reply_info, NETEVENT_NOERROR);
 219         worker_mem_report(worker, NULL);
 220         return 0;
 221 }
 222
 223 int
 224 worker_handle_service_reply(struct comm_point* c, void* arg, int error,
 225         struct comm_reply* reply_info)
 226 {
 227         struct outbound_entry* e = (struct outbound_entry*)arg;
 228         struct worker* worker = e->qstate->env->worker;
 229         struct serviced_query *sq = e->qsent;
 230
 231         verbose(VERB_ALGO, "worker svcd callback for qstate %p", e->qstate);
 232         if(error != 0) {
 233                 mesh_report_reply(worker->env.mesh, e, reply_info, error);
 234                 worker_mem_report(worker, sq);
 235                 return 0;
 236         }
 237         /* sanity check. */
 238         if(!LDNS_QR_WIRE(sldns_buffer_begin(c->buffer))
 239                 || LDNS_OPCODE_WIRE(sldns_buffer_begin(c->buffer)) !=
 240                         LDNS_PACKET_QUERY
 241                 || LDNS_QDCOUNT(sldns_buffer_begin(c->buffer)) > 1) {
 242                 /* error becomes timeout for the module as if this reply
 243                  * never arrived. */
 244                 verbose(VERB_ALGO, "worker: bad reply handled as timeout");
 245                 mesh_report_reply(worker->env.mesh, e, reply_info,
 246                         NETEVENT_TIMEOUT);
 247                 worker_mem_report(worker, sq);
 248                 return 0;
 249         }
 250         mesh_report_reply(worker->env.mesh, e, reply_info, NETEVENT_NOERROR);
 251         worker_mem_report(worker, sq);
 252         return 0;
 253 }
 254
 255 /** ratelimit error replies
 256  * @param worker: the worker struct with ratelimit counter
 257  * @param err: error code that would be wanted.
 258  * @return value of err if okay, or -1 if it should be discarded instead.
 259  */
 260 static int
 261 worker_err_ratelimit(struct worker* worker, int err)
 262 {
 263         if(worker->err_limit_time == *worker->env.now) {
 264                 /* see if limit is exceeded for this second */
 265                 if(worker->err_limit_count++ > ERROR_RATELIMIT)
 266                         return -1;
 267         } else {
 268                 /* new second, new limits */
 269                 worker->err_limit_time = *worker->env.now;
 270                 worker->err_limit_count = 1;
 271         }
 272         return err;
 273 }
 274
 275 /** check request sanity.
 276  * @param pkt: the wire packet to examine for sanity.
 277  * @param worker: parameters for checking.
 278  * @return error code, 0 OK, or -1 discard.
 279 */
 280 static int
 281 worker_check_request(sldns_buffer* pkt, struct worker* worker)
 282 {
 283         if(sldns_buffer_limit(pkt) < LDNS_HEADER_SIZE) {
 284                 verbose(VERB_QUERY, "request too short, discarded");
 285                 return -1;
 286         }
 287         if(sldns_buffer_limit(pkt) > NORMAL_UDP_SIZE &&
 288                 worker->daemon->cfg->harden_large_queries) {
 289                 verbose(VERB_QUERY, "request too large, discarded");
 290                 return -1;
 291         }
 292         if(LDNS_QR_WIRE(sldns_buffer_begin(pkt))) {
 293                 verbose(VERB_QUERY, "request has QR bit on, discarded");
 294                 return -1;
 295         }
 296         if(LDNS_TC_WIRE(sldns_buffer_begin(pkt))) {
 297                 LDNS_TC_CLR(sldns_buffer_begin(pkt));
 298                 verbose(VERB_QUERY, "request bad, has TC bit on");
 299                 return worker_err_ratelimit(worker, LDNS_RCODE_FORMERR);
 300         }
 301         if(LDNS_OPCODE_WIRE(sldns_buffer_begin(pkt)) != LDNS_PACKET_QUERY) {
 302                 verbose(VERB_QUERY, "request unknown opcode %d",
 303                         LDNS_OPCODE_WIRE(sldns_buffer_begin(pkt)));
 304                 return worker_err_ratelimit(worker, LDNS_RCODE_NOTIMPL);
 305         }
 306         if(LDNS_QDCOUNT(sldns_buffer_begin(pkt)) != 1) {
 307                 verbose(VERB_QUERY, "request wrong nr qd=%d",
 308                         LDNS_QDCOUNT(sldns_buffer_begin(pkt)));
 309                 return worker_err_ratelimit(worker, LDNS_RCODE_FORMERR);
 310         }
 311         if(LDNS_ANCOUNT(sldns_buffer_begin(pkt)) != 0) {
 312                 verbose(VERB_QUERY, "request wrong nr an=%d",
 313                         LDNS_ANCOUNT(sldns_buffer_begin(pkt)));
 314                 return worker_err_ratelimit(worker, LDNS_RCODE_FORMERR);
 315         }
 316         if(LDNS_NSCOUNT(sldns_buffer_begin(pkt)) != 0) {
 317                 verbose(VERB_QUERY, "request wrong nr ns=%d",
 318                         LDNS_NSCOUNT(sldns_buffer_begin(pkt)));
 319                 return worker_err_ratelimit(worker, LDNS_RCODE_FORMERR);
 320         }
 321         if(LDNS_ARCOUNT(sldns_buffer_begin(pkt)) > 1) {
 322                 verbose(VERB_QUERY, "request wrong nr ar=%d",
 323                         LDNS_ARCOUNT(sldns_buffer_begin(pkt)));
 324                 return worker_err_ratelimit(worker, LDNS_RCODE_FORMERR);
 325         }
 326         return 0;
 327 }
 328
 329 void
 330 worker_handle_control_cmd(struct tube* ATTR_UNUSED(tube), uint8_t* msg,
 331         size_t len, int error, void* arg)
 332 {
 333         struct worker* worker = (struct worker*)arg;
 334         enum worker_commands cmd;
 335         if(error != NETEVENT_NOERROR) {
 336                 free(msg);
 337                 if(error == NETEVENT_CLOSED)
 338                         comm_base_exit(worker->base);
 339                 else    log_info("control event: %d", error);
 340                 return;
 341         }
 342         if(len != sizeof(uint32_t)) {
 343                 fatal_exit("bad control msg length %d", (int)len);
 344         }
 345         cmd = sldns_read_uint32(msg);
 346         free(msg);
 347         switch(cmd) {
 348         case worker_cmd_quit:
 349                 verbose(VERB_ALGO, "got control cmd quit");
 350                 comm_base_exit(worker->base);
 351                 break;
 352         case worker_cmd_stats:
 353                 verbose(VERB_ALGO, "got control cmd stats");
 354                 server_stats_reply(worker, 1);
 355                 break;
 356         case worker_cmd_stats_noreset:
 357                 verbose(VERB_ALGO, "got control cmd stats_noreset");
 358                 server_stats_reply(worker, 0);
 359                 break;
 360         case worker_cmd_remote:
 361                 verbose(VERB_ALGO, "got control cmd remote");
 362                 daemon_remote_exec(worker);
 363                 break;
 364         default:
 365                 log_err("bad command %d", (int)cmd);
 366                 break;
 367         }
 368 }
 369
 370 /** check if a delegation is secure */
 371 static enum sec_status
 372 check_delegation_secure(struct reply_info *rep)
 373 {
 374         /* return smallest security status */
 375         size_t i;
 376         enum sec_status sec = sec_status_secure;
 377         enum sec_status s;
 378         size_t num = rep->an_numrrsets + rep->ns_numrrsets;
 379         /* check if answer and authority are OK */
 380         for(i=0; i<num; i++) {
 381                 s = ((struct packed_rrset_data*)rep->rrsets[i]->entry.data)
 382                         ->security;
 383                 if(s < sec)
 384                         sec = s;
 385         }
 386         /* in additional, only unchecked triggers revalidation */
 387         for(i=num; i<rep->rrset_count; i++) {
 388                 s = ((struct packed_rrset_data*)rep->rrsets[i]->entry.data)
 389                         ->security;
 390                 if(s == sec_status_unchecked)
 391                         return s;
 392         }
 393         return sec;
 394 }
 395
 396 /** remove nonsecure from a delegation referral additional section */
 397 static void
 398 deleg_remove_nonsecure_additional(struct reply_info* rep)
 399 {
 400         /* we can simply edit it, since we are working in the scratch region */
 401         size_t i;
 402         enum sec_status s;
 403
 404         for(i = rep->an_numrrsets+rep->ns_numrrsets; i<rep->rrset_count; i++) {
 405                 s = ((struct packed_rrset_data*)rep->rrsets[i]->entry.data)
 406                         ->security;
 407                 if(s != sec_status_secure) {
 408                         memmove(rep->rrsets+i, rep->rrsets+i+1,
 409                                 sizeof(struct ub_packed_rrset_key*)*
 410                                 (rep->rrset_count - i - 1));
 411                         rep->ar_numrrsets--;
 412                         rep->rrset_count--;
 413                         i--;
 414                 }
 415         }
 416 }
 417
 418 /** answer nonrecursive query from the cache */
 419 static int
 420 answer_norec_from_cache(struct worker* worker, struct query_info* qinfo,
 421         uint16_t id, uint16_t flags, struct comm_reply* repinfo,
 422         struct edns_data* edns)
 423 {
 424         /* for a nonrecursive query return either:
 425          *      o an error (servfail; we try to avoid this)
 426          *      o a delegation (closest we have; this routine tries that)
 427          *      o the answer (checked by answer_from_cache)
 428          *
 429          * So, grab a delegation from the rrset cache.
