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,
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) {
554 if((fwddp = forwards_find(env->fwds, nm, qclass)) &&
555 /* has_parent_side is turned off for forward_first, where
556 * we are allowed to go to the parent */
557 fwddp->has_parent_side_NS) {
563 /** see if target name is caps-for-id whitelisted */
565 is_caps_whitelisted(struct iter_env* ie, struct iter_qstate* iq)
567 if(!ie->caps_white) return 0; /* no whitelist, or no capsforid */
568 return name_tree_lookup(ie->caps_white, iq->qchase.qname,
569 iq->qchase.qname_len, dname_count_labels(iq->qchase.qname),
570 iq->qchase.qclass) != NULL;
573 /** create target count structure for this query */
575 target_count_create(struct iter_qstate* iq)
577 if(!iq->target_count) {
578 iq->target_count = (int*)calloc(2, sizeof(int));
579 /* if calloc fails we simply do not track this number */
581 iq->target_count[0] = 1;
586 target_count_increase(struct iter_qstate* iq, int num)
588 target_count_create(iq);
590 iq->target_count[1] += num;
594 * Generate a subrequest.
595 * Generate a local request event. Local events are tied to this module, and
596 * have a corresponding (first tier) event that is waiting for this event to
597 * resolve to continue.
599 * @param qname The query name for this request.
600 * @param qnamelen length of qname
601 * @param qtype The query type for this request.
602 * @param qclass The query class for this request.
603 * @param qstate The event that is generating this event.
604 * @param id: module id.
605 * @param iq: The iterator state that is generating this event.
606 * @param initial_state The initial response state (normally this
607 * is QUERY_RESP_STATE, unless it is known that the request won't
608 * need iterative processing
609 * @param finalstate The final state for the response to this request.
610 * @param subq_ret: if newly allocated, the subquerystate, or NULL if it does
611 * not need initialisation.
612 * @param v: if true, validation is done on the subquery.
613 * @return false on error (malloc).
616 generate_sub_request(uint8_t* qname, size_t qnamelen, uint16_t qtype,
617 uint16_t qclass, struct module_qstate* qstate, int id,
618 struct iter_qstate* iq, enum iter_state initial_state,
619 enum iter_state finalstate, struct module_qstate** subq_ret, int v)
621 struct module_qstate* subq = NULL;
622 struct iter_qstate* subiq = NULL;
623 uint16_t qflags = 0; /* OPCODE QUERY, no flags */
624 struct query_info qinf;
625 int prime = (finalstate == PRIME_RESP_STATE)?1:0;
628 qinf.qname_len = qnamelen;
630 qinf.qclass = qclass;
631 qinf.local_alias = NULL;
633 /* RD should be set only when sending the query back through the INIT
635 if(initial_state == INIT_REQUEST_STATE)
637 /* We set the CD flag so we can send this through the "head" of
638 * the resolution chain, which might have a validator. We are
639 * uninterested in validating things not on the direct resolution
646 /* attach subquery, lookup existing or make a new one */
647 fptr_ok(fptr_whitelist_modenv_attach_sub(qstate->env->attach_sub));
648 if(!(*qstate->env->attach_sub)(qstate, &qinf, qflags, prime, valrec,
654 /* initialise the new subquery */
656 subq->ext_state[id] = module_state_initial;
657 subq->minfo[id] = regional_alloc(subq->region,
658 sizeof(struct iter_qstate));
659 if(!subq->minfo[id]) {
660 log_err("init subq: out of memory");
661 fptr_ok(fptr_whitelist_modenv_kill_sub(
662 qstate->env->kill_sub));
663 (*qstate->env->kill_sub)(subq);
666 subiq = (struct iter_qstate*)subq->minfo[id];
667 memset(subiq, 0, sizeof(*subiq));
668 subiq->num_target_queries = 0;
669 target_count_create(iq);
670 subiq->target_count = iq->target_count;
672 iq->target_count[0] ++; /* extra reference */
673 subiq->num_current_queries = 0;
674 subiq->depth = iq->depth+1;
675 outbound_list_init(&subiq->outlist);
676 subiq->state = initial_state;
677 subiq->final_state = finalstate;
678 subiq->qchase = subq->qinfo;
679 subiq->chase_flags = subq->query_flags;
680 subiq->refetch_glue = 0;
681 if(qstate->env->cfg->qname_minimisation)
682 subiq->minimisation_state = INIT_MINIMISE_STATE;
684 subiq->minimisation_state = DONOT_MINIMISE_STATE;
685 memset(&subiq->qinfo_out, 0, sizeof(struct query_info));
691 * Generate and send a root priming request.
692 * @param qstate: the qtstate that triggered the need to prime.
693 * @param iq: iterator query state.
694 * @param id: module id.
695 * @param qclass: the class to prime.
696 * @return 0 on failure
699 prime_root(struct module_qstate* qstate, struct iter_qstate* iq, int id,
703 struct module_qstate* subq;
704 verbose(VERB_DETAIL, "priming . %s NS",
705 sldns_lookup_by_id(sldns_rr_classes, (int)qclass)?
706 sldns_lookup_by_id(sldns_rr_classes, (int)qclass)->name:"??");
707 dp = hints_lookup_root(qstate->env->hints, qclass);
709 verbose(VERB_ALGO, "Cannot prime due to lack of hints");
712 /* Priming requests start at the QUERYTARGETS state, skipping
713 * the normal INIT state logic (which would cause an infloop). */
714 if(!generate_sub_request((uint8_t*)"\000", 1, LDNS_RR_TYPE_NS,
715 qclass, qstate, id, iq, QUERYTARGETS_STATE, PRIME_RESP_STATE,
717 verbose(VERB_ALGO, "could not prime root");
721 struct iter_qstate* subiq =
722 (struct iter_qstate*)subq->minfo[id];
723 /* Set the initial delegation point to the hint.
724 * copy dp, it is now part of the root prime query.
725 * dp was part of in the fixed hints structure. */
726 subiq->dp = delegpt_copy(dp, subq->region);
728 log_err("out of memory priming root, copydp");
729 fptr_ok(fptr_whitelist_modenv_kill_sub(
730 qstate->env->kill_sub));
731 (*qstate->env->kill_sub)(subq);
734 /* there should not be any target queries. */
735 subiq->num_target_queries = 0;
736 subiq->dnssec_expected = iter_indicates_dnssec(
737 qstate->env, subiq->dp, NULL, subq->qinfo.qclass);
740 /* this module stops, our submodule starts, and does the query. */
741 qstate->ext_state[id] = module_wait_subquery;
746 * Generate and process a stub priming request. This method tests for the
747 * need to prime a stub zone, so it is safe to call for every request.
749 * @param qstate: the qtstate that triggered the need to prime.
750 * @param iq: iterator query state.
751 * @param id: module id.
752 * @param qname: request name.
753 * @param qclass: request class.
754 * @return true if a priming subrequest was made, false if not. The will only
755 * issue a priming request if it detects an unprimed stub.
756 * Uses value of 2 to signal during stub-prime in root-prime situation
757 * that a noprime-stub is available and resolution can continue.
760 prime_stub(struct module_qstate* qstate, struct iter_qstate* iq, int id,
761 uint8_t* qname, uint16_t qclass)
763 /* Lookup the stub hint. This will return null if the stub doesn't
764 * need to be re-primed. */
765 struct iter_hints_stub* stub;
766 struct delegpt* stub_dp;
767 struct module_qstate* subq;
770 stub = hints_lookup_stub(qstate->env->hints, qname, qclass, iq->dp);
771 /* The stub (if there is one) does not need priming. */
775 /* if we have an auth_zone dp, and stub is equal, don't prime stub
776 * yet, unless we want to fallback and avoid the auth_zone */
777 if(!iq->auth_zone_avoid && iq->dp && iq->dp->auth_dp &&
778 query_dname_compare(iq->dp->name, stub_dp->name) == 0)
781 /* is it a noprime stub (always use) */
784 if(iq->dp == NULL) r = 2;
785 /* copy the dp out of the fixed hints structure, so that
786 * it can be changed when servicing this query */
787 iq->dp = delegpt_copy(stub_dp, qstate->region);
789 log_err("out of memory priming stub");
790 (void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
791 return 1; /* return 1 to make module stop, with error */
793 log_nametypeclass(VERB_DETAIL, "use stub", stub_dp->name,
794 LDNS_RR_TYPE_NS, qclass);
798 /* Otherwise, we need to (re)prime the stub. */
799 log_nametypeclass(VERB_DETAIL, "priming stub", stub_dp->name,
800 LDNS_RR_TYPE_NS, qclass);
802 /* Stub priming events start at the QUERYTARGETS state to avoid the
803 * redundant INIT state processing. */
804 if(!generate_sub_request(stub_dp->name, stub_dp->namelen,
805 LDNS_RR_TYPE_NS, qclass, qstate, id, iq,
806 QUERYTARGETS_STATE, PRIME_RESP_STATE, &subq, 0)) {
807 verbose(VERB_ALGO, "could not prime stub");
808 (void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
809 return 1; /* return 1 to make module stop, with error */
812 struct iter_qstate* subiq =
813 (struct iter_qstate*)subq->minfo[id];
815 /* Set the initial delegation point to the hint. */
816 /* make copy to avoid use of stub dp by different qs/threads */
817 subiq->dp = delegpt_copy(stub_dp, subq->region);
819 log_err("out of memory priming stub, copydp");
820 fptr_ok(fptr_whitelist_modenv_kill_sub(
821 qstate->env->kill_sub));
822 (*qstate->env->kill_sub)(subq);
823 (void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
824 return 1; /* return 1 to make module stop, with error */
826 /* there should not be any target queries -- although there
827 * wouldn't be anyway, since stub hints never have
828 * missing targets. */
829 subiq->num_target_queries = 0;
830 subiq->wait_priming_stub = 1;
831 subiq->dnssec_expected = iter_indicates_dnssec(
832 qstate->env, subiq->dp, NULL, subq->qinfo.qclass);
835 /* this module stops, our submodule starts, and does the query. */
836 qstate->ext_state[id] = module_wait_subquery;
841 * Generate a delegation point for an auth zone (unless cached dp is better)
842 * false on alloc failure.
845 auth_zone_delegpt(struct module_qstate* qstate, struct iter_qstate* iq,
846 uint8_t* delname, size_t delnamelen)
849 if(iq->auth_zone_avoid)
852 delname = iq->qchase.qname;
853 delnamelen = iq->qchase.qname_len;
855 lock_rw_rdlock(&qstate->env->auth_zones->lock);
856 z = auth_zones_find_zone(qstate->env->auth_zones, delname, delnamelen,
857 qstate->qinfo.qclass);
859 lock_rw_unlock(&qstate->env->auth_zones->lock);
862 lock_rw_rdlock(&z->lock);
863 lock_rw_unlock(&qstate->env->auth_zones->lock);
864 if(z->for_upstream) {
865 if(iq->dp && query_dname_compare(z->name, iq->dp->name) == 0
866 && iq->dp->auth_dp && qstate->blacklist &&
867 z->fallback_enabled) {
868 /* cache is blacklisted and fallback, and we
869 * already have an auth_zone dp */
870 if(verbosity>=VERB_ALGO) {
872 dname_str(z->name, buf);
873 verbose(VERB_ALGO, "auth_zone %s "
874 "fallback because cache blacklisted",
877 lock_rw_unlock(&z->lock);
881 if(iq->dp==NULL || dname_subdomain_c(z->name, iq->dp->name)) {
883 if(qstate->blacklist && z->fallback_enabled) {
884 /* cache is blacklisted because of a DNSSEC
885 * validation failure, and the zone allows
886 * fallback to the internet, query there. */
887 if(verbosity>=VERB_ALGO) {
889 dname_str(z->name, buf);
890 verbose(VERB_ALGO, "auth_zone %s "
891 "fallback because cache blacklisted",
894 lock_rw_unlock(&z->lock);
897 dp = (struct delegpt*)regional_alloc_zero(
898 qstate->region, sizeof(*dp));
900 log_err("alloc failure");
901 if(z->fallback_enabled) {
902 lock_rw_unlock(&z->lock);
903 return 1; /* just fallback */
905 lock_rw_unlock(&z->lock);
908 dp->name = regional_alloc_init(qstate->region,
909 z->name, z->namelen);
911 log_err("alloc failure");
912 if(z->fallback_enabled) {
913 lock_rw_unlock(&z->lock);
914 return 1; /* just fallback */
916 lock_rw_unlock(&z->lock);
919 dp->namelen = z->namelen;
920 dp->namelabs = z->namelabs;
926 lock_rw_unlock(&z->lock);
931 * Generate A and AAAA checks for glue that is in-zone for the referral
932 * we just got to obtain authoritative information on the addresses.
934 * @param qstate: the qtstate that triggered the need to prime.
935 * @param iq: iterator query state.
936 * @param id: module id.
939 generate_a_aaaa_check(struct module_qstate* qstate, struct iter_qstate* iq,
942 struct iter_env* ie = (struct iter_env*)qstate->env->modinfo[id];
943 struct module_qstate* subq;
945 struct reply_info* rep = iq->response->rep;
946 struct ub_packed_rrset_key* s;
949 if(iq->depth == ie->max_dependency_depth)
951 /* walk through additional, and check if in-zone,
952 * only relevant A, AAAA are left after scrub anyway */
953 for(i=rep->an_numrrsets+rep->ns_numrrsets; i<rep->rrset_count; i++) {
955 /* check *ALL* addresses that are transmitted in additional*/
956 /* is it an address ? */
957 if( !(ntohs(s->rk.type)==LDNS_RR_TYPE_A ||
958 ntohs(s->rk.type)==LDNS_RR_TYPE_AAAA)) {
961 /* is this query the same as the A/AAAA check for it */
962 if(qstate->qinfo.qtype == ntohs(s->rk.type) &&
963 qstate->qinfo.qclass == ntohs(s->rk.rrset_class) &&
964 query_dname_compare(qstate->qinfo.qname,
966 (qstate->query_flags&BIT_RD) &&
967 !(qstate->query_flags&BIT_CD))
970 /* generate subrequest for it */
971 log_nametypeclass(VERB_ALGO, "schedule addr fetch",
972 s->rk.dname, ntohs(s->rk.type),
973 ntohs(s->rk.rrset_class));
974 if(!generate_sub_request(s->rk.dname, s->rk.dname_len,
975 ntohs(s->rk.type), ntohs(s->rk.rrset_class),
977 INIT_REQUEST_STATE, FINISHED_STATE, &subq, 1)) {
978 verbose(VERB_ALGO, "could not generate addr check");
981 /* ignore subq - not need for more init */
986 * Generate a NS check request to obtain authoritative information
989 * @param qstate: the qtstate that triggered the need to prime.
