xref: /freebsd/contrib/unbound/iterator/iterator.c (revision b9f654b163bce26de79705e77b872427c9f2afa1)
1 /*
2  * iterator/iterator.c - iterative resolver DNS query response module
3  *
4  * Copyright (c) 2007, NLnet Labs. All rights reserved.
5  *
6  * This software is open source.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  *
12  * Redistributions of source code must retain the above copyright notice,
13  * this list of conditions and the following disclaimer.
14  *
15  * Redistributions in binary form must reproduce the above copyright notice,
16  * this list of conditions and the following disclaimer in the documentation
17  * and/or other materials provided with the distribution.
18  *
19  * Neither the name of the NLNET LABS nor the names of its contributors may
20  * be used to endorse or promote products derived from this software without
21  * specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
26  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
27  * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
28  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
29  * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
30  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
31  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
32  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34  */
35 
36 /**
37  * \file
38  *
39  * This file contains a module that performs recursive iterative DNS query
40  * processing.
41  */
42 
43 #include "config.h"
44 #include "iterator/iterator.h"
45 #include "iterator/iter_utils.h"
46 #include "iterator/iter_hints.h"
47 #include "iterator/iter_fwd.h"
48 #include "iterator/iter_donotq.h"
49 #include "iterator/iter_delegpt.h"
50 #include "iterator/iter_resptype.h"
51 #include "iterator/iter_scrub.h"
52 #include "iterator/iter_priv.h"
53 #include "validator/val_neg.h"
54 #include "services/cache/dns.h"
55 #include "services/cache/infra.h"
56 #include "services/authzone.h"
57 #include "util/module.h"
58 #include "util/netevent.h"
59 #include "util/net_help.h"
60 #include "util/regional.h"
61 #include "util/data/dname.h"
62 #include "util/data/msgencode.h"
63 #include "util/fptr_wlist.h"
64 #include "util/config_file.h"
65 #include "util/random.h"
66 #include "sldns/rrdef.h"
67 #include "sldns/wire2str.h"
68 #include "sldns/str2wire.h"
69 #include "sldns/parseutil.h"
70 #include "sldns/sbuffer.h"
71 
72 int
73 iter_init(struct module_env* env, int id)
74 {
75 	struct iter_env* iter_env = (struct iter_env*)calloc(1,
76 		sizeof(struct iter_env));
77 	if(!iter_env) {
78 		log_err("malloc failure");
79 		return 0;
80 	}
81 	env->modinfo[id] = (void*)iter_env;
82 
83 	lock_basic_init(&iter_env->queries_ratelimit_lock);
84 	lock_protect(&iter_env->queries_ratelimit_lock,
85 			&iter_env->num_queries_ratelimited,
86 		sizeof(iter_env->num_queries_ratelimited));
87 
88 	if(!iter_apply_cfg(iter_env, env->cfg)) {
89 		log_err("iterator: could not apply configuration settings.");
90 		return 0;
91 	}
92 
93 	return 1;
94 }
95 
96 /** delete caps_whitelist element */
97 static void
98 caps_free(struct rbnode_type* n, void* ATTR_UNUSED(d))
99 {
100 	if(n) {
101 		free(((struct name_tree_node*)n)->name);
102 		free(n);
103 	}
104 }
105 
106 void
107 iter_deinit(struct module_env* env, int id)
108 {
109 	struct iter_env* iter_env;
110 	if(!env || !env->modinfo[id])
111 		return;
112 	iter_env = (struct iter_env*)env->modinfo[id];
113 	lock_basic_destroy(&iter_env->queries_ratelimit_lock);
114 	free(iter_env->target_fetch_policy);
115 	priv_delete(iter_env->priv);
116 	donotq_delete(iter_env->donotq);
117 	if(iter_env->caps_white) {
118 		traverse_postorder(iter_env->caps_white, caps_free, NULL);
119 		free(iter_env->caps_white);
120 	}
121 	free(iter_env);
122 	env->modinfo[id] = NULL;
123 }
124 
125 /** new query for iterator */
126 static int
127 iter_new(struct module_qstate* qstate, int id)
128 {
129 	struct iter_qstate* iq = (struct iter_qstate*)regional_alloc(
130 		qstate->region, sizeof(struct iter_qstate));
131 	qstate->minfo[id] = iq;
132 	if(!iq)
133 		return 0;
134 	memset(iq, 0, sizeof(*iq));
135 	iq->state = INIT_REQUEST_STATE;
136 	iq->final_state = FINISHED_STATE;
137 	iq->an_prepend_list = NULL;
138 	iq->an_prepend_last = NULL;
139 	iq->ns_prepend_list = NULL;
140 	iq->ns_prepend_last = NULL;
141 	iq->dp = NULL;
142 	iq->depth = 0;
143 	iq->num_target_queries = 0;
144 	iq->num_current_queries = 0;
145 	iq->query_restart_count = 0;
146 	iq->referral_count = 0;
147 	iq->sent_count = 0;
148 	iq->ratelimit_ok = 0;
149 	iq->target_count = NULL;
150 	iq->wait_priming_stub = 0;
151 	iq->refetch_glue = 0;
152 	iq->dnssec_expected = 0;
153 	iq->dnssec_lame_query = 0;
154 	iq->chase_flags = qstate->query_flags;
155 	/* Start with the (current) qname. */
156 	iq->qchase = qstate->qinfo;
157 	outbound_list_init(&iq->outlist);
158 	iq->minimise_count = 0;
159 	iq->minimise_timeout_count = 0;
160 	if (qstate->env->cfg->qname_minimisation)
161 		iq->minimisation_state = INIT_MINIMISE_STATE;
162 	else
163 		iq->minimisation_state = DONOT_MINIMISE_STATE;
164 
165 	memset(&iq->qinfo_out, 0, sizeof(struct query_info));
166 	return 1;
167 }
168 
169 /**
170  * Transition to the next state. This can be used to advance a currently
171  * processing event. It cannot be used to reactivate a forEvent.
172  *
173  * @param iq: iterator query state
174  * @param nextstate The state to transition to.
175  * @return true. This is so this can be called as the return value for the
176  *         actual process*State() methods. (Transitioning to the next state
177  *         implies further processing).
178  */
179 static int
180 next_state(struct iter_qstate* iq, enum iter_state nextstate)
181 {
182 	/* If transitioning to a "response" state, make sure that there is a
183 	 * response */
184 	if(iter_state_is_responsestate(nextstate)) {
185 		if(iq->response == NULL) {
186 			log_err("transitioning to response state sans "
187 				"response.");
188 		}
189 	}
190 	iq->state = nextstate;
191 	return 1;
192 }
193 
194 /**
195  * Transition an event to its final state. Final states always either return
196  * a result up the module chain, or reactivate a dependent event. Which
197  * final state to transition to is set in the module state for the event when
198  * it was created, and depends on the original purpose of the event.
199  *
200  * The response is stored in the qstate->buf buffer.
201  *
202  * @param iq: iterator query state
203  * @return false. This is so this method can be used as the return value for
204  *         the processState methods. (Transitioning to the final state
205  */
206 static int
207 final_state(struct iter_qstate* iq)
208 {
209 	return next_state(iq, iq->final_state);
210 }
211 
212 /**
213  * Callback routine to handle errors in parent query states
214  * @param qstate: query state that failed.
215  * @param id: module id.
216  * @param super: super state.
217  */
218 static void
219 error_supers(struct module_qstate* qstate, int id, struct module_qstate* super)
220 {
221 	struct iter_qstate* super_iq = (struct iter_qstate*)super->minfo[id];
222 
223 	if(qstate->qinfo.qtype == LDNS_RR_TYPE_A ||
224 		qstate->qinfo.qtype == LDNS_RR_TYPE_AAAA) {
225 		/* mark address as failed. */
226 		struct delegpt_ns* dpns = NULL;
227 		super_iq->num_target_queries--;
228 		if(super_iq->dp)
229 			dpns = delegpt_find_ns(super_iq->dp,
230 				qstate->qinfo.qname, qstate->qinfo.qname_len);
231 		if(!dpns) {
232 			/* not interested */
233 			/* this can happen, for eg. qname minimisation asked
234 			 * for an NXDOMAIN to be validated, and used qtype
235 			 * A for that, and the error of that, the name, is
236 			 * not listed in super_iq->dp */
237 			verbose(VERB_ALGO, "subq error, but not interested");
238 			log_query_info(VERB_ALGO, "superq", &super->qinfo);
239 			return;
240 		} else {
241 			/* see if the failure did get (parent-lame) info */
242 			if(!cache_fill_missing(super->env, super_iq->qchase.qclass,
243 				super->region, super_iq->dp))
244 				log_err("out of memory adding missing");
245 		}
246 		dpns->resolved = 1; /* mark as failed */
247 	}
248 	if(qstate->qinfo.qtype == LDNS_RR_TYPE_NS) {
249 		/* prime failed to get delegation */
250 		super_iq->dp = NULL;
251 	}
252 	/* evaluate targets again */
253 	super_iq->state = QUERYTARGETS_STATE;
254 	/* super becomes runnable, and will process this change */
255 }
256 
257 /**
258  * Return an error to the client
259  * @param qstate: our query state
260  * @param id: module id
261  * @param rcode: error code (DNS errcode).
262  * @return: 0 for use by caller, to make notation easy, like:
263  * 	return error_response(..).
264  */
265 static int
266 error_response(struct module_qstate* qstate, int id, int rcode)
267 {
268 	verbose(VERB_QUERY, "return error response %s",
269 		sldns_lookup_by_id(sldns_rcodes, rcode)?
270 		sldns_lookup_by_id(sldns_rcodes, rcode)->name:"??");
271 	qstate->return_rcode = rcode;
272 	qstate->return_msg = NULL;
273 	qstate->ext_state[id] = module_finished;
274 	return 0;
275 }
276 
277 /**
278  * Return an error to the client and cache the error code in the
279  * message cache (so per qname, qtype, qclass).
280  * @param qstate: our query state
281  * @param id: module id
282  * @param rcode: error code (DNS errcode).
283  * @return: 0 for use by caller, to make notation easy, like:
284  * 	return error_response(..).
285  */
286 static int
287 error_response_cache(struct module_qstate* qstate, int id, int rcode)
288 {
289 	if(!qstate->no_cache_store) {
290 		/* store in cache */
291 		struct reply_info err;
292 		if(qstate->prefetch_leeway > NORR_TTL) {
293 			verbose(VERB_ALGO, "error response for prefetch in cache");
294 			/* attempt to adjust the cache entry prefetch */
295 			if(dns_cache_prefetch_adjust(qstate->env, &qstate->qinfo,
296 				NORR_TTL, qstate->query_flags))
297 				return error_response(qstate, id, rcode);
298 			/* if that fails (not in cache), fall through to store err */
299 		}
300 		if(qstate->env->cfg->serve_expired) {
301 			/* if serving expired contents, and such content is
302 			 * already available, don't overwrite this servfail */
303 			struct msgreply_entry* msg;
304 			if((msg=msg_cache_lookup(qstate->env,
305 				qstate->qinfo.qname, qstate->qinfo.qname_len,
306 				qstate->qinfo.qtype, qstate->qinfo.qclass,
307 				qstate->query_flags, 0,
308 				qstate->env->cfg->serve_expired_ttl_reset))
309 				!= NULL) {
310 				if(qstate->env->cfg->serve_expired_ttl_reset) {
311 					struct reply_info* rep =
312 						(struct reply_info*)msg->entry.data;
313 					if(rep && *qstate->env->now +
314 						qstate->env->cfg->serve_expired_ttl  >
315 						rep->serve_expired_ttl) {
316 						rep->serve_expired_ttl =
317 							*qstate->env->now +
318 							qstate->env->cfg->serve_expired_ttl;
319 					}
320 				}
321 				lock_rw_unlock(&msg->entry.lock);
322 				return error_response(qstate, id, rcode);
323 			}
324 			/* serving expired contents, but nothing is cached
325 			 * at all, so the servfail cache entry is useful
326 			 * (stops waste of time on this servfail NORR_TTL) */
327 		}
328 		memset(&err, 0, sizeof(err));
329 		err.flags = (uint16_t)(BIT_QR | BIT_RA);
330 		FLAGS_SET_RCODE(err.flags, rcode);
331 		err.qdcount = 1;
332 		err.ttl = NORR_TTL;
333 		err.prefetch_ttl = PREFETCH_TTL_CALC(err.ttl);
334 		err.serve_expired_ttl = NORR_TTL;
335 		/* do not waste time trying to validate this servfail */
336 		err.security = sec_status_indeterminate;
337 		verbose(VERB_ALGO, "store error response in message cache");
338 		iter_dns_store(qstate->env, &qstate->qinfo, &err, 0, 0, 0, NULL,
339 			qstate->query_flags);
340 	}
341 	return error_response(qstate, id, rcode);
342 }
343 
344 /** check if prepend item is duplicate item */
345 static int
346 prepend_is_duplicate(struct ub_packed_rrset_key** sets, size_t to,
347 	struct ub_packed_rrset_key* dup)
348 {
349 	size_t i;
350 	for(i=0; i<to; i++) {
351 		if(sets[i]->rk.type == dup->rk.type &&
352 			sets[i]->rk.rrset_class == dup->rk.rrset_class &&
353 			sets[i]->rk.dname_len == dup->rk.dname_len &&
354 			query_dname_compare(sets[i]->rk.dname, dup->rk.dname)
355 			== 0)
356 			return 1;
357 	}
358 	return 0;
359 }
360 
361 /** prepend the prepend list in the answer and authority section of dns_msg */
362 static int
363 iter_prepend(struct iter_qstate* iq, struct dns_msg* msg,
364 	struct regional* region)
365 {
366 	struct iter_prep_list* p;
367 	struct ub_packed_rrset_key** sets;
368 	size_t num_an = 0, num_ns = 0;;
369 	for(p = iq->an_prepend_list; p; p = p->next)
370 		num_an++;
371 	for(p = iq->ns_prepend_list; p; p = p->next)
372 		num_ns++;
373 	if(num_an + num_ns == 0)
374 		return 1;
375 	verbose(VERB_ALGO, "prepending %d rrsets", (int)num_an + (int)num_ns);
376 	if(num_an > RR_COUNT_MAX || num_ns > RR_COUNT_MAX ||
377 		msg->rep->rrset_count > RR_COUNT_MAX) return 0; /* overflow */
378 	sets = regional_alloc(region, (num_an+num_ns+msg->rep->rrset_count) *
379 		sizeof(struct ub_packed_rrset_key*));
380 	if(!sets)
381 		return 0;
382 	/* ANSWER section */
383 	num_an = 0;
384 	for(p = iq->an_prepend_list; p; p = p->next) {
385 		sets[num_an++] = p->rrset;
386 	}
387 	memcpy(sets+num_an, msg->rep->rrsets, msg->rep->an_numrrsets *
388 		sizeof(struct ub_packed_rrset_key*));
389 	/* AUTH section */
390 	num_ns = 0;
391 	for(p = iq->ns_prepend_list; p; p = p->next) {
392 		if(prepend_is_duplicate(sets+msg->rep->an_numrrsets+num_an,
393 			num_ns, p->rrset) || prepend_is_duplicate(
394 			msg->rep->rrsets+msg->rep->an_numrrsets,
395 			msg->rep->ns_numrrsets, p->rrset))
396 			continue;
397 		sets[msg->rep->an_numrrsets + num_an + num_ns++] = p->rrset;
398 	}
399 	memcpy(sets + num_an + msg->rep->an_numrrsets + num_ns,
400 		msg->rep->rrsets + msg->rep->an_numrrsets,
401 		(msg->rep->ns_numrrsets + msg->rep->ar_numrrsets) *
402 		sizeof(struct ub_packed_rrset_key*));
403 
404 	/* NXDOMAIN rcode can stay if we prepended DNAME/CNAMEs, because
405 	 * this is what recursors should give. */
406 	msg->rep->rrset_count += num_an + num_ns;
407 	msg->rep->an_numrrsets += num_an;
408 	msg->rep->ns_numrrsets += num_ns;
409 	msg->rep->rrsets = sets;
410 	return 1;
411 }
412 
413 /**
414  * Find rrset in ANSWER prepend list.
415  * to avoid duplicate DNAMEs when a DNAME is traversed twice.
416  * @param iq: iterator query state.
417  * @param rrset: rrset to add.
418  * @return false if not found
419  */
420 static int
421 iter_find_rrset_in_prepend_answer(struct iter_qstate* iq,
422 	struct ub_packed_rrset_key* rrset)
423 {
424 	struct iter_prep_list* p = iq->an_prepend_list;
425 	while(p) {
426 		if(ub_rrset_compare(p->rrset, rrset) == 0 &&
427 			rrsetdata_equal((struct packed_rrset_data*)p->rrset
428 			->entry.data, (struct packed_rrset_data*)rrset
429 			->entry.data))
430 			return 1;
431 		p = p->next;
432 	}
433 	return 0;
434 }
435 
436 /**
437  * Add rrset to ANSWER prepend list
438  * @param qstate: query state.
439  * @param iq: iterator query state.
440  * @param rrset: rrset to add.
441  * @return false on failure (malloc).
442  */
443 static int
444 iter_add_prepend_answer(struct module_qstate* qstate, struct iter_qstate* iq,
445 	struct ub_packed_rrset_key* rrset)
446 {
447 	struct iter_prep_list* p = (struct iter_prep_list*)regional_alloc(
448 		qstate->region, sizeof(struct iter_prep_list));
449 	if(!p)
450 		return 0;
451 	p->rrset = rrset;
452 	p->next = NULL;
453 	/* add at end */
454 	if(iq->an_prepend_last)
455 		iq->an_prepend_last->next = p;
456 	else	iq->an_prepend_list = p;
457 	iq->an_prepend_last = p;
458 	return 1;
459 }
460 
461 /**
462  * Add rrset to AUTHORITY prepend list
463  * @param qstate: query state.
464  * @param iq: iterator query state.
465  * @param rrset: rrset to add.
466  * @return false on failure (malloc).
467  */
468 static int
469 iter_add_prepend_auth(struct module_qstate* qstate, struct iter_qstate* iq,
470 	struct ub_packed_rrset_key* rrset)
471 {
472 	struct iter_prep_list* p = (struct iter_prep_list*)regional_alloc(
473 		qstate->region, sizeof(struct iter_prep_list));
474 	if(!p)
475 		return 0;
476 	p->rrset = rrset;
477 	p->next = NULL;
478 	/* add at end */
479 	if(iq->ns_prepend_last)
480 		iq->ns_prepend_last->next = p;
481 	else	iq->ns_prepend_list = p;
482 	iq->ns_prepend_last = p;
483 	return 1;
484 }
485 
486 /**
487  * Given a CNAME response (defined as a response containing a CNAME or DNAME
488  * that does not answer the request), process the response, modifying the
489  * state as necessary. This follows the CNAME/DNAME chain and returns the
490  * final query name.
491  *
492  * sets the new query name, after following the CNAME/DNAME chain.
493  * @param qstate: query state.
494  * @param iq: iterator query state.
495  * @param msg: the response.
496  * @param mname: returned target new query name.
497  * @param mname_len: length of mname.
498  * @return false on (malloc) error.
499  */
500 static int
501 handle_cname_response(struct module_qstate* qstate, struct iter_qstate* iq,
502         struct dns_msg* msg, uint8_t** mname, size_t* mname_len)
503 {
504 	size_t i;
505 	/* Start with the (current) qname. */
506 	*mname = iq->qchase.qname;
507 	*mname_len = iq->qchase.qname_len;
508 
509 	/* Iterate over the ANSWER rrsets in order, looking for CNAMEs and
510 	 * DNAMES. */
511 	for(i=0; i<msg->rep->an_numrrsets; i++) {
512 		struct ub_packed_rrset_key* r = msg->rep->rrsets[i];
513 		/* If there is a (relevant) DNAME, add it to the list.
514 		 * We always expect there to be CNAME that was generated
515 		 * by this DNAME following, so we don't process the DNAME
516 		 * directly.  */
517 		if(ntohs(r->rk.type) == LDNS_RR_TYPE_DNAME &&
518 			dname_strict_subdomain_c(*mname, r->rk.dname) &&
519 			!iter_find_rrset_in_prepend_answer(iq, r)) {
520 			if(!iter_add_prepend_answer(qstate, iq, r))
521 				return 0;
522 			continue;
523 		}
524 
525 		if(ntohs(r->rk.type) == LDNS_RR_TYPE_CNAME &&
526 			query_dname_compare(*mname, r->rk.dname) == 0 &&
527 			!iter_find_rrset_in_prepend_answer(iq, r)) {
528 			/* Add this relevant CNAME rrset to the prepend list.*/
529 			if(!iter_add_prepend_answer(qstate, iq, r))
530 				return 0;
531 			get_cname_target(r, mname, mname_len);
532 		}
533 
534 		/* Other rrsets in the section are ignored. */
535 	}
536 	/* add authority rrsets to authority prepend, for wildcarded CNAMEs */
537 	for(i=msg->rep->an_numrrsets; i<msg->rep->an_numrrsets +
538 		msg->rep->ns_numrrsets; i++) {
539 		struct ub_packed_rrset_key* r = msg->rep->rrsets[i];
540 		/* only add NSEC/NSEC3, as they may be needed for validation */
541 		if(ntohs(r->rk.type) == LDNS_RR_TYPE_NSEC ||
542 			ntohs(r->rk.type) == LDNS_RR_TYPE_NSEC3) {
543 			if(!iter_add_prepend_auth(qstate, iq, r))
544 				return 0;
545 		}
546 	}
547 	return 1;
548 }
549 
550 /** see if last resort is possible - does config allow queries to parent */
551 static int
552 can_have_last_resort(struct module_env* env, uint8_t* nm, size_t nmlen,
553 	uint16_t qclass, struct delegpt** retdp)
554 {
555 	struct delegpt* fwddp;
556 	struct iter_hints_stub* stub;
557 	int labs = dname_count_labels(nm);
558 	/* do not process a last resort (the parent side) if a stub
559 	 * or forward is configured, because we do not want to go 'above'
560 	 * the configured servers */
561 	if(!dname_is_root(nm) && (stub = (struct iter_hints_stub*)
562 		name_tree_find(&env->hints->tree, nm, nmlen, labs, qclass)) &&
563 		/* has_parent side is turned off for stub_first, where we
564 		 * are allowed to go to the parent */
565 		stub->dp->has_parent_side_NS) {
566 		if(retdp) *retdp = stub->dp;
567 		return 0;
568 	}
569 	if((fwddp = forwards_find(env->fwds, nm, qclass)) &&
570 		/* has_parent_side is turned off for forward_first, where
571 		 * we are allowed to go to the parent */
572 		fwddp->has_parent_side_NS) {
573 		if(retdp) *retdp = fwddp;
574 		return 0;
575 	}
576 	return 1;
577 }
578 
579 /** see if target name is caps-for-id whitelisted */
580 static int
581 is_caps_whitelisted(struct iter_env* ie, struct iter_qstate* iq)
582 {
583 	if(!ie->caps_white) return 0; /* no whitelist, or no capsforid */
584 	return name_tree_lookup(ie->caps_white, iq->qchase.qname,
585 		iq->qchase.qname_len, dname_count_labels(iq->qchase.qname),
586 		iq->qchase.qclass) != NULL;
587 }
588 
589 /** create target count structure for this query */
590 static void
591 target_count_create(struct iter_qstate* iq)
592 {
593 	if(!iq->target_count) {
594 		iq->target_count = (int*)calloc(2, sizeof(int));
595 		/* if calloc fails we simply do not track this number */
596 		if(iq->target_count)
597 			iq->target_count[0] = 1;
598 	}
599 }
600 
601 static void
602 target_count_increase(struct iter_qstate* iq, int num)
603 {
604 	target_count_create(iq);
605 	if(iq->target_count)
606 		iq->target_count[1] += num;
607 }
608 
609 /**
610  * Generate a subrequest.
