iterator.c revision 291767
1/*
2 * iterator/iterator.c - iterative resolver DNS query response module
3 *
4 * Copyright (c) 2007, NLnet Labs. All rights reserved.
5 *
6 * This software is open source.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 *
12 * Redistributions of source code must retain the above copyright notice,
13 * this list of conditions and the following disclaimer.
14 *
15 * Redistributions in binary form must reproduce the above copyright notice,
16 * this list of conditions and the following disclaimer in the documentation
17 * and/or other materials provided with the distribution.
18 *
19 * Neither the name of the NLNET LABS nor the names of its contributors may
20 * be used to endorse or promote products derived from this software without
21 * specific prior written permission.
22 *
23 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
26 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
27 * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
28 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
29 * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
30 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
31 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
32 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 */
35
36/**
37 * \file
38 *
39 * This file contains a module that performs recusive iterative DNS query
40 * processing.
41 */
42
43#include "config.h"
44#include "iterator/iterator.h"
45#include "iterator/iter_utils.h"
46#include "iterator/iter_hints.h"
47#include "iterator/iter_fwd.h"
48#include "iterator/iter_donotq.h"
49#include "iterator/iter_delegpt.h"
50#include "iterator/iter_resptype.h"
51#include "iterator/iter_scrub.h"
52#include "iterator/iter_priv.h"
53#include "validator/val_neg.h"
54#include "services/cache/dns.h"
55#include "services/cache/infra.h"
56#include "util/module.h"
57#include "util/netevent.h"
58#include "util/net_help.h"
59#include "util/regional.h"
60#include "util/data/dname.h"
61#include "util/data/msgencode.h"
62#include "util/fptr_wlist.h"
63#include "util/config_file.h"
64#include "util/random.h"
65#include "sldns/rrdef.h"
66#include "sldns/wire2str.h"
67#include "sldns/parseutil.h"
68#include "sldns/sbuffer.h"
69
70int
71iter_init(struct module_env* env, int id)
72{
73	struct iter_env* iter_env = (struct iter_env*)calloc(1,
74		sizeof(struct iter_env));
75	if(!iter_env) {
76		log_err("malloc failure");
77		return 0;
78	}
79	env->modinfo[id] = (void*)iter_env;
80	if(!iter_apply_cfg(iter_env, env->cfg)) {
81		log_err("iterator: could not apply configuration settings.");
82		return 0;
83	}
84	return 1;
85}
86
87/** delete caps_whitelist element */
88static void
89caps_free(struct rbnode_t* n, void* ATTR_UNUSED(d))
90{
91	if(n) {
92		free(((struct name_tree_node*)n)->name);
93		free(n);
94	}
95}
96
97void
98iter_deinit(struct module_env* env, int id)
99{
100	struct iter_env* iter_env;
101	if(!env || !env->modinfo[id])
102		return;
103	iter_env = (struct iter_env*)env->modinfo[id];
104	free(iter_env->target_fetch_policy);
105	priv_delete(iter_env->priv);
106	donotq_delete(iter_env->donotq);
107	if(iter_env->caps_white) {
108		traverse_postorder(iter_env->caps_white, caps_free, NULL);
109		free(iter_env->caps_white);
110	}
111	free(iter_env);
112	env->modinfo[id] = NULL;
113}
114
115/** new query for iterator */
116static int
117iter_new(struct module_qstate* qstate, int id)
118{
119	struct iter_qstate* iq = (struct iter_qstate*)regional_alloc(
120		qstate->region, sizeof(struct iter_qstate));
121	qstate->minfo[id] = iq;
122	if(!iq)
123		return 0;
124	memset(iq, 0, sizeof(*iq));
125	iq->state = INIT_REQUEST_STATE;
126	iq->final_state = FINISHED_STATE;
127	iq->an_prepend_list = NULL;
128	iq->an_prepend_last = NULL;
129	iq->ns_prepend_list = NULL;
130	iq->ns_prepend_last = NULL;
131	iq->dp = NULL;
132	iq->depth = 0;
133	iq->num_target_queries = 0;
134	iq->num_current_queries = 0;
135	iq->query_restart_count = 0;
136	iq->referral_count = 0;
137	iq->sent_count = 0;
138	iq->ratelimit_ok = 0;
139	iq->target_count = NULL;
140	iq->wait_priming_stub = 0;
141	iq->refetch_glue = 0;
142	iq->dnssec_expected = 0;
143	iq->dnssec_lame_query = 0;
144	iq->chase_flags = qstate->query_flags;
145	/* Start with the (current) qname. */
146	iq->qchase = qstate->qinfo;
147	outbound_list_init(&iq->outlist);
148	return 1;
149}
150
151/**
152 * Transition to the next state. This can be used to advance a currently
153 * processing event. It cannot be used to reactivate a forEvent.
154 *
155 * @param iq: iterator query state
156 * @param nextstate The state to transition to.
157 * @return true. This is so this can be called as the return value for the
158 *         actual process*State() methods. (Transitioning to the next state
159 *         implies further processing).
160 */
161static int
162next_state(struct iter_qstate* iq, enum iter_state nextstate)
163{
164	/* If transitioning to a "response" state, make sure that there is a
165	 * response */
166	if(iter_state_is_responsestate(nextstate)) {
167		if(iq->response == NULL) {
168			log_err("transitioning to response state sans "
169				"response.");
170		}
171	}
172	iq->state = nextstate;
173	return 1;
174}
175
176/**
177 * Transition an event to its final state. Final states always either return
178 * a result up the module chain, or reactivate a dependent event. Which
179 * final state to transtion to is set in the module state for the event when
180 * it was created, and depends on the original purpose of the event.
181 *
182 * The response is stored in the qstate->buf buffer.
183 *
184 * @param iq: iterator query state
185 * @return false. This is so this method can be used as the return value for
186 *         the processState methods. (Transitioning to the final state
187 */
188static int
189final_state(struct iter_qstate* iq)
190{
191	return next_state(iq, iq->final_state);
192}
193
194/**
195 * Callback routine to handle errors in parent query states
196 * @param qstate: query state that failed.
197 * @param id: module id.
198 * @param super: super state.
199 */
200static void
201error_supers(struct module_qstate* qstate, int id, struct module_qstate* super)
202{
203	struct iter_qstate* super_iq = (struct iter_qstate*)super->minfo[id];
204
205	if(qstate->qinfo.qtype == LDNS_RR_TYPE_A ||
206		qstate->qinfo.qtype == LDNS_RR_TYPE_AAAA) {
207		/* mark address as failed. */
208		struct delegpt_ns* dpns = NULL;
209		if(super_iq->dp)
210			dpns = delegpt_find_ns(super_iq->dp,
211				qstate->qinfo.qname, qstate->qinfo.qname_len);
212		if(!dpns) {
213			/* not interested */
214			verbose(VERB_ALGO, "subq error, but not interested");
215			log_query_info(VERB_ALGO, "superq", &super->qinfo);
216			if(super_iq->dp)
217				delegpt_log(VERB_ALGO, super_iq->dp);
218			log_assert(0);
219			return;
220		} else {
221			/* see if the failure did get (parent-lame) info */
222			if(!cache_fill_missing(super->env,
223				super_iq->qchase.qclass, super->region,
224				super_iq->dp))
225				log_err("out of memory adding missing");
226		}
227		dpns->resolved = 1; /* mark as failed */
228		super_iq->num_target_queries--;
229	}
230	if(qstate->qinfo.qtype == LDNS_RR_TYPE_NS) {
231		/* prime failed to get delegation */
232		super_iq->dp = NULL;
233	}
234	/* evaluate targets again */
235	super_iq->state = QUERYTARGETS_STATE;
236	/* super becomes runnable, and will process this change */
237}
238
239/**
240 * Return an error to the client
241 * @param qstate: our query state
242 * @param id: module id
243 * @param rcode: error code (DNS errcode).
244 * @return: 0 for use by caller, to make notation easy, like:
245 * 	return error_response(..).
246 */
247static int
248error_response(struct module_qstate* qstate, int id, int rcode)
249{
250	verbose(VERB_QUERY, "return error response %s",
251		sldns_lookup_by_id(sldns_rcodes, rcode)?
252		sldns_lookup_by_id(sldns_rcodes, rcode)->name:"??");
253	qstate->return_rcode = rcode;
254	qstate->return_msg = NULL;
255	qstate->ext_state[id] = module_finished;
256	return 0;
257}
258
259/**
260 * Return an error to the client and cache the error code in the
261 * message cache (so per qname, qtype, qclass).
262 * @param qstate: our query state
263 * @param id: module id
264 * @param rcode: error code (DNS errcode).
265 * @return: 0 for use by caller, to make notation easy, like:
266 * 	return error_response(..).
267 */
268static int
269error_response_cache(struct module_qstate* qstate, int id, int rcode)
270{
271	/* store in cache */
272	struct reply_info err;
273	if(qstate->prefetch_leeway > NORR_TTL) {
274		verbose(VERB_ALGO, "error response for prefetch in cache");
275		/* attempt to adjust the cache entry prefetch */
276		if(dns_cache_prefetch_adjust(qstate->env, &qstate->qinfo,
277			NORR_TTL, qstate->query_flags))
278			return error_response(qstate, id, rcode);
279		/* if that fails (not in cache), fall through to store err */
280	}
281	memset(&err, 0, sizeof(err));
282	err.flags = (uint16_t)(BIT_QR | BIT_RA);
283	FLAGS_SET_RCODE(err.flags, rcode);
284	err.qdcount = 1;
285	err.ttl = NORR_TTL;
286	err.prefetch_ttl = PREFETCH_TTL_CALC(err.ttl);
287	/* do not waste time trying to validate this servfail */
288	err.security = sec_status_indeterminate;
289	verbose(VERB_ALGO, "store error response in message cache");
290	iter_dns_store(qstate->env, &qstate->qinfo, &err, 0, 0, 0, NULL,
291		qstate->query_flags);
292	return error_response(qstate, id, rcode);
293}
294
295/** check if prepend item is duplicate item */
296static int
297prepend_is_duplicate(struct ub_packed_rrset_key** sets, size_t to,
298	struct ub_packed_rrset_key* dup)
299{
300	size_t i;
301	for(i=0; i<to; i++) {
302		if(sets[i]->rk.type == dup->rk.type &&
303			sets[i]->rk.rrset_class == dup->rk.rrset_class &&
304			sets[i]->rk.dname_len == dup->rk.dname_len &&
305			query_dname_compare(sets[i]->rk.dname, dup->rk.dname)
306			== 0)
307			return 1;
308	}
309	return 0;
310}
311
312/** prepend the prepend list in the answer and authority section of dns_msg */
313static int
314iter_prepend(struct iter_qstate* iq, struct dns_msg* msg,
315	struct regional* region)
316{
317	struct iter_prep_list* p;
318	struct ub_packed_rrset_key** sets;
319	size_t num_an = 0, num_ns = 0;;
320	for(p = iq->an_prepend_list; p; p = p->next)
321		num_an++;
322	for(p = iq->ns_prepend_list; p; p = p->next)
323		num_ns++;
324	if(num_an + num_ns == 0)
325		return 1;
326	verbose(VERB_ALGO, "prepending %d rrsets", (int)num_an + (int)num_ns);
327	if(num_an > RR_COUNT_MAX || num_ns > RR_COUNT_MAX ||
328		msg->rep->rrset_count > RR_COUNT_MAX) return 0; /* overflow */
329	sets = regional_alloc(region, (num_an+num_ns+msg->rep->rrset_count) *
330		sizeof(struct ub_packed_rrset_key*));
331	if(!sets)
332		return 0;
333	/* ANSWER section */
334	num_an = 0;
335	for(p = iq->an_prepend_list; p; p = p->next) {
336		sets[num_an++] = p->rrset;
337	}
338	memcpy(sets+num_an, msg->rep->rrsets, msg->rep->an_numrrsets *
339		sizeof(struct ub_packed_rrset_key*));
340	/* AUTH section */
341	num_ns = 0;
342	for(p = iq->ns_prepend_list; p; p = p->next) {
343		if(prepend_is_duplicate(sets+msg->rep->an_numrrsets+num_an,
344			num_ns, p->rrset) || prepend_is_duplicate(
345			msg->rep->rrsets+msg->rep->an_numrrsets,
346			msg->rep->ns_numrrsets, p->rrset))
347			continue;
348		sets[msg->rep->an_numrrsets + num_an + num_ns++] = p->rrset;
349	}
350	memcpy(sets + num_an + msg->rep->an_numrrsets + num_ns,
351		msg->rep->rrsets + msg->rep->an_numrrsets,
352		(msg->rep->ns_numrrsets + msg->rep->ar_numrrsets) *
353		sizeof(struct ub_packed_rrset_key*));
354
355	/* NXDOMAIN rcode can stay if we prepended DNAME/CNAMEs, because
356	 * this is what recursors should give. */
357	msg->rep->rrset_count += num_an + num_ns;
358	msg->rep->an_numrrsets += num_an;
359	msg->rep->ns_numrrsets += num_ns;
360	msg->rep->rrsets = sets;
361	return 1;
362}
363
364/**
365 * Add rrset to ANSWER prepend list
366 * @param qstate: query state.
367 * @param iq: iterator query state.
368 * @param rrset: rrset to add.
369 * @return false on failure (malloc).
370 */
371static int
372iter_add_prepend_answer(struct module_qstate* qstate, struct iter_qstate* iq,
373	struct ub_packed_rrset_key* rrset)
374{
375	struct iter_prep_list* p = (struct iter_prep_list*)regional_alloc(
376		qstate->region, sizeof(struct iter_prep_list));
377	if(!p)
378		return 0;
379	p->rrset = rrset;
380	p->next = NULL;
381	/* add at end */
382	if(iq->an_prepend_last)
383		iq->an_prepend_last->next = p;
384	else	iq->an_prepend_list = p;
385	iq->an_prepend_last = p;
386	return 1;
387}
388
389/**
390 * Add rrset to AUTHORITY prepend list
391 * @param qstate: query state.
392 * @param iq: iterator query state.
393 * @param rrset: rrset to add.
394 * @return false on failure (malloc).
395 */
396static int
397iter_add_prepend_auth(struct module_qstate* qstate, struct iter_qstate* iq,
398	struct ub_packed_rrset_key* rrset)
399{
400	struct iter_prep_list* p = (struct iter_prep_list*)regional_alloc(
401		qstate->region, sizeof(struct iter_prep_list));
402	if(!p)
403		return 0;
404	p->rrset = rrset;
405	p->next = NULL;
406	/* add at end */
407	if(iq->ns_prepend_last)
408		iq->ns_prepend_last->next = p;
409	else	iq->ns_prepend_list = p;
410	iq->ns_prepend_last = p;
411	return 1;
412}
413
414/**
415 * Given a CNAME response (defined as a response containing a CNAME or DNAME
416 * that does not answer the request), process the response, modifying the
417 * state as necessary. This follows the CNAME/DNAME chain and returns the
418 * final query name.
419 *
420 * sets the new query name, after following the CNAME/DNAME chain.
421 * @param qstate: query state.
422 * @param iq: iterator query state.
423 * @param msg: the response.
424 * @param mname: returned target new query name.
425 * @param mname_len: length of mname.
426 * @return false on (malloc) error.
427 */
428static int
429handle_cname_response(struct module_qstate* qstate, struct iter_qstate* iq,
430        struct dns_msg* msg, uint8_t** mname, size_t* mname_len)
431{
432	size_t i;
433	/* Start with the (current) qname. */
434	*mname = iq->qchase.qname;
435	*mname_len = iq->qchase.qname_len;
436
437	/* Iterate over the ANSWER rrsets in order, looking for CNAMEs and
438	 * DNAMES. */
439	for(i=0; i<msg->rep->an_numrrsets; i++) {
440		struct ub_packed_rrset_key* r = msg->rep->rrsets[i];
441		/* If there is a (relevant) DNAME, add it to the list.