990 * @param iq: iterator query state.
991 * @param id: module id.
994 generate_ns_check(struct module_qstate* qstate, struct iter_qstate* iq, int id)
996 struct iter_env* ie = (struct iter_env*)qstate->env->modinfo[id];
997 struct module_qstate* subq;
1000 if(iq->depth == ie->max_dependency_depth)
1002 if(!can_have_last_resort(qstate->env, iq->dp->name, iq->dp->namelen,
1005 /* is this query the same as the nscheck? */
1006 if(qstate->qinfo.qtype == LDNS_RR_TYPE_NS &&
1007 query_dname_compare(iq->dp->name, qstate->qinfo.qname)==0 &&
1008 (qstate->query_flags&BIT_RD) && !(qstate->query_flags&BIT_CD)){
1009 /* spawn off A, AAAA queries for in-zone glue to check */
1010 generate_a_aaaa_check(qstate, iq, id);
1013 /* no need to get the NS record for DS, it is above the zonecut */
1014 if(qstate->qinfo.qtype == LDNS_RR_TYPE_DS)
1017 log_nametypeclass(VERB_ALGO, "schedule ns fetch",
1018 iq->dp->name, LDNS_RR_TYPE_NS, iq->qchase.qclass);
1019 if(!generate_sub_request(iq->dp->name, iq->dp->namelen,
1020 LDNS_RR_TYPE_NS, iq->qchase.qclass, qstate, id, iq,
1021 INIT_REQUEST_STATE, FINISHED_STATE, &subq, 1)) {
1022 verbose(VERB_ALGO, "could not generate ns check");
1026 struct iter_qstate* subiq =
1027 (struct iter_qstate*)subq->minfo[id];
1029 /* make copy to avoid use of stub dp by different qs/threads */
1030 /* refetch glue to start higher up the tree */
1031 subiq->refetch_glue = 1;
1032 subiq->dp = delegpt_copy(iq->dp, subq->region);
1034 log_err("out of memory generating ns check, copydp");
1035 fptr_ok(fptr_whitelist_modenv_kill_sub(
1036 qstate->env->kill_sub));
1037 (*qstate->env->kill_sub)(subq);
1044 * Generate a DNSKEY prefetch query to get the DNSKEY for the DS record we
1045 * just got in a referral (where we have dnssec_expected, thus have trust
1046 * anchors above it). Note that right after calling this routine the
1047 * iterator detached subqueries (because of following the referral), and thus
1048 * the DNSKEY query becomes detached, its return stored in the cache for
1049 * later lookup by the validator. This cache lookup by the validator avoids
1050 * the roundtrip incurred by the DNSKEY query. The DNSKEY query is now
1051 * performed at about the same time the original query is sent to the domain,
1052 * thus the two answers are likely to be returned at about the same time,
1053 * saving a roundtrip from the validated lookup.
1055 * @param qstate: the qtstate that triggered the need to prime.
1056 * @param iq: iterator query state.
1057 * @param id: module id.
1060 generate_dnskey_prefetch(struct module_qstate* qstate,
1061 struct iter_qstate* iq, int id)
1063 struct module_qstate* subq;
1066 /* is this query the same as the prefetch? */
1067 if(qstate->qinfo.qtype == LDNS_RR_TYPE_DNSKEY &&
1068 query_dname_compare(iq->dp->name, qstate->qinfo.qname)==0 &&
1069 (qstate->query_flags&BIT_RD) && !(qstate->query_flags&BIT_CD)){
1073 /* if the DNSKEY is in the cache this lookup will stop quickly */
1074 log_nametypeclass(VERB_ALGO, "schedule dnskey prefetch",
1075 iq->dp->name, LDNS_RR_TYPE_DNSKEY, iq->qchase.qclass);
1076 if(!generate_sub_request(iq->dp->name, iq->dp->namelen,
1077 LDNS_RR_TYPE_DNSKEY, iq->qchase.qclass, qstate, id, iq,
1078 INIT_REQUEST_STATE, FINISHED_STATE, &subq, 0)) {
1079 /* we'll be slower, but it'll work */
1080 verbose(VERB_ALGO, "could not generate dnskey prefetch");
1084 struct iter_qstate* subiq =
1085 (struct iter_qstate*)subq->minfo[id];
1086 /* this qstate has the right delegation for the dnskey lookup*/
1087 /* make copy to avoid use of stub dp by different qs/threads */
1088 subiq->dp = delegpt_copy(iq->dp, subq->region);
1089 /* if !subiq->dp, it'll start from the cache, no problem */
1094 * See if the query needs forwarding.
1096 * @param qstate: query state.
1097 * @param iq: iterator query state.
1098 * @return true if the request is forwarded, false if not.
1099 * If returns true but, iq->dp is NULL then a malloc failure occurred.
1102 forward_request(struct module_qstate* qstate, struct iter_qstate* iq)
1105 uint8_t* delname = iq->qchase.qname;
1106 size_t delnamelen = iq->qchase.qname_len;
1107 if(iq->refetch_glue) {
1108 delname = iq->dp->name;
1109 delnamelen = iq->dp->namelen;
1111 /* strip one label off of DS query to lookup higher for it */
1112 if( (iq->qchase.qtype == LDNS_RR_TYPE_DS || iq->refetch_glue)
1113 && !dname_is_root(iq->qchase.qname))
1114 dname_remove_label(&delname, &delnamelen);
1115 dp = forwards_lookup(qstate->env->fwds, delname, iq->qchase.qclass);
1118 /* send recursion desired to forward addr */
1119 iq->chase_flags |= BIT_RD;
1120 iq->dp = delegpt_copy(dp, qstate->region);
1121 /* iq->dp checked by caller */
1122 verbose(VERB_ALGO, "forwarding request");
1127 * Process the initial part of the request handling. This state roughly
1128 * corresponds to resolver algorithms steps 1 (find answer in cache) and 2
1129 * (find the best servers to ask).
1131 * Note that all requests start here, and query restarts revisit this state.
1133 * This state either generates: 1) a response, from cache or error, 2) a
1134 * priming event, or 3) forwards the request to the next state (init2,
1137 * @param qstate: query state.
1138 * @param iq: iterator query state.
1139 * @param ie: iterator shared global environment.
1140 * @param id: module id.
1141 * @return true if the event needs more request processing immediately,
1145 processInitRequest(struct module_qstate* qstate, struct iter_qstate* iq,
1146 struct iter_env* ie, int id)
1150 struct dns_msg* msg = NULL;
1152 log_query_info(VERB_DETAIL, "resolving", &qstate->qinfo);
1155 /* We enforce a maximum number of query restarts. This is primarily a
1156 * cheap way to prevent CNAME loops. */
1157 if(iq->query_restart_count > MAX_RESTART_COUNT) {
1158 verbose(VERB_QUERY, "request has exceeded the maximum number"
1159 " of query restarts with %d", iq->query_restart_count);
1160 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1163 /* We enforce a maximum recursion/dependency depth -- in general,
1164 * this is unnecessary for dependency loops (although it will
1165 * catch those), but it provides a sensible limit to the amount
1166 * of work required to answer a given query. */
1167 verbose(VERB_ALGO, "request has dependency depth of %d", iq->depth);
1168 if(iq->depth > ie->max_dependency_depth) {
1169 verbose(VERB_QUERY, "request has exceeded the maximum "
1170 "dependency depth with depth of %d", iq->depth);
1171 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1174 /* If the request is qclass=ANY, setup to generate each class */
1175 if(qstate->qinfo.qclass == LDNS_RR_CLASS_ANY) {
1176 iq->qchase.qclass = 0;
1177 return next_state(iq, COLLECT_CLASS_STATE);
1181 * If we are restricted by a forward-zone or a stub-zone, we
1182 * can't re-fetch glue for this delegation point.
1183 * we won’t try to re-fetch glue if the iq->dp is null.
1185 if (iq->refetch_glue &&
1187 !can_have_last_resort(qstate->env,
1190 iq->qchase.qclass)) {
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))) {
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 if(!can_have_last_resort(qstate->env, iq->dp->name, iq->dp->namelen, iq->qchase.qclass)) {
1420 verbose(VERB_ALGO, "useless dp "
1421 "but cannot go up, servfail");
1422 return error_response(qstate, id,
1423 LDNS_RCODE_SERVFAIL);
1425 if(dname_is_root(iq->dp->name)) {
1426 /* use safety belt */
1427 verbose(VERB_QUERY, "Cache has root NS but "
1428 "no addresses. Fallback to the safety belt.");
1429 iq->dp = hints_lookup_root(qstate->env->hints,
1431 /* note deleg_msg is from previous lookup,
1432 * but RD is on, so it is not used */
1434 log_err("internal error: no hints dp");
1435 return error_response(qstate, id,
1436 LDNS_RCODE_REFUSED);
1438 iq->dp = delegpt_copy(iq->dp, qstate->region);
1440 log_err("out of memory in safety belt");
1441 return error_response(qstate, id,
1442 LDNS_RCODE_SERVFAIL);
1447 "cache delegation was useless:");
1448 delegpt_log(VERB_ALGO, iq->dp);
1450 delname = iq->dp->name;
1451 delnamelen = iq->dp->namelen;
1452 dname_remove_label(&delname, &delnamelen);
1457 verbose(VERB_ALGO, "cache delegation returns delegpt");
1458 delegpt_log(VERB_ALGO, iq->dp);
1460 /* Otherwise, set the current delegation point and move on to the
1462 return next_state(iq, INIT_REQUEST_2_STATE);
1466 * Process the second part of the initial request handling. This state
1467 * basically exists so that queries that generate root priming events have
1468 * the same init processing as ones that do not. Request events that reach
1469 * this state must have a valid currentDelegationPoint set.
1471 * This part is primarily handling stub zone priming. Events that reach this
1472 * state must have a current delegation point.
1474 * @param qstate: query state.
1475 * @param iq: iterator query state.
1476 * @param id: module id.
1477 * @return true if the event needs more request processing immediately,
1481 processInitRequest2(struct module_qstate* qstate, struct iter_qstate* iq,
1486 log_query_info(VERB_QUERY, "resolving (init part 2): ",
1489 delname = iq->qchase.qname;
1490 delnamelen = iq->qchase.qname_len;
1491 if(iq->refetch_glue) {
1492 struct iter_hints_stub* stub;
1494 log_err("internal or malloc fail: no dp for refetch");
1495 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1497 /* Do not send queries above stub, do not set delname to dp if
1498 * this is above stub without stub-first. */
1499 stub = hints_lookup_stub(
1500 qstate->env->hints, iq->qchase.qname, iq->qchase.qclass,
1502 if(!stub || !stub->dp->has_parent_side_NS ||
1503 dname_subdomain_c(iq->dp->name, stub->dp->name)) {
1504 delname = iq->dp->name;
1505 delnamelen = iq->dp->namelen;
1508 if(iq->qchase.qtype == LDNS_RR_TYPE_DS || iq->refetch_glue) {
1509 if(!dname_is_root(delname))
1510 dname_remove_label(&delname, &delnamelen);
1511 iq->refetch_glue = 0; /* if CNAME causes restart, no refetch */
1514 /* see if we have an auth zone to answer from, improves dp from cache
1515 * (if any dp from cache) with auth zone dp, if that is lower */
1516 if(!auth_zone_delegpt(qstate, iq, delname, delnamelen))
1517 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1519 /* Check to see if we need to prime a stub zone. */
1520 if(prime_stub(qstate, iq, id, delname, iq->qchase.qclass)) {
1521 /* A priming sub request was made */
1525 /* most events just get forwarded to the next state. */
1526 return next_state(iq, INIT_REQUEST_3_STATE);
1530 * Process the third part of the initial request handling. This state exists
1531 * as a separate state so that queries that generate stub priming events
1532 * will get the tail end of the init process but not repeat the stub priming
1535 * @param qstate: query state.
1536 * @param iq: iterator query state.
1537 * @param id: module id.
1538 * @return true, advancing the event to the QUERYTARGETS_STATE.
1541 processInitRequest3(struct module_qstate* qstate, struct iter_qstate* iq,
1544 log_query_info(VERB_QUERY, "resolving (init part 3): ",
1546 /* if the cache reply dp equals a validation anchor or msg has DS,
1547 * then DNSSEC RRSIGs are expected in the reply */
1548 iq->dnssec_expected = iter_indicates_dnssec(qstate->env, iq->dp,
1549 iq->deleg_msg, iq->qchase.qclass);
1551 /* If the RD flag wasn't set, then we just finish with the
1552 * cached referral as the response. */
1553 if(!(qstate->query_flags & BIT_RD) && iq->deleg_msg) {
1554 iq->response = iq->deleg_msg;
1555 if(verbosity >= VERB_ALGO && iq->response)
1556 log_dns_msg("no RD requested, using delegation msg",
1557 &iq->response->qinfo, iq->response->rep);
1558 if(qstate->reply_origin)
1559 sock_list_insert(&qstate->reply_origin, NULL, 0, qstate->region);
1560 return final_state(iq);
1562 /* After this point, unset the RD flag -- this query is going to
1563 * be sent to an auth. server. */
1564 iq->chase_flags &= ~BIT_RD;
1566 /* if dnssec expected, fetch key for the trust-anchor or cached-DS */
1567 if(iq->dnssec_expected && qstate->env->cfg->prefetch_key &&
1568 !(qstate->query_flags&BIT_CD)) {
1569 generate_dnskey_prefetch(qstate, iq, id);
1570 fptr_ok(fptr_whitelist_modenv_detach_subs(
1571 qstate->env->detach_subs));
1572 (*qstate->env->detach_subs)(qstate);
1575 /* Jump to the next state. */
1576 return next_state(iq, QUERYTARGETS_STATE);
1580 * Given a basic query, generate a parent-side "target" query.
