]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/blob - contrib/unbound/daemon/worker.c
MFV r337167: 9442 decrease indirect block size of spacemaps
[FreeBSD/FreeBSD.git] / 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/authzone.h"
62 #include "services/mesh.h"
63 #include "services/localzone.h"
64 #include "util/data/msgparse.h"
65 #include "util/data/msgencode.h"
66 #include "util/data/dname.h"
67 #include "util/fptr_wlist.h"
68 #include "util/tube.h"
69 #include "iterator/iter_fwd.h"
70 #include "iterator/iter_hints.h"
71 #include "validator/autotrust.h"
72 #include "validator/val_anchor.h"
73 #include "respip/respip.h"
74 #include "libunbound/context.h"
75 #include "libunbound/libworker.h"
76 #include "sldns/sbuffer.h"
77 #include "sldns/wire2str.h"
78 #include "util/shm_side/shm_main.h"
79 #include "dnscrypt/dnscrypt.h"
80
81 #ifdef HAVE_SYS_TYPES_H
82 #  include <sys/types.h>
83 #endif
84 #ifdef HAVE_NETDB_H
85 #include <netdb.h>
86 #endif
87 #include <signal.h>
88 #ifdef UB_ON_WINDOWS
89 #include "winrc/win_svc.h"
90 #endif
91
92 /** Size of an UDP datagram */
93 #define NORMAL_UDP_SIZE 512 /* bytes */
94 /** ratelimit for error responses */
95 #define ERROR_RATELIMIT 100 /* qps */
96
97 /** 
98  * seconds to add to prefetch leeway.  This is a TTL that expires old rrsets
99  * earlier than they should in order to put the new update into the cache.
100  * This additional value is to make sure that if not all TTLs are equal in
101  * the message to be updated(and replaced), that rrsets with up to this much
102  * extra TTL are also replaced.  This means that the resulting new message
103  * will have (most likely) this TTL at least, avoiding very small 'split
104  * second' TTLs due to operators choosing relative primes for TTLs (or so).
105  * Also has to be at least one to break ties (and overwrite cached entry).
106  */
107 #define PREFETCH_EXPIRY_ADD 60
108
109 /** Report on memory usage by this thread and global */
110 static void
111 worker_mem_report(struct worker* ATTR_UNUSED(worker), 
112         struct serviced_query* ATTR_UNUSED(cur_serv))
113 {
114 #ifdef UNBOUND_ALLOC_STATS
115         /* measure memory leakage */
116         extern size_t unbound_mem_alloc, unbound_mem_freed;
117         /* debug func in validator module */
118         size_t total, front, back, mesh, msg, rrset, infra, ac, superac;
119         size_t me, iter, val, anch;
120         int i;
121 #ifdef CLIENT_SUBNET
122         size_t subnet = 0;
123 #endif /* CLIENT_SUBNET */
124         if(verbosity < VERB_ALGO) 
125                 return;
126         front = listen_get_mem(worker->front);
127         back = outnet_get_mem(worker->back);
128         msg = slabhash_get_mem(worker->env.msg_cache);
129         rrset = slabhash_get_mem(&worker->env.rrset_cache->table);
130         infra = infra_get_mem(worker->env.infra_cache);
131         mesh = mesh_get_mem(worker->env.mesh);
132         ac = alloc_get_mem(&worker->alloc);
133         superac = alloc_get_mem(&worker->daemon->superalloc);
134         anch = anchors_get_mem(worker->env.anchors);
135         iter = 0;
136         val = 0;
137         for(i=0; i<worker->env.mesh->mods.num; i++) {
138                 fptr_ok(fptr_whitelist_mod_get_mem(worker->env.mesh->
139                         mods.mod[i]->get_mem));
140                 if(strcmp(worker->env.mesh->mods.mod[i]->name, "validator")==0)
141                         val += (*worker->env.mesh->mods.mod[i]->get_mem)
142                                 (&worker->env, i);
143 #ifdef CLIENT_SUBNET
144                 else if(strcmp(worker->env.mesh->mods.mod[i]->name,
145                         "subnet")==0)
146                         subnet += (*worker->env.mesh->mods.mod[i]->get_mem)
147                                 (&worker->env, i);
148 #endif /* CLIENT_SUBNET */
149                 else    iter += (*worker->env.mesh->mods.mod[i]->get_mem)
150                                 (&worker->env, i);
151         }
152         me = sizeof(*worker) + sizeof(*worker->base) + sizeof(*worker->comsig)
153                 + comm_point_get_mem(worker->cmd_com) 
154                 + sizeof(worker->rndstate) 
155                 + regional_get_mem(worker->scratchpad) 
156                 + sizeof(*worker->env.scratch_buffer) 
157                 + sldns_buffer_capacity(worker->env.scratch_buffer)
158                 + forwards_get_mem(worker->env.fwds)
159                 + hints_get_mem(worker->env.hints);
160         if(worker->thread_num == 0)
161                 me += acl_list_get_mem(worker->daemon->acl);
162         if(cur_serv) {
163                 me += serviced_get_mem(cur_serv);
164         }
165         total = front+back+mesh+msg+rrset+infra+iter+val+ac+superac+me;
166 #ifdef CLIENT_SUBNET
167         total += subnet;
168         log_info("Memory conditions: %u front=%u back=%u mesh=%u msg=%u "
169                 "rrset=%u infra=%u iter=%u val=%u subnet=%u anchors=%u "
170                 "alloccache=%u globalalloccache=%u me=%u",
171                 (unsigned)total, (unsigned)front, (unsigned)back, 
172                 (unsigned)mesh, (unsigned)msg, (unsigned)rrset, (unsigned)infra,
173                 (unsigned)iter, (unsigned)val,
174                 (unsigned)subnet, (unsigned)anch, (unsigned)ac,
175                 (unsigned)superac, (unsigned)me);
176 #else /* no CLIENT_SUBNET */
177         log_info("Memory conditions: %u front=%u back=%u mesh=%u msg=%u "
178                 "rrset=%u infra=%u iter=%u val=%u anchors=%u "
179                 "alloccache=%u globalalloccache=%u me=%u",
180                 (unsigned)total, (unsigned)front, (unsigned)back, 
181                 (unsigned)mesh, (unsigned)msg, (unsigned)rrset, 
182                 (unsigned)infra, (unsigned)iter, (unsigned)val, (unsigned)anch,
183                 (unsigned)ac, (unsigned)superac, (unsigned)me);
184 #endif /* CLIENT_SUBNET */
185         log_info("Total heap memory estimate: %u  total-alloc: %u  "
186                 "total-free: %u", (unsigned)total, 
187                 (unsigned)unbound_mem_alloc, (unsigned)unbound_mem_freed);
188 #else /* no UNBOUND_ALLOC_STATS */
189         size_t val = 0;
190 #ifdef CLIENT_SUBNET
191         size_t subnet = 0;
192 #endif /* CLIENT_SUBNET */
193         int i;
194         if(verbosity < VERB_QUERY)
195                 return;
196         for(i=0; i<worker->env.mesh->mods.num; i++) {
197                 fptr_ok(fptr_whitelist_mod_get_mem(worker->env.mesh->
198                         mods.mod[i]->get_mem));
199                 if(strcmp(worker->env.mesh->mods.mod[i]->name, "validator")==0)
200                         val += (*worker->env.mesh->mods.mod[i]->get_mem)
201                                 (&worker->env, i);
202 #ifdef CLIENT_SUBNET
203                 else if(strcmp(worker->env.mesh->mods.mod[i]->name,
204                         "subnet")==0)
205                         subnet += (*worker->env.mesh->mods.mod[i]->get_mem)
206                                 (&worker->env, i);
207 #endif /* CLIENT_SUBNET */
208         }
209 #ifdef CLIENT_SUBNET
210         verbose(VERB_QUERY, "cache memory msg=%u rrset=%u infra=%u val=%u "
211                 "subnet=%u",
212                 (unsigned)slabhash_get_mem(worker->env.msg_cache),
213                 (unsigned)slabhash_get_mem(&worker->env.rrset_cache->table),
214                 (unsigned)infra_get_mem(worker->env.infra_cache),
215                 (unsigned)val, (unsigned)subnet);
216 #else /* no CLIENT_SUBNET */
217         verbose(VERB_QUERY, "cache memory msg=%u rrset=%u infra=%u val=%u",
218                 (unsigned)slabhash_get_mem(worker->env.msg_cache),
219                 (unsigned)slabhash_get_mem(&worker->env.rrset_cache->table),
220                 (unsigned)infra_get_mem(worker->env.infra_cache),
221                 (unsigned)val);
222 #endif /* CLIENT_SUBNET */
223 #endif /* UNBOUND_ALLOC_STATS */
224 }
225
226 void 
227 worker_send_cmd(struct worker* worker, enum worker_commands cmd)
228 {
229         uint32_t c = (uint32_t)htonl(cmd);
230         if(!tube_write_msg(worker->cmd, (uint8_t*)&c, sizeof(c), 0)) {
231                 log_err("worker send cmd %d failed", (int)cmd);
232         }
233 }
234
235 int 
236 worker_handle_reply(struct comm_point* c, void* arg, int error, 
237         struct comm_reply* reply_info)
238 {
239         struct module_qstate* q = (struct module_qstate*)arg;
240         struct worker* worker = q->env->worker;
241         struct outbound_entry e;
242         e.qstate = q;
243         e.qsent = NULL;
244
245         if(error != 0) {
246                 mesh_report_reply(worker->env.mesh, &e, reply_info, error);
247                 worker_mem_report(worker, NULL);
248                 return 0;
249         }
250         /* sanity check. */
251         if(!LDNS_QR_WIRE(sldns_buffer_begin(c->buffer))
252                 || LDNS_OPCODE_WIRE(sldns_buffer_begin(c->buffer)) != 
253                         LDNS_PACKET_QUERY
254                 || LDNS_QDCOUNT(sldns_buffer_begin(c->buffer)) > 1) {
255                 /* error becomes timeout for the module as if this reply
256                  * never arrived. */
257                 mesh_report_reply(worker->env.mesh, &e, reply_info, 
258                         NETEVENT_TIMEOUT);
259                 worker_mem_report(worker, NULL);
260                 return 0;
261         }
262         mesh_report_reply(worker->env.mesh, &e, reply_info, NETEVENT_NOERROR);
263         worker_mem_report(worker, NULL);
264         return 0;
265 }
266
267 int 
268 worker_handle_service_reply(struct comm_point* c, void* arg, int error, 
269         struct comm_reply* reply_info)
270 {
271         struct outbound_entry* e = (struct outbound_entry*)arg;
272         struct worker* worker = e->qstate->env->worker;
273         struct serviced_query *sq = e->qsent;
274
275         verbose(VERB_ALGO, "worker svcd callback for qstate %p", e->qstate);
276         if(error != 0) {
277                 mesh_report_reply(worker->env.mesh, e, reply_info, error);
278                 worker_mem_report(worker, sq);
279                 return 0;
280         }
281         /* sanity check. */
282         if(!LDNS_QR_WIRE(sldns_buffer_begin(c->buffer))
283                 || LDNS_OPCODE_WIRE(sldns_buffer_begin(c->buffer)) != 
284                         LDNS_PACKET_QUERY
285                 || LDNS_QDCOUNT(sldns_buffer_begin(c->buffer)) > 1) {
286                 /* error becomes timeout for the module as if this reply
287                  * never arrived. */
288                 verbose(VERB_ALGO, "worker: bad reply handled as timeout");
289                 mesh_report_reply(worker->env.mesh, e, reply_info, 
290                         NETEVENT_TIMEOUT);
291                 worker_mem_report(worker, sq);
292                 return 0;
293         }
294         mesh_report_reply(worker->env.mesh, e, reply_info, NETEVENT_NOERROR);
295         worker_mem_report(worker, sq);
296         return 0;
297 }
298
299 /** ratelimit error replies
300  * @param worker: the worker struct with ratelimit counter
301  * @param err: error code that would be wanted.