442		 * We always expect there to be CNAME that was generated
443		 * by this DNAME following, so we don't process the DNAME
444		 * directly.  */
445		if(ntohs(r->rk.type) == LDNS_RR_TYPE_DNAME &&
446			dname_strict_subdomain_c(*mname, r->rk.dname)) {
447			if(!iter_add_prepend_answer(qstate, iq, r))
448				return 0;
449			continue;
450		}
451
452		if(ntohs(r->rk.type) == LDNS_RR_TYPE_CNAME &&
453			query_dname_compare(*mname, r->rk.dname) == 0) {
454			/* Add this relevant CNAME rrset to the prepend list.*/
455			if(!iter_add_prepend_answer(qstate, iq, r))
456				return 0;
457			get_cname_target(r, mname, mname_len);
458		}
459
460		/* Other rrsets in the section are ignored. */
461	}
462	/* add authority rrsets to authority prepend, for wildcarded CNAMEs */
463	for(i=msg->rep->an_numrrsets; i<msg->rep->an_numrrsets +
464		msg->rep->ns_numrrsets; i++) {
465		struct ub_packed_rrset_key* r = msg->rep->rrsets[i];
466		/* only add NSEC/NSEC3, as they may be needed for validation */
467		if(ntohs(r->rk.type) == LDNS_RR_TYPE_NSEC ||
468			ntohs(r->rk.type) == LDNS_RR_TYPE_NSEC3) {
469			if(!iter_add_prepend_auth(qstate, iq, r))
470				return 0;
471		}
472	}
473	return 1;
474}
475
476/** see if target name is caps-for-id whitelisted */
477static int
478is_caps_whitelisted(struct iter_env* ie, struct iter_qstate* iq)
479{
480	if(!ie->caps_white) return 0; /* no whitelist, or no capsforid */
481	return name_tree_lookup(ie->caps_white, iq->qchase.qname,
482		iq->qchase.qname_len, dname_count_labels(iq->qchase.qname),
483		iq->qchase.qclass) != NULL;
484}
485
486/** create target count structure for this query */
487static void
488target_count_create(struct iter_qstate* iq)
489{
490	if(!iq->target_count) {
491		iq->target_count = (int*)calloc(2, sizeof(int));
492		/* if calloc fails we simply do not track this number */
493		if(iq->target_count)
494			iq->target_count[0] = 1;
495	}
496}
497
498static void
499target_count_increase(struct iter_qstate* iq, int num)
500{
501	target_count_create(iq);
502	if(iq->target_count)
503		iq->target_count[1] += num;
504}
505
506/**
507 * Generate a subrequest.
508 * Generate a local request event. Local events are tied to this module, and
509 * have a correponding (first tier) event that is waiting for this event to
510 * resolve to continue.
511 *
512 * @param qname The query name for this request.
513 * @param qnamelen length of qname
514 * @param qtype The query type for this request.
515 * @param qclass The query class for this request.
516 * @param qstate The event that is generating this event.
517 * @param id: module id.
518 * @param iq: The iterator state that is generating this event.
519 * @param initial_state The initial response state (normally this
520 *          is QUERY_RESP_STATE, unless it is known that the request won't
521 *          need iterative processing
522 * @param finalstate The final state for the response to this request.
523 * @param subq_ret: if newly allocated, the subquerystate, or NULL if it does
524 * 	not need initialisation.
525 * @param v: if true, validation is done on the subquery.
526 * @return false on error (malloc).
527 */
528static int
529generate_sub_request(uint8_t* qname, size_t qnamelen, uint16_t qtype,
530	uint16_t qclass, struct module_qstate* qstate, int id,
531	struct iter_qstate* iq, enum iter_state initial_state,
532	enum iter_state finalstate, struct module_qstate** subq_ret, int v)
533{
534	struct module_qstate* subq = NULL;
535	struct iter_qstate* subiq = NULL;
536	uint16_t qflags = 0; /* OPCODE QUERY, no flags */
537	struct query_info qinf;
538	int prime = (finalstate == PRIME_RESP_STATE)?1:0;
539	int valrec = 0;
540	qinf.qname = qname;
541	qinf.qname_len = qnamelen;
542	qinf.qtype = qtype;
543	qinf.qclass = qclass;
544
545	/* RD should be set only when sending the query back through the INIT
546	 * state. */
547	if(initial_state == INIT_REQUEST_STATE)
548		qflags |= BIT_RD;
549	/* We set the CD flag so we can send this through the "head" of
550	 * the resolution chain, which might have a validator. We are
551	 * uninterested in validating things not on the direct resolution
552	 * path.  */
553	if(!v) {
554		qflags |= BIT_CD;
555		valrec = 1;
556	}
557
558	/* attach subquery, lookup existing or make a new one */
559	fptr_ok(fptr_whitelist_modenv_attach_sub(qstate->env->attach_sub));
560	if(!(*qstate->env->attach_sub)(qstate, &qinf, qflags, prime, valrec,
561		&subq)) {
562		return 0;
563	}
564	*subq_ret = subq;
565	if(subq) {
566		/* initialise the new subquery */
567		subq->curmod = id;
568		subq->ext_state[id] = module_state_initial;
569		subq->minfo[id] = regional_alloc(subq->region,
570			sizeof(struct iter_qstate));
571		if(!subq->minfo[id]) {
572			log_err("init subq: out of memory");
573			fptr_ok(fptr_whitelist_modenv_kill_sub(
574				qstate->env->kill_sub));
575			(*qstate->env->kill_sub)(subq);
576			return 0;
577		}
578		subiq = (struct iter_qstate*)subq->minfo[id];
579		memset(subiq, 0, sizeof(*subiq));
580		subiq->num_target_queries = 0;
581		target_count_create(iq);
582		subiq->target_count = iq->target_count;
583		if(iq->target_count)
584			iq->target_count[0] ++; /* extra reference */
585		subiq->num_current_queries = 0;
586		subiq->depth = iq->depth+1;
587		outbound_list_init(&subiq->outlist);
588		subiq->state = initial_state;
589		subiq->final_state = finalstate;
590		subiq->qchase = subq->qinfo;
591		subiq->chase_flags = subq->query_flags;
592		subiq->refetch_glue = 0;
593	}
594	return 1;
595}
596
597/**
598 * Generate and send a root priming request.
599 * @param qstate: the qtstate that triggered the need to prime.
600 * @param iq: iterator query state.
601 * @param id: module id.
602 * @param qclass: the class to prime.
603 * @return 0 on failure
604 */
605static int
606prime_root(struct module_qstate* qstate, struct iter_qstate* iq, int id,
607	uint16_t qclass)
608{
609	struct delegpt* dp;
610	struct module_qstate* subq;
611	verbose(VERB_DETAIL, "priming . %s NS",
612		sldns_lookup_by_id(sldns_rr_classes, (int)qclass)?
613		sldns_lookup_by_id(sldns_rr_classes, (int)qclass)->name:"??");
614	dp = hints_lookup_root(qstate->env->hints, qclass);
615	if(!dp) {
616		verbose(VERB_ALGO, "Cannot prime due to lack of hints");
617		return 0;
618	}
619	/* Priming requests start at the QUERYTARGETS state, skipping
620	 * the normal INIT state logic (which would cause an infloop). */
621	if(!generate_sub_request((uint8_t*)"\000", 1, LDNS_RR_TYPE_NS,
622		qclass, qstate, id, iq, QUERYTARGETS_STATE, PRIME_RESP_STATE,
623		&subq, 0)) {
624		verbose(VERB_ALGO, "could not prime root");
625		return 0;
626	}
627	if(subq) {
628		struct iter_qstate* subiq =
629			(struct iter_qstate*)subq->minfo[id];
630		/* Set the initial delegation point to the hint.
631		 * copy dp, it is now part of the root prime query.
632		 * dp was part of in the fixed hints structure. */
633		subiq->dp = delegpt_copy(dp, subq->region);
634		if(!subiq->dp) {
635			log_err("out of memory priming root, copydp");
636			fptr_ok(fptr_whitelist_modenv_kill_sub(
637				qstate->env->kill_sub));
638			(*qstate->env->kill_sub)(subq);
639			return 0;
640		}
641		/* there should not be any target queries. */
642		subiq->num_target_queries = 0;
643		subiq->dnssec_expected = iter_indicates_dnssec(
644			qstate->env, subiq->dp, NULL, subq->qinfo.qclass);
645	}
646
647	/* this module stops, our submodule starts, and does the query. */
648	qstate->ext_state[id] = module_wait_subquery;
649	return 1;
650}
651
652/**
653 * Generate and process a stub priming request. This method tests for the
654 * need to prime a stub zone, so it is safe to call for every request.
655 *
656 * @param qstate: the qtstate that triggered the need to prime.
657 * @param iq: iterator query state.
658 * @param id: module id.
659 * @param qname: request name.
660 * @param qclass: request class.
661 * @return true if a priming subrequest was made, false if not. The will only
662 *         issue a priming request if it detects an unprimed stub.
663 *         Uses value of 2 to signal during stub-prime in root-prime situation
664 *         that a noprime-stub is available and resolution can continue.
665 */
666static int
667prime_stub(struct module_qstate* qstate, struct iter_qstate* iq, int id,
668	uint8_t* qname, uint16_t qclass)
669{
670	/* Lookup the stub hint. This will return null if the stub doesn't
671	 * need to be re-primed. */
672	struct iter_hints_stub* stub;
673	struct delegpt* stub_dp;
674	struct module_qstate* subq;
675
676	if(!qname) return 0;
677	stub = hints_lookup_stub(qstate->env->hints, qname, qclass, iq->dp);
678	/* The stub (if there is one) does not need priming. */
679	if(!stub)
680		return 0;
681	stub_dp = stub->dp;
682
683	/* is it a noprime stub (always use) */
684	if(stub->noprime) {
685		int r = 0;
686		if(iq->dp == NULL) r = 2;
687		/* copy the dp out of the fixed hints structure, so that
688		 * it can be changed when servicing this query */
689		iq->dp = delegpt_copy(stub_dp, qstate->region);
690		if(!iq->dp) {
691			log_err("out of memory priming stub");
692			(void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
693			return 1; /* return 1 to make module stop, with error */
694		}
695		log_nametypeclass(VERB_DETAIL, "use stub", stub_dp->name,
696			LDNS_RR_TYPE_NS, qclass);
697		return r;
698	}
699
700	/* Otherwise, we need to (re)prime the stub. */
701	log_nametypeclass(VERB_DETAIL, "priming stub", stub_dp->name,
702		LDNS_RR_TYPE_NS, qclass);
703
704	/* Stub priming events start at the QUERYTARGETS state to avoid the
705	 * redundant INIT state processing. */
706	if(!generate_sub_request(stub_dp->name, stub_dp->namelen,
707		LDNS_RR_TYPE_NS, qclass, qstate, id, iq,
708		QUERYTARGETS_STATE, PRIME_RESP_STATE, &subq, 0)) {
709		verbose(VERB_ALGO, "could not prime stub");
710		(void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
711		return 1; /* return 1 to make module stop, with error */
712	}
713	if(subq) {
714		struct iter_qstate* subiq =
715			(struct iter_qstate*)subq->minfo[id];
716
717		/* Set the initial delegation point to the hint. */
718		/* make copy to avoid use of stub dp by different qs/threads */
719		subiq->dp = delegpt_copy(stub_dp, subq->region);
720		if(!subiq->dp) {
721			log_err("out of memory priming stub, copydp");
722			fptr_ok(fptr_whitelist_modenv_kill_sub(
723				qstate->env->kill_sub));
724			(*qstate->env->kill_sub)(subq);
725			(void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
726			return 1; /* return 1 to make module stop, with error */
727		}
728		/* there should not be any target queries -- although there
729		 * wouldn't be anyway, since stub hints never have
730		 * missing targets. */
731		subiq->num_target_queries = 0;
732		subiq->wait_priming_stub = 1;
733		subiq->dnssec_expected = iter_indicates_dnssec(
734			qstate->env, subiq->dp, NULL, subq->qinfo.qclass);
735	}
736
737	/* this module stops, our submodule starts, and does the query. */
738	qstate->ext_state[id] = module_wait_subquery;
739	return 1;
740}
741
742/**
743 * Generate A and AAAA checks for glue that is in-zone for the referral
744 * we just got to obtain authoritative information on the adresses.
745 *
746 * @param qstate: the qtstate that triggered the need to prime.
747 * @param iq: iterator query state.
748 * @param id: module id.
749 */
750static void
751generate_a_aaaa_check(struct module_qstate* qstate, struct iter_qstate* iq,
752	int id)
753{
754	struct iter_env* ie = (struct iter_env*)qstate->env->modinfo[id];
755	struct module_qstate* subq;
756	size_t i;
757	struct reply_info* rep = iq->response->rep;
758	struct ub_packed_rrset_key* s;
759	log_assert(iq->dp);
760
761	if(iq->depth == ie->max_dependency_depth)
762		return;
763	/* walk through additional, and check if in-zone,
764	 * only relevant A, AAAA are left after scrub anyway */
765	for(i=rep->an_numrrsets+rep->ns_numrrsets; i<rep->rrset_count; i++) {
766		s = rep->rrsets[i];
767		/* check *ALL* addresses that are transmitted in additional*/
768		/* is it an address ? */
769		if( !(ntohs(s->rk.type)==LDNS_RR_TYPE_A ||
770			ntohs(s->rk.type)==LDNS_RR_TYPE_AAAA)) {
771			continue;
772		}
773		/* is this query the same as the A/AAAA check for it */
774		if(qstate->qinfo.qtype == ntohs(s->rk.type) &&
775			qstate->qinfo.qclass == ntohs(s->rk.rrset_class) &&
776			query_dname_compare(qstate->qinfo.qname,
777				s->rk.dname)==0 &&
778			(qstate->query_flags&BIT_RD) &&
779			!(qstate->query_flags&BIT_CD))
780			continue;
781
782		/* generate subrequest for it */
783		log_nametypeclass(VERB_ALGO, "schedule addr fetch",
784			s->rk.dname, ntohs(s->rk.type),
785			ntohs(s->rk.rrset_class));
786		if(!generate_sub_request(s->rk.dname, s->rk.dname_len,
787			ntohs(s->rk.type), ntohs(s->rk.rrset_class),
788			qstate, id, iq,
789			INIT_REQUEST_STATE, FINISHED_STATE, &subq, 1)) {
790			verbose(VERB_ALGO, "could not generate addr check");
791			return;
792		}
793		/* ignore subq - not need for more init */
794	}
795}
796
797/**
798 * Generate a NS check request to obtain authoritative information
799 * on an NS rrset.
800 *
801 * @param qstate: the qtstate that triggered the need to prime.
802 * @param iq: iterator query state.
803 * @param id: module id.
804 */
805static void
806generate_ns_check(struct module_qstate* qstate, struct iter_qstate* iq, int id)
807{
808	struct iter_env* ie = (struct iter_env*)qstate->env->modinfo[id];
809	struct module_qstate* subq;
810	log_assert(iq->dp);
811
812	if(iq->depth == ie->max_dependency_depth)
813		return;
814	/* is this query the same as the nscheck? */
815	if(qstate->qinfo.qtype == LDNS_RR_TYPE_NS &&
816		query_dname_compare(iq->dp->name, qstate->qinfo.qname)==0 &&
817		(qstate->query_flags&BIT_RD) && !(qstate->query_flags&BIT_CD)){
818		/* spawn off A, AAAA queries for in-zone glue to check */
819		generate_a_aaaa_check(qstate, iq, id);
820		return;
821	}
822
823	log_nametypeclass(VERB_ALGO, "schedule ns fetch",
824		iq->dp->name, LDNS_RR_TYPE_NS, iq->qchase.qclass);
825	if(!generate_sub_request(iq->dp->name, iq->dp->namelen,
826		LDNS_RR_TYPE_NS, iq->qchase.qclass, qstate, id, iq,
827		INIT_REQUEST_STATE, FINISHED_STATE, &subq, 1)) {
828		verbose(VERB_ALGO, "could not generate ns check");
829		return;
830	}
831	if(subq) {
832		struct iter_qstate* subiq =
833			(struct iter_qstate*)subq->minfo[id];
834
835		/* make copy to avoid use of stub dp by different qs/threads */
836		/* refetch glue to start higher up the tree */
837		subiq->refetch_glue = 1;
838		subiq->dp = delegpt_copy(iq->dp, subq->region);
839		if(!subiq->dp) {
840			log_err("out of memory generating ns check, copydp");
841			fptr_ok(fptr_whitelist_modenv_kill_sub(
842				qstate->env->kill_sub));
843			(*qstate->env->kill_sub)(subq);
844			return;
845		}
846	}
847}
848
849/**
850 * Generate a DNSKEY prefetch query to get the DNSKEY for the DS record we
851 * just got in a referral (where we have dnssec_expected, thus have trust
852 * anchors above it).  Note that right after calling this routine the
853 * iterator detached subqueries (because of following the referral), and thus
854 * the DNSKEY query becomes detached, its return stored in the cache for
855 * later lookup by the validator.  This cache lookup by the validator avoids
856 * the roundtrip incurred by the DNSKEY query.  The DNSKEY query is now
857 * performed at about the same time the original query is sent to the domain,
858 * thus the two answers are likely to be returned at about the same time,
859 * saving a roundtrip from the validated lookup.
860 *
861 * @param qstate: the qtstate that triggered the need to prime.
862 * @param iq: iterator query state.
863 * @param id: module id.