1581 * These are subordinate queries for missing delegation point target addresses,
1582 * for which only the parent of the delegation provides correct IP addresses.
1584 * @param qstate: query state.
1585 * @param iq: iterator query state.
1586 * @param id: module id.
1587 * @param name: target qname.
1588 * @param namelen: target qname length.
1589 * @param qtype: target qtype (either A or AAAA).
1590 * @param qclass: target qclass.
1591 * @return true on success, false on failure.
1594 generate_parentside_target_query(struct module_qstate* qstate,
1595 struct iter_qstate* iq, int id, uint8_t* name, size_t namelen,
1596 uint16_t qtype, uint16_t qclass)
1598 struct module_qstate* subq;
1599 if(!generate_sub_request(name, namelen, qtype, qclass, qstate,
1600 id, iq, INIT_REQUEST_STATE, FINISHED_STATE, &subq, 0))
1603 struct iter_qstate* subiq =
1604 (struct iter_qstate*)subq->minfo[id];
1605 /* blacklist the cache - we want to fetch parent stuff */
1606 sock_list_insert(&subq->blacklist, NULL, 0, subq->region);
1607 subiq->query_for_pside_glue = 1;
1608 if(dname_subdomain_c(name, iq->dp->name)) {
1609 subiq->dp = delegpt_copy(iq->dp, subq->region);
1610 subiq->dnssec_expected = iter_indicates_dnssec(
1611 qstate->env, subiq->dp, NULL,
1612 subq->qinfo.qclass);
1613 subiq->refetch_glue = 1;
1615 subiq->dp = dns_cache_find_delegation(qstate->env,
1616 name, namelen, qtype, qclass, subq->region,
1618 *qstate->env->now+subq->prefetch_leeway);
1619 /* if no dp, then it's from root, refetch unneeded */
1621 subiq->dnssec_expected = iter_indicates_dnssec(
1622 qstate->env, subiq->dp, NULL,
1623 subq->qinfo.qclass);
1624 subiq->refetch_glue = 1;
1628 log_nametypeclass(VERB_QUERY, "new pside target", name, qtype, qclass);
1633 * Given a basic query, generate a "target" query. These are subordinate
1634 * queries for missing delegation point target addresses.
1636 * @param qstate: query state.
1637 * @param iq: iterator query state.
1638 * @param id: module id.
1639 * @param name: target qname.
1640 * @param namelen: target qname length.
1641 * @param qtype: target qtype (either A or AAAA).
1642 * @param qclass: target qclass.
1643 * @return true on success, false on failure.
1646 generate_target_query(struct module_qstate* qstate, struct iter_qstate* iq,
1647 int id, uint8_t* name, size_t namelen, uint16_t qtype, uint16_t qclass)
1649 struct module_qstate* subq;
1650 if(!generate_sub_request(name, namelen, qtype, qclass, qstate,
1651 id, iq, INIT_REQUEST_STATE, FINISHED_STATE, &subq, 0))
1653 log_nametypeclass(VERB_QUERY, "new target", name, qtype, qclass);
1658 * Given an event at a certain state, generate zero or more target queries
1659 * for it's current delegation point.
1661 * @param qstate: query state.
1662 * @param iq: iterator query state.
1663 * @param ie: iterator shared global environment.
1664 * @param id: module id.
1665 * @param maxtargets: The maximum number of targets to query for.
1666 * if it is negative, there is no maximum number of targets.
1667 * @param num: returns the number of queries generated and processed,
1668 * which may be zero if there were no missing targets.
1669 * @return false on error.
1672 query_for_targets(struct module_qstate* qstate, struct iter_qstate* iq,
1673 struct iter_env* ie, int id, int maxtargets, int* num)
1675 int query_count = 0;
1676 struct delegpt_ns* ns;
1680 if(iq->depth == ie->max_dependency_depth)
1682 if(iq->depth > 0 && iq->target_count &&
1683 iq->target_count[1] > MAX_TARGET_COUNT) {
1684 char s[LDNS_MAX_DOMAINLEN+1];
1685 dname_str(qstate->qinfo.qname, s);
1686 verbose(VERB_QUERY, "request %s has exceeded the maximum "
1687 "number of glue fetches %d", s, iq->target_count[1]);
1691 iter_mark_cycle_targets(qstate, iq->dp);
1692 missing = (int)delegpt_count_missing_targets(iq->dp);
1693 log_assert(maxtargets != 0); /* that would not be useful */
1695 /* Generate target requests. Basically, any missing targets
1696 * are queried for here, regardless if it is necessary to do
1697 * so to continue processing. */
1698 if(maxtargets < 0 || maxtargets > missing)
1700 else toget = maxtargets;
1705 /* select 'toget' items from the total of 'missing' items */
1706 log_assert(toget <= missing);
1708 /* loop over missing targets */
1709 for(ns = iq->dp->nslist; ns; ns = ns->next) {
1713 /* randomly select this item with probability toget/missing */
1714 if(!iter_ns_probability(qstate->env->rnd, toget, missing)) {
1715 /* do not select this one, next; select toget number
1716 * of items from a list one less in size */
1721 if(ie->supports_ipv6 && !ns->got6) {
1722 /* Send the AAAA request. */
1723 if(!generate_target_query(qstate, iq, id,
1724 ns->name, ns->namelen,
1725 LDNS_RR_TYPE_AAAA, iq->qchase.qclass)) {
1728 qstate->ext_state[id] = module_wait_subquery;
1733 /* Send the A request. */
1734 if(ie->supports_ipv4 && !ns->got4) {
1735 if(!generate_target_query(qstate, iq, id,
1736 ns->name, ns->namelen,
1737 LDNS_RR_TYPE_A, iq->qchase.qclass)) {
1740 qstate->ext_state[id] = module_wait_subquery;
1746 /* mark this target as in progress. */
1755 qstate->ext_state[id] = module_wait_subquery;
1761 * Called by processQueryTargets when it would like extra targets to query
1762 * but it seems to be out of options. At last resort some less appealing
1763 * options are explored. If there are no more options, the result is SERVFAIL
1765 * @param qstate: query state.
1766 * @param iq: iterator query state.
1767 * @param ie: iterator shared global environment.
1768 * @param id: module id.
1769 * @return true if the event requires more request processing immediately,
1773 processLastResort(struct module_qstate* qstate, struct iter_qstate* iq,
1774 struct iter_env* ie, int id)
1776 struct delegpt_ns* ns;
1777 int query_count = 0;
1778 verbose(VERB_ALGO, "No more query targets, attempting last resort");
1781 if(!can_have_last_resort(qstate->env, iq->dp->name, iq->dp->namelen,
1782 iq->qchase.qclass)) {
1783 /* fail -- no more targets, no more hope of targets, no hope
1785 verbose(VERB_QUERY, "configured stub or forward servers failed -- returning SERVFAIL");
1786 return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
1788 if(!iq->dp->has_parent_side_NS && dname_is_root(iq->dp->name)) {
1789 struct delegpt* p = hints_lookup_root(qstate->env->hints,
1792 struct delegpt_ns* ns;
1793 struct delegpt_addr* a;
1794 iq->chase_flags &= ~BIT_RD; /* go to authorities */
1795 for(ns = p->nslist; ns; ns=ns->next) {
1796 (void)delegpt_add_ns(iq->dp, qstate->region,
1797 ns->name, ns->lame);
1799 for(a = p->target_list; a; a=a->next_target) {
1800 (void)delegpt_add_addr(iq->dp, qstate->region,
1801 &a->addr, a->addrlen, a->bogus,
1802 a->lame, a->tls_auth_name);
1805 iq->dp->has_parent_side_NS = 1;
1806 } else if(!iq->dp->has_parent_side_NS) {
1807 if(!iter_lookup_parent_NS_from_cache(qstate->env, iq->dp,
1808 qstate->region, &qstate->qinfo)
1809 || !iq->dp->has_parent_side_NS) {
1810 /* if: malloc failure in lookup go up to try */
1811 /* if: no parent NS in cache - go up one level */
1812 verbose(VERB_ALGO, "try to grab parent NS");
1813 iq->store_parent_NS = iq->dp;
1814 iq->chase_flags &= ~BIT_RD; /* go to authorities */
1815 iq->deleg_msg = NULL;
1816 iq->refetch_glue = 1;
1817 iq->query_restart_count++;
1819 if(qstate->env->cfg->qname_minimisation)
1820 iq->minimisation_state = INIT_MINIMISE_STATE;
1821 return next_state(iq, INIT_REQUEST_STATE);
1824 /* see if that makes new names available */
1825 if(!cache_fill_missing(qstate->env, iq->qchase.qclass,
1826 qstate->region, iq->dp))
1827 log_err("out of memory in cache_fill_missing");
1828 if(iq->dp->usable_list) {
1829 verbose(VERB_ALGO, "try parent-side-name, w. glue from cache");
1830 return next_state(iq, QUERYTARGETS_STATE);
1832 /* try to fill out parent glue from cache */
1833 if(iter_lookup_parent_glue_from_cache(qstate->env, iq->dp,
1834 qstate->region, &qstate->qinfo)) {
1835 /* got parent stuff from cache, see if we can continue */
1836 verbose(VERB_ALGO, "try parent-side glue from cache");
1837 return next_state(iq, QUERYTARGETS_STATE);
1839 /* query for an extra name added by the parent-NS record */
1840 if(delegpt_count_missing_targets(iq->dp) > 0) {
1842 verbose(VERB_ALGO, "try parent-side target name");
1843 if(!query_for_targets(qstate, iq, ie, id, 1, &qs)) {
1844 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1846 iq->num_target_queries += qs;
1847 target_count_increase(iq, qs);
1849 qstate->ext_state[id] = module_wait_subquery;
1850 return 0; /* and wait for them */
1853 if(iq->depth == ie->max_dependency_depth) {
1854 verbose(VERB_QUERY, "maxdepth and need more nameservers, fail");
1855 return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
1857 if(iq->depth > 0 && iq->target_count &&
1858 iq->target_count[1] > MAX_TARGET_COUNT) {
1859 char s[LDNS_MAX_DOMAINLEN+1];
1860 dname_str(qstate->qinfo.qname, s);
1861 verbose(VERB_QUERY, "request %s has exceeded the maximum "
1862 "number of glue fetches %d", s, iq->target_count[1]);
1863 return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
1865 /* mark cycle targets for parent-side lookups */
1866 iter_mark_pside_cycle_targets(qstate, iq->dp);
1867 /* see if we can issue queries to get nameserver addresses */
1868 /* this lookup is not randomized, but sequential. */
1869 for(ns = iq->dp->nslist; ns; ns = ns->next) {
1870 /* if this nameserver is at a delegation point, but that
1871 * delegation point is a stub and we cannot go higher, skip*/
1872 if( ((ie->supports_ipv6 && !ns->done_pside6) ||
1873 (ie->supports_ipv4 && !ns->done_pside4)) &&
1874 !can_have_last_resort(qstate->env, ns->name, ns->namelen,
1875 iq->qchase.qclass)) {
1876 log_nametypeclass(VERB_ALGO, "cannot pside lookup ns "
1877 "because it is also a stub/forward,",
1878 ns->name, LDNS_RR_TYPE_NS, iq->qchase.qclass);
1879 if(ie->supports_ipv6) ns->done_pside6 = 1;
1880 if(ie->supports_ipv4) ns->done_pside4 = 1;
1883 /* query for parent-side A and AAAA for nameservers */
1884 if(ie->supports_ipv6 && !ns->done_pside6) {
1885 /* Send the AAAA request. */
1886 if(!generate_parentside_target_query(qstate, iq, id,
1887 ns->name, ns->namelen,
1888 LDNS_RR_TYPE_AAAA, iq->qchase.qclass))
1889 return error_response(qstate, id,
1890 LDNS_RCODE_SERVFAIL);
1891 ns->done_pside6 = 1;
1894 if(ie->supports_ipv4 && !ns->done_pside4) {
1895 /* Send the A request. */
1896 if(!generate_parentside_target_query(qstate, iq, id,
1897 ns->name, ns->namelen,
1898 LDNS_RR_TYPE_A, iq->qchase.qclass))
1899 return error_response(qstate, id,
1900 LDNS_RCODE_SERVFAIL);
1901 ns->done_pside4 = 1;
1904 if(query_count != 0) { /* suspend to await results */
1905 verbose(VERB_ALGO, "try parent-side glue lookup");
1906 iq->num_target_queries += query_count;
1907 target_count_increase(iq, query_count);
1908 qstate->ext_state[id] = module_wait_subquery;
1913 /* if this was a parent-side glue query itself, then store that
1914 * failure in cache. */
1915 if(!qstate->no_cache_store && iq->query_for_pside_glue
1917 iter_store_parentside_neg(qstate->env, &qstate->qinfo,
1918 iq->deleg_msg?iq->deleg_msg->rep:
1919 (iq->response?iq->response->rep:NULL));
1921 verbose(VERB_QUERY, "out of query targets -- returning SERVFAIL");
1922 /* fail -- no more targets, no more hope of targets, no hope
1924 return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
1928 * Try to find the NS record set that will resolve a qtype DS query. Due
1929 * to grandparent/grandchild reasons we did not get a proper lookup right
1930 * away. We need to create type NS queries until we get the right parent
1931 * for this lookup. We remove labels from the query to find the right point.
1932 * If we end up at the old dp name, then there is no solution.
1934 * @param qstate: query state.
1935 * @param iq: iterator query state.
1936 * @param id: module id.
1937 * @return true if the event requires more immediate processing, false if
1938 * not. This is generally only true when forwarding the request to
1939 * the final state (i.e., on answer).