302  * @return value of err if okay, or -1 if it should be discarded instead.
303  */
304 static int
305 worker_err_ratelimit(struct worker* worker, int err)
306 {
307         if(worker->err_limit_time == *worker->env.now) {
308                 /* see if limit is exceeded for this second */
309                 if(worker->err_limit_count++ > ERROR_RATELIMIT)
310                         return -1;
311         } else {
312                 /* new second, new limits */
313                 worker->err_limit_time = *worker->env.now;
314                 worker->err_limit_count = 1;
315         }
316         return err;
317 }
318
319 /** check request sanity.
320  * @param pkt: the wire packet to examine for sanity.
321  * @param worker: parameters for checking.
322  * @return error code, 0 OK, or -1 discard.
323 */
324 static int 
325 worker_check_request(sldns_buffer* pkt, struct worker* worker)
326 {
327         if(sldns_buffer_limit(pkt) < LDNS_HEADER_SIZE) {
328                 verbose(VERB_QUERY, "request too short, discarded");
329                 return -1;
330         }
331         if(sldns_buffer_limit(pkt) > NORMAL_UDP_SIZE && 
332                 worker->daemon->cfg->harden_large_queries) {
333                 verbose(VERB_QUERY, "request too large, discarded");
334                 return -1;
335         }
336         if(LDNS_QR_WIRE(sldns_buffer_begin(pkt))) {
337                 verbose(VERB_QUERY, "request has QR bit on, discarded");
338                 return -1;
339         }
340         if(LDNS_TC_WIRE(sldns_buffer_begin(pkt))) {
341                 LDNS_TC_CLR(sldns_buffer_begin(pkt));
342                 verbose(VERB_QUERY, "request bad, has TC bit on");
343                 return worker_err_ratelimit(worker, LDNS_RCODE_FORMERR);
344         }
345         if(LDNS_OPCODE_WIRE(sldns_buffer_begin(pkt)) != LDNS_PACKET_QUERY &&
346                 LDNS_OPCODE_WIRE(sldns_buffer_begin(pkt)) != LDNS_PACKET_NOTIFY) {
347                 verbose(VERB_QUERY, "request unknown opcode %d", 
348                         LDNS_OPCODE_WIRE(sldns_buffer_begin(pkt)));
349                 return worker_err_ratelimit(worker, LDNS_RCODE_NOTIMPL);
350         }
351         if(LDNS_QDCOUNT(sldns_buffer_begin(pkt)) != 1) {
352                 verbose(VERB_QUERY, "request wrong nr qd=%d", 
353                         LDNS_QDCOUNT(sldns_buffer_begin(pkt)));
354                 return worker_err_ratelimit(worker, LDNS_RCODE_FORMERR);
355         }
356         if(LDNS_ANCOUNT(sldns_buffer_begin(pkt)) != 0 && 
357                 (LDNS_ANCOUNT(sldns_buffer_begin(pkt)) != 1 ||
358                 LDNS_OPCODE_WIRE(sldns_buffer_begin(pkt)) != LDNS_PACKET_NOTIFY)) {
359                 verbose(VERB_QUERY, "request wrong nr an=%d", 
360                         LDNS_ANCOUNT(sldns_buffer_begin(pkt)));
361                 return worker_err_ratelimit(worker, LDNS_RCODE_FORMERR);
362         }
363         if(LDNS_NSCOUNT(sldns_buffer_begin(pkt)) != 0) {
364                 verbose(VERB_QUERY, "request wrong nr ns=%d", 
365                         LDNS_NSCOUNT(sldns_buffer_begin(pkt)));
366                 return worker_err_ratelimit(worker, LDNS_RCODE_FORMERR);
367         }
368         if(LDNS_ARCOUNT(sldns_buffer_begin(pkt)) > 1) {
369                 verbose(VERB_QUERY, "request wrong nr ar=%d", 
370                         LDNS_ARCOUNT(sldns_buffer_begin(pkt)));
371                 return worker_err_ratelimit(worker, LDNS_RCODE_FORMERR);
372         }
373         return 0;
374 }
375
376 void 
377 worker_handle_control_cmd(struct tube* ATTR_UNUSED(tube), uint8_t* msg,
378         size_t len, int error, void* arg)
379 {
380         struct worker* worker = (struct worker*)arg;
381         enum worker_commands cmd;
382         if(error != NETEVENT_NOERROR) {
383                 free(msg);
384                 if(error == NETEVENT_CLOSED)
385                         comm_base_exit(worker->base);
386                 else    log_info("control event: %d", error);
387                 return;
388         }
389         if(len != sizeof(uint32_t)) {
390                 fatal_exit("bad control msg length %d", (int)len);
391         }
392         cmd = sldns_read_uint32(msg);
393         free(msg);
394         switch(cmd) {
395         case worker_cmd_quit:
396                 verbose(VERB_ALGO, "got control cmd quit");
397                 comm_base_exit(worker->base);
398                 break;
399         case worker_cmd_stats:
400                 verbose(VERB_ALGO, "got control cmd stats");
401                 server_stats_reply(worker, 1);
402                 break;
403         case worker_cmd_stats_noreset:
404                 verbose(VERB_ALGO, "got control cmd stats_noreset");
405                 server_stats_reply(worker, 0);
406                 break;
407         case worker_cmd_remote:
408                 verbose(VERB_ALGO, "got control cmd remote");
409                 daemon_remote_exec(worker);
410                 break;
411         default:
412                 log_err("bad command %d", (int)cmd);
413                 break;
414         }
415 }
416
417 /** check if a delegation is secure */
418 static enum sec_status
419 check_delegation_secure(struct reply_info *rep) 
420 {
421         /* return smallest security status */
422         size_t i;
423         enum sec_status sec = sec_status_secure;
424         enum sec_status s;
425         size_t num = rep->an_numrrsets + rep->ns_numrrsets;
426         /* check if answer and authority are OK */
427         for(i=0; i<num; i++) {
428                 s = ((struct packed_rrset_data*)rep->rrsets[i]->entry.data)
429                         ->security;
430                 if(s < sec)
431                         sec = s;
432         }
433         /* in additional, only unchecked triggers revalidation */
434         for(i=num; i<rep->rrset_count; i++) {
435                 s = ((struct packed_rrset_data*)rep->rrsets[i]->entry.data)
436                         ->security;
437                 if(s == sec_status_unchecked)
438                         return s;
439         }
440         return sec;
441 }
442
443 /** remove nonsecure from a delegation referral additional section */
444 static void
445 deleg_remove_nonsecure_additional(struct reply_info* rep)
446 {
447         /* we can simply edit it, since we are working in the scratch region */
448         size_t i;
449         enum sec_status s;
450
451         for(i = rep->an_numrrsets+rep->ns_numrrsets; i<rep->rrset_count; i++) {
452                 s = ((struct packed_rrset_data*)rep->rrsets[i]->entry.data)
453                         ->security;
454                 if(s != sec_status_secure) {
455                         memmove(rep->rrsets+i, rep->rrsets+i+1, 
456                                 sizeof(struct ub_packed_rrset_key*)* 
457                                 (rep->rrset_count - i - 1));
458                         rep->ar_numrrsets--; 
459                         rep->rrset_count--;
460                         i--;
461                 }
462         }
463 }
464
465 /** answer nonrecursive query from the cache */
466 static int
467 answer_norec_from_cache(struct worker* worker, struct query_info* qinfo,
468         uint16_t id, uint16_t flags, struct comm_reply* repinfo, 
469         struct edns_data* edns)
470 {
471         /* for a nonrecursive query return either:
472          *      o an error (servfail; we try to avoid this)
473          *      o a delegation (closest we have; this routine tries that)
474          *      o the answer (checked by answer_from_cache) 
475          *
476          * So, grab a delegation from the rrset cache. 
477          * Then check if it needs validation, if so, this routine fails,
478          * so that iterator can prime and validator can verify rrsets.
479          */
480         uint16_t udpsize = edns->udp_size;
481         int secure = 0;
482         time_t timenow = *worker->env.now;
483         int must_validate = (!(flags&BIT_CD) || worker->env.cfg->ignore_cd)
484                 && worker->env.need_to_validate;
485         struct dns_msg *msg = NULL;
486         struct delegpt *dp;
487
488         dp = dns_cache_find_delegation(&worker->env, qinfo->qname, 
489                 qinfo->qname_len, qinfo->qtype, qinfo->qclass,
490                 worker->scratchpad, &msg, timenow);
491         if(!dp) { /* no delegation, need to reprime */
492                 return 0;
493         }
494         /* In case we have a local alias, copy it into the delegation message.
495          * Shallow copy should be fine, as we'll be done with msg in this
496          * function. */
497         msg->qinfo.local_alias = qinfo->local_alias;
498         if(must_validate) {
499                 switch(check_delegation_secure(msg->rep)) {
500                 case sec_status_unchecked:
501                         /* some rrsets have not been verified yet, go and 
502                          * let validator do that */
503                         return 0;
504                 case sec_status_bogus:
505                 case sec_status_secure_sentinel_fail:
506                         /* some rrsets are bogus, reply servfail */
507                         edns->edns_version = EDNS_ADVERTISED_VERSION;
508                         edns->udp_size = EDNS_ADVERTISED_SIZE;
509                         edns->ext_rcode = 0;
510                         edns->bits &= EDNS_DO;
511                         if(!inplace_cb_reply_servfail_call(&worker->env, qinfo, NULL,
512                                 msg->rep, LDNS_RCODE_SERVFAIL, edns, worker->scratchpad))
513                                         return 0;
514                         error_encode(repinfo->c->buffer, LDNS_RCODE_SERVFAIL, 
515                                 &msg->qinfo, id, flags, edns);
516                         if(worker->stats.extended) {
517                                 worker->stats.ans_bogus++;
518                                 worker->stats.ans_rcode[LDNS_RCODE_SERVFAIL]++;
519                         }
520                         return 1;
521                 case sec_status_secure:
522                         /* all rrsets are secure */
523                         /* remove non-secure rrsets from the add. section*/
524                         if(worker->env.cfg->val_clean_additional)
525                                 deleg_remove_nonsecure_additional(msg->rep);
526                         secure = 1;
527                         break;
528                 case sec_status_indeterminate:
529                 case sec_status_insecure:
530                 default:
531                         /* not secure */
532                         secure = 0;
533                         break;
534                 }
535         }
536         /* return this delegation from the cache */
537         edns->edns_version = EDNS_ADVERTISED_VERSION;
538         edns->udp_size = EDNS_ADVERTISED_SIZE;
539         edns->ext_rcode = 0;
540         edns->bits &= EDNS_DO;
541         if(!inplace_cb_reply_cache_call(&worker->env, qinfo, NULL, msg->rep,
542                 (int)(flags&LDNS_RCODE_MASK), edns, worker->scratchpad))
543                         return 0;
544         msg->rep->flags |= BIT_QR|BIT_RA;
545         if(!reply_info_answer_encode(&msg->qinfo, msg->rep, id, flags, 
546                 repinfo->c->buffer, 0, 1, worker->scratchpad,
547                 udpsize, edns, (int)(edns->bits & EDNS_DO), secure)) {
548                 if(!inplace_cb_reply_servfail_call(&worker->env, qinfo, NULL, NULL,
549                         LDNS_RCODE_SERVFAIL, edns, worker->scratchpad))
550                                 edns->opt_list = NULL;
551                 error_encode(repinfo->c->buffer, LDNS_RCODE_SERVFAIL, 
552                         &msg->qinfo, id, flags, edns);
553         }
554         if(worker->stats.extended) {
555                 if(secure) worker->stats.ans_secure++;
556                 server_stats_insrcode(&worker->stats, repinfo->c->buffer);
557         }
558         return 1;
559 }
560
561 /** Apply, if applicable, a response IP action to a cached answer.
