2 * iterator/iterator.c - iterative resolver DNS query response module
4 * Copyright (c) 2007, NLnet Labs. All rights reserved.
6 * This software is open source.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
12 * Redistributions of source code must retain the above copyright notice,
13 * this list of conditions and the following disclaimer.
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.
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.
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.
39 * This file contains a module that performs recursive iterative DNS query
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"
73 iter_init(struct module_env* env, int id)
75 struct iter_env* iter_env = (struct iter_env*)calloc(1,
76 sizeof(struct iter_env));
78 log_err("malloc failure");
81 env->modinfo[id] = (void*)iter_env;
83 lock_basic_init(&iter_env->queries_ratelimit_lock);
84 lock_protect(&iter_env->queries_ratelimit_lock,
85 &iter_env->num_queries_ratelimited,
86 sizeof(iter_env->num_queries_ratelimited));
88 if(!iter_apply_cfg(iter_env, env->cfg)) {
89 log_err("iterator: could not apply configuration settings.");
96 /** delete caps_whitelist element */
98 caps_free(struct rbnode_type* n, void* ATTR_UNUSED(d))
101 free(((struct name_tree_node*)n)->name);
107 iter_deinit(struct module_env* env, int id)
109 struct iter_env* iter_env;
110 if(!env || !env->modinfo[id])
112 iter_env = (struct iter_env*)env->modinfo[id];
113 lock_basic_destroy(&iter_env->queries_ratelimit_lock);
114 free(iter_env->target_fetch_policy);
115 priv_delete(iter_env->priv);
116 donotq_delete(iter_env->donotq);
117 if(iter_env->caps_white) {
118 traverse_postorder(iter_env->caps_white, caps_free, NULL);
119 free(iter_env->caps_white);
122 env->modinfo[id] = NULL;
125 /** new query for iterator */
127 iter_new(struct module_qstate* qstate, int id)
129 struct iter_qstate* iq = (struct iter_qstate*)regional_alloc(
130 qstate->region, sizeof(struct iter_qstate));
131 qstate->minfo[id] = iq;
134 memset(iq, 0, sizeof(*iq));
135 iq->state = INIT_REQUEST_STATE;
136 iq->final_state = FINISHED_STATE;
137 iq->an_prepend_list = NULL;
138 iq->an_prepend_last = NULL;
139 iq->ns_prepend_list = NULL;
140 iq->ns_prepend_last = NULL;
143 iq->num_target_queries = 0;
144 iq->num_current_queries = 0;
145 iq->query_restart_count = 0;
146 iq->referral_count = 0;
148 iq->ratelimit_ok = 0;
149 iq->target_count = NULL;
150 iq->wait_priming_stub = 0;
151 iq->refetch_glue = 0;
152 iq->dnssec_expected = 0;
153 iq->dnssec_lame_query = 0;
154 iq->chase_flags = qstate->query_flags;
155 /* Start with the (current) qname. */
156 iq->qchase = qstate->qinfo;
157 outbound_list_init(&iq->outlist);
158 iq->minimise_count = 0;
159 iq->minimise_timeout_count = 0;
160 if (qstate->env->cfg->qname_minimisation)
161 iq->minimisation_state = INIT_MINIMISE_STATE;
163 iq->minimisation_state = DONOT_MINIMISE_STATE;
165 memset(&iq->qinfo_out, 0, sizeof(struct query_info));
170 * Transition to the next state. This can be used to advance a currently
171 * processing event. It cannot be used to reactivate a forEvent.
173 * @param iq: iterator query state
174 * @param nextstate The state to transition to.
175 * @return true. This is so this can be called as the return value for the
176 * actual process*State() methods. (Transitioning to the next state
177 * implies further processing).
180 next_state(struct iter_qstate* iq, enum iter_state nextstate)
182 /* If transitioning to a "response" state, make sure that there is a
184 if(iter_state_is_responsestate(nextstate)) {
185 if(iq->response == NULL) {
186 log_err("transitioning to response state sans "
190 iq->state = nextstate;
195 * Transition an event to its final state. Final states always either return
196 * a result up the module chain, or reactivate a dependent event. Which
197 * final state to transition to is set in the module state for the event when
198 * it was created, and depends on the original purpose of the event.
200 * The response is stored in the qstate->buf buffer.
202 * @param iq: iterator query state
203 * @return false. This is so this method can be used as the return value for
204 * the processState methods. (Transitioning to the final state
207 final_state(struct iter_qstate* iq)
209 return next_state(iq, iq->final_state);
213 * Callback routine to handle errors in parent query states
214 * @param qstate: query state that failed.
215 * @param id: module id.
216 * @param super: super state.
219 error_supers(struct module_qstate* qstate, int id, struct module_qstate* super)
221 struct iter_qstate* super_iq = (struct iter_qstate*)super->minfo[id];
223 if(qstate->qinfo.qtype == LDNS_RR_TYPE_A ||
224 qstate->qinfo.qtype == LDNS_RR_TYPE_AAAA) {
225 /* mark address as failed. */
226 struct delegpt_ns* dpns = NULL;
227 super_iq->num_target_queries--;
229 dpns = delegpt_find_ns(super_iq->dp,
230 qstate->qinfo.qname, qstate->qinfo.qname_len);
233 verbose(VERB_ALGO, "subq error, but not interested");
234 log_query_info(VERB_ALGO, "superq", &super->qinfo);
236 delegpt_log(VERB_ALGO, super_iq->dp);
240 /* see if the failure did get (parent-lame) info */
241 if(!cache_fill_missing(super->env, super_iq->qchase.qclass,
242 super->region, super_iq->dp))
243 log_err("out of memory adding missing");
245 dpns->resolved = 1; /* mark as failed */
247 if(qstate->qinfo.qtype == LDNS_RR_TYPE_NS) {
248 /* prime failed to get delegation */
251 /* evaluate targets again */
252 super_iq->state = QUERYTARGETS_STATE;
253 /* super becomes runnable, and will process this change */
257 * Return an error to the client
258 * @param qstate: our query state
259 * @param id: module id
260 * @param rcode: error code (DNS errcode).
261 * @return: 0 for use by caller, to make notation easy, like:
262 * return error_response(..).
265 error_response(struct module_qstate* qstate, int id, int rcode)
267 verbose(VERB_QUERY, "return error response %s",
268 sldns_lookup_by_id(sldns_rcodes, rcode)?
269 sldns_lookup_by_id(sldns_rcodes, rcode)->name:"??");
270 qstate->return_rcode = rcode;
271 qstate->return_msg = NULL;
272 qstate->ext_state[id] = module_finished;
277 * Return an error to the client and cache the error code in the
278 * message cache (so per qname, qtype, qclass).
279 * @param qstate: our query state
280 * @param id: module id
281 * @param rcode: error code (DNS errcode).
282 * @return: 0 for use by caller, to make notation easy, like:
283 * return error_response(..).
286 error_response_cache(struct module_qstate* qstate, int id, int rcode)
288 if(!qstate->no_cache_store) {
290 struct reply_info err;
291 if(qstate->prefetch_leeway > NORR_TTL) {
292 verbose(VERB_ALGO, "error response for prefetch in cache");
293 /* attempt to adjust the cache entry prefetch */
294 if(dns_cache_prefetch_adjust(qstate->env, &qstate->qinfo,
295 NORR_TTL, qstate->query_flags))
296 return error_response(qstate, id, rcode);
297 /* if that fails (not in cache), fall through to store err */
299 if(qstate->env->cfg->serve_expired) {
300 /* if serving expired contents, and such content is
301 * already available, don't overwrite this servfail */
302 struct msgreply_entry* msg;
303 if((msg=msg_cache_lookup(qstate->env,
304 qstate->qinfo.qname, qstate->qinfo.qname_len,
305 qstate->qinfo.qtype, qstate->qinfo.qclass,
306 qstate->query_flags, 0, 0))
308 lock_rw_unlock(&msg->entry.lock);
309 return error_response(qstate, id, rcode);
311 /* serving expired contents, but nothing is cached
312 * at all, so the servfail cache entry is useful
313 * (stops waste of time on this servfail NORR_TTL) */
315 memset(&err, 0, sizeof(err));
316 err.flags = (uint16_t)(BIT_QR | BIT_RA);
317 FLAGS_SET_RCODE(err.flags, rcode);
320 err.prefetch_ttl = PREFETCH_TTL_CALC(err.ttl);
321 /* do not waste time trying to validate this servfail */
322 err.security = sec_status_indeterminate;
323 verbose(VERB_ALGO, "store error response in message cache");
324 iter_dns_store(qstate->env, &qstate->qinfo, &err, 0, 0, 0, NULL,
325 qstate->query_flags);
327 return error_response(qstate, id, rcode);
330 /** check if prepend item is duplicate item */
332 prepend_is_duplicate(struct ub_packed_rrset_key** sets, size_t to,
333 struct ub_packed_rrset_key* dup)
336 for(i=0; i<to; i++) {
337 if(sets[i]->rk.type == dup->rk.type &&
338 sets[i]->rk.rrset_class == dup->rk.rrset_class &&
339 sets[i]->rk.dname_len == dup->rk.dname_len &&
340 query_dname_compare(sets[i]->rk.dname, dup->rk.dname)
347 /** prepend the prepend list in the answer and authority section of dns_msg */
349 iter_prepend(struct iter_qstate* iq, struct dns_msg* msg,
350 struct regional* region)
352 struct iter_prep_list* p;
353 struct ub_packed_rrset_key** sets;
354 size_t num_an = 0, num_ns = 0;;
355 for(p = iq->an_prepend_list; p; p = p->next)
357 for(p = iq->ns_prepend_list; p; p = p->next)
359 if(num_an + num_ns == 0)
361 verbose(VERB_ALGO, "prepending %d rrsets", (int)num_an + (int)num_ns);
362 if(num_an > RR_COUNT_MAX || num_ns > RR_COUNT_MAX ||
363 msg->rep->rrset_count > RR_COUNT_MAX) return 0; /* overflow */
364 sets = regional_alloc(region, (num_an+num_ns+msg->rep->rrset_count) *
365 sizeof(struct ub_packed_rrset_key*));
370 for(p = iq->an_prepend_list; p; p = p->next) {
371 sets[num_an++] = p->rrset;
373 memcpy(sets+num_an, msg->rep->rrsets, msg->rep->an_numrrsets *
374 sizeof(struct ub_packed_rrset_key*));
377 for(p = iq->ns_prepend_list; p; p = p->next) {
378 if(prepend_is_duplicate(sets+msg->rep->an_numrrsets+num_an,
379 num_ns, p->rrset) || prepend_is_duplicate(
380 msg->rep->rrsets+msg->rep->an_numrrsets,
381 msg->rep->ns_numrrsets, p->rrset))
383 sets[msg->rep->an_numrrsets + num_an + num_ns++] = p->rrset;
385 memcpy(sets + num_an + msg->rep->an_numrrsets + num_ns,
386 msg->rep->rrsets + msg->rep->an_numrrsets,
387 (msg->rep->ns_numrrsets + msg->rep->ar_numrrsets) *
388 sizeof(struct ub_packed_rrset_key*));
390 /* NXDOMAIN rcode can stay if we prepended DNAME/CNAMEs, because
391 * this is what recursors should give. */
392 msg->rep->rrset_count += num_an + num_ns;
393 msg->rep->an_numrrsets += num_an;
394 msg->rep->ns_numrrsets += num_ns;
395 msg->rep->rrsets = sets;
400 * Find rrset in ANSWER prepend list.
401 * to avoid duplicate DNAMEs when a DNAME is traversed twice.
402 * @param iq: iterator query state.
403 * @param rrset: rrset to add.
404 * @return false if not found
407 iter_find_rrset_in_prepend_answer(struct iter_qstate* iq,
408 struct ub_packed_rrset_key* rrset)
410 struct iter_prep_list* p = iq->an_prepend_list;
412 if(ub_rrset_compare(p->rrset, rrset) == 0 &&
413 rrsetdata_equal((struct packed_rrset_data*)p->rrset
414 ->entry.data, (struct packed_rrset_data*)rrset
423 * Add rrset to ANSWER prepend list
424 * @param qstate: query state.
425 * @param iq: iterator query state.
426 * @param rrset: rrset to add.
427 * @return false on failure (malloc).
430 iter_add_prepend_answer(struct module_qstate* qstate, struct iter_qstate* iq,
431 struct ub_packed_rrset_key* rrset)
433 struct iter_prep_list* p = (struct iter_prep_list*)regional_alloc(
434 qstate->region, sizeof(struct iter_prep_list));
440 if(iq->an_prepend_last)
441 iq->an_prepend_last->next = p;
442 else iq->an_prepend_list = p;
443 iq->an_prepend_last = p;
448 * Add rrset to AUTHORITY prepend list
449 * @param qstate: query state.
450 * @param iq: iterator query state.
451 * @param rrset: rrset to add.
452 * @return false on failure (malloc).
455 iter_add_prepend_auth(struct module_qstate* qstate, struct iter_qstate* iq,
456 struct ub_packed_rrset_key* rrset)
458 struct iter_prep_list* p = (struct iter_prep_list*)regional_alloc(
459 qstate->region, sizeof(struct iter_prep_list));
465 if(iq->ns_prepend_last)
466 iq->ns_prepend_last->next = p;
467 else iq->ns_prepend_list = p;
468 iq->ns_prepend_last = p;
473 * Given a CNAME response (defined as a response containing a CNAME or DNAME
474 * that does not answer the request), process the response, modifying the
475 * state as necessary. This follows the CNAME/DNAME chain and returns the
478 * sets the new query name, after following the CNAME/DNAME chain.
479 * @param qstate: query state.
480 * @param iq: iterator query state.
481 * @param msg: the response.
482 * @param mname: returned target new query name.
483 * @param mname_len: length of mname.
484 * @return false on (malloc) error.
487 handle_cname_response(struct module_qstate* qstate, struct iter_qstate* iq,
488 struct dns_msg* msg, uint8_t** mname, size_t* mname_len)
491 /* Start with the (current) qname. */
492 *mname = iq->qchase.qname;
493 *mname_len = iq->qchase.qname_len;
495 /* Iterate over the ANSWER rrsets in order, looking for CNAMEs and
497 for(i=0; i<msg->rep->an_numrrsets; i++) {
498 struct ub_packed_rrset_key* r = msg->rep->rrsets[i];
499 /* If there is a (relevant) DNAME, add it to the list.
500 * We always expect there to be CNAME that was generated
501 * by this DNAME following, so we don't process the DNAME
503 if(ntohs(r->rk.type) == LDNS_RR_TYPE_DNAME &&
504 dname_strict_subdomain_c(*mname, r->rk.dname) &&
505 !iter_find_rrset_in_prepend_answer(iq, r)) {
506 if(!iter_add_prepend_answer(qstate, iq, r))
511 if(ntohs(r->rk.type) == LDNS_RR_TYPE_CNAME &&
512 query_dname_compare(*mname, r->rk.dname) == 0 &&
513 !iter_find_rrset_in_prepend_answer(iq, r)) {
514 /* Add this relevant CNAME rrset to the prepend list.*/
515 if(!iter_add_prepend_answer(qstate, iq, r))
517 get_cname_target(r, mname, mname_len);
520 /* Other rrsets in the section are ignored. */
522 /* add authority rrsets to authority prepend, for wildcarded CNAMEs */
523 for(i=msg->rep->an_numrrsets; i<msg->rep->an_numrrsets +
524 msg->rep->ns_numrrsets; i++) {
525 struct ub_packed_rrset_key* r = msg->rep->rrsets[i];
526 /* only add NSEC/NSEC3, as they may be needed for validation */
527 if(ntohs(r->rk.type) == LDNS_RR_TYPE_NSEC ||
528 ntohs(r->rk.type) == LDNS_RR_TYPE_NSEC3) {
529 if(!iter_add_prepend_auth(qstate, iq, r))
536 /** see if last resort is possible - does config allow queries to parent */
538 can_have_last_resort(struct module_env* env, uint8_t* nm, size_t nmlen,
539 uint16_t qclass, struct delegpt** retdp)
541 struct delegpt* fwddp;
542 struct iter_hints_stub* stub;
543 int labs = dname_count_labels(nm);
544 /* do not process a last resort (the parent side) if a stub
545 * or forward is configured, because we do not want to go 'above'
546 * the configured servers */
547 if(!dname_is_root(nm) && (stub = (struct iter_hints_stub*)
548 name_tree_find(&env->hints->tree, nm, nmlen, labs, qclass)) &&
549 /* has_parent side is turned off for stub_first, where we
550 * are allowed to go to the parent */
551 stub->dp->has_parent_side_NS) {
552 if(retdp) *retdp = stub->dp;
555 if((fwddp = forwards_find(env->fwds, nm, qclass)) &&
556 /* has_parent_side is turned off for forward_first, where
557 * we are allowed to go to the parent */
558 fwddp->has_parent_side_NS) {
559 if(retdp) *retdp = fwddp;
565 /** see if target name is caps-for-id whitelisted */
567 is_caps_whitelisted(struct iter_env* ie, struct iter_qstate* iq)
569 if(!ie->caps_white) return 0; /* no whitelist, or no capsforid */
570 return name_tree_lookup(ie->caps_white, iq->qchase.qname,
571 iq->qchase.qname_len, dname_count_labels(iq->qchase.qname),
572 iq->qchase.qclass) != NULL;
575 /** create target count structure for this query */
577 target_count_create(struct iter_qstate* iq)
579 if(!iq->target_count) {
580 iq->target_count = (int*)calloc(2, sizeof(int));
581 /* if calloc fails we simply do not track this number */
583 iq->target_count[0] = 1;
588 target_count_increase(struct iter_qstate* iq, int num)
590 target_count_create(iq);
592 iq->target_count[1] += num;
596 * Generate a subrequest.