864 */
865static void
866generate_dnskey_prefetch(struct module_qstate* qstate,
867	struct iter_qstate* iq, int id)
868{
869	struct module_qstate* subq;
870	log_assert(iq->dp);
871
872	/* is this query the same as the prefetch? */
873	if(qstate->qinfo.qtype == LDNS_RR_TYPE_DNSKEY &&
874		query_dname_compare(iq->dp->name, qstate->qinfo.qname)==0 &&
875		(qstate->query_flags&BIT_RD) && !(qstate->query_flags&BIT_CD)){
876		return;
877	}
878
879	/* if the DNSKEY is in the cache this lookup will stop quickly */
880	log_nametypeclass(VERB_ALGO, "schedule dnskey prefetch",
881		iq->dp->name, LDNS_RR_TYPE_DNSKEY, iq->qchase.qclass);
882	if(!generate_sub_request(iq->dp->name, iq->dp->namelen,
883		LDNS_RR_TYPE_DNSKEY, iq->qchase.qclass, qstate, id, iq,
884		INIT_REQUEST_STATE, FINISHED_STATE, &subq, 0)) {
885		/* we'll be slower, but it'll work */
886		verbose(VERB_ALGO, "could not generate dnskey prefetch");
887		return;
888	}
889	if(subq) {
890		struct iter_qstate* subiq =
891			(struct iter_qstate*)subq->minfo[id];
892		/* this qstate has the right delegation for the dnskey lookup*/
893		/* make copy to avoid use of stub dp by different qs/threads */
894		subiq->dp = delegpt_copy(iq->dp, subq->region);
895		/* if !subiq->dp, it'll start from the cache, no problem */
896	}
897}
898
899/**
900 * See if the query needs forwarding.
901 *
902 * @param qstate: query state.
903 * @param iq: iterator query state.
904 * @return true if the request is forwarded, false if not.
905 * 	If returns true but, iq->dp is NULL then a malloc failure occurred.
906 */
907static int
908forward_request(struct module_qstate* qstate, struct iter_qstate* iq)
909{
910	struct delegpt* dp;
911	uint8_t* delname = iq->qchase.qname;
912	size_t delnamelen = iq->qchase.qname_len;
913	if(iq->refetch_glue) {
914		delname = iq->dp->name;
915		delnamelen = iq->dp->namelen;
916	}
917	/* strip one label off of DS query to lookup higher for it */
918	if( (iq->qchase.qtype == LDNS_RR_TYPE_DS || iq->refetch_glue)
919		&& !dname_is_root(iq->qchase.qname))
920		dname_remove_label(&delname, &delnamelen);
921	dp = forwards_lookup(qstate->env->fwds, delname, iq->qchase.qclass);
922	if(!dp)
923		return 0;
924	/* send recursion desired to forward addr */
925	iq->chase_flags |= BIT_RD;
926	iq->dp = delegpt_copy(dp, qstate->region);
927	/* iq->dp checked by caller */
928	verbose(VERB_ALGO, "forwarding request");
929	return 1;
930}
931
932/**
933 * Process the initial part of the request handling. This state roughly
934 * corresponds to resolver algorithms steps 1 (find answer in cache) and 2
935 * (find the best servers to ask).
936 *
937 * Note that all requests start here, and query restarts revisit this state.
938 *
939 * This state either generates: 1) a response, from cache or error, 2) a
940 * priming event, or 3) forwards the request to the next state (init2,
941 * generally).
942 *
943 * @param qstate: query state.
944 * @param iq: iterator query state.
945 * @param ie: iterator shared global environment.
946 * @param id: module id.
947 * @return true if the event needs more request processing immediately,
948 *         false if not.
949 */
950static int
951processInitRequest(struct module_qstate* qstate, struct iter_qstate* iq,
952	struct iter_env* ie, int id)
953{
954	uint8_t* delname;
955	size_t delnamelen;
956	struct dns_msg* msg;
957
958	log_query_info(VERB_DETAIL, "resolving", &qstate->qinfo);
959	/* check effort */
960
961	/* We enforce a maximum number of query restarts. This is primarily a
962	 * cheap way to prevent CNAME loops. */
963	if(iq->query_restart_count > MAX_RESTART_COUNT) {
964		verbose(VERB_QUERY, "request has exceeded the maximum number"
965			" of query restarts with %d", iq->query_restart_count);
966		return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
967	}
968
969	/* We enforce a maximum recursion/dependency depth -- in general,
970	 * this is unnecessary for dependency loops (although it will
971	 * catch those), but it provides a sensible limit to the amount
972	 * of work required to answer a given query. */
973	verbose(VERB_ALGO, "request has dependency depth of %d", iq->depth);
974	if(iq->depth > ie->max_dependency_depth) {
975		verbose(VERB_QUERY, "request has exceeded the maximum "
976			"dependency depth with depth of %d", iq->depth);
977		return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
978	}
979
980	/* If the request is qclass=ANY, setup to generate each class */
981	if(qstate->qinfo.qclass == LDNS_RR_CLASS_ANY) {
982		iq->qchase.qclass = 0;
983		return next_state(iq, COLLECT_CLASS_STATE);
984	}
985
986	/* Resolver Algorithm Step 1 -- Look for the answer in local data. */
987
988	/* This either results in a query restart (CNAME cache response), a
989	 * terminating response (ANSWER), or a cache miss (null). */
990
991	if(qstate->blacklist) {
992		/* if cache, or anything else, was blacklisted then
993		 * getting older results from cache is a bad idea, no cache */
994		verbose(VERB_ALGO, "cache blacklisted, going to the network");
995		msg = NULL;
996	} else {
997		msg = dns_cache_lookup(qstate->env, iq->qchase.qname,
998			iq->qchase.qname_len, iq->qchase.qtype,
999			iq->qchase.qclass, qstate->query_flags,
1000			qstate->region, qstate->env->scratch);
1001		if(!msg && qstate->env->neg_cache) {
1002			/* lookup in negative cache; may result in
1003			 * NOERROR/NODATA or NXDOMAIN answers that need validation */
1004			msg = val_neg_getmsg(qstate->env->neg_cache, &iq->qchase,
1005				qstate->region, qstate->env->rrset_cache,
1006				qstate->env->scratch_buffer,
1007				*qstate->env->now, 1/*add SOA*/, NULL);
1008		}
1009		/* item taken from cache does not match our query name, thus
1010		 * security needs to be re-examined later */
1011		if(msg && query_dname_compare(qstate->qinfo.qname,
1012			iq->qchase.qname) != 0)
1013			msg->rep->security = sec_status_unchecked;
1014	}
1015	if(msg) {
1016		/* handle positive cache response */
1017		enum response_type type = response_type_from_cache(msg,
1018			&iq->qchase);
1019		if(verbosity >= VERB_ALGO) {
1020			log_dns_msg("msg from cache lookup", &msg->qinfo,
1021				msg->rep);
1022			verbose(VERB_ALGO, "msg ttl is %d, prefetch ttl %d",
1023				(int)msg->rep->ttl,
1024				(int)msg->rep->prefetch_ttl);
1025		}
1026
1027		if(type == RESPONSE_TYPE_CNAME) {
1028			uint8_t* sname = 0;
1029			size_t slen = 0;
1030			verbose(VERB_ALGO, "returning CNAME response from "
1031				"cache");
1032			if(!handle_cname_response(qstate, iq, msg,
1033				&sname, &slen))
1034				return error_response(qstate, id,
1035					LDNS_RCODE_SERVFAIL);
1036			iq->qchase.qname = sname;
1037			iq->qchase.qname_len = slen;
1038			/* This *is* a query restart, even if it is a cheap
1039			 * one. */
1040			iq->dp = NULL;
1041			iq->refetch_glue = 0;
1042			iq->query_restart_count++;
1043			iq->sent_count = 0;
1044			sock_list_insert(&qstate->reply_origin, NULL, 0, qstate->region);
1045			return next_state(iq, INIT_REQUEST_STATE);
1046		}
1047
1048		/* if from cache, NULL, else insert 'cache IP' len=0 */
1049		if(qstate->reply_origin)
1050			sock_list_insert(&qstate->reply_origin, NULL, 0, qstate->region);
1051		/* it is an answer, response, to final state */
1052		verbose(VERB_ALGO, "returning answer from cache.");
1053		iq->response = msg;
1054		return final_state(iq);
1055	}
1056
1057	/* attempt to forward the request */
1058	if(forward_request(qstate, iq))
1059	{
1060		if(!iq->dp) {
1061			log_err("alloc failure for forward dp");
1062			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1063		}
1064		iq->refetch_glue = 0;
1065		/* the request has been forwarded.
1066		 * forwarded requests need to be immediately sent to the
1067		 * next state, QUERYTARGETS. */
1068		return next_state(iq, QUERYTARGETS_STATE);
1069	}
1070
1071	/* Resolver Algorithm Step 2 -- find the "best" servers. */
1072
1073	/* first, adjust for DS queries. To avoid the grandparent problem,
1074	 * we just look for the closest set of server to the parent of qname.
1075	 * When re-fetching glue we also need to ask the parent.
1076	 */
1077	if(iq->refetch_glue) {
1078		if(!iq->dp) {
1079			log_err("internal or malloc fail: no dp for refetch");
1080			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1081		}
1082		delname = iq->dp->name;
1083		delnamelen = iq->dp->namelen;
1084	} else {
1085		delname = iq->qchase.qname;
1086		delnamelen = iq->qchase.qname_len;
1087	}
1088	if(iq->qchase.qtype == LDNS_RR_TYPE_DS || iq->refetch_glue ||
1089	   (iq->qchase.qtype == LDNS_RR_TYPE_NS && qstate->prefetch_leeway)) {
1090		/* remove first label from delname, root goes to hints,
1091		 * but only to fetch glue, not for qtype=DS. */
1092		/* also when prefetching an NS record, fetch it again from
1093		 * its parent, just as if it expired, so that you do not
1094		 * get stuck on an older nameserver that gives old NSrecords */
1095		if(dname_is_root(delname) && (iq->refetch_glue ||
1096			(iq->qchase.qtype == LDNS_RR_TYPE_NS &&
1097			qstate->prefetch_leeway)))
1098			delname = NULL; /* go to root priming */
1099		else 	dname_remove_label(&delname, &delnamelen);
1100	}
1101	/* delname is the name to lookup a delegation for. If NULL rootprime */
1102	while(1) {
1103
1104		/* Lookup the delegation in the cache. If null, then the
1105		 * cache needs to be primed for the qclass. */
1106		if(delname)
1107		     iq->dp = dns_cache_find_delegation(qstate->env, delname,
1108			delnamelen, iq->qchase.qtype, iq->qchase.qclass,
1109			qstate->region, &iq->deleg_msg,
1110			*qstate->env->now+qstate->prefetch_leeway);
1111		else iq->dp = NULL;
1112
1113		/* If the cache has returned nothing, then we have a
1114		 * root priming situation. */
1115		if(iq->dp == NULL) {
1116			/* if there is a stub, then no root prime needed */
1117			int r = prime_stub(qstate, iq, id, delname,
1118				iq->qchase.qclass);
1119			if(r == 2)
1120				break; /* got noprime-stub-zone, continue */
1121			else if(r)
1122				return 0; /* stub prime request made */
1123			if(forwards_lookup_root(qstate->env->fwds,
1124				iq->qchase.qclass)) {
1125				/* forward zone root, no root prime needed */
1126				/* fill in some dp - safety belt */
1127				iq->dp = hints_lookup_root(qstate->env->hints,
1128					iq->qchase.qclass);
1129				if(!iq->dp) {
1130					log_err("internal error: no hints dp");
1131					return error_response(qstate, id,
1132						LDNS_RCODE_SERVFAIL);
1133				}
1134				iq->dp = delegpt_copy(iq->dp, qstate->region);
1135				if(!iq->dp) {
1136					log_err("out of memory in safety belt");
1137					return error_response(qstate, id,
1138						LDNS_RCODE_SERVFAIL);
1139				}
1140				return next_state(iq, INIT_REQUEST_2_STATE);
1141			}
1142			/* Note that the result of this will set a new
1143			 * DelegationPoint based on the result of priming. */
1144			if(!prime_root(qstate, iq, id, iq->qchase.qclass))
1145				return error_response(qstate, id,
1146					LDNS_RCODE_REFUSED);
1147
1148			/* priming creates and sends a subordinate query, with
1149			 * this query as the parent. So further processing for
1150			 * this event will stop until reactivated by the
1151			 * results of priming. */
1152			return 0;
1153		}
1154		if(!iq->ratelimit_ok && qstate->prefetch_leeway)
1155			iq->ratelimit_ok = 1; /* allow prefetches, this keeps
1156			otherwise valid data in the cache */
1157		if(!iq->ratelimit_ok && infra_ratelimit_exceeded(
1158			qstate->env->infra_cache, iq->dp->name,
1159			iq->dp->namelen, *qstate->env->now)) {
1160			/* and increment the rate, so that the rate for time
1161			 * now will also exceed the rate, keeping cache fresh */
1162			(void)infra_ratelimit_inc(qstate->env->infra_cache,
1163				iq->dp->name, iq->dp->namelen,
1164				*qstate->env->now);
1165			/* see if we are passed through with slip factor */
1166			if(qstate->env->cfg->ratelimit_factor != 0 &&
1167				ub_random_max(qstate->env->rnd,
1168				    qstate->env->cfg->ratelimit_factor) == 1) {
1169				iq->ratelimit_ok = 1;
1170				log_nametypeclass(VERB_ALGO, "ratelimit allowed through for "
1171					"delegation point", iq->dp->name,
1172					LDNS_RR_TYPE_NS, LDNS_RR_CLASS_IN);
1173			} else {
1174				log_nametypeclass(VERB_ALGO, "ratelimit exceeded with "
1175					"delegation point", iq->dp->name,
1176					LDNS_RR_TYPE_NS, LDNS_RR_CLASS_IN);
1177				return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1178			}
1179		}
1180
1181		/* see if this dp not useless.
1182		 * It is useless if:
1183		 *	o all NS items are required glue.
1184		 *	  or the query is for NS item that is required glue.
1185		 *	o no addresses are provided.
1186		 *	o RD qflag is on.
1187		 * Instead, go up one level, and try to get even further
1188		 * If the root was useless, use safety belt information.
1189		 * Only check cache returns, because replies for servers
1190		 * could be useless but lead to loops (bumping into the
1191		 * same server reply) if useless-checked.
1192		 */
1193		if(iter_dp_is_useless(&qstate->qinfo, qstate->query_flags,
1194			iq->dp)) {
1195			if(dname_is_root(iq->dp->name)) {
1196				/* use safety belt */
1197				verbose(VERB_QUERY, "Cache has root NS but "
1198				"no addresses. Fallback to the safety belt.");
1199				iq->dp = hints_lookup_root(qstate->env->hints,
1200					iq->qchase.qclass);
1201				/* note deleg_msg is from previous lookup,
1202				 * but RD is on, so it is not used */
1203				if(!iq->dp) {
1204					log_err("internal error: no hints dp");
1205					return error_response(qstate, id,
1206						LDNS_RCODE_REFUSED);
1207				}
1208				iq->dp = delegpt_copy(iq->dp, qstate->region);
1209				if(!iq->dp) {
1210					log_err("out of memory in safety belt");
1211					return error_response(qstate, id,
1212						LDNS_RCODE_SERVFAIL);
1213				}
1214				break;
1215			} else {
1216				verbose(VERB_ALGO,
1217					"cache delegation was useless:");
1218				delegpt_log(VERB_ALGO, iq->dp);
1219				/* go up */
1220				delname = iq->dp->name;
1221				delnamelen = iq->dp->namelen;
1222				dname_remove_label(&delname, &delnamelen);
1223			}
1224		} else break;
1225	}
1226
1227	verbose(VERB_ALGO, "cache delegation returns delegpt");
1228	delegpt_log(VERB_ALGO, iq->dp);
1229
1230	/* Otherwise, set the current delegation point and move on to the
1231	 * next state. */
1232	return next_state(iq, INIT_REQUEST_2_STATE);
1233}
1234
1235/**
1236 * Process the second part of the initial request handling. This state
1237 * basically exists so that queries that generate root priming events have
1238 * the same init processing as ones that do not. Request events that reach
1239 * this state must have a valid currentDelegationPoint set.
1240 *
1241 * This part is primarly handling stub zone priming. Events that reach this
1242 * state must have a current delegation point.
1243 *
1244 * @param qstate: query state.
1245 * @param iq: iterator query state.
1246 * @param id: module id.
1247 * @return true if the event needs more request processing immediately,
1248 *         false if not.
1249 */
1250static int
1251processInitRequest2(struct module_qstate* qstate, struct iter_qstate* iq,
1252	int id)
1253{
1254	uint8_t* delname;
1255	size_t delnamelen;
1256	log_query_info(VERB_QUERY, "resolving (init part 2): ",
1257		&qstate->qinfo);
1258
1259	if(iq->refetch_glue) {
1260		if(!iq->dp) {
1261			log_err("internal or malloc fail: no dp for refetch");
1262			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1263		}
1264		delname = iq->dp->name;
1265		delnamelen = iq->dp->namelen;
1266	} else {
1267		delname = iq->qchase.qname;
1268		delnamelen = iq->qchase.qname_len;
1269	}
1270	if(iq->qchase.qtype == LDNS_RR_TYPE_DS || iq->refetch_glue) {
1271		if(!dname_is_root(delname))
1272			dname_remove_label(&delname, &delnamelen);
1273		iq->refetch_glue = 0; /* if CNAME causes restart, no refetch */
1274	}
1275	/* Check to see if we need to prime a stub zone. */
1276	if(prime_stub(qstate, iq, id, delname, iq->qchase.qclass)) {
1277		/* A priming sub request was made */
1278		return 0;
1279	}
1280
1281	/* most events just get forwarded to the next state. */
1282	return next_state(iq, INIT_REQUEST_3_STATE);
1283}
1284
1285/**
1286 * Process the third part of the initial request handling. This state exists
1287 * as a separate state so that queries that generate stub priming events
1288 * will get the tail end of the init process but not repeat the stub priming
1289 * check.