562  * If the answer is rewritten as a result of an action, '*encode_repp' will
563  * point to the reply info containing the modified answer.  '*encode_repp' will
564  * be intact otherwise.
565  * It returns 1 on success, 0 otherwise. */
566 static int
567 apply_respip_action(struct worker* worker, const struct query_info* qinfo,
568         struct respip_client_info* cinfo, struct reply_info* rep,
569         struct comm_reply* repinfo, struct ub_packed_rrset_key** alias_rrset,
570         struct reply_info** encode_repp)
571 {
572         struct respip_action_info actinfo = {respip_none, NULL};
573
574         if(qinfo->qtype != LDNS_RR_TYPE_A &&
575                 qinfo->qtype != LDNS_RR_TYPE_AAAA &&
576                 qinfo->qtype != LDNS_RR_TYPE_ANY)
577                 return 1;
578
579         if(!respip_rewrite_reply(qinfo, cinfo, rep, encode_repp, &actinfo,
580                 alias_rrset, 0, worker->scratchpad))
581                 return 0;
582
583         /* xxx_deny actions mean dropping the reply, unless the original reply
584          * was redirected to response-ip data. */
585         if((actinfo.action == respip_deny ||
586                 actinfo.action == respip_inform_deny) &&
587                 *encode_repp == rep)
588                 *encode_repp = NULL;
589
590         /* If address info is returned, it means the action should be an
591          * 'inform' variant and the information should be logged. */
592         if(actinfo.addrinfo) {
593                 respip_inform_print(actinfo.addrinfo, qinfo->qname,
594                         qinfo->qtype, qinfo->qclass, qinfo->local_alias,
595                         repinfo);
596         }
597
598         return 1;
599 }
600
601 /** answer query from the cache.
602  * Normally, the answer message will be built in repinfo->c->buffer; if the
603  * answer is supposed to be suppressed or the answer is supposed to be an
604  * incomplete CNAME chain, the buffer is explicitly cleared to signal the
605  * caller as such.  In the latter case *partial_rep will point to the incomplete
606  * reply, and this function is (possibly) supposed to be called again with that
607  * *partial_rep value to complete the chain.  In addition, if the query should
608  * be completely dropped, '*need_drop' will be set to 1. */
609 static int
610 answer_from_cache(struct worker* worker, struct query_info* qinfo,
611         struct respip_client_info* cinfo, int* need_drop,
612         struct ub_packed_rrset_key** alias_rrset,
613         struct reply_info** partial_repp,
614         struct reply_info* rep, uint16_t id, uint16_t flags, 
615         struct comm_reply* repinfo, struct edns_data* edns)
616 {
617         time_t timenow = *worker->env.now;
618         uint16_t udpsize = edns->udp_size;
619         struct reply_info* encode_rep = rep;
620         struct reply_info* partial_rep = *partial_repp;
621         int secure;
622         int must_validate = (!(flags&BIT_CD) || worker->env.cfg->ignore_cd)
623                 && worker->env.need_to_validate;
624         *partial_repp = NULL;   /* avoid accidental further pass */
625         if(worker->env.cfg->serve_expired) {
626                 /* always lock rrsets, rep->ttl is ignored */
627                 if(!rrset_array_lock(rep->ref, rep->rrset_count, 0))
628                         return 0;
629                 /* below, rrsets with ttl before timenow become TTL 0 in
630                  * the response */
631                 /* This response was served with zero TTL */
632                 if (timenow >= rep->ttl) {
633                         worker->stats.zero_ttl_responses++;
634                 }
635         } else {
636                 /* see if it is possible */
637                 if(rep->ttl < timenow) {
638                         /* the rrsets may have been updated in the meantime.
639                          * we will refetch the message format from the
640                          * authoritative server 
641                          */
642                         return 0;
643                 }
644                 if(!rrset_array_lock(rep->ref, rep->rrset_count, timenow))
645                         return 0;
646                 /* locked and ids and ttls are OK. */
647         }
648         /* check CNAME chain (if any) */
649         if(rep->an_numrrsets > 0 && (rep->rrsets[0]->rk.type == 
650                 htons(LDNS_RR_TYPE_CNAME) || rep->rrsets[0]->rk.type == 
651                 htons(LDNS_RR_TYPE_DNAME))) {
652                 if(!reply_check_cname_chain(qinfo, rep)) {
653                         /* cname chain invalid, redo iterator steps */
654                         verbose(VERB_ALGO, "Cache reply: cname chain broken");
655                 bail_out:
656                         rrset_array_unlock_touch(worker->env.rrset_cache, 
657                                 worker->scratchpad, rep->ref, rep->rrset_count);
658                         return 0;
659                 }
660         }
661         /* check security status of the cached answer */
662         if(must_validate && (rep->security == sec_status_bogus ||
663                 rep->security == sec_status_secure_sentinel_fail)) {
664                 /* BAD cached */
665                 edns->edns_version = EDNS_ADVERTISED_VERSION;
666                 edns->udp_size = EDNS_ADVERTISED_SIZE;
667                 edns->ext_rcode = 0;
668                 edns->bits &= EDNS_DO;
669                 if(!inplace_cb_reply_servfail_call(&worker->env, qinfo, NULL, rep,
670                         LDNS_RCODE_SERVFAIL, edns, worker->scratchpad))
671                         goto bail_out;
672                 error_encode(repinfo->c->buffer, LDNS_RCODE_SERVFAIL, 
673                         qinfo, id, flags, edns);
674                 rrset_array_unlock_touch(worker->env.rrset_cache, 
675                         worker->scratchpad, rep->ref, rep->rrset_count);
676                 if(worker->stats.extended) {
677                         worker->stats.ans_bogus ++;
678                         worker->stats.ans_rcode[LDNS_RCODE_SERVFAIL] ++;
679                 }
680                 return 1;
681         } else if( rep->security == sec_status_unchecked && must_validate) {
682                 verbose(VERB_ALGO, "Cache reply: unchecked entry needs "
683                         "validation");
684                 goto bail_out; /* need to validate cache entry first */
685         } else if(rep->security == sec_status_secure) {
686                 if(reply_all_rrsets_secure(rep))
687                         secure = 1;
688                 else    {
689                         if(must_validate) {
690                                 verbose(VERB_ALGO, "Cache reply: secure entry"
691                                         " changed status");
692                                 goto bail_out; /* rrset changed, re-verify */
693                         }
694                         secure = 0;
695                 }
696         } else  secure = 0;
697
698         edns->edns_version = EDNS_ADVERTISED_VERSION;
699         edns->udp_size = EDNS_ADVERTISED_SIZE;
700         edns->ext_rcode = 0;
701         edns->bits &= EDNS_DO;
702         if(!inplace_cb_reply_cache_call(&worker->env, qinfo, NULL, rep,
703                 (int)(flags&LDNS_RCODE_MASK), edns, worker->scratchpad))
704                 goto bail_out;
705         *alias_rrset = NULL; /* avoid confusion if caller set it to non-NULL */
706         if(worker->daemon->use_response_ip && !partial_rep &&
707            !apply_respip_action(worker, qinfo, cinfo, rep, repinfo, alias_rrset,
708                         &encode_rep)) {
709                 goto bail_out;
710         } else if(partial_rep &&
711                 !respip_merge_cname(partial_rep, qinfo, rep, cinfo,
712                 must_validate, &encode_rep, worker->scratchpad)) {
713                 goto bail_out;
714         }
715         if(encode_rep != rep)
716                 secure = 0; /* if rewritten, it can't be considered "secure" */
717         if(!encode_rep || *alias_rrset) {
718                 sldns_buffer_clear(repinfo->c->buffer);
719                 sldns_buffer_flip(repinfo->c->buffer);
720                 if(!encode_rep)
721                         *need_drop = 1;
722                 else {
723                         /* If a partial CNAME chain is found, we first need to
724                          * make a copy of the reply in the scratchpad so we
725                          * can release the locks and lookup the cache again. */
726                         *partial_repp = reply_info_copy(encode_rep, NULL,
727                                 worker->scratchpad);
728                         if(!*partial_repp)
729                                 goto bail_out;
730                 }
731         } else if(!reply_info_answer_encode(qinfo, encode_rep, id, flags,
732                 repinfo->c->buffer, timenow, 1, worker->scratchpad,
733                 udpsize, edns, (int)(edns->bits & EDNS_DO), secure)) {
734                 if(!inplace_cb_reply_servfail_call(&worker->env, qinfo, NULL, NULL,
735                         LDNS_RCODE_SERVFAIL, edns, worker->scratchpad))
736                                 edns->opt_list = NULL;
737                 error_encode(repinfo->c->buffer, LDNS_RCODE_SERVFAIL, 
738                         qinfo, id, flags, edns);
739         }
740         /* cannot send the reply right now, because blocking network syscall
741          * is bad while holding locks. */
742         rrset_array_unlock_touch(worker->env.rrset_cache, worker->scratchpad,
743                 rep->ref, rep->rrset_count);
744         if(worker->stats.extended) {
745                 if(secure) worker->stats.ans_secure++;
746                 server_stats_insrcode(&worker->stats, repinfo->c->buffer);
747         }
748         /* go and return this buffer to the client */
749         return 1;
750 }
751
752 /** Reply to client and perform prefetch to keep cache up to date.
753  * If the buffer for the reply is empty, it indicates that only prefetch is
754  * necessary and the reply should be suppressed (because it's dropped or
755  * being deferred). */
756 static void
757 reply_and_prefetch(struct worker* worker, struct query_info* qinfo, 
758         uint16_t flags, struct comm_reply* repinfo, time_t leeway)
759 {
760         /* first send answer to client to keep its latency 
761          * as small as a cachereply */
762         if(sldns_buffer_limit(repinfo->c->buffer) != 0)
763                 comm_point_send_reply(repinfo);
764         server_stats_prefetch(&worker->stats, worker);
765         
766         /* create the prefetch in the mesh as a normal lookup without
767          * client addrs waiting, which has the cache blacklisted (to bypass
768          * the cache and go to the network for the data). */
769         /* this (potentially) runs the mesh for the new query */
770         mesh_new_prefetch(worker->env.mesh, qinfo, flags, leeway + 
771                 PREFETCH_EXPIRY_ADD);
772 }
773
774 /**
775  * Fill CH class answer into buffer. Keeps query.
776  * @param pkt: buffer
777  * @param str: string to put into text record (<255).
778  *      array of strings, every string becomes a text record.
779  * @param num: number of strings in array.
780  * @param edns: edns reply information.
781  * @param worker: worker with scratch region.