611  * Generate a local request event. Local events are tied to this module, and
612  * have a corresponding (first tier) event that is waiting for this event to
613  * resolve to continue.
614  *
615  * @param qname The query name for this request.
616  * @param qnamelen length of qname
617  * @param qtype The query type for this request.
618  * @param qclass The query class for this request.
619  * @param qstate The event that is generating this event.
620  * @param id: module id.
621  * @param iq: The iterator state that is generating this event.
622  * @param initial_state The initial response state (normally this
623  *          is QUERY_RESP_STATE, unless it is known that the request won't
624  *          need iterative processing
625  * @param finalstate The final state for the response to this request.
626  * @param subq_ret: if newly allocated, the subquerystate, or NULL if it does
627  * 	not need initialisation.
628  * @param v: if true, validation is done on the subquery.
629  * @return false on error (malloc).
630  */
631 static int
632 generate_sub_request(uint8_t* qname, size_t qnamelen, uint16_t qtype,
633 	uint16_t qclass, struct module_qstate* qstate, int id,
634 	struct iter_qstate* iq, enum iter_state initial_state,
635 	enum iter_state finalstate, struct module_qstate** subq_ret, int v)
636 {
637 	struct module_qstate* subq = NULL;
638 	struct iter_qstate* subiq = NULL;
639 	uint16_t qflags = 0; /* OPCODE QUERY, no flags */
640 	struct query_info qinf;
641 	int prime = (finalstate == PRIME_RESP_STATE)?1:0;
642 	int valrec = 0;
643 	qinf.qname = qname;
644 	qinf.qname_len = qnamelen;
645 	qinf.qtype = qtype;
646 	qinf.qclass = qclass;
647 	qinf.local_alias = NULL;
648 
649 	/* RD should be set only when sending the query back through the INIT
650 	 * state. */
651 	if(initial_state == INIT_REQUEST_STATE)
652 		qflags |= BIT_RD;
653 	/* We set the CD flag so we can send this through the "head" of
654 	 * the resolution chain, which might have a validator. We are
655 	 * uninterested in validating things not on the direct resolution
656 	 * path.  */
657 	if(!v) {
658 		qflags |= BIT_CD;
659 		valrec = 1;
660 	}
661 
662 	/* attach subquery, lookup existing or make a new one */
663 	fptr_ok(fptr_whitelist_modenv_attach_sub(qstate->env->attach_sub));
664 	if(!(*qstate->env->attach_sub)(qstate, &qinf, qflags, prime, valrec,
665 		&subq)) {
666 		return 0;
667 	}
668 	*subq_ret = subq;
669 	if(subq) {
670 		/* initialise the new subquery */
671 		subq->curmod = id;
672 		subq->ext_state[id] = module_state_initial;
673 		subq->minfo[id] = regional_alloc(subq->region,
674 			sizeof(struct iter_qstate));
675 		if(!subq->minfo[id]) {
676 			log_err("init subq: out of memory");
677 			fptr_ok(fptr_whitelist_modenv_kill_sub(
678 				qstate->env->kill_sub));
679 			(*qstate->env->kill_sub)(subq);
680 			return 0;
681 		}
682 		subiq = (struct iter_qstate*)subq->minfo[id];
683 		memset(subiq, 0, sizeof(*subiq));
684 		subiq->num_target_queries = 0;
685 		target_count_create(iq);
686 		subiq->target_count = iq->target_count;
687 		if(iq->target_count)
688 			iq->target_count[0] ++; /* extra reference */
689 		subiq->num_current_queries = 0;
690 		subiq->depth = iq->depth+1;
691 		outbound_list_init(&subiq->outlist);
692 		subiq->state = initial_state;
693 		subiq->final_state = finalstate;
694 		subiq->qchase = subq->qinfo;
695 		subiq->chase_flags = subq->query_flags;
696 		subiq->refetch_glue = 0;
697 		if(qstate->env->cfg->qname_minimisation)
698 			subiq->minimisation_state = INIT_MINIMISE_STATE;
699 		else
700 			subiq->minimisation_state = DONOT_MINIMISE_STATE;
701 		memset(&subiq->qinfo_out, 0, sizeof(struct query_info));
702 	}
703 	return 1;
704 }
705 
706 /**
707  * Generate and send a root priming request.
708  * @param qstate: the qtstate that triggered the need to prime.
709  * @param iq: iterator query state.
710  * @param id: module id.
711  * @param qclass: the class to prime.
712  * @return 0 on failure
713  */
714 static int
715 prime_root(struct module_qstate* qstate, struct iter_qstate* iq, int id,
716 	uint16_t qclass)
717 {
718 	struct delegpt* dp;
719 	struct module_qstate* subq;
720 	verbose(VERB_DETAIL, "priming . %s NS",
721 		sldns_lookup_by_id(sldns_rr_classes, (int)qclass)?
722 		sldns_lookup_by_id(sldns_rr_classes, (int)qclass)->name:"??");
723 	dp = hints_lookup_root(qstate->env->hints, qclass);
724 	if(!dp) {
725 		verbose(VERB_ALGO, "Cannot prime due to lack of hints");
726 		return 0;
727 	}
728 	/* Priming requests start at the QUERYTARGETS state, skipping
729 	 * the normal INIT state logic (which would cause an infloop). */
730 	if(!generate_sub_request((uint8_t*)"\000", 1, LDNS_RR_TYPE_NS,
731 		qclass, qstate, id, iq, QUERYTARGETS_STATE, PRIME_RESP_STATE,
732 		&subq, 0)) {
733 		verbose(VERB_ALGO, "could not prime root");
734 		return 0;
735 	}
736 	if(subq) {
737 		struct iter_qstate* subiq =
738 			(struct iter_qstate*)subq->minfo[id];
739 		/* Set the initial delegation point to the hint.
740 		 * copy dp, it is now part of the root prime query.
741 		 * dp was part of in the fixed hints structure. */
742 		subiq->dp = delegpt_copy(dp, subq->region);
743 		if(!subiq->dp) {
744 			log_err("out of memory priming root, copydp");
745 			fptr_ok(fptr_whitelist_modenv_kill_sub(
746 				qstate->env->kill_sub));
747 			(*qstate->env->kill_sub)(subq);
748 			return 0;
749 		}
750 		/* there should not be any target queries. */
751 		subiq->num_target_queries = 0;
752 		subiq->dnssec_expected = iter_indicates_dnssec(
753 			qstate->env, subiq->dp, NULL, subq->qinfo.qclass);
754 	}
755 
756 	/* this module stops, our submodule starts, and does the query. */
757 	qstate->ext_state[id] = module_wait_subquery;
758 	return 1;
759 }
760 
761 /**
762  * Generate and process a stub priming request. This method tests for the
763  * need to prime a stub zone, so it is safe to call for every request.
764  *
765  * @param qstate: the qtstate that triggered the need to prime.
766  * @param iq: iterator query state.
767  * @param id: module id.
768  * @param qname: request name.
769  * @param qclass: request class.
770  * @return true if a priming subrequest was made, false if not. The will only
771  *         issue a priming request if it detects an unprimed stub.
772  *         Uses value of 2 to signal during stub-prime in root-prime situation
773  *         that a noprime-stub is available and resolution can continue.
774  */
775 static int
776 prime_stub(struct module_qstate* qstate, struct iter_qstate* iq, int id,
777 	uint8_t* qname, uint16_t qclass)
778 {
779 	/* Lookup the stub hint. This will return null if the stub doesn't
780 	 * need to be re-primed. */
781 	struct iter_hints_stub* stub;
782 	struct delegpt* stub_dp;
783 	struct module_qstate* subq;
784 
785 	if(!qname) return 0;
786 	stub = hints_lookup_stub(qstate->env->hints, qname, qclass, iq->dp);
787 	/* The stub (if there is one) does not need priming. */
788 	if(!stub)
789 		return 0;
790 	stub_dp = stub->dp;
791 	/* if we have an auth_zone dp, and stub is equal, don't prime stub
792 	 * yet, unless we want to fallback and avoid the auth_zone */
793 	if(!iq->auth_zone_avoid && iq->dp && iq->dp->auth_dp &&
794 		query_dname_compare(iq->dp->name, stub_dp->name) == 0)
795 		return 0;
796 
797 	/* is it a noprime stub (always use) */
798 	if(stub->noprime) {
799 		int r = 0;
800 		if(iq->dp == NULL) r = 2;
801 		/* copy the dp out of the fixed hints structure, so that
802 		 * it can be changed when servicing this query */
803 		iq->dp = delegpt_copy(stub_dp, qstate->region);
804 		if(!iq->dp) {
805 			log_err("out of memory priming stub");
806 			errinf(qstate, "malloc failure, priming stub");
807 			(void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
808 			return 1; /* return 1 to make module stop, with error */
809 		}
810 		log_nametypeclass(VERB_DETAIL, "use stub", stub_dp->name,
811 			LDNS_RR_TYPE_NS, qclass);
812 		return r;
813 	}
814 
815 	/* Otherwise, we need to (re)prime the stub. */
816 	log_nametypeclass(VERB_DETAIL, "priming stub", stub_dp->name,
817 		LDNS_RR_TYPE_NS, qclass);
818 
819 	/* Stub priming events start at the QUERYTARGETS state to avoid the
820 	 * redundant INIT state processing. */
821 	if(!generate_sub_request(stub_dp->name, stub_dp->namelen,
822 		LDNS_RR_TYPE_NS, qclass, qstate, id, iq,
823 		QUERYTARGETS_STATE, PRIME_RESP_STATE, &subq, 0)) {
824 		verbose(VERB_ALGO, "could not prime stub");
825 		errinf(qstate, "could not generate lookup for stub prime");
826 		(void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
827 		return 1; /* return 1 to make module stop, with error */
828 	}
829 	if(subq) {
830 		struct iter_qstate* subiq =
831 			(struct iter_qstate*)subq->minfo[id];
832 
833 		/* Set the initial delegation point to the hint. */
834 		/* make copy to avoid use of stub dp by different qs/threads */
835 		subiq->dp = delegpt_copy(stub_dp, subq->region);
836 		if(!subiq->dp) {
837 			log_err("out of memory priming stub, copydp");
838 			fptr_ok(fptr_whitelist_modenv_kill_sub(
839 				qstate->env->kill_sub));
840 			(*qstate->env->kill_sub)(subq);
841 			errinf(qstate, "malloc failure, in stub prime");
842 			(void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
843 			return 1; /* return 1 to make module stop, with error */
844 		}
845 		/* there should not be any target queries -- although there
846 		 * wouldn't be anyway, since stub hints never have
847 		 * missing targets. */
848 		subiq->num_target_queries = 0;
849 		subiq->wait_priming_stub = 1;
850 		subiq->dnssec_expected = iter_indicates_dnssec(
851 			qstate->env, subiq->dp, NULL, subq->qinfo.qclass);
852 	}
853 
854 	/* this module stops, our submodule starts, and does the query. */
855 	qstate->ext_state[id] = module_wait_subquery;
856 	return 1;
857 }
858 
859 /**
860  * Generate a delegation point for an auth zone (unless cached dp is better)
861  * false on alloc failure.
862  */
863 static int
864 auth_zone_delegpt(struct module_qstate* qstate, struct iter_qstate* iq,
865 	uint8_t* delname, size_t delnamelen)
866 {
867 	struct auth_zone* z;
868 	if(iq->auth_zone_avoid)
869 		return 1;
870 	if(!delname) {
871 		delname = iq->qchase.qname;
872 		delnamelen = iq->qchase.qname_len;
873 	}
874 	lock_rw_rdlock(&qstate->env->auth_zones->lock);
875 	z = auth_zones_find_zone(qstate->env->auth_zones, delname, delnamelen,
876 		qstate->qinfo.qclass);
877 	if(!z) {
878 		lock_rw_unlock(&qstate->env->auth_zones->lock);
879 		return 1;
880 	}
881 	lock_rw_rdlock(&z->lock);
882 	lock_rw_unlock(&qstate->env->auth_zones->lock);
883 	if(z->for_upstream) {
884 		if(iq->dp && query_dname_compare(z->name, iq->dp->name) == 0
885 			&& iq->dp->auth_dp && qstate->blacklist &&
886 			z->fallback_enabled) {
887 			/* cache is blacklisted and fallback, and we
888 			 * already have an auth_zone dp */
889 			if(verbosity>=VERB_ALGO) {
890 				char buf[255+1];
891 				dname_str(z->name, buf);
892 				verbose(VERB_ALGO, "auth_zone %s "
893 				  "fallback because cache blacklisted",
894 				  buf);
895 			}
896 			lock_rw_unlock(&z->lock);
897 			iq->dp = NULL;
898 			return 1;
899 		}
900 		if(iq->dp==NULL || dname_subdomain_c(z->name, iq->dp->name)) {
901 			struct delegpt* dp;
902 			if(qstate->blacklist && z->fallback_enabled) {
903 				/* cache is blacklisted because of a DNSSEC
904 				 * validation failure, and the zone allows
905 				 * fallback to the internet, query there. */
906 				if(verbosity>=VERB_ALGO) {
907 					char buf[255+1];
908 					dname_str(z->name, buf);
909 					verbose(VERB_ALGO, "auth_zone %s "
910 					  "fallback because cache blacklisted",
911 					  buf);
912 				}
913 				lock_rw_unlock(&z->lock);
914 				return 1;
915 			}
916 			dp = (struct delegpt*)regional_alloc_zero(
917 				qstate->region, sizeof(*dp));
918 			if(!dp) {
919 				log_err("alloc failure");
920 				if(z->fallback_enabled) {
921 					lock_rw_unlock(&z->lock);
922 					return 1; /* just fallback */
923 				}
924 				lock_rw_unlock(&z->lock);
925 				errinf(qstate, "malloc failure");
926 				return 0;
927 			}
928 			dp->name = regional_alloc_init(qstate->region,
929 				z->name, z->namelen);
930 			if(!dp->name) {
931 				log_err("alloc failure");
932 				if(z->fallback_enabled) {
933 					lock_rw_unlock(&z->lock);
934 					return 1; /* just fallback */
935 				}
936 				lock_rw_unlock(&z->lock);
937 				errinf(qstate, "malloc failure");
938 				return 0;
939 			}
940 			dp->namelen = z->namelen;
941 			dp->namelabs = z->namelabs;
942 			dp->auth_dp = 1;
943 			iq->dp = dp;
944 		}
945 	}
946 
947 	lock_rw_unlock(&z->lock);
948 	return 1;
949 }
950 
951 /**
952  * Generate A and AAAA checks for glue that is in-zone for the referral
953  * we just got to obtain authoritative information on the addresses.
954  *
955  * @param qstate: the qtstate that triggered the need to prime.
956  * @param iq: iterator query state.
957  * @param id: module id.
958  */
959 static void
960 generate_a_aaaa_check(struct module_qstate* qstate, struct iter_qstate* iq,
961 	int id)
962 {
963 	struct iter_env* ie = (struct iter_env*)qstate->env->modinfo[id];
964 	struct module_qstate* subq;
965 	size_t i;
966 	struct reply_info* rep = iq->response->rep;
967 	struct ub_packed_rrset_key* s;
968 	log_assert(iq->dp);
969 
970 	if(iq->depth == ie->max_dependency_depth)
971 		return;
972 	/* walk through additional, and check if in-zone,
973 	 * only relevant A, AAAA are left after scrub anyway */
974 	for(i=rep->an_numrrsets+rep->ns_numrrsets; i<rep->rrset_count; i++) {
975 		s = rep->rrsets[i];
976 		/* check *ALL* addresses that are transmitted in additional*/
977 		/* is it an address ? */
978 		if( !(ntohs(s->rk.type)==LDNS_RR_TYPE_A ||
979 			ntohs(s->rk.type)==LDNS_RR_TYPE_AAAA)) {
980 			continue;
981 		}
982 		/* is this query the same as the A/AAAA check for it */
983 		if(qstate->qinfo.qtype == ntohs(s->rk.type) &&
984 			qstate->qinfo.qclass == ntohs(s->rk.rrset_class) &&
985 			query_dname_compare(qstate->qinfo.qname,
986 				s->rk.dname)==0 &&
987 			(qstate->query_flags&BIT_RD) &&
988 			!(qstate->query_flags&BIT_CD))
989 			continue;
990 
991 		/* generate subrequest for it */
992 		log_nametypeclass(VERB_ALGO, "schedule addr fetch",
993 			s->rk.dname, ntohs(s->rk.type),
994 			ntohs(s->rk.rrset_class));
995 		if(!generate_sub_request(s->rk.dname, s->rk.dname_len,
996 			ntohs(s->rk.type), ntohs(s->rk.rrset_class),
997 			qstate, id, iq,
998 			INIT_REQUEST_STATE, FINISHED_STATE, &subq, 1)) {
999 			verbose(VERB_ALGO, "could not generate addr check");
1000 			return;
1001 		}
1002 		/* ignore subq - not need for more init */
1003 	}
1004 }
1005 
1006 /**
1007  * Generate a NS check request to obtain authoritative information
1008  * on an NS rrset.
1009  *
1010  * @param qstate: the qtstate that triggered the need to prime.
1011  * @param iq: iterator query state.
1012  * @param id: module id.
1013  */
1014 static void
1015 generate_ns_check(struct module_qstate* qstate, struct iter_qstate* iq, int id)
1016 {
1017 	struct iter_env* ie = (struct iter_env*)qstate->env->modinfo[id];
1018 	struct module_qstate* subq;
1019 	log_assert(iq->dp);
1020 
1021 	if(iq->depth == ie->max_dependency_depth)
1022 		return;
1023 	if(!can_have_last_resort(qstate->env, iq->dp->name, iq->dp->namelen,
1024 		iq->qchase.qclass, NULL))
1025 		return;
1026 	/* is this query the same as the nscheck? */
1027 	if(qstate->qinfo.qtype == LDNS_RR_TYPE_NS &&
1028 		query_dname_compare(iq->dp->name, qstate->qinfo.qname)==0 &&
1029 		(qstate->query_flags&BIT_RD) && !(qstate->query_flags&BIT_CD)){
1030 		/* spawn off A, AAAA queries for in-zone glue to check */
1031 		generate_a_aaaa_check(qstate, iq, id);
1032 		return;
1033 	}
1034 	/* no need to get the NS record for DS, it is above the zonecut */
1035 	if(qstate->qinfo.qtype == LDNS_RR_TYPE_DS)
1036 		return;
1037 
1038 	log_nametypeclass(VERB_ALGO, "schedule ns fetch",
1039 		iq->dp->name, LDNS_RR_TYPE_NS, iq->qchase.qclass);
1040 	if(!generate_sub_request(iq->dp->name, iq->dp->namelen,
1041 		LDNS_RR_TYPE_NS, iq->qchase.qclass, qstate, id, iq,
1042 		INIT_REQUEST_STATE, FINISHED_STATE, &subq, 1)) {
1043 		verbose(VERB_ALGO, "could not generate ns check");
1044 		return;
1045 	}
1046 	if(subq) {
1047 		struct iter_qstate* subiq =
1048 			(struct iter_qstate*)subq->minfo[id];
1049 
1050 		/* make copy to avoid use of stub dp by different qs/threads */
1051 		/* refetch glue to start higher up the tree */
1052 		subiq->refetch_glue = 1;
1053 		subiq->dp = delegpt_copy(iq->dp, subq->region);
1054 		if(!subiq->dp) {
1055 			log_err("out of memory generating ns check, copydp");
1056 			fptr_ok(fptr_whitelist_modenv_kill_sub(
1057 				qstate->env->kill_sub));
1058 			(*qstate->env->kill_sub)(subq);
1059 			return;
1060 		}
1061 	}
1062 }
1063 
1064 /**
1065  * Generate a DNSKEY prefetch query to get the DNSKEY for the DS record we
1066  * just got in a referral (where we have dnssec_expected, thus have trust
1067  * anchors above it).  Note that right after calling this routine the
1068  * iterator detached subqueries (because of following the referral), and thus
1069  * the DNSKEY query becomes detached, its return stored in the cache for
1070  * later lookup by the validator.  This cache lookup by the validator avoids
1071  * the roundtrip incurred by the DNSKEY query.  The DNSKEY query is now
1072  * performed at about the same time the original query is sent to the domain,
1073  * thus the two answers are likely to be returned at about the same time,
1074  * saving a roundtrip from the validated lookup.
1075  *
1076  * @param qstate: the qtstate that triggered the need to prime.
1077  * @param iq: iterator query state.
1078  * @param id: module id.
