1 /*
2 * iterator/iterator.c - iterative resolver DNS query response module
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 contains a module that performs recursive iterative DNS query
40 * processing.
41 */
42
43 #include "config.h"
44 #include "iterator/iterator.h"
45 #include "iterator/iter_utils.h"
46 #include "iterator/iter_hints.h"
47 #include "iterator/iter_fwd.h"
48 #include "iterator/iter_donotq.h"
49 #include "iterator/iter_delegpt.h"
50 #include "iterator/iter_resptype.h"
51 #include "iterator/iter_scrub.h"
52 #include "iterator/iter_priv.h"
53 #include "validator/val_neg.h"
54 #include "services/cache/dns.h"
55 #include "services/cache/infra.h"
56 #include "services/authzone.h"
57 #include "util/module.h"
58 #include "util/netevent.h"
59 #include "util/net_help.h"
60 #include "util/regional.h"
61 #include "util/data/dname.h"
62 #include "util/data/msgencode.h"
63 #include "util/fptr_wlist.h"
64 #include "util/config_file.h"
65 #include "util/random.h"
66 #include "sldns/rrdef.h"
67 #include "sldns/wire2str.h"
68 #include "sldns/str2wire.h"
69 #include "sldns/parseutil.h"
70 #include "sldns/sbuffer.h"
71
72 /* in msec */
73 int UNKNOWN_SERVER_NICENESS = 376;
74
75 static void target_count_increase_nx(struct iter_qstate* iq, int num);
76
77 int
iter_init(struct module_env * env,int id)78 iter_init(struct module_env* env, int id)
79 {
80 struct iter_env* iter_env = (struct iter_env*)calloc(1,
81 sizeof(struct iter_env));
82 if(!iter_env) {
83 log_err("malloc failure");
84 return 0;
85 }
86 env->modinfo[id] = (void*)iter_env;
87
88 lock_basic_init(&iter_env->queries_ratelimit_lock);
89 lock_protect(&iter_env->queries_ratelimit_lock,
90 &iter_env->num_queries_ratelimited,
91 sizeof(iter_env->num_queries_ratelimited));
92
93 if(!iter_apply_cfg(iter_env, env->cfg)) {
94 log_err("iterator: could not apply configuration settings.");
95 return 0;
96 }
97
98 return 1;
99 }
100
101 /** delete caps_whitelist element */
102 static void
caps_free(struct rbnode_type * n,void * ATTR_UNUSED (d))103 caps_free(struct rbnode_type* n, void* ATTR_UNUSED(d))
104 {
105 if(n) {
106 free(((struct name_tree_node*)n)->name);
107 free(n);
108 }
109 }
110
111 void
iter_deinit(struct module_env * env,int id)112 iter_deinit(struct module_env* env, int id)
113 {
114 struct iter_env* iter_env;
115 if(!env || !env->modinfo[id])
116 return;
117 iter_env = (struct iter_env*)env->modinfo[id];
118 lock_basic_destroy(&iter_env->queries_ratelimit_lock);
119 free(iter_env->target_fetch_policy);
120 priv_delete(iter_env->priv);
121 donotq_delete(iter_env->donotq);
122 if(iter_env->caps_white) {
123 traverse_postorder(iter_env->caps_white, caps_free, NULL);
124 free(iter_env->caps_white);
125 }
126 free(iter_env);
127 env->modinfo[id] = NULL;
128 }
129
130 /** new query for iterator */
131 static int
iter_new(struct module_qstate * qstate,int id)132 iter_new(struct module_qstate* qstate, int id)
133 {
134 struct iter_qstate* iq = (struct iter_qstate*)regional_alloc(
135 qstate->region, sizeof(struct iter_qstate));
136 qstate->minfo[id] = iq;
137 if(!iq)
138 return 0;
139 memset(iq, 0, sizeof(*iq));
140 iq->state = INIT_REQUEST_STATE;
141 iq->final_state = FINISHED_STATE;
142 iq->an_prepend_list = NULL;
143 iq->an_prepend_last = NULL;
144 iq->ns_prepend_list = NULL;
145 iq->ns_prepend_last = NULL;
146 iq->dp = NULL;
147 iq->depth = 0;
148 iq->num_target_queries = 0;
149 iq->num_current_queries = 0;
150 iq->query_restart_count = 0;
151 iq->referral_count = 0;
152 iq->sent_count = 0;
153 iq->ratelimit_ok = 0;
154 iq->target_count = NULL;
155 iq->dp_target_count = 0;
156 iq->wait_priming_stub = 0;
157 iq->refetch_glue = 0;
158 iq->dnssec_expected = 0;
159 iq->dnssec_lame_query = 0;
160 iq->chase_flags = qstate->query_flags;
161 /* Start with the (current) qname. */
162 iq->qchase = qstate->qinfo;
163 outbound_list_init(&iq->outlist);
164 iq->minimise_count = 0;
165 iq->timeout_count = 0;
166 if (qstate->env->cfg->qname_minimisation)
167 iq->minimisation_state = INIT_MINIMISE_STATE;
168 else
169 iq->minimisation_state = DONOT_MINIMISE_STATE;
170
171 memset(&iq->qinfo_out, 0, sizeof(struct query_info));
172 return 1;
173 }
174
175 /**
176 * Transition to the next state. This can be used to advance a currently
177 * processing event. It cannot be used to reactivate a forEvent.
178 *
179 * @param iq: iterator query state
180 * @param nextstate The state to transition to.
181 * @return true. This is so this can be called as the return value for the
182 * actual process*State() methods. (Transitioning to the next state
183 * implies further processing).
184 */
185 static int
next_state(struct iter_qstate * iq,enum iter_state nextstate)186 next_state(struct iter_qstate* iq, enum iter_state nextstate)
187 {
188 /* If transitioning to a "response" state, make sure that there is a
189 * response */
190 if(iter_state_is_responsestate(nextstate)) {
191 if(iq->response == NULL) {
192 log_err("transitioning to response state sans "
193 "response.");
194 }
195 }
196 iq->state = nextstate;
197 return 1;
198 }
199
200 /**
201 * Transition an event to its final state. Final states always either return
202 * a result up the module chain, or reactivate a dependent event. Which
203 * final state to transition to is set in the module state for the event when
204 * it was created, and depends on the original purpose of the event.
205 *
206 * The response is stored in the qstate->buf buffer.
207 *
208 * @param iq: iterator query state
209 * @return false. This is so this method can be used as the return value for
210 * the processState methods. (Transitioning to the final state
211 */
212 static int
final_state(struct iter_qstate * iq)213 final_state(struct iter_qstate* iq)
214 {
215 return next_state(iq, iq->final_state);
216 }
217
218 /**
219 * Callback routine to handle errors in parent query states
220 * @param qstate: query state that failed.
221 * @param id: module id.
222 * @param super: super state.
223 */
224 static void
error_supers(struct module_qstate * qstate,int id,struct module_qstate * super)225 error_supers(struct module_qstate* qstate, int id, struct module_qstate* super)
226 {
227 struct iter_env* ie = (struct iter_env*)qstate->env->modinfo[id];
228 struct iter_qstate* super_iq = (struct iter_qstate*)super->minfo[id];
229
230 if(qstate->qinfo.qtype == LDNS_RR_TYPE_A ||
231 qstate->qinfo.qtype == LDNS_RR_TYPE_AAAA) {
232 /* mark address as failed. */
233 struct delegpt_ns* dpns = NULL;
234 super_iq->num_target_queries--;
235 if(super_iq->dp)
236 dpns = delegpt_find_ns(super_iq->dp,
237 qstate->qinfo.qname, qstate->qinfo.qname_len);
238 if(!dpns) {
239 /* not interested */
240 /* this can happen, for eg. qname minimisation asked
241 * for an NXDOMAIN to be validated, and used qtype
242 * A for that, and the error of that, the name, is
243 * not listed in super_iq->dp */
244 verbose(VERB_ALGO, "subq error, but not interested");
245 log_query_info(VERB_ALGO, "superq", &super->qinfo);
246 return;
247 } else {
248 /* see if the failure did get (parent-lame) info */
249 if(!cache_fill_missing(super->env, super_iq->qchase.qclass,
250 super->region, super_iq->dp))
251 log_err("out of memory adding missing");
252 }
253 delegpt_mark_neg(dpns, qstate->qinfo.qtype);
254 dpns->resolved = 1; /* mark as failed */
255 if((dpns->got4 == 2 || !ie->supports_ipv4) &&
256 (dpns->got6 == 2 || !ie->supports_ipv6))
257 target_count_increase_nx(super_iq, 1);
258 }
259 if(qstate->qinfo.qtype == LDNS_RR_TYPE_NS) {
260 /* prime failed to get delegation */
261 super_iq->dp = NULL;
262 }
263 /* evaluate targets again */
264 super_iq->state = QUERYTARGETS_STATE;
265 /* super becomes runnable, and will process this change */
266 }
267
268 /**
269 * Return an error to the client
270 * @param qstate: our query state
271 * @param id: module id
272 * @param rcode: error code (DNS errcode).
273 * @return: 0 for use by caller, to make notation easy, like:
274 * return error_response(..).
275 */
276 static int
error_response(struct module_qstate * qstate,int id,int rcode)277 error_response(struct module_qstate* qstate, int id, int rcode)
278 {
279 verbose(VERB_QUERY, "return error response %s",
280 sldns_lookup_by_id(sldns_rcodes, rcode)?
281 sldns_lookup_by_id(sldns_rcodes, rcode)->name:"??");
282 qstate->return_rcode = rcode;
283 qstate->return_msg = NULL;
284 qstate->ext_state[id] = module_finished;
285 return 0;
286 }
287
288 /**
289 * Return an error to the client and cache the error code in the
290 * message cache (so per qname, qtype, qclass).
291 * @param qstate: our query state
292 * @param id: module id
293 * @param rcode: error code (DNS errcode).
294 * @return: 0 for use by caller, to make notation easy, like:
295 * return error_response(..).
296 */
297 static int
error_response_cache(struct module_qstate * qstate,int id,int rcode)298 error_response_cache(struct module_qstate* qstate, int id, int rcode)
299 {
300 if(!qstate->no_cache_store) {
301 /* store in cache */
302 struct reply_info err;
303 if(qstate->prefetch_leeway > NORR_TTL) {
304 verbose(VERB_ALGO, "error response for prefetch in cache");
305 /* attempt to adjust the cache entry prefetch */
306 if(dns_cache_prefetch_adjust(qstate->env, &qstate->qinfo,
307 NORR_TTL, qstate->query_flags))
308 return error_response(qstate, id, rcode);
309 /* if that fails (not in cache), fall through to store err */
310 }
311 if(qstate->env->cfg->serve_expired) {
312 /* if serving expired contents, and such content is
313 * already available, don't overwrite this servfail */
314 struct msgreply_entry* msg;
315 if((msg=msg_cache_lookup(qstate->env,
316 qstate->qinfo.qname, qstate->qinfo.qname_len,
317 qstate->qinfo.qtype, qstate->qinfo.qclass,
318 qstate->query_flags, 0,
319 qstate->env->cfg->serve_expired_ttl_reset))
320 != NULL) {
321 if(qstate->env->cfg->serve_expired_ttl_reset) {
322 struct reply_info* rep =
323 (struct reply_info*)msg->entry.data;
324 if(rep && *qstate->env->now +
325 qstate->env->cfg->serve_expired_ttl >
326 rep->serve_expired_ttl) {
327 rep->serve_expired_ttl =
328 *qstate->env->now +
329 qstate->env->cfg->serve_expired_ttl;
330 }
331 }
332 lock_rw_unlock(&msg->entry.lock);
333 return error_response(qstate, id, rcode);
334 }
335 /* serving expired contents, but nothing is cached
336 * at all, so the servfail cache entry is useful
337 * (stops waste of time on this servfail NORR_TTL) */
338 } else {
339 /* don't overwrite existing (non-expired) data in
340 * cache with a servfail */
341 struct msgreply_entry* msg;
342 if((msg=msg_cache_lookup(qstate->env,
343 qstate->qinfo.qname, qstate->qinfo.qname_len,
344 qstate->qinfo.qtype, qstate->qinfo.qclass,
345 qstate->query_flags, *qstate->env->now, 0))
346 != NULL) {
347 struct reply_info* rep = (struct reply_info*)
348 msg->entry.data;
349 if(FLAGS_GET_RCODE(rep->flags) ==
350 LDNS_RCODE_NOERROR ||
351 FLAGS_GET_RCODE(rep->flags) ==
352 LDNS_RCODE_NXDOMAIN) {
353 /* we have a good entry,
354 * don't overwrite */
355 lock_rw_unlock(&msg->entry.lock);
356 return error_response(qstate, id, rcode);
357 }
358 lock_rw_unlock(&msg->entry.lock);
359 }
360
361 }
362 memset(&err, 0, sizeof(err));
363 err.flags = (uint16_t)(BIT_QR | BIT_RA);
364 FLAGS_SET_RCODE(err.flags, rcode);
365 err.qdcount = 1;
366 err.ttl = NORR_TTL;
367 err.prefetch_ttl = PREFETCH_TTL_CALC(err.ttl);
368 err.serve_expired_ttl = NORR_TTL;
369 /* do not waste time trying to validate this servfail */
370 err.security = sec_status_indeterminate;
371 verbose(VERB_ALGO, "store error response in message cache");
372 iter_dns_store(qstate->env, &qstate->qinfo, &err, 0, 0, 0, NULL,
373 qstate->query_flags);
374 }
375 return error_response(qstate, id, rcode);
376 }
377
378 /** check if prepend item is duplicate item */
379 static int
prepend_is_duplicate(struct ub_packed_rrset_key ** sets,size_t to,struct ub_packed_rrset_key * dup)380 prepend_is_duplicate(struct ub_packed_rrset_key** sets, size_t to,
381 struct ub_packed_rrset_key* dup)
382 {
383 size_t i;
384 for(i=0; i<to; i++) {
385 if(sets[i]->rk.type == dup->rk.type &&
386 sets[i]->rk.rrset_class == dup->rk.rrset_class &&
387 sets[i]->rk.dname_len == dup->rk.dname_len &&
388 query_dname_compare(sets[i]->rk.dname, dup->rk.dname)
389 == 0)
390 return 1;
391 }
392 return 0;
393 }
394
395 /** prepend the prepend list in the answer and authority section of dns_msg */
396 static int
iter_prepend(struct iter_qstate * iq,struct dns_msg * msg,struct regional * region)397 iter_prepend(struct iter_qstate* iq, struct dns_msg* msg,
398 struct regional* region)
399 {
400 struct iter_prep_list* p;
401 struct ub_packed_rrset_key** sets;
402 size_t num_an = 0, num_ns = 0;;
403 for(p = iq->an_prepend_list; p; p = p->next)
404 num_an++;
405 for(p = iq->ns_prepend_list; p; p = p->next)
406 num_ns++;
407 if(num_an + num_ns == 0)
408 return 1;
409 verbose(VERB_ALGO, "prepending %d rrsets", (int)num_an + (int)num_ns);
410 if(num_an > RR_COUNT_MAX || num_ns > RR_COUNT_MAX ||
411 msg->rep->rrset_count > RR_COUNT_MAX) return 0; /* overflow */
412 sets = regional_alloc(region, (num_an+num_ns+msg->rep->rrset_count) *
413 sizeof(struct ub_packed_rrset_key*));
414 if(!sets)
415 return 0;
416 /* ANSWER section */
417 num_an = 0;
418 for(p = iq->an_prepend_list; p; p = p->next) {
419 sets[num_an++] = p->rrset;
420 if(ub_packed_rrset_ttl(p->rrset) < msg->rep->ttl)
421 msg->rep->ttl = ub_packed_rrset_ttl(p->rrset);
422 }
423 memcpy(sets+num_an, msg->rep->rrsets, msg->rep->an_numrrsets *
424 sizeof(struct ub_packed_rrset_key*));
425 /* AUTH section */
426 num_ns = 0;
427 for(p = iq->ns_prepend_list; p; p = p->next) {
428 if(prepend_is_duplicate(sets+msg->rep->an_numrrsets+num_an,
429 num_ns, p->rrset) || prepend_is_duplicate(
430 msg->rep->rrsets+msg->rep->an_numrrsets,
431 msg->rep->ns_numrrsets, p->rrset))
432 continue;
433 sets[msg->rep->an_numrrsets + num_an + num_ns++] = p->rrset;
434 if(ub_packed_rrset_ttl(p->rrset) < msg->rep->ttl)
435 msg->rep->ttl = ub_packed_rrset_ttl(p->rrset);
436 }
437 memcpy(sets + num_an + msg->rep->an_numrrsets + num_ns,
438 msg->rep->rrsets + msg->rep->an_numrrsets,
439 (msg->rep->ns_numrrsets + msg->rep->ar_numrrsets) *
440 sizeof(struct ub_packed_rrset_key*));
441
442 /* NXDOMAIN rcode can stay if we prepended DNAME/CNAMEs, because
443 * this is what recursors should give. */
444 msg->rep->rrset_count += num_an + num_ns;
445 msg->rep->an_numrrsets += num_an;
446 msg->rep->ns_numrrsets += num_ns;
447 msg->rep->rrsets = sets;
448 return 1;
449 }
450
451 /**
452 * Find rrset in ANSWER prepend list.
453 * to avoid duplicate DNAMEs when a DNAME is traversed twice.
454 * @param iq: iterator query state.
455 * @param rrset: rrset to add.
456 * @return false if not found
457 */
458 static int
iter_find_rrset_in_prepend_answer(struct iter_qstate * iq,struct ub_packed_rrset_key * rrset)459 iter_find_rrset_in_prepend_answer(struct iter_qstate* iq,
460 struct ub_packed_rrset_key* rrset)
461 {
462 struct iter_prep_list* p = iq->an_prepend_list;
463 while(p) {
464 if(ub_rrset_compare(p->rrset, rrset) == 0 &&
465 rrsetdata_equal((struct packed_rrset_data*)p->rrset
466 ->entry.data, (struct packed_rrset_data*)rrset
467 ->entry.data))
468 return 1;
469 p = p->next;
470 }
471 return 0;
472 }
473
474 /**
475 * Add rrset to ANSWER prepend list
476 * @param qstate: query state.
477 * @param iq: iterator query state.
478 * @param rrset: rrset to add.
479 * @return false on failure (malloc).
480 */
481 static int
iter_add_prepend_answer(struct module_qstate * qstate,struct iter_qstate * iq,struct ub_packed_rrset_key * rrset)482 iter_add_prepend_answer(struct module_qstate* qstate, struct iter_qstate* iq,
483 struct ub_packed_rrset_key* rrset)
484 {
485 struct iter_prep_list* p = (struct iter_prep_list*)regional_alloc(
486 qstate->region, sizeof(struct iter_prep_list));
487 if(!p)
488 return 0;
489 p->rrset = rrset;
490 p->next = NULL;
491 /* add at end */
492 if(iq->an_prepend_last)
493 iq->an_prepend_last->next = p;
494 else iq->an_prepend_list = p;
495 iq->an_prepend_last = p;
496 return 1;
497 }
498
499 /**
500 * Add rrset to AUTHORITY prepend list
501 * @param qstate: query state.
502 * @param iq: iterator query state.
503 * @param rrset: rrset to add.
504 * @return false on failure (malloc).
505 */
506 static int
iter_add_prepend_auth(struct module_qstate * qstate,struct iter_qstate * iq,struct ub_packed_rrset_key * rrset)507 iter_add_prepend_auth(struct module_qstate* qstate, struct iter_qstate* iq,
508 struct ub_packed_rrset_key* rrset)
509 {
510 struct iter_prep_list* p = (struct iter_prep_list*)regional_alloc(
511 qstate->region, sizeof(struct iter_prep_list));
512 if(!p)
513 return 0;
514 p->rrset = rrset;
515 p->next = NULL;
516 /* add at end */
517 if(iq->ns_prepend_last)
518 iq->ns_prepend_last->next = p;
519 else iq->ns_prepend_list = p;
520 iq->ns_prepend_last = p;
521 return 1;
522 }
523
524 /**
525 * Given a CNAME response (defined as a response containing a CNAME or DNAME
526 * that does not answer the request), process the response, modifying the
527 * state as necessary. This follows the CNAME/DNAME chain and returns the
528 * final query name.
529 *
530 * sets the new query name, after following the CNAME/DNAME chain.
531 * @param qstate: query state.
532 * @param iq: iterator query state.
533 * @param msg: the response.
534 * @param mname: returned target new query name.
535 * @param mname_len: length of mname.
536 * @return false on (malloc) error.
537 */
538 static int
handle_cname_response(struct module_qstate * qstate,struct iter_qstate * iq,struct dns_msg * msg,uint8_t ** mname,size_t * mname_len)539 handle_cname_response(struct module_qstate* qstate, struct iter_qstate* iq,
540 struct dns_msg* msg, uint8_t** mname, size_t* mname_len)
541 {
542 size_t i;
543 /* Start with the (current) qname. */
544 *mname = iq->qchase.qname;
545 *mname_len = iq->qchase.qname_len;
546
547 /* Iterate over the ANSWER rrsets in order, looking for CNAMEs and
548 * DNAMES. */
549 for(i=0; i<msg->rep->an_numrrsets; i++) {
550 struct ub_packed_rrset_key* r = msg->rep->rrsets[i];
551 /* If there is a (relevant) DNAME, add it to the list.