 430          * Then check if it needs validation, if so, this routine fails,
 431          * so that iterator can prime and validator can verify rrsets.
 432          */
 433         uint16_t udpsize = edns->udp_size;
 434         int secure = 0;
 435         time_t timenow = *worker->env.now;
 436         int must_validate = (!(flags&BIT_CD) || worker->env.cfg->ignore_cd)
 437                 && worker->env.need_to_validate;
 438         struct dns_msg *msg = NULL;
 439         struct delegpt *dp;
 440
 441         dp = dns_cache_find_delegation(&worker->env, qinfo->qname,
 442                 qinfo->qname_len, qinfo->qtype, qinfo->qclass,
 443                 worker->scratchpad, &msg, timenow);
 444         if(!dp) { /* no delegation, need to reprime */
 445                 return 0;
 446         }
 447         /* In case we have a local alias, copy it into the delegation message.
 448          * Shallow copy should be fine, as we'll be done with msg in this
 449          * function. */
 450         msg->qinfo.local_alias = qinfo->local_alias;
 451         if(must_validate) {
 452                 switch(check_delegation_secure(msg->rep)) {
 453                 case sec_status_unchecked:
 454                         /* some rrsets have not been verified yet, go and
 455                          * let validator do that */
 456                         return 0;
 457                 case sec_status_bogus:
 458                         /* some rrsets are bogus, reply servfail */
 459                         edns->edns_version = EDNS_ADVERTISED_VERSION;
 460                         edns->udp_size = EDNS_ADVERTISED_SIZE;
 461                         edns->ext_rcode = 0;
 462                         edns->bits &= EDNS_DO;
 463                         if(!inplace_cb_reply_servfail_call(&worker->env, qinfo, NULL,
 464                                 msg->rep, LDNS_RCODE_SERVFAIL, edns, worker->scratchpad))
 465                                         return 0;
 466                         error_encode(repinfo->c->buffer, LDNS_RCODE_SERVFAIL,
 467                                 &msg->qinfo, id, flags, edns);
 468                         if(worker->stats.extended) {
 469                                 worker->stats.ans_bogus++;
 470                                 worker->stats.ans_rcode[LDNS_RCODE_SERVFAIL]++;
 471                         }
 472                         return 1;
 473                 case sec_status_secure:
 474                         /* all rrsets are secure */
 475                         /* remove non-secure rrsets from the add. section*/
 476                         if(worker->env.cfg->val_clean_additional)
 477                                 deleg_remove_nonsecure_additional(msg->rep);
 478                         secure = 1;
 479                         break;
 480                 case sec_status_indeterminate:
 481                 case sec_status_insecure:
 482                 default:
 483                         /* not secure */
 484                         secure = 0;
 485                         break;
 486                 }
 487         }
 488         /* return this delegation from the cache */
 489         edns->edns_version = EDNS_ADVERTISED_VERSION;
 490         edns->udp_size = EDNS_ADVERTISED_SIZE;
 491         edns->ext_rcode = 0;
 492         edns->bits &= EDNS_DO;
 493         if(!inplace_cb_reply_cache_call(&worker->env, qinfo, NULL, msg->rep,
 494                 (int)(flags&LDNS_RCODE_MASK), edns, worker->scratchpad))
 495                         return 0;
 496         msg->rep->flags |= BIT_QR|BIT_RA;
 497         if(!reply_info_answer_encode(&msg->qinfo, msg->rep, id, flags,
 498                 repinfo->c->buffer, 0, 1, worker->scratchpad,
 499                 udpsize, edns, (int)(edns->bits & EDNS_DO), secure)) {
 500                 if(!inplace_cb_reply_servfail_call(&worker->env, qinfo, NULL, NULL,
 501                         LDNS_RCODE_SERVFAIL, edns, worker->scratchpad))
 502                                 edns->opt_list = NULL;
 503                 error_encode(repinfo->c->buffer, LDNS_RCODE_SERVFAIL,
 504                         &msg->qinfo, id, flags, edns);
 505         }
 506         if(worker->stats.extended) {
 507                 if(secure) worker->stats.ans_secure++;
 508                 server_stats_insrcode(&worker->stats, repinfo->c->buffer);
 509         }
 510         return 1;
 511 }
 512
 513 /** answer query from the cache */
 514 static int
 515 answer_from_cache(struct worker* worker, struct query_info* qinfo,
 516         struct reply_info* rep, uint16_t id, uint16_t flags,
 517         struct comm_reply* repinfo, struct edns_data* edns)
 518 {
 519         time_t timenow = *worker->env.now;
 520         uint16_t udpsize = edns->udp_size;
 521         int secure;
 522         int must_validate = (!(flags&BIT_CD) || worker->env.cfg->ignore_cd)
 523                 && worker->env.need_to_validate;
 524         if(worker->env.cfg->serve_expired) {
 525                 /* always lock rrsets, rep->ttl is ignored */
 526                 if(!rrset_array_lock(rep->ref, rep->rrset_count, 0))
 527                         return 0;
 528                 /* below, rrsets with ttl before timenow become TTL 0 in
 529                  * the response */
 530                 /* This response was served with zero TTL */
 531                 if (timenow >= rep->ttl) {
 532                         worker->stats.zero_ttl_responses++;
 533                 }
 534         } else {
 535                 /* see if it is possible */
 536                 if(rep->ttl < timenow) {
 537                         /* the rrsets may have been updated in the meantime.