1942 processDSNSFind(struct module_qstate* qstate, struct iter_qstate* iq, int id)
1944 struct module_qstate* subq = NULL;
1945 verbose(VERB_ALGO, "processDSNSFind");
1947 if(!iq->dsns_point) {
1949 iq->dsns_point = iq->qchase.qname;
1950 iq->dsns_point_len = iq->qchase.qname_len;
1952 /* robustcheck for internal error: we are not underneath the dp */
1953 if(!dname_subdomain_c(iq->dsns_point, iq->dp->name)) {
1954 return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
1957 /* go up one (more) step, until we hit the dp, if so, end */
1958 dname_remove_label(&iq->dsns_point, &iq->dsns_point_len);
1959 if(query_dname_compare(iq->dsns_point, iq->dp->name) == 0) {
1960 /* there was no inbetween nameserver, use the old delegation
1961 * point again. And this time, because dsns_point is nonNULL
1962 * we are going to accept the (bad) result */
1963 iq->state = QUERYTARGETS_STATE;
1966 iq->state = DSNS_FIND_STATE;
1968 /* spawn NS lookup (validation not needed, this is for DS lookup) */
1969 log_nametypeclass(VERB_ALGO, "fetch nameservers",
1970 iq->dsns_point, LDNS_RR_TYPE_NS, iq->qchase.qclass);
1971 if(!generate_sub_request(iq->dsns_point, iq->dsns_point_len,
1972 LDNS_RR_TYPE_NS, iq->qchase.qclass, qstate, id, iq,
1973 INIT_REQUEST_STATE, FINISHED_STATE, &subq, 0)) {
1974 return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
1981 * This is the request event state where the request will be sent to one of
1982 * its current query targets. This state also handles issuing target lookup
1983 * queries for missing target IP addresses. Queries typically iterate on
1984 * this state, both when they are just trying different targets for a given
1985 * delegation point, and when they change delegation points. This state
1986 * roughly corresponds to RFC 1034 algorithm steps 3 and 4.
1988 * @param qstate: query state.
1989 * @param iq: iterator query state.
1990 * @param ie: iterator shared global environment.
1991 * @param id: module id.
1992 * @return true if the event requires more request processing immediately,
1993 * false if not. This state only returns true when it is generating
1994 * a SERVFAIL response because the query has hit a dead end.
1997 processQueryTargets(struct module_qstate* qstate, struct iter_qstate* iq,
1998 struct iter_env* ie, int id)
2001 struct delegpt_addr* target;
2002 struct outbound_entry* outq;
2003 int auth_fallback = 0;
2005 /* NOTE: a request will encounter this state for each target it
2006 * needs to send a query to. That is, at least one per referral,
2007 * more if some targets timeout or return throwaway answers. */
2009 log_query_info(VERB_QUERY, "processQueryTargets:", &qstate->qinfo);
2010 verbose(VERB_ALGO, "processQueryTargets: targetqueries %d, "
2011 "currentqueries %d sentcount %d", iq->num_target_queries,
2012 iq->num_current_queries, iq->sent_count);
2014 /* Make sure that we haven't run away */
2015 /* FIXME: is this check even necessary? */
2016 if(iq->referral_count > MAX_REFERRAL_COUNT) {
2017 verbose(VERB_QUERY, "request has exceeded the maximum "
2018 "number of referrrals with %d", iq->referral_count);
2019 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2021 if(iq->sent_count > MAX_SENT_COUNT) {
2022 verbose(VERB_QUERY, "request has exceeded the maximum "
2023 "number of sends with %d", iq->sent_count);
2024 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2027 /* Make sure we have a delegation point, otherwise priming failed
2028 * or another failure occurred */
2030 verbose(VERB_QUERY, "Failed to get a delegation, giving up");
2031 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2033 if(!ie->supports_ipv6)
2034 delegpt_no_ipv6(iq->dp);
2035 if(!ie->supports_ipv4)
2036 delegpt_no_ipv4(iq->dp);
2037 delegpt_log(VERB_ALGO, iq->dp);
2039 if(iq->num_current_queries>0) {
2040 /* already busy answering a query, this restart is because
2041 * more delegpt addrs became available, wait for existing
2043 verbose(VERB_ALGO, "woke up, but wait for outstanding query");
2044 qstate->ext_state[id] = module_wait_reply;
2048 if(iq->minimisation_state == INIT_MINIMISE_STATE) {
2049 /* (Re)set qinfo_out to (new) delegation point, except when
2050 * qinfo_out is already a subdomain of dp. This happens when
2051 * increasing by more than one label at once (QNAMEs with more
2052 * than MAX_MINIMISE_COUNT labels). */
2053 if(!(iq->qinfo_out.qname_len
2054 && dname_subdomain_c(iq->qchase.qname,
2055 iq->qinfo_out.qname)
2056 && dname_subdomain_c(iq->qinfo_out.qname,
2058 iq->qinfo_out.qname = iq->dp->name;
2059 iq->qinfo_out.qname_len = iq->dp->namelen;
2060 iq->qinfo_out.qtype = LDNS_RR_TYPE_A;
2061 iq->qinfo_out.qclass = iq->qchase.qclass;
2062 iq->qinfo_out.local_alias = NULL;
2063 iq->minimise_count = 0;
2066 iq->minimisation_state = MINIMISE_STATE;
2068 if(iq->minimisation_state == MINIMISE_STATE) {
2069 int qchaselabs = dname_count_labels(iq->qchase.qname);
2070 int labdiff = qchaselabs -
2071 dname_count_labels(iq->qinfo_out.qname);
2073 iq->qinfo_out.qname = iq->qchase.qname;
2074 iq->qinfo_out.qname_len = iq->qchase.qname_len;
2075 iq->minimise_count++;
2076 iq->minimise_timeout_count = 0;
2078 iter_dec_attempts(iq->dp, 1);
2080 /* Limit number of iterations for QNAMEs with more
2081 * than MAX_MINIMISE_COUNT labels. Send first MINIMISE_ONE_LAB
2082 * labels of QNAME always individually.
2084 if(qchaselabs > MAX_MINIMISE_COUNT && labdiff > 1 &&
2085 iq->minimise_count > MINIMISE_ONE_LAB) {
2086 if(iq->minimise_count < MAX_MINIMISE_COUNT) {
2087 int multilabs = qchaselabs - 1 -
2089 int extralabs = multilabs /
2090 MINIMISE_MULTIPLE_LABS;
2092 if (MAX_MINIMISE_COUNT - iq->minimise_count >=
2093 multilabs % MINIMISE_MULTIPLE_LABS)
2094 /* Default behaviour is to add 1 label
2095 * every iteration. Therefore, decrement
2096 * the extralabs by 1 */
2098 if (extralabs < labdiff)
2099 labdiff -= extralabs;
2103 /* Last minimised iteration, send all labels with
2110 verbose(VERB_QUERY, "removing %d labels", labdiff-1);
2111 dname_remove_labels(&iq->qinfo_out.qname,
2112 &iq->qinfo_out.qname_len,
2115 if(labdiff < 1 || (labdiff < 2
2116 && (iq->qchase.qtype == LDNS_RR_TYPE_DS
2117 || iq->qchase.qtype == LDNS_RR_TYPE_A)))
2118 /* Stop minimising this query, resolve "as usual" */
2119 iq->minimisation_state = DONOT_MINIMISE_STATE;
2120 else if(!qstate->no_cache_lookup) {
2121 struct dns_msg* msg = dns_cache_lookup(qstate->env,
2122 iq->qinfo_out.qname, iq->qinfo_out.qname_len,
2123 iq->qinfo_out.qtype, iq->qinfo_out.qclass,
2124 qstate->query_flags, qstate->region,
2125 qstate->env->scratch, 0);
2126 if(msg && msg->rep->an_numrrsets == 0
2127 && FLAGS_GET_RCODE(msg->rep->flags) ==
2129 /* no need to send query if it is already
2130 * cached as NOERROR/NODATA */
2134 if(iq->minimisation_state == SKIP_MINIMISE_STATE) {
2135 if(iq->minimise_timeout_count < MAX_MINIMISE_TIMEOUT_COUNT)
2136 /* Do not increment qname, continue incrementing next
2138 iq->minimisation_state = MINIMISE_STATE;
2139 else if(!qstate->env->cfg->qname_minimisation_strict)
2140 /* Too many time-outs detected for this QNAME and QTYPE.
2141 * We give up, disable QNAME minimisation. */
2142 iq->minimisation_state = DONOT_MINIMISE_STATE;
2144 if(iq->minimisation_state == DONOT_MINIMISE_STATE)
2145 iq->qinfo_out = iq->qchase;
2147 /* now find an answer to this query */
2148 /* see if authority zones have an answer */
2149 /* now we know the dp, we can check the auth zone for locally hosted
2151 if(!iq->auth_zone_avoid && qstate->blacklist) {
2152 if(auth_zones_can_fallback(qstate->env->auth_zones,
2153 iq->dp->name, iq->dp->namelen, iq->qinfo_out.qclass)) {
2154 /* if cache is blacklisted and this zone allows us
2155 * to fallback to the internet, then do so, and
2156 * fetch results from the internet servers */
2157 iq->auth_zone_avoid = 1;
2160 if(iq->auth_zone_avoid) {
2161 iq->auth_zone_avoid = 0;
2163 } else if(auth_zones_lookup(qstate->env->auth_zones, &iq->qinfo_out,
2164 qstate->region, &iq->response, &auth_fallback, iq->dp->name,
2166 /* use this as a response to be processed by the iterator */
2167 if(verbosity >= VERB_ALGO) {
2168 log_dns_msg("msg from auth zone",
2169 &iq->response->qinfo, iq->response->rep);
2171 if((iq->chase_flags&BIT_RD) && !(iq->response->rep->flags&BIT_AA)) {
2172 verbose(VERB_ALGO, "forwarder, ignoring referral from auth zone");
2174 lock_rw_wrlock(&qstate->env->auth_zones->lock);
2175 qstate->env->auth_zones->num_query_up++;
2176 lock_rw_unlock(&qstate->env->auth_zones->lock);
2177 iq->num_current_queries++;
2178 iq->chase_to_rd = 0;
2179 iq->dnssec_lame_query = 0;
2180 iq->auth_zone_response = 1;
2181 return next_state(iq, QUERY_RESP_STATE);
2184 iq->auth_zone_response = 0;
2185 if(auth_fallback == 0) {
2186 /* like we got servfail from the auth zone lookup, and
2187 * no internet fallback */
2188 verbose(VERB_ALGO, "auth zone lookup failed, no fallback,"
2190 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2192 if(iq->dp && iq->dp->auth_dp) {
2193 /* we wanted to fallback, but had no delegpt, only the
2194 * auth zone generated delegpt, create an actual one */
2195 iq->auth_zone_avoid = 1;
2196 return next_state(iq, INIT_REQUEST_STATE);
2198 /* but mostly, fallback==1 (like, when no such auth zone exists)
2199 * and we continue with lookups */
2202 /* < not <=, because although the array is large enough for <=, the
2203 * generated query will immediately be discarded due to depth and
2204 * that servfail is cached, which is not good as opportunism goes. */
2205 if(iq->depth < ie->max_dependency_depth
2206 && iq->sent_count < TARGET_FETCH_STOP) {
2207 tf_policy = ie->target_fetch_policy[iq->depth];
2210 /* if in 0x20 fallback get as many targets as possible */
2211 if(iq->caps_fallback) {
2213 size_t naddr, nres, navail;
2214 if(!query_for_targets(qstate, iq, ie, id, -1, &extra)) {
2215 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2217 iq->num_target_queries += extra;
2218 target_count_increase(iq, extra);
2219 if(iq->num_target_queries > 0) {
2220 /* wait to get all targets, we want to try em */
2221 verbose(VERB_ALGO, "wait for all targets for fallback");
2222 qstate->ext_state[id] = module_wait_reply;
2225 /* did we do enough fallback queries already? */
2226 delegpt_count_addr(iq->dp, &naddr, &nres, &navail);
2227 /* the current caps_server is the number of fallbacks sent.
2228 * the original query is one that matched too, so we have
2229 * caps_server+1 number of matching queries now */
2230 if(iq->caps_server+1 >= naddr*3 ||
2231 iq->caps_server*2+2 >= MAX_SENT_COUNT) {
2232 /* *2 on sentcount check because ipv6 may fail */
2233 /* we're done, process the response */
2234 verbose(VERB_ALGO, "0x20 fallback had %d responses "
2235 "match for %d wanted, done.",
2236 (int)iq->caps_server+1, (int)naddr*3);
2237 iq->response = iq->caps_response;
2238 iq->caps_fallback = 0;
2239 iter_dec_attempts(iq->dp, 3); /* space for fallback */
2240 iq->num_current_queries++; /* RespState decrements it*/
2241 iq->referral_count++; /* make sure we don't loop */
2243 iq->state = QUERY_RESP_STATE;
2246 verbose(VERB_ALGO, "0x20 fallback number %d",
2247 (int)iq->caps_server);
2249 /* if there is a policy to fetch missing targets
2250 * opportunistically, do it. we rely on the fact that once a
2251 * query (or queries) for a missing name have been issued,
2252 * they will not show up again. */
2253 } else if(tf_policy != 0) {
2255 verbose(VERB_ALGO, "attempt to get extra %d targets",
2257 (void)query_for_targets(qstate, iq, ie, id, tf_policy, &extra);
2258 /* errors ignored, these targets are not strictly necessary for
2259 * this result, we do not have to reply with SERVFAIL */
2260 iq->num_target_queries += extra;
2261 target_count_increase(iq, extra);
2264 /* Add the current set of unused targets to our queue. */
2265 delegpt_add_unused_targets(iq->dp);
2267 /* Select the next usable target, filtering out unsuitable targets. */
2268 target = iter_server_selection(ie, qstate->env, iq->dp,
2269 iq->dp->name, iq->dp->namelen, iq->qchase.qtype,
2270 &iq->dnssec_lame_query, &iq->chase_to_rd,
2271 iq->num_target_queries, qstate->blacklist,
2272 qstate->prefetch_leeway);
2274 /* If no usable target was selected... */
2276 /* Here we distinguish between three states: generate a new
2277 * target query, just wait, or quit (with a SERVFAIL).
2278 * We have the following information: number of active
2279 * target queries, number of active current queries,
2280 * the presence of missing targets at this delegation
2281 * point, and the given query target policy. */
2283 /* Check for the wait condition. If this is true, then
2284 * an action must be taken. */
2285 if(iq->num_target_queries==0 && iq->num_current_queries==0) {
2286 /* If there is nothing to wait for, then we need
2287 * to distinguish between generating (a) new target
2288 * query, or failing. */
2289 if(delegpt_count_missing_targets(iq->dp) > 0) {
2291 verbose(VERB_ALGO, "querying for next "
2293 if(!query_for_targets(qstate, iq, ie, id,
2295 return error_response(qstate, id,
2296 LDNS_RCODE_SERVFAIL);
2299 delegpt_count_missing_targets(iq->dp) == 0){
2300 /* it looked like there were missing
2301 * targets, but they did not turn up.