782  */
783 static void
784 chaos_replystr(sldns_buffer* pkt, char** str, int num, struct edns_data* edns,
785         struct worker* worker)
786 {
787         int i;
788         unsigned int rd = LDNS_RD_WIRE(sldns_buffer_begin(pkt));
789         unsigned int cd = LDNS_CD_WIRE(sldns_buffer_begin(pkt));
790         sldns_buffer_clear(pkt);
791         sldns_buffer_skip(pkt, (ssize_t)sizeof(uint16_t)); /* skip id */
792         sldns_buffer_write_u16(pkt, (uint16_t)(BIT_QR|BIT_RA));
793         if(rd) LDNS_RD_SET(sldns_buffer_begin(pkt));
794         if(cd) LDNS_CD_SET(sldns_buffer_begin(pkt));
795         sldns_buffer_write_u16(pkt, 1); /* qdcount */
796         sldns_buffer_write_u16(pkt, (uint16_t)num); /* ancount */
797         sldns_buffer_write_u16(pkt, 0); /* nscount */
798         sldns_buffer_write_u16(pkt, 0); /* arcount */
799         (void)query_dname_len(pkt); /* skip qname */
800         sldns_buffer_skip(pkt, (ssize_t)sizeof(uint16_t)); /* skip qtype */
801         sldns_buffer_skip(pkt, (ssize_t)sizeof(uint16_t)); /* skip qclass */
802         for(i=0; i<num; i++) {
803                 size_t len = strlen(str[i]);
804                 if(len>255) len=255; /* cap size of TXT record */
805                 sldns_buffer_write_u16(pkt, 0xc00c); /* compr ptr to query */
806                 sldns_buffer_write_u16(pkt, LDNS_RR_TYPE_TXT);
807                 sldns_buffer_write_u16(pkt, LDNS_RR_CLASS_CH);
808                 sldns_buffer_write_u32(pkt, 0); /* TTL */
809                 sldns_buffer_write_u16(pkt, sizeof(uint8_t) + len);
810                 sldns_buffer_write_u8(pkt, len);
811                 sldns_buffer_write(pkt, str[i], len);
812         }
813         sldns_buffer_flip(pkt);
814         edns->edns_version = EDNS_ADVERTISED_VERSION;
815         edns->udp_size = EDNS_ADVERTISED_SIZE;
816         edns->bits &= EDNS_DO;
817         if(!inplace_cb_reply_local_call(&worker->env, NULL, NULL, NULL,
818                 LDNS_RCODE_NOERROR, edns, worker->scratchpad))
819                         edns->opt_list = NULL;
820         if(sldns_buffer_capacity(pkt) >=
821                 sldns_buffer_limit(pkt)+calc_edns_field_size(edns))
822                 attach_edns_record(pkt, edns);
823 }
824
825 /** Reply with one string */
826 static void
827 chaos_replyonestr(sldns_buffer* pkt, const char* str, struct edns_data* edns,
828         struct worker* worker)
829 {
830         chaos_replystr(pkt, (char**)&str, 1, edns, worker);
831 }
832
833 /**
834  * Create CH class trustanchor answer.
835  * @param pkt: buffer
836  * @param edns: edns reply information.
837  * @param w: worker with scratch region.
838  */
839 static void
840 chaos_trustanchor(sldns_buffer* pkt, struct edns_data* edns, struct worker* w)
841 {
842 #define TA_RESPONSE_MAX_TXT 16 /* max number of TXT records */
843 #define TA_RESPONSE_MAX_TAGS 32 /* max number of tags printed per zone */
844         char* str_array[TA_RESPONSE_MAX_TXT];
845         uint16_t tags[TA_RESPONSE_MAX_TAGS];
846         int num = 0;
847         struct trust_anchor* ta;
848
849         if(!w->env.need_to_validate) {
850                 /* no validator module, reply no trustanchors */
851                 chaos_replystr(pkt, NULL, 0, edns, w);
852                 return;
853         }
854
855         /* fill the string with contents */
856         lock_basic_lock(&w->env.anchors->lock);
857         RBTREE_FOR(ta, struct trust_anchor*, w->env.anchors->tree) {
858                 char* str;
859                 size_t i, numtag, str_len = 255;
860                 if(num == TA_RESPONSE_MAX_TXT) continue;
861                 str = (char*)regional_alloc(w->scratchpad, str_len);
862                 if(!str) continue;
863                 lock_basic_lock(&ta->lock);
864                 numtag = anchor_list_keytags(ta, tags, TA_RESPONSE_MAX_TAGS);
865                 if(numtag == 0) {
866                         /* empty, insecure point */
867                         lock_basic_unlock(&ta->lock);
868                         continue;
869                 }
870                 str_array[num] = str;
871                 num++;
872
873                 /* spool name of anchor */
874                 (void)sldns_wire2str_dname_buf(ta->name, ta->namelen, str, str_len);
875                 str_len -= strlen(str); str += strlen(str);
876                 /* spool tags */
877                 for(i=0; i<numtag; i++) {
878                         snprintf(str, str_len, " %u", (unsigned)tags[i]);
879                         str_len -= strlen(str); str += strlen(str);
880                 }
881                 lock_basic_unlock(&ta->lock);
882         }
883         lock_basic_unlock(&w->env.anchors->lock);
884
885         chaos_replystr(pkt, str_array, num, edns, w);
886         regional_free_all(w->scratchpad);
887 }
888
889 /**
890  * Answer CH class queries.
891  * @param w: worker
892  * @param qinfo: query info. Pointer into packet buffer.
893  * @param edns: edns info from query.
894  * @param pkt: packet buffer.
895  * @return: true if a reply is to be sent.
896  */
897 static int
898 answer_chaos(struct worker* w, struct query_info* qinfo, 
899         struct edns_data* edns, sldns_buffer* pkt)
900 {
901         struct config_file* cfg = w->env.cfg;
902         if(qinfo->qtype != LDNS_RR_TYPE_ANY && qinfo->qtype != LDNS_RR_TYPE_TXT)
903                 return 0;
904         if(query_dname_compare(qinfo->qname, 
905                 (uint8_t*)"\002id\006server") == 0 ||
906                 query_dname_compare(qinfo->qname, 
907                 (uint8_t*)"\010hostname\004bind") == 0)
908         {
909                 if(cfg->hide_identity) 
910                         return 0;
911                 if(cfg->identity==NULL || cfg->identity[0]==0) {
912                         char buf[MAXHOSTNAMELEN+1];
913                         if (gethostname(buf, MAXHOSTNAMELEN) == 0) {
914                                 buf[MAXHOSTNAMELEN] = 0;
915                                 chaos_replyonestr(pkt, buf, edns, w);
916                         } else  {
917                                 log_err("gethostname: %s", strerror(errno));
918                                 chaos_replyonestr(pkt, "no hostname", edns, w);
919                         }
920                 }
921                 else    chaos_replyonestr(pkt, cfg->identity, edns, w);
922                 return 1;
923         }
924         if(query_dname_compare(qinfo->qname, 
925                 (uint8_t*)"\007version\006server") == 0 ||
926                 query_dname_compare(qinfo->qname, 
927                 (uint8_t*)"\007version\004bind") == 0)
928         {
929                 if(cfg->hide_version) 
930                         return 0;
931                 if(cfg->version==NULL || cfg->version[0]==0)
932                         chaos_replyonestr(pkt, PACKAGE_STRING, edns, w);
933                 else    chaos_replyonestr(pkt, cfg->version, edns, w);
934                 return 1;
935         }
936         if(query_dname_compare(qinfo->qname,
937                 (uint8_t*)"\013trustanchor\007unbound") == 0)
938         {
939                 if(cfg->hide_trustanchor)
940                         return 0;
941                 chaos_trustanchor(pkt, edns, w);
942                 return 1;
943         }
944
945         return 0;
946 }
947
948 /**
949  * Answer notify queries.  These are notifies for authoritative zones,
950  * the reply is an ack that the notify has been received.  We need to check
951  * access permission here.
952  * @param w: worker
953  * @param qinfo: query info. Pointer into packet buffer.
954  * @param edns: edns info from query.
955  * @param repinfo: reply info with source address.
956  * @param pkt: packet buffer.