552 * We always expect there to be CNAME that was generated
553 * by this DNAME following, so we don't process the DNAME
554 * directly. */
555 if(ntohs(r->rk.type) == LDNS_RR_TYPE_DNAME &&
556 dname_strict_subdomain_c(*mname, r->rk.dname) &&
557 !iter_find_rrset_in_prepend_answer(iq, r)) {
558 if(!iter_add_prepend_answer(qstate, iq, r))
559 return 0;
560 continue;
561 }
562
563 if(ntohs(r->rk.type) == LDNS_RR_TYPE_CNAME &&
564 query_dname_compare(*mname, r->rk.dname) == 0 &&
565 !iter_find_rrset_in_prepend_answer(iq, r)) {
566 /* Add this relevant CNAME rrset to the prepend list.*/
567 if(!iter_add_prepend_answer(qstate, iq, r))
568 return 0;
569 get_cname_target(r, mname, mname_len);
570 }
571
572 /* Other rrsets in the section are ignored. */
573 }
574 /* add authority rrsets to authority prepend, for wildcarded CNAMEs */
575 for(i=msg->rep->an_numrrsets; i<msg->rep->an_numrrsets +
576 msg->rep->ns_numrrsets; i++) {
577 struct ub_packed_rrset_key* r = msg->rep->rrsets[i];
578 /* only add NSEC/NSEC3, as they may be needed for validation */
579 if(ntohs(r->rk.type) == LDNS_RR_TYPE_NSEC ||
580 ntohs(r->rk.type) == LDNS_RR_TYPE_NSEC3) {
581 if(!iter_add_prepend_auth(qstate, iq, r))
582 return 0;
583 }
584 }
585 return 1;
586 }
587
588 /** add response specific error information for log servfail */
589 static void
errinf_reply(struct module_qstate * qstate,struct iter_qstate * iq)590 errinf_reply(struct module_qstate* qstate, struct iter_qstate* iq)
591 {
592 if(qstate->env->cfg->val_log_level < 2 && !qstate->env->cfg->log_servfail)
593 return;
594 if((qstate->reply && qstate->reply->addrlen != 0) ||
595 (iq->fail_reply && iq->fail_reply->addrlen != 0)) {
596 char from[256], frm[512];
597 if(qstate->reply && qstate->reply->addrlen != 0)
598 addr_to_str(&qstate->reply->addr, qstate->reply->addrlen,
599 from, sizeof(from));
600 else
601 addr_to_str(&iq->fail_reply->addr, iq->fail_reply->addrlen,
602 from, sizeof(from));
603 snprintf(frm, sizeof(frm), "from %s", from);
604 errinf(qstate, frm);
605 }
606 if(iq->scrub_failures || iq->parse_failures) {
607 if(iq->scrub_failures)
608 errinf(qstate, "upstream response failed scrub");
609 if(iq->parse_failures)
610 errinf(qstate, "could not parse upstream response");
611 } else if(iq->response == NULL && iq->timeout_count != 0) {
612 errinf(qstate, "upstream server timeout");
613 } else if(iq->response == NULL) {
614 errinf(qstate, "no server to query");
615 if(iq->dp) {
616 if(iq->dp->target_list == NULL)
617 errinf(qstate, "no addresses for nameservers");
618 else errinf(qstate, "nameserver addresses not usable");
619 if(iq->dp->nslist == NULL)
620 errinf(qstate, "have no nameserver names");
621 if(iq->dp->bogus)
622 errinf(qstate, "NS record was dnssec bogus");
623 }
624 }
625 if(iq->response && iq->response->rep) {
626 if(FLAGS_GET_RCODE(iq->response->rep->flags) != 0) {
627 char rcode[256], rc[32];
628 (void)sldns_wire2str_rcode_buf(
629 FLAGS_GET_RCODE(iq->response->rep->flags),
630 rc, sizeof(rc));
631 snprintf(rcode, sizeof(rcode), "got %s", rc);
632 errinf(qstate, rcode);
633 } else {
634 /* rcode NOERROR */
635 if(iq->response->rep->an_numrrsets == 0) {
636 errinf(qstate, "nodata answer");
637 }
638 }
639 }
640 }
641
642 /** see if last resort is possible - does config allow queries to parent */
643 static int
can_have_last_resort(struct module_env * env,uint8_t * nm,size_t nmlen,uint16_t qclass,struct delegpt ** retdp)644 can_have_last_resort(struct module_env* env, uint8_t* nm, size_t nmlen,
645 uint16_t qclass, struct delegpt** retdp)
646 {
647 struct delegpt* fwddp;
648 struct iter_hints_stub* stub;
649 int labs = dname_count_labels(nm);
650 /* do not process a last resort (the parent side) if a stub
651 * or forward is configured, because we do not want to go 'above'
652 * the configured servers */
653 if(!dname_is_root(nm) && (stub = (struct iter_hints_stub*)
654 name_tree_find(&env->hints->tree, nm, nmlen, labs, qclass)) &&
655 /* has_parent side is turned off for stub_first, where we
656 * are allowed to go to the parent */
657 stub->dp->has_parent_side_NS) {
658 if(retdp) *retdp = stub->dp;
659 return 0;
660 }
661 if((fwddp = forwards_find(env->fwds, nm, qclass)) &&
662 /* has_parent_side is turned off for forward_first, where
663 * we are allowed to go to the parent */
664 fwddp->has_parent_side_NS) {
665 if(retdp) *retdp = fwddp;
666 return 0;
667 }
668 return 1;
669 }
670
671 /** see if target name is caps-for-id whitelisted */
672 static int
is_caps_whitelisted(struct iter_env * ie,struct iter_qstate * iq)673 is_caps_whitelisted(struct iter_env* ie, struct iter_qstate* iq)
674 {
675 if(!ie->caps_white) return 0; /* no whitelist, or no capsforid */
676 return name_tree_lookup(ie->caps_white, iq->qchase.qname,
677 iq->qchase.qname_len, dname_count_labels(iq->qchase.qname),
678 iq->qchase.qclass) != NULL;
679 }
680
681 /** create target count structure for this query */
682 static void
target_count_create(struct iter_qstate * iq)683 target_count_create(struct iter_qstate* iq)
684 {
685 if(!iq->target_count) {
686 iq->target_count = (int*)calloc(3, sizeof(int));
687 /* if calloc fails we simply do not track this number */
688 if(iq->target_count)
689 iq->target_count[0] = 1;
690 }
691 }
692
693 static void
target_count_increase(struct iter_qstate * iq,int num)694 target_count_increase(struct iter_qstate* iq, int num)
695 {
696 target_count_create(iq);
697 if(iq->target_count)
698 iq->target_count[1] += num;
699 iq->dp_target_count++;
700 }
701
702 static void
target_count_increase_nx(struct iter_qstate * iq,int num)703 target_count_increase_nx(struct iter_qstate* iq, int num)
704 {
705 target_count_create(iq);
706 if(iq->target_count)
707 iq->target_count[2] += num;
708 }
709
710 /**
711 * Generate a subrequest.
712 * Generate a local request event. Local events are tied to this module, and
713 * have a corresponding (first tier) event that is waiting for this event to
714 * resolve to continue.
715 *
716 * @param qname The query name for this request.
717 * @param qnamelen length of qname
718 * @param qtype The query type for this request.
719 * @param qclass The query class for this request.
720 * @param qstate The event that is generating this event.
721 * @param id: module id.
722 * @param iq: The iterator state that is generating this event.
723 * @param initial_state The initial response state (normally this
724 * is QUERY_RESP_STATE, unless it is known that the request won't
725 * need iterative processing
726 * @param finalstate The final state for the response to this request.
727 * @param subq_ret: if newly allocated, the subquerystate, or NULL if it does
728 * not need initialisation.
729 * @param v: if true, validation is done on the subquery.
730 * @param detached: true if this qstate should not attach to the subquery
731 * @return false on error (malloc).
732 */
733 static int
generate_sub_request(uint8_t * qname,size_t qnamelen,uint16_t qtype,uint16_t qclass,struct module_qstate * qstate,int id,struct iter_qstate * iq,enum iter_state initial_state,enum iter_state finalstate,struct module_qstate ** subq_ret,int v,int detached)734 generate_sub_request(uint8_t* qname, size_t qnamelen, uint16_t qtype,
735 uint16_t qclass, struct module_qstate* qstate, int id,
736 struct iter_qstate* iq, enum iter_state initial_state,
737 enum iter_state finalstate, struct module_qstate** subq_ret, int v,
738 int detached)
739 {
740 struct module_qstate* subq = NULL;
741 struct iter_qstate* subiq = NULL;
742 uint16_t qflags = 0; /* OPCODE QUERY, no flags */
743 struct query_info qinf;
744 int prime = (finalstate == PRIME_RESP_STATE)?1:0;
745 int valrec = 0;
746 qinf.qname = qname;
747 qinf.qname_len = qnamelen;
748 qinf.qtype = qtype;
749 qinf.qclass = qclass;
750 qinf.local_alias = NULL;
751
752 /* RD should be set only when sending the query back through the INIT
753 * state. */
754 if(initial_state == INIT_REQUEST_STATE)
755 qflags |= BIT_RD;
756 /* We set the CD flag so we can send this through the "head" of
757 * the resolution chain, which might have a validator. We are
758 * uninterested in validating things not on the direct resolution
759 * path. */
760 if(!v) {
761 qflags |= BIT_CD;
762 valrec = 1;
763 }
764
765 if(detached) {
766 struct mesh_state* sub = NULL;
767 fptr_ok(fptr_whitelist_modenv_add_sub(
768 qstate->env->add_sub));
769 if(!(*qstate->env->add_sub)(qstate, &qinf,
770 qflags, prime, valrec, &subq, &sub)){
771 return 0;
772 }
773 }
774 else {
775 /* attach subquery, lookup existing or make a new one */
776 fptr_ok(fptr_whitelist_modenv_attach_sub(
777 qstate->env->attach_sub));
778 if(!(*qstate->env->attach_sub)(qstate, &qinf, qflags, prime,
779 valrec, &subq)) {
780 return 0;
781 }
782 }
783 *subq_ret = subq;
784 if(subq) {
785 /* initialise the new subquery */
786 subq->curmod = id;
787 subq->ext_state[id] = module_state_initial;
788 subq->minfo[id] = regional_alloc(subq->region,
789 sizeof(struct iter_qstate));
790 if(!subq->minfo[id]) {
791 log_err("init subq: out of memory");
792 fptr_ok(fptr_whitelist_modenv_kill_sub(
793 qstate->env->kill_sub));
794 (*qstate->env->kill_sub)(subq);
795 return 0;
796 }
797 subiq = (struct iter_qstate*)subq->minfo[id];
798 memset(subiq, 0, sizeof(*subiq));
799 subiq->num_target_queries = 0;
800 target_count_create(iq);
801 subiq->target_count = iq->target_count;
802 if(iq->target_count)
803 iq->target_count[0] ++; /* extra reference */
804 subiq->dp_target_count = 0;
805 subiq->num_current_queries = 0;
806 subiq->depth = iq->depth+1;
807 outbound_list_init(&subiq->outlist);
808 subiq->state = initial_state;
809 subiq->final_state = finalstate;
810 subiq->qchase = subq->qinfo;
811 subiq->chase_flags = subq->query_flags;
812 subiq->refetch_glue = 0;
813 if(qstate->env->cfg->qname_minimisation)
814 subiq->minimisation_state = INIT_MINIMISE_STATE;
815 else
816 subiq->minimisation_state = DONOT_MINIMISE_STATE;
817 memset(&subiq->qinfo_out, 0, sizeof(struct query_info));
818 }
819 return 1;
820 }
821
822 /**
823 * Generate and send a root priming request.
824 * @param qstate: the qtstate that triggered the need to prime.
825 * @param iq: iterator query state.
826 * @param id: module id.
827 * @param qclass: the class to prime.
828 * @return 0 on failure
829 */
830 static int
prime_root(struct module_qstate * qstate,struct iter_qstate * iq,int id,uint16_t qclass)831 prime_root(struct module_qstate* qstate, struct iter_qstate* iq, int id,
832 uint16_t qclass)
833 {
834 struct delegpt* dp;
835 struct module_qstate* subq;
836 verbose(VERB_DETAIL, "priming . %s NS",
837 sldns_lookup_by_id(sldns_rr_classes, (int)qclass)?
838 sldns_lookup_by_id(sldns_rr_classes, (int)qclass)->name:"??");
839 dp = hints_lookup_root(qstate->env->hints, qclass);
840 if(!dp) {
841 verbose(VERB_ALGO, "Cannot prime due to lack of hints");
842 return 0;
843 }
844 /* Priming requests start at the QUERYTARGETS state, skipping
845 * the normal INIT state logic (which would cause an infloop). */
846 if(!generate_sub_request((uint8_t*)"\000", 1, LDNS_RR_TYPE_NS,
847 qclass, qstate, id, iq, QUERYTARGETS_STATE, PRIME_RESP_STATE,
848 &subq, 0, 0)) {
849 verbose(VERB_ALGO, "could not prime root");
850 return 0;
851 }
852 if(subq) {
853 struct iter_qstate* subiq =
854 (struct iter_qstate*)subq->minfo[id];
855 /* Set the initial delegation point to the hint.
856 * copy dp, it is now part of the root prime query.
857 * dp was part of in the fixed hints structure. */
858 subiq->dp = delegpt_copy(dp, subq->region);
859 if(!subiq->dp) {
860 log_err("out of memory priming root, copydp");
861 fptr_ok(fptr_whitelist_modenv_kill_sub(
862 qstate->env->kill_sub));
863 (*qstate->env->kill_sub)(subq);
864 return 0;
865 }
866 /* there should not be any target queries. */
867 subiq->num_target_queries = 0;
868 subiq->dnssec_expected = iter_indicates_dnssec(
869 qstate->env, subiq->dp, NULL, subq->qinfo.qclass);
870 }
871
872 /* this module stops, our submodule starts, and does the query. */
873 qstate->ext_state[id] = module_wait_subquery;
874 return 1;
875 }
876
877 /**
878 * Generate and process a stub priming request. This method tests for the
879 * need to prime a stub zone, so it is safe to call for every request.
880 *
881 * @param qstate: the qtstate that triggered the need to prime.
882 * @param iq: iterator query state.
883 * @param id: module id.
884 * @param qname: request name.
885 * @param qclass: request class.
886 * @return true if a priming subrequest was made, false if not. The will only
887 * issue a priming request if it detects an unprimed stub.
888 * Uses value of 2 to signal during stub-prime in root-prime situation
889 * that a noprime-stub is available and resolution can continue.
890 */
891 static int
prime_stub(struct module_qstate * qstate,struct iter_qstate * iq,int id,uint8_t * qname,uint16_t qclass)892 prime_stub(struct module_qstate* qstate, struct iter_qstate* iq, int id,
893 uint8_t* qname, uint16_t qclass)
894 {
895 /* Lookup the stub hint. This will return null if the stub doesn't
896 * need to be re-primed. */
897 struct iter_hints_stub* stub;
898 struct delegpt* stub_dp;
899 struct module_qstate* subq;
900
901 if(!qname) return 0;
902 stub = hints_lookup_stub(qstate->env->hints, qname, qclass, iq->dp);
903 /* The stub (if there is one) does not need priming. */
904 if(!stub)
905 return 0;
906 stub_dp = stub->dp;
907 /* if we have an auth_zone dp, and stub is equal, don't prime stub
908 * yet, unless we want to fallback and avoid the auth_zone */
909 if(!iq->auth_zone_avoid && iq->dp && iq->dp->auth_dp &&
910 query_dname_compare(iq->dp->name, stub_dp->name) == 0)
911 return 0;
912
913 /* is it a noprime stub (always use) */
914 if(stub->noprime) {
915 int r = 0;
916 if(iq->dp == NULL) r = 2;
917 /* copy the dp out of the fixed hints structure, so that
918 * it can be changed when servicing this query */
919 iq->dp = delegpt_copy(stub_dp, qstate->region);
920 if(!iq->dp) {
921 log_err("out of memory priming stub");
922 errinf(qstate, "malloc failure, priming stub");
923 (void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
924 return 1; /* return 1 to make module stop, with error */
925 }
926 log_nametypeclass(VERB_DETAIL, "use stub", stub_dp->name,
927 LDNS_RR_TYPE_NS, qclass);
928 return r;
929 }
930
931 /* Otherwise, we need to (re)prime the stub. */
932 log_nametypeclass(VERB_DETAIL, "priming stub", stub_dp->name,
933 LDNS_RR_TYPE_NS, qclass);
934
935 /* Stub priming events start at the QUERYTARGETS state to avoid the
936 * redundant INIT state processing. */
937 if(!generate_sub_request(stub_dp->name, stub_dp->namelen,
938 LDNS_RR_TYPE_NS, qclass, qstate, id, iq,
939 QUERYTARGETS_STATE, PRIME_RESP_STATE, &subq, 0, 0)) {
940 verbose(VERB_ALGO, "could not prime stub");
941 errinf(qstate, "could not generate lookup for stub prime");
942 (void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
943 return 1; /* return 1 to make module stop, with error */
944 }
945 if(subq) {
946 struct iter_qstate* subiq =
947 (struct iter_qstate*)subq->minfo[id];
948
949 /* Set the initial delegation point to the hint. */
950 /* make copy to avoid use of stub dp by different qs/threads */
951 subiq->dp = delegpt_copy(stub_dp, subq->region);
952 if(!subiq->dp) {
953 log_err("out of memory priming stub, copydp");
954 fptr_ok(fptr_whitelist_modenv_kill_sub(
955 qstate->env->kill_sub));
956 (*qstate->env->kill_sub)(subq);
957 errinf(qstate, "malloc failure, in stub prime");
958 (void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
959 return 1; /* return 1 to make module stop, with error */
960 }
961 /* there should not be any target queries -- although there
962 * wouldn't be anyway, since stub hints never have
963 * missing targets. */
964 subiq->num_target_queries = 0;
965 subiq->wait_priming_stub = 1;
966 subiq->dnssec_expected = iter_indicates_dnssec(
967 qstate->env, subiq->dp, NULL, subq->qinfo.qclass);
968 }
969
970 /* this module stops, our submodule starts, and does the query. */
971 qstate->ext_state[id] = module_wait_subquery;
972 return 1;
973 }
974
975 /**
976 * Generate a delegation point for an auth zone (unless cached dp is better)
977 * false on alloc failure.
978 */
979 static int
auth_zone_delegpt(struct module_qstate * qstate,struct iter_qstate * iq,uint8_t * delname,size_t delnamelen)980 auth_zone_delegpt(struct module_qstate* qstate, struct iter_qstate* iq,
981 uint8_t* delname, size_t delnamelen)
982 {
983 struct auth_zone* z;
984 if(iq->auth_zone_avoid)
985 return 1;
986 if(!delname) {
987 delname = iq->qchase.qname;
988 delnamelen = iq->qchase.qname_len;
989 }
990 lock_rw_rdlock(&qstate->env->auth_zones->lock);
991 z = auth_zones_find_zone(qstate->env->auth_zones, delname, delnamelen,
992 qstate->qinfo.qclass);
993 if(!z) {
994 lock_rw_unlock(&qstate->env->auth_zones->lock);
995 return 1;
996 }
997 lock_rw_rdlock(&z->lock);
998 lock_rw_unlock(&qstate->env->auth_zones->lock);
999 if(z->for_upstream) {
1000 if(iq->dp && query_dname_compare(z->name, iq->dp->name) == 0
1001 && iq->dp->auth_dp && qstate->blacklist &&
1002 z->fallback_enabled) {
1003 /* cache is blacklisted and fallback, and we
1004 * already have an auth_zone dp */
1005 if(verbosity>=VERB_ALGO) {
1006 char buf[255+1];
1007 dname_str(z->name, buf);
1008 verbose(VERB_ALGO, "auth_zone %s "
1009 "fallback because cache blacklisted",
1010 buf);
1011 }
1012 lock_rw_unlock(&z->lock);
1013 iq->dp = NULL;
1014 return 1;
1015 }
1016 if(iq->dp==NULL || dname_subdomain_c(z->name, iq->dp->name)) {
1017 struct delegpt* dp;
1018 if(qstate->blacklist && z->fallback_enabled) {
1019 /* cache is blacklisted because of a DNSSEC
1020 * validation failure, and the zone allows
1021 * fallback to the internet, query there. */
1022 if(verbosity>=VERB_ALGO) {
1023 char buf[255+1];
1024 dname_str(z->name, buf);
1025 verbose(VERB_ALGO, "auth_zone %s "
1026 "fallback because cache blacklisted",
1027 buf);
1028 }
1029 lock_rw_unlock(&z->lock);
1030 return 1;
1031 }
1032 dp = (struct delegpt*)regional_alloc_zero(
1033 qstate->region, sizeof(*dp));
1034 if(!dp) {
1035 log_err("alloc failure");
1036 if(z->fallback_enabled) {
1037 lock_rw_unlock(&z->lock);
1038 return 1; /* just fallback */
1039 }
1040 lock_rw_unlock(&z->lock);
1041 errinf(qstate, "malloc failure");
1042 return 0;
1043 }
1044 dp->name = regional_alloc_init(qstate->region,
1045 z->name, z->namelen);
1046 if(!dp->name) {
1047 log_err("alloc failure");
1048 if(z->fallback_enabled) {
1049 lock_rw_unlock(&z->lock);
1050 return 1; /* just fallback */
1051 }
1052 lock_rw_unlock(&z->lock);
1053 errinf(qstate, "malloc failure");
1054 return 0;
1055 }
1056 dp->namelen = z->namelen;
1057 dp->namelabs = z->namelabs;
1058 dp->auth_dp = 1;
1059 iq->dp = dp;
1060 }
1061 }
1062
1063 lock_rw_unlock(&z->lock);
1064 return 1;
1065 }
1066
1067 /**
1068 * Generate A and AAAA checks for glue that is in-zone for the referral
1069 * we just got to obtain authoritative information on the addresses.
1070 *
1071 * @param qstate: the qtstate that triggered the need to prime.
1072 * @param iq: iterator query state.
1073 * @param id: module id.
1074 */
1075 static void
generate_a_aaaa_check(struct module_qstate * qstate,struct iter_qstate * iq,int id)1076 generate_a_aaaa_check(struct module_qstate* qstate, struct iter_qstate* iq,
1077 int id)
1078 {
1079 struct iter_env* ie = (struct iter_env*)qstate->env->modinfo[id];
1080 struct module_qstate* subq;
1081 size_t i;
1082 struct reply_info* rep = iq->response->rep;
1083 struct ub_packed_rrset_key* s;
1084 log_assert(iq->dp);
1085
1086 if(iq->depth == ie->max_dependency_depth)
1087 return;
1088 /* walk through additional, and check if in-zone,
1089 * only relevant A, AAAA are left after scrub anyway */
1090 for(i=rep->an_numrrsets+rep->ns_numrrsets; i<rep->rrset_count; i++) {
1091 s = rep->rrsets[i];
1092 /* check *ALL* addresses that are transmitted in additional*/
1093 /* is it an address ? */
1094 if( !(ntohs(s->rk.type)==LDNS_RR_TYPE_A ||
1095 ntohs(s->rk.type)==LDNS_RR_TYPE_AAAA)) {
1096 continue;
1097 }
1098 /* is this query the same as the A/AAAA check for it */
1099 if(qstate->qinfo.qtype == ntohs(s->rk.type) &&
1100 qstate->qinfo.qclass == ntohs(s->rk.rrset_class) &&
1101 query_dname_compare(qstate->qinfo.qname,
1102 s->rk.dname)==0 &&
1103 (qstate->query_flags&BIT_RD) &&
1104 !(qstate->query_flags&BIT_CD))
1105 continue;
1106
1107 /* generate subrequest for it */
1108 log_nametypeclass(VERB_ALGO, "schedule addr fetch",
1109 s->rk.dname, ntohs(s->rk.type),
1110 ntohs(s->rk.rrset_class));
1111 if(!generate_sub_request(s->rk.dname, s->rk.dname_len,
1112 ntohs(s->rk.type), ntohs(s->rk.rrset_class),
1113 qstate, id, iq,
1114 INIT_REQUEST_STATE, FINISHED_STATE, &subq, 1, 0)) {
1115 verbose(VERB_ALGO, "could not generate addr check");
1116 return;
1117 }
1118 /* ignore subq - not need for more init */
1119 }
1120 }
1121
1122 /**
1123 * Generate a NS check request to obtain authoritative information
1124 * on an NS rrset.
1125 *
1126 * @param qstate: the qtstate that triggered the need to prime.
1127 * @param iq: iterator query state.
1128 * @param id: module id.