 538                          * we will refetch the message format from the
 539                          * authoritative server
 540                          */
 541                         return 0;
 542                 }
 543                 if(!rrset_array_lock(rep->ref, rep->rrset_count, timenow))
 544                         return 0;
 545                 /* locked and ids and ttls are OK. */
 546         }
 547         /* check CNAME chain (if any) */
 548         if(rep->an_numrrsets > 0 && (rep->rrsets[0]->rk.type ==
 549                 htons(LDNS_RR_TYPE_CNAME) || rep->rrsets[0]->rk.type ==
 550                 htons(LDNS_RR_TYPE_DNAME))) {
 551                 if(!reply_check_cname_chain(qinfo, rep)) {
 552                         /* cname chain invalid, redo iterator steps */
 553                         verbose(VERB_ALGO, "Cache reply: cname chain broken");
 554                 bail_out:
 555                         rrset_array_unlock_touch(worker->env.rrset_cache,
 556                                 worker->scratchpad, rep->ref, rep->rrset_count);
 557                         return 0;
 558                 }
 559         }
 560         /* check security status of the cached answer */
 561         if( rep->security == sec_status_bogus && must_validate) {
 562                 /* BAD cached */
 563                 edns->edns_version = EDNS_ADVERTISED_VERSION;
 564                 edns->udp_size = EDNS_ADVERTISED_SIZE;
 565                 edns->ext_rcode = 0;
 566                 edns->bits &= EDNS_DO;
 567                 if(!inplace_cb_reply_servfail_call(&worker->env, qinfo, NULL, rep,
 568                         LDNS_RCODE_SERVFAIL, edns, worker->scratchpad))
 569                         goto bail_out;
 570                 error_encode(repinfo->c->buffer, LDNS_RCODE_SERVFAIL,
 571                         qinfo, id, flags, edns);
 572                 rrset_array_unlock_touch(worker->env.rrset_cache,
 573                         worker->scratchpad, rep->ref, rep->rrset_count);
 574                 if(worker->stats.extended) {
 575                         worker->stats.ans_bogus ++;
 576                         worker->stats.ans_rcode[LDNS_RCODE_SERVFAIL] ++;
 577                 }
 578                 return 1;
 579         } else if( rep->security == sec_status_unchecked && must_validate) {
 580                 verbose(VERB_ALGO, "Cache reply: unchecked entry needs "
 581                         "validation");
 582                 goto bail_out; /* need to validate cache entry first */
 583         } else if(rep->security == sec_status_secure) {
 584                 if(reply_all_rrsets_secure(rep))
 585                         secure = 1;
 586                 else    {
 587                         if(must_validate) {
 588                                 verbose(VERB_ALGO, "Cache reply: secure entry"
 589                                         " changed status");
 590                                 goto bail_out; /* rrset changed, re-verify */
 591                         }
 592                         secure = 0;
 593                 }
 594         } else  secure = 0;
 595
 596         edns->edns_version = EDNS_ADVERTISED_VERSION;
 597         edns->udp_size = EDNS_ADVERTISED_SIZE;
 598         edns->ext_rcode = 0;
 599         edns->bits &= EDNS_DO;
 600         if(!inplace_cb_reply_cache_call(&worker->env, qinfo, NULL, rep,
 601                 (int)(flags&LDNS_RCODE_MASK), edns, worker->scratchpad))
 602                 goto bail_out;
 603         if(!reply_info_answer_encode(qinfo, rep, id, flags,
 604                 repinfo->c->buffer, timenow, 1, worker->scratchpad,
 605                 udpsize, edns, (int)(edns->bits & EDNS_DO), secure)) {
 606                 if(!inplace_cb_reply_servfail_call(&worker->env, qinfo, NULL, NULL,
 607                         LDNS_RCODE_SERVFAIL, edns, worker->scratchpad))
 608                                 edns->opt_list = NULL;
 609                 error_encode(repinfo->c->buffer, LDNS_RCODE_SERVFAIL,
 610                         qinfo, id, flags, edns);
 611         }
 612         /* cannot send the reply right now, because blocking network syscall
 613          * is bad while holding locks. */
 614         rrset_array_unlock_touch(worker->env.rrset_cache, worker->scratchpad,
 615                 rep->ref, rep->rrset_count);
 616         if(worker->stats.extended) {
 617                 if(secure) worker->stats.ans_secure++;
 618                 server_stats_insrcode(&worker->stats, repinfo->c->buffer);
 619         }
 620         /* go and return this buffer to the client */
 621         return 1;
 622 }
 623
 624 /** Reply to client and perform prefetch to keep cache up to date */
 625 static void
 626 reply_and_prefetch(struct worker* worker, struct query_info* qinfo,
 627         uint16_t flags, struct comm_reply* repinfo, time_t leeway)
 628 {
 629         /* first send answer to client to keep its latency
 630          * as small as a cachereply */
 631         comm_point_send_reply(repinfo);
 632         server_stats_prefetch(&worker->stats, worker);
 633
 634         /* create the prefetch in the mesh as a normal lookup without
 635          * client addrs waiting, which has the cache blacklisted (to bypass
 636          * the cache and go to the network for the data). */
 637         /* this (potentially) runs the mesh for the new query */
 638         mesh_new_prefetch(worker->env.mesh, qinfo, flags, leeway +
 639                 PREFETCH_EXPIRY_ADD);
 640 }
 641
 642 /**
 643  * Fill CH class answer into buffer. Keeps query.
 644  * @param pkt: buffer
 645  * @param str: string to put into text record (<255).
 646  * @param edns: edns reply information.
 647  * @param worker: worker with scratch region.
 648  */
 649 static void
 650 chaos_replystr(sldns_buffer* pkt, const char* str, struct edns_data* edns,
 651         struct worker* worker)
 652 {
 653         size_t len = strlen(str);
 654         unsigned int rd = LDNS_RD_WIRE(sldns_buffer_begin(pkt));
 655         unsigned int cd = LDNS_CD_WIRE(sldns_buffer_begin(pkt));
 656         if(len>255) len=255; /* cap size of TXT record */
 657         sldns_buffer_clear(pkt);
 658         sldns_buffer_skip(pkt, (ssize_t)sizeof(uint16_t)); /* skip id */
 659         sldns_buffer_write_u16(pkt, (uint16_t)(BIT_QR|BIT_RA));
 660         if(rd) LDNS_RD_SET(sldns_buffer_begin(pkt));
 661         if(cd) LDNS_CD_SET(sldns_buffer_begin(pkt));
 662         sldns_buffer_write_u16(pkt, 1); /* qdcount */
 663         sldns_buffer_write_u16(pkt, 1); /* ancount */
 664         sldns_buffer_write_u16(pkt, 0); /* nscount */
 665         sldns_buffer_write_u16(pkt, 0); /* arcount */
 666         (void)query_dname_len(pkt); /* skip qname */
 667         sldns_buffer_skip(pkt, (ssize_t)sizeof(uint16_t)); /* skip qtype */
 668         sldns_buffer_skip(pkt, (ssize_t)sizeof(uint16_t)); /* skip qclass */
 669         sldns_buffer_write_u16(pkt, 0xc00c); /* compr ptr to query */
 670         sldns_buffer_write_u16(pkt, LDNS_RR_TYPE_TXT);
 671         sldns_buffer_write_u16(pkt, LDNS_RR_CLASS_CH);
 672         sldns_buffer_write_u32(pkt, 0); /* TTL */
 673         sldns_buffer_write_u16(pkt, sizeof(uint8_t) + len);
 674         sldns_buffer_write_u8(pkt, len);
 675         sldns_buffer_write(pkt, str, len);
 676         sldns_buffer_flip(pkt);
 677         edns->edns_version = EDNS_ADVERTISED_VERSION;
 678         edns->udp_size = EDNS_ADVERTISED_SIZE;
 679         edns->bits &= EDNS_DO;
 680         if(!inplace_cb_reply_local_call(&worker->env, NULL, NULL, NULL,
 681                 LDNS_RCODE_NOERROR, edns, worker->scratchpad))
 682                         edns->opt_list = NULL;
 683         attach_edns_record(pkt, edns);
 684 }
 685
 686 /**
 687  * Answer CH class queries.
 688  * @param w: worker
 689  * @param qinfo: query info. Pointer into packet buffer.
 690  * @param edns: edns info from query.
 691  * @param pkt: packet buffer.
 692  * @return: true if a reply is to be sent.