1079  */
1080 static void
1081 generate_dnskey_prefetch(struct module_qstate* qstate,
1082 	struct iter_qstate* iq, int id)
1083 {
1084 	struct module_qstate* subq;
1085 	log_assert(iq->dp);
1086 
1087 	/* is this query the same as the prefetch? */
1088 	if(qstate->qinfo.qtype == LDNS_RR_TYPE_DNSKEY &&
1089 		query_dname_compare(iq->dp->name, qstate->qinfo.qname)==0 &&
1090 		(qstate->query_flags&BIT_RD) && !(qstate->query_flags&BIT_CD)){
1091 		return;
1092 	}
1093 
1094 	/* if the DNSKEY is in the cache this lookup will stop quickly */
1095 	log_nametypeclass(VERB_ALGO, "schedule dnskey prefetch",
1096 		iq->dp->name, LDNS_RR_TYPE_DNSKEY, iq->qchase.qclass);
1097 	if(!generate_sub_request(iq->dp->name, iq->dp->namelen,
1098 		LDNS_RR_TYPE_DNSKEY, iq->qchase.qclass, qstate, id, iq,
1099 		INIT_REQUEST_STATE, FINISHED_STATE, &subq, 0)) {
1100 		/* we'll be slower, but it'll work */
1101 		verbose(VERB_ALGO, "could not generate dnskey prefetch");
1102 		return;
1103 	}
1104 	if(subq) {
1105 		struct iter_qstate* subiq =
1106 			(struct iter_qstate*)subq->minfo[id];
1107 		/* this qstate has the right delegation for the dnskey lookup*/
1108 		/* make copy to avoid use of stub dp by different qs/threads */
1109 		subiq->dp = delegpt_copy(iq->dp, subq->region);
1110 		/* if !subiq->dp, it'll start from the cache, no problem */
1111 	}
1112 }
1113 
1114 /**
1115  * See if the query needs forwarding.
1116  *
1117  * @param qstate: query state.
1118  * @param iq: iterator query state.
1119  * @return true if the request is forwarded, false if not.
1120  * 	If returns true but, iq->dp is NULL then a malloc failure occurred.
1121  */
1122 static int
1123 forward_request(struct module_qstate* qstate, struct iter_qstate* iq)
1124 {
1125 	struct delegpt* dp;
1126 	uint8_t* delname = iq->qchase.qname;
1127 	size_t delnamelen = iq->qchase.qname_len;
1128 	if(iq->refetch_glue && iq->dp) {
1129 		delname = iq->dp->name;
1130 		delnamelen = iq->dp->namelen;
1131 	}
1132 	/* strip one label off of DS query to lookup higher for it */
1133 	if( (iq->qchase.qtype == LDNS_RR_TYPE_DS || iq->refetch_glue)
1134 		&& !dname_is_root(iq->qchase.qname))
1135 		dname_remove_label(&delname, &delnamelen);
1136 	dp = forwards_lookup(qstate->env->fwds, delname, iq->qchase.qclass);
1137 	if(!dp)
1138 		return 0;
1139 	/* send recursion desired to forward addr */
1140 	iq->chase_flags |= BIT_RD;
1141 	iq->dp = delegpt_copy(dp, qstate->region);
1142 	/* iq->dp checked by caller */
1143 	verbose(VERB_ALGO, "forwarding request");
1144 	return 1;
1145 }
1146 
1147 static int
1148 iter_stub_fwd_no_cache(struct module_qstate *qstate, struct iter_qstate *iq)
1149 {
1150 	struct iter_hints_stub *stub;
1151 	struct delegpt *dp;
1152 
1153 	/* Check for stub. */
1154 	stub = hints_lookup_stub(qstate->env->hints, iq->qchase.qname,
1155 	    iq->qchase.qclass, iq->dp);
1156 	dp = forwards_lookup(qstate->env->fwds, iq->qchase.qname, iq->qchase.qclass);
1157 
1158 	/* see if forward or stub is more pertinent */
1159 	if(stub && stub->dp && dp) {
1160 		if(dname_strict_subdomain(dp->name, dp->namelabs,
1161 			stub->dp->name, stub->dp->namelabs)) {
1162 			stub = NULL; /* ignore stub, forward is lower */
1163 		} else {
1164 			dp = NULL; /* ignore forward, stub is lower */
1165 		}
1166 	}
1167 
1168 	/* check stub */
1169 	if (stub != NULL && stub->dp != NULL) {
1170 		if(stub->dp->no_cache) {
1171 			char qname[255+1];
1172 			char dpname[255+1];
1173 			dname_str(iq->qchase.qname, qname);
1174 			dname_str(stub->dp->name, dpname);
1175 			verbose(VERB_ALGO, "stub for %s %s has no_cache", qname, dpname);
1176 		}
1177 		return (stub->dp->no_cache);
1178 	}
1179 
1180 	/* Check for forward. */
1181 	if (dp) {
1182 		if(dp->no_cache) {
1183 			char qname[255+1];
1184 			char dpname[255+1];
1185 			dname_str(iq->qchase.qname, qname);
1186 			dname_str(dp->name, dpname);
1187 			verbose(VERB_ALGO, "forward for %s %s has no_cache", qname, dpname);
1188 		}
1189 		return (dp->no_cache);
1190 	}
1191 	return 0;
1192 }
1193 
1194 /**
1195  * Process the initial part of the request handling. This state roughly
1196  * corresponds to resolver algorithms steps 1 (find answer in cache) and 2
1197  * (find the best servers to ask).
1198  *
1199  * Note that all requests start here, and query restarts revisit this state.
1200  *
1201  * This state either generates: 1) a response, from cache or error, 2) a
1202  * priming event, or 3) forwards the request to the next state (init2,
1203  * generally).
1204  *
1205  * @param qstate: query state.
1206  * @param iq: iterator query state.
1207  * @param ie: iterator shared global environment.
1208  * @param id: module id.
1209  * @return true if the event needs more request processing immediately,
1210  *         false if not.
1211  */
1212 static int
1213 processInitRequest(struct module_qstate* qstate, struct iter_qstate* iq,
1214 	struct iter_env* ie, int id)
1215 {
1216 	uint8_t* delname;
1217 	size_t delnamelen;
1218 	struct dns_msg* msg = NULL;
1219 
1220 	log_query_info(VERB_DETAIL, "resolving", &qstate->qinfo);
1221 	/* check effort */
1222 
1223 	/* We enforce a maximum number of query restarts. This is primarily a
1224 	 * cheap way to prevent CNAME loops. */
1225 	if(iq->query_restart_count > MAX_RESTART_COUNT) {
1226 		verbose(VERB_QUERY, "request has exceeded the maximum number"
1227 			" of query restarts with %d", iq->query_restart_count);
1228 		errinf(qstate, "request has exceeded the maximum number "
1229 			"restarts (eg. indirections)");
1230 		if(iq->qchase.qname)
1231 			errinf_dname(qstate, "stop at", iq->qchase.qname);
1232 		return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1233 	}
1234 
1235 	/* We enforce a maximum recursion/dependency depth -- in general,
1236 	 * this is unnecessary for dependency loops (although it will
1237 	 * catch those), but it provides a sensible limit to the amount
1238 	 * of work required to answer a given query. */
1239 	verbose(VERB_ALGO, "request has dependency depth of %d", iq->depth);
1240 	if(iq->depth > ie->max_dependency_depth) {
1241 		verbose(VERB_QUERY, "request has exceeded the maximum "
1242 			"dependency depth with depth of %d", iq->depth);
1243 		errinf(qstate, "request has exceeded the maximum dependency "
1244 			"depth (eg. nameserver lookup recursion)");
1245 		return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1246 	}
1247 
1248 	/* If the request is qclass=ANY, setup to generate each class */
1249 	if(qstate->qinfo.qclass == LDNS_RR_CLASS_ANY) {
1250 		iq->qchase.qclass = 0;
1251 		return next_state(iq, COLLECT_CLASS_STATE);
1252 	}
1253 
1254 	/*
1255 	 * If we are restricted by a forward-zone or a stub-zone, we
1256 	 * can't re-fetch glue for this delegation point.
1257 	 * we won’t try to re-fetch glue if the iq->dp is null.
1258 	 */
1259 	if (iq->refetch_glue &&
1260 	        iq->dp &&
1261 	        !can_have_last_resort(qstate->env, iq->dp->name,
1262 	             iq->dp->namelen, iq->qchase.qclass, NULL)) {
1263 	    iq->refetch_glue = 0;
1264 	}
1265 
1266 	/* Resolver Algorithm Step 1 -- Look for the answer in local data. */
1267 
1268 	/* This either results in a query restart (CNAME cache response), a
1269 	 * terminating response (ANSWER), or a cache miss (null). */
1270 
1271 	if (iter_stub_fwd_no_cache(qstate, iq)) {
1272 		/* Asked to not query cache. */
1273 		verbose(VERB_ALGO, "no-cache set, going to the network");
1274 		qstate->no_cache_lookup = 1;
1275 		qstate->no_cache_store = 1;
1276 		msg = NULL;
1277 	} else if(qstate->blacklist) {
1278 		/* if cache, or anything else, was blacklisted then
1279 		 * getting older results from cache is a bad idea, no cache */
1280 		verbose(VERB_ALGO, "cache blacklisted, going to the network");
1281 		msg = NULL;
1282 	} else if(!qstate->no_cache_lookup) {
1283 		msg = dns_cache_lookup(qstate->env, iq->qchase.qname,
1284 			iq->qchase.qname_len, iq->qchase.qtype,
1285 			iq->qchase.qclass, qstate->query_flags,
1286 			qstate->region, qstate->env->scratch, 0);
1287 		if(!msg && qstate->env->neg_cache &&
1288 			iter_qname_indicates_dnssec(qstate->env, &iq->qchase)) {
1289 			/* lookup in negative cache; may result in
1290 			 * NOERROR/NODATA or NXDOMAIN answers that need validation */
1291 			msg = val_neg_getmsg(qstate->env->neg_cache, &iq->qchase,
1292 				qstate->region, qstate->env->rrset_cache,
1293 				qstate->env->scratch_buffer,
1294 				*qstate->env->now, 1/*add SOA*/, NULL,
1295 				qstate->env->cfg);
1296 		}
1297 		/* item taken from cache does not match our query name, thus
1298 		 * security needs to be re-examined later */
1299 		if(msg && query_dname_compare(qstate->qinfo.qname,
1300 			iq->qchase.qname) != 0)
1301 			msg->rep->security = sec_status_unchecked;
1302 	}
1303 	if(msg) {
1304 		/* handle positive cache response */
1305 		enum response_type type = response_type_from_cache(msg,
1306 			&iq->qchase);
1307 		if(verbosity >= VERB_ALGO) {
1308 			log_dns_msg("msg from cache lookup", &msg->qinfo,
1309 				msg->rep);
1310 			verbose(VERB_ALGO, "msg ttl is %d, prefetch ttl %d",
1311 				(int)msg->rep->ttl,
1312 				(int)msg->rep->prefetch_ttl);
1313 		}
1314 
1315 		if(type == RESPONSE_TYPE_CNAME) {
1316 			uint8_t* sname = 0;
1317 			size_t slen = 0;
1318 			verbose(VERB_ALGO, "returning CNAME response from "
1319 				"cache");
1320 			if(!handle_cname_response(qstate, iq, msg,
1321 				&sname, &slen)) {
1322 				errinf(qstate, "failed to prepend CNAME "
1323 					"components, malloc failure");
1324 				return error_response(qstate, id,
1325 					LDNS_RCODE_SERVFAIL);
1326 			}
1327 			iq->qchase.qname = sname;
1328 			iq->qchase.qname_len = slen;
1329 			/* This *is* a query restart, even if it is a cheap
1330 			 * one. */
1331 			iq->dp = NULL;
1332 			iq->refetch_glue = 0;
1333 			iq->query_restart_count++;
1334 			iq->sent_count = 0;
1335 			sock_list_insert(&qstate->reply_origin, NULL, 0, qstate->region);
1336 			if(qstate->env->cfg->qname_minimisation)
1337 				iq->minimisation_state = INIT_MINIMISE_STATE;
1338 			return next_state(iq, INIT_REQUEST_STATE);
1339 		}
1340 
1341 		/* if from cache, NULL, else insert 'cache IP' len=0 */
1342 		if(qstate->reply_origin)
1343 			sock_list_insert(&qstate->reply_origin, NULL, 0, qstate->region);
1344 		if(FLAGS_GET_RCODE(msg->rep->flags) == LDNS_RCODE_SERVFAIL)
1345 			errinf(qstate, "SERVFAIL in cache");
1346 		/* it is an answer, response, to final state */
1347 		verbose(VERB_ALGO, "returning answer from cache.");
1348 		iq->response = msg;
1349 		return final_state(iq);
1350 	}
1351 
1352 	/* attempt to forward the request */
1353 	if(forward_request(qstate, iq))
1354 	{
1355 		if(!iq->dp) {
1356 			log_err("alloc failure for forward dp");
1357 			errinf(qstate, "malloc failure for forward zone");
1358 			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1359 		}
1360 		iq->refetch_glue = 0;
1361 		iq->minimisation_state = DONOT_MINIMISE_STATE;
1362 		/* the request has been forwarded.
1363 		 * forwarded requests need to be immediately sent to the
1364 		 * next state, QUERYTARGETS. */
1365 		return next_state(iq, QUERYTARGETS_STATE);
1366 	}
1367 
1368 	/* Resolver Algorithm Step 2 -- find the "best" servers. */
1369 
1370 	/* first, adjust for DS queries. To avoid the grandparent problem,
1371 	 * we just look for the closest set of server to the parent of qname.
1372 	 * When re-fetching glue we also need to ask the parent.
1373 	 */
1374 	if(iq->refetch_glue) {
1375 		if(!iq->dp) {
1376 			log_err("internal or malloc fail: no dp for refetch");
1377 			errinf(qstate, "malloc failure, for delegation info");
1378 			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1379 		}
1380 		delname = iq->dp->name;
1381 		delnamelen = iq->dp->namelen;
1382 	} else {
1383 		delname = iq->qchase.qname;
1384 		delnamelen = iq->qchase.qname_len;
1385 	}
1386 	if(iq->qchase.qtype == LDNS_RR_TYPE_DS || iq->refetch_glue ||
1387 	   (iq->qchase.qtype == LDNS_RR_TYPE_NS && qstate->prefetch_leeway
1388 	   && can_have_last_resort(qstate->env, delname, delnamelen, iq->qchase.qclass, NULL))) {
1389 		/* remove first label from delname, root goes to hints,
1390 		 * but only to fetch glue, not for qtype=DS. */
1391 		/* also when prefetching an NS record, fetch it again from
1392 		 * its parent, just as if it expired, so that you do not
1393 		 * get stuck on an older nameserver that gives old NSrecords */
1394 		if(dname_is_root(delname) && (iq->refetch_glue ||
1395 			(iq->qchase.qtype == LDNS_RR_TYPE_NS &&
1396 			qstate->prefetch_leeway)))
1397 			delname = NULL; /* go to root priming */
1398 		else 	dname_remove_label(&delname, &delnamelen);
1399 	}
1400 	/* delname is the name to lookup a delegation for. If NULL rootprime */
1401 	while(1) {
1402 
1403 		/* Lookup the delegation in the cache. If null, then the
1404 		 * cache needs to be primed for the qclass. */
1405 		if(delname)
1406 		     iq->dp = dns_cache_find_delegation(qstate->env, delname,
1407 			delnamelen, iq->qchase.qtype, iq->qchase.qclass,
1408 			qstate->region, &iq->deleg_msg,
1409 			*qstate->env->now+qstate->prefetch_leeway);
1410 		else iq->dp = NULL;
1411 
1412 		/* If the cache has returned nothing, then we have a
1413 		 * root priming situation. */
1414 		if(iq->dp == NULL) {
1415 			int r;
1416 			/* if under auth zone, no prime needed */
1417 			if(!auth_zone_delegpt(qstate, iq, delname, delnamelen))
1418 				return error_response(qstate, id,
1419 					LDNS_RCODE_SERVFAIL);
1420 			if(iq->dp) /* use auth zone dp */
1421 				return next_state(iq, INIT_REQUEST_2_STATE);
1422 			/* if there is a stub, then no root prime needed */
1423 			r = prime_stub(qstate, iq, id, delname,
1424 				iq->qchase.qclass);
1425 			if(r == 2)
1426 				break; /* got noprime-stub-zone, continue */
1427 			else if(r)
1428 				return 0; /* stub prime request made */
1429 			if(forwards_lookup_root(qstate->env->fwds,
1430 				iq->qchase.qclass)) {
1431 				/* forward zone root, no root prime needed */
1432 				/* fill in some dp - safety belt */
1433 				iq->dp = hints_lookup_root(qstate->env->hints,
1434 					iq->qchase.qclass);
1435 				if(!iq->dp) {
1436 					log_err("internal error: no hints dp");
1437 					errinf(qstate, "no hints for this class");
1438 					return error_response(qstate, id,
1439 						LDNS_RCODE_SERVFAIL);
1440 				}
1441 				iq->dp = delegpt_copy(iq->dp, qstate->region);
1442 				if(!iq->dp) {
1443 					log_err("out of memory in safety belt");
1444 					errinf(qstate, "malloc failure, in safety belt");
1445 					return error_response(qstate, id,
1446 						LDNS_RCODE_SERVFAIL);
1447 				}
1448 				return next_state(iq, INIT_REQUEST_2_STATE);
1449 			}
1450 			/* Note that the result of this will set a new
1451 			 * DelegationPoint based on the result of priming. */
1452 			if(!prime_root(qstate, iq, id, iq->qchase.qclass))
1453 				return error_response(qstate, id,
1454 					LDNS_RCODE_REFUSED);
1455 
1456 			/* priming creates and sends a subordinate query, with
1457 			 * this query as the parent. So further processing for
1458 			 * this event will stop until reactivated by the
1459 			 * results of priming. */
1460 			return 0;
1461 		}
1462 		if(!iq->ratelimit_ok && qstate->prefetch_leeway)
1463 			iq->ratelimit_ok = 1; /* allow prefetches, this keeps
1464 			otherwise valid data in the cache */
1465 		if(!iq->ratelimit_ok && infra_ratelimit_exceeded(
1466 			qstate->env->infra_cache, iq->dp->name,
1467 			iq->dp->namelen, *qstate->env->now)) {
1468 			/* and increment the rate, so that the rate for time
1469 			 * now will also exceed the rate, keeping cache fresh */
1470 			(void)infra_ratelimit_inc(qstate->env->infra_cache,
1471 				iq->dp->name, iq->dp->namelen,
1472 				*qstate->env->now);
1473 			/* see if we are passed through with slip factor */
1474 			if(qstate->env->cfg->ratelimit_factor != 0 &&
1475 				ub_random_max(qstate->env->rnd,
1476 				    qstate->env->cfg->ratelimit_factor) == 1) {
1477 				iq->ratelimit_ok = 1;
1478 				log_nametypeclass(VERB_ALGO, "ratelimit allowed through for "
1479 					"delegation point", iq->dp->name,
1480 					LDNS_RR_TYPE_NS, LDNS_RR_CLASS_IN);
1481 			} else {
1482 				lock_basic_lock(&ie->queries_ratelimit_lock);
1483 				ie->num_queries_ratelimited++;
1484 				lock_basic_unlock(&ie->queries_ratelimit_lock);
1485 				log_nametypeclass(VERB_ALGO, "ratelimit exceeded with "
1486 					"delegation point", iq->dp->name,
1487 					LDNS_RR_TYPE_NS, LDNS_RR_CLASS_IN);
1488 				qstate->was_ratelimited = 1;
1489 				errinf(qstate, "query was ratelimited");
1490 				errinf_dname(qstate, "for zone", iq->dp->name);
1491 				return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1492 			}
1493 		}
1494 
1495 		/* see if this dp not useless.
1496 		 * It is useless if:
1497 		 *	o all NS items are required glue.
1498 		 *	  or the query is for NS item that is required glue.
1499 		 *	o no addresses are provided.
1500 		 *	o RD qflag is on.
1501 		 * Instead, go up one level, and try to get even further
1502 		 * If the root was useless, use safety belt information.
1503 		 * Only check cache returns, because replies for servers
1504 		 * could be useless but lead to loops (bumping into the
1505 		 * same server reply) if useless-checked.
1506 		 */
1507 		if(iter_dp_is_useless(&qstate->qinfo, qstate->query_flags,
1508 			iq->dp)) {
1509 			struct delegpt* retdp = NULL;
1510 			if(!can_have_last_resort(qstate->env, iq->dp->name, iq->dp->namelen, iq->qchase.qclass, &retdp)) {
1511 				if(retdp) {
1512 					verbose(VERB_QUERY, "cache has stub "
1513 						"or fwd but no addresses, "
1514 						"fallback to config");
1515 					iq->dp = delegpt_copy(retdp,
1516 						qstate->region);
1517 					if(!iq->dp) {
1518 						log_err("out of memory in "
1519 							"stub/fwd fallback");
1520 						errinf(qstate, "malloc failure, for fallback to config");
1521 						return error_response(qstate,
1522 						    id, LDNS_RCODE_SERVFAIL);
1523 					}
1524 					break;
1525 				}
1526 				verbose(VERB_ALGO, "useless dp "
1527 					"but cannot go up, servfail");
1528 				delegpt_log(VERB_ALGO, iq->dp);
1529 				errinf(qstate, "no useful nameservers, "
1530 					"and cannot go up");
1531 				errinf_dname(qstate, "for zone", iq->dp->name);
1532 				return error_response(qstate, id,
1533 					LDNS_RCODE_SERVFAIL);
1534 			}
1535 			if(dname_is_root(iq->dp->name)) {
1536 				/* use safety belt */
1537 				verbose(VERB_QUERY, "Cache has root NS but "
1538 				"no addresses. Fallback to the safety belt.");
1539 				iq->dp = hints_lookup_root(qstate->env->hints,
1540 					iq->qchase.qclass);
1541 				/* note deleg_msg is from previous lookup,
1542 				 * but RD is on, so it is not used */
1543 				if(!iq->dp) {
1544 					log_err("internal error: no hints dp");
1545 					return error_response(qstate, id,
1546 						LDNS_RCODE_REFUSED);
1547 				}
1548 				iq->dp = delegpt_copy(iq->dp, qstate->region);
1549 				if(!iq->dp) {
1550 					log_err("out of memory in safety belt");
1551 					errinf(qstate, "malloc failure, in safety belt, for root");
1552 					return error_response(qstate, id,
1553 						LDNS_RCODE_SERVFAIL);
1554 				}
1555 				break;
1556 			} else {
1557 				verbose(VERB_ALGO,
1558 					"cache delegation was useless:");
1559 				delegpt_log(VERB_ALGO, iq->dp);
1560 				/* go up */
1561 				delname = iq->dp->name;
1562 				delnamelen = iq->dp->namelen;
1563 				dname_remove_label(&delname, &delnamelen);
1564 			}
1565 		} else break;
1566 	}
1567 
1568 	verbose(VERB_ALGO, "cache delegation returns delegpt");
1569 	delegpt_log(VERB_ALGO, iq->dp);
1570 
1571 	/* Otherwise, set the current delegation point and move on to the
1572 	 * next state. */
1573 	return next_state(iq, INIT_REQUEST_2_STATE);
1574 }
1575 
1576 /**
1577  * Process the second part of the initial request handling. This state
1578  * basically exists so that queries that generate root priming events have
1579  * the same init processing as ones that do not. Request events that reach
1580  * this state must have a valid currentDelegationPoint set.