1129 */
1130 static void
generate_ns_check(struct module_qstate * qstate,struct iter_qstate * iq,int id)1131 generate_ns_check(struct module_qstate* qstate, struct iter_qstate* iq, int id)
1132 {
1133 struct iter_env* ie = (struct iter_env*)qstate->env->modinfo[id];
1134 struct module_qstate* subq;
1135 log_assert(iq->dp);
1136
1137 if(iq->depth == ie->max_dependency_depth)
1138 return;
1139 if(!can_have_last_resort(qstate->env, iq->dp->name, iq->dp->namelen,
1140 iq->qchase.qclass, NULL))
1141 return;
1142 /* is this query the same as the nscheck? */
1143 if(qstate->qinfo.qtype == LDNS_RR_TYPE_NS &&
1144 query_dname_compare(iq->dp->name, qstate->qinfo.qname)==0 &&
1145 (qstate->query_flags&BIT_RD) && !(qstate->query_flags&BIT_CD)){
1146 /* spawn off A, AAAA queries for in-zone glue to check */
1147 generate_a_aaaa_check(qstate, iq, id);
1148 return;
1149 }
1150 /* no need to get the NS record for DS, it is above the zonecut */
1151 if(qstate->qinfo.qtype == LDNS_RR_TYPE_DS)
1152 return;
1153
1154 log_nametypeclass(VERB_ALGO, "schedule ns fetch",
1155 iq->dp->name, LDNS_RR_TYPE_NS, iq->qchase.qclass);
1156 if(!generate_sub_request(iq->dp->name, iq->dp->namelen,
1157 LDNS_RR_TYPE_NS, iq->qchase.qclass, qstate, id, iq,
1158 INIT_REQUEST_STATE, FINISHED_STATE, &subq, 1, 0)) {
1159 verbose(VERB_ALGO, "could not generate ns check");
1160 return;
1161 }
1162 if(subq) {
1163 struct iter_qstate* subiq =
1164 (struct iter_qstate*)subq->minfo[id];
1165
1166 /* make copy to avoid use of stub dp by different qs/threads */
1167 /* refetch glue to start higher up the tree */
1168 subiq->refetch_glue = 1;
1169 subiq->dp = delegpt_copy(iq->dp, subq->region);
1170 if(!subiq->dp) {
1171 log_err("out of memory generating ns check, copydp");
1172 fptr_ok(fptr_whitelist_modenv_kill_sub(
1173 qstate->env->kill_sub));
1174 (*qstate->env->kill_sub)(subq);
1175 return;
1176 }
1177 }
1178 }
1179
1180 /**
1181 * Generate a DNSKEY prefetch query to get the DNSKEY for the DS record we
1182 * just got in a referral (where we have dnssec_expected, thus have trust
1183 * anchors above it). Note that right after calling this routine the
1184 * iterator detached subqueries (because of following the referral), and thus
1185 * the DNSKEY query becomes detached, its return stored in the cache for
1186 * later lookup by the validator. This cache lookup by the validator avoids
1187 * the roundtrip incurred by the DNSKEY query. The DNSKEY query is now
1188 * performed at about the same time the original query is sent to the domain,
1189 * thus the two answers are likely to be returned at about the same time,
1190 * saving a roundtrip from the validated lookup.
1191 *
1192 * @param qstate: the qtstate that triggered the need to prime.
1193 * @param iq: iterator query state.
1194 * @param id: module id.
1195 */
1196 static void
generate_dnskey_prefetch(struct module_qstate * qstate,struct iter_qstate * iq,int id)1197 generate_dnskey_prefetch(struct module_qstate* qstate,
1198 struct iter_qstate* iq, int id)
1199 {
1200 struct module_qstate* subq;
1201 log_assert(iq->dp);
1202
1203 /* is this query the same as the prefetch? */
1204 if(qstate->qinfo.qtype == LDNS_RR_TYPE_DNSKEY &&
1205 query_dname_compare(iq->dp->name, qstate->qinfo.qname)==0 &&
1206 (qstate->query_flags&BIT_RD) && !(qstate->query_flags&BIT_CD)){
1207 return;
1208 }
1209
1210 /* if the DNSKEY is in the cache this lookup will stop quickly */
1211 log_nametypeclass(VERB_ALGO, "schedule dnskey prefetch",
1212 iq->dp->name, LDNS_RR_TYPE_DNSKEY, iq->qchase.qclass);
1213 if(!generate_sub_request(iq->dp->name, iq->dp->namelen,
1214 LDNS_RR_TYPE_DNSKEY, iq->qchase.qclass, qstate, id, iq,
1215 INIT_REQUEST_STATE, FINISHED_STATE, &subq, 0, 0)) {
1216 /* we'll be slower, but it'll work */
1217 verbose(VERB_ALGO, "could not generate dnskey prefetch");
1218 return;
1219 }
1220 if(subq) {
1221 struct iter_qstate* subiq =
1222 (struct iter_qstate*)subq->minfo[id];
1223 /* this qstate has the right delegation for the dnskey lookup*/
1224 /* make copy to avoid use of stub dp by different qs/threads */
1225 subiq->dp = delegpt_copy(iq->dp, subq->region);
1226 /* if !subiq->dp, it'll start from the cache, no problem */
1227 }
1228 }
1229
1230 /**
1231 * See if the query needs forwarding.
1232 *
1233 * @param qstate: query state.
1234 * @param iq: iterator query state.
1235 * @return true if the request is forwarded, false if not.
1236 * If returns true but, iq->dp is NULL then a malloc failure occurred.
1237 */
1238 static int
forward_request(struct module_qstate * qstate,struct iter_qstate * iq)1239 forward_request(struct module_qstate* qstate, struct iter_qstate* iq)
1240 {
1241 struct delegpt* dp;
1242 uint8_t* delname = iq->qchase.qname;
1243 size_t delnamelen = iq->qchase.qname_len;
1244 if(iq->refetch_glue && iq->dp) {
1245 delname = iq->dp->name;
1246 delnamelen = iq->dp->namelen;
1247 }
1248 /* strip one label off of DS query to lookup higher for it */
1249 if( (iq->qchase.qtype == LDNS_RR_TYPE_DS || iq->refetch_glue)
1250 && !dname_is_root(iq->qchase.qname))
1251 dname_remove_label(&delname, &delnamelen);
1252 dp = forwards_lookup(qstate->env->fwds, delname, iq->qchase.qclass);
1253 if(!dp)
1254 return 0;
1255 /* send recursion desired to forward addr */
1256 iq->chase_flags |= BIT_RD;
1257 iq->dp = delegpt_copy(dp, qstate->region);
1258 /* iq->dp checked by caller */
1259 verbose(VERB_ALGO, "forwarding request");
1260 return 1;
1261 }
1262
1263 /**
1264 * Process the initial part of the request handling. This state roughly
1265 * corresponds to resolver algorithms steps 1 (find answer in cache) and 2
1266 * (find the best servers to ask).
1267 *
1268 * Note that all requests start here, and query restarts revisit this state.
1269 *
1270 * This state either generates: 1) a response, from cache or error, 2) a
1271 * priming event, or 3) forwards the request to the next state (init2,
1272 * generally).
1273 *
1274 * @param qstate: query state.
1275 * @param iq: iterator query state.
1276 * @param ie: iterator shared global environment.
1277 * @param id: module id.
1278 * @return true if the event needs more request processing immediately,
1279 * false if not.
1280 */
1281 static int
processInitRequest(struct module_qstate * qstate,struct iter_qstate * iq,struct iter_env * ie,int id)1282 processInitRequest(struct module_qstate* qstate, struct iter_qstate* iq,
1283 struct iter_env* ie, int id)
1284 {
1285 uint8_t* delname, *dpname=NULL;
1286 size_t delnamelen, dpnamelen=0;
1287 struct dns_msg* msg = NULL;
1288
1289 log_query_info(VERB_DETAIL, "resolving", &qstate->qinfo);
1290 /* check effort */
1291
1292 /* We enforce a maximum number of query restarts. This is primarily a
1293 * cheap way to prevent CNAME loops. */
1294 if(iq->query_restart_count > MAX_RESTART_COUNT) {
1295 verbose(VERB_QUERY, "request has exceeded the maximum number"
1296 " of query restarts with %d", iq->query_restart_count);
1297 errinf(qstate, "request has exceeded the maximum number "
1298 "restarts (eg. indirections)");
1299 if(iq->qchase.qname)
1300 errinf_dname(qstate, "stop at", iq->qchase.qname);
1301 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1302 }
1303
1304 /* We enforce a maximum recursion/dependency depth -- in general,
1305 * this is unnecessary for dependency loops (although it will
1306 * catch those), but it provides a sensible limit to the amount
1307 * of work required to answer a given query. */
1308 verbose(VERB_ALGO, "request has dependency depth of %d", iq->depth);
1309 if(iq->depth > ie->max_dependency_depth) {
1310 verbose(VERB_QUERY, "request has exceeded the maximum "
1311 "dependency depth with depth of %d", iq->depth);
1312 errinf(qstate, "request has exceeded the maximum dependency "
1313 "depth (eg. nameserver lookup recursion)");
1314 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1315 }
1316
1317 /* If the request is qclass=ANY, setup to generate each class */
1318 if(qstate->qinfo.qclass == LDNS_RR_CLASS_ANY) {
1319 iq->qchase.qclass = 0;
1320 return next_state(iq, COLLECT_CLASS_STATE);
1321 }
1322
1323 /*
1324 * If we are restricted by a forward-zone or a stub-zone, we
1325 * can't re-fetch glue for this delegation point.
1326 * we won’t try to re-fetch glue if the iq->dp is null.
1327 */
1328 if (iq->refetch_glue &&
1329 iq->dp &&
1330 !can_have_last_resort(qstate->env, iq->dp->name,
1331 iq->dp->namelen, iq->qchase.qclass, NULL)) {
1332 iq->refetch_glue = 0;
1333 }
1334
1335 /* Resolver Algorithm Step 1 -- Look for the answer in local data. */
1336
1337 /* This either results in a query restart (CNAME cache response), a
1338 * terminating response (ANSWER), or a cache miss (null). */
1339
1340 if (iter_stub_fwd_no_cache(qstate, &iq->qchase, &dpname, &dpnamelen)) {
1341 /* Asked to not query cache. */
1342 verbose(VERB_ALGO, "no-cache set, going to the network");
1343 qstate->no_cache_lookup = 1;
1344 qstate->no_cache_store = 1;
1345 msg = NULL;
1346 } else if(qstate->blacklist) {
1347 /* if cache, or anything else, was blacklisted then
1348 * getting older results from cache is a bad idea, no cache */
1349 verbose(VERB_ALGO, "cache blacklisted, going to the network");
1350 msg = NULL;
1351 } else if(!qstate->no_cache_lookup) {
1352 msg = dns_cache_lookup(qstate->env, iq->qchase.qname,
1353 iq->qchase.qname_len, iq->qchase.qtype,
1354 iq->qchase.qclass, qstate->query_flags,
1355 qstate->region, qstate->env->scratch, 0, dpname,
1356 dpnamelen);
1357 if(!msg && qstate->env->neg_cache &&
1358 iter_qname_indicates_dnssec(qstate->env, &iq->qchase)) {
1359 /* lookup in negative cache; may result in
1360 * NOERROR/NODATA or NXDOMAIN answers that need validation */
1361 msg = val_neg_getmsg(qstate->env->neg_cache, &iq->qchase,
1362 qstate->region, qstate->env->rrset_cache,
1363 qstate->env->scratch_buffer,
1364 *qstate->env->now, 1/*add SOA*/, NULL,
1365 qstate->env->cfg);
1366 }
1367 /* item taken from cache does not match our query name, thus
1368 * security needs to be re-examined later */
1369 if(msg && query_dname_compare(qstate->qinfo.qname,
1370 iq->qchase.qname) != 0)
1371 msg->rep->security = sec_status_unchecked;
1372 }
1373 if(msg) {
1374 /* handle positive cache response */
1375 enum response_type type = response_type_from_cache(msg,
1376 &iq->qchase);
1377 if(verbosity >= VERB_ALGO) {
1378 log_dns_msg("msg from cache lookup", &msg->qinfo,
1379 msg->rep);
1380 verbose(VERB_ALGO, "msg ttl is %d, prefetch ttl %d",
1381 (int)msg->rep->ttl,
1382 (int)msg->rep->prefetch_ttl);
1383 }
1384
1385 if(type == RESPONSE_TYPE_CNAME) {
1386 uint8_t* sname = 0;
1387 size_t slen = 0;
1388 verbose(VERB_ALGO, "returning CNAME response from "
1389 "cache");
1390 if(!handle_cname_response(qstate, iq, msg,
1391 &sname, &slen)) {
1392 errinf(qstate, "failed to prepend CNAME "
1393 "components, malloc failure");
1394 return error_response(qstate, id,
1395 LDNS_RCODE_SERVFAIL);
1396 }
1397 iq->qchase.qname = sname;
1398 iq->qchase.qname_len = slen;
1399 /* This *is* a query restart, even if it is a cheap
1400 * one. */
1401 iq->dp = NULL;
1402 iq->refetch_glue = 0;
1403 iq->query_restart_count++;
1404 iq->sent_count = 0;
1405 iq->dp_target_count = 0;
1406 sock_list_insert(&qstate->reply_origin, NULL, 0, qstate->region);
1407 if(qstate->env->cfg->qname_minimisation)
1408 iq->minimisation_state = INIT_MINIMISE_STATE;
1409 return next_state(iq, INIT_REQUEST_STATE);
1410 }
1411
1412 /* if from cache, NULL, else insert 'cache IP' len=0 */
1413 if(qstate->reply_origin)
1414 sock_list_insert(&qstate->reply_origin, NULL, 0, qstate->region);
1415 if(FLAGS_GET_RCODE(msg->rep->flags) == LDNS_RCODE_SERVFAIL)
1416 errinf(qstate, "SERVFAIL in cache");
1417 /* it is an answer, response, to final state */
1418 verbose(VERB_ALGO, "returning answer from cache.");
1419 iq->response = msg;
1420 return final_state(iq);
1421 }
1422
1423 /* attempt to forward the request */
1424 if(forward_request(qstate, iq))
1425 {
1426 if(!iq->dp) {
1427 log_err("alloc failure for forward dp");
1428 errinf(qstate, "malloc failure for forward zone");
1429 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1430 }
1431 iq->refetch_glue = 0;
1432 iq->minimisation_state = DONOT_MINIMISE_STATE;
1433 /* the request has been forwarded.
1434 * forwarded requests need to be immediately sent to the
1435 * next state, QUERYTARGETS. */
1436 return next_state(iq, QUERYTARGETS_STATE);
1437 }
1438
1439 /* Resolver Algorithm Step 2 -- find the "best" servers. */
1440
1441 /* first, adjust for DS queries. To avoid the grandparent problem,
1442 * we just look for the closest set of server to the parent of qname.
1443 * When re-fetching glue we also need to ask the parent.
1444 */
1445 if(iq->refetch_glue) {
1446 if(!iq->dp) {
1447 log_err("internal or malloc fail: no dp for refetch");
1448 errinf(qstate, "malloc failure, for delegation info");
1449 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1450 }
1451 delname = iq->dp->name;
1452 delnamelen = iq->dp->namelen;
1453 } else {
1454 delname = iq->qchase.qname;
1455 delnamelen = iq->qchase.qname_len;
1456 }
1457 if(iq->qchase.qtype == LDNS_RR_TYPE_DS || iq->refetch_glue ||
1458 (iq->qchase.qtype == LDNS_RR_TYPE_NS && qstate->prefetch_leeway
1459 && can_have_last_resort(qstate->env, delname, delnamelen, iq->qchase.qclass, NULL))) {
1460 /* remove first label from delname, root goes to hints,
1461 * but only to fetch glue, not for qtype=DS. */
1462 /* also when prefetching an NS record, fetch it again from
1463 * its parent, just as if it expired, so that you do not
1464 * get stuck on an older nameserver that gives old NSrecords */
1465 if(dname_is_root(delname) && (iq->refetch_glue ||
1466 (iq->qchase.qtype == LDNS_RR_TYPE_NS &&
1467 qstate->prefetch_leeway)))
1468 delname = NULL; /* go to root priming */
1469 else dname_remove_label(&delname, &delnamelen);
1470 }
1471 /* delname is the name to lookup a delegation for. If NULL rootprime */
1472 while(1) {
1473
1474 /* Lookup the delegation in the cache. If null, then the
1475 * cache needs to be primed for the qclass. */
1476 if(delname)
1477 iq->dp = dns_cache_find_delegation(qstate->env, delname,
1478 delnamelen, iq->qchase.qtype, iq->qchase.qclass,
1479 qstate->region, &iq->deleg_msg,
1480 *qstate->env->now+qstate->prefetch_leeway);
1481 else iq->dp = NULL;
1482
1483 /* If the cache has returned nothing, then we have a
1484 * root priming situation. */
1485 if(iq->dp == NULL) {
1486 int r;
1487 /* if under auth zone, no prime needed */
1488 if(!auth_zone_delegpt(qstate, iq, delname, delnamelen))
1489 return error_response(qstate, id,
1490 LDNS_RCODE_SERVFAIL);
1491 if(iq->dp) /* use auth zone dp */
1492 return next_state(iq, INIT_REQUEST_2_STATE);
1493 /* if there is a stub, then no root prime needed */
1494 r = prime_stub(qstate, iq, id, delname,
1495 iq->qchase.qclass);
1496 if(r == 2)
1497 break; /* got noprime-stub-zone, continue */
1498 else if(r)
1499 return 0; /* stub prime request made */
1500 if(forwards_lookup_root(qstate->env->fwds,
1501 iq->qchase.qclass)) {
1502 /* forward zone root, no root prime needed */
1503 /* fill in some dp - safety belt */
1504 iq->dp = hints_lookup_root(qstate->env->hints,
1505 iq->qchase.qclass);
1506 if(!iq->dp) {
1507 log_err("internal error: no hints dp");
1508 errinf(qstate, "no hints for this class");
1509 return error_response(qstate, id,
1510 LDNS_RCODE_SERVFAIL);
1511 }
1512 iq->dp = delegpt_copy(iq->dp, qstate->region);
1513 if(!iq->dp) {
1514 log_err("out of memory in safety belt");
1515 errinf(qstate, "malloc failure, in safety belt");
1516 return error_response(qstate, id,
1517 LDNS_RCODE_SERVFAIL);
1518 }
1519 return next_state(iq, INIT_REQUEST_2_STATE);
1520 }
1521 /* Note that the result of this will set a new
1522 * DelegationPoint based on the result of priming. */
1523 if(!prime_root(qstate, iq, id, iq->qchase.qclass))
1524 return error_response(qstate, id,
1525 LDNS_RCODE_REFUSED);
1526
1527 /* priming creates and sends a subordinate query, with
1528 * this query as the parent. So further processing for
1529 * this event will stop until reactivated by the
1530 * results of priming. */
1531 return 0;
1532 }
1533 if(!iq->ratelimit_ok && qstate->prefetch_leeway)
1534 iq->ratelimit_ok = 1; /* allow prefetches, this keeps
1535 otherwise valid data in the cache */
1536 if(!iq->ratelimit_ok && infra_ratelimit_exceeded(
1537 qstate->env->infra_cache, iq->dp->name,
1538 iq->dp->namelen, *qstate->env->now)) {
1539 /* and increment the rate, so that the rate for time
1540 * now will also exceed the rate, keeping cache fresh */
1541 (void)infra_ratelimit_inc(qstate->env->infra_cache,
1542 iq->dp->name, iq->dp->namelen,
1543 *qstate->env->now, &qstate->qinfo,
1544 qstate->reply);
1545 /* see if we are passed through with slip factor */
1546 if(qstate->env->cfg->ratelimit_factor != 0 &&
1547 ub_random_max(qstate->env->rnd,
1548 qstate->env->cfg->ratelimit_factor) == 1) {
1549 iq->ratelimit_ok = 1;
1550 log_nametypeclass(VERB_ALGO, "ratelimit allowed through for "
1551 "delegation point", iq->dp->name,
1552 LDNS_RR_TYPE_NS, LDNS_RR_CLASS_IN);
1553 } else {
1554 lock_basic_lock(&ie->queries_ratelimit_lock);
1555 ie->num_queries_ratelimited++;
1556 lock_basic_unlock(&ie->queries_ratelimit_lock);
1557 log_nametypeclass(VERB_ALGO, "ratelimit exceeded with "
1558 "delegation point", iq->dp->name,
1559 LDNS_RR_TYPE_NS, LDNS_RR_CLASS_IN);
1560 qstate->was_ratelimited = 1;
1561 errinf(qstate, "query was ratelimited");
1562 errinf_dname(qstate, "for zone", iq->dp->name);
1563 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1564 }
1565 }
1566
1567 /* see if this dp not useless.
1568 * It is useless if:
1569 * o all NS items are required glue.
1570 * or the query is for NS item that is required glue.
1571 * o no addresses are provided.
1572 * o RD qflag is on.
1573 * Instead, go up one level, and try to get even further
1574 * If the root was useless, use safety belt information.
1575 * Only check cache returns, because replies for servers
1576 * could be useless but lead to loops (bumping into the
1577 * same server reply) if useless-checked.
1578 */
1579 if(iter_dp_is_useless(&qstate->qinfo, qstate->query_flags,
1580 iq->dp)) {
1581 struct delegpt* retdp = NULL;
1582 if(!can_have_last_resort(qstate->env, iq->dp->name, iq->dp->namelen, iq->qchase.qclass, &retdp)) {
1583 if(retdp) {
1584 verbose(VERB_QUERY, "cache has stub "
1585 "or fwd but no addresses, "
1586 "fallback to config");
1587 iq->dp = delegpt_copy(retdp,
1588 qstate->region);
1589 if(!iq->dp) {
1590 log_err("out of memory in "
1591 "stub/fwd fallback");
1592 errinf(qstate, "malloc failure, for fallback to config");
1593 return error_response(qstate,
1594 id, LDNS_RCODE_SERVFAIL);
1595 }
1596 break;
1597 }
1598 verbose(VERB_ALGO, "useless dp "
1599 "but cannot go up, servfail");
1600 delegpt_log(VERB_ALGO, iq->dp);
1601 errinf(qstate, "no useful nameservers, "
1602 "and cannot go up");
1603 errinf_dname(qstate, "for zone", iq->dp->name);
1604 return error_response(qstate, id,
1605 LDNS_RCODE_SERVFAIL);
1606 }
1607 if(dname_is_root(iq->dp->name)) {
1608 /* use safety belt */
1609 verbose(VERB_QUERY, "Cache has root NS but "
1610 "no addresses. Fallback to the safety belt.");
1611 iq->dp = hints_lookup_root(qstate->env->hints,
1612 iq->qchase.qclass);
1613 /* note deleg_msg is from previous lookup,
1614 * but RD is on, so it is not used */
1615 if(!iq->dp) {
1616 log_err("internal error: no hints dp");
1617 return error_response(qstate, id,
1618 LDNS_RCODE_REFUSED);
1619 }
1620 iq->dp = delegpt_copy(iq->dp, qstate->region);
1621 if(!iq->dp) {
1622 log_err("out of memory in safety belt");
1623 errinf(qstate, "malloc failure, in safety belt, for root");
1624 return error_response(qstate, id,
1625 LDNS_RCODE_SERVFAIL);
1626 }
1627 break;
1628 } else {
1629 verbose(VERB_ALGO,
1630 "cache delegation was useless:");
1631 delegpt_log(VERB_ALGO, iq->dp);
1632 /* go up */
1633 delname = iq->dp->name;
1634 delnamelen = iq->dp->namelen;
1635 dname_remove_label(&delname, &delnamelen);
1636 }
1637 } else break;
1638 }
1639
1640 verbose(VERB_ALGO, "cache delegation returns delegpt");
1641 delegpt_log(VERB_ALGO, iq->dp);
1642
1643 /* Otherwise, set the current delegation point and move on to the
1644 * next state. */
1645 return next_state(iq, INIT_REQUEST_2_STATE);
1646 }
1647
1648 /**
1649 * Process the second part of the initial request handling. This state
1650 * basically exists so that queries that generate root priming events have
1651 * the same init processing as ones that do not. Request events that reach
1652 * this state must have a valid currentDelegationPoint set.