597 * Generate a local request event. Local events are tied to this module, and
598 * have a corresponding (first tier) event that is waiting for this event to
599 * resolve to continue.
601 * @param qname The query name for this request.
602 * @param qnamelen length of qname
603 * @param qtype The query type for this request.
604 * @param qclass The query class for this request.
605 * @param qstate The event that is generating this event.
606 * @param id: module id.
607 * @param iq: The iterator state that is generating this event.
608 * @param initial_state The initial response state (normally this
609 * is QUERY_RESP_STATE, unless it is known that the request won't
610 * need iterative processing
611 * @param finalstate The final state for the response to this request.
612 * @param subq_ret: if newly allocated, the subquerystate, or NULL if it does
613 * not need initialisation.
614 * @param v: if true, validation is done on the subquery.
615 * @return false on error (malloc).
618 generate_sub_request(uint8_t* qname, size_t qnamelen, uint16_t qtype,
619 uint16_t qclass, struct module_qstate* qstate, int id,
620 struct iter_qstate* iq, enum iter_state initial_state,
621 enum iter_state finalstate, struct module_qstate** subq_ret, int v)
623 struct module_qstate* subq = NULL;
624 struct iter_qstate* subiq = NULL;
625 uint16_t qflags = 0; /* OPCODE QUERY, no flags */
626 struct query_info qinf;
627 int prime = (finalstate == PRIME_RESP_STATE)?1:0;
630 qinf.qname_len = qnamelen;
632 qinf.qclass = qclass;
633 qinf.local_alias = NULL;
635 /* RD should be set only when sending the query back through the INIT
637 if(initial_state == INIT_REQUEST_STATE)
639 /* We set the CD flag so we can send this through the "head" of
640 * the resolution chain, which might have a validator. We are
641 * uninterested in validating things not on the direct resolution
648 /* attach subquery, lookup existing or make a new one */
649 fptr_ok(fptr_whitelist_modenv_attach_sub(qstate->env->attach_sub));
650 if(!(*qstate->env->attach_sub)(qstate, &qinf, qflags, prime, valrec,
656 /* initialise the new subquery */
658 subq->ext_state[id] = module_state_initial;
659 subq->minfo[id] = regional_alloc(subq->region,
660 sizeof(struct iter_qstate));
661 if(!subq->minfo[id]) {
662 log_err("init subq: out of memory");
663 fptr_ok(fptr_whitelist_modenv_kill_sub(
664 qstate->env->kill_sub));
665 (*qstate->env->kill_sub)(subq);
668 subiq = (struct iter_qstate*)subq->minfo[id];
669 memset(subiq, 0, sizeof(*subiq));
670 subiq->num_target_queries = 0;
671 target_count_create(iq);
672 subiq->target_count = iq->target_count;
674 iq->target_count[0] ++; /* extra reference */
675 subiq->num_current_queries = 0;
676 subiq->depth = iq->depth+1;
677 outbound_list_init(&subiq->outlist);
678 subiq->state = initial_state;
679 subiq->final_state = finalstate;
680 subiq->qchase = subq->qinfo;
681 subiq->chase_flags = subq->query_flags;
682 subiq->refetch_glue = 0;
683 if(qstate->env->cfg->qname_minimisation)
684 subiq->minimisation_state = INIT_MINIMISE_STATE;
686 subiq->minimisation_state = DONOT_MINIMISE_STATE;
687 memset(&subiq->qinfo_out, 0, sizeof(struct query_info));
693 * Generate and send a root priming request.
694 * @param qstate: the qtstate that triggered the need to prime.
695 * @param iq: iterator query state.
696 * @param id: module id.
697 * @param qclass: the class to prime.
698 * @return 0 on failure
701 prime_root(struct module_qstate* qstate, struct iter_qstate* iq, int id,
705 struct module_qstate* subq;
706 verbose(VERB_DETAIL, "priming . %s NS",
707 sldns_lookup_by_id(sldns_rr_classes, (int)qclass)?
708 sldns_lookup_by_id(sldns_rr_classes, (int)qclass)->name:"??");
709 dp = hints_lookup_root(qstate->env->hints, qclass);
711 verbose(VERB_ALGO, "Cannot prime due to lack of hints");
714 /* Priming requests start at the QUERYTARGETS state, skipping
715 * the normal INIT state logic (which would cause an infloop). */
716 if(!generate_sub_request((uint8_t*)"\000", 1, LDNS_RR_TYPE_NS,
717 qclass, qstate, id, iq, QUERYTARGETS_STATE, PRIME_RESP_STATE,
719 verbose(VERB_ALGO, "could not prime root");
723 struct iter_qstate* subiq =
724 (struct iter_qstate*)subq->minfo[id];
725 /* Set the initial delegation point to the hint.
726 * copy dp, it is now part of the root prime query.
727 * dp was part of in the fixed hints structure. */
728 subiq->dp = delegpt_copy(dp, subq->region);
730 log_err("out of memory priming root, copydp");
731 fptr_ok(fptr_whitelist_modenv_kill_sub(
732 qstate->env->kill_sub));
733 (*qstate->env->kill_sub)(subq);
736 /* there should not be any target queries. */
737 subiq->num_target_queries = 0;
738 subiq->dnssec_expected = iter_indicates_dnssec(
739 qstate->env, subiq->dp, NULL, subq->qinfo.qclass);
742 /* this module stops, our submodule starts, and does the query. */
743 qstate->ext_state[id] = module_wait_subquery;
748 * Generate and process a stub priming request. This method tests for the
749 * need to prime a stub zone, so it is safe to call for every request.
751 * @param qstate: the qtstate that triggered the need to prime.
752 * @param iq: iterator query state.
753 * @param id: module id.
754 * @param qname: request name.
755 * @param qclass: request class.
756 * @return true if a priming subrequest was made, false if not. The will only
757 * issue a priming request if it detects an unprimed stub.
758 * Uses value of 2 to signal during stub-prime in root-prime situation
759 * that a noprime-stub is available and resolution can continue.
762 prime_stub(struct module_qstate* qstate, struct iter_qstate* iq, int id,
763 uint8_t* qname, uint16_t qclass)
765 /* Lookup the stub hint. This will return null if the stub doesn't
766 * need to be re-primed. */
767 struct iter_hints_stub* stub;
768 struct delegpt* stub_dp;
769 struct module_qstate* subq;
772 stub = hints_lookup_stub(qstate->env->hints, qname, qclass, iq->dp);
773 /* The stub (if there is one) does not need priming. */
777 /* if we have an auth_zone dp, and stub is equal, don't prime stub
778 * yet, unless we want to fallback and avoid the auth_zone */
779 if(!iq->auth_zone_avoid && iq->dp && iq->dp->auth_dp &&
780 query_dname_compare(iq->dp->name, stub_dp->name) == 0)
783 /* is it a noprime stub (always use) */
786 if(iq->dp == NULL) r = 2;
787 /* copy the dp out of the fixed hints structure, so that
788 * it can be changed when servicing this query */
789 iq->dp = delegpt_copy(stub_dp, qstate->region);
791 log_err("out of memory priming stub");
792 (void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
793 return 1; /* return 1 to make module stop, with error */
795 log_nametypeclass(VERB_DETAIL, "use stub", stub_dp->name,
796 LDNS_RR_TYPE_NS, qclass);
800 /* Otherwise, we need to (re)prime the stub. */
801 log_nametypeclass(VERB_DETAIL, "priming stub", stub_dp->name,
802 LDNS_RR_TYPE_NS, qclass);
804 /* Stub priming events start at the QUERYTARGETS state to avoid the
805 * redundant INIT state processing. */
806 if(!generate_sub_request(stub_dp->name, stub_dp->namelen,
807 LDNS_RR_TYPE_NS, qclass, qstate, id, iq,
808 QUERYTARGETS_STATE, PRIME_RESP_STATE, &subq, 0)) {
809 verbose(VERB_ALGO, "could not prime stub");
810 (void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
811 return 1; /* return 1 to make module stop, with error */
814 struct iter_qstate* subiq =
815 (struct iter_qstate*)subq->minfo[id];
817 /* Set the initial delegation point to the hint. */
818 /* make copy to avoid use of stub dp by different qs/threads */
819 subiq->dp = delegpt_copy(stub_dp, subq->region);
821 log_err("out of memory priming stub, copydp");
822 fptr_ok(fptr_whitelist_modenv_kill_sub(
823 qstate->env->kill_sub));
824 (*qstate->env->kill_sub)(subq);
825 (void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
826 return 1; /* return 1 to make module stop, with error */
828 /* there should not be any target queries -- although there
829 * wouldn't be anyway, since stub hints never have
830 * missing targets. */
831 subiq->num_target_queries = 0;
832 subiq->wait_priming_stub = 1;
833 subiq->dnssec_expected = iter_indicates_dnssec(
834 qstate->env, subiq->dp, NULL, subq->qinfo.qclass);
837 /* this module stops, our submodule starts, and does the query. */
838 qstate->ext_state[id] = module_wait_subquery;
843 * Generate a delegation point for an auth zone (unless cached dp is better)
844 * false on alloc failure.
847 auth_zone_delegpt(struct module_qstate* qstate, struct iter_qstate* iq,
848 uint8_t* delname, size_t delnamelen)
851 if(iq->auth_zone_avoid)
854 delname = iq->qchase.qname;
855 delnamelen = iq->qchase.qname_len;
857 lock_rw_rdlock(&qstate->env->auth_zones->lock);
858 z = auth_zones_find_zone(qstate->env->auth_zones, delname, delnamelen,
859 qstate->qinfo.qclass);
861 lock_rw_unlock(&qstate->env->auth_zones->lock);
864 lock_rw_rdlock(&z->lock);
865 lock_rw_unlock(&qstate->env->auth_zones->lock);
866 if(z->for_upstream) {
867 if(iq->dp && query_dname_compare(z->name, iq->dp->name) == 0
868 && iq->dp->auth_dp && qstate->blacklist &&
869 z->fallback_enabled) {
870 /* cache is blacklisted and fallback, and we
871 * already have an auth_zone dp */
872 if(verbosity>=VERB_ALGO) {
874 dname_str(z->name, buf);
875 verbose(VERB_ALGO, "auth_zone %s "
876 "fallback because cache blacklisted",
879 lock_rw_unlock(&z->lock);
883 if(iq->dp==NULL || dname_subdomain_c(z->name, iq->dp->name)) {
885 if(qstate->blacklist && z->fallback_enabled) {
886 /* cache is blacklisted because of a DNSSEC
887 * validation failure, and the zone allows
888 * fallback to the internet, query there. */
889 if(verbosity>=VERB_ALGO) {
891 dname_str(z->name, buf);
892 verbose(VERB_ALGO, "auth_zone %s "
893 "fallback because cache blacklisted",
896 lock_rw_unlock(&z->lock);
899 dp = (struct delegpt*)regional_alloc_zero(
900 qstate->region, sizeof(*dp));
902 log_err("alloc failure");
903 if(z->fallback_enabled) {
904 lock_rw_unlock(&z->lock);
905 return 1; /* just fallback */
907 lock_rw_unlock(&z->lock);
910 dp->name = regional_alloc_init(qstate->region,
911 z->name, z->namelen);
913 log_err("alloc failure");
914 if(z->fallback_enabled) {
915 lock_rw_unlock(&z->lock);
916 return 1; /* just fallback */
918 lock_rw_unlock(&z->lock);
921 dp->namelen = z->namelen;
922 dp->namelabs = z->namelabs;
928 lock_rw_unlock(&z->lock);
933 * Generate A and AAAA checks for glue that is in-zone for the referral
934 * we just got to obtain authoritative information on the addresses.
936 * @param qstate: the qtstate that triggered the need to prime.
937 * @param iq: iterator query state.
938 * @param id: module id.
941 generate_a_aaaa_check(struct module_qstate* qstate, struct iter_qstate* iq,
944 struct iter_env* ie = (struct iter_env*)qstate->env->modinfo[id];
945 struct module_qstate* subq;
947 struct reply_info* rep = iq->response->rep;
948 struct ub_packed_rrset_key* s;
951 if(iq->depth == ie->max_dependency_depth)
953 /* walk through additional, and check if in-zone,
954 * only relevant A, AAAA are left after scrub anyway */
955 for(i=rep->an_numrrsets+rep->ns_numrrsets; i<rep->rrset_count; i++) {
957 /* check *ALL* addresses that are transmitted in additional*/
958 /* is it an address ? */
959 if( !(ntohs(s->rk.type)==LDNS_RR_TYPE_A ||
960 ntohs(s->rk.type)==LDNS_RR_TYPE_AAAA)) {
963 /* is this query the same as the A/AAAA check for it */
964 if(qstate->qinfo.qtype == ntohs(s->rk.type) &&
965 qstate->qinfo.qclass == ntohs(s->rk.rrset_class) &&
966 query_dname_compare(qstate->qinfo.qname,
968 (qstate->query_flags&BIT_RD) &&
969 !(qstate->query_flags&BIT_CD))
972 /* generate subrequest for it */
973 log_nametypeclass(VERB_ALGO, "schedule addr fetch",
974 s->rk.dname, ntohs(s->rk.type),
975 ntohs(s->rk.rrset_class));
976 if(!generate_sub_request(s->rk.dname, s->rk.dname_len,
977 ntohs(s->rk.type), ntohs(s->rk.rrset_class),
979 INIT_REQUEST_STATE, FINISHED_STATE, &subq, 1)) {
980 verbose(VERB_ALGO, "could not generate addr check");
983 /* ignore subq - not need for more init */
988 * Generate a NS check request to obtain authoritative information
991 * @param qstate: the qtstate that triggered the need to prime.
992 * @param iq: iterator query state.
993 * @param id: module id.
996 generate_ns_check(struct module_qstate* qstate, struct iter_qstate* iq, int id)
998 struct iter_env* ie = (struct iter_env*)qstate->env->modinfo[id];
999 struct module_qstate* subq;
1002 if(iq->depth == ie->max_dependency_depth)
1004 if(!can_have_last_resort(qstate->env, iq->dp->name, iq->dp->namelen,
1005 iq->qchase.qclass, NULL))
1007 /* is this query the same as the nscheck? */
1008 if(qstate->qinfo.qtype == LDNS_RR_TYPE_NS &&
1009 query_dname_compare(iq->dp->name, qstate->qinfo.qname)==0 &&
1010 (qstate->query_flags&BIT_RD) && !(qstate->query_flags&BIT_CD)){
1011 /* spawn off A, AAAA queries for in-zone glue to check */
1012 generate_a_aaaa_check(qstate, iq, id);
1015 /* no need to get the NS record for DS, it is above the zonecut */
1016 if(qstate->qinfo.qtype == LDNS_RR_TYPE_DS)
1019 log_nametypeclass(VERB_ALGO, "schedule ns fetch",
1020 iq->dp->name, LDNS_RR_TYPE_NS, iq->qchase.qclass);
1021 if(!generate_sub_request(iq->dp->name, iq->dp->namelen,
1022 LDNS_RR_TYPE_NS, iq->qchase.qclass, qstate, id, iq,
1023 INIT_REQUEST_STATE, FINISHED_STATE, &subq, 1)) {
1024 verbose(VERB_ALGO, "could not generate ns check");
1028 struct iter_qstate* subiq =
1029 (struct iter_qstate*)subq->minfo[id];
1031 /* make copy to avoid use of stub dp by different qs/threads */
1032 /* refetch glue to start higher up the tree */
1033 subiq->refetch_glue = 1;
1034 subiq->dp = delegpt_copy(iq->dp, subq->region);
1036 log_err("out of memory generating ns check, copydp");
1037 fptr_ok(fptr_whitelist_modenv_kill_sub(
1038 qstate->env->kill_sub));
1039 (*qstate->env->kill_sub)(subq);
1046 * Generate a DNSKEY prefetch query to get the DNSKEY for the DS record we
1047 * just got in a referral (where we have dnssec_expected, thus have trust
1048 * anchors above it). Note that right after calling this routine the
1049 * iterator detached subqueries (because of following the referral), and thus
1050 * the DNSKEY query becomes detached, its return stored in the cache for
1051 * later lookup by the validator. This cache lookup by the validator avoids
1052 * the roundtrip incurred by the DNSKEY query. The DNSKEY query is now
1053 * performed at about the same time the original query is sent to the domain,
1054 * thus the two answers are likely to be returned at about the same time,
1055 * saving a roundtrip from the validated lookup.
1057 * @param qstate: the qtstate that triggered the need to prime.
1058 * @param iq: iterator query state.
1059 * @param id: module id.