1581  *
1582  * This part is primarily handling stub zone priming. Events that reach this
1583  * state must have a current delegation point.
1584  *
1585  * @param qstate: query state.
1586  * @param iq: iterator query state.
1587  * @param id: module id.
1588  * @return true if the event needs more request processing immediately,
1589  *         false if not.
1590  */
1591 static int
1592 processInitRequest2(struct module_qstate* qstate, struct iter_qstate* iq,
1593 	int id)
1594 {
1595 	uint8_t* delname;
1596 	size_t delnamelen;
1597 	log_query_info(VERB_QUERY, "resolving (init part 2): ",
1598 		&qstate->qinfo);
1599 
1600 	delname = iq->qchase.qname;
1601 	delnamelen = iq->qchase.qname_len;
1602 	if(iq->refetch_glue) {
1603 		struct iter_hints_stub* stub;
1604 		if(!iq->dp) {
1605 			log_err("internal or malloc fail: no dp for refetch");
1606 			errinf(qstate, "malloc failure, no delegation info");
1607 			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1608 		}
1609 		/* Do not send queries above stub, do not set delname to dp if
1610 		 * this is above stub without stub-first. */
1611 		stub = hints_lookup_stub(
1612 			qstate->env->hints, iq->qchase.qname, iq->qchase.qclass,
1613 			iq->dp);
1614 		if(!stub || !stub->dp->has_parent_side_NS ||
1615 			dname_subdomain_c(iq->dp->name, stub->dp->name)) {
1616 			delname = iq->dp->name;
1617 			delnamelen = iq->dp->namelen;
1618 		}
1619 	}
1620 	if(iq->qchase.qtype == LDNS_RR_TYPE_DS || iq->refetch_glue) {
1621 		if(!dname_is_root(delname))
1622 			dname_remove_label(&delname, &delnamelen);
1623 		iq->refetch_glue = 0; /* if CNAME causes restart, no refetch */
1624 	}
1625 
1626 	/* see if we have an auth zone to answer from, improves dp from cache
1627 	 * (if any dp from cache) with auth zone dp, if that is lower */
1628 	if(!auth_zone_delegpt(qstate, iq, delname, delnamelen))
1629 		return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1630 
1631 	/* Check to see if we need to prime a stub zone. */
1632 	if(prime_stub(qstate, iq, id, delname, iq->qchase.qclass)) {
1633 		/* A priming sub request was made */
1634 		return 0;
1635 	}
1636 
1637 	/* most events just get forwarded to the next state. */
1638 	return next_state(iq, INIT_REQUEST_3_STATE);
1639 }
1640 
1641 /**
1642  * Process the third part of the initial request handling. This state exists
1643  * as a separate state so that queries that generate stub priming events
1644  * will get the tail end of the init process but not repeat the stub priming
1645  * check.
1646  *
1647  * @param qstate: query state.
1648  * @param iq: iterator query state.
1649  * @param id: module id.
1650  * @return true, advancing the event to the QUERYTARGETS_STATE.
1651  */
1652 static int
1653 processInitRequest3(struct module_qstate* qstate, struct iter_qstate* iq,
1654 	int id)
1655 {
1656 	log_query_info(VERB_QUERY, "resolving (init part 3): ",
1657 		&qstate->qinfo);
1658 	/* if the cache reply dp equals a validation anchor or msg has DS,
1659 	 * then DNSSEC RRSIGs are expected in the reply */
1660 	iq->dnssec_expected = iter_indicates_dnssec(qstate->env, iq->dp,
1661 		iq->deleg_msg, iq->qchase.qclass);
1662 
1663 	/* If the RD flag wasn't set, then we just finish with the
1664 	 * cached referral as the response. */
1665 	if(!(qstate->query_flags & BIT_RD) && iq->deleg_msg) {
1666 		iq->response = iq->deleg_msg;
1667 		if(verbosity >= VERB_ALGO && iq->response)
1668 			log_dns_msg("no RD requested, using delegation msg",
1669 				&iq->response->qinfo, iq->response->rep);
1670 		if(qstate->reply_origin)
1671 			sock_list_insert(&qstate->reply_origin, NULL, 0, qstate->region);
1672 		return final_state(iq);
1673 	}
1674 	/* After this point, unset the RD flag -- this query is going to
1675 	 * be sent to an auth. server. */
1676 	iq->chase_flags &= ~BIT_RD;
1677 
1678 	/* if dnssec expected, fetch key for the trust-anchor or cached-DS */
1679 	if(iq->dnssec_expected && qstate->env->cfg->prefetch_key &&
1680 		!(qstate->query_flags&BIT_CD)) {
1681 		generate_dnskey_prefetch(qstate, iq, id);
1682 		fptr_ok(fptr_whitelist_modenv_detach_subs(
1683 			qstate->env->detach_subs));
1684 		(*qstate->env->detach_subs)(qstate);
1685 	}
1686 
1687 	/* Jump to the next state. */
1688 	return next_state(iq, QUERYTARGETS_STATE);
1689 }
1690 
1691 /**
1692  * Given a basic query, generate a parent-side "target" query.
1693  * These are subordinate queries for missing delegation point target addresses,
1694  * for which only the parent of the delegation provides correct IP addresses.
1695  *
1696  * @param qstate: query state.
1697  * @param iq: iterator query state.
1698  * @param id: module id.
1699  * @param name: target qname.
1700  * @param namelen: target qname length.
1701  * @param qtype: target qtype (either A or AAAA).
1702  * @param qclass: target qclass.
1703  * @return true on success, false on failure.
1704  */
1705 static int
1706 generate_parentside_target_query(struct module_qstate* qstate,
1707 	struct iter_qstate* iq, int id, uint8_t* name, size_t namelen,
1708 	uint16_t qtype, uint16_t qclass)
1709 {
1710 	struct module_qstate* subq;
1711 	if(!generate_sub_request(name, namelen, qtype, qclass, qstate,
1712 		id, iq, INIT_REQUEST_STATE, FINISHED_STATE, &subq, 0))
1713 		return 0;
1714 	if(subq) {
1715 		struct iter_qstate* subiq =
1716 			(struct iter_qstate*)subq->minfo[id];
1717 		/* blacklist the cache - we want to fetch parent stuff */
1718 		sock_list_insert(&subq->blacklist, NULL, 0, subq->region);
1719 		subiq->query_for_pside_glue = 1;
1720 		if(dname_subdomain_c(name, iq->dp->name)) {
1721 			subiq->dp = delegpt_copy(iq->dp, subq->region);
1722 			subiq->dnssec_expected = iter_indicates_dnssec(
1723 				qstate->env, subiq->dp, NULL,
1724 				subq->qinfo.qclass);
1725 			subiq->refetch_glue = 1;
1726 		} else {
1727 			subiq->dp = dns_cache_find_delegation(qstate->env,
1728 				name, namelen, qtype, qclass, subq->region,
1729 				&subiq->deleg_msg,
1730 				*qstate->env->now+subq->prefetch_leeway);
1731 			/* if no dp, then it's from root, refetch unneeded */
1732 			if(subiq->dp) {
1733 				subiq->dnssec_expected = iter_indicates_dnssec(
1734 					qstate->env, subiq->dp, NULL,
1735 					subq->qinfo.qclass);
1736 				subiq->refetch_glue = 1;
1737 			}
1738 		}
1739 	}
1740 	log_nametypeclass(VERB_QUERY, "new pside target", name, qtype, qclass);
1741 	return 1;
1742 }
1743 
1744 /**
1745  * Given a basic query, generate a "target" query. These are subordinate
1746  * queries for missing delegation point target addresses.
1747  *
1748  * @param qstate: query state.
1749  * @param iq: iterator query state.
1750  * @param id: module id.
1751  * @param name: target qname.
1752  * @param namelen: target qname length.
1753  * @param qtype: target qtype (either A or AAAA).
1754  * @param qclass: target qclass.
1755  * @return true on success, false on failure.
1756  */
1757 static int
1758 generate_target_query(struct module_qstate* qstate, struct iter_qstate* iq,
1759         int id, uint8_t* name, size_t namelen, uint16_t qtype, uint16_t qclass)
1760 {
1761 	struct module_qstate* subq;
1762 	if(!generate_sub_request(name, namelen, qtype, qclass, qstate,
1763 		id, iq, INIT_REQUEST_STATE, FINISHED_STATE, &subq, 0))
1764 		return 0;
1765 	log_nametypeclass(VERB_QUERY, "new target", name, qtype, qclass);
1766 	return 1;
1767 }
1768 
1769 /**
1770  * Given an event at a certain state, generate zero or more target queries
1771  * for it's current delegation point.
1772  *
1773  * @param qstate: query state.
1774  * @param iq: iterator query state.
1775  * @param ie: iterator shared global environment.
1776  * @param id: module id.
1777  * @param maxtargets: The maximum number of targets to query for.
1778  *	if it is negative, there is no maximum number of targets.
1779  * @param num: returns the number of queries generated and processed,
1780  *	which may be zero if there were no missing targets.
1781  * @return false on error.
1782  */
1783 static int
1784 query_for_targets(struct module_qstate* qstate, struct iter_qstate* iq,
1785         struct iter_env* ie, int id, int maxtargets, int* num)
1786 {
1787 	int query_count = 0;
1788 	struct delegpt_ns* ns;
1789 	int missing;
1790 	int toget = 0;
1791 
1792 	if(iq->depth == ie->max_dependency_depth)
1793 		return 0;
1794 	if(iq->depth > 0 && iq->target_count &&
1795 		iq->target_count[1] > MAX_TARGET_COUNT) {
1796 		char s[LDNS_MAX_DOMAINLEN+1];
1797 		dname_str(qstate->qinfo.qname, s);
1798 		verbose(VERB_QUERY, "request %s has exceeded the maximum "
1799 			"number of glue fetches %d", s, iq->target_count[1]);
1800 		return 0;
1801 	}
1802 
1803 	iter_mark_cycle_targets(qstate, iq->dp);
1804 	missing = (int)delegpt_count_missing_targets(iq->dp);
1805 	log_assert(maxtargets != 0); /* that would not be useful */
1806 
1807 	/* Generate target requests. Basically, any missing targets
1808 	 * are queried for here, regardless if it is necessary to do
1809 	 * so to continue processing. */
1810 	if(maxtargets < 0 || maxtargets > missing)
1811 		toget = missing;
1812 	else	toget = maxtargets;
1813 	if(toget == 0) {
1814 		*num = 0;
1815 		return 1;
1816 	}
1817 	/* select 'toget' items from the total of 'missing' items */
1818 	log_assert(toget <= missing);
1819 
1820 	/* loop over missing targets */
1821 	for(ns = iq->dp->nslist; ns; ns = ns->next) {
1822 		if(ns->resolved)
1823 			continue;
1824 
1825 		/* randomly select this item with probability toget/missing */
1826 		if(!iter_ns_probability(qstate->env->rnd, toget, missing)) {
1827 			/* do not select this one, next; select toget number
1828 			 * of items from a list one less in size */
1829 			missing --;
1830 			continue;
1831 		}
1832 
1833 		if(ie->supports_ipv6 && !ns->got6) {
1834 			/* Send the AAAA request. */
1835 			if(!generate_target_query(qstate, iq, id,
1836 				ns->name, ns->namelen,
1837 				LDNS_RR_TYPE_AAAA, iq->qchase.qclass)) {
1838 				*num = query_count;
1839 				if(query_count > 0)
1840 					qstate->ext_state[id] = module_wait_subquery;
1841 				return 0;
1842 			}
1843 			query_count++;
1844 		}
1845 		/* Send the A request. */
1846 		if(ie->supports_ipv4 && !ns->got4) {
1847 			if(!generate_target_query(qstate, iq, id,
1848 				ns->name, ns->namelen,
1849 				LDNS_RR_TYPE_A, iq->qchase.qclass)) {
1850 				*num = query_count;
1851 				if(query_count > 0)
1852 					qstate->ext_state[id] = module_wait_subquery;
1853 				return 0;
1854 			}
1855 			query_count++;
1856 		}
1857 
1858 		/* mark this target as in progress. */
1859 		ns->resolved = 1;
1860 		missing--;
1861 		toget--;
1862 		if(toget == 0)
1863 			break;
1864 	}
1865 	*num = query_count;
1866 	if(query_count > 0)
1867 		qstate->ext_state[id] = module_wait_subquery;
1868 
1869 	return 1;
1870 }
1871 
1872 /**
1873  * Called by processQueryTargets when it would like extra targets to query
1874  * but it seems to be out of options.  At last resort some less appealing
1875  * options are explored.  If there are no more options, the result is SERVFAIL
1876  *
1877  * @param qstate: query state.
1878  * @param iq: iterator query state.
1879  * @param ie: iterator shared global environment.
1880  * @param id: module id.
1881  * @return true if the event requires more request processing immediately,
1882  *         false if not.
1883  */
1884 static int
1885 processLastResort(struct module_qstate* qstate, struct iter_qstate* iq,
1886 	struct iter_env* ie, int id)
1887 {
1888 	struct delegpt_ns* ns;
1889 	int query_count = 0;
1890 	verbose(VERB_ALGO, "No more query targets, attempting last resort");
1891 	log_assert(iq->dp);
1892 
1893 	if(!can_have_last_resort(qstate->env, iq->dp->name, iq->dp->namelen,
1894 		iq->qchase.qclass, NULL)) {
1895 		/* fail -- no more targets, no more hope of targets, no hope
1896 		 * of a response. */
1897 		errinf(qstate, "all the configured stub or forward servers failed,");
1898 		errinf_dname(qstate, "at zone", iq->dp->name);
1899 		verbose(VERB_QUERY, "configured stub or forward servers failed -- returning SERVFAIL");
1900 		return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
1901 	}
1902 	if(!iq->dp->has_parent_side_NS && dname_is_root(iq->dp->name)) {
1903 		struct delegpt* p = hints_lookup_root(qstate->env->hints,
1904 			iq->qchase.qclass);
1905 		if(p) {
1906 			struct delegpt_ns* ns;
1907 			struct delegpt_addr* a;
1908 			iq->chase_flags &= ~BIT_RD; /* go to authorities */
1909 			for(ns = p->nslist; ns; ns=ns->next) {
1910 				(void)delegpt_add_ns(iq->dp, qstate->region,
1911 					ns->name, ns->lame);
1912 			}
1913 			for(a = p->target_list; a; a=a->next_target) {
1914 				(void)delegpt_add_addr(iq->dp, qstate->region,
1915 					&a->addr, a->addrlen, a->bogus,
1916 					a->lame, a->tls_auth_name);
1917 			}
1918 		}
1919 		iq->dp->has_parent_side_NS = 1;
1920 	} else if(!iq->dp->has_parent_side_NS) {
1921 		if(!iter_lookup_parent_NS_from_cache(qstate->env, iq->dp,
1922 			qstate->region, &qstate->qinfo)
1923 			|| !iq->dp->has_parent_side_NS) {
1924 			/* if: malloc failure in lookup go up to try */
1925 			/* if: no parent NS in cache - go up one level */
1926 			verbose(VERB_ALGO, "try to grab parent NS");
1927 			iq->store_parent_NS = iq->dp;
1928 			iq->chase_flags &= ~BIT_RD; /* go to authorities */
1929 			iq->deleg_msg = NULL;
1930 			iq->refetch_glue = 1;
1931 			iq->query_restart_count++;
1932 			iq->sent_count = 0;
1933 			if(qstate->env->cfg->qname_minimisation)
1934 				iq->minimisation_state = INIT_MINIMISE_STATE;
1935 			return next_state(iq, INIT_REQUEST_STATE);
1936 		}
1937 	}
1938 	/* see if that makes new names available */
1939 	if(!cache_fill_missing(qstate->env, iq->qchase.qclass,
1940 		qstate->region, iq->dp))
1941 		log_err("out of memory in cache_fill_missing");
1942 	if(iq->dp->usable_list) {
1943 		verbose(VERB_ALGO, "try parent-side-name, w. glue from cache");
1944 		return next_state(iq, QUERYTARGETS_STATE);
1945 	}
1946 	/* try to fill out parent glue from cache */
1947 	if(iter_lookup_parent_glue_from_cache(qstate->env, iq->dp,
1948 		qstate->region, &qstate->qinfo)) {
1949 		/* got parent stuff from cache, see if we can continue */
1950 		verbose(VERB_ALGO, "try parent-side glue from cache");
1951 		return next_state(iq, QUERYTARGETS_STATE);
1952 	}
1953 	/* query for an extra name added by the parent-NS record */
1954 	if(delegpt_count_missing_targets(iq->dp) > 0) {
1955 		int qs = 0;
1956 		verbose(VERB_ALGO, "try parent-side target name");
1957 		if(!query_for_targets(qstate, iq, ie, id, 1, &qs)) {
1958 			errinf(qstate, "could not fetch nameserver");
1959 			errinf_dname(qstate, "at zone", iq->dp->name);
1960 			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1961 		}
1962 		iq->num_target_queries += qs;
1963 		target_count_increase(iq, qs);
1964 		if(qs != 0) {
1965 			qstate->ext_state[id] = module_wait_subquery;
1966 			return 0; /* and wait for them */
1967 		}
1968 	}
1969 	if(iq->depth == ie->max_dependency_depth) {
1970 		verbose(VERB_QUERY, "maxdepth and need more nameservers, fail");
1971 		errinf(qstate, "cannot fetch more nameservers because at max dependency depth");
1972 		return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
1973 	}
1974 	if(iq->depth > 0 && iq->target_count &&
1975 		iq->target_count[1] > MAX_TARGET_COUNT) {
1976 		char s[LDNS_MAX_DOMAINLEN+1];
1977 		dname_str(qstate->qinfo.qname, s);
1978 		verbose(VERB_QUERY, "request %s has exceeded the maximum "
1979 			"number of glue fetches %d", s, iq->target_count[1]);
1980 		errinf(qstate, "exceeded the maximum number of glue fetches");
1981 		return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
1982 	}
1983 	/* mark cycle targets for parent-side lookups */
1984 	iter_mark_pside_cycle_targets(qstate, iq->dp);
1985 	/* see if we can issue queries to get nameserver addresses */
1986 	/* this lookup is not randomized, but sequential. */
1987 	for(ns = iq->dp->nslist; ns; ns = ns->next) {
1988 		/* if this nameserver is at a delegation point, but that
1989 		 * delegation point is a stub and we cannot go higher, skip*/
1990 		if( ((ie->supports_ipv6 && !ns->done_pside6) ||
1991 		    (ie->supports_ipv4 && !ns->done_pside4)) &&
1992 		    !can_have_last_resort(qstate->env, ns->name, ns->namelen,
1993 			iq->qchase.qclass, NULL)) {
1994 			log_nametypeclass(VERB_ALGO, "cannot pside lookup ns "
1995 				"because it is also a stub/forward,",
1996 				ns->name, LDNS_RR_TYPE_NS, iq->qchase.qclass);
1997 			if(ie->supports_ipv6) ns->done_pside6 = 1;
1998 			if(ie->supports_ipv4) ns->done_pside4 = 1;
1999 			continue;
2000 		}
2001 		/* query for parent-side A and AAAA for nameservers */
2002 		if(ie->supports_ipv6 && !ns->done_pside6) {
2003 			/* Send the AAAA request. */
2004 			if(!generate_parentside_target_query(qstate, iq, id,
2005 				ns->name, ns->namelen,
2006 				LDNS_RR_TYPE_AAAA, iq->qchase.qclass)) {
2007 				errinf_dname(qstate, "could not generate nameserver AAAA lookup for", ns->name);
2008 				return error_response(qstate, id,
2009 					LDNS_RCODE_SERVFAIL);
2010 			}
2011 			ns->done_pside6 = 1;
2012 			query_count++;
2013 		}
2014 		if(ie->supports_ipv4 && !ns->done_pside4) {
2015 			/* Send the A request. */
2016 			if(!generate_parentside_target_query(qstate, iq, id,
2017 				ns->name, ns->namelen,
2018 				LDNS_RR_TYPE_A, iq->qchase.qclass)) {
2019 				errinf_dname(qstate, "could not generate nameserver A lookup for", ns->name);
2020 				return error_response(qstate, id,
2021 					LDNS_RCODE_SERVFAIL);
2022 			}
2023 			ns->done_pside4 = 1;
2024 			query_count++;
2025 		}
2026 		if(query_count != 0) { /* suspend to await results */
2027 			verbose(VERB_ALGO, "try parent-side glue lookup");
2028 			iq->num_target_queries += query_count;
2029 			target_count_increase(iq, query_count);
2030 			qstate->ext_state[id] = module_wait_subquery;
2031 			return 0;
2032 		}
2033 	}
2034 
2035 	/* if this was a parent-side glue query itself, then store that
2036 	 * failure in cache. */
2037 	if(!qstate->no_cache_store && iq->query_for_pside_glue
2038 		&& !iq->pside_glue)
2039 			iter_store_parentside_neg(qstate->env, &qstate->qinfo,
2040 				iq->deleg_msg?iq->deleg_msg->rep:
2041 				(iq->response?iq->response->rep:NULL));
2042 
2043 	errinf(qstate, "all servers for this domain failed,");
2044 	errinf_dname(qstate, "at zone", iq->dp->name);
2045 	verbose(VERB_QUERY, "out of query targets -- returning SERVFAIL");
2046 	/* fail -- no more targets, no more hope of targets, no hope
2047 	 * of a response. */
2048 	return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
2049 }
2050 
2051 /**
2052  * Try to find the NS record set that will resolve a qtype DS query. Due
2053  * to grandparent/grandchild reasons we did not get a proper lookup right
2054  * away.  We need to create type NS queries until we get the right parent
2055  * for this lookup.  We remove labels from the query to find the right point.