 693  */
 694 static int
 695 answer_chaos(struct worker* w, struct query_info* qinfo,
 696         struct edns_data* edns, sldns_buffer* pkt)
 697 {
 698         struct config_file* cfg = w->env.cfg;
 699         if(qinfo->qtype != LDNS_RR_TYPE_ANY && qinfo->qtype != LDNS_RR_TYPE_TXT)
 700                 return 0;
 701         if(query_dname_compare(qinfo->qname,
 702                 (uint8_t*)"\002id\006server") == 0 ||
 703                 query_dname_compare(qinfo->qname,
 704                 (uint8_t*)"\010hostname\004bind") == 0)
 705         {
 706                 if(cfg->hide_identity)
 707                         return 0;
 708                 if(cfg->identity==NULL || cfg->identity[0]==0) {
 709                         char buf[MAXHOSTNAMELEN+1];
 710                         if (gethostname(buf, MAXHOSTNAMELEN) == 0) {
 711                                 buf[MAXHOSTNAMELEN] = 0;
 712                                 chaos_replystr(pkt, buf, edns, w);
 713                         } else  {
 714                                 log_err("gethostname: %s", strerror(errno));
 715                                 chaos_replystr(pkt, "no hostname", edns, w);
 716                         }
 717                 }
 718                 else    chaos_replystr(pkt, cfg->identity, edns, w);
 719                 return 1;
 720         }
 721         if(query_dname_compare(qinfo->qname,
 722                 (uint8_t*)"\007version\006server") == 0 ||
 723                 query_dname_compare(qinfo->qname,
 724                 (uint8_t*)"\007version\004bind") == 0)
 725         {
 726                 if(cfg->hide_version)
 727                         return 0;
 728                 if(cfg->version==NULL || cfg->version[0]==0)
 729                         chaos_replystr(pkt, PACKAGE_STRING, edns, w);
 730                 else    chaos_replystr(pkt, cfg->version, edns, w);
 731                 return 1;
 732         }
 733         return 0;
 734 }
 735
 736 static int
 737 deny_refuse(struct comm_point* c, enum acl_access acl,
 738         enum acl_access deny, enum acl_access refuse,
 739         struct worker* worker, struct comm_reply* repinfo)
 740 {
 741         if(acl == deny) {
 742                 comm_point_drop_reply(repinfo);
 743                 if(worker->stats.extended)
 744                         worker->stats.unwanted_queries++;
 745                 return 0;
 746         } else if(acl == refuse) {
 747                 log_addr(VERB_ALGO, "refused query from",
 748                         &repinfo->addr, repinfo->addrlen);
 749                 log_buf(VERB_ALGO, "refuse", c->buffer);
 750                 if(worker->stats.extended)
 751                         worker->stats.unwanted_queries++;
 752                 if(worker_check_request(c->buffer, worker) == -1) {
 753                         comm_point_drop_reply(repinfo);
 754                         return 0; /* discard this */
 755                 }
 756                 sldns_buffer_set_limit(c->buffer, LDNS_HEADER_SIZE);
 757                 sldns_buffer_write_at(c->buffer, 4,
 758                         (uint8_t*)"\0\0\0\0\0\0\0\0", 8);
 759                 LDNS_QR_SET(sldns_buffer_begin(c->buffer));
 760                 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer),
 761                         LDNS_RCODE_REFUSED);
 762                 sldns_buffer_set_position(c->buffer, LDNS_HEADER_SIZE);
 763                 sldns_buffer_flip(c->buffer);
 764                 return 1;
 765         }
 766
 767         return -1;
 768 }
 769
 770 static int
 771 deny_refuse_all(struct comm_point* c, enum acl_access acl,
 772         struct worker* worker, struct comm_reply* repinfo)
 773 {
 774         return deny_refuse(c, acl, acl_deny, acl_refuse, worker, repinfo);
 775 }
 776
 777 static int
 778 deny_refuse_non_local(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_non_local, acl_refuse_non_local, worker, repinfo);
 782 }
 783
 784 int
 785 worker_handle_request(struct comm_point* c, void* arg, int error,
 786         struct comm_reply* repinfo)
 787 {
 788         struct worker* worker = (struct worker*)arg;
 789         int ret;
 790         hashvalue_type h;
 791         struct lruhash_entry* e;
 792         struct query_info qinfo;
 793         struct edns_data edns;
 794         enum acl_access acl;
 795         struct acl_addr* acladdr;
 796         int rc = 0;
 797
 798         if(error != NETEVENT_NOERROR) {
 799                 /* some bad tcp query DNS formats give these error calls */
 800                 verbose(VERB_ALGO, "handle request called with err=%d", error);
 801                 return 0;
 802         }
 803 #ifdef USE_DNSTAP
 804         if(worker->dtenv.log_client_query_messages)
 805                 dt_msg_send_client_query(&worker->dtenv, &repinfo->addr, c->type,
 806                         c->buffer);
 807 #endif
 808         acladdr = acl_addr_lookup(worker->daemon->acl, &repinfo->addr,
 809                 repinfo->addrlen);
 810         acl = acl_get_control(acladdr);
 811         if((ret=deny_refuse_all(c, acl, worker, repinfo)) != -1)
 812         {
 813                 if(ret == 1)
 814                         goto send_reply;
 815                 return ret;
 816         }
 817         if((ret=worker_check_request(c->buffer, worker)) != 0) {
 818                 verbose(VERB_ALGO, "worker check request: bad query.");
 819                 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen);
 820                 if(ret != -1) {
 821                         LDNS_QR_SET(sldns_buffer_begin(c->buffer));
 822                         LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), ret);
 823                         return 1;
 824                 }
 825                 comm_point_drop_reply(repinfo);
 826                 return 0;
 827         }
 828
 829         worker->stats.num_queries++;
 830
 831         /* check if this query should be dropped based on source ip rate limiting */
 832         if(!infra_ip_ratelimit_inc(worker->env.infra_cache, repinfo,
 833                         *worker->env.now)) {
 834                 /* See if we are passed through with slip factor */
 835                 if(worker->env.cfg->ip_ratelimit_factor != 0 &&
 836                         ub_random_max(worker->env.rnd,
 837                                                   worker->env.cfg->ip_ratelimit_factor) == 1) {
 838
 839                         char addrbuf[128];
 840                         addr_to_str(&repinfo->addr, repinfo->addrlen,
 841                                                 addrbuf, sizeof(addrbuf));
 842                   verbose(VERB_OPS, "ip_ratelimit allowed through for ip address %s ",
 843                                   addrbuf);
 844                 } else {
 845                         worker->stats.num_queries_ip_ratelimited++;
 846                         comm_point_drop_reply(repinfo);
 847                         return 0;
 848                 }
 849         }
 850
 851         /* see if query is in the cache */
 852         if(!query_info_parse(&qinfo, c->buffer)) {
 853                 verbose(VERB_ALGO, "worker parse request: formerror.");
 854                 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen);
 855                 if(worker_err_ratelimit(worker, LDNS_RCODE_FORMERR) == -1) {
 856                         comm_point_drop_reply(repinfo);
 857                         return 0;
 858                 }
 859                 sldns_buffer_rewind(c->buffer);
 860                 LDNS_QR_SET(sldns_buffer_begin(c->buffer));
 861                 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer),
 862                         LDNS_RCODE_FORMERR);
 863                 server_stats_insrcode(&worker->stats, c->buffer);
 864                 goto send_reply;
 865         }
 866         if(worker->env.cfg->log_queries) {
 867                 char ip[128];
 868                 addr_to_str(&repinfo->addr, repinfo->addrlen, ip, sizeof(ip));