957  */
958 static void
959 answer_notify(struct worker* w, struct query_info* qinfo, 
960         struct edns_data* edns, sldns_buffer* pkt, struct comm_reply* repinfo)
961 {
962         int refused = 0;
963         int rcode = LDNS_RCODE_NOERROR;
964         uint32_t serial = 0;
965         int has_serial;
966         if(!w->env.auth_zones) return;
967         has_serial = auth_zone_parse_notify_serial(pkt, &serial);
968         if(auth_zones_notify(w->env.auth_zones, &w->env, qinfo->qname,
969                 qinfo->qname_len, qinfo->qclass, &repinfo->addr,
970                 repinfo->addrlen, has_serial, serial, &refused)) {
971                 rcode = LDNS_RCODE_NOERROR;
972         } else {
973                 if(refused)
974                         rcode = LDNS_RCODE_REFUSED;
975                 else    rcode = LDNS_RCODE_SERVFAIL;
976         }
977
978         if(verbosity >= VERB_DETAIL) {
979                 char buf[380];
980                 char zname[255+1];
981                 char sr[25];
982                 dname_str(qinfo->qname, zname);
983                 sr[0]=0;
984                 if(has_serial)
985                         snprintf(sr, sizeof(sr), "serial %u ",
986                                 (unsigned)serial);
987                 if(rcode == LDNS_RCODE_REFUSED)
988                         snprintf(buf, sizeof(buf),
989                                 "refused NOTIFY %sfor %s from", sr, zname);
990                 else if(rcode == LDNS_RCODE_SERVFAIL)
991                         snprintf(buf, sizeof(buf),
992                                 "servfail for NOTIFY %sfor %s from", sr, zname);
993                 else    snprintf(buf, sizeof(buf),
994                                 "received NOTIFY %sfor %s from", sr, zname);
995                 log_addr(VERB_DETAIL, buf, &repinfo->addr, repinfo->addrlen);
996         }
997         edns->edns_version = EDNS_ADVERTISED_VERSION;
998         edns->udp_size = EDNS_ADVERTISED_SIZE;
999         edns->ext_rcode = 0;
1000         edns->bits &= EDNS_DO;
1001         edns->opt_list = NULL;
1002         error_encode(pkt, rcode, qinfo,
1003                 *(uint16_t*)(void *)sldns_buffer_begin(pkt),
1004                 sldns_buffer_read_u16_at(pkt, 2), edns);
1005         LDNS_OPCODE_SET(sldns_buffer_begin(pkt), LDNS_PACKET_NOTIFY);
1006 }
1007
1008 static int
1009 deny_refuse(struct comm_point* c, enum acl_access acl,
1010         enum acl_access deny, enum acl_access refuse,
1011         struct worker* worker, struct comm_reply* repinfo)
1012 {
1013         if(acl == deny) {
1014                 comm_point_drop_reply(repinfo);
1015                 if(worker->stats.extended)
1016                         worker->stats.unwanted_queries++;
1017                 return 0;
1018         } else if(acl == refuse) {
1019                 log_addr(VERB_ALGO, "refused query from",
1020                         &repinfo->addr, repinfo->addrlen);
1021                 log_buf(VERB_ALGO, "refuse", c->buffer);
1022                 if(worker->stats.extended)
1023                         worker->stats.unwanted_queries++;
1024                 if(worker_check_request(c->buffer, worker) == -1) {
1025                         comm_point_drop_reply(repinfo);
1026                         return 0; /* discard this */
1027                 }
1028                 sldns_buffer_set_limit(c->buffer, LDNS_HEADER_SIZE);
1029                 sldns_buffer_write_at(c->buffer, 4, 
1030                         (uint8_t*)"\0\0\0\0\0\0\0\0", 8);
1031                 LDNS_QR_SET(sldns_buffer_begin(c->buffer));
1032                 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), 
1033                         LDNS_RCODE_REFUSED);
1034                 sldns_buffer_set_position(c->buffer, LDNS_HEADER_SIZE);
1035                 sldns_buffer_flip(c->buffer);
1036                 return 1;
1037         }
1038
1039         return -1;
1040 }
1041
1042 static int
1043 deny_refuse_all(struct comm_point* c, enum acl_access acl,
1044         struct worker* worker, struct comm_reply* repinfo)
1045 {
1046         return deny_refuse(c, acl, acl_deny, acl_refuse, worker, repinfo);
1047 }
1048
1049 static int
1050 deny_refuse_non_local(struct comm_point* c, enum acl_access acl,
1051         struct worker* worker, struct comm_reply* repinfo)
1052 {
1053         return deny_refuse(c, acl, acl_deny_non_local, acl_refuse_non_local, worker, repinfo);
1054 }
1055
1056 int 
1057 worker_handle_request(struct comm_point* c, void* arg, int error,
1058         struct comm_reply* repinfo)
1059 {
1060         struct worker* worker = (struct worker*)arg;
1061         int ret;
1062         hashvalue_type h;
1063         struct lruhash_entry* e;
1064         struct query_info qinfo;
1065         struct edns_data edns;
1066         enum acl_access acl;
1067         struct acl_addr* acladdr;
1068         int rc = 0;
1069         int need_drop = 0;
1070         /* We might have to chase a CNAME chain internally, in which case
1071          * we'll have up to two replies and combine them to build a complete
1072          * answer.  These variables control this case. */
1073         struct ub_packed_rrset_key* alias_rrset = NULL;
1074         struct reply_info* partial_rep = NULL;
1075         struct query_info* lookup_qinfo = &qinfo;
1076         struct query_info qinfo_tmp; /* placeholdoer for lookup_qinfo */
1077         struct respip_client_info* cinfo = NULL, cinfo_tmp;
1078         memset(&qinfo, 0, sizeof(qinfo));
1079
1080         if(error != NETEVENT_NOERROR) {
1081                 /* some bad tcp query DNS formats give these error calls */
1082                 verbose(VERB_ALGO, "handle request called with err=%d", error);
1083                 return 0;
1084         }
1085 #ifdef USE_DNSCRYPT
1086         repinfo->max_udp_size = worker->daemon->cfg->max_udp_size;
1087         if(!dnsc_handle_curved_request(worker->daemon->dnscenv, repinfo)) {
1088                 worker->stats.num_query_dnscrypt_crypted_malformed++;
1089                 return 0;
1090         }
1091         if(c->dnscrypt && !repinfo->is_dnscrypted) {
1092                 char buf[LDNS_MAX_DOMAINLEN+1];
1093                 /* Check if this is unencrypted and asking for certs */
1094                 if(worker_check_request(c->buffer, worker) != 0) {
1095                         verbose(VERB_ALGO,
1096                                 "dnscrypt: worker check request: bad query.");
1097                         log_addr(VERB_CLIENT,"from",&repinfo->addr,
1098                                 repinfo->addrlen);
1099                         comm_point_drop_reply(repinfo);
1100                         return 0;
1101                 }
1102                 if(!query_info_parse(&qinfo, c->buffer)) {
1103                         verbose(VERB_ALGO,
1104                                 "dnscrypt: worker parse request: formerror.");
1105                         log_addr(VERB_CLIENT, "from", &repinfo->addr,
1106                                 repinfo->addrlen);
1107                         comm_point_drop_reply(repinfo);
1108                         return 0;
1109                 }
1110                 dname_str(qinfo.qname, buf);
1111                 if(!(qinfo.qtype == LDNS_RR_TYPE_TXT &&
1112                         strcasecmp(buf,
1113                         worker->daemon->dnscenv->provider_name) == 0)) {
1114                         verbose(VERB_ALGO,
1115                                 "dnscrypt: not TXT \"%s\". Received: %s \"%s\"",
1116                                 worker->daemon->dnscenv->provider_name,
1117                                 sldns_rr_descript(qinfo.qtype)->_name,
1118                                 buf);
1119                         comm_point_drop_reply(repinfo);
1120                         worker->stats.num_query_dnscrypt_cleartext++;
1121                         return 0;
1122                 }
1123                 worker->stats.num_query_dnscrypt_cert++;
1124                 sldns_buffer_rewind(c->buffer);
1125         } else if(c->dnscrypt && repinfo->is_dnscrypted) {
1126                 worker->stats.num_query_dnscrypt_crypted++;
1127         }
1128 #endif
1129 #ifdef USE_DNSTAP
1130         if(worker->dtenv.log_client_query_messages)
1131                 dt_msg_send_client_query(&worker->dtenv, &repinfo->addr, c->type,
1132                         c->buffer);
1133 #endif
1134         acladdr = acl_addr_lookup(worker->daemon->acl, &repinfo->addr, 
1135                 repinfo->addrlen);
1136         acl = acl_get_control(acladdr);
1137         if((ret=deny_refuse_all(c, acl, worker, repinfo)) != -1)
1138         {
1139                 if(ret == 1)
1140                         goto send_reply;
1141                 return ret;
1142         }
1143         if((ret=worker_check_request(c->buffer, worker)) != 0) {
1144                 verbose(VERB_ALGO, "worker check request: bad query.");
1145                 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen);
1146                 if(ret != -1) {
1147                         LDNS_QR_SET(sldns_buffer_begin(c->buffer));
1148                         LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), ret);
1149                         return 1;
1150                 }
1151                 comm_point_drop_reply(repinfo);
1152                 return 0;
1153         }
1154
1155         worker->stats.num_queries++;
1156
1157         /* check if this query should be dropped based on source ip rate limiting */
1158         if(!infra_ip_ratelimit_inc(worker->env.infra_cache, repinfo,
1159                         *worker->env.now)) {
1160                 /* See if we are passed through with slip factor */
1161                 if(worker->env.cfg->ip_ratelimit_factor != 0 &&
1162                         ub_random_max(worker->env.rnd,
1163                                                   worker->env.cfg->ip_ratelimit_factor) == 1) {
1164
1165                         char addrbuf[128];
1166                         addr_to_str(&repinfo->addr, repinfo->addrlen,
1167                                                 addrbuf, sizeof(addrbuf));
1168                   verbose(VERB_OPS, "ip_ratelimit allowed through for ip address %s ",
1169                                   addrbuf);
1170                 } else {
1171                         worker->stats.num_queries_ip_ratelimited++;
1172                         comm_point_drop_reply(repinfo);
1173                         return 0;
1174                 }
1175         }
1176
1177         /* see if query is in the cache */
1178         if(!query_info_parse(&qinfo, c->buffer)) {
1179                 verbose(VERB_ALGO, "worker parse request: formerror.");
1180                 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen);
1181                 memset(&qinfo, 0, sizeof(qinfo)); /* zero qinfo.qname */
1182                 if(worker_err_ratelimit(worker, LDNS_RCODE_FORMERR) == -1) {
1183                         comm_point_drop_reply(repinfo);
1184                         return 0;
1185                 }
1186                 sldns_buffer_rewind(c->buffer);
1187                 LDNS_QR_SET(sldns_buffer_begin(c->buffer));
1188                 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), 
1189                         LDNS_RCODE_FORMERR);
1190                 server_stats_insrcode(&worker->stats, c->buffer);
1191                 goto send_reply;
1192         }
1193         if(worker->env.cfg->log_queries) {
1194                 char ip[128];
1195                 addr_to_str(&repinfo->addr, repinfo->addrlen, ip, sizeof(ip));
1196                 log_nametypeclass(0, ip, qinfo.qname, qinfo.qtype, qinfo.qclass);
1197         }
1198         if(qinfo.qtype == LDNS_RR_TYPE_AXFR || 
1199                 qinfo.qtype == LDNS_RR_TYPE_IXFR) {
1200                 verbose(VERB_ALGO, "worker request: refused zone transfer.");
1201                 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen);
1202                 sldns_buffer_rewind(c->buffer);
1203                 LDNS_QR_SET(sldns_buffer_begin(c->buffer));
1204                 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), 
1205                         LDNS_RCODE_REFUSED);
1206                 if(worker->stats.extended) {