2056  * If we end up at the old dp name, then there is no solution.
2057  *
2058  * @param qstate: query state.
2059  * @param iq: iterator query state.
2060  * @param id: module id.
2061  * @return true if the event requires more immediate processing, false if
2062  *         not. This is generally only true when forwarding the request to
2063  *         the final state (i.e., on answer).
2064  */
2065 static int
2066 processDSNSFind(struct module_qstate* qstate, struct iter_qstate* iq, int id)
2067 {
2068 	struct module_qstate* subq = NULL;
2069 	verbose(VERB_ALGO, "processDSNSFind");
2070 
2071 	if(!iq->dsns_point) {
2072 		/* initialize */
2073 		iq->dsns_point = iq->qchase.qname;
2074 		iq->dsns_point_len = iq->qchase.qname_len;
2075 	}
2076 	/* robustcheck for internal error: we are not underneath the dp */
2077 	if(!dname_subdomain_c(iq->dsns_point, iq->dp->name)) {
2078 		errinf_dname(qstate, "for DS query parent-child nameserver search the query is not under the zone", iq->dp->name);
2079 		return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
2080 	}
2081 
2082 	/* go up one (more) step, until we hit the dp, if so, end */
2083 	dname_remove_label(&iq->dsns_point, &iq->dsns_point_len);
2084 	if(query_dname_compare(iq->dsns_point, iq->dp->name) == 0) {
2085 		/* there was no inbetween nameserver, use the old delegation
2086 		 * point again.  And this time, because dsns_point is nonNULL
2087 		 * we are going to accept the (bad) result */
2088 		iq->state = QUERYTARGETS_STATE;
2089 		return 1;
2090 	}
2091 	iq->state = DSNS_FIND_STATE;
2092 
2093 	/* spawn NS lookup (validation not needed, this is for DS lookup) */
2094 	log_nametypeclass(VERB_ALGO, "fetch nameservers",
2095 		iq->dsns_point, LDNS_RR_TYPE_NS, iq->qchase.qclass);
2096 	if(!generate_sub_request(iq->dsns_point, iq->dsns_point_len,
2097 		LDNS_RR_TYPE_NS, iq->qchase.qclass, qstate, id, iq,
2098 		INIT_REQUEST_STATE, FINISHED_STATE, &subq, 0)) {
2099 		errinf_dname(qstate, "for DS query parent-child nameserver search, could not generate NS lookup for", iq->dsns_point);
2100 		return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
2101 	}
2102 
2103 	return 0;
2104 }
2105 
2106 /**
2107  * This is the request event state where the request will be sent to one of
2108  * its current query targets. This state also handles issuing target lookup
2109  * queries for missing target IP addresses. Queries typically iterate on
2110  * this state, both when they are just trying different targets for a given
2111  * delegation point, and when they change delegation points. This state
2112  * roughly corresponds to RFC 1034 algorithm steps 3 and 4.
2113  *
2114  * @param qstate: query state.
2115  * @param iq: iterator query state.
2116  * @param ie: iterator shared global environment.
2117  * @param id: module id.
2118  * @return true if the event requires more request processing immediately,
2119  *         false if not. This state only returns true when it is generating
2120  *         a SERVFAIL response because the query has hit a dead end.
2121  */
2122 static int
2123 processQueryTargets(struct module_qstate* qstate, struct iter_qstate* iq,
2124 	struct iter_env* ie, int id)
2125 {
2126 	int tf_policy;
2127 	struct delegpt_addr* target;
2128 	struct outbound_entry* outq;
2129 	int auth_fallback = 0;
2130 
2131 	/* NOTE: a request will encounter this state for each target it
2132 	 * needs to send a query to. That is, at least one per referral,
2133 	 * more if some targets timeout or return throwaway answers. */
2134 
2135 	log_query_info(VERB_QUERY, "processQueryTargets:", &qstate->qinfo);
2136 	verbose(VERB_ALGO, "processQueryTargets: targetqueries %d, "
2137 		"currentqueries %d sentcount %d", iq->num_target_queries,
2138 		iq->num_current_queries, iq->sent_count);
2139 
2140 	/* Make sure that we haven't run away */
2141 	/* FIXME: is this check even necessary? */
2142 	if(iq->referral_count > MAX_REFERRAL_COUNT) {
2143 		verbose(VERB_QUERY, "request has exceeded the maximum "
2144 			"number of referrrals with %d", iq->referral_count);
2145 		errinf(qstate, "exceeded the maximum of referrals");
2146 		return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2147 	}
2148 	if(iq->sent_count > MAX_SENT_COUNT) {
2149 		verbose(VERB_QUERY, "request has exceeded the maximum "
2150 			"number of sends with %d", iq->sent_count);
2151 		errinf(qstate, "exceeded the maximum number of sends");
2152 		return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2153 	}
2154 
2155 	/* Make sure we have a delegation point, otherwise priming failed
2156 	 * or another failure occurred */
2157 	if(!iq->dp) {
2158 		verbose(VERB_QUERY, "Failed to get a delegation, giving up");
2159 		errinf(qstate, "failed to get a delegation (eg. prime failure)");
2160 		return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2161 	}
2162 	if(!ie->supports_ipv6)
2163 		delegpt_no_ipv6(iq->dp);
2164 	if(!ie->supports_ipv4)
2165 		delegpt_no_ipv4(iq->dp);
2166 	delegpt_log(VERB_ALGO, iq->dp);
2167 
2168 	if(iq->num_current_queries>0) {
2169 		/* already busy answering a query, this restart is because
2170 		 * more delegpt addrs became available, wait for existing
2171 		 * query. */
2172 		verbose(VERB_ALGO, "woke up, but wait for outstanding query");
2173 		qstate->ext_state[id] = module_wait_reply;
2174 		return 0;
2175 	}
2176 
2177 	if(iq->minimisation_state == INIT_MINIMISE_STATE
2178 		&& !(iq->chase_flags & BIT_RD)) {
2179 		/* (Re)set qinfo_out to (new) delegation point, except when
2180 		 * qinfo_out is already a subdomain of dp. This happens when
2181 		 * increasing by more than one label at once (QNAMEs with more
2182 		 * than MAX_MINIMISE_COUNT labels). */
2183 		if(!(iq->qinfo_out.qname_len
2184 			&& dname_subdomain_c(iq->qchase.qname,
2185 				iq->qinfo_out.qname)
2186 			&& dname_subdomain_c(iq->qinfo_out.qname,
2187 				iq->dp->name))) {
2188 			iq->qinfo_out.qname = iq->dp->name;
2189 			iq->qinfo_out.qname_len = iq->dp->namelen;
2190 			iq->qinfo_out.qtype = LDNS_RR_TYPE_A;
2191 			iq->qinfo_out.qclass = iq->qchase.qclass;
2192 			iq->qinfo_out.local_alias = NULL;
2193 			iq->minimise_count = 0;
2194 		}
2195 
2196 		iq->minimisation_state = MINIMISE_STATE;
2197 	}
2198 	if(iq->minimisation_state == MINIMISE_STATE) {
2199 		int qchaselabs = dname_count_labels(iq->qchase.qname);
2200 		int labdiff = qchaselabs -
2201 			dname_count_labels(iq->qinfo_out.qname);
2202 
2203 		iq->qinfo_out.qname = iq->qchase.qname;
2204 		iq->qinfo_out.qname_len = iq->qchase.qname_len;
2205 		iq->minimise_count++;
2206 		iq->minimise_timeout_count = 0;
2207 
2208 		iter_dec_attempts(iq->dp, 1);
2209 
2210 		/* Limit number of iterations for QNAMEs with more
2211 		 * than MAX_MINIMISE_COUNT labels. Send first MINIMISE_ONE_LAB
2212 		 * labels of QNAME always individually.
2213 		 */
2214 		if(qchaselabs > MAX_MINIMISE_COUNT && labdiff > 1 &&
2215 			iq->minimise_count > MINIMISE_ONE_LAB) {
2216 			if(iq->minimise_count < MAX_MINIMISE_COUNT) {
2217 				int multilabs = qchaselabs - 1 -
2218 					MINIMISE_ONE_LAB;
2219 				int extralabs = multilabs /
2220 					MINIMISE_MULTIPLE_LABS;
2221 
2222 				if (MAX_MINIMISE_COUNT - iq->minimise_count >=
2223 					multilabs % MINIMISE_MULTIPLE_LABS)
2224 					/* Default behaviour is to add 1 label
2225 					 * every iteration. Therefore, decrement
2226 					 * the extralabs by 1 */
2227 					extralabs--;
2228 				if (extralabs < labdiff)
2229 					labdiff -= extralabs;
2230 				else
2231 					labdiff = 1;
2232 			}
2233 			/* Last minimised iteration, send all labels with
2234 			 * QTYPE=NS */
2235 			else
2236 				labdiff = 1;
2237 		}
2238 
2239 		if(labdiff > 1) {
2240 			verbose(VERB_QUERY, "removing %d labels", labdiff-1);
2241 			dname_remove_labels(&iq->qinfo_out.qname,
2242 				&iq->qinfo_out.qname_len,
2243 				labdiff-1);
2244 		}
2245 		if(labdiff < 1 || (labdiff < 2
2246 			&& (iq->qchase.qtype == LDNS_RR_TYPE_DS
2247 			|| iq->qchase.qtype == LDNS_RR_TYPE_A)))
2248 			/* Stop minimising this query, resolve "as usual" */
2249 			iq->minimisation_state = DONOT_MINIMISE_STATE;
2250 		else if(!qstate->no_cache_lookup) {
2251 			struct dns_msg* msg = dns_cache_lookup(qstate->env,
2252 				iq->qinfo_out.qname, iq->qinfo_out.qname_len,
2253 				iq->qinfo_out.qtype, iq->qinfo_out.qclass,
2254 				qstate->query_flags, qstate->region,
2255 				qstate->env->scratch, 0);
2256 			if(msg && msg->rep->an_numrrsets == 0
2257 				&& FLAGS_GET_RCODE(msg->rep->flags) ==
2258 				LDNS_RCODE_NOERROR)
2259 				/* no need to send query if it is already
2260 				 * cached as NOERROR/NODATA */
2261 				return 1;
2262 		}
2263 	}
2264 	if(iq->minimisation_state == SKIP_MINIMISE_STATE) {
2265 		if(iq->minimise_timeout_count < MAX_MINIMISE_TIMEOUT_COUNT)
2266 			/* Do not increment qname, continue incrementing next
2267 			 * iteration */
2268 			iq->minimisation_state = MINIMISE_STATE;
2269 		else if(!qstate->env->cfg->qname_minimisation_strict)
2270 			/* Too many time-outs detected for this QNAME and QTYPE.
2271 			 * We give up, disable QNAME minimisation. */
2272 			iq->minimisation_state = DONOT_MINIMISE_STATE;
2273 	}
2274 	if(iq->minimisation_state == DONOT_MINIMISE_STATE)
2275 		iq->qinfo_out = iq->qchase;
2276 
2277 	/* now find an answer to this query */
2278 	/* see if authority zones have an answer */
2279 	/* now we know the dp, we can check the auth zone for locally hosted
2280 	 * contents */
2281 	if(!iq->auth_zone_avoid && qstate->blacklist) {
2282 		if(auth_zones_can_fallback(qstate->env->auth_zones,
2283 			iq->dp->name, iq->dp->namelen, iq->qinfo_out.qclass)) {
2284 			/* if cache is blacklisted and this zone allows us
2285 			 * to fallback to the internet, then do so, and
2286 			 * fetch results from the internet servers */
2287 			iq->auth_zone_avoid = 1;
2288 		}
2289 	}
2290 	if(iq->auth_zone_avoid) {
2291 		iq->auth_zone_avoid = 0;
2292 		auth_fallback = 1;
2293 	} else if(auth_zones_lookup(qstate->env->auth_zones, &iq->qinfo_out,
2294 		qstate->region, &iq->response, &auth_fallback, iq->dp->name,
2295 		iq->dp->namelen)) {
2296 		/* use this as a response to be processed by the iterator */
2297 		if(verbosity >= VERB_ALGO) {
2298 			log_dns_msg("msg from auth zone",
2299 				&iq->response->qinfo, iq->response->rep);
2300 		}
2301 		if((iq->chase_flags&BIT_RD) && !(iq->response->rep->flags&BIT_AA)) {
2302 			verbose(VERB_ALGO, "forwarder, ignoring referral from auth zone");
2303 		} else {
2304 			lock_rw_wrlock(&qstate->env->auth_zones->lock);
2305 			qstate->env->auth_zones->num_query_up++;
2306 			lock_rw_unlock(&qstate->env->auth_zones->lock);
2307 			iq->num_current_queries++;
2308 			iq->chase_to_rd = 0;
2309 			iq->dnssec_lame_query = 0;
2310 			iq->auth_zone_response = 1;
2311 			return next_state(iq, QUERY_RESP_STATE);
2312 		}
2313 	}
2314 	iq->auth_zone_response = 0;
2315 	if(auth_fallback == 0) {
2316 		/* like we got servfail from the auth zone lookup, and
2317 		 * no internet fallback */
2318 		verbose(VERB_ALGO, "auth zone lookup failed, no fallback,"
2319 			" servfail");
2320 		errinf(qstate, "auth zone lookup failed, fallback is off");
2321 		return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2322 	}
2323 	if(iq->dp && iq->dp->auth_dp) {
2324 		/* we wanted to fallback, but had no delegpt, only the
2325 		 * auth zone generated delegpt, create an actual one */
2326 		iq->auth_zone_avoid = 1;
2327 		return next_state(iq, INIT_REQUEST_STATE);
2328 	}
2329 	/* but mostly, fallback==1 (like, when no such auth zone exists)
2330 	 * and we continue with lookups */
2331 
2332 	tf_policy = 0;
2333 	/* < not <=, because although the array is large enough for <=, the
2334 	 * generated query will immediately be discarded due to depth and
2335 	 * that servfail is cached, which is not good as opportunism goes. */
2336 	if(iq->depth < ie->max_dependency_depth
2337 		&& iq->sent_count < TARGET_FETCH_STOP) {
2338 		tf_policy = ie->target_fetch_policy[iq->depth];
2339 	}
2340 
2341 	/* if in 0x20 fallback get as many targets as possible */
2342 	if(iq->caps_fallback) {
2343 		int extra = 0;
2344 		size_t naddr, nres, navail;
2345 		if(!query_for_targets(qstate, iq, ie, id, -1, &extra)) {
2346 			errinf(qstate, "could not fetch nameservers for 0x20 fallback");
2347 			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2348 		}
2349 		iq->num_target_queries += extra;
2350 		target_count_increase(iq, extra);
2351 		if(iq->num_target_queries > 0) {
2352 			/* wait to get all targets, we want to try em */
2353 			verbose(VERB_ALGO, "wait for all targets for fallback");
2354 			qstate->ext_state[id] = module_wait_reply;
2355 			return 0;
2356 		}
2357 		/* did we do enough fallback queries already? */
2358 		delegpt_count_addr(iq->dp, &naddr, &nres, &navail);
2359 		/* the current caps_server is the number of fallbacks sent.
2360 		 * the original query is one that matched too, so we have
2361 		 * caps_server+1 number of matching queries now */
2362 		if(iq->caps_server+1 >= naddr*3 ||
2363 			iq->caps_server*2+2 >= MAX_SENT_COUNT) {
2364 			/* *2 on sentcount check because ipv6 may fail */
2365 			/* we're done, process the response */
2366 			verbose(VERB_ALGO, "0x20 fallback had %d responses "
2367 				"match for %d wanted, done.",
2368 				(int)iq->caps_server+1, (int)naddr*3);
2369 			iq->response = iq->caps_response;
2370 			iq->caps_fallback = 0;
2371 			iter_dec_attempts(iq->dp, 3); /* space for fallback */
2372 			iq->num_current_queries++; /* RespState decrements it*/
2373 			iq->referral_count++; /* make sure we don't loop */
2374 			iq->sent_count = 0;
2375 			iq->state = QUERY_RESP_STATE;
2376 			return 1;
2377 		}
2378 		verbose(VERB_ALGO, "0x20 fallback number %d",
2379 			(int)iq->caps_server);
2380 
2381 	/* if there is a policy to fetch missing targets
2382 	 * opportunistically, do it. we rely on the fact that once a
2383 	 * query (or queries) for a missing name have been issued,
2384 	 * they will not show up again. */
2385 	} else if(tf_policy != 0) {
2386 		int extra = 0;
2387 		verbose(VERB_ALGO, "attempt to get extra %d targets",
2388 			tf_policy);
2389 		(void)query_for_targets(qstate, iq, ie, id, tf_policy, &extra);
2390 		/* errors ignored, these targets are not strictly necessary for
2391 		 * this result, we do not have to reply with SERVFAIL */
2392 		iq->num_target_queries += extra;
2393 		target_count_increase(iq, extra);
2394 	}
2395 
2396 	/* Add the current set of unused targets to our queue. */
2397 	delegpt_add_unused_targets(iq->dp);
2398 
2399 	/* Select the next usable target, filtering out unsuitable targets. */
2400 	target = iter_server_selection(ie, qstate->env, iq->dp,
2401 		iq->dp->name, iq->dp->namelen, iq->qchase.qtype,
2402 		&iq->dnssec_lame_query, &iq->chase_to_rd,
2403 		iq->num_target_queries, qstate->blacklist,
2404 		qstate->prefetch_leeway);
2405 
2406 	/* If no usable target was selected... */
2407 	if(!target) {
2408 		/* Here we distinguish between three states: generate a new
2409 		 * target query, just wait, or quit (with a SERVFAIL).
2410 		 * We have the following information: number of active
2411 		 * target queries, number of active current queries,
2412 		 * the presence of missing targets at this delegation
2413 		 * point, and the given query target policy. */
2414 
2415 		/* Check for the wait condition. If this is true, then
2416 		 * an action must be taken. */
2417 		if(iq->num_target_queries==0 && iq->num_current_queries==0) {
2418 			/* If there is nothing to wait for, then we need
2419 			 * to distinguish between generating (a) new target
2420 			 * query, or failing. */
2421 			if(delegpt_count_missing_targets(iq->dp) > 0) {
2422 				int qs = 0;
2423 				verbose(VERB_ALGO, "querying for next "
2424 					"missing target");
2425 				if(!query_for_targets(qstate, iq, ie, id,
2426 					1, &qs)) {
2427 					errinf(qstate, "could not fetch nameserver");
2428 					errinf_dname(qstate, "at zone", iq->dp->name);
2429 					return error_response(qstate, id,
2430 						LDNS_RCODE_SERVFAIL);
2431 				}
2432 				if(qs == 0 &&
2433 				   delegpt_count_missing_targets(iq->dp) == 0){
2434 					/* it looked like there were missing
2435 					 * targets, but they did not turn up.