 869                 log_nametypeclass(0, ip, qinfo.qname, qinfo.qtype, qinfo.qclass);
 870         }
 871         if(qinfo.qtype == LDNS_RR_TYPE_AXFR ||
 872                 qinfo.qtype == LDNS_RR_TYPE_IXFR) {
 873                 verbose(VERB_ALGO, "worker request: refused zone transfer.");
 874                 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen);
 875                 sldns_buffer_rewind(c->buffer);
 876                 LDNS_QR_SET(sldns_buffer_begin(c->buffer));
 877                 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer),
 878                         LDNS_RCODE_REFUSED);
 879                 if(worker->stats.extended) {
 880                         worker->stats.qtype[qinfo.qtype]++;
 881                         server_stats_insrcode(&worker->stats, c->buffer);
 882                 }
 883                 goto send_reply;
 884         }
 885         if(qinfo.qtype == LDNS_RR_TYPE_OPT ||
 886                 qinfo.qtype == LDNS_RR_TYPE_TSIG ||
 887                 qinfo.qtype == LDNS_RR_TYPE_TKEY ||
 888                 qinfo.qtype == LDNS_RR_TYPE_MAILA ||
 889                 qinfo.qtype == LDNS_RR_TYPE_MAILB ||
 890                 (qinfo.qtype >= 128 && qinfo.qtype <= 248)) {
 891                 verbose(VERB_ALGO, "worker request: formerror for meta-type.");
 892                 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen);
 893                 if(worker_err_ratelimit(worker, LDNS_RCODE_FORMERR) == -1) {
 894                         comm_point_drop_reply(repinfo);
 895                         return 0;
 896                 }
 897                 sldns_buffer_rewind(c->buffer);
 898                 LDNS_QR_SET(sldns_buffer_begin(c->buffer));
 899                 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer),
 900                         LDNS_RCODE_FORMERR);
 901                 if(worker->stats.extended) {
 902                         worker->stats.qtype[qinfo.qtype]++;
 903                         server_stats_insrcode(&worker->stats, c->buffer);
 904                 }
 905                 goto send_reply;
 906         }
 907         if((ret=parse_edns_from_pkt(c->buffer, &edns, worker->scratchpad)) != 0) {
 908                 struct edns_data reply_edns;
 909                 verbose(VERB_ALGO, "worker parse edns: formerror.");
 910                 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen);
 911                 memset(&reply_edns, 0, sizeof(reply_edns));
 912                 reply_edns.edns_present = 1;
 913                 reply_edns.udp_size = EDNS_ADVERTISED_SIZE;
 914                 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), ret);
 915                 error_encode(c->buffer, ret, &qinfo,
 916                         *(uint16_t*)(void *)sldns_buffer_begin(c->buffer),
 917                         sldns_buffer_read_u16_at(c->buffer, 2), &reply_edns);
 918                 regional_free_all(worker->scratchpad);
 919                 server_stats_insrcode(&worker->stats, c->buffer);
 920                 goto send_reply;
 921         }
 922         if(edns.edns_present && edns.edns_version != 0) {
 923                 edns.ext_rcode = (uint8_t)(EDNS_RCODE_BADVERS>>4);
 924                 edns.edns_version = EDNS_ADVERTISED_VERSION;
 925                 edns.udp_size = EDNS_ADVERTISED_SIZE;
 926                 edns.bits &= EDNS_DO;
 927                 edns.opt_list = NULL;
 928                 verbose(VERB_ALGO, "query with bad edns version.");
 929                 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen);
 930                 error_encode(c->buffer, EDNS_RCODE_BADVERS&0xf, &qinfo,
 931                         *(uint16_t*)(void *)sldns_buffer_begin(c->buffer),
 932                         sldns_buffer_read_u16_at(c->buffer, 2), NULL);
 933                 attach_edns_record(c->buffer, &edns);
 934                 regional_free_all(worker->scratchpad);
 935                 goto send_reply;
 936         }
 937         if(edns.edns_present && edns.udp_size < NORMAL_UDP_SIZE &&
 938                 worker->daemon->cfg->harden_short_bufsize) {
 939                 verbose(VERB_QUERY, "worker request: EDNS bufsize %d ignored",
 940                         (int)edns.udp_size);
 941                 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen);
 942                 edns.udp_size = NORMAL_UDP_SIZE;
 943         }
 944         if(edns.udp_size > worker->daemon->cfg->max_udp_size &&
 945                 c->type == comm_udp) {
 946                 verbose(VERB_QUERY,
 947                         "worker request: max UDP reply size modified"
 948                         " (%d to max-udp-size)", (int)edns.udp_size);
 949                 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen);
 950                 edns.udp_size = worker->daemon->cfg->max_udp_size;
 951         }
 952         if(edns.udp_size < LDNS_HEADER_SIZE) {
 953                 verbose(VERB_ALGO, "worker request: edns is too small.");
 954                 log_addr(VERB_CLIENT, "from", &repinfo->addr, repinfo->addrlen);
 955                 LDNS_QR_SET(sldns_buffer_begin(c->buffer));
 956                 LDNS_TC_SET(sldns_buffer_begin(c->buffer));
 957                 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer),
 958                         LDNS_RCODE_SERVFAIL);
 959                 sldns_buffer_set_position(c->buffer, LDNS_HEADER_SIZE);
 960                 sldns_buffer_write_at(c->buffer, 4,
 961                         (uint8_t*)"\0\0\0\0\0\0\0\0", 8);
 962                 sldns_buffer_flip(c->buffer);
 963                 regional_free_all(worker->scratchpad);
 964                 goto send_reply;
 965         }
 966         if(worker->stats.extended)
 967                 server_stats_insquery(&worker->stats, c, qinfo.qtype,
 968                         qinfo.qclass, &edns, repinfo);
 969         if(c->type != comm_udp)
 970                 edns.udp_size = 65535; /* max size for TCP replies */
 971         if(qinfo.qclass == LDNS_RR_CLASS_CH && answer_chaos(worker, &qinfo,
 972                 &edns, c->buffer)) {
 973                 server_stats_insrcode(&worker->stats, c->buffer);
 974                 regional_free_all(worker->scratchpad);
 975                 goto send_reply;
 976         }
 977         if(local_zones_answer(worker->daemon->local_zones, &worker->env, &qinfo,
 978                 &edns, c->buffer, worker->scratchpad, repinfo, acladdr->taglist,
 979                 acladdr->taglen, acladdr->tag_actions,
 980                 acladdr->tag_actions_size, acladdr->tag_datas,
 981                 acladdr->tag_datas_size, worker->daemon->cfg->tagname,
 982                 worker->daemon->cfg->num_tags, acladdr->view)) {
 983                 regional_free_all(worker->scratchpad);
 984                 if(sldns_buffer_limit(c->buffer) == 0) {
 985                         comm_point_drop_reply(repinfo);
 986                         return 0;
 987                 }
 988                 server_stats_insrcode(&worker->stats, c->buffer);
 989                 goto send_reply;
 990         }
 991
 992         /* We've looked in our local zones. If the answer isn't there, we
 993          * might need to bail out based on ACLs now. */
 994         if((ret=deny_refuse_non_local(c, acl, worker, repinfo)) != -1)
 995         {
 996                 regional_free_all(worker->scratchpad);
 997                 if(ret == 1)
 998                         goto send_reply;
 999                 return ret;
1000         }
1001
1002         /* If this request does not have the recursion bit set, verify
1003          * ACLs allow the snooping. */
1004         if(!(LDNS_RD_WIRE(sldns_buffer_begin(c->buffer))) &&
1005                 acl != acl_allow_snoop ) {
1006                 sldns_buffer_set_limit(c->buffer, LDNS_HEADER_SIZE);
1007                 sldns_buffer_write_at(c->buffer, 4,
1008                         (uint8_t*)"\0\0\0\0\0\0\0\0", 8);
1009                 LDNS_QR_SET(sldns_buffer_begin(c->buffer));