1653 *
1654 * This part is primarily handling stub zone priming. Events that reach this
1655 * state must have a current delegation point.
1656 *
1657 * @param qstate: query state.
1658 * @param iq: iterator query state.
1659 * @param id: module id.
1660 * @return true if the event needs more request processing immediately,
1661 * false if not.
1662 */
1663 static int
processInitRequest2(struct module_qstate * qstate,struct iter_qstate * iq,int id)1664 processInitRequest2(struct module_qstate* qstate, struct iter_qstate* iq,
1665 int id)
1666 {
1667 uint8_t* delname;
1668 size_t delnamelen;
1669 log_query_info(VERB_QUERY, "resolving (init part 2): ",
1670 &qstate->qinfo);
1671
1672 delname = iq->qchase.qname;
1673 delnamelen = iq->qchase.qname_len;
1674 if(iq->refetch_glue) {
1675 struct iter_hints_stub* stub;
1676 if(!iq->dp) {
1677 log_err("internal or malloc fail: no dp for refetch");
1678 errinf(qstate, "malloc failure, no delegation info");
1679 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1680 }
1681 /* Do not send queries above stub, do not set delname to dp if
1682 * this is above stub without stub-first. */
1683 stub = hints_lookup_stub(
1684 qstate->env->hints, iq->qchase.qname, iq->qchase.qclass,
1685 iq->dp);
1686 if(!stub || !stub->dp->has_parent_side_NS ||
1687 dname_subdomain_c(iq->dp->name, stub->dp->name)) {
1688 delname = iq->dp->name;
1689 delnamelen = iq->dp->namelen;
1690 }
1691 }
1692 if(iq->qchase.qtype == LDNS_RR_TYPE_DS || iq->refetch_glue) {
1693 if(!dname_is_root(delname))
1694 dname_remove_label(&delname, &delnamelen);
1695 iq->refetch_glue = 0; /* if CNAME causes restart, no refetch */
1696 }
1697
1698 /* see if we have an auth zone to answer from, improves dp from cache
1699 * (if any dp from cache) with auth zone dp, if that is lower */
1700 if(!auth_zone_delegpt(qstate, iq, delname, delnamelen))
1701 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1702
1703 /* Check to see if we need to prime a stub zone. */
1704 if(prime_stub(qstate, iq, id, delname, iq->qchase.qclass)) {
1705 /* A priming sub request was made */
1706 return 0;
1707 }
1708
1709 /* most events just get forwarded to the next state. */
1710 return next_state(iq, INIT_REQUEST_3_STATE);
1711 }
1712
1713 /**
1714 * Process the third part of the initial request handling. This state exists
1715 * as a separate state so that queries that generate stub priming events
1716 * will get the tail end of the init process but not repeat the stub priming
1717 * check.
1718 *
1719 * @param qstate: query state.
1720 * @param iq: iterator query state.
1721 * @param id: module id.
1722 * @return true, advancing the event to the QUERYTARGETS_STATE.
1723 */
1724 static int
processInitRequest3(struct module_qstate * qstate,struct iter_qstate * iq,int id)1725 processInitRequest3(struct module_qstate* qstate, struct iter_qstate* iq,
1726 int id)
1727 {
1728 log_query_info(VERB_QUERY, "resolving (init part 3): ",
1729 &qstate->qinfo);
1730 /* if the cache reply dp equals a validation anchor or msg has DS,
1731 * then DNSSEC RRSIGs are expected in the reply */
1732 iq->dnssec_expected = iter_indicates_dnssec(qstate->env, iq->dp,
1733 iq->deleg_msg, iq->qchase.qclass);
1734
1735 /* If the RD flag wasn't set, then we just finish with the
1736 * cached referral as the response. */
1737 if(!(qstate->query_flags & BIT_RD) && iq->deleg_msg) {
1738 iq->response = iq->deleg_msg;
1739 if(verbosity >= VERB_ALGO && iq->response)
1740 log_dns_msg("no RD requested, using delegation msg",
1741 &iq->response->qinfo, iq->response->rep);
1742 if(qstate->reply_origin)
1743 sock_list_insert(&qstate->reply_origin, NULL, 0, qstate->region);
1744 return final_state(iq);
1745 }
1746 /* After this point, unset the RD flag -- this query is going to
1747 * be sent to an auth. server. */
1748 iq->chase_flags &= ~BIT_RD;
1749
1750 /* if dnssec expected, fetch key for the trust-anchor or cached-DS */
1751 if(iq->dnssec_expected && qstate->env->cfg->prefetch_key &&
1752 !(qstate->query_flags&BIT_CD)) {
1753 generate_dnskey_prefetch(qstate, iq, id);
1754 fptr_ok(fptr_whitelist_modenv_detach_subs(
1755 qstate->env->detach_subs));
1756 (*qstate->env->detach_subs)(qstate);
1757 }
1758
1759 /* Jump to the next state. */
1760 return next_state(iq, QUERYTARGETS_STATE);
1761 }
1762
1763 /**
1764 * Given a basic query, generate a parent-side "target" query.
1765 * These are subordinate queries for missing delegation point target addresses,
1766 * for which only the parent of the delegation provides correct IP addresses.
1767 *
1768 * @param qstate: query state.
1769 * @param iq: iterator query state.
1770 * @param id: module id.
1771 * @param name: target qname.
1772 * @param namelen: target qname length.
1773 * @param qtype: target qtype (either A or AAAA).
1774 * @param qclass: target qclass.
1775 * @return true on success, false on failure.
1776 */
1777 static int
generate_parentside_target_query(struct module_qstate * qstate,struct iter_qstate * iq,int id,uint8_t * name,size_t namelen,uint16_t qtype,uint16_t qclass)1778 generate_parentside_target_query(struct module_qstate* qstate,
1779 struct iter_qstate* iq, int id, uint8_t* name, size_t namelen,
1780 uint16_t qtype, uint16_t qclass)
1781 {
1782 struct module_qstate* subq;
1783 if(!generate_sub_request(name, namelen, qtype, qclass, qstate,
1784 id, iq, INIT_REQUEST_STATE, FINISHED_STATE, &subq, 0, 0))
1785 return 0;
1786 if(subq) {
1787 struct iter_qstate* subiq =
1788 (struct iter_qstate*)subq->minfo[id];
1789 /* blacklist the cache - we want to fetch parent stuff */
1790 sock_list_insert(&subq->blacklist, NULL, 0, subq->region);
1791 subiq->query_for_pside_glue = 1;
1792 if(dname_subdomain_c(name, iq->dp->name)) {
1793 subiq->dp = delegpt_copy(iq->dp, subq->region);
1794 subiq->dnssec_expected = iter_indicates_dnssec(
1795 qstate->env, subiq->dp, NULL,
1796 subq->qinfo.qclass);
1797 subiq->refetch_glue = 1;
1798 } else {
1799 subiq->dp = dns_cache_find_delegation(qstate->env,
1800 name, namelen, qtype, qclass, subq->region,
1801 &subiq->deleg_msg,
1802 *qstate->env->now+subq->prefetch_leeway);
1803 /* if no dp, then it's from root, refetch unneeded */
1804 if(subiq->dp) {
1805 subiq->dnssec_expected = iter_indicates_dnssec(
1806 qstate->env, subiq->dp, NULL,
1807 subq->qinfo.qclass);
1808 subiq->refetch_glue = 1;
1809 }
1810 }
1811 }
1812 log_nametypeclass(VERB_QUERY, "new pside target", name, qtype, qclass);
1813 return 1;
1814 }
1815
1816 /**
1817 * Given a basic query, generate a "target" query. These are subordinate
1818 * queries for missing delegation point target addresses.
1819 *
1820 * @param qstate: query state.
1821 * @param iq: iterator query state.
1822 * @param id: module id.
1823 * @param name: target qname.
1824 * @param namelen: target qname length.
1825 * @param qtype: target qtype (either A or AAAA).
1826 * @param qclass: target qclass.
1827 * @return true on success, false on failure.
1828 */
1829 static int
generate_target_query(struct module_qstate * qstate,struct iter_qstate * iq,int id,uint8_t * name,size_t namelen,uint16_t qtype,uint16_t qclass)1830 generate_target_query(struct module_qstate* qstate, struct iter_qstate* iq,
1831 int id, uint8_t* name, size_t namelen, uint16_t qtype, uint16_t qclass)
1832 {
1833 struct module_qstate* subq;
1834 if(!generate_sub_request(name, namelen, qtype, qclass, qstate,
1835 id, iq, INIT_REQUEST_STATE, FINISHED_STATE, &subq, 0, 0))
1836 return 0;
1837 log_nametypeclass(VERB_QUERY, "new target", name, qtype, qclass);
1838 return 1;
1839 }
1840
1841 /**
1842 * Given an event at a certain state, generate zero or more target queries
1843 * for it's current delegation point.
1844 *
1845 * @param qstate: query state.
1846 * @param iq: iterator query state.
1847 * @param ie: iterator shared global environment.
1848 * @param id: module id.
1849 * @param maxtargets: The maximum number of targets to query for.
1850 * if it is negative, there is no maximum number of targets.
1851 * @param num: returns the number of queries generated and processed,
1852 * which may be zero if there were no missing targets.
1853 * @return false on error.
1854 */
1855 static int
query_for_targets(struct module_qstate * qstate,struct iter_qstate * iq,struct iter_env * ie,int id,int maxtargets,int * num)1856 query_for_targets(struct module_qstate* qstate, struct iter_qstate* iq,
1857 struct iter_env* ie, int id, int maxtargets, int* num)
1858 {
1859 int query_count = 0;
1860 struct delegpt_ns* ns;
1861 int missing;
1862 int toget = 0;
1863
1864 if(iq->depth == ie->max_dependency_depth)
1865 return 0;
1866 if(iq->depth > 0 && iq->target_count &&
1867 iq->target_count[1] > MAX_TARGET_COUNT) {
1868 char s[LDNS_MAX_DOMAINLEN+1];
1869 dname_str(qstate->qinfo.qname, s);
1870 verbose(VERB_QUERY, "request %s has exceeded the maximum "
1871 "number of glue fetches %d", s, iq->target_count[1]);
1872 return 0;
1873 }
1874 if(iq->dp_target_count > MAX_DP_TARGET_COUNT) {
1875 char s[LDNS_MAX_DOMAINLEN+1];
1876 dname_str(qstate->qinfo.qname, s);
1877 verbose(VERB_QUERY, "request %s has exceeded the maximum "
1878 "number of glue fetches %d to a single delegation point",
1879 s, iq->dp_target_count);
1880 return 0;
1881 }
1882
1883 iter_mark_cycle_targets(qstate, iq->dp);
1884 missing = (int)delegpt_count_missing_targets(iq->dp);
1885 log_assert(maxtargets != 0); /* that would not be useful */
1886
1887 /* Generate target requests. Basically, any missing targets
1888 * are queried for here, regardless if it is necessary to do
1889 * so to continue processing. */
1890 if(maxtargets < 0 || maxtargets > missing)
1891 toget = missing;
1892 else toget = maxtargets;
1893 if(toget == 0) {
1894 *num = 0;
1895 return 1;
1896 }
1897 /* select 'toget' items from the total of 'missing' items */
1898 log_assert(toget <= missing);
1899
1900 /* loop over missing targets */
1901 for(ns = iq->dp->nslist; ns; ns = ns->next) {
1902 if(ns->resolved)
1903 continue;
1904
1905 /* randomly select this item with probability toget/missing */
1906 if(!iter_ns_probability(qstate->env->rnd, toget, missing)) {
1907 /* do not select this one, next; select toget number
1908 * of items from a list one less in size */
1909 missing --;
1910 continue;
1911 }
1912
1913 if(ie->supports_ipv6 && !ns->got6) {
1914 /* Send the AAAA request. */
1915 if(!generate_target_query(qstate, iq, id,
1916 ns->name, ns->namelen,
1917 LDNS_RR_TYPE_AAAA, iq->qchase.qclass)) {
1918 *num = query_count;
1919 if(query_count > 0)
1920 qstate->ext_state[id] = module_wait_subquery;
1921 return 0;
1922 }
1923 query_count++;
1924 }
1925 /* Send the A request. */
1926 if(ie->supports_ipv4 && !ns->got4) {
1927 if(!generate_target_query(qstate, iq, id,
1928 ns->name, ns->namelen,
1929 LDNS_RR_TYPE_A, iq->qchase.qclass)) {
1930 *num = query_count;
1931 if(query_count > 0)
1932 qstate->ext_state[id] = module_wait_subquery;
1933 return 0;
1934 }
1935 query_count++;
1936 }
1937
1938 /* mark this target as in progress. */
1939 ns->resolved = 1;
1940 missing--;
1941 toget--;
1942 if(toget == 0)
1943 break;
1944 }
1945 *num = query_count;
1946 if(query_count > 0)
1947 qstate->ext_state[id] = module_wait_subquery;
1948
1949 return 1;
1950 }
1951
1952 /**
1953 * Called by processQueryTargets when it would like extra targets to query
1954 * but it seems to be out of options. At last resort some less appealing
1955 * options are explored. If there are no more options, the result is SERVFAIL
1956 *
1957 * @param qstate: query state.
1958 * @param iq: iterator query state.
1959 * @param ie: iterator shared global environment.
1960 * @param id: module id.
1961 * @return true if the event requires more request processing immediately,
1962 * false if not.
1963 */
1964 static int
processLastResort(struct module_qstate * qstate,struct iter_qstate * iq,struct iter_env * ie,int id)1965 processLastResort(struct module_qstate* qstate, struct iter_qstate* iq,
1966 struct iter_env* ie, int id)
1967 {
1968 struct delegpt_ns* ns;
1969 int query_count = 0;
1970 verbose(VERB_ALGO, "No more query targets, attempting last resort");
1971 log_assert(iq->dp);
1972
1973 if(!can_have_last_resort(qstate->env, iq->dp->name, iq->dp->namelen,
1974 iq->qchase.qclass, NULL)) {
1975 /* fail -- no more targets, no more hope of targets, no hope
1976 * of a response. */
1977 errinf(qstate, "all the configured stub or forward servers failed,");
1978 errinf_dname(qstate, "at zone", iq->dp->name);
1979 errinf_reply(qstate, iq);
1980 verbose(VERB_QUERY, "configured stub or forward servers failed -- returning SERVFAIL");
1981 return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
1982 }
1983 if(!iq->dp->has_parent_side_NS && dname_is_root(iq->dp->name)) {
1984 struct delegpt* p = hints_lookup_root(qstate->env->hints,
1985 iq->qchase.qclass);
1986 if(p) {
1987 struct delegpt_addr* a;
1988 iq->chase_flags &= ~BIT_RD; /* go to authorities */
1989 for(ns = p->nslist; ns; ns=ns->next) {
1990 (void)delegpt_add_ns(iq->dp, qstate->region,
1991 ns->name, ns->lame);
1992 }
1993 for(a = p->target_list; a; a=a->next_target) {
1994 (void)delegpt_add_addr(iq->dp, qstate->region,
1995 &a->addr, a->addrlen, a->bogus,
1996 a->lame, a->tls_auth_name, NULL);
1997 }
1998 }
1999 iq->dp->has_parent_side_NS = 1;
2000 } else if(!iq->dp->has_parent_side_NS) {
2001 if(!iter_lookup_parent_NS_from_cache(qstate->env, iq->dp,
2002 qstate->region, &qstate->qinfo)
2003 || !iq->dp->has_parent_side_NS) {
2004 /* if: malloc failure in lookup go up to try */
2005 /* if: no parent NS in cache - go up one level */
2006 verbose(VERB_ALGO, "try to grab parent NS");
2007 iq->store_parent_NS = iq->dp;
2008 iq->chase_flags &= ~BIT_RD; /* go to authorities */
2009 iq->deleg_msg = NULL;
2010 iq->refetch_glue = 1;
2011 iq->query_restart_count++;
2012 iq->sent_count = 0;
2013 iq->dp_target_count = 0;
2014 if(qstate->env->cfg->qname_minimisation)
2015 iq->minimisation_state = INIT_MINIMISE_STATE;
2016 return next_state(iq, INIT_REQUEST_STATE);
2017 }
2018 }
2019 /* see if that makes new names available */
2020 if(!cache_fill_missing(qstate->env, iq->qchase.qclass,
2021 qstate->region, iq->dp))
2022 log_err("out of memory in cache_fill_missing");
2023 if(iq->dp->usable_list) {
2024 verbose(VERB_ALGO, "try parent-side-name, w. glue from cache");
2025 return next_state(iq, QUERYTARGETS_STATE);
2026 }
2027 /* try to fill out parent glue from cache */
2028 if(iter_lookup_parent_glue_from_cache(qstate->env, iq->dp,
2029 qstate->region, &qstate->qinfo)) {
2030 /* got parent stuff from cache, see if we can continue */
2031 verbose(VERB_ALGO, "try parent-side glue from cache");
2032 return next_state(iq, QUERYTARGETS_STATE);
2033 }
2034 /* query for an extra name added by the parent-NS record */
2035 if(delegpt_count_missing_targets(iq->dp) > 0) {
2036 int qs = 0;
2037 verbose(VERB_ALGO, "try parent-side target name");
2038 if(!query_for_targets(qstate, iq, ie, id, 1, &qs)) {
2039 errinf(qstate, "could not fetch nameserver");
2040 errinf_dname(qstate, "at zone", iq->dp->name);
2041 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2042 }
2043 iq->num_target_queries += qs;
2044 target_count_increase(iq, qs);
2045 if(qs != 0) {
2046 qstate->ext_state[id] = module_wait_subquery;
2047 return 0; /* and wait for them */
2048 }
2049 }
2050 if(iq->depth == ie->max_dependency_depth) {
2051 verbose(VERB_QUERY, "maxdepth and need more nameservers, fail");
2052 errinf(qstate, "cannot fetch more nameservers because at max dependency depth");
2053 return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
2054 }
2055 if(iq->depth > 0 && iq->target_count &&
2056 iq->target_count[1] > MAX_TARGET_COUNT) {
2057 char s[LDNS_MAX_DOMAINLEN+1];
2058 dname_str(qstate->qinfo.qname, s);
2059 verbose(VERB_QUERY, "request %s has exceeded the maximum "
2060 "number of glue fetches %d", s, iq->target_count[1]);
2061 errinf(qstate, "exceeded the maximum number of glue fetches");
2062 return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
2063 }
2064 /* mark cycle targets for parent-side lookups */
2065 iter_mark_pside_cycle_targets(qstate, iq->dp);
2066 /* see if we can issue queries to get nameserver addresses */
2067 /* this lookup is not randomized, but sequential. */
2068 for(ns = iq->dp->nslist; ns; ns = ns->next) {
2069 /* if this nameserver is at a delegation point, but that
2070 * delegation point is a stub and we cannot go higher, skip*/
2071 if( ((ie->supports_ipv6 && !ns->done_pside6) ||
2072 (ie->supports_ipv4 && !ns->done_pside4)) &&
2073 !can_have_last_resort(qstate->env, ns->name, ns->namelen,
2074 iq->qchase.qclass, NULL)) {
2075 log_nametypeclass(VERB_ALGO, "cannot pside lookup ns "
2076 "because it is also a stub/forward,",
2077 ns->name, LDNS_RR_TYPE_NS, iq->qchase.qclass);
2078 if(ie->supports_ipv6) ns->done_pside6 = 1;
2079 if(ie->supports_ipv4) ns->done_pside4 = 1;
2080 continue;
2081 }
2082 /* query for parent-side A and AAAA for nameservers */
2083 if(ie->supports_ipv6 && !ns->done_pside6) {
2084 /* Send the AAAA request. */
2085 if(!generate_parentside_target_query(qstate, iq, id,
2086 ns->name, ns->namelen,
2087 LDNS_RR_TYPE_AAAA, iq->qchase.qclass)) {
2088 errinf_dname(qstate, "could not generate nameserver AAAA lookup for", ns->name);
2089 return error_response(qstate, id,
2090 LDNS_RCODE_SERVFAIL);
2091 }
2092 ns->done_pside6 = 1;
2093 query_count++;
2094 }
2095 if(ie->supports_ipv4 && !ns->done_pside4) {
2096 /* Send the A request. */
2097 if(!generate_parentside_target_query(qstate, iq, id,
2098 ns->name, ns->namelen,
2099 LDNS_RR_TYPE_A, iq->qchase.qclass)) {
2100 errinf_dname(qstate, "could not generate nameserver A lookup for", ns->name);
2101 return error_response(qstate, id,
2102 LDNS_RCODE_SERVFAIL);
2103 }
2104 ns->done_pside4 = 1;
2105 query_count++;
2106 }
2107 if(query_count != 0) { /* suspend to await results */
2108 verbose(VERB_ALGO, "try parent-side glue lookup");
2109 iq->num_target_queries += query_count;
2110 target_count_increase(iq, query_count);
2111 qstate->ext_state[id] = module_wait_subquery;
2112 return 0;
2113 }
2114 }
2115
2116 /* if this was a parent-side glue query itself, then store that
2117 * failure in cache. */
2118 if(!qstate->no_cache_store && iq->query_for_pside_glue
2119 && !iq->pside_glue)
2120 iter_store_parentside_neg(qstate->env, &qstate->qinfo,
2121 iq->deleg_msg?iq->deleg_msg->rep:
2122 (iq->response?iq->response->rep:NULL));
2123
2124 errinf(qstate, "all servers for this domain failed,");
2125 errinf_dname(qstate, "at zone", iq->dp->name);
2126 errinf_reply(qstate, iq);
2127 verbose(VERB_QUERY, "out of query targets -- returning SERVFAIL");
2128 /* fail -- no more targets, no more hope of targets, no hope
2129 * of a response. */
2130 return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
2131 }
2132
2133 /**
2134 * Try to find the NS record set that will resolve a qtype DS query. Due
2135 * to grandparent/grandchild reasons we did not get a proper lookup right
2136 * away. We need to create type NS queries until we get the right parent
2137 * for this lookup. We remove labels from the query to find the right point.