1290 *
1291 * @param qstate: query state.
1292 * @param iq: iterator query state.
1293 * @param id: module id.
1294 * @return true, advancing the event to the QUERYTARGETS_STATE.
1295 */
1296static int
1297processInitRequest3(struct module_qstate* qstate, struct iter_qstate* iq,
1298	int id)
1299{
1300	log_query_info(VERB_QUERY, "resolving (init part 3): ",
1301		&qstate->qinfo);
1302	/* if the cache reply dp equals a validation anchor or msg has DS,
1303	 * then DNSSEC RRSIGs are expected in the reply */
1304	iq->dnssec_expected = iter_indicates_dnssec(qstate->env, iq->dp,
1305		iq->deleg_msg, iq->qchase.qclass);
1306
1307	/* If the RD flag wasn't set, then we just finish with the
1308	 * cached referral as the response. */
1309	if(!(qstate->query_flags & BIT_RD)) {
1310		iq->response = iq->deleg_msg;
1311		if(verbosity >= VERB_ALGO && iq->response)
1312			log_dns_msg("no RD requested, using delegation msg",
1313				&iq->response->qinfo, iq->response->rep);
1314		if(qstate->reply_origin)
1315			sock_list_insert(&qstate->reply_origin, NULL, 0, qstate->region);
1316		return final_state(iq);
1317	}
1318	/* After this point, unset the RD flag -- this query is going to
1319	 * be sent to an auth. server. */
1320	iq->chase_flags &= ~BIT_RD;
1321
1322	/* if dnssec expected, fetch key for the trust-anchor or cached-DS */
1323	if(iq->dnssec_expected && qstate->env->cfg->prefetch_key &&
1324		!(qstate->query_flags&BIT_CD)) {
1325		generate_dnskey_prefetch(qstate, iq, id);
1326		fptr_ok(fptr_whitelist_modenv_detach_subs(
1327			qstate->env->detach_subs));
1328		(*qstate->env->detach_subs)(qstate);
1329	}
1330
1331	/* Jump to the next state. */
1332	return next_state(iq, QUERYTARGETS_STATE);
1333}
1334
1335/**
1336 * Given a basic query, generate a parent-side "target" query.
1337 * These are subordinate queries for missing delegation point target addresses,
1338 * for which only the parent of the delegation provides correct IP addresses.
1339 *
1340 * @param qstate: query state.
1341 * @param iq: iterator query state.
1342 * @param id: module id.
1343 * @param name: target qname.
1344 * @param namelen: target qname length.
1345 * @param qtype: target qtype (either A or AAAA).
1346 * @param qclass: target qclass.
1347 * @return true on success, false on failure.
1348 */
1349static int
1350generate_parentside_target_query(struct module_qstate* qstate,
1351	struct iter_qstate* iq, int id, uint8_t* name, size_t namelen,
1352	uint16_t qtype, uint16_t qclass)
1353{
1354	struct module_qstate* subq;
1355	if(!generate_sub_request(name, namelen, qtype, qclass, qstate,
1356		id, iq, INIT_REQUEST_STATE, FINISHED_STATE, &subq, 0))
1357		return 0;
1358	if(subq) {
1359		struct iter_qstate* subiq =
1360			(struct iter_qstate*)subq->minfo[id];
1361		/* blacklist the cache - we want to fetch parent stuff */
1362		sock_list_insert(&subq->blacklist, NULL, 0, subq->region);
1363		subiq->query_for_pside_glue = 1;
1364		if(dname_subdomain_c(name, iq->dp->name)) {
1365			subiq->dp = delegpt_copy(iq->dp, subq->region);
1366			subiq->dnssec_expected = iter_indicates_dnssec(
1367				qstate->env, subiq->dp, NULL,
1368				subq->qinfo.qclass);
1369			subiq->refetch_glue = 1;
1370		} else {
1371			subiq->dp = dns_cache_find_delegation(qstate->env,
1372				name, namelen, qtype, qclass, subq->region,
1373				&subiq->deleg_msg,
1374				*qstate->env->now+subq->prefetch_leeway);
1375			/* if no dp, then it's from root, refetch unneeded */
1376			if(subiq->dp) {
1377				subiq->dnssec_expected = iter_indicates_dnssec(
1378					qstate->env, subiq->dp, NULL,
1379					subq->qinfo.qclass);
1380				subiq->refetch_glue = 1;
1381			}
1382		}
1383	}
1384	log_nametypeclass(VERB_QUERY, "new pside target", name, qtype, qclass);
1385	return 1;
1386}
1387
1388/**
1389 * Given a basic query, generate a "target" query. These are subordinate
1390 * queries for missing delegation point target addresses.
1391 *
1392 * @param qstate: query state.
1393 * @param iq: iterator query state.
1394 * @param id: module id.
1395 * @param name: target qname.
1396 * @param namelen: target qname length.
1397 * @param qtype: target qtype (either A or AAAA).
1398 * @param qclass: target qclass.
1399 * @return true on success, false on failure.
1400 */
1401static int
1402generate_target_query(struct module_qstate* qstate, struct iter_qstate* iq,
1403        int id, uint8_t* name, size_t namelen, uint16_t qtype, uint16_t qclass)
1404{
1405	struct module_qstate* subq;
1406	if(!generate_sub_request(name, namelen, qtype, qclass, qstate,
1407		id, iq, INIT_REQUEST_STATE, FINISHED_STATE, &subq, 0))
1408		return 0;
1409	log_nametypeclass(VERB_QUERY, "new target", name, qtype, qclass);
1410	return 1;
1411}
1412
1413/**
1414 * Given an event at a certain state, generate zero or more target queries
1415 * for it's current delegation point.
1416 *
1417 * @param qstate: query state.
1418 * @param iq: iterator query state.
1419 * @param ie: iterator shared global environment.
1420 * @param id: module id.
1421 * @param maxtargets: The maximum number of targets to query for.
1422 *	if it is negative, there is no maximum number of targets.
1423 * @param num: returns the number of queries generated and processed,
1424 *	which may be zero if there were no missing targets.
1425 * @return false on error.
1426 */
1427static int
1428query_for_targets(struct module_qstate* qstate, struct iter_qstate* iq,
1429        struct iter_env* ie, int id, int maxtargets, int* num)
1430{
1431	int query_count = 0;
1432	struct delegpt_ns* ns;
1433	int missing;
1434	int toget = 0;
1435
1436	if(iq->depth == ie->max_dependency_depth)
1437		return 0;
1438	if(iq->depth > 0 && iq->target_count &&
1439		iq->target_count[1] > MAX_TARGET_COUNT) {
1440		char s[LDNS_MAX_DOMAINLEN+1];
1441		dname_str(qstate->qinfo.qname, s);
1442		verbose(VERB_QUERY, "request %s has exceeded the maximum "
1443			"number of glue fetches %d", s, iq->target_count[1]);
1444		return 0;
1445	}
1446
1447	iter_mark_cycle_targets(qstate, iq->dp);
1448	missing = (int)delegpt_count_missing_targets(iq->dp);
1449	log_assert(maxtargets != 0); /* that would not be useful */
1450
1451	/* Generate target requests. Basically, any missing targets
1452	 * are queried for here, regardless if it is necessary to do
1453	 * so to continue processing. */
1454	if(maxtargets < 0 || maxtargets > missing)
1455		toget = missing;
1456	else	toget = maxtargets;
1457	if(toget == 0) {
1458		*num = 0;
1459		return 1;
1460	}
1461	/* select 'toget' items from the total of 'missing' items */
1462	log_assert(toget <= missing);
1463
1464	/* loop over missing targets */
1465	for(ns = iq->dp->nslist; ns; ns = ns->next) {
1466		if(ns->resolved)
1467			continue;
1468
1469		/* randomly select this item with probability toget/missing */
1470		if(!iter_ns_probability(qstate->env->rnd, toget, missing)) {
1471			/* do not select this one, next; select toget number
1472			 * of items from a list one less in size */
1473			missing --;
1474			continue;
1475		}
1476
1477		if(ie->supports_ipv6 && !ns->got6) {
1478			/* Send the AAAA request. */
1479			if(!generate_target_query(qstate, iq, id,
1480				ns->name, ns->namelen,
1481				LDNS_RR_TYPE_AAAA, iq->qchase.qclass)) {
1482				*num = query_count;
1483				if(query_count > 0)
1484					qstate->ext_state[id] = module_wait_subquery;
1485				return 0;
1486			}
1487			query_count++;
1488		}
1489		/* Send the A request. */
1490		if(ie->supports_ipv4 && !ns->got4) {
1491			if(!generate_target_query(qstate, iq, id,
1492				ns->name, ns->namelen,
1493				LDNS_RR_TYPE_A, iq->qchase.qclass)) {
1494				*num = query_count;
1495				if(query_count > 0)
1496					qstate->ext_state[id] = module_wait_subquery;
1497				return 0;
1498			}
1499			query_count++;
1500		}
1501
1502		/* mark this target as in progress. */
1503		ns->resolved = 1;
1504		missing--;
1505		toget--;
1506		if(toget == 0)
1507			break;
1508	}
1509	*num = query_count;
1510	if(query_count > 0)
1511		qstate->ext_state[id] = module_wait_subquery;
1512
1513	return 1;
1514}
1515
1516/** see if last resort is possible - does config allow queries to parent */
1517static int
1518can_have_last_resort(struct module_env* env, struct delegpt* dp,
1519	struct iter_qstate* iq)
1520{
1521	struct delegpt* fwddp;
1522	struct iter_hints_stub* stub;
1523	/* do not process a last resort (the parent side) if a stub
1524	 * or forward is configured, because we do not want to go 'above'
1525	 * the configured servers */
1526	if(!dname_is_root(dp->name) && (stub = (struct iter_hints_stub*)
1527		name_tree_find(&env->hints->tree, dp->name, dp->namelen,
1528		dp->namelabs, iq->qchase.qclass)) &&
1529		/* has_parent side is turned off for stub_first, where we
1530		 * are allowed to go to the parent */
1531		stub->dp->has_parent_side_NS) {
1532		verbose(VERB_QUERY, "configured stub servers failed -- returning SERVFAIL");
1533		return 0;
1534	}
1535	if((fwddp = forwards_find(env->fwds, dp->name, iq->qchase.qclass)) &&
1536		/* has_parent_side is turned off for forward_first, where
1537		 * we are allowed to go to the parent */
1538		fwddp->has_parent_side_NS) {
1539		verbose(VERB_QUERY, "configured forward servers failed -- returning SERVFAIL");
1540		return 0;
1541	}
1542	return 1;
1543}
1544
1545/**
1546 * Called by processQueryTargets when it would like extra targets to query
1547 * but it seems to be out of options.  At last resort some less appealing
1548 * options are explored.  If there are no more options, the result is SERVFAIL
1549 *
1550 * @param qstate: query state.
1551 * @param iq: iterator query state.
1552 * @param ie: iterator shared global environment.
1553 * @param id: module id.
1554 * @return true if the event requires more request processing immediately,
1555 *         false if not.
1556 */
1557static int
1558processLastResort(struct module_qstate* qstate, struct iter_qstate* iq,
1559	struct iter_env* ie, int id)
1560{
1561	struct delegpt_ns* ns;
1562	int query_count = 0;
1563	verbose(VERB_ALGO, "No more query targets, attempting last resort");
1564	log_assert(iq->dp);
1565
1566	if(!can_have_last_resort(qstate->env, iq->dp, iq)) {
1567		/* fail -- no more targets, no more hope of targets, no hope
1568		 * of a response. */
1569		return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
1570	}
1571	if(!iq->dp->has_parent_side_NS && dname_is_root(iq->dp->name)) {
1572		struct delegpt* p = hints_lookup_root(qstate->env->hints,
1573			iq->qchase.qclass);
1574		if(p) {
1575			struct delegpt_ns* ns;
1576			struct delegpt_addr* a;
1577			iq->chase_flags &= ~BIT_RD; /* go to authorities */
1578			for(ns = p->nslist; ns; ns=ns->next) {
1579				(void)delegpt_add_ns(iq->dp, qstate->region,
1580					ns->name, ns->lame);
1581			}
1582			for(a = p->target_list; a; a=a->next_target) {
1583				(void)delegpt_add_addr(iq->dp, qstate->region,
1584					&a->addr, a->addrlen, a->bogus,
1585					a->lame);
1586			}
1587		}
1588		iq->dp->has_parent_side_NS = 1;
1589	} else if(!iq->dp->has_parent_side_NS) {
1590		if(!iter_lookup_parent_NS_from_cache(qstate->env, iq->dp,
1591			qstate->region, &qstate->qinfo)
1592			|| !iq->dp->has_parent_side_NS) {
1593			/* if: malloc failure in lookup go up to try */
1594			/* if: no parent NS in cache - go up one level */
1595			verbose(VERB_ALGO, "try to grab parent NS");
1596			iq->store_parent_NS = iq->dp;
1597			iq->chase_flags &= ~BIT_RD; /* go to authorities */
1598			iq->deleg_msg = NULL;
1599			iq->refetch_glue = 1;
1600			iq->query_restart_count++;
1601			iq->sent_count = 0;
1602			return next_state(iq, INIT_REQUEST_STATE);
1603		}
1604	}
1605	/* see if that makes new names available */
1606	if(!cache_fill_missing(qstate->env, iq->qchase.qclass,
1607		qstate->region, iq->dp))
1608		log_err("out of memory in cache_fill_missing");
1609	if(iq->dp->usable_list) {
1610		verbose(VERB_ALGO, "try parent-side-name, w. glue from cache");
1611		return next_state(iq, QUERYTARGETS_STATE);
1612	}
1613	/* try to fill out parent glue from cache */
1614	if(iter_lookup_parent_glue_from_cache(qstate->env, iq->dp,
1615		qstate->region, &qstate->qinfo)) {
1616		/* got parent stuff from cache, see if we can continue */
1617		verbose(VERB_ALGO, "try parent-side glue from cache");
1618		return next_state(iq, QUERYTARGETS_STATE);
1619	}
1620	/* query for an extra name added by the parent-NS record */
1621	if(delegpt_count_missing_targets(iq->dp) > 0) {
1622		int qs = 0;
1623		verbose(VERB_ALGO, "try parent-side target name");
1624		if(!query_for_targets(qstate, iq, ie, id, 1, &qs)) {
1625			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1626		}
1627		iq->num_target_queries += qs;
1628		target_count_increase(iq, qs);
1629		if(qs != 0) {
1630			qstate->ext_state[id] = module_wait_subquery;
1631			return 0; /* and wait for them */
1632		}
1633	}
1634	if(iq->depth == ie->max_dependency_depth) {
1635		verbose(VERB_QUERY, "maxdepth and need more nameservers, fail");
1636		return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
1637	}
1638	if(iq->depth > 0 && iq->target_count &&
1639		iq->target_count[1] > MAX_TARGET_COUNT) {
1640		char s[LDNS_MAX_DOMAINLEN+1];
1641		dname_str(qstate->qinfo.qname, s);
1642		verbose(VERB_QUERY, "request %s has exceeded the maximum "
1643			"number of glue fetches %d", s, iq->target_count[1]);
1644		return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
1645	}
1646	/* mark cycle targets for parent-side lookups */
1647	iter_mark_pside_cycle_targets(qstate, iq->dp);
1648	/* see if we can issue queries to get nameserver addresses */
1649	/* this lookup is not randomized, but sequential. */
1650	for(ns = iq->dp->nslist; ns; ns = ns->next) {
1651		/* query for parent-side A and AAAA for nameservers */
1652		if(ie->supports_ipv6 && !ns->done_pside6) {
1653			/* Send the AAAA request. */
1654			if(!generate_parentside_target_query(qstate, iq, id,
1655				ns->name, ns->namelen,
1656				LDNS_RR_TYPE_AAAA, iq->qchase.qclass))
1657				return error_response(qstate, id,
1658					LDNS_RCODE_SERVFAIL);
1659			ns->done_pside6 = 1;
1660			query_count++;
1661		}
1662		if(ie->supports_ipv4 && !ns->done_pside4) {
1663			/* Send the A request. */
1664			if(!generate_parentside_target_query(qstate, iq, id,
1665				ns->name, ns->namelen,
1666				LDNS_RR_TYPE_A, iq->qchase.qclass))
1667				return error_response(qstate, id,
1668					LDNS_RCODE_SERVFAIL);
1669			ns->done_pside4 = 1;
1670			query_count++;
1671		}
1672		if(query_count != 0) { /* suspend to await results */
1673			verbose(VERB_ALGO, "try parent-side glue lookup");
1674			iq->num_target_queries += query_count;
1675			target_count_increase(iq, query_count);
1676			qstate->ext_state[id] = module_wait_subquery;
1677			return 0;
1678		}
1679	}
1680
1681	/* if this was a parent-side glue query itself, then store that
1682	 * failure in cache. */
1683	if(iq->query_for_pside_glue && !iq->pside_glue)
1684		iter_store_parentside_neg(qstate->env, &qstate->qinfo,
1685			iq->deleg_msg?iq->deleg_msg->rep:
1686			(iq->response?iq->response->rep:NULL));
1687
1688	verbose(VERB_QUERY, "out of query targets -- returning SERVFAIL");
1689	/* fail -- no more targets, no more hope of targets, no hope
1690	 * of a response. */
1691	return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
1692}
1693
1694/**
1695 * Try to find the NS record set that will resolve a qtype DS query. Due
1696 * to grandparent/grandchild reasons we did not get a proper lookup right
1697 * away.  We need to create type NS queries until we get the right parent
1698 * for this lookup.  We remove labels from the query to find the right point.