1010                 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer),
1011                         LDNS_RCODE_REFUSED);
1012                 sldns_buffer_flip(c->buffer);
1013                 regional_free_all(worker->scratchpad);
1014                 server_stats_insrcode(&worker->stats, c->buffer);
1015                 log_addr(VERB_ALGO, "refused nonrec (cache snoop) query from",
1016                         &repinfo->addr, repinfo->addrlen);
1017                 goto send_reply;
1018         }
1019
1020         /* If we've found a local alias, replace the qname with the alias
1021          * target before resolving it. */
1022         if(qinfo.local_alias) {
1023                 struct ub_packed_rrset_key* rrset = qinfo.local_alias->rrset;
1024                 struct packed_rrset_data* d = rrset->entry.data;
1025
1026                 /* Sanity check: our current implementation only supports
1027                  * a single CNAME RRset as a local alias. */
1028                 if(qinfo.local_alias->next ||
1029                         rrset->rk.type != htons(LDNS_RR_TYPE_CNAME) ||
1030                         d->count != 1) {
1031                         log_err("assumption failure: unexpected local alias");
1032                         regional_free_all(worker->scratchpad);
1033                         return 0; /* drop it */
1034                 }
1035                 qinfo.qname = d->rr_data[0] + 2;
1036                 qinfo.qname_len = d->rr_len[0] - 2;
1037         }
1038
1039         if(!edns_bypass_cache_stage(edns.opt_list, &worker->env)) {
1040                 h = query_info_hash(&qinfo, sldns_buffer_read_u16_at(c->buffer, 2));
1041                 if((e=slabhash_lookup(worker->env.msg_cache, h, &qinfo, 0))) {
1042                         /* answer from cache - we have acquired a readlock on it */
1043                         if(answer_from_cache(worker, &qinfo,
1044                                 (struct reply_info*)e->data,
1045                                 *(uint16_t*)(void *)sldns_buffer_begin(c->buffer),
1046                                 sldns_buffer_read_u16_at(c->buffer, 2), repinfo,
1047                                 &edns)) {
1048                                 /* prefetch it if the prefetch TTL expired */
1049                                 if((worker->env.cfg->prefetch || worker->env.cfg->serve_expired)
1050                                         && *worker->env.now >=
1051                                         ((struct reply_info*)e->data)->prefetch_ttl) {
1052                                         time_t leeway = ((struct reply_info*)e->
1053                                                 data)->ttl - *worker->env.now;
1054                                         if(((struct reply_info*)e->data)->ttl
1055                                                 < *worker->env.now)
1056                                                 leeway = 0;
1057                                         lock_rw_unlock(&e->lock);
1058                                         reply_and_prefetch(worker, &qinfo,
1059                                                 sldns_buffer_read_u16_at(c->buffer, 2),
1060                                                 repinfo, leeway);
1061                                         rc = 0;
1062                                         regional_free_all(worker->scratchpad);
1063                                         goto send_reply_rc;
1064                                 }
1065                                 lock_rw_unlock(&e->lock);
1066                                 regional_free_all(worker->scratchpad);
1067                                 goto send_reply;
1068                         }
1069                         verbose(VERB_ALGO, "answer from the cache failed");
1070                         lock_rw_unlock(&e->lock);
1071                 }
1072                 if(!LDNS_RD_WIRE(sldns_buffer_begin(c->buffer))) {
1073                         if(answer_norec_from_cache(worker, &qinfo,
1074                                 *(uint16_t*)(void *)sldns_buffer_begin(c->buffer),
1075                                 sldns_buffer_read_u16_at(c->buffer, 2), repinfo,
1076                                 &edns)) {
1077                                 regional_free_all(worker->scratchpad);
1078                                 goto send_reply;
1079                         }
1080                         verbose(VERB_ALGO, "answer norec from cache -- "
1081                                 "need to validate or not primed");
1082                 }
1083         }
1084         sldns_buffer_rewind(c->buffer);
1085         server_stats_querymiss(&worker->stats, worker);
1086
1087         if(verbosity >= VERB_CLIENT) {
1088                 if(c->type == comm_udp)
1089                         log_addr(VERB_CLIENT, "udp request from",
1090                                 &repinfo->addr, repinfo->addrlen);
1091                 else    log_addr(VERB_CLIENT, "tcp request from",
1092                                 &repinfo->addr, repinfo->addrlen);
1093         }
1094
1095         /* grab a work request structure for this new request */
1096         mesh_new_client(worker->env.mesh, &qinfo,
1097                 sldns_buffer_read_u16_at(c->buffer, 2),
1098                 &edns, repinfo, *(uint16_t*)(void *)sldns_buffer_begin(c->buffer));
1099         regional_free_all(worker->scratchpad);
1100         worker_mem_report(worker, NULL);
1101         return 0;
1102
1103 send_reply:
1104         rc = 1;
1105 send_reply_rc:
1106 #ifdef USE_DNSTAP
1107         if(worker->dtenv.log_client_response_messages)
1108                 dt_msg_send_client_response(&worker->dtenv, &repinfo->addr,
1109                         c->type, c->buffer);
1110 #endif
1111         if(worker->env.cfg->log_replies)
1112         {
1113                 struct timeval tv = {0, 0};
1114                 log_reply_info(0, &qinfo, &repinfo->addr, repinfo->addrlen,
1115                         tv, 1, c->buffer);
1116         }
1117         return rc;
1118 }
1119
1120 void
1121 worker_sighandler(int sig, void* arg)
1122 {
1123         /* note that log, print, syscalls here give race conditions.
1124          * And cause hangups if the log-lock is held by the application. */
1125         struct worker* worker = (struct worker*)arg;
1126         switch(sig) {
1127 #ifdef SIGHUP
1128                 case SIGHUP:
1129                         comm_base_exit(worker->base);
1130                         break;
1131 #endif
1132                 case SIGINT:
1133                         worker->need_to_exit = 1;
1134                         comm_base_exit(worker->base);
1135                         break;
1136 #ifdef SIGQUIT
1137                 case SIGQUIT:
1138                         worker->need_to_exit = 1;
1139                         comm_base_exit(worker->base);
1140                         break;
1141 #endif
1142                 case SIGTERM:
1143                         worker->need_to_exit = 1;
1144                         comm_base_exit(worker->base);
1145                         break;
1146                 default:
1147                         /* unknown signal, ignored */
1148                         break;
1149         }
1150 }
1151
1152 /** restart statistics timer for worker, if enabled */
1153 static void
1154 worker_restart_timer(struct worker* worker)
1155 {
1156         if(worker->env.cfg->stat_interval > 0) {
1157                 struct timeval tv;
1158 #ifndef S_SPLINT_S
1159                 tv.tv_sec = worker->env.cfg->stat_interval;
1160                 tv.tv_usec = 0;
1161 #endif
1162                 comm_timer_set(worker->stat_timer, &tv);
1163         }
1164 }
1165
1166 void worker_stat_timer_cb(void* arg)
1167 {
1168         struct worker* worker = (struct worker*)arg;
1169         server_stats_log(&worker->stats, worker, worker->thread_num);
1170         mesh_stats(worker->env.mesh, "mesh has");
1171         worker_mem_report(worker, NULL);