2138 * If we end up at the old dp name, then there is no solution.
2139 *
2140 * @param qstate: query state.
2141 * @param iq: iterator query state.
2142 * @param id: module id.
2143 * @return true if the event requires more immediate processing, false if
2144 * not. This is generally only true when forwarding the request to
2145 * the final state (i.e., on answer).
2146 */
2147 static int
processDSNSFind(struct module_qstate * qstate,struct iter_qstate * iq,int id)2148 processDSNSFind(struct module_qstate* qstate, struct iter_qstate* iq, int id)
2149 {
2150 struct module_qstate* subq = NULL;
2151 verbose(VERB_ALGO, "processDSNSFind");
2152
2153 if(!iq->dsns_point) {
2154 /* initialize */
2155 iq->dsns_point = iq->qchase.qname;
2156 iq->dsns_point_len = iq->qchase.qname_len;
2157 }
2158 /* robustcheck for internal error: we are not underneath the dp */
2159 if(!dname_subdomain_c(iq->dsns_point, iq->dp->name)) {
2160 errinf_dname(qstate, "for DS query parent-child nameserver search the query is not under the zone", iq->dp->name);
2161 return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
2162 }
2163
2164 /* go up one (more) step, until we hit the dp, if so, end */
2165 dname_remove_label(&iq->dsns_point, &iq->dsns_point_len);
2166 if(query_dname_compare(iq->dsns_point, iq->dp->name) == 0) {
2167 /* there was no inbetween nameserver, use the old delegation
2168 * point again. And this time, because dsns_point is nonNULL
2169 * we are going to accept the (bad) result */
2170 iq->state = QUERYTARGETS_STATE;
2171 return 1;
2172 }
2173 iq->state = DSNS_FIND_STATE;
2174
2175 /* spawn NS lookup (validation not needed, this is for DS lookup) */
2176 log_nametypeclass(VERB_ALGO, "fetch nameservers",
2177 iq->dsns_point, LDNS_RR_TYPE_NS, iq->qchase.qclass);
2178 if(!generate_sub_request(iq->dsns_point, iq->dsns_point_len,
2179 LDNS_RR_TYPE_NS, iq->qchase.qclass, qstate, id, iq,
2180 INIT_REQUEST_STATE, FINISHED_STATE, &subq, 0, 0)) {
2181 errinf_dname(qstate, "for DS query parent-child nameserver search, could not generate NS lookup for", iq->dsns_point);
2182 return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
2183 }
2184
2185 return 0;
2186 }
2187
2188 /**
2189 * This is the request event state where the request will be sent to one of
2190 * its current query targets. This state also handles issuing target lookup
2191 * queries for missing target IP addresses. Queries typically iterate on
2192 * this state, both when they are just trying different targets for a given
2193 * delegation point, and when they change delegation points. This state
2194 * roughly corresponds to RFC 1034 algorithm steps 3 and 4.
2195 *
2196 * @param qstate: query state.
2197 * @param iq: iterator query state.
2198 * @param ie: iterator shared global environment.
2199 * @param id: module id.
2200 * @return true if the event requires more request processing immediately,
2201 * false if not. This state only returns true when it is generating
2202 * a SERVFAIL response because the query has hit a dead end.
2203 */
2204 static int
processQueryTargets(struct module_qstate * qstate,struct iter_qstate * iq,struct iter_env * ie,int id)2205 processQueryTargets(struct module_qstate* qstate, struct iter_qstate* iq,
2206 struct iter_env* ie, int id)
2207 {
2208 int tf_policy;
2209 struct delegpt_addr* target;
2210 struct outbound_entry* outq;
2211 int auth_fallback = 0;
2212 uint8_t* qout_orig = NULL;
2213 size_t qout_orig_len = 0;
2214
2215 /* NOTE: a request will encounter this state for each target it
2216 * needs to send a query to. That is, at least one per referral,
2217 * more if some targets timeout or return throwaway answers. */
2218
2219 log_query_info(VERB_QUERY, "processQueryTargets:", &qstate->qinfo);
2220 verbose(VERB_ALGO, "processQueryTargets: targetqueries %d, "
2221 "currentqueries %d sentcount %d", iq->num_target_queries,
2222 iq->num_current_queries, iq->sent_count);
2223
2224 /* Make sure that we haven't run away */
2225 /* FIXME: is this check even necessary? */
2226 if(iq->referral_count > MAX_REFERRAL_COUNT) {
2227 verbose(VERB_QUERY, "request has exceeded the maximum "
2228 "number of referrrals with %d", iq->referral_count);
2229 errinf(qstate, "exceeded the maximum of referrals");
2230 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2231 }
2232 if(iq->sent_count > MAX_SENT_COUNT) {
2233 verbose(VERB_QUERY, "request has exceeded the maximum "
2234 "number of sends with %d", iq->sent_count);
2235 errinf(qstate, "exceeded the maximum number of sends");
2236 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2237 }
2238 if(iq->target_count && iq->target_count[2] > MAX_TARGET_NX) {
2239 verbose(VERB_QUERY, "request has exceeded the maximum "
2240 " number of nxdomain nameserver lookups with %d",
2241 iq->target_count[2]);
2242 errinf(qstate, "exceeded the maximum nameserver nxdomains");
2243 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2244 }
2245
2246 /* Make sure we have a delegation point, otherwise priming failed
2247 * or another failure occurred */
2248 if(!iq->dp) {
2249 verbose(VERB_QUERY, "Failed to get a delegation, giving up");
2250 errinf(qstate, "failed to get a delegation (eg. prime failure)");
2251 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2252 }
2253 if(!ie->supports_ipv6)
2254 delegpt_no_ipv6(iq->dp);
2255 if(!ie->supports_ipv4)
2256 delegpt_no_ipv4(iq->dp);
2257 delegpt_log(VERB_ALGO, iq->dp);
2258
2259 if(iq->num_current_queries>0) {
2260 /* already busy answering a query, this restart is because
2261 * more delegpt addrs became available, wait for existing
2262 * query. */
2263 verbose(VERB_ALGO, "woke up, but wait for outstanding query");
2264 qstate->ext_state[id] = module_wait_reply;
2265 return 0;
2266 }
2267
2268 if(iq->minimisation_state == INIT_MINIMISE_STATE
2269 && !(iq->chase_flags & BIT_RD)) {
2270 /* (Re)set qinfo_out to (new) delegation point, except when
2271 * qinfo_out is already a subdomain of dp. This happens when
2272 * increasing by more than one label at once (QNAMEs with more
2273 * than MAX_MINIMISE_COUNT labels). */
2274 if(!(iq->qinfo_out.qname_len
2275 && dname_subdomain_c(iq->qchase.qname,
2276 iq->qinfo_out.qname)
2277 && dname_subdomain_c(iq->qinfo_out.qname,
2278 iq->dp->name))) {
2279 iq->qinfo_out.qname = iq->dp->name;
2280 iq->qinfo_out.qname_len = iq->dp->namelen;
2281 iq->qinfo_out.qtype = LDNS_RR_TYPE_A;
2282 iq->qinfo_out.qclass = iq->qchase.qclass;
2283 iq->qinfo_out.local_alias = NULL;
2284 iq->minimise_count = 0;
2285 }
2286
2287 iq->minimisation_state = MINIMISE_STATE;
2288 }
2289 if(iq->minimisation_state == MINIMISE_STATE) {
2290 int qchaselabs = dname_count_labels(iq->qchase.qname);
2291 int labdiff = qchaselabs -
2292 dname_count_labels(iq->qinfo_out.qname);
2293
2294 qout_orig = iq->qinfo_out.qname;
2295 qout_orig_len = iq->qinfo_out.qname_len;
2296 iq->qinfo_out.qname = iq->qchase.qname;
2297 iq->qinfo_out.qname_len = iq->qchase.qname_len;
2298 iq->minimise_count++;
2299 iq->timeout_count = 0;
2300
2301 iter_dec_attempts(iq->dp, 1, ie->outbound_msg_retry);
2302
2303 /* Limit number of iterations for QNAMEs with more
2304 * than MAX_MINIMISE_COUNT labels. Send first MINIMISE_ONE_LAB
2305 * labels of QNAME always individually.
2306 */
2307 if(qchaselabs > MAX_MINIMISE_COUNT && labdiff > 1 &&
2308 iq->minimise_count > MINIMISE_ONE_LAB) {
2309 if(iq->minimise_count < MAX_MINIMISE_COUNT) {
2310 int multilabs = qchaselabs - 1 -
2311 MINIMISE_ONE_LAB;
2312 int extralabs = multilabs /
2313 MINIMISE_MULTIPLE_LABS;
2314
2315 if (MAX_MINIMISE_COUNT - iq->minimise_count >=
2316 multilabs % MINIMISE_MULTIPLE_LABS)
2317 /* Default behaviour is to add 1 label
2318 * every iteration. Therefore, decrement
2319 * the extralabs by 1 */
2320 extralabs--;
2321 if (extralabs < labdiff)
2322 labdiff -= extralabs;
2323 else
2324 labdiff = 1;
2325 }
2326 /* Last minimised iteration, send all labels with
2327 * QTYPE=NS */
2328 else
2329 labdiff = 1;
2330 }
2331
2332 if(labdiff > 1) {
2333 verbose(VERB_QUERY, "removing %d labels", labdiff-1);
2334 dname_remove_labels(&iq->qinfo_out.qname,
2335 &iq->qinfo_out.qname_len,
2336 labdiff-1);
2337 }
2338 if(labdiff < 1 || (labdiff < 2
2339 && (iq->qchase.qtype == LDNS_RR_TYPE_DS
2340 || iq->qchase.qtype == LDNS_RR_TYPE_A)))
2341 /* Stop minimising this query, resolve "as usual" */
2342 iq->minimisation_state = DONOT_MINIMISE_STATE;
2343 else if(!qstate->no_cache_lookup) {
2344 struct dns_msg* msg = dns_cache_lookup(qstate->env,
2345 iq->qinfo_out.qname, iq->qinfo_out.qname_len,
2346 iq->qinfo_out.qtype, iq->qinfo_out.qclass,
2347 qstate->query_flags, qstate->region,
2348 qstate->env->scratch, 0, iq->dp->name,
2349 iq->dp->namelen);
2350 if(msg && FLAGS_GET_RCODE(msg->rep->flags) ==
2351 LDNS_RCODE_NOERROR)
2352 /* no need to send query if it is already
2353 * cached as NOERROR */
2354 return 1;
2355 if(msg && FLAGS_GET_RCODE(msg->rep->flags) ==
2356 LDNS_RCODE_NXDOMAIN &&
2357 qstate->env->need_to_validate &&
2358 qstate->env->cfg->harden_below_nxdomain) {
2359 if(msg->rep->security == sec_status_secure) {
2360 iq->response = msg;
2361 return final_state(iq);
2362 }
2363 if(msg->rep->security == sec_status_unchecked) {
2364 struct module_qstate* subq = NULL;
2365 if(!generate_sub_request(
2366 iq->qinfo_out.qname,
2367 iq->qinfo_out.qname_len,
2368 iq->qinfo_out.qtype,
2369 iq->qinfo_out.qclass,
2370 qstate, id, iq,
2371 INIT_REQUEST_STATE,
2372 FINISHED_STATE, &subq, 1, 1))
2373 verbose(VERB_ALGO,
2374 "could not validate NXDOMAIN "
2375 "response");
2376 }
2377 }
2378 if(msg && FLAGS_GET_RCODE(msg->rep->flags) ==
2379 LDNS_RCODE_NXDOMAIN) {
2380 /* return and add a label in the next
2381 * minimisation iteration.
2382 */
2383 return 1;
2384 }
2385 }
2386 }
2387 if(iq->minimisation_state == SKIP_MINIMISE_STATE) {
2388 if(iq->timeout_count < MAX_MINIMISE_TIMEOUT_COUNT)
2389 /* Do not increment qname, continue incrementing next
2390 * iteration */
2391 iq->minimisation_state = MINIMISE_STATE;
2392 else if(!qstate->env->cfg->qname_minimisation_strict)
2393 /* Too many time-outs detected for this QNAME and QTYPE.
2394 * We give up, disable QNAME minimisation. */
2395 iq->minimisation_state = DONOT_MINIMISE_STATE;
2396 }
2397 if(iq->minimisation_state == DONOT_MINIMISE_STATE)
2398 iq->qinfo_out = iq->qchase;
2399
2400 /* now find an answer to this query */
2401 /* see if authority zones have an answer */
2402 /* now we know the dp, we can check the auth zone for locally hosted
2403 * contents */
2404 if(!iq->auth_zone_avoid && qstate->blacklist) {
2405 if(auth_zones_can_fallback(qstate->env->auth_zones,
2406 iq->dp->name, iq->dp->namelen, iq->qinfo_out.qclass)) {
2407 /* if cache is blacklisted and this zone allows us
2408 * to fallback to the internet, then do so, and
2409 * fetch results from the internet servers */
2410 iq->auth_zone_avoid = 1;
2411 }
2412 }
2413 if(iq->auth_zone_avoid) {
2414 iq->auth_zone_avoid = 0;
2415 auth_fallback = 1;
2416 } else if(auth_zones_lookup(qstate->env->auth_zones, &iq->qinfo_out,
2417 qstate->region, &iq->response, &auth_fallback, iq->dp->name,
2418 iq->dp->namelen)) {
2419 /* use this as a response to be processed by the iterator */
2420 if(verbosity >= VERB_ALGO) {
2421 log_dns_msg("msg from auth zone",
2422 &iq->response->qinfo, iq->response->rep);
2423 }
2424 if((iq->chase_flags&BIT_RD) && !(iq->response->rep->flags&BIT_AA)) {
2425 verbose(VERB_ALGO, "forwarder, ignoring referral from auth zone");
2426 } else {
2427 lock_rw_wrlock(&qstate->env->auth_zones->lock);
2428 qstate->env->auth_zones->num_query_up++;
2429 lock_rw_unlock(&qstate->env->auth_zones->lock);
2430 iq->num_current_queries++;
2431 iq->chase_to_rd = 0;
2432 iq->dnssec_lame_query = 0;
2433 iq->auth_zone_response = 1;
2434 return next_state(iq, QUERY_RESP_STATE);
2435 }
2436 }
2437 iq->auth_zone_response = 0;
2438 if(auth_fallback == 0) {
2439 /* like we got servfail from the auth zone lookup, and
2440 * no internet fallback */
2441 verbose(VERB_ALGO, "auth zone lookup failed, no fallback,"
2442 " servfail");
2443 errinf(qstate, "auth zone lookup failed, fallback is off");
2444 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2445 }
2446 if(iq->dp->auth_dp) {
2447 /* we wanted to fallback, but had no delegpt, only the
2448 * auth zone generated delegpt, create an actual one */
2449 iq->auth_zone_avoid = 1;
2450 return next_state(iq, INIT_REQUEST_STATE);
2451 }
2452 /* but mostly, fallback==1 (like, when no such auth zone exists)
2453 * and we continue with lookups */
2454
2455 tf_policy = 0;
2456 /* < not <=, because although the array is large enough for <=, the
2457 * generated query will immediately be discarded due to depth and
2458 * that servfail is cached, which is not good as opportunism goes. */
2459 if(iq->depth < ie->max_dependency_depth
2460 && iq->num_target_queries == 0
2461 && (!iq->target_count || iq->target_count[2]==0)
2462 && iq->sent_count < TARGET_FETCH_STOP) {
2463 tf_policy = ie->target_fetch_policy[iq->depth];
2464 }
2465
2466 /* if in 0x20 fallback get as many targets as possible */
2467 if(iq->caps_fallback) {
2468 int extra = 0;
2469 size_t naddr, nres, navail;
2470 if(!query_for_targets(qstate, iq, ie, id, -1, &extra)) {
2471 errinf(qstate, "could not fetch nameservers for 0x20 fallback");
2472 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2473 }
2474 iq->num_target_queries += extra;
2475 target_count_increase(iq, extra);
2476 if(iq->num_target_queries > 0) {
2477 /* wait to get all targets, we want to try em */
2478 verbose(VERB_ALGO, "wait for all targets for fallback");
2479 qstate->ext_state[id] = module_wait_reply;
2480 /* undo qname minimise step because we'll get back here
2481 * to do it again */
2482 if(qout_orig && iq->minimise_count > 0) {
2483 iq->minimise_count--;
2484 iq->qinfo_out.qname = qout_orig;
2485 iq->qinfo_out.qname_len = qout_orig_len;
2486 }
2487 return 0;
2488 }
2489 /* did we do enough fallback queries already? */
2490 delegpt_count_addr(iq->dp, &naddr, &nres, &navail);
2491 /* the current caps_server is the number of fallbacks sent.
2492 * the original query is one that matched too, so we have
2493 * caps_server+1 number of matching queries now */
2494 if(iq->caps_server+1 >= naddr*3 ||
2495 iq->caps_server*2+2 >= MAX_SENT_COUNT) {
2496 /* *2 on sentcount check because ipv6 may fail */
2497 /* we're done, process the response */
2498 verbose(VERB_ALGO, "0x20 fallback had %d responses "
2499 "match for %d wanted, done.",
2500 (int)iq->caps_server+1, (int)naddr*3);
2501 iq->response = iq->caps_response;
2502 iq->caps_fallback = 0;
2503 iter_dec_attempts(iq->dp, 3, ie->outbound_msg_retry); /* space for fallback */
2504 iq->num_current_queries++; /* RespState decrements it*/
2505 iq->referral_count++; /* make sure we don't loop */
2506 iq->sent_count = 0;
2507 iq->dp_target_count = 0;
2508 iq->state = QUERY_RESP_STATE;
2509 return 1;
2510 }
2511 verbose(VERB_ALGO, "0x20 fallback number %d",
2512 (int)iq->caps_server);
2513
2514 /* if there is a policy to fetch missing targets
2515 * opportunistically, do it. we rely on the fact that once a
2516 * query (or queries) for a missing name have been issued,
2517 * they will not show up again. */
2518 } else if(tf_policy != 0) {
2519 int extra = 0;
2520 verbose(VERB_ALGO, "attempt to get extra %d targets",
2521 tf_policy);
2522 (void)query_for_targets(qstate, iq, ie, id, tf_policy, &extra);
2523 /* errors ignored, these targets are not strictly necessary for
2524 * this result, we do not have to reply with SERVFAIL */
2525 iq->num_target_queries += extra;
2526 target_count_increase(iq, extra);
2527 }
2528
2529 /* Add the current set of unused targets to our queue. */
2530 delegpt_add_unused_targets(iq->dp);
2531
2532 if(qstate->env->auth_zones) {
2533 /* apply rpz triggers at query time */
2534 struct dns_msg* forged_response = rpz_callback_from_iterator_module(qstate, iq);
2535 if(forged_response != NULL) {
2536 qstate->ext_state[id] = module_finished;
2537 qstate->return_rcode = FLAGS_GET_RCODE(forged_response->rep->flags);
2538 qstate->return_msg = forged_response;
2539 iq->response = forged_response;
2540 next_state(iq, FINISHED_STATE);
2541 if(!iter_prepend(iq, qstate->return_msg, qstate->region)) {
2542 log_err("rpz, prepend rrsets: out of memory");
2543 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2544 }
2545 return 0;
2546 }
2547 }
2548
2549 /* Select the next usable target, filtering out unsuitable targets. */
2550 target = iter_server_selection(ie, qstate->env, iq->dp,
2551 iq->dp->name, iq->dp->namelen, iq->qchase.qtype,
2552 &iq->dnssec_lame_query, &iq->chase_to_rd,
2553 iq->num_target_queries, qstate->blacklist,
2554 qstate->prefetch_leeway);
2555
2556 /* If no usable target was selected... */
2557 if(!target) {
2558 /* Here we distinguish between three states: generate a new
2559 * target query, just wait, or quit (with a SERVFAIL).
2560 * We have the following information: number of active
2561 * target queries, number of active current queries,
2562 * the presence of missing targets at this delegation
2563 * point, and the given query target policy. */
2564
2565 /* Check for the wait condition. If this is true, then
2566 * an action must be taken. */
2567 if(iq->num_target_queries==0 && iq->num_current_queries==0) {
2568 /* If there is nothing to wait for, then we need
2569 * to distinguish between generating (a) new target
2570 * query, or failing. */
2571 if(delegpt_count_missing_targets(iq->dp) > 0) {
2572 int qs = 0;
2573 verbose(VERB_ALGO, "querying for next "
2574 "missing target");
2575 if(!query_for_targets(qstate, iq, ie, id,
2576 1, &qs)) {
2577 errinf(qstate, "could not fetch nameserver");
2578 errinf_dname(qstate, "at zone", iq->dp->name);
2579 return error_response(qstate, id,
2580 LDNS_RCODE_SERVFAIL);
2581 }
2582 if(qs == 0 &&
2583 delegpt_count_missing_targets(iq->dp) == 0){
2584 /* it looked like there were missing
2585 * targets, but they did not turn up.