2436 					 * Try the bad choices again (if any),
2437 					 * when we get back here missing==0,
2438 					 * so this is not a loop. */
2439 					return 1;
2440 				}
2441 				iq->num_target_queries += qs;
2442 				target_count_increase(iq, qs);
2443 			}
2444 			/* Since a target query might have been made, we
2445 			 * need to check again. */
2446 			if(iq->num_target_queries == 0) {
2447 				/* if in capsforid fallback, instead of last
2448 				 * resort, we agree with the current reply
2449 				 * we have (if any) (our count of addrs bad)*/
2450 				if(iq->caps_fallback && iq->caps_reply) {
2451 					/* we're done, process the response */
2452 					verbose(VERB_ALGO, "0x20 fallback had %d responses, "
2453 						"but no more servers except "
2454 						"last resort, done.",
2455 						(int)iq->caps_server+1);
2456 					iq->response = iq->caps_response;
2457 					iq->caps_fallback = 0;
2458 					iter_dec_attempts(iq->dp, 3); /* space for fallback */
2459 					iq->num_current_queries++; /* RespState decrements it*/
2460 					iq->referral_count++; /* make sure we don't loop */
2461 					iq->sent_count = 0;
2462 					iq->state = QUERY_RESP_STATE;
2463 					return 1;
2464 				}
2465 				return processLastResort(qstate, iq, ie, id);
2466 			}
2467 		}
2468 
2469 		/* otherwise, we have no current targets, so submerge
2470 		 * until one of the target or direct queries return. */
2471 		if(iq->num_target_queries>0 && iq->num_current_queries>0) {
2472 			verbose(VERB_ALGO, "no current targets -- waiting "
2473 				"for %d targets to resolve or %d outstanding"
2474 				" queries to respond", iq->num_target_queries,
2475 				iq->num_current_queries);
2476 			qstate->ext_state[id] = module_wait_reply;
2477 		} else if(iq->num_target_queries>0) {
2478 			verbose(VERB_ALGO, "no current targets -- waiting "
2479 				"for %d targets to resolve.",
2480 				iq->num_target_queries);
2481 			qstate->ext_state[id] = module_wait_subquery;
2482 		} else {
2483 			verbose(VERB_ALGO, "no current targets -- waiting "
2484 				"for %d outstanding queries to respond.",
2485 				iq->num_current_queries);
2486 			qstate->ext_state[id] = module_wait_reply;
2487 		}
2488 		return 0;
2489 	}
2490 
2491 	/* if not forwarding, check ratelimits per delegationpoint name */
2492 	if(!(iq->chase_flags & BIT_RD) && !iq->ratelimit_ok) {
2493 		if(!infra_ratelimit_inc(qstate->env->infra_cache, iq->dp->name,
2494 			iq->dp->namelen, *qstate->env->now)) {
2495 			lock_basic_lock(&ie->queries_ratelimit_lock);
2496 			ie->num_queries_ratelimited++;
2497 			lock_basic_unlock(&ie->queries_ratelimit_lock);
2498 			verbose(VERB_ALGO, "query exceeded ratelimits");
2499 			qstate->was_ratelimited = 1;
2500 			errinf_dname(qstate, "exceeded ratelimit for zone",
2501 				iq->dp->name);
2502 			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2503 		}
2504 	}
2505 
2506 	/* We have a valid target. */
2507 	if(verbosity >= VERB_QUERY) {
2508 		log_query_info(VERB_QUERY, "sending query:", &iq->qinfo_out);
2509 		log_name_addr(VERB_QUERY, "sending to target:", iq->dp->name,
2510 			&target->addr, target->addrlen);
2511 		verbose(VERB_ALGO, "dnssec status: %s%s",
2512 			iq->dnssec_expected?"expected": "not expected",
2513 			iq->dnssec_lame_query?" but lame_query anyway": "");
2514 	}
2515 	fptr_ok(fptr_whitelist_modenv_send_query(qstate->env->send_query));
2516 	outq = (*qstate->env->send_query)(&iq->qinfo_out,
2517 		iq->chase_flags | (iq->chase_to_rd?BIT_RD:0),
2518 		/* unset CD if to forwarder(RD set) and not dnssec retry
2519 		 * (blacklist nonempty) and no trust-anchors are configured
2520 		 * above the qname or on the first attempt when dnssec is on */
2521 		EDNS_DO| ((iq->chase_to_rd||(iq->chase_flags&BIT_RD)!=0)&&
2522 		!qstate->blacklist&&(!iter_qname_indicates_dnssec(qstate->env,
2523 		&iq->qinfo_out)||target->attempts==1)?0:BIT_CD),
2524 		iq->dnssec_expected, iq->caps_fallback || is_caps_whitelisted(
2525 		ie, iq), &target->addr, target->addrlen,
2526 		iq->dp->name, iq->dp->namelen,
2527 		(iq->dp->ssl_upstream || qstate->env->cfg->ssl_upstream),
2528 		target->tls_auth_name, qstate);
2529 	if(!outq) {
2530 		log_addr(VERB_DETAIL, "error sending query to auth server",
2531 			&target->addr, target->addrlen);
2532 		if(!(iq->chase_flags & BIT_RD) && !iq->ratelimit_ok)
2533 		    infra_ratelimit_dec(qstate->env->infra_cache, iq->dp->name,
2534 			iq->dp->namelen, *qstate->env->now);
2535 		if(qstate->env->cfg->qname_minimisation)
2536 			iq->minimisation_state = SKIP_MINIMISE_STATE;
2537 		return next_state(iq, QUERYTARGETS_STATE);
2538 	}
2539 	outbound_list_insert(&iq->outlist, outq);
2540 	iq->num_current_queries++;
2541 	iq->sent_count++;
2542 	qstate->ext_state[id] = module_wait_reply;
2543 
2544 	return 0;
2545 }
2546 
2547 /** find NS rrset in given list */
2548 static struct ub_packed_rrset_key*
2549 find_NS(struct reply_info* rep, size_t from, size_t to)
2550 {
2551 	size_t i;
2552 	for(i=from; i<to; i++) {
2553 		if(ntohs(rep->rrsets[i]->rk.type) == LDNS_RR_TYPE_NS)
2554 			return rep->rrsets[i];
2555 	}
2556 	return NULL;
2557 }
2558 
2559 
2560 /**
2561  * Process the query response. All queries end up at this state first. This
2562  * process generally consists of analyzing the response and routing the
2563  * event to the next state (either bouncing it back to a request state, or
2564  * terminating the processing for this event).
2565  *
2566  * @param qstate: query state.
2567  * @param iq: iterator query state.
2568  * @param id: module id.
2569  * @return true if the event requires more immediate processing, false if
2570  *         not. This is generally only true when forwarding the request to
2571  *         the final state (i.e., on answer).
2572  */
2573 static int
2574 processQueryResponse(struct module_qstate* qstate, struct iter_qstate* iq,
2575 	int id)
2576 {
2577 	int dnsseclame = 0;
2578 	enum response_type type;
2579 	iq->num_current_queries--;
2580 
2581 	if(!inplace_cb_query_response_call(qstate->env, qstate, iq->response))
2582 		log_err("unable to call query_response callback");
2583 
2584 	if(iq->response == NULL) {
2585 		/* Don't increment qname when QNAME minimisation is enabled */
2586 		if(qstate->env->cfg->qname_minimisation) {
2587 			iq->minimise_timeout_count++;
2588 			iq->minimisation_state = SKIP_MINIMISE_STATE;
2589 		}
2590 		iq->chase_to_rd = 0;
2591 		iq->dnssec_lame_query = 0;
2592 		verbose(VERB_ALGO, "query response was timeout");
2593 		return next_state(iq, QUERYTARGETS_STATE);
2594 	}
2595 	type = response_type_from_server(
2596 		(int)((iq->chase_flags&BIT_RD) || iq->chase_to_rd),
2597 		iq->response, &iq->qinfo_out, iq->dp);
2598 	iq->chase_to_rd = 0;
2599 	if(type == RESPONSE_TYPE_REFERRAL && (iq->chase_flags&BIT_RD) &&
2600 		!iq->auth_zone_response) {
2601 		/* When forwarding (RD bit is set), we handle referrals
2602 		 * differently. No queries should be sent elsewhere */
2603 		type = RESPONSE_TYPE_ANSWER;
2604 	}
2605 	if(!qstate->env->cfg->disable_dnssec_lame_check && iq->dnssec_expected
2606                 && !iq->dnssec_lame_query &&
2607 		!(iq->chase_flags&BIT_RD)
2608 		&& iq->sent_count < DNSSEC_LAME_DETECT_COUNT
2609 		&& type != RESPONSE_TYPE_LAME
2610 		&& type != RESPONSE_TYPE_REC_LAME
2611 		&& type != RESPONSE_TYPE_THROWAWAY
2612 		&& type != RESPONSE_TYPE_UNTYPED) {
2613 		/* a possible answer, see if it is missing DNSSEC */
2614 		/* but not when forwarding, so we dont mark fwder lame */
2615 		if(!iter_msg_has_dnssec(iq->response)) {
2616 			/* Mark this address as dnsseclame in this dp,
2617 			 * because that will make serverselection disprefer
2618 			 * it, but also, once it is the only final option,
2619 			 * use dnssec-lame-bypass if it needs to query there.*/
2620 			if(qstate->reply) {
2621 				struct delegpt_addr* a = delegpt_find_addr(
2622 					iq->dp, &qstate->reply->addr,
2623 					qstate->reply->addrlen);
2624 				if(a) a->dnsseclame = 1;
2625 			}
2626 			/* test the answer is from the zone we expected,
2627 		 	 * otherwise, (due to parent,child on same server), we
2628 		 	 * might mark the server,zone lame inappropriately */
2629 			if(!iter_msg_from_zone(iq->response, iq->dp, type,
2630 				iq->qchase.qclass))
2631 				qstate->reply = NULL;
2632 			type = RESPONSE_TYPE_LAME;
2633 			dnsseclame = 1;
2634 		}
2635 	} else iq->dnssec_lame_query = 0;
2636 	/* see if referral brings us close to the target */
2637 	if(type == RESPONSE_TYPE_REFERRAL) {
2638 		struct ub_packed_rrset_key* ns = find_NS(
2639 			iq->response->rep, iq->response->rep->an_numrrsets,
2640 			iq->response->rep->an_numrrsets
2641 			+ iq->response->rep->ns_numrrsets);
2642 		if(!ns) ns = find_NS(iq->response->rep, 0,
2643 				iq->response->rep->an_numrrsets);
2644 		if(!ns || !dname_strict_subdomain_c(ns->rk.dname, iq->dp->name)
2645 			|| !dname_subdomain_c(iq->qchase.qname, ns->rk.dname)){
2646 			verbose(VERB_ALGO, "bad referral, throwaway");
2647 			type = RESPONSE_TYPE_THROWAWAY;
2648 		} else
2649 			iter_scrub_ds(iq->response, ns, iq->dp->name);
2650 	} else iter_scrub_ds(iq->response, NULL, NULL);
2651 	if(type == RESPONSE_TYPE_THROWAWAY &&
2652 		FLAGS_GET_RCODE(iq->response->rep->flags) == LDNS_RCODE_YXDOMAIN) {
2653 		/* YXDOMAIN is a permanent error, no need to retry */
2654 		type = RESPONSE_TYPE_ANSWER;
2655 	}
2656 	if(type == RESPONSE_TYPE_CNAME && iq->response->rep->an_numrrsets >= 1
2657 		&& ntohs(iq->response->rep->rrsets[0]->rk.type) == LDNS_RR_TYPE_DNAME) {
2658 		uint8_t* sname = NULL;
2659 		size_t snamelen = 0;
2660 		get_cname_target(iq->response->rep->rrsets[0], &sname,
2661 			&snamelen);
2662 		if(snamelen && dname_subdomain_c(sname, iq->response->rep->rrsets[0]->rk.dname)) {
2663 			/* DNAME to a subdomain loop; do not recurse */
2664 			type = RESPONSE_TYPE_ANSWER;
2665 		}
2666 	} else if(type == RESPONSE_TYPE_CNAME &&
2667 		iq->qchase.qtype == LDNS_RR_TYPE_CNAME &&
2668 		iq->minimisation_state == MINIMISE_STATE &&
2669 		query_dname_compare(iq->qchase.qname, iq->qinfo_out.qname) == 0) {
2670 		/* The minimised query for full QTYPE and hidden QTYPE can be
2671 		 * classified as CNAME response type, even when the original
2672 		 * QTYPE=CNAME. This should be treated as answer response type.
2673 		 */
2674 		type = RESPONSE_TYPE_ANSWER;
2675 	}
2676 
2677 	/* handle each of the type cases */
2678 	if(type == RESPONSE_TYPE_ANSWER) {
2679 		/* ANSWER type responses terminate the query algorithm,
2680 		 * so they sent on their */
2681 		if(verbosity >= VERB_DETAIL) {
2682 			verbose(VERB_DETAIL, "query response was %s",
2683 				FLAGS_GET_RCODE(iq->response->rep->flags)
2684 				==LDNS_RCODE_NXDOMAIN?"NXDOMAIN ANSWER":
2685 				(iq->response->rep->an_numrrsets?"ANSWER":
2686 				"nodata ANSWER"));
2687 		}
2688 		/* if qtype is DS, check we have the right level of answer,
2689 		 * like grandchild answer but we need the middle, reject it */
2690 		if(iq->qchase.qtype == LDNS_RR_TYPE_DS && !iq->dsns_point
2691 			&& !(iq->chase_flags&BIT_RD)
2692 			&& iter_ds_toolow(iq->response, iq->dp)
2693 			&& iter_dp_cangodown(&iq->qchase, iq->dp)) {
2694 			/* close down outstanding requests to be discarded */
2695 			outbound_list_clear(&iq->outlist);
2696 			iq->num_current_queries = 0;
2697 			fptr_ok(fptr_whitelist_modenv_detach_subs(
2698 				qstate->env->detach_subs));
2699 			(*qstate->env->detach_subs)(qstate);
2700 			iq->num_target_queries = 0;
2701 			return processDSNSFind(qstate, iq, id);
2702 		}
2703 		if(!qstate->no_cache_store)
2704 			iter_dns_store(qstate->env, &iq->response->qinfo,
2705 				iq->response->rep, 0, qstate->prefetch_leeway,
2706 				iq->dp&&iq->dp->has_parent_side_NS,
2707 				qstate->region, qstate->query_flags);
2708 		/* close down outstanding requests to be discarded */
2709 		outbound_list_clear(&iq->outlist);
2710 		iq->num_current_queries = 0;
2711 		fptr_ok(fptr_whitelist_modenv_detach_subs(
2712 			qstate->env->detach_subs));
2713 		(*qstate->env->detach_subs)(qstate);
2714 		iq->num_target_queries = 0;
2715 		if(qstate->reply)
2716 			sock_list_insert(&qstate->reply_origin,
2717 				&qstate->reply->addr, qstate->reply->addrlen,
2718 				qstate->region);
2719 		if(iq->minimisation_state != DONOT_MINIMISE_STATE
2720 			&& !(iq->chase_flags & BIT_RD)) {
2721 			if(FLAGS_GET_RCODE(iq->response->rep->flags) !=
2722 				LDNS_RCODE_NOERROR) {
2723 				if(qstate->env->cfg->qname_minimisation_strict)
2724 					return final_state(iq);
2725 				/* Best effort qname-minimisation.
2726 				 * Stop minimising and send full query when
2727 				 * RCODE is not NOERROR. */
2728 				iq->minimisation_state = DONOT_MINIMISE_STATE;
2729 			}
2730 			if(FLAGS_GET_RCODE(iq->response->rep->flags) ==
2731 				LDNS_RCODE_NXDOMAIN) {
2732 				/* Stop resolving when NXDOMAIN is DNSSEC
2733 				 * signed. Based on assumption that nameservers
2734 				 * serving signed zones do not return NXDOMAIN
2735 				 * for empty-non-terminals. */
2736 				if(iq->dnssec_expected)
2737 					return final_state(iq);
2738 				/* Make subrequest to validate intermediate
2739 				 * NXDOMAIN if harden-below-nxdomain is
2740 				 * enabled. */
2741 				if(qstate->env->cfg->harden_below_nxdomain) {
2742 					struct module_qstate* subq = NULL;
2743 					log_query_info(VERB_QUERY,
2744 						"schedule NXDOMAIN validation:",
2745 						&iq->response->qinfo);
2746 					if(!generate_sub_request(
2747 						iq->response->qinfo.qname,
2748 						iq->response->qinfo.qname_len,
2749 						iq->response->qinfo.qtype,
2750 						iq->response->qinfo.qclass,
2751 						qstate, id, iq,
2752 						INIT_REQUEST_STATE,
2753 						FINISHED_STATE, &subq, 1))
2754 						verbose(VERB_ALGO,
2755 						"could not validate NXDOMAIN "
2756 						"response");
2757 					outbound_list_clear(&iq->outlist);
2758 					iq->num_current_queries = 0;
2759 					fptr_ok(fptr_whitelist_modenv_detach_subs(
2760 						qstate->env->detach_subs));
2761 					(*qstate->env->detach_subs)(qstate);
2762 					iq->num_target_queries = 0;
2763 				}
2764 			}
2765 			return next_state(iq, QUERYTARGETS_STATE);
2766 		}
2767 		return final_state(iq);
2768 	} else if(type == RESPONSE_TYPE_REFERRAL) {
2769 		/* REFERRAL type responses get a reset of the
2770 		 * delegation point, and back to the QUERYTARGETS_STATE. */
2771 		verbose(VERB_DETAIL, "query response was REFERRAL");
2772 
2773 		if(!(iq->chase_flags & BIT_RD) && !iq->ratelimit_ok) {
2774 			/* we have a referral, no ratelimit, we can send
2775 			 * our queries to the given name */
2776 			infra_ratelimit_dec(qstate->env->infra_cache,
2777 				iq->dp->name, iq->dp->namelen,
2778 				*qstate->env->now);
2779 		}
2780 
2781 		/* if hardened, only store referral if we asked for it */
2782 		if(!qstate->no_cache_store &&
2783 		(!qstate->env->cfg->harden_referral_path ||
2784 		    (  qstate->qinfo.qtype == LDNS_RR_TYPE_NS
2785 			&& (qstate->query_flags&BIT_RD)
2786 			&& !(qstate->query_flags&BIT_CD)
2787 			   /* we know that all other NS rrsets are scrubbed
2788 			    * away, thus on referral only one is left.
2789 			    * see if that equals the query name... */
2790 			&& ( /* auth section, but sometimes in answer section*/
2791 			  reply_find_rrset_section_ns(iq->response->rep,
2792 				iq->qchase.qname, iq->qchase.qname_len,
2793 				LDNS_RR_TYPE_NS, iq->qchase.qclass)
2794 			  || reply_find_rrset_section_an(iq->response->rep,
2795 				iq->qchase.qname, iq->qchase.qname_len,
2796 				LDNS_RR_TYPE_NS, iq->qchase.qclass)
2797 			  )
2798 		    ))) {
2799 			/* Store the referral under the current query */
2800 			/* no prefetch-leeway, since its not the answer */
2801 			iter_dns_store(qstate->env, &iq->response->qinfo,
2802 				iq->response->rep, 1, 0, 0, NULL, 0);
2803 			if(iq->store_parent_NS)
2804 				iter_store_parentside_NS(qstate->env,
2805 					iq->response->rep);
2806 			if(qstate->env->neg_cache)
2807 				val_neg_addreferral(qstate->env->neg_cache,
2808 					iq->response->rep, iq->dp->name);
2809 		}
2810 		/* store parent-side-in-zone-glue, if directly queried for */
2811 		if(!qstate->no_cache_store && iq->query_for_pside_glue
2812 			&& !iq->pside_glue) {
2813 				iq->pside_glue = reply_find_rrset(iq->response->rep,
2814 					iq->qchase.qname, iq->qchase.qname_len,
2815 					iq->qchase.qtype, iq->qchase.qclass);
2816 				if(iq->pside_glue) {
2817 					log_rrset_key(VERB_ALGO, "found parent-side "
2818 						"glue", iq->pside_glue);
2819 					iter_store_parentside_rrset(qstate->env,
2820 						iq->pside_glue);
2821 				}
2822 		}
2823 
2824 		/* Reset the event state, setting the current delegation
2825 		 * point to the referral. */
2826 		iq->deleg_msg = iq->response;
2827 		iq->dp = delegpt_from_message(iq->response, qstate->region);
2828 		if (qstate->env->cfg->qname_minimisation)
2829 			iq->minimisation_state = INIT_MINIMISE_STATE;
2830 		if(!iq->dp) {
2831 			errinf(qstate, "malloc failure, for delegation point");
2832 			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2833 		}
2834 		if(!cache_fill_missing(qstate->env, iq->qchase.qclass,
2835 			qstate->region, iq->dp)) {
2836 			errinf(qstate, "malloc failure, copy extra info into delegation point");
2837 			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2838 		}
2839 		if(iq->store_parent_NS && query_dname_compare(iq->dp->name,
2840 			iq->store_parent_NS->name) == 0)
2841 			iter_merge_retry_counts(iq->dp, iq->store_parent_NS);
2842 		delegpt_log(VERB_ALGO, iq->dp);
2843 		/* Count this as a referral. */
2844 		iq->referral_count++;
2845 		iq->sent_count = 0;
2846 		/* see if the next dp is a trust anchor, or a DS was sent
2847 		 * along, indicating dnssec is expected for next zone */
2848 		iq->dnssec_expected = iter_indicates_dnssec(qstate->env,
2849 			iq->dp, iq->response, iq->qchase.qclass);
2850 		/* if dnssec, validating then also fetch the key for the DS */
2851 		if(iq->dnssec_expected && qstate->env->cfg->prefetch_key &&
2852 			!(qstate->query_flags&BIT_CD))
2853 			generate_dnskey_prefetch(qstate, iq, id);
2854 
2855 		/* spawn off NS and addr to auth servers for the NS we just
2856 		 * got in the referral. This gets authoritative answer
2857 		 * (answer section trust level) rrset.
2858 		 * right after, we detach the subs, answer goes to cache. */
2859 		if(qstate->env->cfg->harden_referral_path)
2860 			generate_ns_check(qstate, iq, id);
2861 
2862 		/* stop current outstanding queries.
2863 		 * FIXME: should the outstanding queries be waited for and
2864 		 * handled? Say by a subquery that inherits the outbound_entry.
2865 		 */
2866 		outbound_list_clear(&iq->outlist);
2867 		iq->num_current_queries = 0;
2868 		fptr_ok(fptr_whitelist_modenv_detach_subs(
2869 			qstate->env->detach_subs));
2870 		(*qstate->env->detach_subs)(qstate);
2871 		iq->num_target_queries = 0;
2872 		verbose(VERB_ALGO, "cleared outbound list for next round");
2873 		return next_state(iq, QUERYTARGETS_STATE);
2874 	} else if(type == RESPONSE_TYPE_CNAME) {
2875 		uint8_t* sname = NULL;
2876 		size_t snamelen = 0;
2877 		/* CNAME type responses get a query restart (i.e., get a
2878 		 * reset of the query state and go back to INIT_REQUEST_STATE).
2879 		 */
2880 		verbose(VERB_DETAIL, "query response was CNAME");
2881 		if(verbosity >= VERB_ALGO)
2882 			log_dns_msg("cname msg", &iq->response->qinfo,
2883 				iq->response->rep);
2884 		/* if qtype is DS, check we have the right level of answer,
2885 		 * like grandchild answer but we need the middle, reject it */
2886 		if(iq->qchase.qtype == LDNS_RR_TYPE_DS && !iq->dsns_point
2887 			&& !(iq->chase_flags&BIT_RD)
2888 			&& iter_ds_toolow(iq->response, iq->dp)
2889 			&& iter_dp_cangodown(&iq->qchase, iq->dp)) {
2890 			outbound_list_clear(&iq->outlist);
2891 			iq->num_current_queries = 0;
2892 			fptr_ok(fptr_whitelist_modenv_detach_subs(
2893 				qstate->env->detach_subs));
2894 			(*qstate->env->detach_subs)(qstate);
2895 			iq->num_target_queries = 0;
2896 			return processDSNSFind(qstate, iq, id);
2897 		}
2898 		/* Process the CNAME response. */
2899 		if(!handle_cname_response(qstate, iq, iq->response,
2900 			&sname, &snamelen)) {
2901 			errinf(qstate, "malloc failure, CNAME info");
2902 			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2903 		}
2904 		/* cache the CNAME response under the current query */
2905 		/* NOTE : set referral=1, so that rrsets get stored but not
2906 		 * the partial query answer (CNAME only). */
2907 		/* prefetchleeway applied because this updates answer parts */
2908 		if(!qstate->no_cache_store)
2909 			iter_dns_store(qstate->env, &iq->response->qinfo,
2910 				iq->response->rep, 1, qstate->prefetch_leeway,
2911 				iq->dp&&iq->dp->has_parent_side_NS, NULL,
2912 				qstate->query_flags);
2913 		/* set the current request's qname to the new value. */
2914 		iq->qchase.qname = sname;
2915 		iq->qchase.qname_len = snamelen;
2916 		/* Clear the query state, since this is a query restart. */
2917 		iq->deleg_msg = NULL;
2918 		iq->dp = NULL;
2919 		iq->dsns_point = NULL;
2920 		iq->auth_zone_response = 0;
2921 		iq->sent_count = 0;
2922 		if(iq->minimisation_state != MINIMISE_STATE)
2923 			/* Only count as query restart when it is not an extra
2924 			 * query as result of qname minimisation. */
2925 			iq->query_restart_count++;
2926 		if(qstate->env->cfg->qname_minimisation)
2927 			iq->minimisation_state = INIT_MINIMISE_STATE;
2928 
2929 		/* stop current outstanding queries.