1207                         worker->stats.qtype[qinfo.qtype]++;
1208                         server_stats_insrcode(&worker->stats, c->buffer);
1209                 }
1210                 goto send_reply;
1211         }
1212         if(qinfo.qtype == LDNS_RR_TYPE_OPT || 
1213                 qinfo.qtype == LDNS_RR_TYPE_TSIG ||
1214                 qinfo.qtype == LDNS_RR_TYPE_TKEY ||
1215                 qinfo.qtype == LDNS_RR_TYPE_MAILA ||
1216                 qinfo.qtype == LDNS_RR_TYPE_MAILB ||
1217                 (qinfo.qtype >= 128 && qinfo.qtype <= 248)) {
1218                 verbose(VERB_ALGO, "worker request: formerror for meta-type.");
1219                 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen);
1220                 if(worker_err_ratelimit(worker, LDNS_RCODE_FORMERR) == -1) {
1221                         comm_point_drop_reply(repinfo);
1222                         return 0;
1223                 }
1224                 sldns_buffer_rewind(c->buffer);
1225                 LDNS_QR_SET(sldns_buffer_begin(c->buffer));
1226                 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), 
1227                         LDNS_RCODE_FORMERR);
1228                 if(worker->stats.extended) {
1229                         worker->stats.qtype[qinfo.qtype]++;
1230                         server_stats_insrcode(&worker->stats, c->buffer);
1231                 }
1232                 goto send_reply;
1233         }
1234         if((ret=parse_edns_from_pkt(c->buffer, &edns, worker->scratchpad)) != 0) {
1235                 struct edns_data reply_edns;
1236                 verbose(VERB_ALGO, "worker parse edns: formerror.");
1237                 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen);
1238                 memset(&reply_edns, 0, sizeof(reply_edns));
1239                 reply_edns.edns_present = 1;
1240                 reply_edns.udp_size = EDNS_ADVERTISED_SIZE;
1241                 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), ret);
1242                 error_encode(c->buffer, ret, &qinfo,
1243                         *(uint16_t*)(void *)sldns_buffer_begin(c->buffer),
1244                         sldns_buffer_read_u16_at(c->buffer, 2), &reply_edns);
1245                 regional_free_all(worker->scratchpad);
1246                 server_stats_insrcode(&worker->stats, c->buffer);
1247                 goto send_reply;
1248         }
1249         if(edns.edns_present && edns.edns_version != 0) {
1250                 edns.ext_rcode = (uint8_t)(EDNS_RCODE_BADVERS>>4);
1251                 edns.edns_version = EDNS_ADVERTISED_VERSION;
1252                 edns.udp_size = EDNS_ADVERTISED_SIZE;
1253                 edns.bits &= EDNS_DO;
1254                 edns.opt_list = NULL;
1255                 verbose(VERB_ALGO, "query with bad edns version.");
1256                 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen);
1257                 error_encode(c->buffer, EDNS_RCODE_BADVERS&0xf, &qinfo,
1258                         *(uint16_t*)(void *)sldns_buffer_begin(c->buffer),
1259                         sldns_buffer_read_u16_at(c->buffer, 2), NULL);
1260                 if(sldns_buffer_capacity(c->buffer) >=
1261                         sldns_buffer_limit(c->buffer)+calc_edns_field_size(&edns))
1262                         attach_edns_record(c->buffer, &edns);
1263                 regional_free_all(worker->scratchpad);
1264                 goto send_reply;
1265         }
1266         if(edns.edns_present && edns.udp_size < NORMAL_UDP_SIZE &&
1267                 worker->daemon->cfg->harden_short_bufsize) {
1268                 verbose(VERB_QUERY, "worker request: EDNS bufsize %d ignored",
1269                         (int)edns.udp_size);
1270                 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen);
1271                 edns.udp_size = NORMAL_UDP_SIZE;
1272         }
1273         if(edns.udp_size > worker->daemon->cfg->max_udp_size &&
1274                 c->type == comm_udp) {
1275                 verbose(VERB_QUERY,
1276                         "worker request: max UDP reply size modified"
1277                         " (%d to max-udp-size)", (int)edns.udp_size);
1278                 log_addr(VERB_CLIENT,"from",&repinfo->addr, repinfo->addrlen);
1279                 edns.udp_size = worker->daemon->cfg->max_udp_size;
1280         }
1281         if(edns.udp_size < LDNS_HEADER_SIZE) {
1282                 verbose(VERB_ALGO, "worker request: edns is too small.");
1283                 log_addr(VERB_CLIENT, "from", &repinfo->addr, repinfo->addrlen);
1284                 LDNS_QR_SET(sldns_buffer_begin(c->buffer));
1285                 LDNS_TC_SET(sldns_buffer_begin(c->buffer));
1286                 LDNS_RCODE_SET(sldns_buffer_begin(c->buffer), 
1287                         LDNS_RCODE_SERVFAIL);
1288                 sldns_buffer_set_position(c->buffer, LDNS_HEADER_SIZE);
1289                 sldns_buffer_write_at(c->buffer, 4, 
1290                         (uint8_t*)"\0\0\0\0\0\0\0\0", 8);
1291                 sldns_buffer_flip(c->buffer);
1292                 regional_free_all(worker->scratchpad);
1293                 goto send_reply;
1294         }
1295         if(worker->stats.extended)
1296                 server_stats_insquery(&worker->stats, c, qinfo.qtype,
1297                         qinfo.qclass, &edns, repinfo);
1298         if(c->type != comm_udp)
1299                 edns.udp_size = 65535; /* max size for TCP replies */
1300         if(qinfo.qclass == LDNS_RR_CLASS_CH && answer_chaos(worker, &qinfo,
1301                 &edns, c->buffer)) {
1302                 server_stats_insrcode(&worker->stats, c->buffer);
1303                 regional_free_all(worker->scratchpad);
1304                 goto send_reply;
1305         }
1306         if(LDNS_OPCODE_WIRE(sldns_buffer_begin(c->buffer)) ==
1307                 LDNS_PACKET_NOTIFY) {
1308                 answer_notify(worker, &qinfo, &edns, c->buffer, repinfo);
1309                 regional_free_all(worker->scratchpad);
1310                 goto send_reply;
1311         }
1312         if(local_zones_answer(worker->daemon->local_zones, &worker->env, &qinfo,
1313                 &edns, c->buffer, worker->scratchpad, repinfo, acladdr->taglist,
1314                 acladdr->taglen, acladdr->tag_actions,
1315                 acladdr->tag_actions_size, acladdr->tag_datas,
1316                 acladdr->tag_datas_size, worker->daemon->cfg->tagname,
1317                 worker->daemon->cfg->num_tags, acladdr->view)) {
1318                 regional_free_all(worker->scratchpad);
1319                 if(sldns_buffer_limit(c->buffer) == 0) {
1320                         comm_point_drop_reply(repinfo);
1321                         return 0;
1322                 }
1323                 server_stats_insrcode(&worker->stats, c->buffer);
1324                 goto send_reply;
1325         }
1326         if(worker->env.auth_zones &&
1327                 auth_zones_answer(worker->env.auth_zones, &worker->env,
1328                 &qinfo, &edns, c->buffer, worker->scratchpad)) {
1329                 regional_free_all(worker->scratchpad);
1330                 if(sldns_buffer_limit(c->buffer) == 0) {
1331                         comm_point_drop_reply(repinfo);
1332                         return 0;
1333                 }
1334                 /* set RA for everyone that can have recursion (based on
1335                  * access control list) */
1336                 if(LDNS_RD_WIRE(sldns_buffer_begin(c->buffer)) &&
1337                    acl != acl_deny_non_local && acl != acl_refuse_non_local)
1338                         LDNS_RA_SET(sldns_buffer_begin(c->buffer));
1339                 server_stats_insrcode(&worker->stats, c->buffer);
1340                 goto send_reply;
1341         }
1342
1343         /* We've looked in our local zones. If the answer isn't there, we
1344          * might need to bail out based on ACLs now. */
1345         if((ret=deny_refuse_non_local(c, acl, worker, repinfo)) != -1)
1346         {
1347                 regional_free_all(worker->scratchpad);
1348                 if(ret == 1)
1349                         goto send_reply;
1350                 return ret;
1351         }
1352
1353         /* If this request does not have the recursion bit set, verify
1354          * ACLs allow the snooping. */
1355         if(!(LDNS_RD_WIRE(sldns_buffer_begin(c->buffer))) &&
1356                 acl != acl_allow_snoop ) {
1357                 error_encode(c->buffer, LDNS_RCODE_REFUSED, &qinfo,
1358                         *(uint16_t*)(void *)sldns_buffer_begin(c->buffer),
1359                         sldns_buffer_read_u16_at(c->buffer, 2), NULL);
1360                 regional_free_all(worker->scratchpad);
1361                 server_stats_insrcode(&worker->stats, c->buffer);
1362                 log_addr(VERB_ALGO, "refused nonrec (cache snoop) query from",
1363                         &repinfo->addr, repinfo->addrlen);
1364                 goto send_reply;
1365         }
1366
1367         /* If we've found a local alias, replace the qname with the alias
1368          * target before resolving it. */
1369         if(qinfo.local_alias) {
1370                 struct ub_packed_rrset_key* rrset = qinfo.local_alias->rrset;
1371                 struct packed_rrset_data* d = rrset->entry.data;
1372
1373                 /* Sanity check: our current implementation only supports
1374                  * a single CNAME RRset as a local alias. */
1375                 if(qinfo.local_alias->next ||
1376                         rrset->rk.type != htons(LDNS_RR_TYPE_CNAME) ||
1377                         d->count != 1) {
1378                         log_err("assumption failure: unexpected local alias");
1379                         regional_free_all(worker->scratchpad);
1380                         return 0; /* drop it */
1381                 }
1382                 qinfo.qname = d->rr_data[0] + 2;
1383                 qinfo.qname_len = d->rr_len[0] - 2;
1384         }
1385
1386         /* If we may apply IP-based actions to the answer, build the client
1387          * information.  As this can be expensive, skip it if there is
1388          * absolutely no possibility of it. */
1389         if(worker->daemon->use_response_ip &&
1390                 (qinfo.qtype == LDNS_RR_TYPE_A ||
1391                 qinfo.qtype == LDNS_RR_TYPE_AAAA ||
1392                 qinfo.qtype == LDNS_RR_TYPE_ANY)) {
1393                 cinfo_tmp.taglist = acladdr->taglist;
1394                 cinfo_tmp.taglen = acladdr->taglen;
1395                 cinfo_tmp.tag_actions = acladdr->tag_actions;
1396                 cinfo_tmp.tag_actions_size = acladdr->tag_actions_size;
1397                 cinfo_tmp.tag_datas = acladdr->tag_datas;
1398                 cinfo_tmp.tag_datas_size = acladdr->tag_datas_size;
1399                 cinfo_tmp.view = acladdr->view;
1400                 cinfo_tmp.respip_set = worker->daemon->respip_set;
1401                 cinfo = &cinfo_tmp;
1402         }
1403
1404 lookup_cache:
1405         /* Lookup the cache.  In case we chase an intermediate CNAME chain
1406          * this is a two-pass operation, and lookup_qinfo is different for
1407          * each pass.  We should still pass the original qinfo to
1408          * answer_from_cache(), however, since it's used to build the reply. */
1409         if(!edns_bypass_cache_stage(edns.opt_list, &worker->env)) {
1410                 h = query_info_hash(lookup_qinfo, sldns_buffer_read_u16_at(c->buffer, 2));
1411                 if((e=slabhash_lookup(worker->env.msg_cache, h, lookup_qinfo, 0))) {
1412                         /* answer from cache - we have acquired a readlock on it */
1413                         if(answer_from_cache(worker, &qinfo,
1414                                 cinfo, &need_drop, &alias_rrset, &partial_rep,
1415                                 (struct reply_info*)e->data,
1416                                 *(uint16_t*)(void *)sldns_buffer_begin(c->buffer),
1417                                 sldns_buffer_read_u16_at(c->buffer, 2), repinfo,
1418                                 &edns)) {
1419                                 /* prefetch it if the prefetch TTL expired.