2586 * Try the bad choices again (if any),
2587 * when we get back here missing==0,
2588 * so this is not a loop. */
2589 return 1;
2590 }
2591 iq->num_target_queries += qs;
2592 target_count_increase(iq, qs);
2593 }
2594 /* Since a target query might have been made, we
2595 * need to check again. */
2596 if(iq->num_target_queries == 0) {
2597 /* if in capsforid fallback, instead of last
2598 * resort, we agree with the current reply
2599 * we have (if any) (our count of addrs bad)*/
2600 if(iq->caps_fallback && iq->caps_reply) {
2601 /* we're done, process the response */
2602 verbose(VERB_ALGO, "0x20 fallback had %d responses, "
2603 "but no more servers except "
2604 "last resort, done.",
2605 (int)iq->caps_server+1);
2606 iq->response = iq->caps_response;
2607 iq->caps_fallback = 0;
2608 iter_dec_attempts(iq->dp, 3, ie->outbound_msg_retry); /* space for fallback */
2609 iq->num_current_queries++; /* RespState decrements it*/
2610 iq->referral_count++; /* make sure we don't loop */
2611 iq->sent_count = 0;
2612 iq->dp_target_count = 0;
2613 iq->state = QUERY_RESP_STATE;
2614 return 1;
2615 }
2616 return processLastResort(qstate, iq, ie, id);
2617 }
2618 }
2619
2620 /* otherwise, we have no current targets, so submerge
2621 * until one of the target or direct queries return. */
2622 if(iq->num_target_queries>0 && iq->num_current_queries>0) {
2623 verbose(VERB_ALGO, "no current targets -- waiting "
2624 "for %d targets to resolve or %d outstanding"
2625 " queries to respond", iq->num_target_queries,
2626 iq->num_current_queries);
2627 qstate->ext_state[id] = module_wait_reply;
2628 } else if(iq->num_target_queries>0) {
2629 verbose(VERB_ALGO, "no current targets -- waiting "
2630 "for %d targets to resolve.",
2631 iq->num_target_queries);
2632 qstate->ext_state[id] = module_wait_subquery;
2633 } else {
2634 verbose(VERB_ALGO, "no current targets -- waiting "
2635 "for %d outstanding queries to respond.",
2636 iq->num_current_queries);
2637 qstate->ext_state[id] = module_wait_reply;
2638 }
2639 /* undo qname minimise step because we'll get back here
2640 * to do it again */
2641 if(qout_orig && iq->minimise_count > 0) {
2642 iq->minimise_count--;
2643 iq->qinfo_out.qname = qout_orig;
2644 iq->qinfo_out.qname_len = qout_orig_len;
2645 }
2646 return 0;
2647 }
2648
2649 /* if not forwarding, check ratelimits per delegationpoint name */
2650 if(!(iq->chase_flags & BIT_RD) && !iq->ratelimit_ok) {
2651 if(!infra_ratelimit_inc(qstate->env->infra_cache, iq->dp->name,
2652 iq->dp->namelen, *qstate->env->now, &qstate->qinfo,
2653 qstate->reply)) {
2654 lock_basic_lock(&ie->queries_ratelimit_lock);
2655 ie->num_queries_ratelimited++;
2656 lock_basic_unlock(&ie->queries_ratelimit_lock);
2657 verbose(VERB_ALGO, "query exceeded ratelimits");
2658 qstate->was_ratelimited = 1;
2659 errinf_dname(qstate, "exceeded ratelimit for zone",
2660 iq->dp->name);
2661 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2662 }
2663 }
2664
2665 /* We have a valid target. */
2666 if(verbosity >= VERB_QUERY) {
2667 log_query_info(VERB_QUERY, "sending query:", &iq->qinfo_out);
2668 log_name_addr(VERB_QUERY, "sending to target:", iq->dp->name,
2669 &target->addr, target->addrlen);
2670 verbose(VERB_ALGO, "dnssec status: %s%s",
2671 iq->dnssec_expected?"expected": "not expected",
2672 iq->dnssec_lame_query?" but lame_query anyway": "");
2673 }
2674 fptr_ok(fptr_whitelist_modenv_send_query(qstate->env->send_query));
2675 outq = (*qstate->env->send_query)(&iq->qinfo_out,
2676 iq->chase_flags | (iq->chase_to_rd?BIT_RD:0),
2677 /* unset CD if to forwarder(RD set) and not dnssec retry
2678 * (blacklist nonempty) and no trust-anchors are configured
2679 * above the qname or on the first attempt when dnssec is on */
2680 EDNS_DO| ((iq->chase_to_rd||(iq->chase_flags&BIT_RD)!=0)&&
2681 !qstate->blacklist&&(!iter_qname_indicates_dnssec(qstate->env,
2682 &iq->qinfo_out)||target->attempts==1)?0:BIT_CD),
2683 iq->dnssec_expected, iq->caps_fallback || is_caps_whitelisted(
2684 ie, iq), &target->addr, target->addrlen,
2685 iq->dp->name, iq->dp->namelen,
2686 (iq->dp->tcp_upstream || qstate->env->cfg->tcp_upstream),
2687 (iq->dp->ssl_upstream || qstate->env->cfg->ssl_upstream),
2688 target->tls_auth_name, qstate);
2689 if(!outq) {
2690 log_addr(VERB_QUERY, "error sending query to auth server",
2691 &target->addr, target->addrlen);
2692 if(!(iq->chase_flags & BIT_RD) && !iq->ratelimit_ok)
2693 infra_ratelimit_dec(qstate->env->infra_cache, iq->dp->name,
2694 iq->dp->namelen, *qstate->env->now);
2695 if(qstate->env->cfg->qname_minimisation)
2696 iq->minimisation_state = SKIP_MINIMISE_STATE;
2697 return next_state(iq, QUERYTARGETS_STATE);
2698 }
2699 outbound_list_insert(&iq->outlist, outq);
2700 iq->num_current_queries++;
2701 iq->sent_count++;
2702 qstate->ext_state[id] = module_wait_reply;
2703
2704 return 0;
2705 }
2706
2707 /** find NS rrset in given list */
2708 static struct ub_packed_rrset_key*
find_NS(struct reply_info * rep,size_t from,size_t to)2709 find_NS(struct reply_info* rep, size_t from, size_t to)
2710 {
2711 size_t i;
2712 for(i=from; i<to; i++) {
2713 if(ntohs(rep->rrsets[i]->rk.type) == LDNS_RR_TYPE_NS)
2714 return rep->rrsets[i];
2715 }
2716 return NULL;
2717 }
2718
2719
2720 /**
2721 * Process the query response. All queries end up at this state first. This
2722 * process generally consists of analyzing the response and routing the
2723 * event to the next state (either bouncing it back to a request state, or
2724 * terminating the processing for this event).
2725 *
2726 * @param qstate: query state.
2727 * @param iq: iterator query state.
2728 * @param ie: iterator shared global environment.
2729 * @param id: module id.
2730 * @return true if the event requires more immediate processing, false if
2731 * not. This is generally only true when forwarding the request to
2732 * the final state (i.e., on answer).
2733 */
2734 static int
processQueryResponse(struct module_qstate * qstate,struct iter_qstate * iq,struct iter_env * ie,int id)2735 processQueryResponse(struct module_qstate* qstate, struct iter_qstate* iq,
2736 struct iter_env* ie, int id)
2737 {
2738 int dnsseclame = 0;
2739 enum response_type type;
2740
2741 iq->num_current_queries--;
2742
2743 if(!inplace_cb_query_response_call(qstate->env, qstate, iq->response))
2744 log_err("unable to call query_response callback");
2745
2746 if(iq->response == NULL) {
2747 /* Don't increment qname when QNAME minimisation is enabled */
2748 if(qstate->env->cfg->qname_minimisation) {
2749 iq->minimisation_state = SKIP_MINIMISE_STATE;
2750 }
2751 iq->timeout_count++;
2752 iq->chase_to_rd = 0;
2753 iq->dnssec_lame_query = 0;
2754 verbose(VERB_ALGO, "query response was timeout");
2755 return next_state(iq, QUERYTARGETS_STATE);
2756 }
2757 iq->timeout_count = 0;
2758 type = response_type_from_server(
2759 (int)((iq->chase_flags&BIT_RD) || iq->chase_to_rd),
2760 iq->response, &iq->qinfo_out, iq->dp);
2761 iq->chase_to_rd = 0;
2762 if(type == RESPONSE_TYPE_REFERRAL && (iq->chase_flags&BIT_RD) &&
2763 !iq->auth_zone_response) {
2764 /* When forwarding (RD bit is set), we handle referrals
2765 * differently. No queries should be sent elsewhere */
2766 type = RESPONSE_TYPE_ANSWER;
2767 }
2768 if(!qstate->env->cfg->disable_dnssec_lame_check && iq->dnssec_expected
2769 && !iq->dnssec_lame_query &&
2770 !(iq->chase_flags&BIT_RD)
2771 && iq->sent_count < DNSSEC_LAME_DETECT_COUNT
2772 && type != RESPONSE_TYPE_LAME
2773 && type != RESPONSE_TYPE_REC_LAME
2774 && type != RESPONSE_TYPE_THROWAWAY
2775 && type != RESPONSE_TYPE_UNTYPED) {
2776 /* a possible answer, see if it is missing DNSSEC */
2777 /* but not when forwarding, so we dont mark fwder lame */
2778 if(!iter_msg_has_dnssec(iq->response)) {
2779 /* Mark this address as dnsseclame in this dp,
2780 * because that will make serverselection disprefer
2781 * it, but also, once it is the only final option,
2782 * use dnssec-lame-bypass if it needs to query there.*/
2783 if(qstate->reply) {
2784 struct delegpt_addr* a = delegpt_find_addr(
2785 iq->dp, &qstate->reply->addr,
2786 qstate->reply->addrlen);
2787 if(a) a->dnsseclame = 1;
2788 }
2789 /* test the answer is from the zone we expected,
2790 * otherwise, (due to parent,child on same server), we
2791 * might mark the server,zone lame inappropriately */
2792 if(!iter_msg_from_zone(iq->response, iq->dp, type,
2793 iq->qchase.qclass))
2794 qstate->reply = NULL;
2795 type = RESPONSE_TYPE_LAME;
2796 dnsseclame = 1;
2797 }
2798 } else iq->dnssec_lame_query = 0;
2799 /* see if referral brings us close to the target */
2800 if(type == RESPONSE_TYPE_REFERRAL) {
2801 struct ub_packed_rrset_key* ns = find_NS(
2802 iq->response->rep, iq->response->rep->an_numrrsets,
2803 iq->response->rep->an_numrrsets
2804 + iq->response->rep->ns_numrrsets);
2805 if(!ns) ns = find_NS(iq->response->rep, 0,
2806 iq->response->rep->an_numrrsets);
2807 if(!ns || !dname_strict_subdomain_c(ns->rk.dname, iq->dp->name)
2808 || !dname_subdomain_c(iq->qchase.qname, ns->rk.dname)){
2809 verbose(VERB_ALGO, "bad referral, throwaway");
2810 type = RESPONSE_TYPE_THROWAWAY;
2811 } else
2812 iter_scrub_ds(iq->response, ns, iq->dp->name);
2813 } else iter_scrub_ds(iq->response, NULL, NULL);
2814 if(type == RESPONSE_TYPE_THROWAWAY &&
2815 FLAGS_GET_RCODE(iq->response->rep->flags) == LDNS_RCODE_YXDOMAIN) {
2816 /* YXDOMAIN is a permanent error, no need to retry */
2817 type = RESPONSE_TYPE_ANSWER;
2818 }
2819 if(type == RESPONSE_TYPE_CNAME && iq->response->rep->an_numrrsets >= 1
2820 && ntohs(iq->response->rep->rrsets[0]->rk.type) == LDNS_RR_TYPE_DNAME) {
2821 uint8_t* sname = NULL;
2822 size_t snamelen = 0;
2823 get_cname_target(iq->response->rep->rrsets[0], &sname,
2824 &snamelen);
2825 if(snamelen && dname_subdomain_c(sname, iq->response->rep->rrsets[0]->rk.dname)) {
2826 /* DNAME to a subdomain loop; do not recurse */
2827 type = RESPONSE_TYPE_ANSWER;
2828 }
2829 } else if(type == RESPONSE_TYPE_CNAME &&
2830 iq->qchase.qtype == LDNS_RR_TYPE_CNAME &&
2831 iq->minimisation_state == MINIMISE_STATE &&
2832 query_dname_compare(iq->qchase.qname, iq->qinfo_out.qname) == 0) {
2833 /* The minimised query for full QTYPE and hidden QTYPE can be
2834 * classified as CNAME response type, even when the original
2835 * QTYPE=CNAME. This should be treated as answer response type.
2836 */
2837 type = RESPONSE_TYPE_ANSWER;
2838 }
2839
2840 /* handle each of the type cases */
2841 if(type == RESPONSE_TYPE_ANSWER) {
2842 /* ANSWER type responses terminate the query algorithm,
2843 * so they sent on their */
2844 if(verbosity >= VERB_DETAIL) {
2845 verbose(VERB_DETAIL, "query response was %s",
2846 FLAGS_GET_RCODE(iq->response->rep->flags)
2847 ==LDNS_RCODE_NXDOMAIN?"NXDOMAIN ANSWER":
2848 (iq->response->rep->an_numrrsets?"ANSWER":
2849 "nodata ANSWER"));
2850 }
2851 /* if qtype is DS, check we have the right level of answer,
2852 * like grandchild answer but we need the middle, reject it */
2853 if(iq->qchase.qtype == LDNS_RR_TYPE_DS && !iq->dsns_point
2854 && !(iq->chase_flags&BIT_RD)
2855 && iter_ds_toolow(iq->response, iq->dp)
2856 && iter_dp_cangodown(&iq->qchase, iq->dp)) {
2857 /* close down outstanding requests to be discarded */
2858 outbound_list_clear(&iq->outlist);
2859 iq->num_current_queries = 0;
2860 fptr_ok(fptr_whitelist_modenv_detach_subs(
2861 qstate->env->detach_subs));
2862 (*qstate->env->detach_subs)(qstate);
2863 iq->num_target_queries = 0;
2864 return processDSNSFind(qstate, iq, id);
2865 }
2866 if(!qstate->no_cache_store)
2867 iter_dns_store(qstate->env, &iq->response->qinfo,
2868 iq->response->rep, 0, qstate->prefetch_leeway,
2869 iq->dp&&iq->dp->has_parent_side_NS,
2870 qstate->region, qstate->query_flags);
2871 /* close down outstanding requests to be discarded */
2872 outbound_list_clear(&iq->outlist);
2873 iq->num_current_queries = 0;
2874 fptr_ok(fptr_whitelist_modenv_detach_subs(
2875 qstate->env->detach_subs));
2876 (*qstate->env->detach_subs)(qstate);
2877 iq->num_target_queries = 0;
2878 if(qstate->reply)
2879 sock_list_insert(&qstate->reply_origin,
2880 &qstate->reply->addr, qstate->reply->addrlen,
2881 qstate->region);
2882 if(iq->minimisation_state != DONOT_MINIMISE_STATE
2883 && !(iq->chase_flags & BIT_RD)) {
2884 if(FLAGS_GET_RCODE(iq->response->rep->flags) !=
2885 LDNS_RCODE_NOERROR) {
2886 if(qstate->env->cfg->qname_minimisation_strict) {
2887 if(FLAGS_GET_RCODE(iq->response->rep->flags) ==
2888 LDNS_RCODE_NXDOMAIN) {
2889 iter_scrub_nxdomain(iq->response);
2890 return final_state(iq);
2891 }
2892 return error_response(qstate, id,
2893 LDNS_RCODE_SERVFAIL);
2894 }
2895 /* Best effort qname-minimisation.
2896 * Stop minimising and send full query when
2897 * RCODE is not NOERROR. */
2898 iq->minimisation_state = DONOT_MINIMISE_STATE;
2899 }
2900 if(FLAGS_GET_RCODE(iq->response->rep->flags) ==
2901 LDNS_RCODE_NXDOMAIN) {
2902 /* Stop resolving when NXDOMAIN is DNSSEC
2903 * signed. Based on assumption that nameservers
2904 * serving signed zones do not return NXDOMAIN
2905 * for empty-non-terminals. */
2906 if(iq->dnssec_expected)
2907 return final_state(iq);
2908 /* Make subrequest to validate intermediate
2909 * NXDOMAIN if harden-below-nxdomain is
2910 * enabled. */
2911 if(qstate->env->cfg->harden_below_nxdomain &&
2912 qstate->env->need_to_validate) {
2913 struct module_qstate* subq = NULL;
2914 log_query_info(VERB_QUERY,
2915 "schedule NXDOMAIN validation:",
2916 &iq->response->qinfo);
2917 if(!generate_sub_request(
2918 iq->response->qinfo.qname,
2919 iq->response->qinfo.qname_len,
2920 iq->response->qinfo.qtype,
2921 iq->response->qinfo.qclass,
2922 qstate, id, iq,
2923 INIT_REQUEST_STATE,
2924 FINISHED_STATE, &subq, 1, 1))
2925 verbose(VERB_ALGO,
2926 "could not validate NXDOMAIN "
2927 "response");
2928 }
2929 }
2930 return next_state(iq, QUERYTARGETS_STATE);
2931 }
2932 return final_state(iq);
2933 } else if(type == RESPONSE_TYPE_REFERRAL) {
2934 /* REFERRAL type responses get a reset of the
2935 * delegation point, and back to the QUERYTARGETS_STATE. */
2936 verbose(VERB_DETAIL, "query response was REFERRAL");
2937
2938 if(!(iq->chase_flags & BIT_RD) && !iq->ratelimit_ok) {
2939 /* we have a referral, no ratelimit, we can send
2940 * our queries to the given name */
2941 infra_ratelimit_dec(qstate->env->infra_cache,
2942 iq->dp->name, iq->dp->namelen,
2943 *qstate->env->now);
2944 }
2945
2946 /* if hardened, only store referral if we asked for it */
2947 if(!qstate->no_cache_store &&
2948 (!qstate->env->cfg->harden_referral_path ||
2949 ( qstate->qinfo.qtype == LDNS_RR_TYPE_NS
2950 && (qstate->query_flags&BIT_RD)
2951 && !(qstate->query_flags&BIT_CD)
2952 /* we know that all other NS rrsets are scrubbed
2953 * away, thus on referral only one is left.
2954 * see if that equals the query name... */
2955 && ( /* auth section, but sometimes in answer section*/
2956 reply_find_rrset_section_ns(iq->response->rep,
2957 iq->qchase.qname, iq->qchase.qname_len,
2958 LDNS_RR_TYPE_NS, iq->qchase.qclass)
2959 || reply_find_rrset_section_an(iq->response->rep,
2960 iq->qchase.qname, iq->qchase.qname_len,
2961 LDNS_RR_TYPE_NS, iq->qchase.qclass)
2962 )
2963 ))) {
2964 /* Store the referral under the current query */
2965 /* no prefetch-leeway, since its not the answer */
2966 iter_dns_store(qstate->env, &iq->response->qinfo,
2967 iq->response->rep, 1, 0, 0, NULL, 0);
2968 if(iq->store_parent_NS)
2969 iter_store_parentside_NS(qstate->env,
2970 iq->response->rep);
2971 if(qstate->env->neg_cache)
2972 val_neg_addreferral(qstate->env->neg_cache,
2973 iq->response->rep, iq->dp->name);
2974 }
2975 /* store parent-side-in-zone-glue, if directly queried for */
2976 if(!qstate->no_cache_store && iq->query_for_pside_glue
2977 && !iq->pside_glue) {
2978 iq->pside_glue = reply_find_rrset(iq->response->rep,
2979 iq->qchase.qname, iq->qchase.qname_len,
2980 iq->qchase.qtype, iq->qchase.qclass);
2981 if(iq->pside_glue) {
2982 log_rrset_key(VERB_ALGO, "found parent-side "
2983 "glue", iq->pside_glue);
2984 iter_store_parentside_rrset(qstate->env,
2985 iq->pside_glue);
2986 }
2987 }
2988
2989 /* Reset the event state, setting the current delegation
2990 * point to the referral. */
2991 iq->deleg_msg = iq->response;
2992 iq->dp = delegpt_from_message(iq->response, qstate->region);
2993 if (qstate->env->cfg->qname_minimisation)
2994 iq->minimisation_state = INIT_MINIMISE_STATE;
2995 if(!iq->dp) {
2996 errinf(qstate, "malloc failure, for delegation point");
2997 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2998 }
2999 if(!cache_fill_missing(qstate->env, iq->qchase.qclass,
3000 qstate->region, iq->dp)) {
3001 errinf(qstate, "malloc failure, copy extra info into delegation point");
3002 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3003 }
3004 if(iq->store_parent_NS && query_dname_compare(iq->dp->name,
3005 iq->store_parent_NS->name) == 0)
3006 iter_merge_retry_counts(iq->dp, iq->store_parent_NS,
3007 ie->outbound_msg_retry);
3008 delegpt_log(VERB_ALGO, iq->dp);
3009 /* Count this as a referral. */
3010 iq->referral_count++;
3011 iq->sent_count = 0;
3012 iq->dp_target_count = 0;
3013 /* see if the next dp is a trust anchor, or a DS was sent
3014 * along, indicating dnssec is expected for next zone */
3015 iq->dnssec_expected = iter_indicates_dnssec(qstate->env,
3016 iq->dp, iq->response, iq->qchase.qclass);
3017 /* if dnssec, validating then also fetch the key for the DS */
3018 if(iq->dnssec_expected && qstate->env->cfg->prefetch_key &&
3019 !(qstate->query_flags&BIT_CD))
3020 generate_dnskey_prefetch(qstate, iq, id);
3021
3022 /* spawn off NS and addr to auth servers for the NS we just
3023 * got in the referral. This gets authoritative answer
3024 * (answer section trust level) rrset.
3025 * right after, we detach the subs, answer goes to cache. */
3026 if(qstate->env->cfg->harden_referral_path)
3027 generate_ns_check(qstate, iq, id);
3028
3029 /* stop current outstanding queries.
3030 * FIXME: should the outstanding queries be waited for and
3031 * handled? Say by a subquery that inherits the outbound_entry.
3032 */
3033 outbound_list_clear(&iq->outlist);
3034 iq->num_current_queries = 0;
3035 fptr_ok(fptr_whitelist_modenv_detach_subs(
3036 qstate->env->detach_subs));
3037 (*qstate->env->detach_subs)(qstate);
3038 iq->num_target_queries = 0;
3039 iq->response = NULL;
3040 iq->fail_reply = NULL;
3041 verbose(VERB_ALGO, "cleared outbound list for next round");
3042 return next_state(iq, QUERYTARGETS_STATE);
3043 } else if(type == RESPONSE_TYPE_CNAME) {
3044 uint8_t* sname = NULL;
3045 size_t snamelen = 0;
3046 /* CNAME type responses get a query restart (i.e., get a
3047 * reset of the query state and go back to INIT_REQUEST_STATE).