1062 generate_dnskey_prefetch(struct module_qstate* qstate,
1063 struct iter_qstate* iq, int id)
1065 struct module_qstate* subq;
1068 /* is this query the same as the prefetch? */
1069 if(qstate->qinfo.qtype == LDNS_RR_TYPE_DNSKEY &&
1070 query_dname_compare(iq->dp->name, qstate->qinfo.qname)==0 &&
1071 (qstate->query_flags&BIT_RD) && !(qstate->query_flags&BIT_CD)){
1075 /* if the DNSKEY is in the cache this lookup will stop quickly */
1076 log_nametypeclass(VERB_ALGO, "schedule dnskey prefetch",
1077 iq->dp->name, LDNS_RR_TYPE_DNSKEY, iq->qchase.qclass);
1078 if(!generate_sub_request(iq->dp->name, iq->dp->namelen,
1079 LDNS_RR_TYPE_DNSKEY, iq->qchase.qclass, qstate, id, iq,
1080 INIT_REQUEST_STATE, FINISHED_STATE, &subq, 0)) {
1081 /* we'll be slower, but it'll work */
1082 verbose(VERB_ALGO, "could not generate dnskey prefetch");
1086 struct iter_qstate* subiq =
1087 (struct iter_qstate*)subq->minfo[id];
1088 /* this qstate has the right delegation for the dnskey lookup*/
1089 /* make copy to avoid use of stub dp by different qs/threads */
1090 subiq->dp = delegpt_copy(iq->dp, subq->region);
1091 /* if !subiq->dp, it'll start from the cache, no problem */
1096 * See if the query needs forwarding.
1098 * @param qstate: query state.
1099 * @param iq: iterator query state.
1100 * @return true if the request is forwarded, false if not.
1101 * If returns true but, iq->dp is NULL then a malloc failure occurred.
1104 forward_request(struct module_qstate* qstate, struct iter_qstate* iq)
1107 uint8_t* delname = iq->qchase.qname;
1108 size_t delnamelen = iq->qchase.qname_len;
1109 if(iq->refetch_glue) {
1110 delname = iq->dp->name;
1111 delnamelen = iq->dp->namelen;
1113 /* strip one label off of DS query to lookup higher for it */
1114 if( (iq->qchase.qtype == LDNS_RR_TYPE_DS || iq->refetch_glue)
1115 && !dname_is_root(iq->qchase.qname))
1116 dname_remove_label(&delname, &delnamelen);
1117 dp = forwards_lookup(qstate->env->fwds, delname, iq->qchase.qclass);
1120 /* send recursion desired to forward addr */
1121 iq->chase_flags |= BIT_RD;
1122 iq->dp = delegpt_copy(dp, qstate->region);
1123 /* iq->dp checked by caller */
1124 verbose(VERB_ALGO, "forwarding request");
1129 * Process the initial part of the request handling. This state roughly
1130 * corresponds to resolver algorithms steps 1 (find answer in cache) and 2
1131 * (find the best servers to ask).
1133 * Note that all requests start here, and query restarts revisit this state.
1135 * This state either generates: 1) a response, from cache or error, 2) a
1136 * priming event, or 3) forwards the request to the next state (init2,
1139 * @param qstate: query state.
1140 * @param iq: iterator query state.
1141 * @param ie: iterator shared global environment.
1142 * @param id: module id.
1143 * @return true if the event needs more request processing immediately,
1147 processInitRequest(struct module_qstate* qstate, struct iter_qstate* iq,
1148 struct iter_env* ie, int id)
1152 struct dns_msg* msg = NULL;
1154 log_query_info(VERB_DETAIL, "resolving", &qstate->qinfo);
1157 /* We enforce a maximum number of query restarts. This is primarily a
1158 * cheap way to prevent CNAME loops. */
1159 if(iq->query_restart_count > MAX_RESTART_COUNT) {
1160 verbose(VERB_QUERY, "request has exceeded the maximum number"
1161 " of query restarts with %d", iq->query_restart_count);
1162 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1165 /* We enforce a maximum recursion/dependency depth -- in general,
1166 * this is unnecessary for dependency loops (although it will
1167 * catch those), but it provides a sensible limit to the amount
1168 * of work required to answer a given query. */
1169 verbose(VERB_ALGO, "request has dependency depth of %d", iq->depth);
1170 if(iq->depth > ie->max_dependency_depth) {
1171 verbose(VERB_QUERY, "request has exceeded the maximum "
1172 "dependency depth with depth of %d", iq->depth);
1173 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1176 /* If the request is qclass=ANY, setup to generate each class */
1177 if(qstate->qinfo.qclass == LDNS_RR_CLASS_ANY) {
1178 iq->qchase.qclass = 0;
1179 return next_state(iq, COLLECT_CLASS_STATE);
1183 * If we are restricted by a forward-zone or a stub-zone, we
1184 * can't re-fetch glue for this delegation point.
1185 * we won’t try to re-fetch glue if the iq->dp is null.
1187 if (iq->refetch_glue &&
1189 !can_have_last_resort(qstate->env, iq->dp->name,
1190 iq->dp->namelen, iq->qchase.qclass, NULL)) {
1191 iq->refetch_glue = 0;
1194 /* Resolver Algorithm Step 1 -- Look for the answer in local data. */
1196 /* This either results in a query restart (CNAME cache response), a
1197 * terminating response (ANSWER), or a cache miss (null). */
1199 if(qstate->blacklist) {
1200 /* if cache, or anything else, was blacklisted then
1201 * getting older results from cache is a bad idea, no cache */
1202 verbose(VERB_ALGO, "cache blacklisted, going to the network");
1204 } else if(!qstate->no_cache_lookup) {
1205 msg = dns_cache_lookup(qstate->env, iq->qchase.qname,
1206 iq->qchase.qname_len, iq->qchase.qtype,
1207 iq->qchase.qclass, qstate->query_flags,
1208 qstate->region, qstate->env->scratch, 0);
1209 if(!msg && qstate->env->neg_cache &&
1210 iter_qname_indicates_dnssec(qstate->env, &iq->qchase)) {
1211 /* lookup in negative cache; may result in
1212 * NOERROR/NODATA or NXDOMAIN answers that need validation */
1213 msg = val_neg_getmsg(qstate->env->neg_cache, &iq->qchase,
1214 qstate->region, qstate->env->rrset_cache,
1215 qstate->env->scratch_buffer,
1216 *qstate->env->now, 1/*add SOA*/, NULL,
1219 /* item taken from cache does not match our query name, thus
1220 * security needs to be re-examined later */
1221 if(msg && query_dname_compare(qstate->qinfo.qname,
1222 iq->qchase.qname) != 0)
1223 msg->rep->security = sec_status_unchecked;
1226 /* handle positive cache response */
1227 enum response_type type = response_type_from_cache(msg,
1229 if(verbosity >= VERB_ALGO) {
1230 log_dns_msg("msg from cache lookup", &msg->qinfo,
1232 verbose(VERB_ALGO, "msg ttl is %d, prefetch ttl %d",
1234 (int)msg->rep->prefetch_ttl);
1237 if(type == RESPONSE_TYPE_CNAME) {
1240 verbose(VERB_ALGO, "returning CNAME response from "
1242 if(!handle_cname_response(qstate, iq, msg,
1244 return error_response(qstate, id,
1245 LDNS_RCODE_SERVFAIL);
1246 iq->qchase.qname = sname;
1247 iq->qchase.qname_len = slen;
1248 /* This *is* a query restart, even if it is a cheap
1251 iq->refetch_glue = 0;
1252 iq->query_restart_count++;
1254 sock_list_insert(&qstate->reply_origin, NULL, 0, qstate->region);
1255 if(qstate->env->cfg->qname_minimisation)
1256 iq->minimisation_state = INIT_MINIMISE_STATE;
1257 return next_state(iq, INIT_REQUEST_STATE);
1260 /* if from cache, NULL, else insert 'cache IP' len=0 */
1261 if(qstate->reply_origin)
1262 sock_list_insert(&qstate->reply_origin, NULL, 0, qstate->region);
1263 /* it is an answer, response, to final state */
1264 verbose(VERB_ALGO, "returning answer from cache.");
1266 return final_state(iq);
1269 /* attempt to forward the request */
1270 if(forward_request(qstate, iq))
1273 log_err("alloc failure for forward dp");
1274 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1276 iq->refetch_glue = 0;
1277 iq->minimisation_state = DONOT_MINIMISE_STATE;
1278 /* the request has been forwarded.
1279 * forwarded requests need to be immediately sent to the
1280 * next state, QUERYTARGETS. */
1281 return next_state(iq, QUERYTARGETS_STATE);
1284 /* Resolver Algorithm Step 2 -- find the "best" servers. */
1286 /* first, adjust for DS queries. To avoid the grandparent problem,
1287 * we just look for the closest set of server to the parent of qname.
1288 * When re-fetching glue we also need to ask the parent.
1290 if(iq->refetch_glue) {
1292 log_err("internal or malloc fail: no dp for refetch");
1293 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1295 delname = iq->dp->name;
1296 delnamelen = iq->dp->namelen;
1298 delname = iq->qchase.qname;
1299 delnamelen = iq->qchase.qname_len;
1301 if(iq->qchase.qtype == LDNS_RR_TYPE_DS || iq->refetch_glue ||
1302 (iq->qchase.qtype == LDNS_RR_TYPE_NS && qstate->prefetch_leeway
1303 && can_have_last_resort(qstate->env, delname, delnamelen, iq->qchase.qclass, NULL))) {
1304 /* remove first label from delname, root goes to hints,
1305 * but only to fetch glue, not for qtype=DS. */
1306 /* also when prefetching an NS record, fetch it again from
1307 * its parent, just as if it expired, so that you do not
1308 * get stuck on an older nameserver that gives old NSrecords */
1309 if(dname_is_root(delname) && (iq->refetch_glue ||
1310 (iq->qchase.qtype == LDNS_RR_TYPE_NS &&
1311 qstate->prefetch_leeway)))
1312 delname = NULL; /* go to root priming */
1313 else dname_remove_label(&delname, &delnamelen);
1315 /* delname is the name to lookup a delegation for. If NULL rootprime */
1318 /* Lookup the delegation in the cache. If null, then the
1319 * cache needs to be primed for the qclass. */
1321 iq->dp = dns_cache_find_delegation(qstate->env, delname,
1322 delnamelen, iq->qchase.qtype, iq->qchase.qclass,
1323 qstate->region, &iq->deleg_msg,
1324 *qstate->env->now+qstate->prefetch_leeway);
1327 /* If the cache has returned nothing, then we have a
1328 * root priming situation. */
1329 if(iq->dp == NULL) {
1331 /* if under auth zone, no prime needed */
1332 if(!auth_zone_delegpt(qstate, iq, delname, delnamelen))
1333 return error_response(qstate, id,
1334 LDNS_RCODE_SERVFAIL);
1335 if(iq->dp) /* use auth zone dp */
1336 return next_state(iq, INIT_REQUEST_2_STATE);
1337 /* if there is a stub, then no root prime needed */
1338 r = prime_stub(qstate, iq, id, delname,
1341 break; /* got noprime-stub-zone, continue */
1343 return 0; /* stub prime request made */
1344 if(forwards_lookup_root(qstate->env->fwds,
1345 iq->qchase.qclass)) {
1346 /* forward zone root, no root prime needed */
1347 /* fill in some dp - safety belt */
1348 iq->dp = hints_lookup_root(qstate->env->hints,
1351 log_err("internal error: no hints dp");
1352 return error_response(qstate, id,
1353 LDNS_RCODE_SERVFAIL);
1355 iq->dp = delegpt_copy(iq->dp, qstate->region);
1357 log_err("out of memory in safety belt");
1358 return error_response(qstate, id,
1359 LDNS_RCODE_SERVFAIL);
1361 return next_state(iq, INIT_REQUEST_2_STATE);
1363 /* Note that the result of this will set a new
1364 * DelegationPoint based on the result of priming. */
1365 if(!prime_root(qstate, iq, id, iq->qchase.qclass))
1366 return error_response(qstate, id,
1367 LDNS_RCODE_REFUSED);
1369 /* priming creates and sends a subordinate query, with
1370 * this query as the parent. So further processing for
1371 * this event will stop until reactivated by the
1372 * results of priming. */
1375 if(!iq->ratelimit_ok && qstate->prefetch_leeway)
1376 iq->ratelimit_ok = 1; /* allow prefetches, this keeps
1377 otherwise valid data in the cache */
1378 if(!iq->ratelimit_ok && infra_ratelimit_exceeded(
1379 qstate->env->infra_cache, iq->dp->name,
1380 iq->dp->namelen, *qstate->env->now)) {
1381 /* and increment the rate, so that the rate for time
1382 * now will also exceed the rate, keeping cache fresh */
1383 (void)infra_ratelimit_inc(qstate->env->infra_cache,
1384 iq->dp->name, iq->dp->namelen,
1386 /* see if we are passed through with slip factor */
1387 if(qstate->env->cfg->ratelimit_factor != 0 &&
1388 ub_random_max(qstate->env->rnd,
1389 qstate->env->cfg->ratelimit_factor) == 1) {
1390 iq->ratelimit_ok = 1;
1391 log_nametypeclass(VERB_ALGO, "ratelimit allowed through for "
1392 "delegation point", iq->dp->name,
1393 LDNS_RR_TYPE_NS, LDNS_RR_CLASS_IN);
1395 lock_basic_lock(&ie->queries_ratelimit_lock);
1396 ie->num_queries_ratelimited++;
1397 lock_basic_unlock(&ie->queries_ratelimit_lock);
1398 log_nametypeclass(VERB_ALGO, "ratelimit exceeded with "
1399 "delegation point", iq->dp->name,
1400 LDNS_RR_TYPE_NS, LDNS_RR_CLASS_IN);
1401 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1405 /* see if this dp not useless.
1407 * o all NS items are required glue.
1408 * or the query is for NS item that is required glue.
1409 * o no addresses are provided.
1411 * Instead, go up one level, and try to get even further
1412 * If the root was useless, use safety belt information.
1413 * Only check cache returns, because replies for servers
1414 * could be useless but lead to loops (bumping into the
1415 * same server reply) if useless-checked.
1417 if(iter_dp_is_useless(&qstate->qinfo, qstate->query_flags,
1419 struct delegpt* retdp = NULL;
1420 if(!can_have_last_resort(qstate->env, iq->dp->name, iq->dp->namelen, iq->qchase.qclass, &retdp)) {
1422 verbose(VERB_QUERY, "cache has stub "
1423 "or fwd but no addresses, "
1424 "fallback to config");
1425 iq->dp = delegpt_copy(retdp,
1428 log_err("out of memory in "
1429 "stub/fwd fallback");
1430 return error_response(qstate,
1431 id, LDNS_RCODE_SERVFAIL);
1435 verbose(VERB_ALGO, "useless dp "
1436 "but cannot go up, servfail");
1437 delegpt_log(VERB_ALGO, iq->dp);
1438 return error_response(qstate, id,
1439 LDNS_RCODE_SERVFAIL);
1441 if(dname_is_root(iq->dp->name)) {
1442 /* use safety belt */
1443 verbose(VERB_QUERY, "Cache has root NS but "
1444 "no addresses. Fallback to the safety belt.");
1445 iq->dp = hints_lookup_root(qstate->env->hints,
1447 /* note deleg_msg is from previous lookup,
1448 * but RD is on, so it is not used */
1450 log_err("internal error: no hints dp");
1451 return error_response(qstate, id,
1452 LDNS_RCODE_REFUSED);
1454 iq->dp = delegpt_copy(iq->dp, qstate->region);
1456 log_err("out of memory in safety belt");
1457 return error_response(qstate, id,
1458 LDNS_RCODE_SERVFAIL);
1463 "cache delegation was useless:");
1464 delegpt_log(VERB_ALGO, iq->dp);
1466 delname = iq->dp->name;
1467 delnamelen = iq->dp->namelen;
1468 dname_remove_label(&delname, &delnamelen);
1473 verbose(VERB_ALGO, "cache delegation returns delegpt");
1474 delegpt_log(VERB_ALGO, iq->dp);
1476 /* Otherwise, set the current delegation point and move on to the
1478 return next_state(iq, INIT_REQUEST_2_STATE);
1482 * Process the second part of the initial request handling. This state
1483 * basically exists so that queries that generate root priming events have
1484 * the same init processing as ones that do not. Request events that reach
1485 * this state must have a valid currentDelegationPoint set.
1487 * This part is primarily handling stub zone priming. Events that reach this
1488 * state must have a current delegation point.
1490 * @param qstate: query state.
1491 * @param iq: iterator query state.
1492 * @param id: module id.
1493 * @return true if the event needs more request processing immediately,
1497 processInitRequest2(struct module_qstate* qstate, struct iter_qstate* iq,
1502 log_query_info(VERB_QUERY, "resolving (init part 2): ",
1505 delname = iq->qchase.qname;
1506 delnamelen = iq->qchase.qname_len;
1507 if(iq->refetch_glue) {
1508 struct iter_hints_stub* stub;
1510 log_err("internal or malloc fail: no dp for refetch");
1511 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1513 /* Do not send queries above stub, do not set delname to dp if
1514 * this is above stub without stub-first. */
1515 stub = hints_lookup_stub(
1516 qstate->env->hints, iq->qchase.qname, iq->qchase.qclass,
1518 if(!stub || !stub->dp->has_parent_side_NS ||
1519 dname_subdomain_c(iq->dp->name, stub->dp->name)) {
1520 delname = iq->dp->name;
1521 delnamelen = iq->dp->namelen;
1524 if(iq->qchase.qtype == LDNS_RR_TYPE_DS || iq->refetch_glue) {
1525 if(!dname_is_root(delname))
1526 dname_remove_label(&delname, &delnamelen);
1527 iq->refetch_glue = 0; /* if CNAME causes restart, no refetch */
1530 /* see if we have an auth zone to answer from, improves dp from cache
1531 * (if any dp from cache) with auth zone dp, if that is lower */
1532 if(!auth_zone_delegpt(qstate, iq, delname, delnamelen))
1533 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1535 /* Check to see if we need to prime a stub zone. */
1536 if(prime_stub(qstate, iq, id, delname, iq->qchase.qclass)) {
1537 /* A priming sub request was made */
1541 /* most events just get forwarded to the next state. */
1542 return next_state(iq, INIT_REQUEST_3_STATE);
1546 * Process the third part of the initial request handling. This state exists
1547 * as a separate state so that queries that generate stub priming events
1548 * will get the tail end of the init process but not repeat the stub priming
1551 * @param qstate: query state.