2302 * Try the bad choices again (if any),
2303 * when we get back here missing==0,
2304 * so this is not a loop. */
2307 iq->num_target_queries += qs;
2308 target_count_increase(iq, qs);
2310 /* Since a target query might have been made, we
2311 * need to check again. */
2312 if(iq->num_target_queries == 0) {
2313 /* if in capsforid fallback, instead of last
2314 * resort, we agree with the current reply
2315 * we have (if any) (our count of addrs bad)*/
2316 if(iq->caps_fallback && iq->caps_reply) {
2317 /* we're done, process the response */
2318 verbose(VERB_ALGO, "0x20 fallback had %d responses, "
2319 "but no more servers except "
2320 "last resort, done.",
2321 (int)iq->caps_server+1);
2322 iq->response = iq->caps_response;
2323 iq->caps_fallback = 0;
2324 iter_dec_attempts(iq->dp, 3); /* space for fallback */
2325 iq->num_current_queries++; /* RespState decrements it*/
2326 iq->referral_count++; /* make sure we don't loop */
2328 iq->state = QUERY_RESP_STATE;
2331 return processLastResort(qstate, iq, ie, id);
2335 /* otherwise, we have no current targets, so submerge
2336 * until one of the target or direct queries return. */
2337 if(iq->num_target_queries>0 && iq->num_current_queries>0) {
2338 verbose(VERB_ALGO, "no current targets -- waiting "
2339 "for %d targets to resolve or %d outstanding"
2340 " queries to respond", iq->num_target_queries,
2341 iq->num_current_queries);
2342 qstate->ext_state[id] = module_wait_reply;
2343 } else if(iq->num_target_queries>0) {
2344 verbose(VERB_ALGO, "no current targets -- waiting "
2345 "for %d targets to resolve.",
2346 iq->num_target_queries);
2347 qstate->ext_state[id] = module_wait_subquery;
2349 verbose(VERB_ALGO, "no current targets -- waiting "
2350 "for %d outstanding queries to respond.",
2351 iq->num_current_queries);
2352 qstate->ext_state[id] = module_wait_reply;
2357 /* if not forwarding, check ratelimits per delegationpoint name */
2358 if(!(iq->chase_flags & BIT_RD) && !iq->ratelimit_ok) {
2359 if(!infra_ratelimit_inc(qstate->env->infra_cache, iq->dp->name,
2360 iq->dp->namelen, *qstate->env->now)) {
2361 lock_basic_lock(&ie->queries_ratelimit_lock);
2362 ie->num_queries_ratelimited++;
2363 lock_basic_unlock(&ie->queries_ratelimit_lock);
2364 verbose(VERB_ALGO, "query exceeded ratelimits");
2365 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2369 /* We have a valid target. */
2370 if(verbosity >= VERB_QUERY) {
2371 log_query_info(VERB_QUERY, "sending query:", &iq->qinfo_out);
2372 log_name_addr(VERB_QUERY, "sending to target:", iq->dp->name,
2373 &target->addr, target->addrlen);
2374 verbose(VERB_ALGO, "dnssec status: %s%s",
2375 iq->dnssec_expected?"expected": "not expected",
2376 iq->dnssec_lame_query?" but lame_query anyway": "");
2378 fptr_ok(fptr_whitelist_modenv_send_query(qstate->env->send_query));
2379 outq = (*qstate->env->send_query)(&iq->qinfo_out,
2380 iq->chase_flags | (iq->chase_to_rd?BIT_RD:0),
2381 /* unset CD if to forwarder(RD set) and not dnssec retry
2382 * (blacklist nonempty) and no trust-anchors are configured
2383 * above the qname or on the first attempt when dnssec is on */
2384 EDNS_DO| ((iq->chase_to_rd||(iq->chase_flags&BIT_RD)!=0)&&
2385 !qstate->blacklist&&(!iter_qname_indicates_dnssec(qstate->env,
2386 &iq->qinfo_out)||target->attempts==1)?0:BIT_CD),
2387 iq->dnssec_expected, iq->caps_fallback || is_caps_whitelisted(
2388 ie, iq), &target->addr, target->addrlen,
2389 iq->dp->name, iq->dp->namelen,
2390 (iq->dp->ssl_upstream || qstate->env->cfg->ssl_upstream),
2391 target->tls_auth_name, qstate);
2393 log_addr(VERB_DETAIL, "error sending query to auth server",
2394 &target->addr, target->addrlen);
2395 if(!(iq->chase_flags & BIT_RD) && !iq->ratelimit_ok)
2396 infra_ratelimit_dec(qstate->env->infra_cache, iq->dp->name,
2397 iq->dp->namelen, *qstate->env->now);
2398 if(qstate->env->cfg->qname_minimisation)
2399 iq->minimisation_state = SKIP_MINIMISE_STATE;
2400 return next_state(iq, QUERYTARGETS_STATE);
2402 outbound_list_insert(&iq->outlist, outq);
2403 iq->num_current_queries++;
2405 qstate->ext_state[id] = module_wait_reply;
2410 /** find NS rrset in given list */
2411 static struct ub_packed_rrset_key*
2412 find_NS(struct reply_info* rep, size_t from, size_t to)
2415 for(i=from; i<to; i++) {
2416 if(ntohs(rep->rrsets[i]->rk.type) == LDNS_RR_TYPE_NS)
2417 return rep->rrsets[i];
2424 * Process the query response. All queries end up at this state first. This
2425 * process generally consists of analyzing the response and routing the
2426 * event to the next state (either bouncing it back to a request state, or
2427 * terminating the processing for this event).
2429 * @param qstate: query state.
2430 * @param iq: iterator query state.
2431 * @param id: module id.
2432 * @return true if the event requires more immediate processing, false if
2433 * not. This is generally only true when forwarding the request to
2434 * the final state (i.e., on answer).
2437 processQueryResponse(struct module_qstate* qstate, struct iter_qstate* iq,
2441 enum response_type type;
2442 iq->num_current_queries--;
2444 if(!inplace_cb_query_response_call(qstate->env, qstate, iq->response))
2445 log_err("unable to call query_response callback");
2447 if(iq->response == NULL) {
2448 /* Don't increment qname when QNAME minimisation is enabled */
2449 if(qstate->env->cfg->qname_minimisation) {
2450 iq->minimise_timeout_count++;
2451 iq->minimisation_state = SKIP_MINIMISE_STATE;
2453 iq->chase_to_rd = 0;
2454 iq->dnssec_lame_query = 0;
2455 verbose(VERB_ALGO, "query response was timeout");
2456 return next_state(iq, QUERYTARGETS_STATE);
2458 type = response_type_from_server(
2459 (int)((iq->chase_flags&BIT_RD) || iq->chase_to_rd),
2460 iq->response, &iq->qinfo_out, iq->dp);
2461 iq->chase_to_rd = 0;
2462 if(type == RESPONSE_TYPE_REFERRAL && (iq->chase_flags&BIT_RD) &&
2463 !iq->auth_zone_response) {
2464 /* When forwarding (RD bit is set), we handle referrals
2465 * differently. No queries should be sent elsewhere */
2466 type = RESPONSE_TYPE_ANSWER;
2468 if(!qstate->env->cfg->disable_dnssec_lame_check && iq->dnssec_expected
2469 && !iq->dnssec_lame_query &&
2470 !(iq->chase_flags&BIT_RD)
2471 && iq->sent_count < DNSSEC_LAME_DETECT_COUNT
2472 && type != RESPONSE_TYPE_LAME
2473 && type != RESPONSE_TYPE_REC_LAME
2474 && type != RESPONSE_TYPE_THROWAWAY
2475 && type != RESPONSE_TYPE_UNTYPED) {
2476 /* a possible answer, see if it is missing DNSSEC */
2477 /* but not when forwarding, so we dont mark fwder lame */
2478 if(!iter_msg_has_dnssec(iq->response)) {
2479 /* Mark this address as dnsseclame in this dp,
2480 * because that will make serverselection disprefer
2481 * it, but also, once it is the only final option,
2482 * use dnssec-lame-bypass if it needs to query there.*/
2484 struct delegpt_addr* a = delegpt_find_addr(
2485 iq->dp, &qstate->reply->addr,
2486 qstate->reply->addrlen);
2487 if(a) a->dnsseclame = 1;
2489 /* test the answer is from the zone we expected,
2490 * otherwise, (due to parent,child on same server), we
2491 * might mark the server,zone lame inappropriately */
2492 if(!iter_msg_from_zone(iq->response, iq->dp, type,
2494 qstate->reply = NULL;
2495 type = RESPONSE_TYPE_LAME;
2498 } else iq->dnssec_lame_query = 0;
2499 /* see if referral brings us close to the target */
2500 if(type == RESPONSE_TYPE_REFERRAL) {
2501 struct ub_packed_rrset_key* ns = find_NS(
2502 iq->response->rep, iq->response->rep->an_numrrsets,
2503 iq->response->rep->an_numrrsets
2504 + iq->response->rep->ns_numrrsets);
2505 if(!ns) ns = find_NS(iq->response->rep, 0,
2506 iq->response->rep->an_numrrsets);
2507 if(!ns || !dname_strict_subdomain_c(ns->rk.dname, iq->dp->name)
2508 || !dname_subdomain_c(iq->qchase.qname, ns->rk.dname)){
2509 verbose(VERB_ALGO, "bad referral, throwaway");
2510 type = RESPONSE_TYPE_THROWAWAY;
2512 iter_scrub_ds(iq->response, ns, iq->dp->name);
2513 } else iter_scrub_ds(iq->response, NULL, NULL);
2514 if(type == RESPONSE_TYPE_THROWAWAY &&
2515 FLAGS_GET_RCODE(iq->response->rep->flags) == LDNS_RCODE_YXDOMAIN) {
2516 /* YXDOMAIN is a permanent error, no need to retry */
2517 type = RESPONSE_TYPE_ANSWER;
2519 if(type == RESPONSE_TYPE_CNAME && iq->response->rep->an_numrrsets >= 1
2520 && ntohs(iq->response->rep->rrsets[0]->rk.type) == LDNS_RR_TYPE_DNAME) {
2521 uint8_t* sname = NULL;
2522 size_t snamelen = 0;
2523 get_cname_target(iq->response->rep->rrsets[0], &sname,
2525 if(snamelen && dname_subdomain_c(sname, iq->response->rep->rrsets[0]->rk.dname)) {
2526 /* DNAME to a subdomain loop; do not recurse */
2527 type = RESPONSE_TYPE_ANSWER;
2531 /* handle each of the type cases */
2532 if(type == RESPONSE_TYPE_ANSWER) {
2533 /* ANSWER type responses terminate the query algorithm,
2534 * so they sent on their */
2535 if(verbosity >= VERB_DETAIL) {
2536 verbose(VERB_DETAIL, "query response was %s",
2537 FLAGS_GET_RCODE(iq->response->rep->flags)
2538 ==LDNS_RCODE_NXDOMAIN?"NXDOMAIN ANSWER":
2539 (iq->response->rep->an_numrrsets?"ANSWER":
2542 /* if qtype is DS, check we have the right level of answer,
2543 * like grandchild answer but we need the middle, reject it */
2544 if(iq->qchase.qtype == LDNS_RR_TYPE_DS && !iq->dsns_point
2545 && !(iq->chase_flags&BIT_RD)
2546 && iter_ds_toolow(iq->response, iq->dp)
2547 && iter_dp_cangodown(&iq->qchase, iq->dp)) {
2548 /* close down outstanding requests to be discarded */
2549 outbound_list_clear(&iq->outlist);
2550 iq->num_current_queries = 0;
2551 fptr_ok(fptr_whitelist_modenv_detach_subs(
2552 qstate->env->detach_subs));
2553 (*qstate->env->detach_subs)(qstate);
2554 iq->num_target_queries = 0;
2555 return processDSNSFind(qstate, iq, id);
2557 if(!qstate->no_cache_store)
2558 iter_dns_store(qstate->env, &iq->response->qinfo,
2559 iq->response->rep, 0, qstate->prefetch_leeway,
2560 iq->dp&&iq->dp->has_parent_side_NS,
2561 qstate->region, qstate->query_flags);
2562 /* close down outstanding requests to be discarded */
2563 outbound_list_clear(&iq->outlist);
2564 iq->num_current_queries = 0;
2565 fptr_ok(fptr_whitelist_modenv_detach_subs(
2566 qstate->env->detach_subs));
2567 (*qstate->env->detach_subs)(qstate);
2568 iq->num_target_queries = 0;
2570 sock_list_insert(&qstate->reply_origin,
2571 &qstate->reply->addr, qstate->reply->addrlen,
2573 if(iq->minimisation_state != DONOT_MINIMISE_STATE) {
2574 if(FLAGS_GET_RCODE(iq->response->rep->flags) !=
2575 LDNS_RCODE_NOERROR) {
2576 if(qstate->env->cfg->qname_minimisation_strict)
2577 return final_state(iq);
2578 /* Best effort qname-minimisation.