1172         if(!worker->daemon->cfg->stat_cumulative) {
1173                 worker_stats_clear(worker);
1174         }
1175         /* start next timer */
1176         worker_restart_timer(worker);
1177 }
1178
1179 void worker_probe_timer_cb(void* arg)
1180 {
1181         struct worker* worker = (struct worker*)arg;
1182         struct timeval tv;
1183 #ifndef S_SPLINT_S
1184         tv.tv_sec = (time_t)autr_probe_timer(&worker->env);
1185         tv.tv_usec = 0;
1186 #endif
1187         if(tv.tv_sec != 0)
1188                 comm_timer_set(worker->env.probe_timer, &tv);
1189 }
1190
1191 struct worker*
1192 worker_create(struct daemon* daemon, int id, int* ports, int n)
1193 {
1194         unsigned int seed;
1195         struct worker* worker = (struct worker*)calloc(1,
1196                 sizeof(struct worker));
1197         if(!worker)
1198                 return NULL;
1199         worker->numports = n;
1200         worker->ports = (int*)memdup(ports, sizeof(int)*n);
1201         if(!worker->ports) {
1202                 free(worker);
1203                 return NULL;
1204         }
1205         worker->daemon = daemon;
1206         worker->thread_num = id;
1207         if(!(worker->cmd = tube_create())) {
1208                 free(worker->ports);
1209                 free(worker);
1210                 return NULL;
1211         }
1212         /* create random state here to avoid locking trouble in RAND_bytes */
1213         seed = (unsigned int)time(NULL) ^ (unsigned int)getpid() ^
1214                 (((unsigned int)worker->thread_num)<<17);
1215                 /* shift thread_num so it does not match out pid bits */
1216         if(!(worker->rndstate = ub_initstate(seed, daemon->rand))) {
1217                 seed = 0;
1218                 log_err("could not init random numbers.");
1219                 tube_delete(worker->cmd);
1220                 free(worker->ports);
1221                 free(worker);
1222                 return NULL;
1223         }
1224         seed = 0;
1225 #ifdef USE_DNSTAP
1226         if(daemon->cfg->dnstap) {
1227                 log_assert(daemon->dtenv != NULL);
1228                 memcpy(&worker->dtenv, daemon->dtenv, sizeof(struct dt_env));
1229                 if(!dt_init(&worker->dtenv))
1230                         fatal_exit("dt_init failed");
1231         }
1232 #endif
1233         return worker;
1234 }
1235
1236 int
1237 worker_init(struct worker* worker, struct config_file *cfg,
1238         struct listen_port* ports, int do_sigs)
1239 {
1240 #ifdef USE_DNSTAP
1241         struct dt_env* dtenv = &worker->dtenv;
1242 #else
1243         void* dtenv = NULL;
1244 #endif
1245         worker->need_to_exit = 0;
1246         worker->base = comm_base_create(do_sigs);
1247         if(!worker->base) {
1248                 log_err("could not create event handling base");
1249                 worker_delete(worker);
1250                 return 0;
1251         }
1252         comm_base_set_slow_accept_handlers(worker->base, &worker_stop_accept,
1253                 &worker_start_accept, worker);
1254         if(do_sigs) {
1255 #ifdef SIGHUP
1256                 ub_thread_sig_unblock(SIGHUP);
1257 #endif
1258                 ub_thread_sig_unblock(SIGINT);
1259 #ifdef SIGQUIT
1260                 ub_thread_sig_unblock(SIGQUIT);
1261 #endif
1262                 ub_thread_sig_unblock(SIGTERM);
1263 #ifndef LIBEVENT_SIGNAL_PROBLEM
1264                 worker->comsig = comm_signal_create(worker->base,
1265                         worker_sighandler, worker);
1266                 if(!worker->comsig
1267 #ifdef SIGHUP
1268                         || !comm_signal_bind(worker->comsig, SIGHUP)
1269 #endif
1270 #ifdef SIGQUIT
1271                         || !comm_signal_bind(worker->comsig, SIGQUIT)
1272 #endif
1273                         || !comm_signal_bind(worker->comsig, SIGTERM)
1274                         || !comm_signal_bind(worker->comsig, SIGINT)) {
1275                         log_err("could not create signal handlers");
1276                         worker_delete(worker);
1277                         return 0;
1278                 }
1279 #endif /* LIBEVENT_SIGNAL_PROBLEM */
1280                 if(!daemon_remote_open_accept(worker->daemon->rc,
1281                         worker->daemon->rc_ports, worker)) {
1282                         worker_delete(worker);
1283                         return 0;
1284                 }
1285 #ifdef UB_ON_WINDOWS
1286                 wsvc_setup_worker(worker);
1287 #endif /* UB_ON_WINDOWS */
1288         } else { /* !do_sigs */
1289                 worker->comsig = NULL;
1290         }
1291         worker->front = listen_create(worker->base, ports,
1292                 cfg->msg_buffer_size, (int)cfg->incoming_num_tcp,
1293                 worker->daemon->listen_sslctx, dtenv, worker_handle_request,
1294                 worker);
1295         if(!worker->front) {
1296                 log_err("could not create listening sockets");
1297                 worker_delete(worker);
1298                 return 0;
1299         }
1300         worker->back = outside_network_create(worker->base,
1301                 cfg->msg_buffer_size, (size_t)cfg->outgoing_num_ports,
1302                 cfg->out_ifs, cfg->num_out_ifs, cfg->do_ip4, cfg->do_ip6,
1303                 cfg->do_tcp?cfg->outgoing_num_tcp:0,
1304                 worker->daemon->env->infra_cache, worker->rndstate,
1305                 cfg->use_caps_bits_for_id, worker->ports, worker->numports,
1306                 cfg->unwanted_threshold, cfg->outgoing_tcp_mss,
1307                 &worker_alloc_cleanup, worker,
1308                 cfg->do_udp, worker->daemon->connect_sslctx, cfg->delay_close,
1309                 dtenv);
1310         if(!worker->back) {
1311                 log_err("could not create outgoing sockets");
1312                 worker_delete(worker);
1313                 return 0;
1314         }
1315         /* start listening to commands */
1316         if(!tube_setup_bg_listen(worker->cmd, worker->base,
1317                 &worker_handle_control_cmd, worker)) {
1318                 log_err("could not create control compt.");
1319                 worker_delete(worker);
1320                 return 0;
1321         }
1322         worker->stat_timer = comm_timer_create(worker->base,
1323                 worker_stat_timer_cb, worker);
1324         if(!worker->stat_timer) {
1325                 log_err("could not create statistics timer");
1326         }
1327
1328         /* we use the msg_buffer_size as a good estimate for what the
1329          * user wants for memory usage sizes */
1330         worker->scratchpad = regional_create_custom(cfg->msg_buffer_size);
1331         if(!worker->scratchpad) {
1332                 log_err("malloc failure");
1333                 worker_delete(worker);
1334                 return 0;
1335         }
1336
1337         server_stats_init(&worker->stats, cfg);
1338         alloc_init(&worker->alloc, &worker->daemon->superalloc,
1339                 worker->thread_num);
1340         alloc_set_id_cleanup(&worker->alloc, &worker_alloc_cleanup, worker);
1341         worker->env = *worker->daemon->env;
1342         comm_base_timept(worker->base, &worker->env.now, &worker->env.now_tv);
1343         if(worker->thread_num == 0)
1344                 log_set_time(worker->env.now);
1345         worker->env.worker = worker;
1346         worker->env.send_query = &worker_send_query;
1347         worker->env.alloc = &worker->alloc;
1348         worker->env.rnd = worker->rndstate;
1349         worker->env.scratch = worker->scratchpad;
1350         worker->env.mesh = mesh_create(&worker->daemon->mods, &worker->env);
1351         worker->env.detach_subs = &mesh_detach_subs;
1352         worker->env.attach_sub = &mesh_attach_sub;
1353         worker->env.kill_sub = &mesh_state_delete;
1354         worker->env.detect_cycle = &mesh_detect_cycle;
1355         worker->env.scratch_buffer = sldns_buffer_new(cfg->msg_buffer_size);
1356         if(!(worker->env.fwds = forwards_create()) ||