1699 * If we end up at the old dp name, then there is no solution.
1700 *
1701 * @param qstate: query state.
1702 * @param iq: iterator query state.
1703 * @param id: module id.
1704 * @return true if the event requires more immediate processing, false if
1705 *         not. This is generally only true when forwarding the request to
1706 *         the final state (i.e., on answer).
1707 */
1708static int
1709processDSNSFind(struct module_qstate* qstate, struct iter_qstate* iq, int id)
1710{
1711	struct module_qstate* subq = NULL;
1712	verbose(VERB_ALGO, "processDSNSFind");
1713
1714	if(!iq->dsns_point) {
1715		/* initialize */
1716		iq->dsns_point = iq->qchase.qname;
1717		iq->dsns_point_len = iq->qchase.qname_len;
1718	}
1719	/* robustcheck for internal error: we are not underneath the dp */
1720	if(!dname_subdomain_c(iq->dsns_point, iq->dp->name)) {
1721		return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
1722	}
1723
1724	/* go up one (more) step, until we hit the dp, if so, end */
1725	dname_remove_label(&iq->dsns_point, &iq->dsns_point_len);
1726	if(query_dname_compare(iq->dsns_point, iq->dp->name) == 0) {
1727		/* there was no inbetween nameserver, use the old delegation
1728		 * point again.  And this time, because dsns_point is nonNULL
1729		 * we are going to accept the (bad) result */
1730		iq->state = QUERYTARGETS_STATE;
1731		return 1;
1732	}
1733	iq->state = DSNS_FIND_STATE;
1734
1735	/* spawn NS lookup (validation not needed, this is for DS lookup) */
1736	log_nametypeclass(VERB_ALGO, "fetch nameservers",
1737		iq->dsns_point, LDNS_RR_TYPE_NS, iq->qchase.qclass);
1738	if(!generate_sub_request(iq->dsns_point, iq->dsns_point_len,
1739		LDNS_RR_TYPE_NS, iq->qchase.qclass, qstate, id, iq,
1740		INIT_REQUEST_STATE, FINISHED_STATE, &subq, 0)) {
1741		return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
1742	}
1743
1744	return 0;
1745}
1746
1747/**
1748 * This is the request event state where the request will be sent to one of
1749 * its current query targets. This state also handles issuing target lookup
1750 * queries for missing target IP addresses. Queries typically iterate on
1751 * this state, both when they are just trying different targets for a given
1752 * delegation point, and when they change delegation points. This state
1753 * roughly corresponds to RFC 1034 algorithm steps 3 and 4.
1754 *
1755 * @param qstate: query state.
1756 * @param iq: iterator query state.
1757 * @param ie: iterator shared global environment.
1758 * @param id: module id.
1759 * @return true if the event requires more request processing immediately,
1760 *         false if not. This state only returns true when it is generating
1761 *         a SERVFAIL response because the query has hit a dead end.
1762 */
1763static int
1764processQueryTargets(struct module_qstate* qstate, struct iter_qstate* iq,
1765	struct iter_env* ie, int id)
1766{
1767	int tf_policy;
1768	struct delegpt_addr* target;
1769	struct outbound_entry* outq;
1770
1771	/* NOTE: a request will encounter this state for each target it
1772	 * needs to send a query to. That is, at least one per referral,
1773	 * more if some targets timeout or return throwaway answers. */
1774
1775	log_query_info(VERB_QUERY, "processQueryTargets:", &qstate->qinfo);
1776	verbose(VERB_ALGO, "processQueryTargets: targetqueries %d, "
1777		"currentqueries %d sentcount %d", iq->num_target_queries,
1778		iq->num_current_queries, iq->sent_count);
1779
1780	/* Make sure that we haven't run away */
1781	/* FIXME: is this check even necessary? */
1782	if(iq->referral_count > MAX_REFERRAL_COUNT) {
1783		verbose(VERB_QUERY, "request has exceeded the maximum "
1784			"number of referrrals with %d", iq->referral_count);
1785		return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1786	}
1787	if(iq->sent_count > MAX_SENT_COUNT) {
1788		verbose(VERB_QUERY, "request has exceeded the maximum "
1789			"number of sends with %d", iq->sent_count);
1790		return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1791	}
1792
1793	/* Make sure we have a delegation point, otherwise priming failed
1794	 * or another failure occurred */
1795	if(!iq->dp) {
1796		verbose(VERB_QUERY, "Failed to get a delegation, giving up");
1797		return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1798	}
1799	if(!ie->supports_ipv6)
1800		delegpt_no_ipv6(iq->dp);
1801	if(!ie->supports_ipv4)
1802		delegpt_no_ipv4(iq->dp);
1803	delegpt_log(VERB_ALGO, iq->dp);
1804
1805	if(iq->num_current_queries>0) {
1806		/* already busy answering a query, this restart is because
1807		 * more delegpt addrs became available, wait for existing
1808		 * query. */
1809		verbose(VERB_ALGO, "woke up, but wait for outstanding query");
1810		qstate->ext_state[id] = module_wait_reply;
1811		return 0;
1812	}
1813
1814	tf_policy = 0;
1815	/* < not <=, because although the array is large enough for <=, the
1816	 * generated query will immediately be discarded due to depth and
1817	 * that servfail is cached, which is not good as opportunism goes. */
1818	if(iq->depth < ie->max_dependency_depth
1819		&& iq->sent_count < TARGET_FETCH_STOP) {
1820		tf_policy = ie->target_fetch_policy[iq->depth];
1821	}
1822
1823	/* if in 0x20 fallback get as many targets as possible */
1824	if(iq->caps_fallback) {
1825		int extra = 0;
1826		size_t naddr, nres, navail;
1827		if(!query_for_targets(qstate, iq, ie, id, -1, &extra)) {
1828			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1829		}
1830		iq->num_target_queries += extra;
1831		target_count_increase(iq, extra);
1832		if(iq->num_target_queries > 0) {
1833			/* wait to get all targets, we want to try em */
1834			verbose(VERB_ALGO, "wait for all targets for fallback");
1835			qstate->ext_state[id] = module_wait_reply;
1836			return 0;
1837		}
1838		/* did we do enough fallback queries already? */
1839		delegpt_count_addr(iq->dp, &naddr, &nres, &navail);
1840		/* the current caps_server is the number of fallbacks sent.
1841		 * the original query is one that matched too, so we have
1842		 * caps_server+1 number of matching queries now */
1843		if(iq->caps_server+1 >= naddr*3 ||
1844			iq->caps_server*2+2 >= MAX_SENT_COUNT) {
1845			/* *2 on sentcount check because ipv6 may fail */
1846			/* we're done, process the response */
1847			verbose(VERB_ALGO, "0x20 fallback had %d responses "
1848				"match for %d wanted, done.",
1849				(int)iq->caps_server+1, (int)naddr*3);
1850			iq->response = iq->caps_response;
1851			iq->caps_fallback = 0;
1852			iter_dec_attempts(iq->dp, 3); /* space for fallback */
1853			iq->num_current_queries++; /* RespState decrements it*/
1854			iq->referral_count++; /* make sure we don't loop */
1855			iq->sent_count = 0;
1856			iq->state = QUERY_RESP_STATE;
1857			return 1;
1858		}
1859		verbose(VERB_ALGO, "0x20 fallback number %d",
1860			(int)iq->caps_server);
1861
1862	/* if there is a policy to fetch missing targets
1863	 * opportunistically, do it. we rely on the fact that once a
1864	 * query (or queries) for a missing name have been issued,
1865	 * they will not show up again. */
1866	} else if(tf_policy != 0) {
1867		int extra = 0;
1868		verbose(VERB_ALGO, "attempt to get extra %d targets",
1869			tf_policy);
1870		(void)query_for_targets(qstate, iq, ie, id, tf_policy, &extra);
1871		/* errors ignored, these targets are not strictly necessary for
1872		 * this result, we do not have to reply with SERVFAIL */
1873		iq->num_target_queries += extra;
1874		target_count_increase(iq, extra);
1875	}
1876
1877	/* Add the current set of unused targets to our queue. */
1878	delegpt_add_unused_targets(iq->dp);
1879
1880	/* Select the next usable target, filtering out unsuitable targets. */
1881	target = iter_server_selection(ie, qstate->env, iq->dp,
1882		iq->dp->name, iq->dp->namelen, iq->qchase.qtype,
1883		&iq->dnssec_lame_query, &iq->chase_to_rd,
1884		iq->num_target_queries, qstate->blacklist);
1885
1886	/* If no usable target was selected... */
1887	if(!target) {
1888		/* Here we distinguish between three states: generate a new
1889		 * target query, just wait, or quit (with a SERVFAIL).
1890		 * We have the following information: number of active
1891		 * target queries, number of active current queries,
1892		 * the presence of missing targets at this delegation
1893		 * point, and the given query target policy. */
1894
1895		/* Check for the wait condition. If this is true, then
1896		 * an action must be taken. */
1897		if(iq->num_target_queries==0 && iq->num_current_queries==0) {
1898			/* If there is nothing to wait for, then we need
1899			 * to distinguish between generating (a) new target
1900			 * query, or failing. */
1901			if(delegpt_count_missing_targets(iq->dp) > 0) {
1902				int qs = 0;
1903				verbose(VERB_ALGO, "querying for next "
1904					"missing target");
1905				if(!query_for_targets(qstate, iq, ie, id,
1906					1, &qs)) {
1907					return error_response(qstate, id,
1908						LDNS_RCODE_SERVFAIL);
1909				}
1910				if(qs == 0 &&
1911				   delegpt_count_missing_targets(iq->dp) == 0){
1912					/* it looked like there were missing
1913					 * targets, but they did not turn up.
1914					 * Try the bad choices again (if any),
1915					 * when we get back here missing==0,
1916					 * so this is not a loop. */
1917					return 1;
1918				}
1919				iq->num_target_queries += qs;
1920				target_count_increase(iq, qs);
1921			}
1922			/* Since a target query might have been made, we
1923			 * need to check again. */
1924			if(iq->num_target_queries == 0) {
1925				/* if in capsforid fallback, instead of last
1926				 * resort, we agree with the current reply
1927				 * we have (if any) (our count of addrs bad)*/
1928				if(iq->caps_fallback && iq->caps_reply) {
1929					/* we're done, process the response */
1930					verbose(VERB_ALGO, "0x20 fallback had %d responses, "
1931						"but no more servers except "
1932						"last resort, done.",
1933						(int)iq->caps_server+1);
1934					iq->response = iq->caps_response;
1935					iq->caps_fallback = 0;
1936					iter_dec_attempts(iq->dp, 3); /* space for fallback */
1937					iq->num_current_queries++; /* RespState decrements it*/
1938					iq->referral_count++; /* make sure we don't loop */
1939					iq->sent_count = 0;
1940					iq->state = QUERY_RESP_STATE;
1941					return 1;
1942				}
1943				return processLastResort(qstate, iq, ie, id);
1944			}
1945		}
1946
1947		/* otherwise, we have no current targets, so submerge
1948		 * until one of the target or direct queries return. */
1949		if(iq->num_target_queries>0 && iq->num_current_queries>0) {
1950			verbose(VERB_ALGO, "no current targets -- waiting "
1951				"for %d targets to resolve or %d outstanding"
1952				" queries to respond", iq->num_target_queries,
1953				iq->num_current_queries);
1954			qstate->ext_state[id] = module_wait_reply;
1955		} else if(iq->num_target_queries>0) {
1956			verbose(VERB_ALGO, "no current targets -- waiting "
1957				"for %d targets to resolve.",
1958				iq->num_target_queries);
1959			qstate->ext_state[id] = module_wait_subquery;
1960		} else {
1961			verbose(VERB_ALGO, "no current targets -- waiting "
1962				"for %d outstanding queries to respond.",
1963				iq->num_current_queries);
1964			qstate->ext_state[id] = module_wait_reply;
1965		}
1966		return 0;
1967	}
1968
1969	/* if not forwarding, check ratelimits per delegationpoint name */
1970	if(!(iq->chase_flags & BIT_RD) && !iq->ratelimit_ok) {
1971		if(!infra_ratelimit_inc(qstate->env->infra_cache, iq->dp->name,
1972			iq->dp->namelen, *qstate->env->now)) {
1973			verbose(VERB_ALGO, "query exceeded ratelimits");
1974			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1975		}
1976	}
1977
1978	/* We have a valid target. */
1979	if(verbosity >= VERB_QUERY) {
1980		log_query_info(VERB_QUERY, "sending query:", &iq->qchase);
1981		log_name_addr(VERB_QUERY, "sending to target:", iq->dp->name,
1982			&target->addr, target->addrlen);
1983		verbose(VERB_ALGO, "dnssec status: %s%s",
1984			iq->dnssec_expected?"expected": "not expected",
1985			iq->dnssec_lame_query?" but lame_query anyway": "");
1986	}
1987	fptr_ok(fptr_whitelist_modenv_send_query(qstate->env->send_query));
1988	outq = (*qstate->env->send_query)(
1989		iq->qchase.qname, iq->qchase.qname_len,
1990		iq->qchase.qtype, iq->qchase.qclass,
1991		iq->chase_flags | (iq->chase_to_rd?BIT_RD:0), EDNS_DO|BIT_CD,
1992		iq->dnssec_expected, iq->caps_fallback || is_caps_whitelisted(
1993		ie, iq), &target->addr, target->addrlen, iq->dp->name,
1994		iq->dp->namelen, qstate);
1995	if(!outq) {
1996		log_addr(VERB_DETAIL, "error sending query to auth server",
1997			&target->addr, target->addrlen);
1998		if(!(iq->chase_flags & BIT_RD) && !iq->ratelimit_ok)
1999		    infra_ratelimit_dec(qstate->env->infra_cache, iq->dp->name,
2000			iq->dp->namelen, *qstate->env->now);
2001		return next_state(iq, QUERYTARGETS_STATE);
2002	}
2003	outbound_list_insert(&iq->outlist, outq);
2004	iq->num_current_queries++;
2005	iq->sent_count++;
2006	qstate->ext_state[id] = module_wait_reply;
2007
2008	return 0;
2009}
2010
2011/** find NS rrset in given list */
2012static struct ub_packed_rrset_key*
2013find_NS(struct reply_info* rep, size_t from, size_t to)
2014{
2015	size_t i;
2016	for(i=from; i<to; i++) {
2017		if(ntohs(rep->rrsets[i]->rk.type) == LDNS_RR_TYPE_NS)
2018			return rep->rrsets[i];
2019	}
2020	return NULL;
2021}
2022
2023
2024/**
2025 * Process the query response. All queries end up at this state first. This
2026 * process generally consists of analyzing the response and routing the
2027 * event to the next state (either bouncing it back to a request state, or
2028 * terminating the processing for this event).
2029 *
2030 * @param qstate: query state.
2031 * @param iq: iterator query state.
2032 * @param id: module id.
2033 * @return true if the event requires more immediate processing, false if
2034 *         not. This is generally only true when forwarding the request to
2035 *         the final state (i.e., on answer).