2579 * Stop minimising and send full query when
2580 * RCODE is not NOERROR. */
2581 iq->minimisation_state = DONOT_MINIMISE_STATE;
2583 if(FLAGS_GET_RCODE(iq->response->rep->flags) ==
2584 LDNS_RCODE_NXDOMAIN) {
2585 /* Stop resolving when NXDOMAIN is DNSSEC
2586 * signed. Based on assumption that nameservers
2587 * serving signed zones do not return NXDOMAIN
2588 * for empty-non-terminals. */
2589 if(iq->dnssec_expected)
2590 return final_state(iq);
2591 /* Make subrequest to validate intermediate
2592 * NXDOMAIN if harden-below-nxdomain is
2594 if(qstate->env->cfg->harden_below_nxdomain) {
2595 struct module_qstate* subq = NULL;
2596 log_query_info(VERB_QUERY,
2597 "schedule NXDOMAIN validation:",
2598 &iq->response->qinfo);
2599 if(!generate_sub_request(
2600 iq->response->qinfo.qname,
2601 iq->response->qinfo.qname_len,
2602 iq->response->qinfo.qtype,
2603 iq->response->qinfo.qclass,
2606 FINISHED_STATE, &subq, 1))
2608 "could not validate NXDOMAIN "
2612 return next_state(iq, QUERYTARGETS_STATE);
2614 return final_state(iq);
2615 } else if(type == RESPONSE_TYPE_REFERRAL) {
2616 /* REFERRAL type responses get a reset of the
2617 * delegation point, and back to the QUERYTARGETS_STATE. */
2618 verbose(VERB_DETAIL, "query response was REFERRAL");
2620 if(!(iq->chase_flags & BIT_RD) && !iq->ratelimit_ok) {
2621 /* we have a referral, no ratelimit, we can send
2622 * our queries to the given name */
2623 infra_ratelimit_dec(qstate->env->infra_cache,
2624 iq->dp->name, iq->dp->namelen,
2628 /* if hardened, only store referral if we asked for it */
2629 if(!qstate->no_cache_store &&
2630 (!qstate->env->cfg->harden_referral_path ||
2631 ( qstate->qinfo.qtype == LDNS_RR_TYPE_NS
2632 && (qstate->query_flags&BIT_RD)
2633 && !(qstate->query_flags&BIT_CD)
2634 /* we know that all other NS rrsets are scrubbed
2635 * away, thus on referral only one is left.
2636 * see if that equals the query name... */
2637 && ( /* auth section, but sometimes in answer section*/
2638 reply_find_rrset_section_ns(iq->response->rep,
2639 iq->qchase.qname, iq->qchase.qname_len,
2640 LDNS_RR_TYPE_NS, iq->qchase.qclass)
2641 || reply_find_rrset_section_an(iq->response->rep,
2642 iq->qchase.qname, iq->qchase.qname_len,
2643 LDNS_RR_TYPE_NS, iq->qchase.qclass)
2646 /* Store the referral under the current query */
2647 /* no prefetch-leeway, since its not the answer */
2648 iter_dns_store(qstate->env, &iq->response->qinfo,
2649 iq->response->rep, 1, 0, 0, NULL, 0);
2650 if(iq->store_parent_NS)
2651 iter_store_parentside_NS(qstate->env,
2653 if(qstate->env->neg_cache)
2654 val_neg_addreferral(qstate->env->neg_cache,
2655 iq->response->rep, iq->dp->name);
2657 /* store parent-side-in-zone-glue, if directly queried for */
2658 if(!qstate->no_cache_store && iq->query_for_pside_glue
2659 && !iq->pside_glue) {
2660 iq->pside_glue = reply_find_rrset(iq->response->rep,
2661 iq->qchase.qname, iq->qchase.qname_len,
2662 iq->qchase.qtype, iq->qchase.qclass);
2663 if(iq->pside_glue) {
2664 log_rrset_key(VERB_ALGO, "found parent-side "
2665 "glue", iq->pside_glue);
2666 iter_store_parentside_rrset(qstate->env,
2671 /* Reset the event state, setting the current delegation
2672 * point to the referral. */
2673 iq->deleg_msg = iq->response;
2674 iq->dp = delegpt_from_message(iq->response, qstate->region);
2675 if (qstate->env->cfg->qname_minimisation)
2676 iq->minimisation_state = INIT_MINIMISE_STATE;
2678 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2679 if(!cache_fill_missing(qstate->env, iq->qchase.qclass,
2680 qstate->region, iq->dp))
2681 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2682 if(iq->store_parent_NS && query_dname_compare(iq->dp->name,
2683 iq->store_parent_NS->name) == 0)
2684 iter_merge_retry_counts(iq->dp, iq->store_parent_NS);
2685 delegpt_log(VERB_ALGO, iq->dp);
2686 /* Count this as a referral. */
2687 iq->referral_count++;
2689 /* see if the next dp is a trust anchor, or a DS was sent
2690 * along, indicating dnssec is expected for next zone */
2691 iq->dnssec_expected = iter_indicates_dnssec(qstate->env,
2692 iq->dp, iq->response, iq->qchase.qclass);
2693 /* if dnssec, validating then also fetch the key for the DS */
2694 if(iq->dnssec_expected && qstate->env->cfg->prefetch_key &&
2695 !(qstate->query_flags&BIT_CD))
2696 generate_dnskey_prefetch(qstate, iq, id);
2698 /* spawn off NS and addr to auth servers for the NS we just
2699 * got in the referral. This gets authoritative answer
2700 * (answer section trust level) rrset.
2701 * right after, we detach the subs, answer goes to cache. */
2702 if(qstate->env->cfg->harden_referral_path)
2703 generate_ns_check(qstate, iq, id);
2705 /* stop current outstanding queries.
2706 * FIXME: should the outstanding queries be waited for and
2707 * handled? Say by a subquery that inherits the outbound_entry.
2709 outbound_list_clear(&iq->outlist);
2710 iq->num_current_queries = 0;
2711 fptr_ok(fptr_whitelist_modenv_detach_subs(
2712 qstate->env->detach_subs));
2713 (*qstate->env->detach_subs)(qstate);
2714 iq->num_target_queries = 0;
2715 verbose(VERB_ALGO, "cleared outbound list for next round");
2716 return next_state(iq, QUERYTARGETS_STATE);
2717 } else if(type == RESPONSE_TYPE_CNAME) {
2718 uint8_t* sname = NULL;
2719 size_t snamelen = 0;
2720 /* CNAME type responses get a query restart (i.e., get a
2721 * reset of the query state and go back to INIT_REQUEST_STATE).
2723 verbose(VERB_DETAIL, "query response was CNAME");
2724 if(verbosity >= VERB_ALGO)
2725 log_dns_msg("cname msg", &iq->response->qinfo,
2727 /* if qtype is DS, check we have the right level of answer,
2728 * like grandchild answer but we need the middle, reject it */
2729 if(iq->qchase.qtype == LDNS_RR_TYPE_DS && !iq->dsns_point
2730 && !(iq->chase_flags&BIT_RD)
2731 && iter_ds_toolow(iq->response, iq->dp)
2732 && iter_dp_cangodown(&iq->qchase, iq->dp)) {
2733 outbound_list_clear(&iq->outlist);
2734 iq->num_current_queries = 0;
2735 fptr_ok(fptr_whitelist_modenv_detach_subs(
2736 qstate->env->detach_subs));
2737 (*qstate->env->detach_subs)(qstate);
2738 iq->num_target_queries = 0;
2739 return processDSNSFind(qstate, iq, id);
2741 /* Process the CNAME response. */
2742 if(!handle_cname_response(qstate, iq, iq->response,
2744 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2745 /* cache the CNAME response under the current query */
2746 /* NOTE : set referral=1, so that rrsets get stored but not
2747 * the partial query answer (CNAME only). */
2748 /* prefetchleeway applied because this updates answer parts */
2749 if(!qstate->no_cache_store)
2750 iter_dns_store(qstate->env, &iq->response->qinfo,
2751 iq->response->rep, 1, qstate->prefetch_leeway,
2752 iq->dp&&iq->dp->has_parent_side_NS, NULL,
2753 qstate->query_flags);
2754 /* set the current request's qname to the new value. */
2755 iq->qchase.qname = sname;
2756 iq->qchase.qname_len = snamelen;
2757 if (qstate->env->cfg->qname_minimisation)
2758 iq->minimisation_state = INIT_MINIMISE_STATE;
2759 /* Clear the query state, since this is a query restart. */
2760 iq->deleg_msg = NULL;
2762 iq->dsns_point = NULL;
2763 iq->auth_zone_response = 0;
2764 /* Note the query restart. */
2765 iq->query_restart_count++;
2768 /* stop current outstanding queries.
2769 * FIXME: should the outstanding queries be waited for and
2770 * handled? Say by a subquery that inherits the outbound_entry.
2772 outbound_list_clear(&iq->outlist);
2773 iq->num_current_queries = 0;
2774 fptr_ok(fptr_whitelist_modenv_detach_subs(
2775 qstate->env->detach_subs));
2776 (*qstate->env->detach_subs)(qstate);
2777 iq->num_target_queries = 0;
2779 sock_list_insert(&qstate->reply_origin,
2780 &qstate->reply->addr, qstate->reply->addrlen,
2782 verbose(VERB_ALGO, "cleared outbound list for query restart");
2783 /* go to INIT_REQUEST_STATE for new qname. */
2784 return next_state(iq, INIT_REQUEST_STATE);
2785 } else if(type == RESPONSE_TYPE_LAME) {
2786 /* Cache the LAMEness. */
2787 verbose(VERB_DETAIL, "query response was %sLAME",
2788 dnsseclame?"DNSSEC ":"");
2789 if(!dname_subdomain_c(iq->qchase.qname, iq->dp->name)) {
2790 log_err("mark lame: mismatch in qname and dpname");
2791 /* throwaway this reply below */
2792 } else if(qstate->reply) {
2793 /* need addr for lameness cache, but we may have
2794 * gotten this from cache, so test to be sure */
2795 if(!infra_set_lame(qstate->env->infra_cache,
2796 &qstate->reply->addr, qstate->reply->addrlen,
2797 iq->dp->name, iq->dp->namelen,
2798 *qstate->env->now, dnsseclame, 0,
2800 log_err("mark host lame: out of memory");
2802 } else if(type == RESPONSE_TYPE_REC_LAME) {
2803 /* Cache the LAMEness. */
2804 verbose(VERB_DETAIL, "query response REC_LAME: "
2805 "recursive but not authoritative server");
2806 if(!dname_subdomain_c(iq->qchase.qname, iq->dp->name)) {
2807 log_err("mark rec_lame: mismatch in qname and dpname");
2808 /* throwaway this reply below */
2809 } else if(qstate->reply) {
2810 /* need addr for lameness cache, but we may have
2811 * gotten this from cache, so test to be sure */
2812 verbose(VERB_DETAIL, "mark as REC_LAME");
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, 0, 1, iq->qchase.qtype))
2817 log_err("mark host lame: out of memory");
2819 } else if(type == RESPONSE_TYPE_THROWAWAY) {
2820 /* LAME and THROWAWAY responses are handled the same way.
2821 * In this case, the event is just sent directly back to
2822 * the QUERYTARGETS_STATE without resetting anything,
2823 * because, clearly, the next target must be tried. */
2824 verbose(VERB_DETAIL, "query response was THROWAWAY");
2826 log_warn("A query response came back with an unknown type: %d",
2830 /* LAME, THROWAWAY and "unknown" all end up here.
2831 * Recycle to the QUERYTARGETS state to hopefully try a
2832 * different target. */
2833 if (qstate->env->cfg->qname_minimisation &&
2834 !qstate->env->cfg->qname_minimisation_strict)
2835 iq->minimisation_state = DONOT_MINIMISE_STATE;
2836 if(iq->auth_zone_response) {
2837 /* can we fallback? */
2838 iq->auth_zone_response = 0;
2839 if(!auth_zones_can_fallback(qstate->env->auth_zones,
2840 iq->dp->name, iq->dp->namelen, qstate->qinfo.qclass)) {
2841 verbose(VERB_ALGO, "auth zone response bad, and no"
2842 " fallback possible, servfail");
2843 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2845 verbose(VERB_ALGO, "auth zone response was bad, "
2846 "fallback enabled");
2847 iq->auth_zone_avoid = 1;
2848 if(iq->dp->auth_dp) {
2849 /* we are using a dp for the auth zone, with no
2850 * nameservers, get one first */
2852 return next_state(iq, INIT_REQUEST_STATE);
2855 return next_state(iq, QUERYTARGETS_STATE);
2859 * Return priming query results to interested super querystates.
2861 * Sets the delegation point and delegation message (not nonRD queries).
2862 * This is a callback from walk_supers.
2864 * @param qstate: priming query state that finished.
2865 * @param id: module id.
2866 * @param forq: the qstate for which priming has been done.
2869 prime_supers(struct module_qstate* qstate, int id, struct module_qstate* forq)
2871 struct iter_qstate* foriq = (struct iter_qstate*)forq->minfo[id];
2872 struct delegpt* dp = NULL;
2874 log_assert(qstate->is_priming || foriq->wait_priming_stub);
2875 log_assert(qstate->return_rcode == LDNS_RCODE_NOERROR);
2876 /* Convert our response to a delegation point */
2877 dp = delegpt_from_message(qstate->return_msg, forq->region);
2879 /* if there is no convertable delegation point, then
2880 * the ANSWER type was (presumably) a negative answer. */
2881 verbose(VERB_ALGO, "prime response was not a positive "
2884 foriq->state = QUERYTARGETS_STATE;
2888 log_query_info(VERB_DETAIL, "priming successful for", &qstate->qinfo);
2889 delegpt_log(VERB_ALGO, dp);
2891 foriq->deleg_msg = dns_copy_msg(qstate->return_msg, forq->region);
2892 if(!foriq->deleg_msg) {
2893 log_err("copy prime response: out of memory");
2895 foriq->state = QUERYTARGETS_STATE;
2899 /* root priming responses go to init stage 2, priming stub
2900 * responses to to stage 3. */
2901 if(foriq->wait_priming_stub) {
2902 foriq->state = INIT_REQUEST_3_STATE;
2903 foriq->wait_priming_stub = 0;
2904 } else foriq->state = INIT_REQUEST_2_STATE;
2905 /* because we are finished, the parent will be reactivated */
2909 * This handles the response to a priming query. This is used to handle both
2910 * root and stub priming responses. This is basically the equivalent of the
2911 * QUERY_RESP_STATE, but will not handle CNAME responses and will treat
2912 * REFERRALs as ANSWERS. It will also update and reactivate the originating
2915 * @param qstate: query state.
2916 * @param id: module id.
2917 * @return true if the event needs more immediate processing, false if not.
2918 * This state always returns false.
2921 processPrimeResponse(struct module_qstate* qstate, int id)
2923 struct iter_qstate* iq = (struct iter_qstate*)qstate->minfo[id];
2924 enum response_type type;
2925 iq->response->rep->flags &= ~(BIT_RD|BIT_RA); /* ignore rec-lame */
2926 type = response_type_from_server(
2927 (int)((iq->chase_flags&BIT_RD) || iq->chase_to_rd),
2928 iq->response, &iq->qchase, iq->dp);
2929 if(type == RESPONSE_TYPE_ANSWER) {
2930 qstate->return_rcode = LDNS_RCODE_NOERROR;
2931 qstate->return_msg = iq->response;
2933 qstate->return_rcode = LDNS_RCODE_SERVFAIL;
2934 qstate->return_msg = NULL;
2937 /* validate the root or stub after priming (if enabled).