1552 * @param iq: iterator query state.
1553 * @param id: module id.
1554 * @return true, advancing the event to the QUERYTARGETS_STATE.
1557 processInitRequest3(struct module_qstate* qstate, struct iter_qstate* iq,
1560 log_query_info(VERB_QUERY, "resolving (init part 3): ",
1562 /* if the cache reply dp equals a validation anchor or msg has DS,
1563 * then DNSSEC RRSIGs are expected in the reply */
1564 iq->dnssec_expected = iter_indicates_dnssec(qstate->env, iq->dp,
1565 iq->deleg_msg, iq->qchase.qclass);
1567 /* If the RD flag wasn't set, then we just finish with the
1568 * cached referral as the response. */
1569 if(!(qstate->query_flags & BIT_RD) && iq->deleg_msg) {
1570 iq->response = iq->deleg_msg;
1571 if(verbosity >= VERB_ALGO && iq->response)
1572 log_dns_msg("no RD requested, using delegation msg",
1573 &iq->response->qinfo, iq->response->rep);
1574 if(qstate->reply_origin)
1575 sock_list_insert(&qstate->reply_origin, NULL, 0, qstate->region);
1576 return final_state(iq);
1578 /* After this point, unset the RD flag -- this query is going to
1579 * be sent to an auth. server. */
1580 iq->chase_flags &= ~BIT_RD;
1582 /* if dnssec expected, fetch key for the trust-anchor or cached-DS */
1583 if(iq->dnssec_expected && qstate->env->cfg->prefetch_key &&
1584 !(qstate->query_flags&BIT_CD)) {
1585 generate_dnskey_prefetch(qstate, iq, id);
1586 fptr_ok(fptr_whitelist_modenv_detach_subs(
1587 qstate->env->detach_subs));
1588 (*qstate->env->detach_subs)(qstate);
1591 /* Jump to the next state. */
1592 return next_state(iq, QUERYTARGETS_STATE);
1596 * Given a basic query, generate a parent-side "target" query.
1597 * These are subordinate queries for missing delegation point target addresses,
1598 * for which only the parent of the delegation provides correct IP addresses.
1600 * @param qstate: query state.
1601 * @param iq: iterator query state.
1602 * @param id: module id.
1603 * @param name: target qname.
1604 * @param namelen: target qname length.
1605 * @param qtype: target qtype (either A or AAAA).
1606 * @param qclass: target qclass.
1607 * @return true on success, false on failure.
1610 generate_parentside_target_query(struct module_qstate* qstate,
1611 struct iter_qstate* iq, int id, uint8_t* name, size_t namelen,
1612 uint16_t qtype, uint16_t qclass)
1614 struct module_qstate* subq;
1615 if(!generate_sub_request(name, namelen, qtype, qclass, qstate,
1616 id, iq, INIT_REQUEST_STATE, FINISHED_STATE, &subq, 0))
1619 struct iter_qstate* subiq =
1620 (struct iter_qstate*)subq->minfo[id];
1621 /* blacklist the cache - we want to fetch parent stuff */
1622 sock_list_insert(&subq->blacklist, NULL, 0, subq->region);
1623 subiq->query_for_pside_glue = 1;
1624 if(dname_subdomain_c(name, iq->dp->name)) {
1625 subiq->dp = delegpt_copy(iq->dp, subq->region);
1626 subiq->dnssec_expected = iter_indicates_dnssec(
1627 qstate->env, subiq->dp, NULL,
1628 subq->qinfo.qclass);
1629 subiq->refetch_glue = 1;
1631 subiq->dp = dns_cache_find_delegation(qstate->env,
1632 name, namelen, qtype, qclass, subq->region,
1634 *qstate->env->now+subq->prefetch_leeway);
1635 /* if no dp, then it's from root, refetch unneeded */
1637 subiq->dnssec_expected = iter_indicates_dnssec(
1638 qstate->env, subiq->dp, NULL,
1639 subq->qinfo.qclass);
1640 subiq->refetch_glue = 1;
1644 log_nametypeclass(VERB_QUERY, "new pside target", name, qtype, qclass);
1649 * Given a basic query, generate a "target" query. These are subordinate
1650 * queries for missing delegation point target addresses.
1652 * @param qstate: query state.
1653 * @param iq: iterator query state.
1654 * @param id: module id.
1655 * @param name: target qname.
1656 * @param namelen: target qname length.
1657 * @param qtype: target qtype (either A or AAAA).
1658 * @param qclass: target qclass.
1659 * @return true on success, false on failure.
1662 generate_target_query(struct module_qstate* qstate, struct iter_qstate* iq,
1663 int id, uint8_t* name, size_t namelen, uint16_t qtype, uint16_t qclass)
1665 struct module_qstate* subq;
1666 if(!generate_sub_request(name, namelen, qtype, qclass, qstate,
1667 id, iq, INIT_REQUEST_STATE, FINISHED_STATE, &subq, 0))
1669 log_nametypeclass(VERB_QUERY, "new target", name, qtype, qclass);
1674 * Given an event at a certain state, generate zero or more target queries
1675 * for it's current delegation point.
1677 * @param qstate: query state.
1678 * @param iq: iterator query state.
1679 * @param ie: iterator shared global environment.
1680 * @param id: module id.
1681 * @param maxtargets: The maximum number of targets to query for.
1682 * if it is negative, there is no maximum number of targets.
1683 * @param num: returns the number of queries generated and processed,
1684 * which may be zero if there were no missing targets.
1685 * @return false on error.
1688 query_for_targets(struct module_qstate* qstate, struct iter_qstate* iq,
1689 struct iter_env* ie, int id, int maxtargets, int* num)
1691 int query_count = 0;
1692 struct delegpt_ns* ns;
1696 if(iq->depth == ie->max_dependency_depth)
1698 if(iq->depth > 0 && iq->target_count &&
1699 iq->target_count[1] > MAX_TARGET_COUNT) {
1700 char s[LDNS_MAX_DOMAINLEN+1];
1701 dname_str(qstate->qinfo.qname, s);
1702 verbose(VERB_QUERY, "request %s has exceeded the maximum "
1703 "number of glue fetches %d", s, iq->target_count[1]);
1707 iter_mark_cycle_targets(qstate, iq->dp);
1708 missing = (int)delegpt_count_missing_targets(iq->dp);
1709 log_assert(maxtargets != 0); /* that would not be useful */
1711 /* Generate target requests. Basically, any missing targets
1712 * are queried for here, regardless if it is necessary to do
1713 * so to continue processing. */
1714 if(maxtargets < 0 || maxtargets > missing)
1716 else toget = maxtargets;
1721 /* select 'toget' items from the total of 'missing' items */
1722 log_assert(toget <= missing);
1724 /* loop over missing targets */
1725 for(ns = iq->dp->nslist; ns; ns = ns->next) {
1729 /* randomly select this item with probability toget/missing */
1730 if(!iter_ns_probability(qstate->env->rnd, toget, missing)) {
1731 /* do not select this one, next; select toget number
1732 * of items from a list one less in size */
1737 if(ie->supports_ipv6 && !ns->got6) {
1738 /* Send the AAAA request. */
1739 if(!generate_target_query(qstate, iq, id,
1740 ns->name, ns->namelen,
1741 LDNS_RR_TYPE_AAAA, iq->qchase.qclass)) {
1744 qstate->ext_state[id] = module_wait_subquery;
1749 /* Send the A request. */
1750 if(ie->supports_ipv4 && !ns->got4) {
1751 if(!generate_target_query(qstate, iq, id,
1752 ns->name, ns->namelen,
1753 LDNS_RR_TYPE_A, iq->qchase.qclass)) {
1756 qstate->ext_state[id] = module_wait_subquery;
1762 /* mark this target as in progress. */
1771 qstate->ext_state[id] = module_wait_subquery;
1777 * Called by processQueryTargets when it would like extra targets to query
1778 * but it seems to be out of options. At last resort some less appealing
1779 * options are explored. If there are no more options, the result is SERVFAIL
1781 * @param qstate: query state.
1782 * @param iq: iterator query state.
1783 * @param ie: iterator shared global environment.
1784 * @param id: module id.
1785 * @return true if the event requires more request processing immediately,
1789 processLastResort(struct module_qstate* qstate, struct iter_qstate* iq,
1790 struct iter_env* ie, int id)
1792 struct delegpt_ns* ns;
1793 int query_count = 0;
1794 verbose(VERB_ALGO, "No more query targets, attempting last resort");
1797 if(!can_have_last_resort(qstate->env, iq->dp->name, iq->dp->namelen,
1798 iq->qchase.qclass, NULL)) {
1799 /* fail -- no more targets, no more hope of targets, no hope
1801 verbose(VERB_QUERY, "configured stub or forward servers failed -- returning SERVFAIL");
1802 return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
1804 if(!iq->dp->has_parent_side_NS && dname_is_root(iq->dp->name)) {
1805 struct delegpt* p = hints_lookup_root(qstate->env->hints,
1808 struct delegpt_ns* ns;
1809 struct delegpt_addr* a;
1810 iq->chase_flags &= ~BIT_RD; /* go to authorities */
1811 for(ns = p->nslist; ns; ns=ns->next) {
1812 (void)delegpt_add_ns(iq->dp, qstate->region,
1813 ns->name, ns->lame);
1815 for(a = p->target_list; a; a=a->next_target) {
1816 (void)delegpt_add_addr(iq->dp, qstate->region,
1817 &a->addr, a->addrlen, a->bogus,
1818 a->lame, a->tls_auth_name);
1821 iq->dp->has_parent_side_NS = 1;
1822 } else if(!iq->dp->has_parent_side_NS) {
1823 if(!iter_lookup_parent_NS_from_cache(qstate->env, iq->dp,
1824 qstate->region, &qstate->qinfo)
1825 || !iq->dp->has_parent_side_NS) {
1826 /* if: malloc failure in lookup go up to try */
1827 /* if: no parent NS in cache - go up one level */
1828 verbose(VERB_ALGO, "try to grab parent NS");
1829 iq->store_parent_NS = iq->dp;
1830 iq->chase_flags &= ~BIT_RD; /* go to authorities */
1831 iq->deleg_msg = NULL;
1832 iq->refetch_glue = 1;
1833 iq->query_restart_count++;
1835 if(qstate->env->cfg->qname_minimisation)
1836 iq->minimisation_state = INIT_MINIMISE_STATE;
1837 return next_state(iq, INIT_REQUEST_STATE);
1840 /* see if that makes new names available */
1841 if(!cache_fill_missing(qstate->env, iq->qchase.qclass,
1842 qstate->region, iq->dp))
1843 log_err("out of memory in cache_fill_missing");
1844 if(iq->dp->usable_list) {
1845 verbose(VERB_ALGO, "try parent-side-name, w. glue from cache");
1846 return next_state(iq, QUERYTARGETS_STATE);
1848 /* try to fill out parent glue from cache */
1849 if(iter_lookup_parent_glue_from_cache(qstate->env, iq->dp,
1850 qstate->region, &qstate->qinfo)) {
1851 /* got parent stuff from cache, see if we can continue */
1852 verbose(VERB_ALGO, "try parent-side glue from cache");
1853 return next_state(iq, QUERYTARGETS_STATE);
1855 /* query for an extra name added by the parent-NS record */
1856 if(delegpt_count_missing_targets(iq->dp) > 0) {
1858 verbose(VERB_ALGO, "try parent-side target name");
1859 if(!query_for_targets(qstate, iq, ie, id, 1, &qs)) {
1860 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1862 iq->num_target_queries += qs;
1863 target_count_increase(iq, qs);
1865 qstate->ext_state[id] = module_wait_subquery;
1866 return 0; /* and wait for them */
1869 if(iq->depth == ie->max_dependency_depth) {
1870 verbose(VERB_QUERY, "maxdepth and need more nameservers, fail");
1871 return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
1873 if(iq->depth > 0 && iq->target_count &&
1874 iq->target_count[1] > MAX_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", s, iq->target_count[1]);
1879 return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
1881 /* mark cycle targets for parent-side lookups */
1882 iter_mark_pside_cycle_targets(qstate, iq->dp);
1883 /* see if we can issue queries to get nameserver addresses */
1884 /* this lookup is not randomized, but sequential. */
1885 for(ns = iq->dp->nslist; ns; ns = ns->next) {
1886 /* if this nameserver is at a delegation point, but that
1887 * delegation point is a stub and we cannot go higher, skip*/
1888 if( ((ie->supports_ipv6 && !ns->done_pside6) ||
1889 (ie->supports_ipv4 && !ns->done_pside4)) &&
1890 !can_have_last_resort(qstate->env, ns->name, ns->namelen,
1891 iq->qchase.qclass, NULL)) {
1892 log_nametypeclass(VERB_ALGO, "cannot pside lookup ns "
1893 "because it is also a stub/forward,",
1894 ns->name, LDNS_RR_TYPE_NS, iq->qchase.qclass);
1895 if(ie->supports_ipv6) ns->done_pside6 = 1;
1896 if(ie->supports_ipv4) ns->done_pside4 = 1;
1899 /* query for parent-side A and AAAA for nameservers */
1900 if(ie->supports_ipv6 && !ns->done_pside6) {
1901 /* Send the AAAA request. */
1902 if(!generate_parentside_target_query(qstate, iq, id,
1903 ns->name, ns->namelen,
1904 LDNS_RR_TYPE_AAAA, iq->qchase.qclass))
1905 return error_response(qstate, id,
1906 LDNS_RCODE_SERVFAIL);
1907 ns->done_pside6 = 1;
1910 if(ie->supports_ipv4 && !ns->done_pside4) {
1911 /* Send the A request. */
1912 if(!generate_parentside_target_query(qstate, iq, id,
1913 ns->name, ns->namelen,
1914 LDNS_RR_TYPE_A, iq->qchase.qclass))
1915 return error_response(qstate, id,
1916 LDNS_RCODE_SERVFAIL);
1917 ns->done_pside4 = 1;
1920 if(query_count != 0) { /* suspend to await results */
1921 verbose(VERB_ALGO, "try parent-side glue lookup");
1922 iq->num_target_queries += query_count;
1923 target_count_increase(iq, query_count);
1924 qstate->ext_state[id] = module_wait_subquery;
1929 /* if this was a parent-side glue query itself, then store that
1930 * failure in cache. */
1931 if(!qstate->no_cache_store && iq->query_for_pside_glue
1933 iter_store_parentside_neg(qstate->env, &qstate->qinfo,
1934 iq->deleg_msg?iq->deleg_msg->rep:
1935 (iq->response?iq->response->rep:NULL));
1937 verbose(VERB_QUERY, "out of query targets -- returning SERVFAIL");
1938 /* fail -- no more targets, no more hope of targets, no hope
1940 return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
1944 * Try to find the NS record set that will resolve a qtype DS query. Due
1945 * to grandparent/grandchild reasons we did not get a proper lookup right
1946 * away. We need to create type NS queries until we get the right parent
1947 * for this lookup. We remove labels from the query to find the right point.
1948 * If we end up at the old dp name, then there is no solution.
1950 * @param qstate: query state.
1951 * @param iq: iterator query state.
1952 * @param id: module id.
1953 * @return true if the event requires more immediate processing, false if
1954 * not. This is generally only true when forwarding the request to
1955 * the final state (i.e., on answer).
1958 processDSNSFind(struct module_qstate* qstate, struct iter_qstate* iq, int id)
1960 struct module_qstate* subq = NULL;
1961 verbose(VERB_ALGO, "processDSNSFind");
1963 if(!iq->dsns_point) {
1965 iq->dsns_point = iq->qchase.qname;
1966 iq->dsns_point_len = iq->qchase.qname_len;
1968 /* robustcheck for internal error: we are not underneath the dp */
1969 if(!dname_subdomain_c(iq->dsns_point, iq->dp->name)) {
1970 return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
1973 /* go up one (more) step, until we hit the dp, if so, end */
1974 dname_remove_label(&iq->dsns_point, &iq->dsns_point_len);
1975 if(query_dname_compare(iq->dsns_point, iq->dp->name) == 0) {
1976 /* there was no inbetween nameserver, use the old delegation
1977 * point again. And this time, because dsns_point is nonNULL
1978 * we are going to accept the (bad) result */
1979 iq->state = QUERYTARGETS_STATE;
1982 iq->state = DSNS_FIND_STATE;
1984 /* spawn NS lookup (validation not needed, this is for DS lookup) */
1985 log_nametypeclass(VERB_ALGO, "fetch nameservers",
1986 iq->dsns_point, LDNS_RR_TYPE_NS, iq->qchase.qclass);
1987 if(!generate_sub_request(iq->dsns_point, iq->dsns_point_len,
1988 LDNS_RR_TYPE_NS, iq->qchase.qclass, qstate, id, iq,
1989 INIT_REQUEST_STATE, FINISHED_STATE, &subq, 0)) {
1990 return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
1997 * This is the request event state where the request will be sent to one of
1998 * its current query targets. This state also handles issuing target lookup
1999 * queries for missing target IP addresses. Queries typically iterate on
2000 * this state, both when they are just trying different targets for a given
2001 * delegation point, and when they change delegation points. This state
2002 * roughly corresponds to RFC 1034 algorithm steps 3 and 4.