1420                                  * Note that if there is more than one pass
1421                                  * its qname must be that used for cache
1422                                  * lookup. */
1423                                 if((worker->env.cfg->prefetch || worker->env.cfg->serve_expired)
1424                                         && *worker->env.now >=
1425                                         ((struct reply_info*)e->data)->prefetch_ttl) {
1426                                         time_t leeway = ((struct reply_info*)e->
1427                                                 data)->ttl - *worker->env.now;
1428                                         if(((struct reply_info*)e->data)->ttl
1429                                                 < *worker->env.now)
1430                                                 leeway = 0;
1431                                         lock_rw_unlock(&e->lock);
1432                                         reply_and_prefetch(worker, lookup_qinfo,
1433                                                 sldns_buffer_read_u16_at(c->buffer, 2),
1434                                                 repinfo, leeway);
1435                                         if(!partial_rep) {
1436                                                 rc = 0;
1437                                                 regional_free_all(worker->scratchpad);
1438                                                 goto send_reply_rc;
1439                                         }
1440                                 } else if(!partial_rep) {
1441                                         lock_rw_unlock(&e->lock);
1442                                         regional_free_all(worker->scratchpad);
1443                                         goto send_reply;
1444                                 } else {
1445                                         /* Note that we've already released the
1446                                          * lock if we're here after prefetch. */
1447                                         lock_rw_unlock(&e->lock);
1448                                 }
1449                                 /* We've found a partial reply ending with an
1450                                  * alias.  Replace the lookup qinfo for the
1451                                  * alias target and lookup the cache again to
1452                                  * (possibly) complete the reply.  As we're
1453                                  * passing the "base" reply, there will be no
1454                                  * more alias chasing. */
1455                                 memset(&qinfo_tmp, 0, sizeof(qinfo_tmp));
1456                                 get_cname_target(alias_rrset, &qinfo_tmp.qname,
1457                                         &qinfo_tmp.qname_len);
1458                                 if(!qinfo_tmp.qname) {
1459                                         log_err("unexpected: invalid answer alias");
1460                                         regional_free_all(worker->scratchpad);
1461                                         return 0; /* drop query */
1462                                 }
1463                                 qinfo_tmp.qtype = qinfo.qtype;
1464                                 qinfo_tmp.qclass = qinfo.qclass;
1465                                 lookup_qinfo = &qinfo_tmp;
1466                                 goto lookup_cache;
1467                         }
1468                         verbose(VERB_ALGO, "answer from the cache failed");
1469                         lock_rw_unlock(&e->lock);
1470                 }
1471                 if(!LDNS_RD_WIRE(sldns_buffer_begin(c->buffer))) {
1472                         if(answer_norec_from_cache(worker, &qinfo,
1473                                 *(uint16_t*)(void *)sldns_buffer_begin(c->buffer), 
1474                                 sldns_buffer_read_u16_at(c->buffer, 2), repinfo, 
1475                                 &edns)) {
1476                                 regional_free_all(worker->scratchpad);
1477                                 goto send_reply;
1478                         }
1479                         verbose(VERB_ALGO, "answer norec from cache -- "
1480                                 "need to validate or not primed");
1481                 }
1482         }
1483         sldns_buffer_rewind(c->buffer);
1484         server_stats_querymiss(&worker->stats, worker);
1485
1486         if(verbosity >= VERB_CLIENT) {
1487                 if(c->type == comm_udp)
1488                         log_addr(VERB_CLIENT, "udp request from",
1489                                 &repinfo->addr, repinfo->addrlen);
1490                 else    log_addr(VERB_CLIENT, "tcp request from",
1491                                 &repinfo->addr, repinfo->addrlen);
1492         }
1493
1494         /* grab a work request structure for this new request */
1495         mesh_new_client(worker->env.mesh, &qinfo, cinfo,
1496                 sldns_buffer_read_u16_at(c->buffer, 2),
1497                 &edns, repinfo, *(uint16_t*)(void *)sldns_buffer_begin(c->buffer));
1498         regional_free_all(worker->scratchpad);
1499         worker_mem_report(worker, NULL);
1500         return 0;
1501
1502 send_reply:
1503         rc = 1;
1504 send_reply_rc:
1505         if(need_drop) {
1506                 comm_point_drop_reply(repinfo);
1507                 return 0;
1508         }
1509 #ifdef USE_DNSTAP
1510         if(worker->dtenv.log_client_response_messages)
1511                 dt_msg_send_client_response(&worker->dtenv, &repinfo->addr,
1512                         c->type, c->buffer);
1513 #endif
1514         if(worker->env.cfg->log_replies)
1515         {
1516                 struct timeval tv = {0, 0};
1517                 log_reply_info(0, &qinfo, &repinfo->addr, repinfo->addrlen,
1518                         tv, 1, c->buffer);
1519         }
1520 #ifdef USE_DNSCRYPT
1521         if(!dnsc_handle_uncurved_request(repinfo)) {
1522                 return 0;
1523         }
1524 #endif
1525         return rc;
1526 }
1527
1528 void 
1529 worker_sighandler(int sig, void* arg)
1530 {
1531         /* note that log, print, syscalls here give race conditions. 
1532          * And cause hangups if the log-lock is held by the application. */
1533         struct worker* worker = (struct worker*)arg;
1534         switch(sig) {
1535 #ifdef SIGHUP
1536                 case SIGHUP:
1537                         comm_base_exit(worker->base);
1538                         break;
1539 #endif
1540                 case SIGINT:
1541                         worker->need_to_exit = 1;
1542                         comm_base_exit(worker->base);
1543                         break;
1544 #ifdef SIGQUIT
1545                 case SIGQUIT:
1546                         worker->need_to_exit = 1;
1547                         comm_base_exit(worker->base);
1548                         break;
1549 #endif
1550                 case SIGTERM:
1551                         worker->need_to_exit = 1;
1552                         comm_base_exit(worker->base);
1553                         break;
1554                 default:
1555                         /* unknown signal, ignored */
1556                         break;
1557         }
1558 }
1559
1560 /** restart statistics timer for worker, if enabled */
1561 static void
1562 worker_restart_timer(struct worker* worker)
1563 {
1564         if(worker->env.cfg->stat_interval > 0) {
1565                 struct timeval tv;
1566 #ifndef S_SPLINT_S
1567                 tv.tv_sec = worker->env.cfg->stat_interval;
1568                 tv.tv_usec = 0;
1569 #endif
1570                 comm_timer_set(worker->stat_timer, &tv);
1571         }
1572 }
1573
1574 void worker_stat_timer_cb(void* arg)
1575 {
1576         struct worker* worker = (struct worker*)arg;
1577         server_stats_log(&worker->stats, worker, worker->thread_num);
1578         mesh_stats(worker->env.mesh, "mesh has");
1579         worker_mem_report(worker, NULL);
1580         /* SHM is enabled, process data to SHM */
1581         if (worker->daemon->cfg->shm_enable) {
1582                 shm_main_run(worker);
1583         }
1584         if(!worker->daemon->cfg->stat_cumulative) {
1585                 worker_stats_clear(worker);
1586         }
1587         /* start next timer */
1588         worker_restart_timer(worker);
1589 }
1590
1591 void worker_probe_timer_cb(void* arg)
1592 {
1593         struct worker* worker = (struct worker*)arg;
1594         struct timeval tv;
1595 #ifndef S_SPLINT_S
1596         tv.tv_sec = (time_t)autr_probe_timer(&worker->env);
1597         tv.tv_usec = 0;
1598 #endif
1599         if(tv.tv_sec != 0)
1600                 comm_timer_set(worker->env.probe_timer, &tv);
1601 }
1602
1603 struct worker* 
1604 worker_create(struct daemon* daemon, int id, int* ports, int n)
1605 {
1606         unsigned int seed;
1607         struct worker* worker = (struct worker*)calloc(1, 
1608                 sizeof(struct worker));
1609         if(!worker) 
1610                 return NULL;
1611         worker->numports = n;
1612         worker->ports = (int*)memdup(ports, sizeof(int)*n);
1613         if(!worker->ports) {
1614                 free(worker);
1615                 return NULL;
1616         }
1617         worker->daemon = daemon;
1618         worker->thread_num = id;
1619         if(!(worker->cmd = tube_create())) {
1620                 free(worker->ports);
1621                 free(worker);
1622                 return NULL;
1623         }
1624         /* create random state here to avoid locking trouble in RAND_bytes */
1625         seed = (unsigned int)time(NULL) ^ (unsigned int)getpid() ^
1626                 (((unsigned int)worker->thread_num)<<17);
1627                 /* shift thread_num so it does not match out pid bits */
1628         if(!(worker->rndstate = ub_initstate(seed, daemon->rand))) {
1629                 seed = 0;
1630                 log_err("could not init random numbers.");
1631                 tube_delete(worker->cmd);
1632                 free(worker->ports);
1633                 free(worker);
1634                 return NULL;
1635         }
1636         seed = 0;
1637 #ifdef USE_DNSTAP
1638         if(daemon->cfg->dnstap) {
1639                 log_assert(daemon->dtenv != NULL);
1640                 memcpy(&worker->dtenv, daemon->dtenv, sizeof(struct dt_env));
1641                 if(!dt_init(&worker->dtenv))
1642                         fatal_exit("dt_init failed");
1643         }
1644 #endif
1645         return worker;
1646 }
1647
1648 int
1649 worker_init(struct worker* worker, struct config_file *cfg, 
1650         struct listen_port* ports, int do_sigs)
1651 {
1652 #ifdef USE_DNSTAP
1653         struct dt_env* dtenv = &worker->dtenv;
1654 #else
1655         void* dtenv = NULL;
1656 #endif
1657         worker->need_to_exit = 0;
1658         worker->base = comm_base_create(do_sigs);
1659         if(!worker->base) {
1660                 log_err("could not create event handling base");
1661                 worker_delete(worker);
1662                 return 0;
1663         }
1664         comm_base_set_slow_accept_handlers(worker->base, &worker_stop_accept,
1665                 &worker_start_accept, worker);
1666         if(do_sigs) {
1667 #ifdef SIGHUP
1668                 ub_thread_sig_unblock(SIGHUP);
1669 #endif
1670                 ub_thread_sig_unblock(SIGINT);
1671 #ifdef SIGQUIT
1672                 ub_thread_sig_unblock(SIGQUIT);
1673 #endif
1674                 ub_thread_sig_unblock(SIGTERM);
1675 #ifndef LIBEVENT_SIGNAL_PROBLEM
1676                 worker->comsig = comm_signal_create(worker->base, 
1677                         worker_sighandler, worker);
1678                 if(!worker->comsig 
1679 #ifdef SIGHUP
1680                         || !comm_signal_bind(worker->comsig, SIGHUP)
1681 #endif
1682 #ifdef SIGQUIT
1683                         || !comm_signal_bind(worker->comsig, SIGQUIT)
1684 #endif
1685                         || !comm_signal_bind(worker->comsig, SIGTERM)
1686                         || !comm_signal_bind(worker->comsig, SIGINT)) {
1687                         log_err("could not create signal handlers");
1688                         worker_delete(worker);
1689                         return 0;
1690                 }
1691 #endif /* LIBEVENT_SIGNAL_PROBLEM */
1692                 if(!daemon_remote_open_accept(worker->daemon->rc, 
1693                         worker->daemon->rc_ports, worker)) {
1694                         worker_delete(worker);
1695                         return 0;
1696                 }
1697 #ifdef UB_ON_WINDOWS
1698                 wsvc_setup_worker(worker);
1699 #endif /* UB_ON_WINDOWS */
1700         } else { /* !do_sigs */
1701                 worker->comsig = NULL;
1702         }
1703         worker->front = listen_create(worker->base, ports,
1704                 cfg->msg_buffer_size, (int)cfg->incoming_num_tcp, 
1705                 worker->daemon->listen_sslctx, dtenv, worker_handle_request,
1706                 worker);
1707         if(!worker->front) {
1708                 log_err("could not create listening sockets");
1709                 worker_delete(worker);
1710                 return 0;
1711         }
1712         worker->back = outside_network_create(worker->base,
1713                 cfg->msg_buffer_size, (size_t)cfg->outgoing_num_ports, 
1714                 cfg->out_ifs, cfg->num_out_ifs, cfg->do_ip4, cfg->do_ip6, 
1715                 cfg->do_tcp?cfg->outgoing_num_tcp:0, 
1716                 worker->daemon->env->infra_cache, worker->rndstate,
1717                 cfg->use_caps_bits_for_id, worker->ports, worker->numports,
1718                 cfg->unwanted_threshold, cfg->outgoing_tcp_mss,
1719                 &worker_alloc_cleanup, worker,
1720                 cfg->do_udp || cfg->udp_upstream_without_downstream,
1721                 worker->daemon->connect_sslctx, cfg->delay_close,
1722                 dtenv);
1723         if(!worker->back) {
1724                 log_err("could not create outgoing sockets");
1725                 worker_delete(worker);
1726                 return 0;
1727         }
1728         /* start listening to commands */
1729         if(!tube_setup_bg_listen(worker->cmd, worker->base,
1730                 &worker_handle_control_cmd, worker)) {
1731                 log_err("could not create control compt.");
1732                 worker_delete(worker);
1733                 return 0;
1734         }
1735         worker->stat_timer = comm_timer_create(worker->base, 
1736                 worker_stat_timer_cb, worker);
1737         if(!worker->stat_timer) {
1738                 log_err("could not create statistics timer");
1739         }
1740
1741         /* we use the msg_buffer_size as a good estimate for what the 
1742          * user wants for memory usage sizes */
1743         worker->scratchpad = regional_create_custom(cfg->msg_buffer_size);
1744         if(!worker->scratchpad) {
1745                 log_err("malloc failure");
1746                 worker_delete(worker);
1747                 return 0;
1748         }
1749
1750         server_stats_init(&worker->stats, cfg);
1751         alloc_init(&worker->alloc, &worker->daemon->superalloc, 
1752                 worker->thread_num);
1753         alloc_set_id_cleanup(&worker->alloc, &worker_alloc_cleanup, worker);
1754         worker->env = *worker->daemon->env;
1755         comm_base_timept(worker->base, &worker->env.now, &worker->env.now_tv);
1756         if(worker->thread_num == 0)
1757                 log_set_time(worker->env.now);
1758         worker->env.worker = worker;
1759         worker->env.worker_base = worker->base;
1760         worker->env.send_query = &worker_send_query;
1761         worker->env.alloc = &worker->alloc;
1762         worker->env.outnet = worker->back;
1763         worker->env.rnd = worker->rndstate;
1764         /* If case prefetch is triggered, the corresponding mesh will clear
1765          * the scratchpad for the module env in the middle of request handling.