2938 * This is the same query as the prime query, but with validation.
2939 * Now that we are primed, the additional queries that validation
2940 * may need can be resolved, such as DLV. */
2941 if(qstate->env->cfg->harden_referral_path) {
2942 struct module_qstate* subq = NULL;
2943 log_nametypeclass(VERB_ALGO, "schedule prime validation",
2944 qstate->qinfo.qname, qstate->qinfo.qtype,
2945 qstate->qinfo.qclass);
2946 if(!generate_sub_request(qstate->qinfo.qname,
2947 qstate->qinfo.qname_len, qstate->qinfo.qtype,
2948 qstate->qinfo.qclass, qstate, id, iq,
2949 INIT_REQUEST_STATE, FINISHED_STATE, &subq, 1)) {
2950 verbose(VERB_ALGO, "could not generate prime check");
2952 generate_a_aaaa_check(qstate, iq, id);
2955 /* This event is finished. */
2956 qstate->ext_state[id] = module_finished;
2961 * Do final processing on responses to target queries. Events reach this
2962 * state after the iterative resolution algorithm terminates. This state is
2963 * responsible for reactivating the original event, and housekeeping related
2964 * to received target responses (caching, updating the current delegation
2966 * Callback from walk_supers for every super state that is interested in
2967 * the results from this query.
2969 * @param qstate: query state.
2970 * @param id: module id.
2971 * @param forq: super query state.
2974 processTargetResponse(struct module_qstate* qstate, int id,
2975 struct module_qstate* forq)
2977 struct iter_qstate* iq = (struct iter_qstate*)qstate->minfo[id];
2978 struct iter_qstate* foriq = (struct iter_qstate*)forq->minfo[id];
2979 struct ub_packed_rrset_key* rrset;
2980 struct delegpt_ns* dpns;
2981 log_assert(qstate->return_rcode == LDNS_RCODE_NOERROR);
2983 foriq->state = QUERYTARGETS_STATE;
2984 log_query_info(VERB_ALGO, "processTargetResponse", &qstate->qinfo);
2985 log_query_info(VERB_ALGO, "processTargetResponse super", &forq->qinfo);
2987 /* Tell the originating event that this target query has finished
2988 * (regardless if it succeeded or not). */
2989 foriq->num_target_queries--;
2991 /* check to see if parent event is still interested (in orig name). */
2993 verbose(VERB_ALGO, "subq: parent not interested, was reset");
2994 return; /* not interested anymore */
2996 dpns = delegpt_find_ns(foriq->dp, qstate->qinfo.qname,
2997 qstate->qinfo.qname_len);
2999 /* If not interested, just stop processing this event */
3000 verbose(VERB_ALGO, "subq: parent not interested anymore");
3001 /* could be because parent was jostled out of the cache,
3002 and a new identical query arrived, that does not want it*/
3006 /* if iq->query_for_pside_glue then add the pside_glue (marked lame) */
3007 if(iq->pside_glue) {
3008 /* if the pside_glue is NULL, then it could not be found,
3009 * the done_pside is already set when created and a cache
3010 * entry created in processFinished so nothing to do here */
3011 log_rrset_key(VERB_ALGO, "add parentside glue to dp",
3013 if(!delegpt_add_rrset(foriq->dp, forq->region,
3015 log_err("out of memory adding pside glue");
3018 /* This response is relevant to the current query, so we
3019 * add (attempt to add, anyway) this target(s) and reactivate
3020 * the original event.
3021 * NOTE: we could only look for the AnswerRRset if the
3022 * response type was ANSWER. */
3023 rrset = reply_find_answer_rrset(&iq->qchase, qstate->return_msg->rep);
3025 /* if CNAMEs have been followed - add new NS to delegpt. */
3026 /* BTW. RFC 1918 says NS should not have got CNAMEs. Robust. */
3027 if(!delegpt_find_ns(foriq->dp, rrset->rk.dname,
3028 rrset->rk.dname_len)) {
3029 /* if dpns->lame then set newcname ns lame too */
3030 if(!delegpt_add_ns(foriq->dp, forq->region,
3031 rrset->rk.dname, dpns->lame))
3032 log_err("out of memory adding cnamed-ns");
3034 /* if dpns->lame then set the address(es) lame too */
3035 if(!delegpt_add_rrset(foriq->dp, forq->region, rrset,
3037 log_err("out of memory adding targets");
3038 verbose(VERB_ALGO, "added target response");
3039 delegpt_log(VERB_ALGO, foriq->dp);
3041 verbose(VERB_ALGO, "iterator TargetResponse failed");
3042 dpns->resolved = 1; /* fail the target */
3047 * Process response for DS NS Find queries, that attempt to find the delegation
3048 * point where we ask the DS query from.
3050 * @param qstate: query state.
3051 * @param id: module id.
3052 * @param forq: super query state.
3055 processDSNSResponse(struct module_qstate* qstate, int id,
3056 struct module_qstate* forq)
3058 struct iter_qstate* foriq = (struct iter_qstate*)forq->minfo[id];
3060 /* if the finished (iq->response) query has no NS set: continue
3061 * up to look for the right dp; nothing to change, do DPNSstate */
3062 if(qstate->return_rcode != LDNS_RCODE_NOERROR)
3063 return; /* seek further */
3064 /* find the NS RRset (without allowing CNAMEs) */
3065 if(!reply_find_rrset(qstate->return_msg->rep, qstate->qinfo.qname,
3066 qstate->qinfo.qname_len, LDNS_RR_TYPE_NS,
3067 qstate->qinfo.qclass)){
3068 return; /* seek further */
3071 /* else, store as DP and continue at querytargets */
3072 foriq->state = QUERYTARGETS_STATE;
3073 foriq->dp = delegpt_from_message(qstate->return_msg, forq->region);
3075 log_err("out of memory in dsns dp alloc");
3076 return; /* dp==NULL in QUERYTARGETS makes SERVFAIL */
3078 /* success, go query the querytargets in the new dp (and go down) */
3082 * Process response for qclass=ANY queries for a particular class.
3083 * Append to result or error-exit.
3085 * @param qstate: query state.
3086 * @param id: module id.
3087 * @param forq: super query state.
3090 processClassResponse(struct module_qstate* qstate, int id,
3091 struct module_qstate* forq)
3093 struct iter_qstate* foriq = (struct iter_qstate*)forq->minfo[id];
3094 struct dns_msg* from = qstate->return_msg;
3095 log_query_info(VERB_ALGO, "processClassResponse", &qstate->qinfo);
3096 log_query_info(VERB_ALGO, "processClassResponse super", &forq->qinfo);
3097 if(qstate->return_rcode != LDNS_RCODE_NOERROR) {
3098 /* cause servfail for qclass ANY query */
3099 foriq->response = NULL;
3100 foriq->state = FINISHED_STATE;
3104 if(!foriq->response) {
3105 /* allocate the response: copy RCODE, sec_state */
3106 foriq->response = dns_copy_msg(from, forq->region);
3107 if(!foriq->response) {
3108 log_err("malloc failed for qclass ANY response");
3109 foriq->state = FINISHED_STATE;
3112 foriq->response->qinfo.qclass = forq->qinfo.qclass;
3113 /* qclass ANY does not receive the AA flag on replies */
3114 foriq->response->rep->authoritative = 0;
3116 struct dns_msg* to = foriq->response;
3117 /* add _from_ this response _to_ existing collection */
3118 /* if there are records, copy RCODE */
3119 /* lower sec_state if this message is lower */
3120 if(from->rep->rrset_count != 0) {
3121 size_t n = from->rep->rrset_count+to->rep->rrset_count;
3122 struct ub_packed_rrset_key** dest, **d;
3123 /* copy appropriate rcode */
3124 to->rep->flags = from->rep->flags;
3126 if(from->rep->rrset_count > RR_COUNT_MAX ||
3127 to->rep->rrset_count > RR_COUNT_MAX) {
3128 log_err("malloc failed (too many rrsets) in collect ANY");
3129 foriq->state = FINISHED_STATE;
3130 return; /* integer overflow protection */
3132 dest = regional_alloc(forq->region, sizeof(dest[0])*n);
3134 log_err("malloc failed in collect ANY");
3135 foriq->state = FINISHED_STATE;
3140 memcpy(dest, to->rep->rrsets, to->rep->an_numrrsets
3142 dest += to->rep->an_numrrsets;
3143 memcpy(dest, from->rep->rrsets, from->rep->an_numrrsets
3145 dest += from->rep->an_numrrsets;
3147 memcpy(dest, to->rep->rrsets+to->rep->an_numrrsets,
3148 to->rep->ns_numrrsets * sizeof(dest[0]));
3149 dest += to->rep->ns_numrrsets;
3150 memcpy(dest, from->rep->rrsets+from->rep->an_numrrsets,
3151 from->rep->ns_numrrsets * sizeof(dest[0]));
3152 dest += from->rep->ns_numrrsets;
3154 memcpy(dest, to->rep->rrsets+to->rep->an_numrrsets+
3155 to->rep->ns_numrrsets,
3156 to->rep->ar_numrrsets * sizeof(dest[0]));
3157 dest += to->rep->ar_numrrsets;
3158 memcpy(dest, from->rep->rrsets+from->rep->an_numrrsets+
3159 from->rep->ns_numrrsets,
3160 from->rep->ar_numrrsets * sizeof(dest[0]));
3162 to->rep->rrsets = d;
3163 to->rep->an_numrrsets += from->rep->an_numrrsets;
3164 to->rep->ns_numrrsets += from->rep->ns_numrrsets;
3165 to->rep->ar_numrrsets += from->rep->ar_numrrsets;
3166 to->rep->rrset_count = n;
3168 if(from->rep->security < to->rep->security) /* lowest sec */
3169 to->rep->security = from->rep->security;
3170 if(from->rep->qdcount != 0) /* insert qd if appropriate */
3171 to->rep->qdcount = from->rep->qdcount;
3172 if(from->rep->ttl < to->rep->ttl) /* use smallest TTL */
3173 to->rep->ttl = from->rep->ttl;
3174 if(from->rep->prefetch_ttl < to->rep->prefetch_ttl)
3175 to->rep->prefetch_ttl = from->rep->prefetch_ttl;
3178 foriq->num_current_queries --;
3179 if(foriq->num_current_queries == 0)
3180 foriq->state = FINISHED_STATE;
3184 * Collect class ANY responses and make them into one response. This
3185 * state is started and it creates queries for all classes (that have
3186 * root hints). The answers are then collected.
3188 * @param qstate: query state.
3189 * @param id: module id.
3190 * @return true if the event needs more immediate processing, false if not.
3193 processCollectClass(struct module_qstate* qstate, int id)
3195 struct iter_qstate* iq = (struct iter_qstate*)qstate->minfo[id];
3196 struct module_qstate* subq;
3197 /* If qchase.qclass == 0 then send out queries for all classes.
3198 * Otherwise, do nothing (wait for all answers to arrive and the
3199 * processClassResponse to put them together, and that moves us
3200 * towards the Finished state when done. */
3201 if(iq->qchase.qclass == 0) {
3203 iq->qchase.qclass = LDNS_RR_CLASS_ANY;
3204 while(iter_get_next_root(qstate->env->hints,
3205 qstate->env->fwds, &c)) {
3206 /* generate query for this class */
3207 log_nametypeclass(VERB_ALGO, "spawn collect query",
3208 qstate->qinfo.qname, qstate->qinfo.qtype, c);
3209 if(!generate_sub_request(qstate->qinfo.qname,
3210 qstate->qinfo.qname_len, qstate->qinfo.qtype,
3211 c, qstate, id, iq, INIT_REQUEST_STATE,
3212 FINISHED_STATE, &subq,
3213 (int)!(qstate->query_flags&BIT_CD))) {
3214 return error_response(qstate, id,
3215 LDNS_RCODE_SERVFAIL);
3217 /* ignore subq, no special init required */
3218 iq->num_current_queries ++;
3223 /* if no roots are configured at all, return */
3224 if(iq->num_current_queries == 0) {
3225 verbose(VERB_ALGO, "No root hints or fwds, giving up "
3227 return error_response(qstate, id, LDNS_RCODE_REFUSED);
3229 /* return false, wait for queries to return */
3231 /* if woke up here because of an answer, wait for more answers */
3236 * This handles the final state for first-tier responses (i.e., responses to
3237 * externally generated queries).
3239 * @param qstate: query state.
3240 * @param iq: iterator query state.
3241 * @param id: module id.
3242 * @return true if the event needs more processing, false if not. Since this
3243 * is the final state for an event, it always returns false.
3246 processFinished(struct module_qstate* qstate, struct iter_qstate* iq,
3249 log_query_info(VERB_QUERY, "finishing processing for",
3252 /* store negative cache element for parent side glue. */
3253 if(!qstate->no_cache_store && iq->query_for_pside_glue
3255 iter_store_parentside_neg(qstate->env, &qstate->qinfo,
3256 iq->deleg_msg?iq->deleg_msg->rep:
3257 (iq->response?iq->response->rep:NULL));
3259 verbose(VERB_ALGO, "No response is set, servfail");
3260 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3263 /* Make sure that the RA flag is set (since the presence of
3264 * this module means that recursion is available) */
3265 iq->response->rep->flags |= BIT_RA;
3267 /* Clear the AA flag */
3268 /* FIXME: does this action go here or in some other module? */
3269 iq->response->rep->flags &= ~BIT_AA;
3271 /* make sure QR flag is on */
3272 iq->response->rep->flags |= BIT_QR;
3274 /* we have finished processing this query */
3275 qstate->ext_state[id] = module_finished;
3277 /* TODO: we are using a private TTL, trim the response. */
3278 /* if (mPrivateTTL > 0){IterUtils.setPrivateTTL(resp, mPrivateTTL); } */
3280 /* prepend any items we have accumulated */
3281 if(iq->an_prepend_list || iq->ns_prepend_list) {
3282 if(!iter_prepend(iq, iq->response, qstate->region)) {
3283 log_err("prepend rrsets: out of memory");
3284 return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3286 /* reset the query name back */
3287 iq->response->qinfo = qstate->qinfo;
3288 /* the security state depends on the combination */
3289 iq->response->rep->security = sec_status_unchecked;
3290 /* store message with the finished prepended items,
3291 * but only if we did recursion. The nonrecursion referral
3292 * from cache does not need to be stored in the msg cache. */
3293 if(!qstate->no_cache_store && qstate->query_flags&BIT_RD) {
3294 iter_dns_store(qstate->env, &qstate->qinfo,
3295 iq->response->rep, 0, qstate->prefetch_leeway,
3296 iq->dp&&iq->dp->has_parent_side_NS,
3297 qstate->region, qstate->query_flags);
3300 qstate->return_rcode = LDNS_RCODE_NOERROR;
3301 qstate->return_msg = iq->response;
3306 * Return priming query results to interested super querystates.