2004 * @param qstate: query state.
2005 * @param iq: iterator query state.
2006 * @param ie: iterator shared global environment.
2007 * @param id: module id.
2008 * @return true if the event requires more request processing immediately,
2009 * false if not. This state only returns true when it is generating
2010 * a SERVFAIL response because the query has hit a dead end.
2013 processQueryTargets(struct module_qstate* qstate, struct iter_qstate* iq,
2014 struct iter_env* ie, int id)
2017 struct delegpt_addr* target;
2018 struct outbound_entry* outq;
2019 int auth_fallback = 0;
2021 /* NOTE: a request will encounter this state for each target it
2022 * needs to send a query to. That is, at least one per referral,
2023 * more if some targets timeout or return throwaway answers. */
2025 log_query_info(VERB_QUERY, "processQueryTargets:", &qstate->qinfo);
2026 verbose(VERB_ALGO, "processQueryTargets: targetqueries %d, "
2027 "currentqueries %d sentcount %d", iq->num_target_queries,
2028 iq->num_current_queries, iq->sent_count);
2030 /* Make sure that we haven't run away */
2031 /* FIXME: is this check even necessary? */
2032 if(iq->referral_count > MAX_REFERRAL_COUNT) {
2033 verbose(VERB_QUERY, "request has exceeded the maximum "
2034 "number of referrrals with %d", iq->referral_count);
2035 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2037 if(iq->sent_count > MAX_SENT_COUNT) {
2038 verbose(VERB_QUERY, "request has exceeded the maximum "
2039 "number of sends with %d", iq->sent_count);
2040 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2043 /* Make sure we have a delegation point, otherwise priming failed
2044 * or another failure occurred */
2046 verbose(VERB_QUERY, "Failed to get a delegation, giving up");
2047 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2049 if(!ie->supports_ipv6)
2050 delegpt_no_ipv6(iq->dp);
2051 if(!ie->supports_ipv4)
2052 delegpt_no_ipv4(iq->dp);
2053 delegpt_log(VERB_ALGO, iq->dp);
2055 if(iq->num_current_queries>0) {
2056 /* already busy answering a query, this restart is because
2057 * more delegpt addrs became available, wait for existing
2059 verbose(VERB_ALGO, "woke up, but wait for outstanding query");
2060 qstate->ext_state[id] = module_wait_reply;
2064 if(iq->minimisation_state == INIT_MINIMISE_STATE) {
2065 /* (Re)set qinfo_out to (new) delegation point, except when
2066 * qinfo_out is already a subdomain of dp. This happens when
2067 * increasing by more than one label at once (QNAMEs with more
2068 * than MAX_MINIMISE_COUNT labels). */
2069 if(!(iq->qinfo_out.qname_len
2070 && dname_subdomain_c(iq->qchase.qname,
2071 iq->qinfo_out.qname)
2072 && dname_subdomain_c(iq->qinfo_out.qname,
2074 iq->qinfo_out.qname = iq->dp->name;
2075 iq->qinfo_out.qname_len = iq->dp->namelen;
2076 iq->qinfo_out.qtype = LDNS_RR_TYPE_A;
2077 iq->qinfo_out.qclass = iq->qchase.qclass;
2078 iq->qinfo_out.local_alias = NULL;
2079 iq->minimise_count = 0;
2082 iq->minimisation_state = MINIMISE_STATE;
2084 if(iq->minimisation_state == MINIMISE_STATE) {
2085 int qchaselabs = dname_count_labels(iq->qchase.qname);
2086 int labdiff = qchaselabs -
2087 dname_count_labels(iq->qinfo_out.qname);
2089 iq->qinfo_out.qname = iq->qchase.qname;
2090 iq->qinfo_out.qname_len = iq->qchase.qname_len;
2091 iq->minimise_count++;
2092 iq->minimise_timeout_count = 0;
2094 iter_dec_attempts(iq->dp, 1);
2096 /* Limit number of iterations for QNAMEs with more
2097 * than MAX_MINIMISE_COUNT labels. Send first MINIMISE_ONE_LAB
2098 * labels of QNAME always individually.
2100 if(qchaselabs > MAX_MINIMISE_COUNT && labdiff > 1 &&
2101 iq->minimise_count > MINIMISE_ONE_LAB) {
2102 if(iq->minimise_count < MAX_MINIMISE_COUNT) {
2103 int multilabs = qchaselabs - 1 -
2105 int extralabs = multilabs /
2106 MINIMISE_MULTIPLE_LABS;
2108 if (MAX_MINIMISE_COUNT - iq->minimise_count >=
2109 multilabs % MINIMISE_MULTIPLE_LABS)
2110 /* Default behaviour is to add 1 label
2111 * every iteration. Therefore, decrement
2112 * the extralabs by 1 */
2114 if (extralabs < labdiff)
2115 labdiff -= extralabs;
2119 /* Last minimised iteration, send all labels with
2126 verbose(VERB_QUERY, "removing %d labels", labdiff-1);
2127 dname_remove_labels(&iq->qinfo_out.qname,
2128 &iq->qinfo_out.qname_len,
2131 if(labdiff < 1 || (labdiff < 2
2132 && (iq->qchase.qtype == LDNS_RR_TYPE_DS
2133 || iq->qchase.qtype == LDNS_RR_TYPE_A)))
2134 /* Stop minimising this query, resolve "as usual" */
2135 iq->minimisation_state = DONOT_MINIMISE_STATE;
2136 else if(!qstate->no_cache_lookup) {
2137 struct dns_msg* msg = dns_cache_lookup(qstate->env,
2138 iq->qinfo_out.qname, iq->qinfo_out.qname_len,
2139 iq->qinfo_out.qtype, iq->qinfo_out.qclass,
2140 qstate->query_flags, qstate->region,
2141 qstate->env->scratch, 0);
2142 if(msg && msg->rep->an_numrrsets == 0
2143 && FLAGS_GET_RCODE(msg->rep->flags) ==
2145 /* no need to send query if it is already
2146 * cached as NOERROR/NODATA */
2150 if(iq->minimisation_state == SKIP_MINIMISE_STATE) {
2151 if(iq->minimise_timeout_count < MAX_MINIMISE_TIMEOUT_COUNT)
2152 /* Do not increment qname, continue incrementing next
2154 iq->minimisation_state = MINIMISE_STATE;
2155 else if(!qstate->env->cfg->qname_minimisation_strict)
2156 /* Too many time-outs detected for this QNAME and QTYPE.
2157 * We give up, disable QNAME minimisation. */
2158 iq->minimisation_state = DONOT_MINIMISE_STATE;
2160 if(iq->minimisation_state == DONOT_MINIMISE_STATE)
2161 iq->qinfo_out = iq->qchase;
2163 /* now find an answer to this query */
2164 /* see if authority zones have an answer */
2165 /* now we know the dp, we can check the auth zone for locally hosted
2167 if(!iq->auth_zone_avoid && qstate->blacklist) {
2168 if(auth_zones_can_fallback(qstate->env->auth_zones,
2169 iq->dp->name, iq->dp->namelen, iq->qinfo_out.qclass)) {
2170 /* if cache is blacklisted and this zone allows us
2171 * to fallback to the internet, then do so, and
2172 * fetch results from the internet servers */
2173 iq->auth_zone_avoid = 1;
2176 if(iq->auth_zone_avoid) {
2177 iq->auth_zone_avoid = 0;
2179 } else if(auth_zones_lookup(qstate->env->auth_zones, &iq->qinfo_out,
2180 qstate->region, &iq->response, &auth_fallback, iq->dp->name,
2182 /* use this as a response to be processed by the iterator */
2183 if(verbosity >= VERB_ALGO) {
2184 log_dns_msg("msg from auth zone",
2185 &iq->response->qinfo, iq->response->rep);
2187 if((iq->chase_flags&BIT_RD) && !(iq->response->rep->flags&BIT_AA)) {
2188 verbose(VERB_ALGO, "forwarder, ignoring referral from auth zone");
2190 lock_rw_wrlock(&qstate->env->auth_zones->lock);
2191 qstate->env->auth_zones->num_query_up++;
2192 lock_rw_unlock(&qstate->env->auth_zones->lock);
2193 iq->num_current_queries++;
2194 iq->chase_to_rd = 0;
2195 iq->dnssec_lame_query = 0;
2196 iq->auth_zone_response = 1;
2197 return next_state(iq, QUERY_RESP_STATE);
2200 iq->auth_zone_response = 0;
2201 if(auth_fallback == 0) {
2202 /* like we got servfail from the auth zone lookup, and
2203 * no internet fallback */
2204 verbose(VERB_ALGO, "auth zone lookup failed, no fallback,"
2206 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2208 if(iq->dp && iq->dp->auth_dp) {
2209 /* we wanted to fallback, but had no delegpt, only the
2210 * auth zone generated delegpt, create an actual one */
2211 iq->auth_zone_avoid = 1;
2212 return next_state(iq, INIT_REQUEST_STATE);
2214 /* but mostly, fallback==1 (like, when no such auth zone exists)
2215 * and we continue with lookups */
2218 /* < not <=, because although the array is large enough for <=, the
2219 * generated query will immediately be discarded due to depth and
2220 * that servfail is cached, which is not good as opportunism goes. */
2221 if(iq->depth < ie->max_dependency_depth
2222 && iq->sent_count < TARGET_FETCH_STOP) {
2223 tf_policy = ie->target_fetch_policy[iq->depth];
2226 /* if in 0x20 fallback get as many targets as possible */
2227 if(iq->caps_fallback) {
2229 size_t naddr, nres, navail;
2230 if(!query_for_targets(qstate, iq, ie, id, -1, &extra)) {
2231 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2233 iq->num_target_queries += extra;
2234 target_count_increase(iq, extra);
2235 if(iq->num_target_queries > 0) {
2236 /* wait to get all targets, we want to try em */
2237 verbose(VERB_ALGO, "wait for all targets for fallback");
2238 qstate->ext_state[id] = module_wait_reply;
2241 /* did we do enough fallback queries already? */
2242 delegpt_count_addr(iq->dp, &naddr, &nres, &navail);
2243 /* the current caps_server is the number of fallbacks sent.
2244 * the original query is one that matched too, so we have
2245 * caps_server+1 number of matching queries now */
2246 if(iq->caps_server+1 >= naddr*3 ||
2247 iq->caps_server*2+2 >= MAX_SENT_COUNT) {
2248 /* *2 on sentcount check because ipv6 may fail */
2249 /* we're done, process the response */
2250 verbose(VERB_ALGO, "0x20 fallback had %d responses "
2251 "match for %d wanted, done.",
2252 (int)iq->caps_server+1, (int)naddr*3);
2253 iq->response = iq->caps_response;
2254 iq->caps_fallback = 0;
2255 iter_dec_attempts(iq->dp, 3); /* space for fallback */
2256 iq->num_current_queries++; /* RespState decrements it*/
2257 iq->referral_count++; /* make sure we don't loop */
2259 iq->state = QUERY_RESP_STATE;
2262 verbose(VERB_ALGO, "0x20 fallback number %d",
2263 (int)iq->caps_server);
2265 /* if there is a policy to fetch missing targets
2266 * opportunistically, do it. we rely on the fact that once a
2267 * query (or queries) for a missing name have been issued,
2268 * they will not show up again. */
2269 } else if(tf_policy != 0) {
2271 verbose(VERB_ALGO, "attempt to get extra %d targets",
2273 (void)query_for_targets(qstate, iq, ie, id, tf_policy, &extra);
2274 /* errors ignored, these targets are not strictly necessary for
2275 * this result, we do not have to reply with SERVFAIL */
2276 iq->num_target_queries += extra;
2277 target_count_increase(iq, extra);
2280 /* Add the current set of unused targets to our queue. */
2281 delegpt_add_unused_targets(iq->dp);
2283 /* Select the next usable target, filtering out unsuitable targets. */
2284 target = iter_server_selection(ie, qstate->env, iq->dp,
2285 iq->dp->name, iq->dp->namelen, iq->qchase.qtype,
2286 &iq->dnssec_lame_query, &iq->chase_to_rd,
2287 iq->num_target_queries, qstate->blacklist,
2288 qstate->prefetch_leeway);
2290 /* If no usable target was selected... */
2292 /* Here we distinguish between three states: generate a new
2293 * target query, just wait, or quit (with a SERVFAIL).
2294 * We have the following information: number of active
2295 * target queries, number of active current queries,
2296 * the presence of missing targets at this delegation
2297 * point, and the given query target policy. */
2299 /* Check for the wait condition. If this is true, then
2300 * an action must be taken. */
2301 if(iq->num_target_queries==0 && iq->num_current_queries==0) {
2302 /* If there is nothing to wait for, then we need
2303 * to distinguish between generating (a) new target
2304 * query, or failing. */
2305 if(delegpt_count_missing_targets(iq->dp) > 0) {
2307 verbose(VERB_ALGO, "querying for next "
2309 if(!query_for_targets(qstate, iq, ie, id,
2311 return error_response(qstate, id,
2312 LDNS_RCODE_SERVFAIL);
2315 delegpt_count_missing_targets(iq->dp) == 0){
2316 /* it looked like there were missing
2317 * targets, but they did not turn up.
2318 * Try the bad choices again (if any),
2319 * when we get back here missing==0,
2320 * so this is not a loop. */
2323 iq->num_target_queries += qs;
2324 target_count_increase(iq, qs);
2326 /* Since a target query might have been made, we
2327 * need to check again. */
2328 if(iq->num_target_queries == 0) {
2329 /* if in capsforid fallback, instead of last
2330 * resort, we agree with the current reply
2331 * we have (if any) (our count of addrs bad)*/
2332 if(iq->caps_fallback && iq->caps_reply) {
2333 /* we're done, process the response */
2334 verbose(VERB_ALGO, "0x20 fallback had %d responses, "
2335 "but no more servers except "
2336 "last resort, done.",
2337 (int)iq->caps_server+1);
2338 iq->response = iq->caps_response;
2339 iq->caps_fallback = 0;
2340 iter_dec_attempts(iq->dp, 3); /* space for fallback */
2341 iq->num_current_queries++; /* RespState decrements it*/
2342 iq->referral_count++; /* make sure we don't loop */
2344 iq->state = QUERY_RESP_STATE;
2347 return processLastResort(qstate, iq, ie, id);
2351 /* otherwise, we have no current targets, so submerge
2352 * until one of the target or direct queries return. */
2353 if(iq->num_target_queries>0 && iq->num_current_queries>0) {
2354 verbose(VERB_ALGO, "no current targets -- waiting "
2355 "for %d targets to resolve or %d outstanding"
2356 " queries to respond", iq->num_target_queries,
2357 iq->num_current_queries);
2358 qstate->ext_state[id] = module_wait_reply;
2359 } else if(iq->num_target_queries>0) {
2360 verbose(VERB_ALGO, "no current targets -- waiting "
2361 "for %d targets to resolve.",
2362 iq->num_target_queries);
2363 qstate->ext_state[id] = module_wait_subquery;
2365 verbose(VERB_ALGO, "no current targets -- waiting "
2366 "for %d outstanding queries to respond.",
2367 iq->num_current_queries);
2368 qstate->ext_state[id] = module_wait_reply;
2373 /* if not forwarding, check ratelimits per delegationpoint name */
2374 if(!(iq->chase_flags & BIT_RD) && !iq->ratelimit_ok) {
2375 if(!infra_ratelimit_inc(qstate->env->infra_cache, iq->dp->name,
2376 iq->dp->namelen, *qstate->env->now)) {
2377 lock_basic_lock(&ie->queries_ratelimit_lock);
2378 ie->num_queries_ratelimited++;
2379 lock_basic_unlock(&ie->queries_ratelimit_lock);
2380 verbose(VERB_ALGO, "query exceeded ratelimits");
2381 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2385 /* We have a valid target. */
2386 if(verbosity >= VERB_QUERY) {
2387 log_query_info(VERB_QUERY, "sending query:", &iq->qinfo_out);
2388 log_name_addr(VERB_QUERY, "sending to target:", iq->dp->name,
2389 &target->addr, target->addrlen);
2390 verbose(VERB_ALGO, "dnssec status: %s%s",
2391 iq->dnssec_expected?"expected": "not expected",
2392 iq->dnssec_lame_query?" but lame_query anyway": "");
2394 fptr_ok(fptr_whitelist_modenv_send_query(qstate->env->send_query));
2395 outq = (*qstate->env->send_query)(&iq->qinfo_out,
2396 iq->chase_flags | (iq->chase_to_rd?BIT_RD:0),
2397 /* unset CD if to forwarder(RD set) and not dnssec retry
2398 * (blacklist nonempty) and no trust-anchors are configured
2399 * above the qname or on the first attempt when dnssec is on */
2400 EDNS_DO| ((iq->chase_to_rd||(iq->chase_flags&BIT_RD)!=0)&&
2401 !qstate->blacklist&&(!iter_qname_indicates_dnssec(qstate->env,
2402 &iq->qinfo_out)||target->attempts==1)?0:BIT_CD),
2403 iq->dnssec_expected, iq->caps_fallback || is_caps_whitelisted(
2404 ie, iq), &target->addr, target->addrlen,
2405 iq->dp->name, iq->dp->namelen,
2406 (iq->dp->ssl_upstream || qstate->env->cfg->ssl_upstream),
2407 target->tls_auth_name, qstate);
2409 log_addr(VERB_DETAIL, "error sending query to auth server",
2410 &target->addr, target->addrlen);
2411 if(!(iq->chase_flags & BIT_RD) && !iq->ratelimit_ok)
2412 infra_ratelimit_dec(qstate->env->infra_cache, iq->dp->name,
2413 iq->dp->namelen, *qstate->env->now);
2414 if(qstate->env->cfg->qname_minimisation)
2415 iq->minimisation_state = SKIP_MINIMISE_STATE;
2416 return next_state(iq, QUERYTARGETS_STATE);
2418 outbound_list_insert(&iq->outlist, outq);
2419 iq->num_current_queries++;
2421 qstate->ext_state[id] = module_wait_reply;
2426 /** find NS rrset in given list */
2427 static struct ub_packed_rrset_key*
2428 find_NS(struct reply_info* rep, size_t from, size_t to)
2431 for(i=from; i<to; i++) {
2432 if(ntohs(rep->rrsets[i]->rk.type) == LDNS_RR_TYPE_NS)
2433 return rep->rrsets[i];
2440 * Process the query response. All queries end up at this state first. This
2441 * process generally consists of analyzing the response and routing the
2442 * event to the next state (either bouncing it back to a request state, or
2443 * terminating the processing for this event).