3048 */
3049 verbose(VERB_DETAIL, "query response was CNAME");
3050 if(verbosity >= VERB_ALGO)
3051 log_dns_msg("cname msg", &iq->response->qinfo,
3052 iq->response->rep);
3053 /* if qtype is DS, check we have the right level of answer,
3054 * like grandchild answer but we need the middle, reject it */
3055 if(iq->qchase.qtype == LDNS_RR_TYPE_DS && !iq->dsns_point
3056 && !(iq->chase_flags&BIT_RD)
3057 && iter_ds_toolow(iq->response, iq->dp)
3058 && iter_dp_cangodown(&iq->qchase, iq->dp)) {
3059 outbound_list_clear(&iq->outlist);
3060 iq->num_current_queries = 0;
3061 fptr_ok(fptr_whitelist_modenv_detach_subs(
3062 qstate->env->detach_subs));
3063 (*qstate->env->detach_subs)(qstate);
3064 iq->num_target_queries = 0;
3065 return processDSNSFind(qstate, iq, id);
3066 }
3067 /* Process the CNAME response. */
3068 if(!handle_cname_response(qstate, iq, iq->response,
3069 &sname, &snamelen)) {
3070 errinf(qstate, "malloc failure, CNAME info");
3071 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3072 }
3073 /* cache the CNAME response under the current query */
3074 /* NOTE : set referral=1, so that rrsets get stored but not
3075 * the partial query answer (CNAME only). */
3076 /* prefetchleeway applied because this updates answer parts */
3077 if(!qstate->no_cache_store)
3078 iter_dns_store(qstate->env, &iq->response->qinfo,
3079 iq->response->rep, 1, qstate->prefetch_leeway,
3080 iq->dp&&iq->dp->has_parent_side_NS, NULL,
3081 qstate->query_flags);
3082 /* set the current request's qname to the new value. */
3083 iq->qchase.qname = sname;
3084 iq->qchase.qname_len = snamelen;
3085 if(qstate->env->auth_zones) {
3086 /* apply rpz qname triggers after cname */
3087 struct dns_msg* forged_response =
3088 rpz_callback_from_iterator_cname(qstate, iq);
3089 while(forged_response && reply_find_rrset_section_an(
3090 forged_response->rep, iq->qchase.qname,
3091 iq->qchase.qname_len, LDNS_RR_TYPE_CNAME,
3092 iq->qchase.qclass)) {
3093 /* another cname to follow */
3094 if(!handle_cname_response(qstate, iq, forged_response,
3095 &sname, &snamelen)) {
3096 errinf(qstate, "malloc failure, CNAME info");
3097 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3098 }
3099 iq->qchase.qname = sname;
3100 iq->qchase.qname_len = snamelen;
3101 forged_response =
3102 rpz_callback_from_iterator_cname(qstate, iq);
3103 }
3104 if(forged_response != NULL) {
3105 qstate->ext_state[id] = module_finished;
3106 qstate->return_rcode = FLAGS_GET_RCODE(forged_response->rep->flags);
3107 qstate->return_msg = forged_response;
3108 iq->response = forged_response;
3109 next_state(iq, FINISHED_STATE);
3110 if(!iter_prepend(iq, qstate->return_msg, qstate->region)) {
3111 log_err("rpz after cname, prepend rrsets: out of memory");
3112 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3113 }
3114 qstate->return_msg->qinfo = qstate->qinfo;
3115 return 0;
3116 }
3117 }
3118 /* Clear the query state, since this is a query restart. */
3119 iq->deleg_msg = NULL;
3120 iq->dp = NULL;
3121 iq->dsns_point = NULL;
3122 iq->auth_zone_response = 0;
3123 iq->sent_count = 0;
3124 iq->dp_target_count = 0;
3125 if(iq->minimisation_state != MINIMISE_STATE)
3126 /* Only count as query restart when it is not an extra
3127 * query as result of qname minimisation. */
3128 iq->query_restart_count++;
3129 if(qstate->env->cfg->qname_minimisation)
3130 iq->minimisation_state = INIT_MINIMISE_STATE;
3131
3132 /* stop current outstanding queries.
3133 * FIXME: should the outstanding queries be waited for and
3134 * handled? Say by a subquery that inherits the outbound_entry.
3135 */
3136 outbound_list_clear(&iq->outlist);
3137 iq->num_current_queries = 0;
3138 fptr_ok(fptr_whitelist_modenv_detach_subs(
3139 qstate->env->detach_subs));
3140 (*qstate->env->detach_subs)(qstate);
3141 iq->num_target_queries = 0;
3142 if(qstate->reply)
3143 sock_list_insert(&qstate->reply_origin,
3144 &qstate->reply->addr, qstate->reply->addrlen,
3145 qstate->region);
3146 verbose(VERB_ALGO, "cleared outbound list for query restart");
3147 /* go to INIT_REQUEST_STATE for new qname. */
3148 return next_state(iq, INIT_REQUEST_STATE);
3149 } else if(type == RESPONSE_TYPE_LAME) {
3150 /* Cache the LAMEness. */
3151 verbose(VERB_DETAIL, "query response was %sLAME",
3152 dnsseclame?"DNSSEC ":"");
3153 if(!dname_subdomain_c(iq->qchase.qname, iq->dp->name)) {
3154 log_err("mark lame: mismatch in qname and dpname");
3155 /* throwaway this reply below */
3156 } else if(qstate->reply) {
3157 /* need addr for lameness cache, but we may have
3158 * gotten this from cache, so test to be sure */
3159 if(!infra_set_lame(qstate->env->infra_cache,
3160 &qstate->reply->addr, qstate->reply->addrlen,
3161 iq->dp->name, iq->dp->namelen,
3162 *qstate->env->now, dnsseclame, 0,
3163 iq->qchase.qtype))
3164 log_err("mark host lame: out of memory");
3165 }
3166 } else if(type == RESPONSE_TYPE_REC_LAME) {
3167 /* Cache the LAMEness. */
3168 verbose(VERB_DETAIL, "query response REC_LAME: "
3169 "recursive but not authoritative server");
3170 if(!dname_subdomain_c(iq->qchase.qname, iq->dp->name)) {
3171 log_err("mark rec_lame: mismatch in qname and dpname");
3172 /* throwaway this reply below */
3173 } else if(qstate->reply) {
3174 /* need addr for lameness cache, but we may have
3175 * gotten this from cache, so test to be sure */
3176 verbose(VERB_DETAIL, "mark as REC_LAME");
3177 if(!infra_set_lame(qstate->env->infra_cache,
3178 &qstate->reply->addr, qstate->reply->addrlen,
3179 iq->dp->name, iq->dp->namelen,
3180 *qstate->env->now, 0, 1, iq->qchase.qtype))
3181 log_err("mark host lame: out of memory");
3182 }
3183 } else if(type == RESPONSE_TYPE_THROWAWAY) {
3184 /* LAME and THROWAWAY responses are handled the same way.
3185 * In this case, the event is just sent directly back to
3186 * the QUERYTARGETS_STATE without resetting anything,
3187 * because, clearly, the next target must be tried. */
3188 verbose(VERB_DETAIL, "query response was THROWAWAY");
3189 } else {
3190 log_warn("A query response came back with an unknown type: %d",
3191 (int)type);
3192 }
3193
3194 /* LAME, THROWAWAY and "unknown" all end up here.
3195 * Recycle to the QUERYTARGETS state to hopefully try a
3196 * different target. */
3197 if (qstate->env->cfg->qname_minimisation &&
3198 !qstate->env->cfg->qname_minimisation_strict)
3199 iq->minimisation_state = DONOT_MINIMISE_STATE;
3200 if(iq->auth_zone_response) {
3201 /* can we fallback? */
3202 iq->auth_zone_response = 0;
3203 if(!auth_zones_can_fallback(qstate->env->auth_zones,
3204 iq->dp->name, iq->dp->namelen, qstate->qinfo.qclass)) {
3205 verbose(VERB_ALGO, "auth zone response bad, and no"
3206 " fallback possible, servfail");
3207 errinf_dname(qstate, "response is bad, no fallback, "
3208 "for auth zone", iq->dp->name);
3209 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3210 }
3211 verbose(VERB_ALGO, "auth zone response was bad, "
3212 "fallback enabled");
3213 iq->auth_zone_avoid = 1;
3214 if(iq->dp->auth_dp) {
3215 /* we are using a dp for the auth zone, with no
3216 * nameservers, get one first */
3217 iq->dp = NULL;
3218 return next_state(iq, INIT_REQUEST_STATE);
3219 }
3220 }
3221 return next_state(iq, QUERYTARGETS_STATE);
3222 }
3223
3224 /**
3225 * Return priming query results to interested super querystates.
3226 *
3227 * Sets the delegation point and delegation message (not nonRD queries).
3228 * This is a callback from walk_supers.
3229 *
3230 * @param qstate: priming query state that finished.
3231 * @param id: module id.
3232 * @param forq: the qstate for which priming has been done.
3233 */
3234 static void
prime_supers(struct module_qstate * qstate,int id,struct module_qstate * forq)3235 prime_supers(struct module_qstate* qstate, int id, struct module_qstate* forq)
3236 {
3237 struct iter_qstate* foriq = (struct iter_qstate*)forq->minfo[id];
3238 struct delegpt* dp = NULL;
3239
3240 log_assert(qstate->is_priming || foriq->wait_priming_stub);
3241 log_assert(qstate->return_rcode == LDNS_RCODE_NOERROR);
3242 /* Convert our response to a delegation point */
3243 dp = delegpt_from_message(qstate->return_msg, forq->region);
3244 if(!dp) {
3245 /* if there is no convertible delegation point, then
3246 * the ANSWER type was (presumably) a negative answer. */
3247 verbose(VERB_ALGO, "prime response was not a positive "
3248 "ANSWER; failing");
3249 foriq->dp = NULL;
3250 foriq->state = QUERYTARGETS_STATE;
3251 return;
3252 }
3253
3254 log_query_info(VERB_DETAIL, "priming successful for", &qstate->qinfo);
3255 delegpt_log(VERB_ALGO, dp);
3256 foriq->dp = dp;
3257 foriq->deleg_msg = dns_copy_msg(qstate->return_msg, forq->region);
3258 if(!foriq->deleg_msg) {
3259 log_err("copy prime response: out of memory");
3260 foriq->dp = NULL;
3261 foriq->state = QUERYTARGETS_STATE;
3262 return;
3263 }
3264
3265 /* root priming responses go to init stage 2, priming stub
3266 * responses to to stage 3. */
3267 if(foriq->wait_priming_stub) {
3268 foriq->state = INIT_REQUEST_3_STATE;
3269 foriq->wait_priming_stub = 0;
3270 } else foriq->state = INIT_REQUEST_2_STATE;
3271 /* because we are finished, the parent will be reactivated */
3272 }
3273
3274 /**
3275 * This handles the response to a priming query. This is used to handle both
3276 * root and stub priming responses. This is basically the equivalent of the
3277 * QUERY_RESP_STATE, but will not handle CNAME responses and will treat
3278 * REFERRALs as ANSWERS. It will also update and reactivate the originating
3279 * event.
3280 *
3281 * @param qstate: query state.
3282 * @param id: module id.
3283 * @return true if the event needs more immediate processing, false if not.
3284 * This state always returns false.
3285 */
3286 static int
processPrimeResponse(struct module_qstate * qstate,int id)3287 processPrimeResponse(struct module_qstate* qstate, int id)
3288 {
3289 struct iter_qstate* iq = (struct iter_qstate*)qstate->minfo[id];
3290 enum response_type type;
3291 iq->response->rep->flags &= ~(BIT_RD|BIT_RA); /* ignore rec-lame */
3292 type = response_type_from_server(
3293 (int)((iq->chase_flags&BIT_RD) || iq->chase_to_rd),
3294 iq->response, &iq->qchase, iq->dp);
3295 if(type == RESPONSE_TYPE_ANSWER) {
3296 qstate->return_rcode = LDNS_RCODE_NOERROR;
3297 qstate->return_msg = iq->response;
3298 } else {
3299 errinf(qstate, "prime response did not get an answer");
3300 errinf_dname(qstate, "for", qstate->qinfo.qname);
3301 qstate->return_rcode = LDNS_RCODE_SERVFAIL;
3302 qstate->return_msg = NULL;
3303 }
3304
3305 /* validate the root or stub after priming (if enabled).
3306 * This is the same query as the prime query, but with validation.
3307 * Now that we are primed, the additional queries that validation
3308 * may need can be resolved. */
3309 if(qstate->env->cfg->harden_referral_path) {
3310 struct module_qstate* subq = NULL;
3311 log_nametypeclass(VERB_ALGO, "schedule prime validation",
3312 qstate->qinfo.qname, qstate->qinfo.qtype,
3313 qstate->qinfo.qclass);
3314 if(!generate_sub_request(qstate->qinfo.qname,
3315 qstate->qinfo.qname_len, qstate->qinfo.qtype,
3316 qstate->qinfo.qclass, qstate, id, iq,
3317 INIT_REQUEST_STATE, FINISHED_STATE, &subq, 1, 0)) {
3318 verbose(VERB_ALGO, "could not generate prime check");
3319 }
3320 generate_a_aaaa_check(qstate, iq, id);
3321 }
3322
3323 /* This event is finished. */
3324 qstate->ext_state[id] = module_finished;
3325 return 0;
3326 }
3327
3328 /**
3329 * Do final processing on responses to target queries. Events reach this
3330 * state after the iterative resolution algorithm terminates. This state is
3331 * responsible for reactivating the original event, and housekeeping related
3332 * to received target responses (caching, updating the current delegation
3333 * point, etc).
3334 * Callback from walk_supers for every super state that is interested in
3335 * the results from this query.
3336 *
3337 * @param qstate: query state.
3338 * @param id: module id.
3339 * @param forq: super query state.
3340 */
3341 static void
processTargetResponse(struct module_qstate * qstate,int id,struct module_qstate * forq)3342 processTargetResponse(struct module_qstate* qstate, int id,
3343 struct module_qstate* forq)
3344 {
3345 struct iter_env* ie = (struct iter_env*)qstate->env->modinfo[id];
3346 struct iter_qstate* iq = (struct iter_qstate*)qstate->minfo[id];
3347 struct iter_qstate* foriq = (struct iter_qstate*)forq->minfo[id];
3348 struct ub_packed_rrset_key* rrset;
3349 struct delegpt_ns* dpns;
3350 log_assert(qstate->return_rcode == LDNS_RCODE_NOERROR);
3351
3352 foriq->state = QUERYTARGETS_STATE;
3353 log_query_info(VERB_ALGO, "processTargetResponse", &qstate->qinfo);
3354 log_query_info(VERB_ALGO, "processTargetResponse super", &forq->qinfo);
3355
3356 /* Tell the originating event that this target query has finished
3357 * (regardless if it succeeded or not). */
3358 foriq->num_target_queries--;
3359
3360 /* check to see if parent event is still interested (in orig name). */
3361 if(!foriq->dp) {
3362 verbose(VERB_ALGO, "subq: parent not interested, was reset");
3363 return; /* not interested anymore */
3364 }
3365 dpns = delegpt_find_ns(foriq->dp, qstate->qinfo.qname,
3366 qstate->qinfo.qname_len);
3367 if(!dpns) {
3368 /* If not interested, just stop processing this event */
3369 verbose(VERB_ALGO, "subq: parent not interested anymore");
3370 /* could be because parent was jostled out of the cache,
3371 and a new identical query arrived, that does not want it*/
3372 return;
3373 }
3374
3375 /* if iq->query_for_pside_glue then add the pside_glue (marked lame) */
3376 if(iq->pside_glue) {
3377 /* if the pside_glue is NULL, then it could not be found,
3378 * the done_pside is already set when created and a cache
3379 * entry created in processFinished so nothing to do here */
3380 log_rrset_key(VERB_ALGO, "add parentside glue to dp",
3381 iq->pside_glue);
3382 if(!delegpt_add_rrset(foriq->dp, forq->region,
3383 iq->pside_glue, 1, NULL))
3384 log_err("out of memory adding pside glue");
3385 }
3386
3387 /* This response is relevant to the current query, so we
3388 * add (attempt to add, anyway) this target(s) and reactivate
3389 * the original event.
3390 * NOTE: we could only look for the AnswerRRset if the
3391 * response type was ANSWER. */
3392 rrset = reply_find_answer_rrset(&iq->qchase, qstate->return_msg->rep);
3393 if(rrset) {
3394 int additions = 0;
3395 /* if CNAMEs have been followed - add new NS to delegpt. */
3396 /* BTW. RFC 1918 says NS should not have got CNAMEs. Robust. */
3397 if(!delegpt_find_ns(foriq->dp, rrset->rk.dname,
3398 rrset->rk.dname_len)) {
3399 /* if dpns->lame then set newcname ns lame too */
3400 if(!delegpt_add_ns(foriq->dp, forq->region,
3401 rrset->rk.dname, dpns->lame))
3402 log_err("out of memory adding cnamed-ns");
3403 }
3404 /* if dpns->lame then set the address(es) lame too */
3405 if(!delegpt_add_rrset(foriq->dp, forq->region, rrset,
3406 dpns->lame, &additions))
3407 log_err("out of memory adding targets");
3408 if(!additions) {
3409 /* no new addresses, increase the nxns counter, like
3410 * this could be a list of wildcards with no new
3411 * addresses */
3412 target_count_increase_nx(foriq, 1);
3413 }
3414 verbose(VERB_ALGO, "added target response");
3415 delegpt_log(VERB_ALGO, foriq->dp);
3416 } else {
3417 verbose(VERB_ALGO, "iterator TargetResponse failed");
3418 delegpt_mark_neg(dpns, qstate->qinfo.qtype);
3419 dpns->resolved = 1; /* fail the target */
3420 if((dpns->got4 == 2 || !ie->supports_ipv4) &&
3421 (dpns->got6 == 2 || !ie->supports_ipv6))
3422 target_count_increase_nx(foriq, 1);
3423 }
3424 }
3425
3426 /**
3427 * Process response for DS NS Find queries, that attempt to find the delegation
3428 * point where we ask the DS query from.
3429 *
3430 * @param qstate: query state.
3431 * @param id: module id.
3432 * @param forq: super query state.
3433 */
3434 static void
processDSNSResponse(struct module_qstate * qstate,int id,struct module_qstate * forq)3435 processDSNSResponse(struct module_qstate* qstate, int id,
3436 struct module_qstate* forq)
3437 {
3438 struct iter_qstate* foriq = (struct iter_qstate*)forq->minfo[id];
3439
3440 /* if the finished (iq->response) query has no NS set: continue
3441 * up to look for the right dp; nothing to change, do DPNSstate */
3442 if(qstate->return_rcode != LDNS_RCODE_NOERROR)
3443 return; /* seek further */
3444 /* find the NS RRset (without allowing CNAMEs) */
3445 if(!reply_find_rrset(qstate->return_msg->rep, qstate->qinfo.qname,
3446 qstate->qinfo.qname_len, LDNS_RR_TYPE_NS,
3447 qstate->qinfo.qclass)){
3448 return; /* seek further */
3449 }
3450
3451 /* else, store as DP and continue at querytargets */
3452 foriq->state = QUERYTARGETS_STATE;
3453 foriq->dp = delegpt_from_message(qstate->return_msg, forq->region);
3454 if(!foriq->dp) {
3455 log_err("out of memory in dsns dp alloc");
3456 errinf(qstate, "malloc failure, in DS search");
3457 return; /* dp==NULL in QUERYTARGETS makes SERVFAIL */
3458 }
3459 /* success, go query the querytargets in the new dp (and go down) */
3460 }
3461
3462 /**
3463 * Process response for qclass=ANY queries for a particular class.
3464 * Append to result or error-exit.
3465 *
3466 * @param qstate: query state.
3467 * @param id: module id.
3468 * @param forq: super query state.
3469 */
3470 static void
processClassResponse(struct module_qstate * qstate,int id,struct module_qstate * forq)3471 processClassResponse(struct module_qstate* qstate, int id,
3472 struct module_qstate* forq)
3473 {
3474 struct iter_qstate* foriq = (struct iter_qstate*)forq->minfo[id];
3475 struct dns_msg* from = qstate->return_msg;
3476 log_query_info(VERB_ALGO, "processClassResponse", &qstate->qinfo);
3477 log_query_info(VERB_ALGO, "processClassResponse super", &forq->qinfo);
3478 if(qstate->return_rcode != LDNS_RCODE_NOERROR) {
3479 /* cause servfail for qclass ANY query */
3480 foriq->response = NULL;
3481 foriq->state = FINISHED_STATE;
3482 return;
3483 }
3484 /* append result */
3485 if(!foriq->response) {
3486 /* allocate the response: copy RCODE, sec_state */
3487 foriq->response = dns_copy_msg(from, forq->region);
3488 if(!foriq->response) {
3489 log_err("malloc failed for qclass ANY response");
3490 foriq->state = FINISHED_STATE;
3491 return;
3492 }
3493 foriq->response->qinfo.qclass = forq->qinfo.qclass;
3494 /* qclass ANY does not receive the AA flag on replies */
3495 foriq->response->rep->authoritative = 0;
3496 } else {
3497 struct dns_msg* to = foriq->response;
3498 /* add _from_ this response _to_ existing collection */
3499 /* if there are records, copy RCODE */
3500 /* lower sec_state if this message is lower */
3501 if(from->rep->rrset_count != 0) {
3502 size_t n = from->rep->rrset_count+to->rep->rrset_count;
3503 struct ub_packed_rrset_key** dest, **d;
3504 /* copy appropriate rcode */
3505 to->rep->flags = from->rep->flags;
3506 /* copy rrsets */
3507 if(from->rep->rrset_count > RR_COUNT_MAX ||
3508 to->rep->rrset_count > RR_COUNT_MAX) {
3509 log_err("malloc failed (too many rrsets) in collect ANY");
3510 foriq->state = FINISHED_STATE;
3511 return; /* integer overflow protection */
3512 }
3513 dest = regional_alloc(forq->region, sizeof(dest[0])*n);
3514 if(!dest) {
3515 log_err("malloc failed in collect ANY");
3516 foriq->state = FINISHED_STATE;
3517 return;
3518 }
3519 d = dest;
3520 /* copy AN */
3521 memcpy(dest, to->rep->rrsets, to->rep->an_numrrsets
3522 * sizeof(dest[0]));
3523 dest += to->rep->an_numrrsets;
3524 memcpy(dest, from->rep->rrsets, from->rep->an_numrrsets
3525 * sizeof(dest[0]));
3526 dest += from->rep->an_numrrsets;
3527 /* copy NS */
3528 memcpy(dest, to->rep->rrsets+to->rep->an_numrrsets,
3529 to->rep->ns_numrrsets * sizeof(dest[0]));
3530 dest += to->rep->ns_numrrsets;
3531 memcpy(dest, from->rep->rrsets+from->rep->an_numrrsets,
3532 from->rep->ns_numrrsets * sizeof(dest[0]));
3533 dest += from->rep->ns_numrrsets;
3534 /* copy AR */
3535 memcpy(dest, to->rep->rrsets+to->rep->an_numrrsets+
3536 to->rep->ns_numrrsets,
3537 to->rep->ar_numrrsets * sizeof(dest[0]));
3538 dest += to->rep->ar_numrrsets;
3539 memcpy(dest, from->rep->rrsets+from->rep->an_numrrsets+
3540 from->rep->ns_numrrsets,
3541 from->rep->ar_numrrsets * sizeof(dest[0]));
3542 /* update counts */
3543 to->rep->rrsets = d;
3544 to->rep->an_numrrsets += from->rep->an_numrrsets;
3545 to->rep->ns_numrrsets += from->rep->ns_numrrsets;
3546 to->rep->ar_numrrsets += from->rep->ar_numrrsets;
3547 to->rep->rrset_count = n;
3548 }
3549 if(from->rep->security < to->rep->security) /* lowest sec */
3550 to->rep->security = from->rep->security;
3551 if(from->rep->qdcount != 0) /* insert qd if appropriate */
3552 to->rep->qdcount = from->rep->qdcount;
3553 if(from->rep->ttl < to->rep->ttl) /* use smallest TTL */
3554 to->rep->ttl = from->rep->ttl;
3555 if(from->rep->prefetch_ttl < to->rep->prefetch_ttl)
3556 to->rep->prefetch_ttl = from->rep->prefetch_ttl;
3557 if(from->rep->serve_expired_ttl < to->rep->serve_expired_ttl)
3558 to->rep->serve_expired_ttl = from->rep->serve_expired_ttl;
3559 }
3560 /* are we done? */
3561 foriq->num_current_queries --;
3562 if(foriq->num_current_queries == 0)
3563 foriq->state = FINISHED_STATE;
3564 }
3565
3566 /**
3567 * Collect class ANY responses and make them into one response. This
3568 * state is started and it creates queries for all classes (that have
3569 * root hints). The answers are then collected.
3570 *
3571 * @param qstate: query state.