2036 */
2037static int
2038processQueryResponse(struct module_qstate* qstate, struct iter_qstate* iq,
2039	int id)
2040{
2041	int dnsseclame = 0;
2042	enum response_type type;
2043	iq->num_current_queries--;
2044	if(iq->response == NULL) {
2045		iq->chase_to_rd = 0;
2046		iq->dnssec_lame_query = 0;
2047		verbose(VERB_ALGO, "query response was timeout");
2048		return next_state(iq, QUERYTARGETS_STATE);
2049	}
2050	type = response_type_from_server(
2051		(int)((iq->chase_flags&BIT_RD) || iq->chase_to_rd),
2052		iq->response, &iq->qchase, iq->dp);
2053	iq->chase_to_rd = 0;
2054	if(type == RESPONSE_TYPE_REFERRAL && (iq->chase_flags&BIT_RD)) {
2055		/* When forwarding (RD bit is set), we handle referrals
2056		 * differently. No queries should be sent elsewhere */
2057		type = RESPONSE_TYPE_ANSWER;
2058	}
2059	if(iq->dnssec_expected && !iq->dnssec_lame_query &&
2060		!(iq->chase_flags&BIT_RD)
2061		&& type != RESPONSE_TYPE_LAME
2062		&& type != RESPONSE_TYPE_REC_LAME
2063		&& type != RESPONSE_TYPE_THROWAWAY
2064		&& type != RESPONSE_TYPE_UNTYPED) {
2065		/* a possible answer, see if it is missing DNSSEC */
2066		/* but not when forwarding, so we dont mark fwder lame */
2067		if(!iter_msg_has_dnssec(iq->response)) {
2068			/* Mark this address as dnsseclame in this dp,
2069			 * because that will make serverselection disprefer
2070			 * it, but also, once it is the only final option,
2071			 * use dnssec-lame-bypass if it needs to query there.*/
2072			if(qstate->reply) {
2073				struct delegpt_addr* a = delegpt_find_addr(
2074					iq->dp, &qstate->reply->addr,
2075					qstate->reply->addrlen);
2076				if(a) a->dnsseclame = 1;
2077			}
2078			/* test the answer is from the zone we expected,
2079		 	 * otherwise, (due to parent,child on same server), we
2080		 	 * might mark the server,zone lame inappropriately */
2081			if(!iter_msg_from_zone(iq->response, iq->dp, type,
2082				iq->qchase.qclass))
2083				qstate->reply = NULL;
2084			type = RESPONSE_TYPE_LAME;
2085			dnsseclame = 1;
2086		}
2087	} else iq->dnssec_lame_query = 0;
2088	/* see if referral brings us close to the target */
2089	if(type == RESPONSE_TYPE_REFERRAL) {
2090		struct ub_packed_rrset_key* ns = find_NS(
2091			iq->response->rep, iq->response->rep->an_numrrsets,
2092			iq->response->rep->an_numrrsets
2093			+ iq->response->rep->ns_numrrsets);
2094		if(!ns) ns = find_NS(iq->response->rep, 0,
2095				iq->response->rep->an_numrrsets);
2096		if(!ns || !dname_strict_subdomain_c(ns->rk.dname, iq->dp->name)
2097			|| !dname_subdomain_c(iq->qchase.qname, ns->rk.dname)){
2098			verbose(VERB_ALGO, "bad referral, throwaway");
2099			type = RESPONSE_TYPE_THROWAWAY;
2100		} else
2101			iter_scrub_ds(iq->response, ns, iq->dp->name);
2102	} else iter_scrub_ds(iq->response, NULL, NULL);
2103
2104	/* handle each of the type cases */
2105	if(type == RESPONSE_TYPE_ANSWER) {
2106		/* ANSWER type responses terminate the query algorithm,
2107		 * so they sent on their */
2108		if(verbosity >= VERB_DETAIL) {
2109			verbose(VERB_DETAIL, "query response was %s",
2110				FLAGS_GET_RCODE(iq->response->rep->flags)
2111				==LDNS_RCODE_NXDOMAIN?"NXDOMAIN ANSWER":
2112				(iq->response->rep->an_numrrsets?"ANSWER":
2113				"nodata ANSWER"));
2114		}
2115		/* if qtype is DS, check we have the right level of answer,
2116		 * like grandchild answer but we need the middle, reject it */
2117		if(iq->qchase.qtype == LDNS_RR_TYPE_DS && !iq->dsns_point
2118			&& !(iq->chase_flags&BIT_RD)
2119			&& iter_ds_toolow(iq->response, iq->dp)
2120			&& iter_dp_cangodown(&iq->qchase, iq->dp)) {
2121			/* close down outstanding requests to be discarded */
2122			outbound_list_clear(&iq->outlist);
2123			iq->num_current_queries = 0;
2124			fptr_ok(fptr_whitelist_modenv_detach_subs(
2125				qstate->env->detach_subs));
2126			(*qstate->env->detach_subs)(qstate);
2127			iq->num_target_queries = 0;
2128			return processDSNSFind(qstate, iq, id);
2129		}
2130		iter_dns_store(qstate->env, &iq->response->qinfo,
2131			iq->response->rep, 0, qstate->prefetch_leeway,
2132			iq->dp&&iq->dp->has_parent_side_NS,
2133			qstate->region, qstate->query_flags);
2134		/* close down outstanding requests to be discarded */
2135		outbound_list_clear(&iq->outlist);
2136		iq->num_current_queries = 0;
2137		fptr_ok(fptr_whitelist_modenv_detach_subs(
2138			qstate->env->detach_subs));
2139		(*qstate->env->detach_subs)(qstate);
2140		iq->num_target_queries = 0;
2141		if(qstate->reply)
2142			sock_list_insert(&qstate->reply_origin,
2143				&qstate->reply->addr, qstate->reply->addrlen,
2144				qstate->region);
2145		return final_state(iq);
2146	} else if(type == RESPONSE_TYPE_REFERRAL) {
2147		/* REFERRAL type responses get a reset of the
2148		 * delegation point, and back to the QUERYTARGETS_STATE. */
2149		verbose(VERB_DETAIL, "query response was REFERRAL");
2150
2151		if(!(iq->chase_flags & BIT_RD) && !iq->ratelimit_ok) {
2152			/* we have a referral, no ratelimit, we can send
2153			 * our queries to the given name */
2154			infra_ratelimit_dec(qstate->env->infra_cache,
2155				iq->dp->name, iq->dp->namelen,
2156				*qstate->env->now);
2157		}
2158
2159		/* if hardened, only store referral if we asked for it */
2160		if(!qstate->env->cfg->harden_referral_path ||
2161		    (  qstate->qinfo.qtype == LDNS_RR_TYPE_NS
2162			&& (qstate->query_flags&BIT_RD)
2163			&& !(qstate->query_flags&BIT_CD)
2164			   /* we know that all other NS rrsets are scrubbed
2165			    * away, thus on referral only one is left.
2166			    * see if that equals the query name... */
2167			&& ( /* auth section, but sometimes in answer section*/
2168			  reply_find_rrset_section_ns(iq->response->rep,
2169				iq->qchase.qname, iq->qchase.qname_len,
2170				LDNS_RR_TYPE_NS, iq->qchase.qclass)
2171			  || reply_find_rrset_section_an(iq->response->rep,
2172				iq->qchase.qname, iq->qchase.qname_len,
2173				LDNS_RR_TYPE_NS, iq->qchase.qclass)
2174			  )
2175		    )) {
2176			/* Store the referral under the current query */
2177			/* no prefetch-leeway, since its not the answer */
2178			iter_dns_store(qstate->env, &iq->response->qinfo,
2179				iq->response->rep, 1, 0, 0, NULL, 0);
2180			if(iq->store_parent_NS)
2181				iter_store_parentside_NS(qstate->env,
2182					iq->response->rep);
2183			if(qstate->env->neg_cache)
2184				val_neg_addreferral(qstate->env->neg_cache,
2185					iq->response->rep, iq->dp->name);
2186		}
2187		/* store parent-side-in-zone-glue, if directly queried for */
2188		if(iq->query_for_pside_glue && !iq->pside_glue) {
2189			iq->pside_glue = reply_find_rrset(iq->response->rep,
2190				iq->qchase.qname, iq->qchase.qname_len,
2191				iq->qchase.qtype, iq->qchase.qclass);
2192			if(iq->pside_glue) {
2193				log_rrset_key(VERB_ALGO, "found parent-side "
2194					"glue", iq->pside_glue);
2195				iter_store_parentside_rrset(qstate->env,
2196					iq->pside_glue);
2197			}
2198		}
2199
2200		/* Reset the event state, setting the current delegation
2201		 * point to the referral. */
2202		iq->deleg_msg = iq->response;
2203		iq->dp = delegpt_from_message(iq->response, qstate->region);
2204		if(!iq->dp)
2205			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2206		if(!cache_fill_missing(qstate->env, iq->qchase.qclass,
2207			qstate->region, iq->dp))
2208			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2209		if(iq->store_parent_NS && query_dname_compare(iq->dp->name,
2210			iq->store_parent_NS->name) == 0)
2211			iter_merge_retry_counts(iq->dp, iq->store_parent_NS);
2212		delegpt_log(VERB_ALGO, iq->dp);
2213		/* Count this as a referral. */
2214		iq->referral_count++;
2215		iq->sent_count = 0;
2216		/* see if the next dp is a trust anchor, or a DS was sent
2217		 * along, indicating dnssec is expected for next zone */
2218		iq->dnssec_expected = iter_indicates_dnssec(qstate->env,
2219			iq->dp, iq->response, iq->qchase.qclass);
2220		/* if dnssec, validating then also fetch the key for the DS */
2221		if(iq->dnssec_expected && qstate->env->cfg->prefetch_key &&
2222			!(qstate->query_flags&BIT_CD))
2223			generate_dnskey_prefetch(qstate, iq, id);
2224
2225		/* spawn off NS and addr to auth servers for the NS we just
2226		 * got in the referral. This gets authoritative answer
2227		 * (answer section trust level) rrset.
2228		 * right after, we detach the subs, answer goes to cache. */
2229		if(qstate->env->cfg->harden_referral_path)
2230			generate_ns_check(qstate, iq, id);
2231
2232		/* stop current outstanding queries.
2233		 * FIXME: should the outstanding queries be waited for and
2234		 * handled? Say by a subquery that inherits the outbound_entry.
2235		 */
2236		outbound_list_clear(&iq->outlist);
2237		iq->num_current_queries = 0;
2238		fptr_ok(fptr_whitelist_modenv_detach_subs(
2239			qstate->env->detach_subs));
2240		(*qstate->env->detach_subs)(qstate);
2241		iq->num_target_queries = 0;
2242		verbose(VERB_ALGO, "cleared outbound list for next round");
2243		return next_state(iq, QUERYTARGETS_STATE);
2244	} else if(type == RESPONSE_TYPE_CNAME) {
2245		uint8_t* sname = NULL;
2246		size_t snamelen = 0;
2247		/* CNAME type responses get a query restart (i.e., get a
2248		 * reset of the query state and go back to INIT_REQUEST_STATE).
2249		 */
2250		verbose(VERB_DETAIL, "query response was CNAME");
2251		if(verbosity >= VERB_ALGO)
2252			log_dns_msg("cname msg", &iq->response->qinfo,
2253				iq->response->rep);
2254		/* if qtype is DS, check we have the right level of answer,
2255		 * like grandchild answer but we need the middle, reject it */
2256		if(iq->qchase.qtype == LDNS_RR_TYPE_DS && !iq->dsns_point
2257			&& !(iq->chase_flags&BIT_RD)
2258			&& iter_ds_toolow(iq->response, iq->dp)
2259			&& iter_dp_cangodown(&iq->qchase, iq->dp)) {
2260			outbound_list_clear(&iq->outlist);
2261			iq->num_current_queries = 0;
2262			fptr_ok(fptr_whitelist_modenv_detach_subs(
2263				qstate->env->detach_subs));
2264			(*qstate->env->detach_subs)(qstate);
2265			iq->num_target_queries = 0;
2266			return processDSNSFind(qstate, iq, id);
2267		}
2268		/* Process the CNAME response. */
2269		if(!handle_cname_response(qstate, iq, iq->response,
2270			&sname, &snamelen))
2271			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2272		/* cache the CNAME response under the current query */
2273		/* NOTE : set referral=1, so that rrsets get stored but not
2274		 * the partial query answer (CNAME only). */
2275		/* prefetchleeway applied because this updates answer parts */
2276		iter_dns_store(qstate->env, &iq->response->qinfo,
2277			iq->response->rep, 1, qstate->prefetch_leeway,
2278			iq->dp&&iq->dp->has_parent_side_NS, NULL,
2279			qstate->query_flags);
2280		/* set the current request's qname to the new value. */
2281		iq->qchase.qname = sname;
2282		iq->qchase.qname_len = snamelen;
2283		/* Clear the query state, since this is a query restart. */
2284		iq->deleg_msg = NULL;
2285		iq->dp = NULL;
2286		iq->dsns_point = NULL;
2287		/* Note the query restart. */
2288		iq->query_restart_count++;
2289		iq->sent_count = 0;
2290
2291		/* stop current outstanding queries.
2292		 * FIXME: should the outstanding queries be waited for and
2293		 * handled? Say by a subquery that inherits the outbound_entry.
2294		 */
2295		outbound_list_clear(&iq->outlist);
2296		iq->num_current_queries = 0;
2297		fptr_ok(fptr_whitelist_modenv_detach_subs(
2298			qstate->env->detach_subs));
2299		(*qstate->env->detach_subs)(qstate);
2300		iq->num_target_queries = 0;
2301		if(qstate->reply)
2302			sock_list_insert(&qstate->reply_origin,
2303				&qstate->reply->addr, qstate->reply->addrlen,
2304				qstate->region);
2305		verbose(VERB_ALGO, "cleared outbound list for query restart");
2306		/* go to INIT_REQUEST_STATE for new qname. */
2307		return next_state(iq, INIT_REQUEST_STATE);
2308	} else if(type == RESPONSE_TYPE_LAME) {
2309		/* Cache the LAMEness. */
2310		verbose(VERB_DETAIL, "query response was %sLAME",
2311			dnsseclame?"DNSSEC ":"");
2312		if(!dname_subdomain_c(iq->qchase.qname, iq->dp->name)) {
2313			log_err("mark lame: mismatch in qname and dpname");
2314			/* throwaway this reply below */
2315		} else if(qstate->reply) {
2316			/* need addr for lameness cache, but we may have
2317			 * gotten this from cache, so test to be sure */
2318			if(!infra_set_lame(qstate->env->infra_cache,
2319				&qstate->reply->addr, qstate->reply->addrlen,
2320				iq->dp->name, iq->dp->namelen,
2321				*qstate->env->now, dnsseclame, 0,
2322				iq->qchase.qtype))
2323				log_err("mark host lame: out of memory");
2324		}
2325	} else if(type == RESPONSE_TYPE_REC_LAME) {
2326		/* Cache the LAMEness. */
2327		verbose(VERB_DETAIL, "query response REC_LAME: "
2328			"recursive but not authoritative server");
2329		if(!dname_subdomain_c(iq->qchase.qname, iq->dp->name)) {
2330			log_err("mark rec_lame: mismatch in qname and dpname");
2331			/* throwaway this reply below */
2332		} else if(qstate->reply) {
2333			/* need addr for lameness cache, but we may have
2334			 * gotten this from cache, so test to be sure */
2335			verbose(VERB_DETAIL, "mark as REC_LAME");
2336			if(!infra_set_lame(qstate->env->infra_cache,
2337				&qstate->reply->addr, qstate->reply->addrlen,
2338				iq->dp->name, iq->dp->namelen,
2339				*qstate->env->now, 0, 1, iq->qchase.qtype))
2340				log_err("mark host lame: out of memory");
2341		}
2342	} else if(type == RESPONSE_TYPE_THROWAWAY) {
2343		/* LAME and THROWAWAY responses are handled the same way.
2344		 * In this case, the event is just sent directly back to
2345		 * the QUERYTARGETS_STATE without resetting anything,
2346		 * because, clearly, the next target must be tried. */
2347		verbose(VERB_DETAIL, "query response was THROWAWAY");
2348	} else {
2349		log_warn("A query response came back with an unknown type: %d",
2350			(int)type);
2351	}
2352
2353	/* LAME, THROWAWAY and "unknown" all end up here.
2354	 * Recycle to the QUERYTARGETS state to hopefully try a
2355	 * different target. */
2356	return next_state(iq, QUERYTARGETS_STATE);
2357}
2358
2359/**
2360 * Return priming query results to interested super querystates.
2361 *
2362 * Sets the delegation point and delegation message (not nonRD queries).
2363 * This is a callback from walk_supers.
2364 *
2365 * @param qstate: priming query state that finished.
2366 * @param id: module id.
2367 * @param forq: the qstate for which priming has been done.