1766          * It would be prone to a use-after-free kind of bug, so we avoid
1767          * sharing it with worker's own scratchpad at the cost of having
1768          * one more pad per worker. */
1769         worker->env.scratch = regional_create_custom(cfg->msg_buffer_size);
1770         if(!worker->env.scratch) {
1771                 log_err("malloc failure");
1772                 worker_delete(worker);
1773                 return 0;
1774         }
1775         worker->env.mesh = mesh_create(&worker->daemon->mods, &worker->env);
1776         worker->env.detach_subs = &mesh_detach_subs;
1777         worker->env.attach_sub = &mesh_attach_sub;
1778         worker->env.add_sub = &mesh_add_sub;
1779         worker->env.kill_sub = &mesh_state_delete;
1780         worker->env.detect_cycle = &mesh_detect_cycle;
1781         worker->env.scratch_buffer = sldns_buffer_new(cfg->msg_buffer_size);
1782         if(!(worker->env.fwds = forwards_create()) ||
1783                 !forwards_apply_cfg(worker->env.fwds, cfg)) {
1784                 log_err("Could not set forward zones");
1785                 worker_delete(worker);
1786                 return 0;
1787         }
1788         if(!(worker->env.hints = hints_create()) ||
1789                 !hints_apply_cfg(worker->env.hints, cfg)) {
1790                 log_err("Could not set root or stub hints");
1791                 worker_delete(worker);
1792                 return 0;
1793         }
1794         /* one probe timer per process -- if we have 5011 anchors */
1795         if(autr_get_num_anchors(worker->env.anchors) > 0
1796 #ifndef THREADS_DISABLED
1797                 && worker->thread_num == 0
1798 #endif
1799                 ) {
1800                 struct timeval tv;
1801                 tv.tv_sec = 0;
1802                 tv.tv_usec = 0;
1803                 worker->env.probe_timer = comm_timer_create(worker->base,
1804                         worker_probe_timer_cb, worker);
1805                 if(!worker->env.probe_timer) {
1806                         log_err("could not create 5011-probe timer");
1807                 } else {
1808                         /* let timer fire, then it can reset itself */
1809                         comm_timer_set(worker->env.probe_timer, &tv);
1810                 }
1811         }
1812         /* zone transfer tasks, setup once per process, if any */
1813         if(worker->env.auth_zones
1814 #ifndef THREADS_DISABLED
1815                 && worker->thread_num == 0
1816 #endif
1817                 ) {
1818                 auth_xfer_pickup_initial(worker->env.auth_zones, &worker->env);
1819         }
1820         if(!worker->env.mesh || !worker->env.scratch_buffer) {
1821                 worker_delete(worker);
1822                 return 0;
1823         }
1824         worker_mem_report(worker, NULL);
1825         /* if statistics enabled start timer */
1826         if(worker->env.cfg->stat_interval > 0) {
1827                 verbose(VERB_ALGO, "set statistics interval %d secs", 
1828                         worker->env.cfg->stat_interval);
1829                 worker_restart_timer(worker);
1830         }
1831         return 1;
1832 }
1833
1834 void 
1835 worker_work(struct worker* worker)
1836 {
1837         comm_base_dispatch(worker->base);
1838 }
1839
1840 void 
1841 worker_delete(struct worker* worker)
1842 {
1843         if(!worker) 
1844                 return;
1845         if(worker->env.mesh && verbosity >= VERB_OPS) {
1846                 server_stats_log(&worker->stats, worker, worker->thread_num);
1847                 mesh_stats(worker->env.mesh, "mesh has");
1848                 worker_mem_report(worker, NULL);
1849         }
1850         outside_network_quit_prepare(worker->back);
1851         mesh_delete(worker->env.mesh);
1852         sldns_buffer_free(worker->env.scratch_buffer);
1853         forwards_delete(worker->env.fwds);
1854         hints_delete(worker->env.hints);
1855         listen_delete(worker->front);
1856         outside_network_delete(worker->back);
1857         comm_signal_delete(worker->comsig);
1858         tube_delete(worker->cmd);
1859         comm_timer_delete(worker->stat_timer);
1860         comm_timer_delete(worker->env.probe_timer);
1861         free(worker->ports);
1862         if(worker->thread_num == 0) {
1863                 log_set_time(NULL);
1864 #ifdef UB_ON_WINDOWS
1865                 wsvc_desetup_worker(worker);
1866 #endif /* UB_ON_WINDOWS */
1867         }
1868         comm_base_delete(worker->base);
1869         ub_randfree(worker->rndstate);
1870         alloc_clear(&worker->alloc);
1871         regional_destroy(worker->env.scratch);
1872         regional_destroy(worker->scratchpad);
1873         free(worker);
1874 }
1875
1876 struct outbound_entry*
1877 worker_send_query(struct query_info* qinfo, uint16_t flags, int dnssec,
1878         int want_dnssec, int nocaps, struct sockaddr_storage* addr,
1879         socklen_t addrlen, uint8_t* zone, size_t zonelen, int ssl_upstream,
1880         char* tls_auth_name, struct module_qstate* q)
1881 {
1882         struct worker* worker = q->env->worker;
1883         struct outbound_entry* e = (struct outbound_entry*)regional_alloc(
1884                 q->region, sizeof(*e));
1885         if(!e) 
1886                 return NULL;
1887         e->qstate = q;
1888         e->qsent = outnet_serviced_query(worker->back, qinfo, flags, dnssec,
1889                 want_dnssec, nocaps, q->env->cfg->tcp_upstream,
1890                 ssl_upstream, tls_auth_name, addr, addrlen, zone, zonelen, q,
1891                 worker_handle_service_reply, e, worker->back->udp_buff, q->env);
1892         if(!e->qsent) {
1893                 return NULL;
1894         }
1895         return e;
1896 }
1897
1898 void 
1899 worker_alloc_cleanup(void* arg)
1900 {
1901         struct worker* worker = (struct worker*)arg;
1902         slabhash_clear(&worker->env.rrset_cache->table);
1903         slabhash_clear(worker->env.msg_cache);
1904 }
1905
1906 void worker_stats_clear(struct worker* worker)
1907 {
1908         server_stats_init(&worker->stats, worker->env.cfg);
1909         mesh_stats_clear(worker->env.mesh);
1910         worker->back->unwanted_replies = 0;
1911         worker->back->num_tcp_outgoing = 0;
1912 }
1913
1914 void worker_start_accept(void* arg)
1915 {
1916         struct worker* worker = (struct worker*)arg;
1917         listen_start_accept(worker->front);
1918         if(worker->thread_num == 0)
1919                 daemon_remote_start_accept(worker->daemon->rc);
1920 }
1921
1922 void worker_stop_accept(void* arg)
1923 {
1924         struct worker* worker = (struct worker*)arg;
1925         listen_stop_accept(worker->front);
1926         if(worker->thread_num == 0)
1927                 daemon_remote_stop_accept(worker->daemon->rc);
1928 }
1929
1930 /* --- fake callbacks for fptr_wlist to work --- */
1931 struct outbound_entry* libworker_send_query(
1932         struct query_info* ATTR_UNUSED(qinfo),
1933         uint16_t ATTR_UNUSED(flags), int ATTR_UNUSED(dnssec),
1934         int ATTR_UNUSED(want_dnssec), int ATTR_UNUSED(nocaps),
1935         struct sockaddr_storage* ATTR_UNUSED(addr), socklen_t ATTR_UNUSED(addrlen),
1936         uint8_t* ATTR_UNUSED(zone), size_t ATTR_UNUSED(zonelen),
1937         int ATTR_UNUSED(ssl_upstream), char* ATTR_UNUSED(tls_auth_name),
1938         struct module_qstate* ATTR_UNUSED(q))
1939 {
1940         log_assert(0);
1941         return 0;
1942 }
1943
1944 int libworker_handle_reply(struct comm_point* ATTR_UNUSED(c), 
1945         void* ATTR_UNUSED(arg), int ATTR_UNUSED(error),
1946         struct comm_reply* ATTR_UNUSED(reply_info))
1947 {
1948         log_assert(0);
1949         return 0;
1950 }
1951
1952 int libworker_handle_service_reply(struct comm_point* ATTR_UNUSED(c), 
1953         void* ATTR_UNUSED(arg), int ATTR_UNUSED(error),
1954         struct comm_reply* ATTR_UNUSED(reply_info))
1955 {
1956         log_assert(0);
1957         return 0;
1958 }
1959
1960 void libworker_handle_control_cmd(struct tube* ATTR_UNUSED(tube),
1961         uint8_t* ATTR_UNUSED(buffer), size_t ATTR_UNUSED(len),
1962         int ATTR_UNUSED(error), void* ATTR_UNUSED(arg))
1963 {
1964         log_assert(0);
1965 }
1966
1967 void libworker_fg_done_cb(void* ATTR_UNUSED(arg), int ATTR_UNUSED(rcode),
1968         sldns_buffer* ATTR_UNUSED(buf), enum sec_status ATTR_UNUSED(s),
1969         char* ATTR_UNUSED(why_bogus))
1970 {
1971         log_assert(0);
1972 }
1973
1974 void libworker_bg_done_cb(void* ATTR_UNUSED(arg), int ATTR_UNUSED(rcode),
1975         sldns_buffer* ATTR_UNUSED(buf), enum sec_status ATTR_UNUSED(s),
1976         char* ATTR_UNUSED(why_bogus))
1977 {
1978         log_assert(0);
1979 }
1980
1981 void libworker_event_done_cb(void* ATTR_UNUSED(arg), int ATTR_UNUSED(rcode),
1982         sldns_buffer* ATTR_UNUSED(buf), enum sec_status ATTR_UNUSED(s),
1983         char* ATTR_UNUSED(why_bogus))
1984 {
1985         log_assert(0);
1986 }
1987
1988 int context_query_cmp(const void* ATTR_UNUSED(a), const void* ATTR_UNUSED(b))
1989 {
1990         log_assert(0);
1991         return 0;
1992 }
1993
1994 int order_lock_cmp(const void* ATTR_UNUSED(e1), const void* ATTR_UNUSED(e2))
1995 {
1996         log_assert(0);
1997         return 0;
1998 }
1999
2000 int codeline_cmp(const void* ATTR_UNUSED(a), const void* ATTR_UNUSED(b))
2001 {
2002         log_assert(0);
2003         return 0;
2004 }
2005