3572 * @param id: module id.
3573 * @return true if the event needs more immediate processing, false if not.
3574 */
3575 static int
processCollectClass(struct module_qstate * qstate,int id)3576 processCollectClass(struct module_qstate* qstate, int id)
3577 {
3578 struct iter_qstate* iq = (struct iter_qstate*)qstate->minfo[id];
3579 struct module_qstate* subq;
3580 /* If qchase.qclass == 0 then send out queries for all classes.
3581 * Otherwise, do nothing (wait for all answers to arrive and the
3582 * processClassResponse to put them together, and that moves us
3583 * towards the Finished state when done. */
3584 if(iq->qchase.qclass == 0) {
3585 uint16_t c = 0;
3586 iq->qchase.qclass = LDNS_RR_CLASS_ANY;
3587 while(iter_get_next_root(qstate->env->hints,
3588 qstate->env->fwds, &c)) {
3589 /* generate query for this class */
3590 log_nametypeclass(VERB_ALGO, "spawn collect query",
3591 qstate->qinfo.qname, qstate->qinfo.qtype, c);
3592 if(!generate_sub_request(qstate->qinfo.qname,
3593 qstate->qinfo.qname_len, qstate->qinfo.qtype,
3594 c, qstate, id, iq, INIT_REQUEST_STATE,
3595 FINISHED_STATE, &subq,
3596 (int)!(qstate->query_flags&BIT_CD), 0)) {
3597 errinf(qstate, "could not generate class ANY"
3598 " lookup query");
3599 return error_response(qstate, id,
3600 LDNS_RCODE_SERVFAIL);
3601 }
3602 /* ignore subq, no special init required */
3603 iq->num_current_queries ++;
3604 if(c == 0xffff)
3605 break;
3606 else c++;
3607 }
3608 /* if no roots are configured at all, return */
3609 if(iq->num_current_queries == 0) {
3610 verbose(VERB_ALGO, "No root hints or fwds, giving up "
3611 "on qclass ANY");
3612 return error_response(qstate, id, LDNS_RCODE_REFUSED);
3613 }
3614 /* return false, wait for queries to return */
3615 }
3616 /* if woke up here because of an answer, wait for more answers */
3617 return 0;
3618 }
3619
3620 /**
3621 * This handles the final state for first-tier responses (i.e., responses to
3622 * externally generated queries).
3623 *
3624 * @param qstate: query state.
3625 * @param iq: iterator query state.
3626 * @param id: module id.
3627 * @return true if the event needs more processing, false if not. Since this
3628 * is the final state for an event, it always returns false.
3629 */
3630 static int
processFinished(struct module_qstate * qstate,struct iter_qstate * iq,int id)3631 processFinished(struct module_qstate* qstate, struct iter_qstate* iq,
3632 int id)
3633 {
3634 log_query_info(VERB_QUERY, "finishing processing for",
3635 &qstate->qinfo);
3636
3637 /* store negative cache element for parent side glue. */
3638 if(!qstate->no_cache_store && iq->query_for_pside_glue
3639 && !iq->pside_glue)
3640 iter_store_parentside_neg(qstate->env, &qstate->qinfo,
3641 iq->deleg_msg?iq->deleg_msg->rep:
3642 (iq->response?iq->response->rep:NULL));
3643 if(!iq->response) {
3644 verbose(VERB_ALGO, "No response is set, servfail");
3645 errinf(qstate, "(no response found at query finish)");
3646 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3647 }
3648
3649 /* Make sure that the RA flag is set (since the presence of
3650 * this module means that recursion is available) */
3651 iq->response->rep->flags |= BIT_RA;
3652
3653 /* Clear the AA flag */
3654 /* FIXME: does this action go here or in some other module? */
3655 iq->response->rep->flags &= ~BIT_AA;
3656
3657 /* make sure QR flag is on */
3658 iq->response->rep->flags |= BIT_QR;
3659
3660 /* we have finished processing this query */
3661 qstate->ext_state[id] = module_finished;
3662
3663 /* TODO: we are using a private TTL, trim the response. */
3664 /* if (mPrivateTTL > 0){IterUtils.setPrivateTTL(resp, mPrivateTTL); } */
3665
3666 /* prepend any items we have accumulated */
3667 if(iq->an_prepend_list || iq->ns_prepend_list) {
3668 if(!iter_prepend(iq, iq->response, qstate->region)) {
3669 log_err("prepend rrsets: out of memory");
3670 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3671 }
3672 /* reset the query name back */
3673 iq->response->qinfo = qstate->qinfo;
3674 /* the security state depends on the combination */
3675 iq->response->rep->security = sec_status_unchecked;
3676 /* store message with the finished prepended items,
3677 * but only if we did recursion. The nonrecursion referral
3678 * from cache does not need to be stored in the msg cache. */
3679 if(!qstate->no_cache_store && qstate->query_flags&BIT_RD) {
3680 iter_dns_store(qstate->env, &qstate->qinfo,
3681 iq->response->rep, 0, qstate->prefetch_leeway,
3682 iq->dp&&iq->dp->has_parent_side_NS,
3683 qstate->region, qstate->query_flags);
3684 }
3685 }
3686 qstate->return_rcode = LDNS_RCODE_NOERROR;
3687 qstate->return_msg = iq->response;
3688 return 0;
3689 }
3690
3691 /*
3692 * Return priming query results to interested super querystates.
3693 *
3694 * Sets the delegation point and delegation message (not nonRD queries).
3695 * This is a callback from walk_supers.
3696 *
3697 * @param qstate: query state that finished.
3698 * @param id: module id.
3699 * @param super: the qstate to inform.
3700 */
3701 void
iter_inform_super(struct module_qstate * qstate,int id,struct module_qstate * super)3702 iter_inform_super(struct module_qstate* qstate, int id,
3703 struct module_qstate* super)
3704 {
3705 if(!qstate->is_priming && super->qinfo.qclass == LDNS_RR_CLASS_ANY)
3706 processClassResponse(qstate, id, super);
3707 else if(super->qinfo.qtype == LDNS_RR_TYPE_DS && ((struct iter_qstate*)
3708 super->minfo[id])->state == DSNS_FIND_STATE)
3709 processDSNSResponse(qstate, id, super);
3710 else if(qstate->return_rcode != LDNS_RCODE_NOERROR)
3711 error_supers(qstate, id, super);
3712 else if(qstate->is_priming)
3713 prime_supers(qstate, id, super);
3714 else processTargetResponse(qstate, id, super);
3715 }
3716
3717 /**
3718 * Handle iterator state.
3719 * Handle events. This is the real processing loop for events, responsible
3720 * for moving events through the various states. If a processing method
3721 * returns true, then it will be advanced to the next state. If false, then
3722 * processing will stop.
3723 *
3724 * @param qstate: query state.
3725 * @param ie: iterator shared global environment.
3726 * @param iq: iterator query state.
3727 * @param id: module id.
3728 */
3729 static void
iter_handle(struct module_qstate * qstate,struct iter_qstate * iq,struct iter_env * ie,int id)3730 iter_handle(struct module_qstate* qstate, struct iter_qstate* iq,
3731 struct iter_env* ie, int id)
3732 {
3733 int cont = 1;
3734 while(cont) {
3735 verbose(VERB_ALGO, "iter_handle processing q with state %s",
3736 iter_state_to_string(iq->state));
3737 switch(iq->state) {
3738 case INIT_REQUEST_STATE:
3739 cont = processInitRequest(qstate, iq, ie, id);
3740 break;
3741 case INIT_REQUEST_2_STATE:
3742 cont = processInitRequest2(qstate, iq, id);
3743 break;
3744 case INIT_REQUEST_3_STATE:
3745 cont = processInitRequest3(qstate, iq, id);
3746 break;
3747 case QUERYTARGETS_STATE:
3748 cont = processQueryTargets(qstate, iq, ie, id);
3749 break;
3750 case QUERY_RESP_STATE:
3751 cont = processQueryResponse(qstate, iq, ie, id);
3752 break;
3753 case PRIME_RESP_STATE:
3754 cont = processPrimeResponse(qstate, id);
3755 break;
3756 case COLLECT_CLASS_STATE:
3757 cont = processCollectClass(qstate, id);
3758 break;
3759 case DSNS_FIND_STATE:
3760 cont = processDSNSFind(qstate, iq, id);
3761 break;
3762 case FINISHED_STATE:
3763 cont = processFinished(qstate, iq, id);
3764 break;
3765 default:
3766 log_warn("iterator: invalid state: %d",
3767 iq->state);
3768 cont = 0;
3769 break;
3770 }
3771 }
3772 }
3773
3774 /**
3775 * This is the primary entry point for processing request events. Note that
3776 * this method should only be used by external modules.
3777 * @param qstate: query state.
3778 * @param ie: iterator shared global environment.
3779 * @param iq: iterator query state.
3780 * @param id: module id.
3781 */
3782 static void
process_request(struct module_qstate * qstate,struct iter_qstate * iq,struct iter_env * ie,int id)3783 process_request(struct module_qstate* qstate, struct iter_qstate* iq,
3784 struct iter_env* ie, int id)
3785 {
3786 /* external requests start in the INIT state, and finish using the
3787 * FINISHED state. */
3788 iq->state = INIT_REQUEST_STATE;
3789 iq->final_state = FINISHED_STATE;
3790 verbose(VERB_ALGO, "process_request: new external request event");
3791 iter_handle(qstate, iq, ie, id);
3792 }
3793
3794 /** process authoritative server reply */
3795 static void
process_response(struct module_qstate * qstate,struct iter_qstate * iq,struct iter_env * ie,int id,struct outbound_entry * outbound,enum module_ev event)3796 process_response(struct module_qstate* qstate, struct iter_qstate* iq,
3797 struct iter_env* ie, int id, struct outbound_entry* outbound,
3798 enum module_ev event)
3799 {
3800 struct msg_parse* prs;
3801 struct edns_data edns;
3802 sldns_buffer* pkt;
3803
3804 verbose(VERB_ALGO, "process_response: new external response event");
3805 iq->response = NULL;
3806 iq->state = QUERY_RESP_STATE;
3807 if(event == module_event_noreply || event == module_event_error) {
3808 if(event == module_event_noreply && iq->timeout_count >= 3 &&
3809 qstate->env->cfg->use_caps_bits_for_id &&
3810 !iq->caps_fallback && !is_caps_whitelisted(ie, iq)) {
3811 /* start fallback */
3812 iq->caps_fallback = 1;
3813 iq->caps_server = 0;
3814 iq->caps_reply = NULL;
3815 iq->caps_response = NULL;
3816 iq->caps_minimisation_state = DONOT_MINIMISE_STATE;
3817 iq->state = QUERYTARGETS_STATE;
3818 iq->num_current_queries--;
3819 /* need fresh attempts for the 0x20 fallback, if
3820 * that was the cause for the failure */
3821 iter_dec_attempts(iq->dp, 3, ie->outbound_msg_retry);
3822 verbose(VERB_DETAIL, "Capsforid: timeouts, starting fallback");
3823 goto handle_it;
3824 }
3825 goto handle_it;
3826 }
3827 if( (event != module_event_reply && event != module_event_capsfail)
3828 || !qstate->reply) {
3829 log_err("Bad event combined with response");
3830 outbound_list_remove(&iq->outlist, outbound);
3831 errinf(qstate, "module iterator received wrong internal event with a response message");
3832 (void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3833 return;
3834 }
3835
3836 /* parse message */
3837 iq->fail_reply = qstate->reply;
3838 prs = (struct msg_parse*)regional_alloc(qstate->env->scratch,
3839 sizeof(struct msg_parse));
3840 if(!prs) {
3841 log_err("out of memory on incoming message");
3842 /* like packet got dropped */
3843 goto handle_it;
3844 }
3845 memset(prs, 0, sizeof(*prs));
3846 memset(&edns, 0, sizeof(edns));
3847 pkt = qstate->reply->c->buffer;
3848 sldns_buffer_set_position(pkt, 0);
3849 if(parse_packet(pkt, prs, qstate->env->scratch) != LDNS_RCODE_NOERROR) {
3850 verbose(VERB_ALGO, "parse error on reply packet");
3851 iq->parse_failures++;
3852 goto handle_it;
3853 }
3854 /* edns is not examined, but removed from message to help cache */
3855 if(parse_extract_edns_from_response_msg(prs, &edns, qstate->env->scratch) !=
3856 LDNS_RCODE_NOERROR) {
3857 iq->parse_failures++;
3858 goto handle_it;
3859 }
3860
3861 /* Copy the edns options we may got from the back end */
3862 if(edns.opt_list_in) {
3863 qstate->edns_opts_back_in = edns_opt_copy_region(edns.opt_list_in,
3864 qstate->region);
3865 if(!qstate->edns_opts_back_in) {
3866 log_err("out of memory on incoming message");
3867 /* like packet got dropped */
3868 goto handle_it;
3869 }
3870 if(!inplace_cb_edns_back_parsed_call(qstate->env, qstate)) {
3871 log_err("unable to call edns_back_parsed callback");
3872 goto handle_it;
3873 }
3874 }
3875
3876 /* remove CD-bit, we asked for in case we handle validation ourself */
3877 prs->flags &= ~BIT_CD;
3878
3879 /* normalize and sanitize: easy to delete items from linked lists */
3880 if(!scrub_message(pkt, prs, &iq->qinfo_out, iq->dp->name,
3881 qstate->env->scratch, qstate->env, ie)) {
3882 /* if 0x20 enabled, start fallback, but we have no message */
3883 if(event == module_event_capsfail && !iq->caps_fallback) {
3884 iq->caps_fallback = 1;
3885 iq->caps_server = 0;
3886 iq->caps_reply = NULL;
3887 iq->caps_response = NULL;
3888 iq->caps_minimisation_state = DONOT_MINIMISE_STATE;
3889 iq->state = QUERYTARGETS_STATE;
3890 iq->num_current_queries--;
3891 verbose(VERB_DETAIL, "Capsforid: scrub failed, starting fallback with no response");
3892 }
3893 iq->scrub_failures++;
3894 goto handle_it;
3895 }
3896
3897 /* allocate response dns_msg in region */
3898 iq->response = dns_alloc_msg(pkt, prs, qstate->region);
3899 if(!iq->response)
3900 goto handle_it;
3901 log_query_info(VERB_DETAIL, "response for", &qstate->qinfo);
3902 log_name_addr(VERB_DETAIL, "reply from", iq->dp->name,
3903 &qstate->reply->addr, qstate->reply->addrlen);
3904 if(verbosity >= VERB_ALGO)
3905 log_dns_msg("incoming scrubbed packet:", &iq->response->qinfo,
3906 iq->response->rep);
3907
3908 if(event == module_event_capsfail || iq->caps_fallback) {
3909 if(qstate->env->cfg->qname_minimisation &&
3910 iq->minimisation_state != DONOT_MINIMISE_STATE) {
3911 /* Skip QNAME minimisation for next query, since that
3912 * one has to match the current query. */
3913 iq->minimisation_state = SKIP_MINIMISE_STATE;
3914 }
3915 /* for fallback we care about main answer, not additionals */
3916 /* removing that makes comparison more likely to succeed */
3917 caps_strip_reply(iq->response->rep);
3918
3919 if(iq->caps_fallback &&
3920 iq->caps_minimisation_state != iq->minimisation_state) {
3921 /* QNAME minimisation state has changed, restart caps
3922 * fallback. */
3923 iq->caps_fallback = 0;
3924 }
3925
3926 if(!iq->caps_fallback) {
3927 /* start fallback */
3928 iq->caps_fallback = 1;
3929 iq->caps_server = 0;
3930 iq->caps_reply = iq->response->rep;
3931 iq->caps_response = iq->response;
3932 iq->caps_minimisation_state = iq->minimisation_state;
3933 iq->state = QUERYTARGETS_STATE;
3934 iq->num_current_queries--;
3935 verbose(VERB_DETAIL, "Capsforid: starting fallback");
3936 goto handle_it;
3937 } else {
3938 /* check if reply is the same, otherwise, fail */
3939 if(!iq->caps_reply) {
3940 iq->caps_reply = iq->response->rep;
3941 iq->caps_response = iq->response;
3942 iq->caps_server = -1; /*become zero at ++,
3943 so that we start the full set of trials */
3944 } else if(caps_failed_rcode(iq->caps_reply) &&
3945 !caps_failed_rcode(iq->response->rep)) {
3946 /* prefer to upgrade to non-SERVFAIL */
3947 iq->caps_reply = iq->response->rep;
3948 iq->caps_response = iq->response;
3949 } else if(!caps_failed_rcode(iq->caps_reply) &&
3950 caps_failed_rcode(iq->response->rep)) {
3951 /* if we have non-SERVFAIL as answer then
3952 * we can ignore SERVFAILs for the equality
3953 * comparison */
3954 /* no instructions here, skip other else */
3955 } else if(caps_failed_rcode(iq->caps_reply) &&
3956 caps_failed_rcode(iq->response->rep)) {
3957 /* failure is same as other failure in fallbk*/
3958 /* no instructions here, skip other else */
3959 } else if(!reply_equal(iq->response->rep, iq->caps_reply,
3960 qstate->env->scratch)) {
3961 verbose(VERB_DETAIL, "Capsforid fallback: "
3962 "getting different replies, failed");
3963 outbound_list_remove(&iq->outlist, outbound);
3964 errinf(qstate, "0x20 failed, then got different replies in fallback");
3965 (void)error_response(qstate, id,
3966 LDNS_RCODE_SERVFAIL);
3967 return;
3968 }
3969 /* continue the fallback procedure at next server */
3970 iq->caps_server++;
3971 iq->state = QUERYTARGETS_STATE;
3972 iq->num_current_queries--;
3973 verbose(VERB_DETAIL, "Capsforid: reply is equal. "
3974 "go to next fallback");
3975 goto handle_it;
3976 }
3977 }
3978 iq->caps_fallback = 0; /* if we were in fallback, 0x20 is OK now */
3979
3980 handle_it:
3981 outbound_list_remove(&iq->outlist, outbound);
3982 iter_handle(qstate, iq, ie, id);
3983 }
3984
3985 void
iter_operate(struct module_qstate * qstate,enum module_ev event,int id,struct outbound_entry * outbound)3986 iter_operate(struct module_qstate* qstate, enum module_ev event, int id,
3987 struct outbound_entry* outbound)
3988 {
3989 struct iter_env* ie = (struct iter_env*)qstate->env->modinfo[id];
3990 struct iter_qstate* iq = (struct iter_qstate*)qstate->minfo[id];
3991 verbose(VERB_QUERY, "iterator[module %d] operate: extstate:%s event:%s",
3992 id, strextstate(qstate->ext_state[id]), strmodulevent(event));
3993 if(iq) log_query_info(VERB_QUERY, "iterator operate: query",
3994 &qstate->qinfo);
3995 if(iq && qstate->qinfo.qname != iq->qchase.qname)
3996 log_query_info(VERB_QUERY, "iterator operate: chased to",
3997 &iq->qchase);
3998
3999 /* perform iterator state machine */
4000 if((event == module_event_new || event == module_event_pass) &&
4001 iq == NULL) {
4002 if(!iter_new(qstate, id)) {
4003 errinf(qstate, "malloc failure, new iterator module allocation");
4004 (void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
4005 return;
4006 }
4007 iq = (struct iter_qstate*)qstate->minfo[id];
4008 process_request(qstate, iq, ie, id);
4009 return;
4010 }
4011 if(iq && event == module_event_pass) {
4012 iter_handle(qstate, iq, ie, id);
4013 return;
4014 }
4015 if(iq && outbound) {
4016 process_response(qstate, iq, ie, id, outbound, event);
4017 return;
4018 }
4019 if(event == module_event_error) {
4020 verbose(VERB_ALGO, "got called with event error, giving up");
4021 errinf(qstate, "iterator module got the error event");
4022 (void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
4023 return;
4024 }
4025
4026 log_err("bad event for iterator");
4027 errinf(qstate, "iterator module received wrong event");
4028 (void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
4029 }
4030
4031 void
iter_clear(struct module_qstate * qstate,int id)4032 iter_clear(struct module_qstate* qstate, int id)
4033 {
4034 struct iter_qstate* iq;
4035 if(!qstate)
4036 return;
4037 iq = (struct iter_qstate*)qstate->minfo[id];
4038 if(iq) {
4039 outbound_list_clear(&iq->outlist);
4040 if(iq->target_count && --iq->target_count[0] == 0)
4041 free(iq->target_count);
4042 iq->num_current_queries = 0;
4043 }
4044 qstate->minfo[id] = NULL;
4045 }
4046
4047 size_t
iter_get_mem(struct module_env * env,int id)4048 iter_get_mem(struct module_env* env, int id)
4049 {
4050 struct iter_env* ie = (struct iter_env*)env->modinfo[id];
4051 if(!ie)
4052 return 0;
4053 return sizeof(*ie) + sizeof(int)*((size_t)ie->max_dependency_depth+1)
4054 + donotq_get_mem(ie->donotq) + priv_get_mem(ie->priv);
4055 }
4056
4057 /**
4058 * The iterator function block
4059 */
4060 static struct module_func_block iter_block = {
4061 "iterator",
4062 &iter_init, &iter_deinit, &iter_operate, &iter_inform_super,
4063 &iter_clear, &iter_get_mem
4064 };
4065
4066 struct module_func_block*
iter_get_funcblock(void)4067 iter_get_funcblock(void)
4068 {
4069 return &iter_block;
4070 }
4071
4072 const char*
iter_state_to_string(enum iter_state state)4073 iter_state_to_string(enum iter_state state)
4074 {
4075 switch (state)
4076 {
4077 case INIT_REQUEST_STATE :
4078 return "INIT REQUEST STATE";
4079 case INIT_REQUEST_2_STATE :
4080 return "INIT REQUEST STATE (stage 2)";
4081 case INIT_REQUEST_3_STATE:
4082 return "INIT REQUEST STATE (stage 3)";
4083 case QUERYTARGETS_STATE :
4084 return "QUERY TARGETS STATE";
4085 case PRIME_RESP_STATE :
4086 return "PRIME RESPONSE STATE";
4087 case COLLECT_CLASS_STATE :
4088 return "COLLECT CLASS STATE";
4089 case DSNS_FIND_STATE :
4090 return "DSNS FIND STATE";
4091 case QUERY_RESP_STATE :
4092 return "QUERY RESPONSE STATE";
4093 case FINISHED_STATE :
4094 return "FINISHED RESPONSE STATE";
4095 default :
4096 return "UNKNOWN ITER STATE";
4097 }
4098 }
4099
4100 int
iter_state_is_responsestate(enum iter_state s)4101 iter_state_is_responsestate(enum iter_state s)
4102 {
4103 switch(s) {
4104 case INIT_REQUEST_STATE :
4105 case INIT_REQUEST_2_STATE :
4106 case INIT_REQUEST_3_STATE :
4107 case QUERYTARGETS_STATE :
4108 case COLLECT_CLASS_STATE :
4109 return 0;
4110 default:
4111 break;
4112 }
4113 return 1;
4114 }
4115