2368 */
2369static void
2370prime_supers(struct module_qstate* qstate, int id, struct module_qstate* forq)
2371{
2372	struct iter_qstate* foriq = (struct iter_qstate*)forq->minfo[id];
2373	struct delegpt* dp = NULL;
2374
2375	log_assert(qstate->is_priming || foriq->wait_priming_stub);
2376	log_assert(qstate->return_rcode == LDNS_RCODE_NOERROR);
2377	/* Convert our response to a delegation point */
2378	dp = delegpt_from_message(qstate->return_msg, forq->region);
2379	if(!dp) {
2380		/* if there is no convertable delegation point, then
2381		 * the ANSWER type was (presumably) a negative answer. */
2382		verbose(VERB_ALGO, "prime response was not a positive "
2383			"ANSWER; failing");
2384		foriq->dp = NULL;
2385		foriq->state = QUERYTARGETS_STATE;
2386		return;
2387	}
2388
2389	log_query_info(VERB_DETAIL, "priming successful for", &qstate->qinfo);
2390	delegpt_log(VERB_ALGO, dp);
2391	foriq->dp = dp;
2392	foriq->deleg_msg = dns_copy_msg(qstate->return_msg, forq->region);
2393	if(!foriq->deleg_msg) {
2394		log_err("copy prime response: out of memory");
2395		foriq->dp = NULL;
2396		foriq->state = QUERYTARGETS_STATE;
2397		return;
2398	}
2399
2400	/* root priming responses go to init stage 2, priming stub
2401	 * responses to to stage 3. */
2402	if(foriq->wait_priming_stub) {
2403		foriq->state = INIT_REQUEST_3_STATE;
2404		foriq->wait_priming_stub = 0;
2405	} else	foriq->state = INIT_REQUEST_2_STATE;
2406	/* because we are finished, the parent will be reactivated */
2407}
2408
2409/**
2410 * This handles the response to a priming query. This is used to handle both
2411 * root and stub priming responses. This is basically the equivalent of the
2412 * QUERY_RESP_STATE, but will not handle CNAME responses and will treat
2413 * REFERRALs as ANSWERS. It will also update and reactivate the originating
2414 * event.
2415 *
2416 * @param qstate: query state.
2417 * @param id: module id.
2418 * @return true if the event needs more immediate processing, false if not.
2419 *         This state always returns false.
2420 */
2421static int
2422processPrimeResponse(struct module_qstate* qstate, int id)
2423{
2424	struct iter_qstate* iq = (struct iter_qstate*)qstate->minfo[id];
2425	enum response_type type;
2426	iq->response->rep->flags &= ~(BIT_RD|BIT_RA); /* ignore rec-lame */
2427	type = response_type_from_server(
2428		(int)((iq->chase_flags&BIT_RD) || iq->chase_to_rd),
2429		iq->response, &iq->qchase, iq->dp);
2430	if(type == RESPONSE_TYPE_ANSWER) {
2431		qstate->return_rcode = LDNS_RCODE_NOERROR;
2432		qstate->return_msg = iq->response;
2433	} else {
2434		qstate->return_rcode = LDNS_RCODE_SERVFAIL;
2435		qstate->return_msg = NULL;
2436	}
2437
2438	/* validate the root or stub after priming (if enabled).
2439	 * This is the same query as the prime query, but with validation.
2440	 * Now that we are primed, the additional queries that validation
2441	 * may need can be resolved, such as DLV. */
2442	if(qstate->env->cfg->harden_referral_path) {
2443		struct module_qstate* subq = NULL;
2444		log_nametypeclass(VERB_ALGO, "schedule prime validation",
2445			qstate->qinfo.qname, qstate->qinfo.qtype,
2446			qstate->qinfo.qclass);
2447		if(!generate_sub_request(qstate->qinfo.qname,
2448			qstate->qinfo.qname_len, qstate->qinfo.qtype,
2449			qstate->qinfo.qclass, qstate, id, iq,
2450			INIT_REQUEST_STATE, FINISHED_STATE, &subq, 1)) {
2451			verbose(VERB_ALGO, "could not generate prime check");
2452		}
2453		generate_a_aaaa_check(qstate, iq, id);
2454	}
2455
2456	/* This event is finished. */
2457	qstate->ext_state[id] = module_finished;
2458	return 0;
2459}
2460
2461/**
2462 * Do final processing on responses to target queries. Events reach this
2463 * state after the iterative resolution algorithm terminates. This state is
2464 * responsible for reactiving the original event, and housekeeping related
2465 * to received target responses (caching, updating the current delegation
2466 * point, etc).
2467 * Callback from walk_supers for every super state that is interested in
2468 * the results from this query.
2469 *
2470 * @param qstate: query state.
2471 * @param id: module id.
2472 * @param forq: super query state.
2473 */
2474static void
2475processTargetResponse(struct module_qstate* qstate, int id,
2476	struct module_qstate* forq)
2477{
2478	struct iter_qstate* iq = (struct iter_qstate*)qstate->minfo[id];
2479	struct iter_qstate* foriq = (struct iter_qstate*)forq->minfo[id];
2480	struct ub_packed_rrset_key* rrset;
2481	struct delegpt_ns* dpns;
2482	log_assert(qstate->return_rcode == LDNS_RCODE_NOERROR);
2483
2484	foriq->state = QUERYTARGETS_STATE;
2485	log_query_info(VERB_ALGO, "processTargetResponse", &qstate->qinfo);
2486	log_query_info(VERB_ALGO, "processTargetResponse super", &forq->qinfo);
2487
2488	/* check to see if parent event is still interested (in orig name).  */
2489	if(!foriq->dp) {
2490		verbose(VERB_ALGO, "subq: parent not interested, was reset");
2491		return; /* not interested anymore */
2492	}
2493	dpns = delegpt_find_ns(foriq->dp, qstate->qinfo.qname,
2494			qstate->qinfo.qname_len);
2495	if(!dpns) {
2496		/* If not interested, just stop processing this event */
2497		verbose(VERB_ALGO, "subq: parent not interested anymore");
2498		/* could be because parent was jostled out of the cache,
2499		   and a new identical query arrived, that does not want it*/
2500		return;
2501	}
2502
2503	/* Tell the originating event that this target query has finished
2504	 * (regardless if it succeeded or not). */
2505	foriq->num_target_queries--;
2506
2507	/* if iq->query_for_pside_glue then add the pside_glue (marked lame) */
2508	if(iq->pside_glue) {
2509		/* if the pside_glue is NULL, then it could not be found,
2510		 * the done_pside is already set when created and a cache
2511		 * entry created in processFinished so nothing to do here */
2512		log_rrset_key(VERB_ALGO, "add parentside glue to dp",
2513			iq->pside_glue);
2514		if(!delegpt_add_rrset(foriq->dp, forq->region,
2515			iq->pside_glue, 1))
2516			log_err("out of memory adding pside glue");
2517	}
2518
2519	/* This response is relevant to the current query, so we
2520	 * add (attempt to add, anyway) this target(s) and reactivate
2521	 * the original event.
2522	 * NOTE: we could only look for the AnswerRRset if the
2523	 * response type was ANSWER. */
2524	rrset = reply_find_answer_rrset(&iq->qchase, qstate->return_msg->rep);
2525	if(rrset) {
2526		/* if CNAMEs have been followed - add new NS to delegpt. */
2527		/* BTW. RFC 1918 says NS should not have got CNAMEs. Robust. */
2528		if(!delegpt_find_ns(foriq->dp, rrset->rk.dname,
2529			rrset->rk.dname_len)) {
2530			/* if dpns->lame then set newcname ns lame too */
2531			if(!delegpt_add_ns(foriq->dp, forq->region,
2532				rrset->rk.dname, dpns->lame))
2533				log_err("out of memory adding cnamed-ns");
2534		}
2535		/* if dpns->lame then set the address(es) lame too */
2536		if(!delegpt_add_rrset(foriq->dp, forq->region, rrset,
2537			dpns->lame))
2538			log_err("out of memory adding targets");
2539		verbose(VERB_ALGO, "added target response");
2540		delegpt_log(VERB_ALGO, foriq->dp);
2541	} else {
2542		verbose(VERB_ALGO, "iterator TargetResponse failed");
2543		dpns->resolved = 1; /* fail the target */
2544	}
2545}
2546
2547/**
2548 * Process response for DS NS Find queries, that attempt to find the delegation
2549 * point where we ask the DS query from.
2550 *
2551 * @param qstate: query state.
2552 * @param id: module id.
2553 * @param forq: super query state.
2554 */
2555static void
2556processDSNSResponse(struct module_qstate* qstate, int id,
2557	struct module_qstate* forq)
2558{
2559	struct iter_qstate* foriq = (struct iter_qstate*)forq->minfo[id];
2560
2561	/* if the finished (iq->response) query has no NS set: continue
2562	 * up to look for the right dp; nothing to change, do DPNSstate */
2563	if(qstate->return_rcode != LDNS_RCODE_NOERROR)
2564		return; /* seek further */
2565	/* find the NS RRset (without allowing CNAMEs) */
2566	if(!reply_find_rrset(qstate->return_msg->rep, qstate->qinfo.qname,
2567		qstate->qinfo.qname_len, LDNS_RR_TYPE_NS,
2568		qstate->qinfo.qclass)){
2569		return; /* seek further */
2570	}
2571
2572	/* else, store as DP and continue at querytargets */
2573	foriq->state = QUERYTARGETS_STATE;
2574	foriq->dp = delegpt_from_message(qstate->return_msg, forq->region);
2575	if(!foriq->dp) {
2576		log_err("out of memory in dsns dp alloc");
2577		return; /* dp==NULL in QUERYTARGETS makes SERVFAIL */
2578	}
2579	/* success, go query the querytargets in the new dp (and go down) */
2580}
2581
2582/**
2583 * Process response for qclass=ANY queries for a particular class.
2584 * Append to result or error-exit.
2585 *
2586 * @param qstate: query state.
2587 * @param id: module id.
2588 * @param forq: super query state.
2589 */
2590static void
2591processClassResponse(struct module_qstate* qstate, int id,
2592	struct module_qstate* forq)
2593{
2594	struct iter_qstate* foriq = (struct iter_qstate*)forq->minfo[id];
2595	struct dns_msg* from = qstate->return_msg;
2596	log_query_info(VERB_ALGO, "processClassResponse", &qstate->qinfo);
2597	log_query_info(VERB_ALGO, "processClassResponse super", &forq->qinfo);
2598	if(qstate->return_rcode != LDNS_RCODE_NOERROR) {
2599		/* cause servfail for qclass ANY query */
2600		foriq->response = NULL;
2601		foriq->state = FINISHED_STATE;
2602		return;
2603	}
2604	/* append result */
2605	if(!foriq->response) {
2606		/* allocate the response: copy RCODE, sec_state */
2607		foriq->response = dns_copy_msg(from, forq->region);
2608		if(!foriq->response) {
2609			log_err("malloc failed for qclass ANY response");
2610			foriq->state = FINISHED_STATE;
2611			return;
2612		}
2613		foriq->response->qinfo.qclass = forq->qinfo.qclass;
2614		/* qclass ANY does not receive the AA flag on replies */
2615		foriq->response->rep->authoritative = 0;
2616	} else {
2617		struct dns_msg* to = foriq->response;
2618		/* add _from_ this response _to_ existing collection */
2619		/* if there are records, copy RCODE */
2620		/* lower sec_state if this message is lower */
2621		if(from->rep->rrset_count != 0) {
2622			size_t n = from->rep->rrset_count+to->rep->rrset_count;
2623			struct ub_packed_rrset_key** dest, **d;
2624			/* copy appropriate rcode */
2625			to->rep->flags = from->rep->flags;
2626			/* copy rrsets */
2627			if(from->rep->rrset_count > RR_COUNT_MAX ||
2628				to->rep->rrset_count > RR_COUNT_MAX) {
2629				log_err("malloc failed (too many rrsets) in collect ANY");
2630				foriq->state = FINISHED_STATE;
2631				return; /* integer overflow protection */
2632			}
2633			dest = regional_alloc(forq->region, sizeof(dest[0])*n);
2634			if(!dest) {
2635				log_err("malloc failed in collect ANY");
2636				foriq->state = FINISHED_STATE;
2637				return;
2638			}
2639			d = dest;
2640			/* copy AN */
2641			memcpy(dest, to->rep->rrsets, to->rep->an_numrrsets
2642				* sizeof(dest[0]));
2643			dest += to->rep->an_numrrsets;
2644			memcpy(dest, from->rep->rrsets, from->rep->an_numrrsets
2645				* sizeof(dest[0]));
2646			dest += from->rep->an_numrrsets;
2647			/* copy NS */
2648			memcpy(dest, to->rep->rrsets+to->rep->an_numrrsets,
2649				to->rep->ns_numrrsets * sizeof(dest[0]));
2650			dest += to->rep->ns_numrrsets;
2651			memcpy(dest, from->rep->rrsets+from->rep->an_numrrsets,
2652				from->rep->ns_numrrsets * sizeof(dest[0]));
2653			dest += from->rep->ns_numrrsets;
2654			/* copy AR */
2655			memcpy(dest, to->rep->rrsets+to->rep->an_numrrsets+
2656				to->rep->ns_numrrsets,
2657				to->rep->ar_numrrsets * sizeof(dest[0]));
2658			dest += to->rep->ar_numrrsets;
2659			memcpy(dest, from->rep->rrsets+from->rep->an_numrrsets+
2660				from->rep->ns_numrrsets,
2661				from->rep->ar_numrrsets * sizeof(dest[0]));
2662			/* update counts */
2663			to->rep->rrsets = d;
2664			to->rep->an_numrrsets += from->rep->an_numrrsets;
2665			to->rep->ns_numrrsets += from->rep->ns_numrrsets;
2666			to->rep->ar_numrrsets += from->rep->ar_numrrsets;
2667			to->rep->rrset_count = n;
2668		}
2669		if(from->rep->security < to->rep->security) /* lowest sec */
2670			to->rep->security = from->rep->security;
2671		if(from->rep->qdcount != 0) /* insert qd if appropriate */
2672			to->rep->qdcount = from->rep->qdcount;
2673		if(from->rep->ttl < to->rep->ttl) /* use smallest TTL */
2674			to->rep->ttl = from->rep->ttl;
2675		if(from->rep->prefetch_ttl < to->rep->prefetch_ttl)
2676			to->rep->prefetch_ttl = from->rep->prefetch_ttl;
2677	}
2678	/* are we done? */
2679	foriq->num_current_queries --;
2680	if(foriq->num_current_queries == 0)
2681		foriq->state = FINISHED_STATE;
2682}
2683
2684/**
2685 * Collect class ANY responses and make them into one response.  This
2686 * state is started and it creates queries for all classes (that have
2687 * root hints).  The answers are then collected.
2688 *
2689 * @param qstate: query state.
2690 * @param id: module id.
2691 * @return true if the event needs more immediate processing, false if not.
2692 */
2693static int
2694processCollectClass(struct module_qstate* qstate, int id)
2695{
2696	struct iter_qstate* iq = (struct iter_qstate*)qstate->minfo[id];
2697	struct module_qstate* subq;
2698	/* If qchase.qclass == 0 then send out queries for all classes.
2699	 * Otherwise, do nothing (wait for all answers to arrive and the
2700	 * processClassResponse to put them together, and that moves us
2701	 * towards the Finished state when done. */
2702	if(iq->qchase.qclass == 0) {
2703		uint16_t c = 0;
2704		iq->qchase.qclass = LDNS_RR_CLASS_ANY;
2705		while(iter_get_next_root(qstate->env->hints,
2706			qstate->env->fwds, &c)) {
2707			/* generate query for this class */
2708			log_nametypeclass(VERB_ALGO, "spawn collect query",
2709				qstate->qinfo.qname, qstate->qinfo.qtype, c);
2710			if(!generate_sub_request(qstate->qinfo.qname,
2711				qstate->qinfo.qname_len, qstate->qinfo.qtype,
2712				c, qstate, id, iq, INIT_REQUEST_STATE,
2713				FINISHED_STATE, &subq,
2714				(int)!(qstate->query_flags&BIT_CD))) {
2715				return error_response(qstate, id,
2716					LDNS_RCODE_SERVFAIL);
2717			}
2718			/* ignore subq, no special init required */
2719			iq->num_current_queries ++;
2720			if(c == 0xffff)
2721				break;
2722			else c++;
2723		}
2724		/* if no roots are configured at all, return */
2725		if(iq->num_current_queries == 0) {
2726			verbose(VERB_ALGO, "No root hints or fwds, giving up "
2727				"on qclass ANY");
2728			return error_response(qstate, id, LDNS_RCODE_REFUSED);
2729		}
2730		/* return false, wait for queries to return */
2731	}
2732	/* if woke up here because of an answer, wait for more answers */
2733	return 0;
2734}
2735
2736/**
2737 * This handles the final state for first-tier responses (i.e., responses to
2738 * externally generated queries).
2739 *
2740 * @param qstate: query state.
2741 * @param iq: iterator query state.
2742 * @param id: module id.
2743 * @return true if the event needs more processing, false if not. Since this
2744 *         is the final state for an event, it always returns false.
