1 /*
2  * iterator/iterator.c - iterative resolver DNS query response module
3  *
4  * Copyright (c) 2007, NLnet Labs. All rights reserved.
5  *
6  * This software is open source.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  *
12  * Redistributions of source code must retain the above copyright notice,
13  * this list of conditions and the following disclaimer.
14  *
15  * Redistributions in binary form must reproduce the above copyright notice,
16  * this list of conditions and the following disclaimer in the documentation
17  * and/or other materials provided with the distribution.
18  *
19  * Neither the name of the NLNET LABS nor the names of its contributors may
20  * be used to endorse or promote products derived from this software without
21  * specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
26  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
27  * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
28  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
29  * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
30  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
31  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
32  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34  */
35 
36 /**
37  * \file
38  *
39  * This file contains a module that performs recursive iterative DNS query
40  * processing.
41  */
42 
43 #include "config.h"
44 #include "iterator/iterator.h"
45 #include "iterator/iter_utils.h"
46 #include "iterator/iter_hints.h"
47 #include "iterator/iter_fwd.h"
48 #include "iterator/iter_donotq.h"
49 #include "iterator/iter_delegpt.h"
50 #include "iterator/iter_resptype.h"
51 #include "iterator/iter_scrub.h"
52 #include "iterator/iter_priv.h"
53 #include "validator/val_neg.h"
54 #include "services/cache/dns.h"
55 #include "services/cache/infra.h"
56 #include "services/authzone.h"
57 #include "util/module.h"
58 #include "util/netevent.h"
59 #include "util/net_help.h"
60 #include "util/regional.h"
61 #include "util/data/dname.h"
62 #include "util/data/msgencode.h"
63 #include "util/fptr_wlist.h"
64 #include "util/config_file.h"
65 #include "util/random.h"
66 #include "sldns/rrdef.h"
67 #include "sldns/wire2str.h"
68 #include "sldns/str2wire.h"
69 #include "sldns/parseutil.h"
70 #include "sldns/sbuffer.h"
71 
72 /* in msec */
73 int UNKNOWN_SERVER_NICENESS = 376;
74 /* in msec */
75 int USEFUL_SERVER_TOP_TIMEOUT = 120000;
76 /* Equals USEFUL_SERVER_TOP_TIMEOUT*4 */
77 int BLACKLIST_PENALTY = (120000*4);
78 
79 static void target_count_increase_nx(struct iter_qstate* iq, int num);
80 
81 int
iter_init(struct module_env * env,int id)82 iter_init(struct module_env* env, int id)
83 {
84 	struct iter_env* iter_env = (struct iter_env*)calloc(1,
85 		sizeof(struct iter_env));
86 	if(!iter_env) {
87 		log_err("malloc failure");
88 		return 0;
89 	}
90 	env->modinfo[id] = (void*)iter_env;
91 
92 	lock_basic_init(&iter_env->queries_ratelimit_lock);
93 	lock_protect(&iter_env->queries_ratelimit_lock,
94 			&iter_env->num_queries_ratelimited,
95 		sizeof(iter_env->num_queries_ratelimited));
96 
97 	if(!iter_apply_cfg(iter_env, env->cfg)) {
98 		log_err("iterator: could not apply configuration settings.");
99 		return 0;
100 	}
101 
102 	return 1;
103 }
104 
105 /** delete caps_whitelist element */
106 static void
caps_free(struct rbnode_type * n,void * ATTR_UNUSED (d))107 caps_free(struct rbnode_type* n, void* ATTR_UNUSED(d))
108 {
109 	if(n) {
110 		free(((struct name_tree_node*)n)->name);
111 		free(n);
112 	}
113 }
114 
115 void
iter_deinit(struct module_env * env,int id)116 iter_deinit(struct module_env* env, int id)
117 {
118 	struct iter_env* iter_env;
119 	if(!env || !env->modinfo[id])
120 		return;
121 	iter_env = (struct iter_env*)env->modinfo[id];
122 	lock_basic_destroy(&iter_env->queries_ratelimit_lock);
123 	free(iter_env->target_fetch_policy);
124 	priv_delete(iter_env->priv);
125 	donotq_delete(iter_env->donotq);
126 	if(iter_env->caps_white) {
127 		traverse_postorder(iter_env->caps_white, caps_free, NULL);
128 		free(iter_env->caps_white);
129 	}
130 	free(iter_env);
131 	env->modinfo[id] = NULL;
132 }
133 
134 /** new query for iterator */
135 static int
iter_new(struct module_qstate * qstate,int id)136 iter_new(struct module_qstate* qstate, int id)
137 {
138 	struct iter_qstate* iq = (struct iter_qstate*)regional_alloc(
139 		qstate->region, sizeof(struct iter_qstate));
140 	qstate->minfo[id] = iq;
141 	if(!iq)
142 		return 0;
143 	memset(iq, 0, sizeof(*iq));
144 	iq->state = INIT_REQUEST_STATE;
145 	iq->final_state = FINISHED_STATE;
146 	iq->an_prepend_list = NULL;
147 	iq->an_prepend_last = NULL;
148 	iq->ns_prepend_list = NULL;
149 	iq->ns_prepend_last = NULL;
150 	iq->dp = NULL;
151 	iq->depth = 0;
152 	iq->num_target_queries = 0;
153 	iq->num_current_queries = 0;
154 	iq->query_restart_count = 0;
155 	iq->referral_count = 0;
156 	iq->sent_count = 0;
157 	iq->ratelimit_ok = 0;
158 	iq->target_count = NULL;
159 	iq->dp_target_count = 0;
160 	iq->wait_priming_stub = 0;
161 	iq->refetch_glue = 0;
162 	iq->dnssec_expected = 0;
163 	iq->dnssec_lame_query = 0;
164 	iq->chase_flags = qstate->query_flags;
165 	/* Start with the (current) qname. */
166 	iq->qchase = qstate->qinfo;
167 	outbound_list_init(&iq->outlist);
168 	iq->minimise_count = 0;
169 	iq->timeout_count = 0;
170 	if (qstate->env->cfg->qname_minimisation)
171 		iq->minimisation_state = INIT_MINIMISE_STATE;
172 	else
173 		iq->minimisation_state = DONOT_MINIMISE_STATE;
174 
175 	memset(&iq->qinfo_out, 0, sizeof(struct query_info));
176 	return 1;
177 }
178 
179 /**
180  * Transition to the next state. This can be used to advance a currently
181  * processing event. It cannot be used to reactivate a forEvent.