2930 		 * FIXME: should the outstanding queries be waited for and
2931 		 * handled? Say by a subquery that inherits the outbound_entry.
2932 		 */
2933 		outbound_list_clear(&iq->outlist);
2934 		iq->num_current_queries = 0;
2935 		fptr_ok(fptr_whitelist_modenv_detach_subs(
2936 			qstate->env->detach_subs));
2937 		(*qstate->env->detach_subs)(qstate);
2938 		iq->num_target_queries = 0;
2939 		if(qstate->reply)
2940 			sock_list_insert(&qstate->reply_origin,
2941 				&qstate->reply->addr, qstate->reply->addrlen,
2942 				qstate->region);
2943 		verbose(VERB_ALGO, "cleared outbound list for query restart");
2944 		/* go to INIT_REQUEST_STATE for new qname. */
2945 		return next_state(iq, INIT_REQUEST_STATE);
2946 	} else if(type == RESPONSE_TYPE_LAME) {
2947 		/* Cache the LAMEness. */
2948 		verbose(VERB_DETAIL, "query response was %sLAME",
2949 			dnsseclame?"DNSSEC ":"");
2950 		if(!dname_subdomain_c(iq->qchase.qname, iq->dp->name)) {
2951 			log_err("mark lame: mismatch in qname and dpname");
2952 			/* throwaway this reply below */
2953 		} else if(qstate->reply) {
2954 			/* need addr for lameness cache, but we may have
2955 			 * gotten this from cache, so test to be sure */
2956 			if(!infra_set_lame(qstate->env->infra_cache,
2957 				&qstate->reply->addr, qstate->reply->addrlen,
2958 				iq->dp->name, iq->dp->namelen,
2959 				*qstate->env->now, dnsseclame, 0,
2960 				iq->qchase.qtype))
2961 				log_err("mark host lame: out of memory");
2962 		}
2963 	} else if(type == RESPONSE_TYPE_REC_LAME) {
2964 		/* Cache the LAMEness. */
2965 		verbose(VERB_DETAIL, "query response REC_LAME: "
2966 			"recursive but not authoritative server");
2967 		if(!dname_subdomain_c(iq->qchase.qname, iq->dp->name)) {
2968 			log_err("mark rec_lame: mismatch in qname and dpname");
2969 			/* throwaway this reply below */
2970 		} else if(qstate->reply) {
2971 			/* need addr for lameness cache, but we may have
2972 			 * gotten this from cache, so test to be sure */
2973 			verbose(VERB_DETAIL, "mark as REC_LAME");
2974 			if(!infra_set_lame(qstate->env->infra_cache,
2975 				&qstate->reply->addr, qstate->reply->addrlen,
2976 				iq->dp->name, iq->dp->namelen,
2977 				*qstate->env->now, 0, 1, iq->qchase.qtype))
2978 				log_err("mark host lame: out of memory");
2979 		}
2980 	} else if(type == RESPONSE_TYPE_THROWAWAY) {
2981 		/* LAME and THROWAWAY responses are handled the same way.
2982 		 * In this case, the event is just sent directly back to
2983 		 * the QUERYTARGETS_STATE without resetting anything,
2984 		 * because, clearly, the next target must be tried. */
2985 		verbose(VERB_DETAIL, "query response was THROWAWAY");
2986 	} else {
2987 		log_warn("A query response came back with an unknown type: %d",
2988 			(int)type);
2989 	}
2990 
2991 	/* LAME, THROWAWAY and "unknown" all end up here.
2992 	 * Recycle to the QUERYTARGETS state to hopefully try a
2993 	 * different target. */
2994 	if (qstate->env->cfg->qname_minimisation &&
2995 		!qstate->env->cfg->qname_minimisation_strict)
2996 		iq->minimisation_state = DONOT_MINIMISE_STATE;
2997 	if(iq->auth_zone_response) {
2998 		/* can we fallback? */
2999 		iq->auth_zone_response = 0;
3000 		if(!auth_zones_can_fallback(qstate->env->auth_zones,
3001 			iq->dp->name, iq->dp->namelen, qstate->qinfo.qclass)) {
3002 			verbose(VERB_ALGO, "auth zone response bad, and no"
3003 				" fallback possible, servfail");
3004 			errinf_dname(qstate, "response is bad, no fallback, "
3005 				"for auth zone", iq->dp->name);
3006 			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3007 		}
3008 		verbose(VERB_ALGO, "auth zone response was bad, "
3009 			"fallback enabled");
3010 		iq->auth_zone_avoid = 1;
3011 		if(iq->dp->auth_dp) {
3012 			/* we are using a dp for the auth zone, with no
3013 			 * nameservers, get one first */
3014 			iq->dp = NULL;
3015 			return next_state(iq, INIT_REQUEST_STATE);
3016 		}
3017 	}
3018 	return next_state(iq, QUERYTARGETS_STATE);
3019 }
3020 
3021 /**
3022  * Return priming query results to interested super querystates.
3023  *
3024  * Sets the delegation point and delegation message (not nonRD queries).
3025  * This is a callback from walk_supers.
3026  *
3027  * @param qstate: priming query state that finished.
3028  * @param id: module id.
3029  * @param forq: the qstate for which priming has been done.
3030  */
3031 static void
3032 prime_supers(struct module_qstate* qstate, int id, struct module_qstate* forq)
3033 {
3034 	struct iter_qstate* foriq = (struct iter_qstate*)forq->minfo[id];
3035 	struct delegpt* dp = NULL;
3036 
3037 	log_assert(qstate->is_priming || foriq->wait_priming_stub);
3038 	log_assert(qstate->return_rcode == LDNS_RCODE_NOERROR);
3039 	/* Convert our response to a delegation point */
3040 	dp = delegpt_from_message(qstate->return_msg, forq->region);
3041 	if(!dp) {
3042 		/* if there is no convertable delegation point, then
3043 		 * the ANSWER type was (presumably) a negative answer. */
3044 		verbose(VERB_ALGO, "prime response was not a positive "
3045 			"ANSWER; failing");
3046 		foriq->dp = NULL;
3047 		foriq->state = QUERYTARGETS_STATE;
3048 		return;
3049 	}
3050 
3051 	log_query_info(VERB_DETAIL, "priming successful for", &qstate->qinfo);
3052 	delegpt_log(VERB_ALGO, dp);
3053 	foriq->dp = dp;
3054 	foriq->deleg_msg = dns_copy_msg(qstate->return_msg, forq->region);
3055 	if(!foriq->deleg_msg) {
3056 		log_err("copy prime response: out of memory");
3057 		foriq->dp = NULL;
3058 		foriq->state = QUERYTARGETS_STATE;
3059 		return;
3060 	}
3061 
3062 	/* root priming responses go to init stage 2, priming stub
3063 	 * responses to to stage 3. */
3064 	if(foriq->wait_priming_stub) {
3065 		foriq->state = INIT_REQUEST_3_STATE;
3066 		foriq->wait_priming_stub = 0;
3067 	} else	foriq->state = INIT_REQUEST_2_STATE;
3068 	/* because we are finished, the parent will be reactivated */
3069 }
3070 
3071 /**
3072  * This handles the response to a priming query. This is used to handle both
3073  * root and stub priming responses. This is basically the equivalent of the
3074  * QUERY_RESP_STATE, but will not handle CNAME responses and will treat
3075  * REFERRALs as ANSWERS. It will also update and reactivate the originating
3076  * event.
3077  *
3078  * @param qstate: query state.
3079  * @param id: module id.
3080  * @return true if the event needs more immediate processing, false if not.
3081  *         This state always returns false.
3082  */
3083 static int
3084 processPrimeResponse(struct module_qstate* qstate, int id)
3085 {
3086 	struct iter_qstate* iq = (struct iter_qstate*)qstate->minfo[id];
3087 	enum response_type type;
3088 	iq->response->rep->flags &= ~(BIT_RD|BIT_RA); /* ignore rec-lame */
3089 	type = response_type_from_server(
3090 		(int)((iq->chase_flags&BIT_RD) || iq->chase_to_rd),
3091 		iq->response, &iq->qchase, iq->dp);
3092 	if(type == RESPONSE_TYPE_ANSWER) {
3093 		qstate->return_rcode = LDNS_RCODE_NOERROR;
3094 		qstate->return_msg = iq->response;
3095 	} else {
3096 		errinf(qstate, "prime response did not get an answer");
3097 		errinf_dname(qstate, "for", qstate->qinfo.qname);
3098 		qstate->return_rcode = LDNS_RCODE_SERVFAIL;
3099 		qstate->return_msg = NULL;
3100 	}
3101 
3102 	/* validate the root or stub after priming (if enabled).
3103 	 * This is the same query as the prime query, but with validation.
3104 	 * Now that we are primed, the additional queries that validation
3105 	 * may need can be resolved, such as DLV. */
3106 	if(qstate->env->cfg->harden_referral_path) {
3107 		struct module_qstate* subq = NULL;
3108 		log_nametypeclass(VERB_ALGO, "schedule prime validation",
3109 			qstate->qinfo.qname, qstate->qinfo.qtype,
3110 			qstate->qinfo.qclass);
3111 		if(!generate_sub_request(qstate->qinfo.qname,
3112 			qstate->qinfo.qname_len, qstate->qinfo.qtype,
3113 			qstate->qinfo.qclass, qstate, id, iq,
3114 			INIT_REQUEST_STATE, FINISHED_STATE, &subq, 1)) {
3115 			verbose(VERB_ALGO, "could not generate prime check");
3116 		}
3117 		generate_a_aaaa_check(qstate, iq, id);
3118 	}
3119 
3120 	/* This event is finished. */
3121 	qstate->ext_state[id] = module_finished;
3122 	return 0;
3123 }
3124 
3125 /**
3126  * Do final processing on responses to target queries. Events reach this
3127  * state after the iterative resolution algorithm terminates. This state is
3128  * responsible for reactivating the original event, and housekeeping related
3129  * to received target responses (caching, updating the current delegation
3130  * point, etc).
3131  * Callback from walk_supers for every super state that is interested in
3132  * the results from this query.
3133  *
3134  * @param qstate: query state.
3135  * @param id: module id.
3136  * @param forq: super query state.
3137  */
3138 static void
3139 processTargetResponse(struct module_qstate* qstate, int id,
3140 	struct module_qstate* forq)
3141 {
3142 	struct iter_qstate* iq = (struct iter_qstate*)qstate->minfo[id];
3143 	struct iter_qstate* foriq = (struct iter_qstate*)forq->minfo[id];
3144 	struct ub_packed_rrset_key* rrset;
3145 	struct delegpt_ns* dpns;
3146 	log_assert(qstate->return_rcode == LDNS_RCODE_NOERROR);
3147 
3148 	foriq->state = QUERYTARGETS_STATE;
3149 	log_query_info(VERB_ALGO, "processTargetResponse", &qstate->qinfo);
3150 	log_query_info(VERB_ALGO, "processTargetResponse super", &forq->qinfo);
3151 
3152 	/* Tell the originating event that this target query has finished
3153 	 * (regardless if it succeeded or not). */
3154 	foriq->num_target_queries--;
3155 
3156 	/* check to see if parent event is still interested (in orig name).  */
3157 	if(!foriq->dp) {
3158 		verbose(VERB_ALGO, "subq: parent not interested, was reset");
3159 		return; /* not interested anymore */
3160 	}
3161 	dpns = delegpt_find_ns(foriq->dp, qstate->qinfo.qname,
3162 			qstate->qinfo.qname_len);
3163 	if(!dpns) {
3164 		/* If not interested, just stop processing this event */
3165 		verbose(VERB_ALGO, "subq: parent not interested anymore");
3166 		/* could be because parent was jostled out of the cache,
3167 		   and a new identical query arrived, that does not want it*/
3168 		return;
3169 	}
3170 
3171 	/* if iq->query_for_pside_glue then add the pside_glue (marked lame) */
3172 	if(iq->pside_glue) {
3173 		/* if the pside_glue is NULL, then it could not be found,
3174 		 * the done_pside is already set when created and a cache
3175 		 * entry created in processFinished so nothing to do here */
3176 		log_rrset_key(VERB_ALGO, "add parentside glue to dp",
3177 			iq->pside_glue);
3178 		if(!delegpt_add_rrset(foriq->dp, forq->region,
3179 			iq->pside_glue, 1))
3180 			log_err("out of memory adding pside glue");
3181 	}
3182 
3183 	/* This response is relevant to the current query, so we
3184 	 * add (attempt to add, anyway) this target(s) and reactivate
3185 	 * the original event.
3186 	 * NOTE: we could only look for the AnswerRRset if the
3187 	 * response type was ANSWER. */
3188 	rrset = reply_find_answer_rrset(&iq->qchase, qstate->return_msg->rep);
3189 	if(rrset) {
3190 		/* if CNAMEs have been followed - add new NS to delegpt. */
3191 		/* BTW. RFC 1918 says NS should not have got CNAMEs. Robust. */
3192 		if(!delegpt_find_ns(foriq->dp, rrset->rk.dname,
3193 			rrset->rk.dname_len)) {
3194 			/* if dpns->lame then set newcname ns lame too */
3195 			if(!delegpt_add_ns(foriq->dp, forq->region,
3196 				rrset->rk.dname, dpns->lame))
3197 				log_err("out of memory adding cnamed-ns");
3198 		}
3199 		/* if dpns->lame then set the address(es) lame too */
3200 		if(!delegpt_add_rrset(foriq->dp, forq->region, rrset,
3201 			dpns->lame))
3202 			log_err("out of memory adding targets");
3203 		verbose(VERB_ALGO, "added target response");
3204 		delegpt_log(VERB_ALGO, foriq->dp);
3205 	} else {
3206 		verbose(VERB_ALGO, "iterator TargetResponse failed");
3207 		dpns->resolved = 1; /* fail the target */
3208 	}
3209 }
3210 
3211 /**
3212  * Process response for DS NS Find queries, that attempt to find the delegation
3213  * point where we ask the DS query from.
3214  *
3215  * @param qstate: query state.
3216  * @param id: module id.
3217  * @param forq: super query state.
3218  */
3219 static void
3220 processDSNSResponse(struct module_qstate* qstate, int id,
3221 	struct module_qstate* forq)
3222 {
3223 	struct iter_qstate* foriq = (struct iter_qstate*)forq->minfo[id];
3224 
3225 	/* if the finished (iq->response) query has no NS set: continue
3226 	 * up to look for the right dp; nothing to change, do DPNSstate */
3227 	if(qstate->return_rcode != LDNS_RCODE_NOERROR)
3228 		return; /* seek further */
3229 	/* find the NS RRset (without allowing CNAMEs) */
3230 	if(!reply_find_rrset(qstate->return_msg->rep, qstate->qinfo.qname,
3231 		qstate->qinfo.qname_len, LDNS_RR_TYPE_NS,
3232 		qstate->qinfo.qclass)){
3233 		return; /* seek further */
3234 	}
3235 
3236 	/* else, store as DP and continue at querytargets */
3237 	foriq->state = QUERYTARGETS_STATE;
3238 	foriq->dp = delegpt_from_message(qstate->return_msg, forq->region);
3239 	if(!foriq->dp) {
3240 		log_err("out of memory in dsns dp alloc");
3241 		errinf(qstate, "malloc failure, in DS search");
3242 		return; /* dp==NULL in QUERYTARGETS makes SERVFAIL */
3243 	}
3244 	/* success, go query the querytargets in the new dp (and go down) */
3245 }
3246 
3247 /**
3248  * Process response for qclass=ANY queries for a particular class.
3249  * Append to result or error-exit.
3250  *
3251  * @param qstate: query state.
3252  * @param id: module id.
3253  * @param forq: super query state.
3254  */
3255 static void
3256 processClassResponse(struct module_qstate* qstate, int id,
3257 	struct module_qstate* forq)
3258 {
3259 	struct iter_qstate* foriq = (struct iter_qstate*)forq->minfo[id];
3260 	struct dns_msg* from = qstate->return_msg;
3261 	log_query_info(VERB_ALGO, "processClassResponse", &qstate->qinfo);
3262 	log_query_info(VERB_ALGO, "processClassResponse super", &forq->qinfo);
3263 	if(qstate->return_rcode != LDNS_RCODE_NOERROR) {
3264 		/* cause servfail for qclass ANY query */
3265 		foriq->response = NULL;
3266 		foriq->state = FINISHED_STATE;
3267 		return;
3268 	}
3269 	/* append result */
3270 	if(!foriq->response) {
3271 		/* allocate the response: copy RCODE, sec_state */
3272 		foriq->response = dns_copy_msg(from, forq->region);
3273 		if(!foriq->response) {
3274 			log_err("malloc failed for qclass ANY response");
3275 			foriq->state = FINISHED_STATE;
3276 			return;
3277 		}
3278 		foriq->response->qinfo.qclass = forq->qinfo.qclass;
3279 		/* qclass ANY does not receive the AA flag on replies */
3280 		foriq->response->rep->authoritative = 0;
3281 	} else {
3282 		struct dns_msg* to = foriq->response;
3283 		/* add _from_ this response _to_ existing collection */
3284 		/* if there are records, copy RCODE */
3285 		/* lower sec_state if this message is lower */
3286 		if(from->rep->rrset_count != 0) {
3287 			size_t n = from->rep->rrset_count+to->rep->rrset_count;
3288 			struct ub_packed_rrset_key** dest, **d;
3289 			/* copy appropriate rcode */
3290 			to->rep->flags = from->rep->flags;
3291 			/* copy rrsets */
3292 			if(from->rep->rrset_count > RR_COUNT_MAX ||
3293 				to->rep->rrset_count > RR_COUNT_MAX) {
3294 				log_err("malloc failed (too many rrsets) in collect ANY");
3295 				foriq->state = FINISHED_STATE;
3296 				return; /* integer overflow protection */
3297 			}
3298 			dest = regional_alloc(forq->region, sizeof(dest[0])*n);
3299 			if(!dest) {
3300 				log_err("malloc failed in collect ANY");
3301 				foriq->state = FINISHED_STATE;
3302 				return;
3303 			}
3304 			d = dest;
3305 			/* copy AN */
3306 			memcpy(dest, to->rep->rrsets, to->rep->an_numrrsets
3307 				* sizeof(dest[0]));
3308 			dest += to->rep->an_numrrsets;
3309 			memcpy(dest, from->rep->rrsets, from->rep->an_numrrsets
3310 				* sizeof(dest[0]));
3311 			dest += from->rep->an_numrrsets;
3312 			/* copy NS */
3313 			memcpy(dest, to->rep->rrsets+to->rep->an_numrrsets,
3314 				to->rep->ns_numrrsets * sizeof(dest[0]));
3315 			dest += to->rep->ns_numrrsets;
3316 			memcpy(dest, from->rep->rrsets+from->rep->an_numrrsets,
3317 				from->rep->ns_numrrsets * sizeof(dest[0]));
3318 			dest += from->rep->ns_numrrsets;
3319 			/* copy AR */
3320 			memcpy(dest, to->rep->rrsets+to->rep->an_numrrsets+
3321 				to->rep->ns_numrrsets,
3322 				to->rep->ar_numrrsets * sizeof(dest[0]));
3323 			dest += to->rep->ar_numrrsets;
3324 			memcpy(dest, from->rep->rrsets+from->rep->an_numrrsets+
3325 				from->rep->ns_numrrsets,
3326 				from->rep->ar_numrrsets * sizeof(dest[0]));
3327 			/* update counts */
3328 			to->rep->rrsets = d;
3329 			to->rep->an_numrrsets += from->rep->an_numrrsets;
3330 			to->rep->ns_numrrsets += from->rep->ns_numrrsets;
3331 			to->rep->ar_numrrsets += from->rep->ar_numrrsets;
3332 			to->rep->rrset_count = n;
3333 		}
3334 		if(from->rep->security < to->rep->security) /* lowest sec */
3335 			to->rep->security = from->rep->security;
3336 		if(from->rep->qdcount != 0) /* insert qd if appropriate */
3337 			to->rep->qdcount = from->rep->qdcount;
3338 		if(from->rep->ttl < to->rep->ttl) /* use smallest TTL */
3339 			to->rep->ttl = from->rep->ttl;
3340 		if(from->rep->prefetch_ttl < to->rep->prefetch_ttl)
3341 			to->rep->prefetch_ttl = from->rep->prefetch_ttl;
3342 		if(from->rep->serve_expired_ttl < to->rep->serve_expired_ttl)
3343 			to->rep->serve_expired_ttl = from->rep->serve_expired_ttl;
3344 	}
3345 	/* are we done? */
3346 	foriq->num_current_queries --;
3347 	if(foriq->num_current_queries == 0)
3348 		foriq->state = FINISHED_STATE;
3349 }
3350 
3351 /**
3352  * Collect class ANY responses and make them into one response.  This
3353  * state is started and it creates queries for all classes (that have
3354  * root hints).  The answers are then collected.
3355  *
3356  * @param qstate: query state.
3357  * @param id: module id.
3358  * @return true if the event needs more immediate processing, false if not.
3359  */
3360 static int
3361 processCollectClass(struct module_qstate* qstate, int id)
3362 {
3363 	struct iter_qstate* iq = (struct iter_qstate*)qstate->minfo[id];
3364 	struct module_qstate* subq;
3365 	/* If qchase.qclass == 0 then send out queries for all classes.