2445 * @param qstate: query state.
2446 * @param iq: iterator query state.
2447 * @param id: module id.
2448 * @return true if the event requires more immediate processing, false if
2449 * not. This is generally only true when forwarding the request to
2450 * the final state (i.e., on answer).
2453 processQueryResponse(struct module_qstate* qstate, struct iter_qstate* iq,
2457 enum response_type type;
2458 iq->num_current_queries--;
2460 if(!inplace_cb_query_response_call(qstate->env, qstate, iq->response))
2461 log_err("unable to call query_response callback");
2463 if(iq->response == NULL) {
2464 /* Don't increment qname when QNAME minimisation is enabled */
2465 if(qstate->env->cfg->qname_minimisation) {
2466 iq->minimise_timeout_count++;
2467 iq->minimisation_state = SKIP_MINIMISE_STATE;
2469 iq->chase_to_rd = 0;
2470 iq->dnssec_lame_query = 0;
2471 verbose(VERB_ALGO, "query response was timeout");
2472 return next_state(iq, QUERYTARGETS_STATE);
2474 type = response_type_from_server(
2475 (int)((iq->chase_flags&BIT_RD) || iq->chase_to_rd),
2476 iq->response, &iq->qinfo_out, iq->dp);
2477 iq->chase_to_rd = 0;
2478 if(type == RESPONSE_TYPE_REFERRAL && (iq->chase_flags&BIT_RD) &&
2479 !iq->auth_zone_response) {
2480 /* When forwarding (RD bit is set), we handle referrals
2481 * differently. No queries should be sent elsewhere */
2482 type = RESPONSE_TYPE_ANSWER;
2484 if(!qstate->env->cfg->disable_dnssec_lame_check && iq->dnssec_expected
2485 && !iq->dnssec_lame_query &&
2486 !(iq->chase_flags&BIT_RD)
2487 && iq->sent_count < DNSSEC_LAME_DETECT_COUNT
2488 && type != RESPONSE_TYPE_LAME
2489 && type != RESPONSE_TYPE_REC_LAME
2490 && type != RESPONSE_TYPE_THROWAWAY
2491 && type != RESPONSE_TYPE_UNTYPED) {
2492 /* a possible answer, see if it is missing DNSSEC */
2493 /* but not when forwarding, so we dont mark fwder lame */
2494 if(!iter_msg_has_dnssec(iq->response)) {
2495 /* Mark this address as dnsseclame in this dp,
2496 * because that will make serverselection disprefer
2497 * it, but also, once it is the only final option,
2498 * use dnssec-lame-bypass if it needs to query there.*/
2500 struct delegpt_addr* a = delegpt_find_addr(
2501 iq->dp, &qstate->reply->addr,
2502 qstate->reply->addrlen);
2503 if(a) a->dnsseclame = 1;
2505 /* test the answer is from the zone we expected,
2506 * otherwise, (due to parent,child on same server), we
2507 * might mark the server,zone lame inappropriately */
2508 if(!iter_msg_from_zone(iq->response, iq->dp, type,
2510 qstate->reply = NULL;
2511 type = RESPONSE_TYPE_LAME;
2514 } else iq->dnssec_lame_query = 0;
2515 /* see if referral brings us close to the target */
2516 if(type == RESPONSE_TYPE_REFERRAL) {
2517 struct ub_packed_rrset_key* ns = find_NS(
2518 iq->response->rep, iq->response->rep->an_numrrsets,
2519 iq->response->rep->an_numrrsets
2520 + iq->response->rep->ns_numrrsets);
2521 if(!ns) ns = find_NS(iq->response->rep, 0,
2522 iq->response->rep->an_numrrsets);
2523 if(!ns || !dname_strict_subdomain_c(ns->rk.dname, iq->dp->name)
2524 || !dname_subdomain_c(iq->qchase.qname, ns->rk.dname)){
2525 verbose(VERB_ALGO, "bad referral, throwaway");
2526 type = RESPONSE_TYPE_THROWAWAY;
2528 iter_scrub_ds(iq->response, ns, iq->dp->name);
2529 } else iter_scrub_ds(iq->response, NULL, NULL);
2530 if(type == RESPONSE_TYPE_THROWAWAY &&
2531 FLAGS_GET_RCODE(iq->response->rep->flags) == LDNS_RCODE_YXDOMAIN) {
2532 /* YXDOMAIN is a permanent error, no need to retry */
2533 type = RESPONSE_TYPE_ANSWER;
2535 if(type == RESPONSE_TYPE_CNAME && iq->response->rep->an_numrrsets >= 1
2536 && ntohs(iq->response->rep->rrsets[0]->rk.type) == LDNS_RR_TYPE_DNAME) {
2537 uint8_t* sname = NULL;
2538 size_t snamelen = 0;
2539 get_cname_target(iq->response->rep->rrsets[0], &sname,
2541 if(snamelen && dname_subdomain_c(sname, iq->response->rep->rrsets[0]->rk.dname)) {
2542 /* DNAME to a subdomain loop; do not recurse */
2543 type = RESPONSE_TYPE_ANSWER;
2547 /* handle each of the type cases */
2548 if(type == RESPONSE_TYPE_ANSWER) {
2549 /* ANSWER type responses terminate the query algorithm,
2550 * so they sent on their */
2551 if(verbosity >= VERB_DETAIL) {
2552 verbose(VERB_DETAIL, "query response was %s",
2553 FLAGS_GET_RCODE(iq->response->rep->flags)
2554 ==LDNS_RCODE_NXDOMAIN?"NXDOMAIN ANSWER":
2555 (iq->response->rep->an_numrrsets?"ANSWER":
2558 /* if qtype is DS, check we have the right level of answer,
2559 * like grandchild answer but we need the middle, reject it */
2560 if(iq->qchase.qtype == LDNS_RR_TYPE_DS && !iq->dsns_point
2561 && !(iq->chase_flags&BIT_RD)
2562 && iter_ds_toolow(iq->response, iq->dp)
2563 && iter_dp_cangodown(&iq->qchase, iq->dp)) {
2564 /* close down outstanding requests to be discarded */
2565 outbound_list_clear(&iq->outlist);
2566 iq->num_current_queries = 0;
2567 fptr_ok(fptr_whitelist_modenv_detach_subs(
2568 qstate->env->detach_subs));
2569 (*qstate->env->detach_subs)(qstate);
2570 iq->num_target_queries = 0;
2571 return processDSNSFind(qstate, iq, id);
2573 if(!qstate->no_cache_store)
2574 iter_dns_store(qstate->env, &iq->response->qinfo,
2575 iq->response->rep, 0, qstate->prefetch_leeway,
2576 iq->dp&&iq->dp->has_parent_side_NS,
2577 qstate->region, qstate->query_flags);
2578 /* close down outstanding requests to be discarded */
2579 outbound_list_clear(&iq->outlist);
2580 iq->num_current_queries = 0;
2581 fptr_ok(fptr_whitelist_modenv_detach_subs(
2582 qstate->env->detach_subs));
2583 (*qstate->env->detach_subs)(qstate);
2584 iq->num_target_queries = 0;
2586 sock_list_insert(&qstate->reply_origin,
2587 &qstate->reply->addr, qstate->reply->addrlen,
2589 if(iq->minimisation_state != DONOT_MINIMISE_STATE) {
2590 if(FLAGS_GET_RCODE(iq->response->rep->flags) !=
2591 LDNS_RCODE_NOERROR) {
2592 if(qstate->env->cfg->qname_minimisation_strict)
2593 return final_state(iq);
2594 /* Best effort qname-minimisation.
2595 * Stop minimising and send full query when
2596 * RCODE is not NOERROR. */
2597 iq->minimisation_state = DONOT_MINIMISE_STATE;
2599 if(FLAGS_GET_RCODE(iq->response->rep->flags) ==
2600 LDNS_RCODE_NXDOMAIN) {
2601 /* Stop resolving when NXDOMAIN is DNSSEC
2602 * signed. Based on assumption that nameservers
2603 * serving signed zones do not return NXDOMAIN
2604 * for empty-non-terminals. */
2605 if(iq->dnssec_expected)
2606 return final_state(iq);
2607 /* Make subrequest to validate intermediate
2608 * NXDOMAIN if harden-below-nxdomain is
2610 if(qstate->env->cfg->harden_below_nxdomain) {
2611 struct module_qstate* subq = NULL;
2612 log_query_info(VERB_QUERY,
2613 "schedule NXDOMAIN validation:",
2614 &iq->response->qinfo);
2615 if(!generate_sub_request(
2616 iq->response->qinfo.qname,
2617 iq->response->qinfo.qname_len,
2618 iq->response->qinfo.qtype,
2619 iq->response->qinfo.qclass,
2622 FINISHED_STATE, &subq, 1))
2624 "could not validate NXDOMAIN "
2628 return next_state(iq, QUERYTARGETS_STATE);
2630 return final_state(iq);
2631 } else if(type == RESPONSE_TYPE_REFERRAL) {
2632 /* REFERRAL type responses get a reset of the
2633 * delegation point, and back to the QUERYTARGETS_STATE. */
2634 verbose(VERB_DETAIL, "query response was REFERRAL");
2636 if(!(iq->chase_flags & BIT_RD) && !iq->ratelimit_ok) {
2637 /* we have a referral, no ratelimit, we can send
2638 * our queries to the given name */
2639 infra_ratelimit_dec(qstate->env->infra_cache,
2640 iq->dp->name, iq->dp->namelen,
2644 /* if hardened, only store referral if we asked for it */
2645 if(!qstate->no_cache_store &&
2646 (!qstate->env->cfg->harden_referral_path ||
2647 ( qstate->qinfo.qtype == LDNS_RR_TYPE_NS
2648 && (qstate->query_flags&BIT_RD)
2649 && !(qstate->query_flags&BIT_CD)
2650 /* we know that all other NS rrsets are scrubbed
2651 * away, thus on referral only one is left.
2652 * see if that equals the query name... */
2653 && ( /* auth section, but sometimes in answer section*/
2654 reply_find_rrset_section_ns(iq->response->rep,
2655 iq->qchase.qname, iq->qchase.qname_len,
2656 LDNS_RR_TYPE_NS, iq->qchase.qclass)
2657 || reply_find_rrset_section_an(iq->response->rep,
2658 iq->qchase.qname, iq->qchase.qname_len,
2659 LDNS_RR_TYPE_NS, iq->qchase.qclass)
2662 /* Store the referral under the current query */
2663 /* no prefetch-leeway, since its not the answer */
2664 iter_dns_store(qstate->env, &iq->response->qinfo,
2665 iq->response->rep, 1, 0, 0, NULL, 0);
2666 if(iq->store_parent_NS)
2667 iter_store_parentside_NS(qstate->env,
2669 if(qstate->env->neg_cache)
2670 val_neg_addreferral(qstate->env->neg_cache,
2671 iq->response->rep, iq->dp->name);
2673 /* store parent-side-in-zone-glue, if directly queried for */
2674 if(!qstate->no_cache_store && iq->query_for_pside_glue
2675 && !iq->pside_glue) {
2676 iq->pside_glue = reply_find_rrset(iq->response->rep,
2677 iq->qchase.qname, iq->qchase.qname_len,
2678 iq->qchase.qtype, iq->qchase.qclass);
2679 if(iq->pside_glue) {
2680 log_rrset_key(VERB_ALGO, "found parent-side "
2681 "glue", iq->pside_glue);
2682 iter_store_parentside_rrset(qstate->env,
2687 /* Reset the event state, setting the current delegation
2688 * point to the referral. */
2689 iq->deleg_msg = iq->response;
2690 iq->dp = delegpt_from_message(iq->response, qstate->region);
2691 if (qstate->env->cfg->qname_minimisation)
2692 iq->minimisation_state = INIT_MINIMISE_STATE;
2694 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2695 if(!cache_fill_missing(qstate->env, iq->qchase.qclass,
2696 qstate->region, iq->dp))
2697 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2698 if(iq->store_parent_NS && query_dname_compare(iq->dp->name,
2699 iq->store_parent_NS->name) == 0)
2700 iter_merge_retry_counts(iq->dp, iq->store_parent_NS);
2701 delegpt_log(VERB_ALGO, iq->dp);
2702 /* Count this as a referral. */
2703 iq->referral_count++;
2705 /* see if the next dp is a trust anchor, or a DS was sent
2706 * along, indicating dnssec is expected for next zone */
2707 iq->dnssec_expected = iter_indicates_dnssec(qstate->env,
2708 iq->dp, iq->response, iq->qchase.qclass);
2709 /* if dnssec, validating then also fetch the key for the DS */
2710 if(iq->dnssec_expected && qstate->env->cfg->prefetch_key &&
2711 !(qstate->query_flags&BIT_CD))
2712 generate_dnskey_prefetch(qstate, iq, id);
2714 /* spawn off NS and addr to auth servers for the NS we just
2715 * got in the referral. This gets authoritative answer
2716 * (answer section trust level) rrset.
2717 * right after, we detach the subs, answer goes to cache. */
2718 if(qstate->env->cfg->harden_referral_path)
2719 generate_ns_check(qstate, iq, id);
2721 /* stop current outstanding queries.
2722 * FIXME: should the outstanding queries be waited for and
2723 * handled? Say by a subquery that inherits the outbound_entry.
2725 outbound_list_clear(&iq->outlist);
2726 iq->num_current_queries = 0;
2727 fptr_ok(fptr_whitelist_modenv_detach_subs(
2728 qstate->env->detach_subs));
2729 (*qstate->env->detach_subs)(qstate);
2730 iq->num_target_queries = 0;
2731 verbose(VERB_ALGO, "cleared outbound list for next round");
2732 return next_state(iq, QUERYTARGETS_STATE);
2733 } else if(type == RESPONSE_TYPE_CNAME) {
2734 uint8_t* sname = NULL;
2735 size_t snamelen = 0;
2736 /* CNAME type responses get a query restart (i.e., get a
2737 * reset of the query state and go back to INIT_REQUEST_STATE).
2739 verbose(VERB_DETAIL, "query response was CNAME");
2740 if(verbosity >= VERB_ALGO)
2741 log_dns_msg("cname msg", &iq->response->qinfo,
2743 /* if qtype is DS, check we have the right level of answer,
2744 * like grandchild answer but we need the middle, reject it */
2745 if(iq->qchase.qtype == LDNS_RR_TYPE_DS && !iq->dsns_point
2746 && !(iq->chase_flags&BIT_RD)
2747 && iter_ds_toolow(iq->response, iq->dp)
2748 && iter_dp_cangodown(&iq->qchase, iq->dp)) {
2749 outbound_list_clear(&iq->outlist);
2750 iq->num_current_queries = 0;
2751 fptr_ok(fptr_whitelist_modenv_detach_subs(
2752 qstate->env->detach_subs));
2753 (*qstate->env->detach_subs)(qstate);
2754 iq->num_target_queries = 0;
2755 return processDSNSFind(qstate, iq, id);
2757 /* Process the CNAME response. */
2758 if(!handle_cname_response(qstate, iq, iq->response,
2760 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2761 /* cache the CNAME response under the current query */
2762 /* NOTE : set referral=1, so that rrsets get stored but not
2763 * the partial query answer (CNAME only). */
2764 /* prefetchleeway applied because this updates answer parts */
2765 if(!qstate->no_cache_store)
2766 iter_dns_store(qstate->env, &iq->response->qinfo,
2767 iq->response->rep, 1, qstate->prefetch_leeway,
2768 iq->dp&&iq->dp->has_parent_side_NS, NULL,
2769 qstate->query_flags);
2770 /* set the current request's qname to the new value. */
2771 iq->qchase.qname = sname;
2772 iq->qchase.qname_len = snamelen;
2773 /* Clear the query state, since this is a query restart. */
2774 iq->deleg_msg = NULL;
2776 iq->dsns_point = NULL;
2777 iq->auth_zone_response = 0;
2779 if(iq->minimisation_state != MINIMISE_STATE)
2780 /* Only count as query restart when it is not an extra
2781 * query as result of qname minimisation. */
2782 iq->query_restart_count++;
2783 if(qstate->env->cfg->qname_minimisation)
2784 iq->minimisation_state = INIT_MINIMISE_STATE;
2786 /* stop current outstanding queries.
2787 * FIXME: should the outstanding queries be waited for and
2788 * handled? Say by a subquery that inherits the outbound_entry.