3308 * Sets the delegation point and delegation message (not nonRD queries).
3309 * This is a callback from walk_supers.
3311 * @param qstate: query state that finished.
3312 * @param id: module id.
3313 * @param super: the qstate to inform.
3316 iter_inform_super(struct module_qstate* qstate, int id,
3317 struct module_qstate* super)
3319 if(!qstate->is_priming && super->qinfo.qclass == LDNS_RR_CLASS_ANY)
3320 processClassResponse(qstate, id, super);
3321 else if(super->qinfo.qtype == LDNS_RR_TYPE_DS && ((struct iter_qstate*)
3322 super->minfo[id])->state == DSNS_FIND_STATE)
3323 processDSNSResponse(qstate, id, super);
3324 else if(qstate->return_rcode != LDNS_RCODE_NOERROR)
3325 error_supers(qstate, id, super);
3326 else if(qstate->is_priming)
3327 prime_supers(qstate, id, super);
3328 else processTargetResponse(qstate, id, super);
3332 * Handle iterator state.
3333 * Handle events. This is the real processing loop for events, responsible
3334 * for moving events through the various states. If a processing method
3335 * returns true, then it will be advanced to the next state. If false, then
3336 * processing will stop.
3338 * @param qstate: query state.
3339 * @param ie: iterator shared global environment.
3340 * @param iq: iterator query state.
3341 * @param id: module id.
3344 iter_handle(struct module_qstate* qstate, struct iter_qstate* iq,
3345 struct iter_env* ie, int id)
3349 verbose(VERB_ALGO, "iter_handle processing q with state %s",
3350 iter_state_to_string(iq->state));
3352 case INIT_REQUEST_STATE:
3353 cont = processInitRequest(qstate, iq, ie, id);
3355 case INIT_REQUEST_2_STATE:
3356 cont = processInitRequest2(qstate, iq, id);
3358 case INIT_REQUEST_3_STATE:
3359 cont = processInitRequest3(qstate, iq, id);
3361 case QUERYTARGETS_STATE:
3362 cont = processQueryTargets(qstate, iq, ie, id);
3364 case QUERY_RESP_STATE:
3365 cont = processQueryResponse(qstate, iq, id);
3367 case PRIME_RESP_STATE:
3368 cont = processPrimeResponse(qstate, id);
3370 case COLLECT_CLASS_STATE:
3371 cont = processCollectClass(qstate, id);
3373 case DSNS_FIND_STATE:
3374 cont = processDSNSFind(qstate, iq, id);
3376 case FINISHED_STATE:
3377 cont = processFinished(qstate, iq, id);
3380 log_warn("iterator: invalid state: %d",
3389 * This is the primary entry point for processing request events. Note that
3390 * this method should only be used by external modules.
3391 * @param qstate: query state.
3392 * @param ie: iterator shared global environment.
3393 * @param iq: iterator query state.
3394 * @param id: module id.
3397 process_request(struct module_qstate* qstate, struct iter_qstate* iq,
3398 struct iter_env* ie, int id)
3400 /* external requests start in the INIT state, and finish using the
3401 * FINISHED state. */
3402 iq->state = INIT_REQUEST_STATE;
3403 iq->final_state = FINISHED_STATE;
3404 verbose(VERB_ALGO, "process_request: new external request event");
3405 iter_handle(qstate, iq, ie, id);
3408 /** process authoritative server reply */
3410 process_response(struct module_qstate* qstate, struct iter_qstate* iq,
3411 struct iter_env* ie, int id, struct outbound_entry* outbound,
3412 enum module_ev event)
3414 struct msg_parse* prs;
3415 struct edns_data edns;
3418 verbose(VERB_ALGO, "process_response: new external response event");
3419 iq->response = NULL;
3420 iq->state = QUERY_RESP_STATE;
3421 if(event == module_event_noreply || event == module_event_error) {
3422 if(event == module_event_noreply && iq->sent_count >= 3 &&
3423 qstate->env->cfg->use_caps_bits_for_id &&
3424 !iq->caps_fallback) {
3425 /* start fallback */
3426 iq->caps_fallback = 1;
3427 iq->caps_server = 0;
3428 iq->caps_reply = NULL;
3429 iq->caps_response = NULL;
3430 iq->state = QUERYTARGETS_STATE;
3431 iq->num_current_queries--;
3432 /* need fresh attempts for the 0x20 fallback, if
3433 * that was the cause for the failure */
3434 iter_dec_attempts(iq->dp, 3);
3435 verbose(VERB_DETAIL, "Capsforid: timeouts, starting fallback");
3440 if( (event != module_event_reply && event != module_event_capsfail)
3441 || !qstate->reply) {
3442 log_err("Bad event combined with response");
3443 outbound_list_remove(&iq->outlist, outbound);
3444 (void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3449 prs = (struct msg_parse*)regional_alloc(qstate->env->scratch,
3450 sizeof(struct msg_parse));
3452 log_err("out of memory on incoming message");
3453 /* like packet got dropped */
3456 memset(prs, 0, sizeof(*prs));
3457 memset(&edns, 0, sizeof(edns));
3458 pkt = qstate->reply->c->buffer;
3459 sldns_buffer_set_position(pkt, 0);
3460 if(parse_packet(pkt, prs, qstate->env->scratch) != LDNS_RCODE_NOERROR) {
3461 verbose(VERB_ALGO, "parse error on reply packet");
3464 /* edns is not examined, but removed from message to help cache */
3465 if(parse_extract_edns(prs, &edns, qstate->env->scratch) !=
3469 /* Copy the edns options we may got from the back end */
3471 qstate->edns_opts_back_in = edns_opt_copy_region(edns.opt_list,
3473 if(!qstate->edns_opts_back_in) {
3474 log_err("out of memory on incoming message");
3475 /* like packet got dropped */
3478 if(!inplace_cb_edns_back_parsed_call(qstate->env, qstate)) {
3479 log_err("unable to call edns_back_parsed callback");
3484 /* remove CD-bit, we asked for in case we handle validation ourself */
3485 prs->flags &= ~BIT_CD;
3487 /* normalize and sanitize: easy to delete items from linked lists */
3488 if(!scrub_message(pkt, prs, &iq->qinfo_out, iq->dp->name,
3489 qstate->env->scratch, qstate->env, ie)) {
3490 /* if 0x20 enabled, start fallback, but we have no message */
3491 if(event == module_event_capsfail && !iq->caps_fallback) {
3492 iq->caps_fallback = 1;
3493 iq->caps_server = 0;
3494 iq->caps_reply = NULL;
3495 iq->caps_response = NULL;
3496 iq->state = QUERYTARGETS_STATE;
3497 iq->num_current_queries--;
3498 verbose(VERB_DETAIL, "Capsforid: scrub failed, starting fallback with no response");
3503 /* allocate response dns_msg in region */
3504 iq->response = dns_alloc_msg(pkt, prs, qstate->region);
3507 log_query_info(VERB_DETAIL, "response for", &qstate->qinfo);
3508 log_name_addr(VERB_DETAIL, "reply from", iq->dp->name,
3509 &qstate->reply->addr, qstate->reply->addrlen);
3510 if(verbosity >= VERB_ALGO)
3511 log_dns_msg("incoming scrubbed packet:", &iq->response->qinfo,
3514 if(event == module_event_capsfail || iq->caps_fallback) {
3515 /* for fallback we care about main answer, not additionals */
3516 /* removing that makes comparison more likely to succeed */
3517 caps_strip_reply(iq->response->rep);
3518 if(!iq->caps_fallback) {
3519 /* start fallback */
3520 iq->caps_fallback = 1;
3521 iq->caps_server = 0;
3522 iq->caps_reply = iq->response->rep;
3523 iq->caps_response = iq->response;
3524 iq->state = QUERYTARGETS_STATE;
3525 iq->num_current_queries--;
3526 verbose(VERB_DETAIL, "Capsforid: starting fallback");
3529 /* check if reply is the same, otherwise, fail */
3530 if(!iq->caps_reply) {
3531 iq->caps_reply = iq->response->rep;
3532 iq->caps_response = iq->response;
3533 iq->caps_server = -1; /*become zero at ++,
3534 so that we start the full set of trials */
3535 } else if(caps_failed_rcode(iq->caps_reply) &&
3536 !caps_failed_rcode(iq->response->rep)) {
3537 /* prefer to upgrade to non-SERVFAIL */
3538 iq->caps_reply = iq->response->rep;
3539 iq->caps_response = iq->response;
3540 } else if(!caps_failed_rcode(iq->caps_reply) &&
3541 caps_failed_rcode(iq->response->rep)) {
3542 /* if we have non-SERVFAIL as answer then
3543 * we can ignore SERVFAILs for the equality
3545 /* no instructions here, skip other else */
3546 } else if(caps_failed_rcode(iq->caps_reply) &&
3547 caps_failed_rcode(iq->response->rep)) {
3548 /* failure is same as other failure in fallbk*/
3549 /* no instructions here, skip other else */
3550 } else if(!reply_equal(iq->response->rep, iq->caps_reply,
3551 qstate->env->scratch)) {
3552 verbose(VERB_DETAIL, "Capsforid fallback: "
3553 "getting different replies, failed");
3554 outbound_list_remove(&iq->outlist, outbound);
3555 (void)error_response(qstate, id,
3556 LDNS_RCODE_SERVFAIL);
3559 /* continue the fallback procedure at next server */
3561 iq->state = QUERYTARGETS_STATE;
3562 iq->num_current_queries--;
3563 verbose(VERB_DETAIL, "Capsforid: reply is equal. "
3564 "go to next fallback");
3568 iq->caps_fallback = 0; /* if we were in fallback, 0x20 is OK now */
3571 outbound_list_remove(&iq->outlist, outbound);
3572 iter_handle(qstate, iq, ie, id);
3576 iter_operate(struct module_qstate* qstate, enum module_ev event, int id,
3577 struct outbound_entry* outbound)
3579 struct iter_env* ie = (struct iter_env*)qstate->env->modinfo[id];
3580 struct iter_qstate* iq = (struct iter_qstate*)qstate->minfo[id];
3581 verbose(VERB_QUERY, "iterator[module %d] operate: extstate:%s event:%s",
3582 id, strextstate(qstate->ext_state[id]), strmodulevent(event));
3583 if(iq) log_query_info(VERB_QUERY, "iterator operate: query",
3585 if(iq && qstate->qinfo.qname != iq->qchase.qname)
3586 log_query_info(VERB_QUERY, "iterator operate: chased to",
3589 /* perform iterator state machine */
3590 if((event == module_event_new || event == module_event_pass) &&
3592 if(!iter_new(qstate, id)) {
3593 (void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3596 iq = (struct iter_qstate*)qstate->minfo[id];
3597 process_request(qstate, iq, ie, id);
3600 if(iq && event == module_event_pass) {
3601 iter_handle(qstate, iq, ie, id);
3604 if(iq && outbound) {
3605 process_response(qstate, iq, ie, id, outbound, event);
3608 if(event == module_event_error) {
3609 verbose(VERB_ALGO, "got called with event error, giving up");
3610 (void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3614 log_err("bad event for iterator");
3615 (void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3619 iter_clear(struct module_qstate* qstate, int id)
3621 struct iter_qstate* iq;
3624 iq = (struct iter_qstate*)qstate->minfo[id];
3626 outbound_list_clear(&iq->outlist);
3627 if(iq->target_count && --iq->target_count[0] == 0)
3628 free(iq->target_count);
3629 iq->num_current_queries = 0;
3631 qstate->minfo[id] = NULL;
3635 iter_get_mem(struct module_env* env, int id)
3637 struct iter_env* ie = (struct iter_env*)env->modinfo[id];
3640 return sizeof(*ie) + sizeof(int)*((size_t)ie->max_dependency_depth+1)
3641 + donotq_get_mem(ie->donotq) + priv_get_mem(ie->priv);
3645 * The iterator function block
3647 static struct module_func_block iter_block = {
3649 &iter_init, &iter_deinit, &iter_operate, &iter_inform_super,
3650 &iter_clear, &iter_get_mem
3653 struct module_func_block*
3654 iter_get_funcblock(void)
3660 iter_state_to_string(enum iter_state state)
3664 case INIT_REQUEST_STATE :
3665 return "INIT REQUEST STATE";
3666 case INIT_REQUEST_2_STATE :
3667 return "INIT REQUEST STATE (stage 2)";
3668 case INIT_REQUEST_3_STATE:
3669 return "INIT REQUEST STATE (stage 3)";
3670 case QUERYTARGETS_STATE :
3671 return "QUERY TARGETS STATE";
3672 case PRIME_RESP_STATE :
3673 return "PRIME RESPONSE STATE";
3674 case COLLECT_CLASS_STATE :
3675 return "COLLECT CLASS STATE";
3676 case DSNS_FIND_STATE :
3677 return "DSNS FIND STATE";
3678 case QUERY_RESP_STATE :
3679 return "QUERY RESPONSE STATE";
3680 case FINISHED_STATE :
3681 return "FINISHED RESPONSE STATE";
3683 return "UNKNOWN ITER STATE";
3688 iter_state_is_responsestate(enum iter_state s)
3691 case INIT_REQUEST_STATE :
3692 case INIT_REQUEST_2_STATE :
3693 case INIT_REQUEST_3_STATE :
3694 case QUERYTARGETS_STATE :
3695 case COLLECT_CLASS_STATE :