3366 	 * Otherwise, do nothing (wait for all answers to arrive and the
3367 	 * processClassResponse to put them together, and that moves us
3368 	 * towards the Finished state when done. */
3369 	if(iq->qchase.qclass == 0) {
3370 		uint16_t c = 0;
3371 		iq->qchase.qclass = LDNS_RR_CLASS_ANY;
3372 		while(iter_get_next_root(qstate->env->hints,
3373 			qstate->env->fwds, &c)) {
3374 			/* generate query for this class */
3375 			log_nametypeclass(VERB_ALGO, "spawn collect query",
3376 				qstate->qinfo.qname, qstate->qinfo.qtype, c);
3377 			if(!generate_sub_request(qstate->qinfo.qname,
3378 				qstate->qinfo.qname_len, qstate->qinfo.qtype,
3379 				c, qstate, id, iq, INIT_REQUEST_STATE,
3380 				FINISHED_STATE, &subq,
3381 				(int)!(qstate->query_flags&BIT_CD))) {
3382 				errinf(qstate, "could not generate class ANY"
3383 					" lookup query");
3384 				return error_response(qstate, id,
3385 					LDNS_RCODE_SERVFAIL);
3386 			}
3387 			/* ignore subq, no special init required */
3388 			iq->num_current_queries ++;
3389 			if(c == 0xffff)
3390 				break;
3391 			else c++;
3392 		}
3393 		/* if no roots are configured at all, return */
3394 		if(iq->num_current_queries == 0) {
3395 			verbose(VERB_ALGO, "No root hints or fwds, giving up "
3396 				"on qclass ANY");
3397 			return error_response(qstate, id, LDNS_RCODE_REFUSED);
3398 		}
3399 		/* return false, wait for queries to return */
3400 	}
3401 	/* if woke up here because of an answer, wait for more answers */
3402 	return 0;
3403 }
3404 
3405 /**
3406  * This handles the final state for first-tier responses (i.e., responses to
3407  * externally generated queries).
3408  *
3409  * @param qstate: query state.
3410  * @param iq: iterator query state.
3411  * @param id: module id.
3412  * @return true if the event needs more processing, false if not. Since this
3413  *         is the final state for an event, it always returns false.
3414  */
3415 static int
3416 processFinished(struct module_qstate* qstate, struct iter_qstate* iq,
3417 	int id)
3418 {
3419 	log_query_info(VERB_QUERY, "finishing processing for",
3420 		&qstate->qinfo);
3421 
3422 	/* store negative cache element for parent side glue. */
3423 	if(!qstate->no_cache_store && iq->query_for_pside_glue
3424 		&& !iq->pside_glue)
3425 			iter_store_parentside_neg(qstate->env, &qstate->qinfo,
3426 				iq->deleg_msg?iq->deleg_msg->rep:
3427 				(iq->response?iq->response->rep:NULL));
3428 	if(!iq->response) {
3429 		verbose(VERB_ALGO, "No response is set, servfail");
3430 		errinf(qstate, "(no response found at query finish)");
3431 		return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3432 	}
3433 
3434 	/* Make sure that the RA flag is set (since the presence of
3435 	 * this module means that recursion is available) */
3436 	iq->response->rep->flags |= BIT_RA;
3437 
3438 	/* Clear the AA flag */
3439 	/* FIXME: does this action go here or in some other module? */
3440 	iq->response->rep->flags &= ~BIT_AA;
3441 
3442 	/* make sure QR flag is on */
3443 	iq->response->rep->flags |= BIT_QR;
3444 
3445 	/* we have finished processing this query */
3446 	qstate->ext_state[id] = module_finished;
3447 
3448 	/* TODO:  we are using a private TTL, trim the response. */
3449 	/* if (mPrivateTTL > 0){IterUtils.setPrivateTTL(resp, mPrivateTTL); } */
3450 
3451 	/* prepend any items we have accumulated */
3452 	if(iq->an_prepend_list || iq->ns_prepend_list) {
3453 		if(!iter_prepend(iq, iq->response, qstate->region)) {
3454 			log_err("prepend rrsets: out of memory");
3455 			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3456 		}
3457 		/* reset the query name back */
3458 		iq->response->qinfo = qstate->qinfo;
3459 		/* the security state depends on the combination */
3460 		iq->response->rep->security = sec_status_unchecked;
3461 		/* store message with the finished prepended items,
3462 		 * but only if we did recursion. The nonrecursion referral
3463 		 * from cache does not need to be stored in the msg cache. */
3464 		if(!qstate->no_cache_store && qstate->query_flags&BIT_RD) {
3465 			iter_dns_store(qstate->env, &qstate->qinfo,
3466 				iq->response->rep, 0, qstate->prefetch_leeway,
3467 				iq->dp&&iq->dp->has_parent_side_NS,
3468 				qstate->region, qstate->query_flags);
3469 		}
3470 	}
3471 	qstate->return_rcode = LDNS_RCODE_NOERROR;
3472 	qstate->return_msg = iq->response;
3473 	return 0;
3474 }
3475 
3476 /*
3477  * Return priming query results to interested super querystates.
3478  *
3479  * Sets the delegation point and delegation message (not nonRD queries).
3480  * This is a callback from walk_supers.
3481  *
3482  * @param qstate: query state that finished.
3483  * @param id: module id.
3484  * @param super: the qstate to inform.
3485  */
3486 void
3487 iter_inform_super(struct module_qstate* qstate, int id,
3488 	struct module_qstate* super)
3489 {
3490 	if(!qstate->is_priming && super->qinfo.qclass == LDNS_RR_CLASS_ANY)
3491 		processClassResponse(qstate, id, super);
3492 	else if(super->qinfo.qtype == LDNS_RR_TYPE_DS && ((struct iter_qstate*)
3493 		super->minfo[id])->state == DSNS_FIND_STATE)
3494 		processDSNSResponse(qstate, id, super);
3495 	else if(qstate->return_rcode != LDNS_RCODE_NOERROR)
3496 		error_supers(qstate, id, super);
3497 	else if(qstate->is_priming)
3498 		prime_supers(qstate, id, super);
3499 	else	processTargetResponse(qstate, id, super);
3500 }
3501 
3502 /**
3503  * Handle iterator state.
3504  * Handle events. This is the real processing loop for events, responsible
3505  * for moving events through the various states. If a processing method
3506  * returns true, then it will be advanced to the next state. If false, then
3507  * processing will stop.
3508  *
3509  * @param qstate: query state.
3510  * @param ie: iterator shared global environment.
3511  * @param iq: iterator query state.
3512  * @param id: module id.
3513  */
3514 static void
3515 iter_handle(struct module_qstate* qstate, struct iter_qstate* iq,
3516 	struct iter_env* ie, int id)
3517 {
3518 	int cont = 1;
3519 	while(cont) {
3520 		verbose(VERB_ALGO, "iter_handle processing q with state %s",
3521 			iter_state_to_string(iq->state));
3522 		switch(iq->state) {
3523 			case INIT_REQUEST_STATE:
3524 				cont = processInitRequest(qstate, iq, ie, id);
3525 				break;
3526 			case INIT_REQUEST_2_STATE:
3527 				cont = processInitRequest2(qstate, iq, id);
3528 				break;
3529 			case INIT_REQUEST_3_STATE:
3530 				cont = processInitRequest3(qstate, iq, id);
3531 				break;
3532 			case QUERYTARGETS_STATE:
3533 				cont = processQueryTargets(qstate, iq, ie, id);
3534 				break;
3535 			case QUERY_RESP_STATE:
3536 				cont = processQueryResponse(qstate, iq, id);
3537 				break;
3538 			case PRIME_RESP_STATE:
3539 				cont = processPrimeResponse(qstate, id);
3540 				break;
3541 			case COLLECT_CLASS_STATE:
3542 				cont = processCollectClass(qstate, id);
3543 				break;
3544 			case DSNS_FIND_STATE:
3545 				cont = processDSNSFind(qstate, iq, id);
3546 				break;
3547 			case FINISHED_STATE:
3548 				cont = processFinished(qstate, iq, id);
3549 				break;
3550 			default:
3551 				log_warn("iterator: invalid state: %d",
3552 					iq->state);
3553 				cont = 0;
3554 				break;
3555 		}
3556 	}
3557 }
3558 
3559 /**
3560  * This is the primary entry point for processing request events. Note that
3561  * this method should only be used by external modules.
3562  * @param qstate: query state.
3563  * @param ie: iterator shared global environment.
3564  * @param iq: iterator query state.
3565  * @param id: module id.
3566  */
3567 static void
3568 process_request(struct module_qstate* qstate, struct iter_qstate* iq,
3569 	struct iter_env* ie, int id)
3570 {
3571 	/* external requests start in the INIT state, and finish using the
3572 	 * FINISHED state. */
3573 	iq->state = INIT_REQUEST_STATE;
3574 	iq->final_state = FINISHED_STATE;
3575 	verbose(VERB_ALGO, "process_request: new external request event");
3576 	iter_handle(qstate, iq, ie, id);
3577 }
3578 
3579 /** process authoritative server reply */
3580 static void
3581 process_response(struct module_qstate* qstate, struct iter_qstate* iq,
3582 	struct iter_env* ie, int id, struct outbound_entry* outbound,
3583 	enum module_ev event)
3584 {
3585 	struct msg_parse* prs;
3586 	struct edns_data edns;
3587 	sldns_buffer* pkt;
3588 
3589 	verbose(VERB_ALGO, "process_response: new external response event");
3590 	iq->response = NULL;
3591 	iq->state = QUERY_RESP_STATE;
3592 	if(event == module_event_noreply || event == module_event_error) {
3593 		if(event == module_event_noreply && iq->sent_count >= 3 &&
3594 			qstate->env->cfg->use_caps_bits_for_id &&
3595 			!iq->caps_fallback) {
3596 			/* start fallback */
3597 			iq->caps_fallback = 1;
3598 			iq->caps_server = 0;
3599 			iq->caps_reply = NULL;
3600 			iq->caps_response = NULL;
3601 			iq->caps_minimisation_state = DONOT_MINIMISE_STATE;
3602 			iq->state = QUERYTARGETS_STATE;
3603 			iq->num_current_queries--;
3604 			/* need fresh attempts for the 0x20 fallback, if
3605 			 * that was the cause for the failure */
3606 			iter_dec_attempts(iq->dp, 3);
3607 			verbose(VERB_DETAIL, "Capsforid: timeouts, starting fallback");
3608 			goto handle_it;
3609 		}
3610 		goto handle_it;
3611 	}
3612 	if( (event != module_event_reply && event != module_event_capsfail)
3613 		|| !qstate->reply) {
3614 		log_err("Bad event combined with response");
3615 		outbound_list_remove(&iq->outlist, outbound);
3616 		errinf(qstate, "module iterator received wrong internal event with a response message");
3617 		(void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3618 		return;
3619 	}
3620 
3621 	/* parse message */
3622 	prs = (struct msg_parse*)regional_alloc(qstate->env->scratch,
3623 		sizeof(struct msg_parse));
3624 	if(!prs) {
3625 		log_err("out of memory on incoming message");
3626 		/* like packet got dropped */
3627 		goto handle_it;
3628 	}
3629 	memset(prs, 0, sizeof(*prs));
3630 	memset(&edns, 0, sizeof(edns));
3631 	pkt = qstate->reply->c->buffer;
3632 	sldns_buffer_set_position(pkt, 0);
3633 	if(parse_packet(pkt, prs, qstate->env->scratch) != LDNS_RCODE_NOERROR) {
3634 		verbose(VERB_ALGO, "parse error on reply packet");
3635 		goto handle_it;
3636 	}
3637 	/* edns is not examined, but removed from message to help cache */
3638 	if(parse_extract_edns(prs, &edns, qstate->env->scratch) !=
3639 		LDNS_RCODE_NOERROR)
3640 		goto handle_it;
3641 
3642 	/* Copy the edns options we may got from the back end */
3643 	if(edns.opt_list) {
3644 		qstate->edns_opts_back_in = edns_opt_copy_region(edns.opt_list,
3645 			qstate->region);
3646 		if(!qstate->edns_opts_back_in) {
3647 			log_err("out of memory on incoming message");
3648 			/* like packet got dropped */
3649 			goto handle_it;
3650 		}
3651 		if(!inplace_cb_edns_back_parsed_call(qstate->env, qstate)) {
3652 			log_err("unable to call edns_back_parsed callback");
3653 			goto handle_it;
3654 		}
3655 	}
3656 
3657 	/* remove CD-bit, we asked for in case we handle validation ourself */
3658 	prs->flags &= ~BIT_CD;
3659 
3660 	/* normalize and sanitize: easy to delete items from linked lists */
3661 	if(!scrub_message(pkt, prs, &iq->qinfo_out, iq->dp->name,
3662 		qstate->env->scratch, qstate->env, ie)) {
3663 		/* if 0x20 enabled, start fallback, but we have no message */
3664 		if(event == module_event_capsfail && !iq->caps_fallback) {
3665 			iq->caps_fallback = 1;
3666 			iq->caps_server = 0;
3667 			iq->caps_reply = NULL;
3668 			iq->caps_response = NULL;
3669 			iq->caps_minimisation_state = DONOT_MINIMISE_STATE;
3670 			iq->state = QUERYTARGETS_STATE;
3671 			iq->num_current_queries--;
3672 			verbose(VERB_DETAIL, "Capsforid: scrub failed, starting fallback with no response");
3673 		}
3674 		goto handle_it;
3675 	}
3676 
3677 	/* allocate response dns_msg in region */
3678 	iq->response = dns_alloc_msg(pkt, prs, qstate->region);
3679 	if(!iq->response)
3680 		goto handle_it;
3681 	log_query_info(VERB_DETAIL, "response for", &qstate->qinfo);
3682 	log_name_addr(VERB_DETAIL, "reply from", iq->dp->name,
3683 		&qstate->reply->addr, qstate->reply->addrlen);
3684 	if(verbosity >= VERB_ALGO)
3685 		log_dns_msg("incoming scrubbed packet:", &iq->response->qinfo,
3686 			iq->response->rep);
3687 
3688 	if(event == module_event_capsfail || iq->caps_fallback) {
3689 		if(qstate->env->cfg->qname_minimisation &&
3690 			iq->minimisation_state != DONOT_MINIMISE_STATE) {
3691 			/* Skip QNAME minimisation for next query, since that
3692 			 * one has to match the current query. */
3693 			iq->minimisation_state = SKIP_MINIMISE_STATE;
3694 		}
3695 		/* for fallback we care about main answer, not additionals */
3696 		/* removing that makes comparison more likely to succeed */
3697 		caps_strip_reply(iq->response->rep);
3698 
3699 		if(iq->caps_fallback &&
3700 			iq->caps_minimisation_state != iq->minimisation_state) {
3701 			/* QNAME minimisation state has changed, restart caps
3702 			 * fallback. */
3703 			iq->caps_fallback = 0;
3704 		}
3705 
3706 		if(!iq->caps_fallback) {
3707 			/* start fallback */
3708 			iq->caps_fallback = 1;
3709 			iq->caps_server = 0;
3710 			iq->caps_reply = iq->response->rep;
3711 			iq->caps_response = iq->response;
3712 			iq->caps_minimisation_state = iq->minimisation_state;
3713 			iq->state = QUERYTARGETS_STATE;
3714 			iq->num_current_queries--;
3715 			verbose(VERB_DETAIL, "Capsforid: starting fallback");
3716 			goto handle_it;
3717 		} else {
3718 			/* check if reply is the same, otherwise, fail */
3719 			if(!iq->caps_reply) {
3720 				iq->caps_reply = iq->response->rep;
3721 				iq->caps_response = iq->response;
3722 				iq->caps_server = -1; /*become zero at ++,
3723 				so that we start the full set of trials */
3724 			} else if(caps_failed_rcode(iq->caps_reply) &&
3725 				!caps_failed_rcode(iq->response->rep)) {
3726 				/* prefer to upgrade to non-SERVFAIL */
3727 				iq->caps_reply = iq->response->rep;
3728 				iq->caps_response = iq->response;
3729 			} else if(!caps_failed_rcode(iq->caps_reply) &&
3730 				caps_failed_rcode(iq->response->rep)) {
3731 				/* if we have non-SERVFAIL as answer then
3732 				 * we can ignore SERVFAILs for the equality
3733 				 * comparison */
3734 				/* no instructions here, skip other else */
3735 			} else if(caps_failed_rcode(iq->caps_reply) &&
3736 				caps_failed_rcode(iq->response->rep)) {
3737 				/* failure is same as other failure in fallbk*/
3738 				/* no instructions here, skip other else */
3739 			} else if(!reply_equal(iq->response->rep, iq->caps_reply,
3740 				qstate->env->scratch)) {
3741 				verbose(VERB_DETAIL, "Capsforid fallback: "
3742 					"getting different replies, failed");
3743 				outbound_list_remove(&iq->outlist, outbound);
3744 				errinf(qstate, "0x20 failed, then got different replies in fallback");
3745 				(void)error_response(qstate, id,
3746 					LDNS_RCODE_SERVFAIL);
3747 				return;
3748 			}
3749 			/* continue the fallback procedure at next server */
3750 			iq->caps_server++;
3751 			iq->state = QUERYTARGETS_STATE;
3752 			iq->num_current_queries--;
3753 			verbose(VERB_DETAIL, "Capsforid: reply is equal. "
3754 				"go to next fallback");
3755 			goto handle_it;
3756 		}
3757 	}
3758 	iq->caps_fallback = 0; /* if we were in fallback, 0x20 is OK now */
3759 
3760 handle_it:
3761 	outbound_list_remove(&iq->outlist, outbound);
3762 	iter_handle(qstate, iq, ie, id);
3763 }
3764 
3765 void
3766 iter_operate(struct module_qstate* qstate, enum module_ev event, int id,
3767 	struct outbound_entry* outbound)
3768 {
3769 	struct iter_env* ie = (struct iter_env*)qstate->env->modinfo[id];
3770 	struct iter_qstate* iq = (struct iter_qstate*)qstate->minfo[id];
3771 	verbose(VERB_QUERY, "iterator[module %d] operate: extstate:%s event:%s",
3772 		id, strextstate(qstate->ext_state[id]), strmodulevent(event));
3773 	if(iq) log_query_info(VERB_QUERY, "iterator operate: query",
3774 		&qstate->qinfo);
3775 	if(iq && qstate->qinfo.qname != iq->qchase.qname)
3776 		log_query_info(VERB_QUERY, "iterator operate: chased to",
3777 			&iq->qchase);
3778 
3779 	/* perform iterator state machine */
3780 	if((event == module_event_new || event == module_event_pass) &&
3781 		iq == NULL) {
3782 		if(!iter_new(qstate, id)) {
3783 			errinf(qstate, "malloc failure, new iterator module allocation");
3784 			(void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3785 			return;
3786 		}
3787 		iq = (struct iter_qstate*)qstate->minfo[id];
3788 		process_request(qstate, iq, ie, id);
3789 		return;
3790 	}
3791 	if(iq && event == module_event_pass) {
3792 		iter_handle(qstate, iq, ie, id);
3793 		return;
3794 	}
3795 	if(iq && outbound) {
3796 		process_response(qstate, iq, ie, id, outbound, event);
3797 		return;
3798 	}
3799 	if(event == module_event_error) {
3800 		verbose(VERB_ALGO, "got called with event error, giving up");
3801 		errinf(qstate, "iterator module got the error event");
3802 		(void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3803 		return;
3804 	}
3805 
3806 	log_err("bad event for iterator");
3807 	errinf(qstate, "iterator module received wrong event");
3808 	(void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3809 }
3810 
3811 void
3812 iter_clear(struct module_qstate* qstate, int id)
3813 {
3814 	struct iter_qstate* iq;
3815 	if(!qstate)
3816 		return;
3817 	iq = (struct iter_qstate*)qstate->minfo[id];
3818 	if(iq) {
3819 		outbound_list_clear(&iq->outlist);
3820 		if(iq->target_count && --iq->target_count[0] == 0)
3821 			free(iq->target_count);
3822 		iq->num_current_queries = 0;
3823 	}
3824 	qstate->minfo[id] = NULL;
3825 }
3826 
3827 size_t
3828 iter_get_mem(struct module_env* env, int id)
3829 {
3830 	struct iter_env* ie = (struct iter_env*)env->modinfo[id];
3831 	if(!ie)
3832 		return 0;
3833 	return sizeof(*ie) + sizeof(int)*((size_t)ie->max_dependency_depth+1)
3834 		+ donotq_get_mem(ie->donotq) + priv_get_mem(ie->priv);
3835 }
3836 
3837 /**
3838  * The iterator function block
3839  */
3840 static struct module_func_block iter_block = {
3841 	"iterator",
3842 	&iter_init, &iter_deinit, &iter_operate, &iter_inform_super,
3843 	&iter_clear, &iter_get_mem
3844 };
3845 
3846 struct module_func_block*
3847 iter_get_funcblock(void)
3848 {
3849 	return &iter_block;
3850 }
3851 
3852 const char*
3853 iter_state_to_string(enum iter_state state)
3854 {
3855 	switch (state)
3856 	{
3857 	case INIT_REQUEST_STATE :
3858 		return "INIT REQUEST STATE";
3859 	case INIT_REQUEST_2_STATE :
3860 		return "INIT REQUEST STATE (stage 2)";
3861 	case INIT_REQUEST_3_STATE:
3862 		return "INIT REQUEST STATE (stage 3)";
3863 	case QUERYTARGETS_STATE :
3864 		return "QUERY TARGETS STATE";
3865 	case PRIME_RESP_STATE :
3866 		return "PRIME RESPONSE STATE";
3867 	case COLLECT_CLASS_STATE :
3868 		return "COLLECT CLASS STATE";
3869 	case DSNS_FIND_STATE :
3870 		return "DSNS FIND STATE";
3871 	case QUERY_RESP_STATE :
3872 		return "QUERY RESPONSE STATE";
3873 	case FINISHED_STATE :
3874 		return "FINISHED RESPONSE STATE";
3875 	default :
3876 		return "UNKNOWN ITER STATE";
3877 	}
3878 }
3879 
3880 int
3881 iter_state_is_responsestate(enum iter_state s)
3882 {
3883 	switch(s) {
3884 		case INIT_REQUEST_STATE :
3885 		case INIT_REQUEST_2_STATE :
3886 		case INIT_REQUEST_3_STATE :
3887 		case QUERYTARGETS_STATE :
3888 		case COLLECT_CLASS_STATE :
3889 			return 0;
3890 		default:
3891 			break;
3892 	}
3893 	return 1;
3894 }
3895