2790 outbound_list_clear(&iq->outlist);
2791 iq->num_current_queries = 0;
2792 fptr_ok(fptr_whitelist_modenv_detach_subs(
2793 qstate->env->detach_subs));
2794 (*qstate->env->detach_subs)(qstate);
2795 iq->num_target_queries = 0;
2797 sock_list_insert(&qstate->reply_origin,
2798 &qstate->reply->addr, qstate->reply->addrlen,
2800 verbose(VERB_ALGO, "cleared outbound list for query restart");
2801 /* go to INIT_REQUEST_STATE for new qname. */
2802 return next_state(iq, INIT_REQUEST_STATE);
2803 } else if(type == RESPONSE_TYPE_LAME) {
2804 /* Cache the LAMEness. */
2805 verbose(VERB_DETAIL, "query response was %sLAME",
2806 dnsseclame?"DNSSEC ":"");
2807 if(!dname_subdomain_c(iq->qchase.qname, iq->dp->name)) {
2808 log_err("mark lame: mismatch in qname and dpname");
2809 /* throwaway this reply below */
2810 } else if(qstate->reply) {
2811 /* need addr for lameness cache, but we may have
2812 * gotten this from cache, so test to be sure */
2813 if(!infra_set_lame(qstate->env->infra_cache,
2814 &qstate->reply->addr, qstate->reply->addrlen,
2815 iq->dp->name, iq->dp->namelen,
2816 *qstate->env->now, dnsseclame, 0,
2818 log_err("mark host lame: out of memory");
2820 } else if(type == RESPONSE_TYPE_REC_LAME) {
2821 /* Cache the LAMEness. */
2822 verbose(VERB_DETAIL, "query response REC_LAME: "
2823 "recursive but not authoritative server");
2824 if(!dname_subdomain_c(iq->qchase.qname, iq->dp->name)) {
2825 log_err("mark rec_lame: mismatch in qname and dpname");
2826 /* throwaway this reply below */
2827 } else if(qstate->reply) {
2828 /* need addr for lameness cache, but we may have
2829 * gotten this from cache, so test to be sure */
2830 verbose(VERB_DETAIL, "mark as REC_LAME");
2831 if(!infra_set_lame(qstate->env->infra_cache,
2832 &qstate->reply->addr, qstate->reply->addrlen,
2833 iq->dp->name, iq->dp->namelen,
2834 *qstate->env->now, 0, 1, iq->qchase.qtype))
2835 log_err("mark host lame: out of memory");
2837 } else if(type == RESPONSE_TYPE_THROWAWAY) {
2838 /* LAME and THROWAWAY responses are handled the same way.
2839 * In this case, the event is just sent directly back to
2840 * the QUERYTARGETS_STATE without resetting anything,
2841 * because, clearly, the next target must be tried. */
2842 verbose(VERB_DETAIL, "query response was THROWAWAY");
2844 log_warn("A query response came back with an unknown type: %d",
2848 /* LAME, THROWAWAY and "unknown" all end up here.
2849 * Recycle to the QUERYTARGETS state to hopefully try a
2850 * different target. */
2851 if (qstate->env->cfg->qname_minimisation &&
2852 !qstate->env->cfg->qname_minimisation_strict)
2853 iq->minimisation_state = DONOT_MINIMISE_STATE;
2854 if(iq->auth_zone_response) {
2855 /* can we fallback? */
2856 iq->auth_zone_response = 0;
2857 if(!auth_zones_can_fallback(qstate->env->auth_zones,
2858 iq->dp->name, iq->dp->namelen, qstate->qinfo.qclass)) {
2859 verbose(VERB_ALGO, "auth zone response bad, and no"
2860 " fallback possible, servfail");
2861 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2863 verbose(VERB_ALGO, "auth zone response was bad, "
2864 "fallback enabled");
2865 iq->auth_zone_avoid = 1;
2866 if(iq->dp->auth_dp) {
2867 /* we are using a dp for the auth zone, with no
2868 * nameservers, get one first */
2870 return next_state(iq, INIT_REQUEST_STATE);
2873 return next_state(iq, QUERYTARGETS_STATE);
2877 * Return priming query results to interested super querystates.
2879 * Sets the delegation point and delegation message (not nonRD queries).
2880 * This is a callback from walk_supers.
2882 * @param qstate: priming query state that finished.
2883 * @param id: module id.
2884 * @param forq: the qstate for which priming has been done.
2887 prime_supers(struct module_qstate* qstate, int id, struct module_qstate* forq)
2889 struct iter_qstate* foriq = (struct iter_qstate*)forq->minfo[id];
2890 struct delegpt* dp = NULL;
2892 log_assert(qstate->is_priming || foriq->wait_priming_stub);
2893 log_assert(qstate->return_rcode == LDNS_RCODE_NOERROR);
2894 /* Convert our response to a delegation point */
2895 dp = delegpt_from_message(qstate->return_msg, forq->region);
2897 /* if there is no convertable delegation point, then
2898 * the ANSWER type was (presumably) a negative answer. */
2899 verbose(VERB_ALGO, "prime response was not a positive "
2902 foriq->state = QUERYTARGETS_STATE;
2906 log_query_info(VERB_DETAIL, "priming successful for", &qstate->qinfo);
2907 delegpt_log(VERB_ALGO, dp);
2909 foriq->deleg_msg = dns_copy_msg(qstate->return_msg, forq->region);
2910 if(!foriq->deleg_msg) {
2911 log_err("copy prime response: out of memory");
2913 foriq->state = QUERYTARGETS_STATE;
2917 /* root priming responses go to init stage 2, priming stub
2918 * responses to to stage 3. */
2919 if(foriq->wait_priming_stub) {
2920 foriq->state = INIT_REQUEST_3_STATE;
2921 foriq->wait_priming_stub = 0;
2922 } else foriq->state = INIT_REQUEST_2_STATE;
2923 /* because we are finished, the parent will be reactivated */
2927 * This handles the response to a priming query. This is used to handle both
2928 * root and stub priming responses. This is basically the equivalent of the
2929 * QUERY_RESP_STATE, but will not handle CNAME responses and will treat
2930 * REFERRALs as ANSWERS. It will also update and reactivate the originating
2933 * @param qstate: query state.
2934 * @param id: module id.
2935 * @return true if the event needs more immediate processing, false if not.
2936 * This state always returns false.
2939 processPrimeResponse(struct module_qstate* qstate, int id)
2941 struct iter_qstate* iq = (struct iter_qstate*)qstate->minfo[id];
2942 enum response_type type;
2943 iq->response->rep->flags &= ~(BIT_RD|BIT_RA); /* ignore rec-lame */
2944 type = response_type_from_server(
2945 (int)((iq->chase_flags&BIT_RD) || iq->chase_to_rd),
2946 iq->response, &iq->qchase, iq->dp);
2947 if(type == RESPONSE_TYPE_ANSWER) {
2948 qstate->return_rcode = LDNS_RCODE_NOERROR;
2949 qstate->return_msg = iq->response;
2951 qstate->return_rcode = LDNS_RCODE_SERVFAIL;
2952 qstate->return_msg = NULL;
2955 /* validate the root or stub after priming (if enabled).
2956 * This is the same query as the prime query, but with validation.
2957 * Now that we are primed, the additional queries that validation
2958 * may need can be resolved, such as DLV. */
2959 if(qstate->env->cfg->harden_referral_path) {
2960 struct module_qstate* subq = NULL;
2961 log_nametypeclass(VERB_ALGO, "schedule prime validation",
2962 qstate->qinfo.qname, qstate->qinfo.qtype,
2963 qstate->qinfo.qclass);
2964 if(!generate_sub_request(qstate->qinfo.qname,
2965 qstate->qinfo.qname_len, qstate->qinfo.qtype,
2966 qstate->qinfo.qclass, qstate, id, iq,
2967 INIT_REQUEST_STATE, FINISHED_STATE, &subq, 1)) {
2968 verbose(VERB_ALGO, "could not generate prime check");
2970 generate_a_aaaa_check(qstate, iq, id);
2973 /* This event is finished. */
2974 qstate->ext_state[id] = module_finished;
2979 * Do final processing on responses to target queries. Events reach this
2980 * state after the iterative resolution algorithm terminates. This state is
2981 * responsible for reactivating the original event, and housekeeping related
2982 * to received target responses (caching, updating the current delegation
2984 * Callback from walk_supers for every super state that is interested in
2985 * the results from this query.
2987 * @param qstate: query state.
2988 * @param id: module id.
2989 * @param forq: super query state.
2992 processTargetResponse(struct module_qstate* qstate, int id,
2993 struct module_qstate* forq)
2995 struct iter_qstate* iq = (struct iter_qstate*)qstate->minfo[id];
2996 struct iter_qstate* foriq = (struct iter_qstate*)forq->minfo[id];
2997 struct ub_packed_rrset_key* rrset;
2998 struct delegpt_ns* dpns;
2999 log_assert(qstate->return_rcode == LDNS_RCODE_NOERROR);
3001 foriq->state = QUERYTARGETS_STATE;
3002 log_query_info(VERB_ALGO, "processTargetResponse", &qstate->qinfo);
3003 log_query_info(VERB_ALGO, "processTargetResponse super", &forq->qinfo);
3005 /* Tell the originating event that this target query has finished
3006 * (regardless if it succeeded or not). */
3007 foriq->num_target_queries--;
3009 /* check to see if parent event is still interested (in orig name). */
3011 verbose(VERB_ALGO, "subq: parent not interested, was reset");
3012 return; /* not interested anymore */
3014 dpns = delegpt_find_ns(foriq->dp, qstate->qinfo.qname,
3015 qstate->qinfo.qname_len);
3017 /* If not interested, just stop processing this event */
3018 verbose(VERB_ALGO, "subq: parent not interested anymore");
3019 /* could be because parent was jostled out of the cache,
3020 and a new identical query arrived, that does not want it*/
3024 /* if iq->query_for_pside_glue then add the pside_glue (marked lame) */
3025 if(iq->pside_glue) {
3026 /* if the pside_glue is NULL, then it could not be found,
3027 * the done_pside is already set when created and a cache
3028 * entry created in processFinished so nothing to do here */
3029 log_rrset_key(VERB_ALGO, "add parentside glue to dp",
3031 if(!delegpt_add_rrset(foriq->dp, forq->region,
3033 log_err("out of memory adding pside glue");
3036 /* This response is relevant to the current query, so we
3037 * add (attempt to add, anyway) this target(s) and reactivate
3038 * the original event.
3039 * NOTE: we could only look for the AnswerRRset if the
3040 * response type was ANSWER. */
3041 rrset = reply_find_answer_rrset(&iq->qchase, qstate->return_msg->rep);
3043 /* if CNAMEs have been followed - add new NS to delegpt. */
3044 /* BTW. RFC 1918 says NS should not have got CNAMEs. Robust. */
3045 if(!delegpt_find_ns(foriq->dp, rrset->rk.dname,
3046 rrset->rk.dname_len)) {
3047 /* if dpns->lame then set newcname ns lame too */
3048 if(!delegpt_add_ns(foriq->dp, forq->region,
3049 rrset->rk.dname, dpns->lame))
3050 log_err("out of memory adding cnamed-ns");
3052 /* if dpns->lame then set the address(es) lame too */
3053 if(!delegpt_add_rrset(foriq->dp, forq->region, rrset,
3055 log_err("out of memory adding targets");
3056 verbose(VERB_ALGO, "added target response");
3057 delegpt_log(VERB_ALGO, foriq->dp);
3059 verbose(VERB_ALGO, "iterator TargetResponse failed");
3060 dpns->resolved = 1; /* fail the target */
3065 * Process response for DS NS Find queries, that attempt to find the delegation
3066 * point where we ask the DS query from.
3068 * @param qstate: query state.
3069 * @param id: module id.
3070 * @param forq: super query state.
3073 processDSNSResponse(struct module_qstate* qstate, int id,
3074 struct module_qstate* forq)
3076 struct iter_qstate* foriq = (struct iter_qstate*)forq->minfo[id];
3078 /* if the finished (iq->response) query has no NS set: continue
3079 * up to look for the right dp; nothing to change, do DPNSstate */
3080 if(qstate->return_rcode != LDNS_RCODE_NOERROR)
3081 return; /* seek further */
3082 /* find the NS RRset (without allowing CNAMEs) */
3083 if(!reply_find_rrset(qstate->return_msg->rep, qstate->qinfo.qname,
3084 qstate->qinfo.qname_len, LDNS_RR_TYPE_NS,
3085 qstate->qinfo.qclass)){
3086 return; /* seek further */
3089 /* else, store as DP and continue at querytargets */
3090 foriq->state = QUERYTARGETS_STATE;
3091 foriq->dp = delegpt_from_message(qstate->return_msg, forq->region);
3093 log_err("out of memory in dsns dp alloc");
3094 return; /* dp==NULL in QUERYTARGETS makes SERVFAIL */
3096 /* success, go query the querytargets in the new dp (and go down) */
3100 * Process response for qclass=ANY queries for a particular class.
3101 * Append to result or error-exit.
3103 * @param qstate: query state.
3104 * @param id: module id.
3105 * @param forq: super query state.
3108 processClassResponse(struct module_qstate* qstate, int id,
3109 struct module_qstate* forq)
3111 struct iter_qstate* foriq = (struct iter_qstate*)forq->minfo[id];
3112 struct dns_msg* from = qstate->return_msg;
3113 log_query_info(VERB_ALGO, "processClassResponse", &qstate->qinfo);
3114 log_query_info(VERB_ALGO, "processClassResponse super", &forq->qinfo);
3115 if(qstate->return_rcode != LDNS_RCODE_NOERROR) {
3116 /* cause servfail for qclass ANY query */
3117 foriq->response = NULL;
3118 foriq->state = FINISHED_STATE;
3122 if(!foriq->response) {
3123 /* allocate the response: copy RCODE, sec_state */
3124 foriq->response = dns_copy_msg(from, forq->region);
3125 if(!foriq->response) {
3126 log_err("malloc failed for qclass ANY response");
3127 foriq->state = FINISHED_STATE;
3130 foriq->response->qinfo.qclass = forq->qinfo.qclass;
3131 /* qclass ANY does not receive the AA flag on replies */
3132 foriq->response->rep->authoritative = 0;
3134 struct dns_msg* to = foriq->response;
3135 /* add _from_ this response _to_ existing collection */
3136 /* if there are records, copy RCODE */
3137 /* lower sec_state if this message is lower */
3138 if(from->rep->rrset_count != 0) {
3139 size_t n = from->rep->rrset_count+to->rep->rrset_count;
3140 struct ub_packed_rrset_key** dest, **d;
3141 /* copy appropriate rcode */
3142 to->rep->flags = from->rep->flags;
3144 if(from->rep->rrset_count > RR_COUNT_MAX ||
3145 to->rep->rrset_count > RR_COUNT_MAX) {
3146 log_err("malloc failed (too many rrsets) in collect ANY");
3147 foriq->state = FINISHED_STATE;
3148 return; /* integer overflow protection */
3150 dest = regional_alloc(forq->region, sizeof(dest[0])*n);
3152 log_err("malloc failed in collect ANY");
3153 foriq->state = FINISHED_STATE;
3158 memcpy(dest, to->rep->rrsets, to->rep->an_numrrsets
3160 dest += to->rep->an_numrrsets;
3161 memcpy(dest, from->rep->rrsets, from->rep->an_numrrsets
3163 dest += from->rep->an_numrrsets;
3165 memcpy(dest, to->rep->rrsets+to->rep->an_numrrsets,
3166 to->rep->ns_numrrsets * sizeof(dest[0]));
3167 dest += to->rep->ns_numrrsets;
3168 memcpy(dest, from->rep->rrsets+from->rep->an_numrrsets,
3169 from->rep->ns_numrrsets * sizeof(dest[0]));
3170 dest += from->rep->ns_numrrsets;
3172 memcpy(dest, to->rep->rrsets+to->rep->an_numrrsets+
3173 to->rep->ns_numrrsets,
3174 to->rep->ar_numrrsets * sizeof(dest[0]));
3175 dest += to->rep->ar_numrrsets;
3176 memcpy(dest, from->rep->rrsets+from->rep->an_numrrsets+
3177 from->rep->ns_numrrsets,
3178 from->rep->ar_numrrsets * sizeof(dest[0]));
3180 to->rep->rrsets = d;
3181 to->rep->an_numrrsets += from->rep->an_numrrsets;
3182 to->rep->ns_numrrsets += from->rep->ns_numrrsets;
3183 to->rep->ar_numrrsets += from->rep->ar_numrrsets;
3184 to->rep->rrset_count = n;
3186 if(from->rep->security < to->rep->security) /* lowest sec */
3187 to->rep->security = from->rep->security;
3188 if(from->rep->qdcount != 0) /* insert qd if appropriate */
3189 to->rep->qdcount = from->rep->qdcount;
3190 if(from->rep->ttl < to->rep->ttl) /* use smallest TTL */
3191 to->rep->ttl = from->rep->ttl;
3192 if(from->rep->prefetch_ttl < to->rep->prefetch_ttl)
3193 to->rep->prefetch_ttl = from->rep->prefetch_ttl;
3196 foriq->num_current_queries --;
3197 if(foriq->num_current_queries == 0)
3198 foriq->state = FINISHED_STATE;
3202 * Collect class ANY responses and make them into one response. This
3203 * state is started and it creates queries for all classes (that have
3204 * root hints). The answers are then collected.
3206 * @param qstate: query state.
3207 * @param id: module id.
3208 * @return true if the event needs more immediate processing, false if not.