2745 */
2746static int
2747processFinished(struct module_qstate* qstate, struct iter_qstate* iq,
2748	int id)
2749{
2750	log_query_info(VERB_QUERY, "finishing processing for",
2751		&qstate->qinfo);
2752
2753	/* store negative cache element for parent side glue. */
2754	if(iq->query_for_pside_glue && !iq->pside_glue)
2755		iter_store_parentside_neg(qstate->env, &qstate->qinfo,
2756			iq->deleg_msg?iq->deleg_msg->rep:
2757			(iq->response?iq->response->rep:NULL));
2758	if(!iq->response) {
2759		verbose(VERB_ALGO, "No response is set, servfail");
2760		return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2761	}
2762
2763	/* Make sure that the RA flag is set (since the presence of
2764	 * this module means that recursion is available) */
2765	iq->response->rep->flags |= BIT_RA;
2766
2767	/* Clear the AA flag */
2768	/* FIXME: does this action go here or in some other module? */
2769	iq->response->rep->flags &= ~BIT_AA;
2770
2771	/* make sure QR flag is on */
2772	iq->response->rep->flags |= BIT_QR;
2773
2774	/* we have finished processing this query */
2775	qstate->ext_state[id] = module_finished;
2776
2777	/* TODO:  we are using a private TTL, trim the response. */
2778	/* if (mPrivateTTL > 0){IterUtils.setPrivateTTL(resp, mPrivateTTL); } */
2779
2780	/* prepend any items we have accumulated */
2781	if(iq->an_prepend_list || iq->ns_prepend_list) {
2782		if(!iter_prepend(iq, iq->response, qstate->region)) {
2783			log_err("prepend rrsets: out of memory");
2784			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2785		}
2786		/* reset the query name back */
2787		iq->response->qinfo = qstate->qinfo;
2788		/* the security state depends on the combination */
2789		iq->response->rep->security = sec_status_unchecked;
2790		/* store message with the finished prepended items,
2791		 * but only if we did recursion. The nonrecursion referral
2792		 * from cache does not need to be stored in the msg cache. */
2793		if(qstate->query_flags&BIT_RD) {
2794			iter_dns_store(qstate->env, &qstate->qinfo,
2795				iq->response->rep, 0, qstate->prefetch_leeway,
2796				iq->dp&&iq->dp->has_parent_side_NS,
2797				qstate->region, qstate->query_flags);
2798		}
2799	}
2800	qstate->return_rcode = LDNS_RCODE_NOERROR;
2801	qstate->return_msg = iq->response;
2802	return 0;
2803}
2804
2805/*
2806 * Return priming query results to interestes super querystates.
2807 *
2808 * Sets the delegation point and delegation message (not nonRD queries).
2809 * This is a callback from walk_supers.
2810 *
2811 * @param qstate: query state that finished.
2812 * @param id: module id.
2813 * @param super: the qstate to inform.
2814 */
2815void
2816iter_inform_super(struct module_qstate* qstate, int id,
2817	struct module_qstate* super)
2818{
2819	if(!qstate->is_priming && super->qinfo.qclass == LDNS_RR_CLASS_ANY)
2820		processClassResponse(qstate, id, super);
2821	else if(super->qinfo.qtype == LDNS_RR_TYPE_DS && ((struct iter_qstate*)
2822		super->minfo[id])->state == DSNS_FIND_STATE)
2823		processDSNSResponse(qstate, id, super);
2824	else if(qstate->return_rcode != LDNS_RCODE_NOERROR)
2825		error_supers(qstate, id, super);
2826	else if(qstate->is_priming)
2827		prime_supers(qstate, id, super);
2828	else	processTargetResponse(qstate, id, super);
2829}
2830
2831/**
2832 * Handle iterator state.
2833 * Handle events. This is the real processing loop for events, responsible
2834 * for moving events through the various states. If a processing method
2835 * returns true, then it will be advanced to the next state. If false, then
2836 * processing will stop.
2837 *
2838 * @param qstate: query state.
2839 * @param ie: iterator shared global environment.
2840 * @param iq: iterator query state.
2841 * @param id: module id.
2842 */
2843static void
2844iter_handle(struct module_qstate* qstate, struct iter_qstate* iq,
2845	struct iter_env* ie, int id)
2846{
2847	int cont = 1;
2848	while(cont) {
2849		verbose(VERB_ALGO, "iter_handle processing q with state %s",
2850			iter_state_to_string(iq->state));
2851		switch(iq->state) {
2852			case INIT_REQUEST_STATE:
2853				cont = processInitRequest(qstate, iq, ie, id);
2854				break;
2855			case INIT_REQUEST_2_STATE:
2856				cont = processInitRequest2(qstate, iq, id);
2857				break;
2858			case INIT_REQUEST_3_STATE:
2859				cont = processInitRequest3(qstate, iq, id);
2860				break;
2861			case QUERYTARGETS_STATE:
2862				cont = processQueryTargets(qstate, iq, ie, id);
2863				break;
2864			case QUERY_RESP_STATE:
2865				cont = processQueryResponse(qstate, iq, id);
2866				break;
2867			case PRIME_RESP_STATE:
2868				cont = processPrimeResponse(qstate, id);
2869				break;
2870			case COLLECT_CLASS_STATE:
2871				cont = processCollectClass(qstate, id);
2872				break;
2873			case DSNS_FIND_STATE:
2874				cont = processDSNSFind(qstate, iq, id);
2875				break;
2876			case FINISHED_STATE:
2877				cont = processFinished(qstate, iq, id);
2878				break;
2879			default:
2880				log_warn("iterator: invalid state: %d",
2881					iq->state);
2882				cont = 0;
2883				break;
2884		}
2885	}
2886}
2887
2888/**
2889 * This is the primary entry point for processing request events. Note that
2890 * this method should only be used by external modules.
2891 * @param qstate: query state.
2892 * @param ie: iterator shared global environment.
2893 * @param iq: iterator query state.
2894 * @param id: module id.
2895 */
2896static void
2897process_request(struct module_qstate* qstate, struct iter_qstate* iq,
2898	struct iter_env* ie, int id)
2899{
2900	/* external requests start in the INIT state, and finish using the
2901	 * FINISHED state. */
2902	iq->state = INIT_REQUEST_STATE;
2903	iq->final_state = FINISHED_STATE;
2904	verbose(VERB_ALGO, "process_request: new external request event");
2905	iter_handle(qstate, iq, ie, id);
2906}
2907
2908/** process authoritative server reply */
2909static void
2910process_response(struct module_qstate* qstate, struct iter_qstate* iq,
2911	struct iter_env* ie, int id, struct outbound_entry* outbound,
2912	enum module_ev event)
2913{
2914	struct msg_parse* prs;
2915	struct edns_data edns;
2916	sldns_buffer* pkt;
2917
2918	verbose(VERB_ALGO, "process_response: new external response event");
2919	iq->response = NULL;
2920	iq->state = QUERY_RESP_STATE;
2921	if(event == module_event_noreply || event == module_event_error) {
2922		if(event == module_event_noreply && iq->sent_count >= 3 &&
2923			qstate->env->cfg->use_caps_bits_for_id &&
2924			!iq->caps_fallback) {
2925			/* start fallback */
2926			iq->caps_fallback = 1;
2927			iq->caps_server = 0;
2928			iq->caps_reply = NULL;
2929			iq->caps_response = NULL;
2930			iq->state = QUERYTARGETS_STATE;
2931			iq->num_current_queries--;
2932			/* need fresh attempts for the 0x20 fallback, if
2933			 * that was the cause for the failure */
2934			iter_dec_attempts(iq->dp, 3);
2935			verbose(VERB_DETAIL, "Capsforid: timeouts, starting fallback");
2936			goto handle_it;
2937		}
2938		goto handle_it;
2939	}
2940	if( (event != module_event_reply && event != module_event_capsfail)
2941		|| !qstate->reply) {
2942		log_err("Bad event combined with response");
2943		outbound_list_remove(&iq->outlist, outbound);
2944		(void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2945		return;
2946	}
2947
2948	/* parse message */
2949	prs = (struct msg_parse*)regional_alloc(qstate->env->scratch,
2950		sizeof(struct msg_parse));
2951	if(!prs) {
2952		log_err("out of memory on incoming message");
2953		/* like packet got dropped */
2954		goto handle_it;
2955	}
2956	memset(prs, 0, sizeof(*prs));
2957	memset(&edns, 0, sizeof(edns));
2958	pkt = qstate->reply->c->buffer;
2959	sldns_buffer_set_position(pkt, 0);
2960	if(parse_packet(pkt, prs, qstate->env->scratch) != LDNS_RCODE_NOERROR) {
2961		verbose(VERB_ALGO, "parse error on reply packet");
2962		goto handle_it;
2963	}
2964	/* edns is not examined, but removed from message to help cache */
2965	if(parse_extract_edns(prs, &edns) != LDNS_RCODE_NOERROR)
2966		goto handle_it;
2967	/* remove CD-bit, we asked for in case we handle validation ourself */
2968	prs->flags &= ~BIT_CD;
2969
2970	/* normalize and sanitize: easy to delete items from linked lists */
2971	if(!scrub_message(pkt, prs, &iq->qchase, iq->dp->name,
2972		qstate->env->scratch, qstate->env, ie)) {
2973		/* if 0x20 enabled, start fallback, but we have no message */
2974		if(event == module_event_capsfail && !iq->caps_fallback) {
2975			iq->caps_fallback = 1;
2976			iq->caps_server = 0;
2977			iq->caps_reply = NULL;
2978			iq->caps_response = NULL;
2979			iq->state = QUERYTARGETS_STATE;
2980			iq->num_current_queries--;
2981			verbose(VERB_DETAIL, "Capsforid: scrub failed, starting fallback with no response");
2982		}
2983		goto handle_it;
2984	}
2985
2986	/* allocate response dns_msg in region */
2987	iq->response = dns_alloc_msg(pkt, prs, qstate->region);
2988	if(!iq->response)
2989		goto handle_it;
2990	log_query_info(VERB_DETAIL, "response for", &qstate->qinfo);
2991	log_name_addr(VERB_DETAIL, "reply from", iq->dp->name,
2992		&qstate->reply->addr, qstate->reply->addrlen);
2993	if(verbosity >= VERB_ALGO)
2994		log_dns_msg("incoming scrubbed packet:", &iq->response->qinfo,
2995			iq->response->rep);
2996
2997	if(event == module_event_capsfail || iq->caps_fallback) {
2998		/* for fallback we care about main answer, not additionals */
2999		/* removing that makes comparison more likely to succeed */
3000		caps_strip_reply(iq->response->rep);
3001		if(!iq->caps_fallback) {
3002			/* start fallback */
3003			iq->caps_fallback = 1;
3004			iq->caps_server = 0;
3005			iq->caps_reply = iq->response->rep;
3006			iq->caps_response = iq->response;
3007			iq->state = QUERYTARGETS_STATE;
3008			iq->num_current_queries--;
3009			verbose(VERB_DETAIL, "Capsforid: starting fallback");
3010			goto handle_it;
3011		} else {
3012			/* check if reply is the same, otherwise, fail */
3013			if(!iq->caps_reply) {
3014				iq->caps_reply = iq->response->rep;
3015				iq->caps_response = iq->response;
3016				iq->caps_server = -1; /*become zero at ++,
3017				so that we start the full set of trials */
3018			} else if(caps_failed_rcode(iq->caps_reply) &&
3019				!caps_failed_rcode(iq->response->rep)) {
3020				/* prefer to upgrade to non-SERVFAIL */
3021				iq->caps_reply = iq->response->rep;
3022				iq->caps_response = iq->response;
3023			} else if(!caps_failed_rcode(iq->caps_reply) &&
3024				caps_failed_rcode(iq->response->rep)) {
3025				/* if we have non-SERVFAIL as answer then
3026				 * we can ignore SERVFAILs for the equality
3027				 * comparison */
3028				/* no instructions here, skip other else */
3029			} else if(caps_failed_rcode(iq->caps_reply) &&
3030				caps_failed_rcode(iq->response->rep)) {
3031				/* failure is same as other failure in fallbk*/
3032				/* no instructions here, skip other else */
3033			} else if(!reply_equal(iq->response->rep, iq->caps_reply,
3034				qstate->env->scratch)) {
3035				verbose(VERB_DETAIL, "Capsforid fallback: "
3036					"getting different replies, failed");
3037				outbound_list_remove(&iq->outlist, outbound);
3038				(void)error_response(qstate, id,
3039					LDNS_RCODE_SERVFAIL);
3040				return;
3041			}
3042			/* continue the fallback procedure at next server */
3043			iq->caps_server++;
3044			iq->state = QUERYTARGETS_STATE;
3045			iq->num_current_queries--;
3046			verbose(VERB_DETAIL, "Capsforid: reply is equal. "
3047				"go to next fallback");
3048			goto handle_it;
3049		}
3050	}
3051	iq->caps_fallback = 0; /* if we were in fallback, 0x20 is OK now */
3052
3053handle_it:
3054	outbound_list_remove(&iq->outlist, outbound);
3055	iter_handle(qstate, iq, ie, id);
3056}
3057
3058void
3059iter_operate(struct module_qstate* qstate, enum module_ev event, int id,
3060	struct outbound_entry* outbound)
3061{
3062	struct iter_env* ie = (struct iter_env*)qstate->env->modinfo[id];
3063	struct iter_qstate* iq = (struct iter_qstate*)qstate->minfo[id];
3064	verbose(VERB_QUERY, "iterator[module %d] operate: extstate:%s event:%s",
3065		id, strextstate(qstate->ext_state[id]), strmodulevent(event));
3066	if(iq) log_query_info(VERB_QUERY, "iterator operate: query",
3067		&qstate->qinfo);
3068	if(iq && qstate->qinfo.qname != iq->qchase.qname)
3069		log_query_info(VERB_QUERY, "iterator operate: chased to",
3070			&iq->qchase);
3071
3072	/* perform iterator state machine */
3073	if((event == module_event_new || event == module_event_pass) &&
3074		iq == NULL) {
3075		if(!iter_new(qstate, id)) {
3076			(void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3077			return;
3078		}
3079		iq = (struct iter_qstate*)qstate->minfo[id];
3080		process_request(qstate, iq, ie, id);
3081		return;
3082	}
3083	if(iq && event == module_event_pass) {
3084		iter_handle(qstate, iq, ie, id);
3085		return;
3086	}
3087	if(iq && outbound) {
3088		process_response(qstate, iq, ie, id, outbound, event);
3089		return;
3090	}
3091	if(event == module_event_error) {
3092		verbose(VERB_ALGO, "got called with event error, giving up");
3093		(void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3094		return;
3095	}
3096
3097	log_err("bad event for iterator");
3098	(void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3099}
3100
3101void
3102iter_clear(struct module_qstate* qstate, int id)
3103{
3104	struct iter_qstate* iq;
3105	if(!qstate)
3106		return;
3107	iq = (struct iter_qstate*)qstate->minfo[id];
3108	if(iq) {
3109		outbound_list_clear(&iq->outlist);
3110		if(iq->target_count && --iq->target_count[0] == 0)
3111			free(iq->target_count);
3112		iq->num_current_queries = 0;
3113	}
3114	qstate->minfo[id] = NULL;
3115}
3116
3117size_t
3118iter_get_mem(struct module_env* env, int id)
3119{
3120	struct iter_env* ie = (struct iter_env*)env->modinfo[id];
3121	if(!ie)
3122		return 0;
3123	return sizeof(*ie) + sizeof(int)*((size_t)ie->max_dependency_depth+1)
3124		+ donotq_get_mem(ie->donotq) + priv_get_mem(ie->priv);
3125}
3126
3127/**
3128 * The iterator function block
3129 */
3130static struct module_func_block iter_block = {
3131	"iterator",
3132	&iter_init, &iter_deinit, &iter_operate, &iter_inform_super,
3133	&iter_clear, &iter_get_mem
3134};
3135
3136struct module_func_block*
3137iter_get_funcblock(void)
3138{
3139	return &iter_block;
3140}
3141
3142const char*
3143iter_state_to_string(enum iter_state state)
3144{
3145	switch (state)
3146	{
3147	case INIT_REQUEST_STATE :
3148		return "INIT REQUEST STATE";
3149	case INIT_REQUEST_2_STATE :
3150		return "INIT REQUEST STATE (stage 2)";
3151	case INIT_REQUEST_3_STATE:
3152		return "INIT REQUEST STATE (stage 3)";
3153	case QUERYTARGETS_STATE :
3154		return "QUERY TARGETS STATE";
3155	case PRIME_RESP_STATE :
3156		return "PRIME RESPONSE STATE";
3157	case COLLECT_CLASS_STATE :
3158		return "COLLECT CLASS STATE";
3159	case DSNS_FIND_STATE :
3160		return "DSNS FIND STATE";
3161	case QUERY_RESP_STATE :
3162		return "QUERY RESPONSE STATE";
3163	case FINISHED_STATE :
3164		return "FINISHED RESPONSE STATE";
3165	default :
3166		return "UNKNOWN ITER STATE";
3167	}
3168}
3169
3170int
3171iter_state_is_responsestate(enum iter_state s)
3172{
3173	switch(s) {
3174		case INIT_REQUEST_STATE :
3175		case INIT_REQUEST_2_STATE :
3176		case INIT_REQUEST_3_STATE :
3177		case QUERYTARGETS_STATE :
3178		case COLLECT_CLASS_STATE :
3179			return 0;
3180		default:
3181			break;
3182	}
3183	return 1;
3184}
3185