1357                 !forwards_apply_cfg(worker->env.fwds, cfg)) {
1358                 log_err("Could not set forward zones");
1359                 worker_delete(worker);
1360                 return 0;
1361         }
1362         if(!(worker->env.hints = hints_create()) ||
1363                 !hints_apply_cfg(worker->env.hints, cfg)) {
1364                 log_err("Could not set root or stub hints");
1365                 worker_delete(worker);
1366                 return 0;
1367         }
1368         /* one probe timer per process -- if we have 5011 anchors */
1369         if(autr_get_num_anchors(worker->env.anchors) > 0
1370 #ifndef THREADS_DISABLED
1371                 && worker->thread_num == 0
1372 #endif
1373                 ) {
1374                 struct timeval tv;
1375                 tv.tv_sec = 0;
1376                 tv.tv_usec = 0;
1377                 worker->env.probe_timer = comm_timer_create(worker->base,
1378                         worker_probe_timer_cb, worker);
1379                 if(!worker->env.probe_timer) {
1380                         log_err("could not create 5011-probe timer");
1381                 } else {
1382                         /* let timer fire, then it can reset itself */
1383                         comm_timer_set(worker->env.probe_timer, &tv);
1384                 }
1385         }
1386         if(!worker->env.mesh || !worker->env.scratch_buffer) {
1387                 worker_delete(worker);
1388                 return 0;
1389         }
1390         worker_mem_report(worker, NULL);
1391         /* if statistics enabled start timer */
1392         if(worker->env.cfg->stat_interval > 0) {
1393                 verbose(VERB_ALGO, "set statistics interval %d secs",
1394                         worker->env.cfg->stat_interval);
1395                 worker_restart_timer(worker);
1396         }
1397         return 1;
1398 }
1399
1400 void
1401 worker_work(struct worker* worker)
1402 {
1403         comm_base_dispatch(worker->base);
1404 }
1405
1406 void
1407 worker_delete(struct worker* worker)
1408 {
1409         if(!worker)
1410                 return;
1411         if(worker->env.mesh && verbosity >= VERB_OPS) {
1412                 server_stats_log(&worker->stats, worker, worker->thread_num);
1413                 mesh_stats(worker->env.mesh, "mesh has");
1414                 worker_mem_report(worker, NULL);
1415         }
1416         outside_network_quit_prepare(worker->back);
1417         mesh_delete(worker->env.mesh);
1418         sldns_buffer_free(worker->env.scratch_buffer);
1419         forwards_delete(worker->env.fwds);
1420         hints_delete(worker->env.hints);
1421         listen_delete(worker->front);
1422         outside_network_delete(worker->back);
1423         comm_signal_delete(worker->comsig);
1424         tube_delete(worker->cmd);
1425         comm_timer_delete(worker->stat_timer);
1426         comm_timer_delete(worker->env.probe_timer);
1427         free(worker->ports);
1428         if(worker->thread_num == 0) {
1429                 log_set_time(NULL);
1430 #ifdef UB_ON_WINDOWS
1431                 wsvc_desetup_worker(worker);
1432 #endif /* UB_ON_WINDOWS */
1433         }
1434         comm_base_delete(worker->base);
1435         ub_randfree(worker->rndstate);
1436         alloc_clear(&worker->alloc);
1437         regional_destroy(worker->scratchpad);
1438         free(worker);
1439 }
1440
1441 struct outbound_entry*
1442 worker_send_query(struct query_info* qinfo, uint16_t flags, int dnssec,
1443         int want_dnssec, int nocaps, struct sockaddr_storage* addr,
1444         socklen_t addrlen, uint8_t* zone, size_t zonelen, int ssl_upstream,
1445         struct module_qstate* q)
1446 {
1447         struct worker* worker = q->env->worker;
1448         struct outbound_entry* e = (struct outbound_entry*)regional_alloc(
1449                 q->region, sizeof(*e));
1450         if(!e)
1451                 return NULL;
1452         e->qstate = q;
1453         e->qsent = outnet_serviced_query(worker->back, qinfo, flags, dnssec,
1454                 want_dnssec, nocaps, q->env->cfg->tcp_upstream,
1455                 ssl_upstream, addr, addrlen, zone, zonelen, q,
1456                 worker_handle_service_reply, e, worker->back->udp_buff, q->env);
1457         if(!e->qsent) {
1458                 return NULL;
1459         }
1460         return e;
1461 }
1462
1463 void
1464 worker_alloc_cleanup(void* arg)
1465 {
1466         struct worker* worker = (struct worker*)arg;
1467         slabhash_clear(&worker->env.rrset_cache->table);
1468         slabhash_clear(worker->env.msg_cache);
1469 }
1470
1471 void worker_stats_clear(struct worker* worker)
1472 {
1473         server_stats_init(&worker->stats, worker->env.cfg);
1474         mesh_stats_clear(worker->env.mesh);
1475         worker->back->unwanted_replies = 0;
1476         worker->back->num_tcp_outgoing = 0;
1477 }
1478
1479 void worker_start_accept(void* arg)
1480 {
1481         struct worker* worker = (struct worker*)arg;
1482         listen_start_accept(worker->front);
1483         if(worker->thread_num == 0)
1484                 daemon_remote_start_accept(worker->daemon->rc);
1485 }
1486
1487 void worker_stop_accept(void* arg)
1488 {
1489         struct worker* worker = (struct worker*)arg;
1490         listen_stop_accept(worker->front);
1491         if(worker->thread_num == 0)
1492                 daemon_remote_stop_accept(worker->daemon->rc);
1493 }
1494
1495 /* --- fake callbacks for fptr_wlist to work --- */
1496 struct outbound_entry* libworker_send_query(
1497         struct query_info* ATTR_UNUSED(qinfo),
1498         uint16_t ATTR_UNUSED(flags), int ATTR_UNUSED(dnssec),
1499         int ATTR_UNUSED(want_dnssec), int ATTR_UNUSED(nocaps),
1500         struct sockaddr_storage* ATTR_UNUSED(addr), socklen_t ATTR_UNUSED(addrlen),
1501         uint8_t* ATTR_UNUSED(zone), size_t ATTR_UNUSED(zonelen),
1502         int ATTR_UNUSED(ssl_upstream), struct module_qstate* ATTR_UNUSED(q))
1503 {
1504         log_assert(0);
1505         return 0;
1506 }
1507
1508 int libworker_handle_reply(struct comm_point* ATTR_UNUSED(c),
1509         void* ATTR_UNUSED(arg), int ATTR_UNUSED(error),
1510         struct comm_reply* ATTR_UNUSED(reply_info))
1511 {
1512         log_assert(0);
1513         return 0;
1514 }
1515
1516 int libworker_handle_service_reply(struct comm_point* ATTR_UNUSED(c),
1517         void* ATTR_UNUSED(arg), int ATTR_UNUSED(error),
1518         struct comm_reply* ATTR_UNUSED(reply_info))
1519 {
1520         log_assert(0);
1521         return 0;
1522 }
1523
1524 void libworker_handle_control_cmd(struct tube* ATTR_UNUSED(tube),
1525         uint8_t* ATTR_UNUSED(buffer), size_t ATTR_UNUSED(len),
1526         int ATTR_UNUSED(error), void* ATTR_UNUSED(arg))
1527 {
1528         log_assert(0);
1529 }
1530
1531 void libworker_fg_done_cb(void* ATTR_UNUSED(arg), int ATTR_UNUSED(rcode),
1532         sldns_buffer* ATTR_UNUSED(buf), enum sec_status ATTR_UNUSED(s),
1533         char* ATTR_UNUSED(why_bogus))
1534 {
1535         log_assert(0);
1536 }
1537
1538 void libworker_bg_done_cb(void* ATTR_UNUSED(arg), int ATTR_UNUSED(rcode),
1539         sldns_buffer* ATTR_UNUSED(buf), enum sec_status ATTR_UNUSED(s),
1540         char* ATTR_UNUSED(why_bogus))
1541 {
1542         log_assert(0);
1543 }
1544
1545 void libworker_event_done_cb(void* ATTR_UNUSED(arg), int ATTR_UNUSED(rcode),
1546         sldns_buffer* ATTR_UNUSED(buf), enum sec_status ATTR_UNUSED(s),
1547         char* ATTR_UNUSED(why_bogus))
1548 {
1549         log_assert(0);
1550 }
1551
1552 int context_query_cmp(const void* ATTR_UNUSED(a), const void* ATTR_UNUSED(b))
1553 {
1554         log_assert(0);
1555         return 0;
1556 }
1557
1558 int order_lock_cmp(const void* ATTR_UNUSED(e1), const void* ATTR_UNUSED(e2))
1559 {
1560         log_assert(0);
1561         return 0;
1562 }
1563
1564 int codeline_cmp(const void* ATTR_UNUSED(a), const void* ATTR_UNUSED(b))
1565 {
1566         log_assert(0);
1567         return 0;
1568 }
1569