3211 processCollectClass(struct module_qstate* qstate, int id)
3213 struct iter_qstate* iq = (struct iter_qstate*)qstate->minfo[id];
3214 struct module_qstate* subq;
3215 /* If qchase.qclass == 0 then send out queries for all classes.
3216 * Otherwise, do nothing (wait for all answers to arrive and the
3217 * processClassResponse to put them together, and that moves us
3218 * towards the Finished state when done. */
3219 if(iq->qchase.qclass == 0) {
3221 iq->qchase.qclass = LDNS_RR_CLASS_ANY;
3222 while(iter_get_next_root(qstate->env->hints,
3223 qstate->env->fwds, &c)) {
3224 /* generate query for this class */
3225 log_nametypeclass(VERB_ALGO, "spawn collect query",
3226 qstate->qinfo.qname, qstate->qinfo.qtype, c);
3227 if(!generate_sub_request(qstate->qinfo.qname,
3228 qstate->qinfo.qname_len, qstate->qinfo.qtype,
3229 c, qstate, id, iq, INIT_REQUEST_STATE,
3230 FINISHED_STATE, &subq,
3231 (int)!(qstate->query_flags&BIT_CD))) {
3232 return error_response(qstate, id,
3233 LDNS_RCODE_SERVFAIL);
3235 /* ignore subq, no special init required */
3236 iq->num_current_queries ++;
3241 /* if no roots are configured at all, return */
3242 if(iq->num_current_queries == 0) {
3243 verbose(VERB_ALGO, "No root hints or fwds, giving up "
3245 return error_response(qstate, id, LDNS_RCODE_REFUSED);
3247 /* return false, wait for queries to return */
3249 /* if woke up here because of an answer, wait for more answers */
3254 * This handles the final state for first-tier responses (i.e., responses to
3255 * externally generated queries).
3257 * @param qstate: query state.
3258 * @param iq: iterator query state.
3259 * @param id: module id.
3260 * @return true if the event needs more processing, false if not. Since this
3261 * is the final state for an event, it always returns false.
3264 processFinished(struct module_qstate* qstate, struct iter_qstate* iq,
3267 log_query_info(VERB_QUERY, "finishing processing for",
3270 /* store negative cache element for parent side glue. */
3271 if(!qstate->no_cache_store && iq->query_for_pside_glue
3273 iter_store_parentside_neg(qstate->env, &qstate->qinfo,
3274 iq->deleg_msg?iq->deleg_msg->rep:
3275 (iq->response?iq->response->rep:NULL));
3277 verbose(VERB_ALGO, "No response is set, servfail");
3278 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3281 /* Make sure that the RA flag is set (since the presence of
3282 * this module means that recursion is available) */
3283 iq->response->rep->flags |= BIT_RA;
3285 /* Clear the AA flag */
3286 /* FIXME: does this action go here or in some other module? */
3287 iq->response->rep->flags &= ~BIT_AA;
3289 /* make sure QR flag is on */
3290 iq->response->rep->flags |= BIT_QR;
3292 /* we have finished processing this query */
3293 qstate->ext_state[id] = module_finished;
3295 /* TODO: we are using a private TTL, trim the response. */
3296 /* if (mPrivateTTL > 0){IterUtils.setPrivateTTL(resp, mPrivateTTL); } */
3298 /* prepend any items we have accumulated */
3299 if(iq->an_prepend_list || iq->ns_prepend_list) {
3300 if(!iter_prepend(iq, iq->response, qstate->region)) {
3301 log_err("prepend rrsets: out of memory");
3302 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3304 /* reset the query name back */
3305 iq->response->qinfo = qstate->qinfo;
3306 /* the security state depends on the combination */
3307 iq->response->rep->security = sec_status_unchecked;
3308 /* store message with the finished prepended items,
3309 * but only if we did recursion. The nonrecursion referral
3310 * from cache does not need to be stored in the msg cache. */
3311 if(!qstate->no_cache_store && qstate->query_flags&BIT_RD) {
3312 iter_dns_store(qstate->env, &qstate->qinfo,
3313 iq->response->rep, 0, qstate->prefetch_leeway,
3314 iq->dp&&iq->dp->has_parent_side_NS,
3315 qstate->region, qstate->query_flags);
3318 qstate->return_rcode = LDNS_RCODE_NOERROR;
3319 qstate->return_msg = iq->response;
3324 * Return priming query results to interested super querystates.
3326 * Sets the delegation point and delegation message (not nonRD queries).
3327 * This is a callback from walk_supers.
3329 * @param qstate: query state that finished.
3330 * @param id: module id.
3331 * @param super: the qstate to inform.
3334 iter_inform_super(struct module_qstate* qstate, int id,
3335 struct module_qstate* super)
3337 if(!qstate->is_priming && super->qinfo.qclass == LDNS_RR_CLASS_ANY)
3338 processClassResponse(qstate, id, super);
3339 else if(super->qinfo.qtype == LDNS_RR_TYPE_DS && ((struct iter_qstate*)
3340 super->minfo[id])->state == DSNS_FIND_STATE)
3341 processDSNSResponse(qstate, id, super);
3342 else if(qstate->return_rcode != LDNS_RCODE_NOERROR)
3343 error_supers(qstate, id, super);
3344 else if(qstate->is_priming)
3345 prime_supers(qstate, id, super);
3346 else processTargetResponse(qstate, id, super);
3350 * Handle iterator state.
3351 * Handle events. This is the real processing loop for events, responsible
3352 * for moving events through the various states. If a processing method
3353 * returns true, then it will be advanced to the next state. If false, then
3354 * processing will stop.
3356 * @param qstate: query state.
3357 * @param ie: iterator shared global environment.
3358 * @param iq: iterator query state.
3359 * @param id: module id.
3362 iter_handle(struct module_qstate* qstate, struct iter_qstate* iq,
3363 struct iter_env* ie, int id)
3367 verbose(VERB_ALGO, "iter_handle processing q with state %s",
3368 iter_state_to_string(iq->state));
3370 case INIT_REQUEST_STATE:
3371 cont = processInitRequest(qstate, iq, ie, id);
3373 case INIT_REQUEST_2_STATE:
3374 cont = processInitRequest2(qstate, iq, id);
3376 case INIT_REQUEST_3_STATE:
3377 cont = processInitRequest3(qstate, iq, id);
3379 case QUERYTARGETS_STATE:
3380 cont = processQueryTargets(qstate, iq, ie, id);
3382 case QUERY_RESP_STATE:
3383 cont = processQueryResponse(qstate, iq, id);
3385 case PRIME_RESP_STATE:
3386 cont = processPrimeResponse(qstate, id);
3388 case COLLECT_CLASS_STATE:
3389 cont = processCollectClass(qstate, id);
3391 case DSNS_FIND_STATE:
3392 cont = processDSNSFind(qstate, iq, id);
3394 case FINISHED_STATE:
3395 cont = processFinished(qstate, iq, id);
3398 log_warn("iterator: invalid state: %d",
3407 * This is the primary entry point for processing request events. Note that
3408 * this method should only be used by external modules.
3409 * @param qstate: query state.
3410 * @param ie: iterator shared global environment.
3411 * @param iq: iterator query state.
3412 * @param id: module id.
3415 process_request(struct module_qstate* qstate, struct iter_qstate* iq,
3416 struct iter_env* ie, int id)
3418 /* external requests start in the INIT state, and finish using the
3419 * FINISHED state. */
3420 iq->state = INIT_REQUEST_STATE;
3421 iq->final_state = FINISHED_STATE;
3422 verbose(VERB_ALGO, "process_request: new external request event");
3423 iter_handle(qstate, iq, ie, id);
3426 /** process authoritative server reply */
3428 process_response(struct module_qstate* qstate, struct iter_qstate* iq,
3429 struct iter_env* ie, int id, struct outbound_entry* outbound,
3430 enum module_ev event)
3432 struct msg_parse* prs;
3433 struct edns_data edns;
3436 verbose(VERB_ALGO, "process_response: new external response event");
3437 iq->response = NULL;
3438 iq->state = QUERY_RESP_STATE;
3439 if(event == module_event_noreply || event == module_event_error) {
3440 if(event == module_event_noreply && iq->sent_count >= 3 &&
3441 qstate->env->cfg->use_caps_bits_for_id &&
3442 !iq->caps_fallback) {
3443 /* start fallback */
3444 iq->caps_fallback = 1;
3445 iq->caps_server = 0;
3446 iq->caps_reply = NULL;
3447 iq->caps_response = NULL;
3448 iq->state = QUERYTARGETS_STATE;
3449 iq->num_current_queries--;
3450 /* need fresh attempts for the 0x20 fallback, if
3451 * that was the cause for the failure */
3452 iter_dec_attempts(iq->dp, 3);
3453 verbose(VERB_DETAIL, "Capsforid: timeouts, starting fallback");
3458 if( (event != module_event_reply && event != module_event_capsfail)
3459 || !qstate->reply) {
3460 log_err("Bad event combined with response");
3461 outbound_list_remove(&iq->outlist, outbound);
3462 (void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3467 prs = (struct msg_parse*)regional_alloc(qstate->env->scratch,
3468 sizeof(struct msg_parse));
3470 log_err("out of memory on incoming message");
3471 /* like packet got dropped */
3474 memset(prs, 0, sizeof(*prs));
3475 memset(&edns, 0, sizeof(edns));
3476 pkt = qstate->reply->c->buffer;
3477 sldns_buffer_set_position(pkt, 0);
3478 if(parse_packet(pkt, prs, qstate->env->scratch) != LDNS_RCODE_NOERROR) {
3479 verbose(VERB_ALGO, "parse error on reply packet");
3482 /* edns is not examined, but removed from message to help cache */
3483 if(parse_extract_edns(prs, &edns, qstate->env->scratch) !=
3487 /* Copy the edns options we may got from the back end */
3489 qstate->edns_opts_back_in = edns_opt_copy_region(edns.opt_list,
3491 if(!qstate->edns_opts_back_in) {
3492 log_err("out of memory on incoming message");
3493 /* like packet got dropped */
3496 if(!inplace_cb_edns_back_parsed_call(qstate->env, qstate)) {
3497 log_err("unable to call edns_back_parsed callback");
3502 /* remove CD-bit, we asked for in case we handle validation ourself */
3503 prs->flags &= ~BIT_CD;
3505 /* normalize and sanitize: easy to delete items from linked lists */
3506 if(!scrub_message(pkt, prs, &iq->qinfo_out, iq->dp->name,
3507 qstate->env->scratch, qstate->env, ie)) {
3508 /* if 0x20 enabled, start fallback, but we have no message */
3509 if(event == module_event_capsfail && !iq->caps_fallback) {
3510 iq->caps_fallback = 1;
3511 iq->caps_server = 0;
3512 iq->caps_reply = NULL;
3513 iq->caps_response = NULL;
3514 iq->state = QUERYTARGETS_STATE;
3515 iq->num_current_queries--;
3516 verbose(VERB_DETAIL, "Capsforid: scrub failed, starting fallback with no response");
3521 /* allocate response dns_msg in region */
3522 iq->response = dns_alloc_msg(pkt, prs, qstate->region);
3525 log_query_info(VERB_DETAIL, "response for", &qstate->qinfo);
3526 log_name_addr(VERB_DETAIL, "reply from", iq->dp->name,
3527 &qstate->reply->addr, qstate->reply->addrlen);
3528 if(verbosity >= VERB_ALGO)
3529 log_dns_msg("incoming scrubbed packet:", &iq->response->qinfo,
3532 if(event == module_event_capsfail || iq->caps_fallback) {
3533 /* for fallback we care about main answer, not additionals */
3534 /* removing that makes comparison more likely to succeed */
3535 caps_strip_reply(iq->response->rep);
3536 if(!iq->caps_fallback) {
3537 /* start fallback */
3538 iq->caps_fallback = 1;
3539 iq->caps_server = 0;
3540 iq->caps_reply = iq->response->rep;
3541 iq->caps_response = iq->response;
3542 iq->state = QUERYTARGETS_STATE;
3543 iq->num_current_queries--;
3544 verbose(VERB_DETAIL, "Capsforid: starting fallback");
3547 /* check if reply is the same, otherwise, fail */
3548 if(!iq->caps_reply) {
3549 iq->caps_reply = iq->response->rep;
3550 iq->caps_response = iq->response;
3551 iq->caps_server = -1; /*become zero at ++,
3552 so that we start the full set of trials */
3553 } else if(caps_failed_rcode(iq->caps_reply) &&
3554 !caps_failed_rcode(iq->response->rep)) {
3555 /* prefer to upgrade to non-SERVFAIL */
3556 iq->caps_reply = iq->response->rep;
3557 iq->caps_response = iq->response;
3558 } else if(!caps_failed_rcode(iq->caps_reply) &&
3559 caps_failed_rcode(iq->response->rep)) {
3560 /* if we have non-SERVFAIL as answer then
3561 * we can ignore SERVFAILs for the equality
3563 /* no instructions here, skip other else */
3564 } else if(caps_failed_rcode(iq->caps_reply) &&
3565 caps_failed_rcode(iq->response->rep)) {
3566 /* failure is same as other failure in fallbk*/
3567 /* no instructions here, skip other else */
3568 } else if(!reply_equal(iq->response->rep, iq->caps_reply,
3569 qstate->env->scratch)) {
3570 verbose(VERB_DETAIL, "Capsforid fallback: "
3571 "getting different replies, failed");
3572 outbound_list_remove(&iq->outlist, outbound);
3573 (void)error_response(qstate, id,
3574 LDNS_RCODE_SERVFAIL);
3577 /* continue the fallback procedure at next server */
3579 iq->state = QUERYTARGETS_STATE;
3580 iq->num_current_queries--;
3581 verbose(VERB_DETAIL, "Capsforid: reply is equal. "
3582 "go to next fallback");
3586 iq->caps_fallback = 0; /* if we were in fallback, 0x20 is OK now */
3589 outbound_list_remove(&iq->outlist, outbound);
3590 iter_handle(qstate, iq, ie, id);
3594 iter_operate(struct module_qstate* qstate, enum module_ev event, int id,
3595 struct outbound_entry* outbound)
3597 struct iter_env* ie = (struct iter_env*)qstate->env->modinfo[id];
3598 struct iter_qstate* iq = (struct iter_qstate*)qstate->minfo[id];
3599 verbose(VERB_QUERY, "iterator[module %d] operate: extstate:%s event:%s",
3600 id, strextstate(qstate->ext_state[id]), strmodulevent(event));
3601 if(iq) log_query_info(VERB_QUERY, "iterator operate: query",
3603 if(iq && qstate->qinfo.qname != iq->qchase.qname)
3604 log_query_info(VERB_QUERY, "iterator operate: chased to",
3607 /* perform iterator state machine */
3608 if((event == module_event_new || event == module_event_pass) &&
3610 if(!iter_new(qstate, id)) {
3611 (void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3614 iq = (struct iter_qstate*)qstate->minfo[id];
3615 process_request(qstate, iq, ie, id);
3618 if(iq && event == module_event_pass) {
3619 iter_handle(qstate, iq, ie, id);
3622 if(iq && outbound) {
3623 process_response(qstate, iq, ie, id, outbound, event);
3626 if(event == module_event_error) {
3627 verbose(VERB_ALGO, "got called with event error, giving up");
3628 (void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3632 log_err("bad event for iterator");
3633 (void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3637 iter_clear(struct module_qstate* qstate, int id)
3639 struct iter_qstate* iq;
3642 iq = (struct iter_qstate*)qstate->minfo[id];
3644 outbound_list_clear(&iq->outlist);
3645 if(iq->target_count && --iq->target_count[0] == 0)
3646 free(iq->target_count);
3647 iq->num_current_queries = 0;
3649 qstate->minfo[id] = NULL;
3653 iter_get_mem(struct module_env* env, int id)
3655 struct iter_env* ie = (struct iter_env*)env->modinfo[id];
3658 return sizeof(*ie) + sizeof(int)*((size_t)ie->max_dependency_depth+1)
3659 + donotq_get_mem(ie->donotq) + priv_get_mem(ie->priv);
3663 * The iterator function block
3665 static struct module_func_block iter_block = {
3667 &iter_init, &iter_deinit, &iter_operate, &iter_inform_super,
3668 &iter_clear, &iter_get_mem
3671 struct module_func_block*
3672 iter_get_funcblock(void)
3678 iter_state_to_string(enum iter_state state)
3682 case INIT_REQUEST_STATE :
3683 return "INIT REQUEST STATE";
3684 case INIT_REQUEST_2_STATE :
3685 return "INIT REQUEST STATE (stage 2)";
3686 case INIT_REQUEST_3_STATE:
3687 return "INIT REQUEST STATE (stage 3)";
3688 case QUERYTARGETS_STATE :
3689 return "QUERY TARGETS STATE";
3690 case PRIME_RESP_STATE :
3691 return "PRIME RESPONSE STATE";
3692 case COLLECT_CLASS_STATE :
3693 return "COLLECT CLASS STATE";
3694 case DSNS_FIND_STATE :
3695 return "DSNS FIND STATE";
3696 case QUERY_RESP_STATE :
3697 return "QUERY RESPONSE STATE";
3698 case FINISHED_STATE :
3699 return "FINISHED RESPONSE STATE";
3701 return "UNKNOWN ITER STATE";
3706 iter_state_is_responsestate(enum iter_state s)
3709 case INIT_REQUEST_STATE :
3710 case INIT_REQUEST_2_STATE :
3711 case INIT_REQUEST_3_STATE :
3712 case QUERYTARGETS_STATE :
3713 case COLLECT_CLASS_STATE :