182  *
183  * @param iq: iterator query state
184  * @param nextstate The state to transition to.
185  * @return true. This is so this can be called as the return value for the
186  *         actual process*State() methods. (Transitioning to the next state
187  *         implies further processing).
188  */
189 static int
next_state(struct iter_qstate * iq,enum iter_state nextstate)190 next_state(struct iter_qstate* iq, enum iter_state nextstate)
191 {
192 	/* If transitioning to a "response" state, make sure that there is a
193 	 * response */
194 	if(iter_state_is_responsestate(nextstate)) {
195 		if(iq->response == NULL) {
196 			log_err("transitioning to response state sans "
197 				"response.");
198 		}
199 	}
200 	iq->state = nextstate;
201 	return 1;
202 }
203 
204 /**
205  * Transition an event to its final state. Final states always either return
206  * a result up the module chain, or reactivate a dependent event. Which
207  * final state to transition to is set in the module state for the event when
208  * it was created, and depends on the original purpose of the event.
209  *
210  * The response is stored in the qstate->buf buffer.
211  *
212  * @param iq: iterator query state
213  * @return false. This is so this method can be used as the return value for
214  *         the processState methods. (Transitioning to the final state
215  */
216 static int
final_state(struct iter_qstate * iq)217 final_state(struct iter_qstate* iq)
218 {
219 	return next_state(iq, iq->final_state);
220 }
221 
222 /**
223  * Callback routine to handle errors in parent query states
224  * @param qstate: query state that failed.
225  * @param id: module id.
226  * @param super: super state.
227  */
228 static void
error_supers(struct module_qstate * qstate,int id,struct module_qstate * super)229 error_supers(struct module_qstate* qstate, int id, struct module_qstate* super)
230 {
231 	struct iter_env* ie = (struct iter_env*)qstate->env->modinfo[id];
232 	struct iter_qstate* super_iq = (struct iter_qstate*)super->minfo[id];
233 
234 	if(qstate->qinfo.qtype == LDNS_RR_TYPE_A ||
235 		qstate->qinfo.qtype == LDNS_RR_TYPE_AAAA) {
236 		/* mark address as failed. */
237 		struct delegpt_ns* dpns = NULL;
238 		super_iq->num_target_queries--;
239 		if(super_iq->dp)
240 			dpns = delegpt_find_ns(super_iq->dp,
241 				qstate->qinfo.qname, qstate->qinfo.qname_len);
242 		if(!dpns) {
243 			/* not interested */
244 			/* this can happen, for eg. qname minimisation asked
245 			 * for an NXDOMAIN to be validated, and used qtype
246 			 * A for that, and the error of that, the name, is
247 			 * not listed in super_iq->dp */
248 			verbose(VERB_ALGO, "subq error, but not interested");
249 			log_query_info(VERB_ALGO, "superq", &super->qinfo);
250 			return;
251 		} else {
252 			/* see if the failure did get (parent-lame) info */
253 			if(!cache_fill_missing(super->env, super_iq->qchase.qclass,
254 				super->region, super_iq->dp))
255 				log_err("out of memory adding missing");
256 		}
257 		delegpt_mark_neg(dpns, qstate->qinfo.qtype);
258 		if((dpns->got4 == 2 || (!ie->supports_ipv4 && !ie->use_nat64)) &&
259 			(dpns->got6 == 2 || !ie->supports_ipv6)) {
260 			dpns->resolved = 1; /* mark as failed */
261 			target_count_increase_nx(super_iq, 1);
262 		}
263 	}
264 	if(qstate->qinfo.qtype == LDNS_RR_TYPE_NS) {
265 		/* prime failed to get delegation */
266 		super_iq->dp = NULL;
267 	}
268 	/* evaluate targets again */
269 	super_iq->state = QUERYTARGETS_STATE;
270 	/* super becomes runnable, and will process this change */
271 }
272 
273 /**
274  * Return an error to the client
275  * @param qstate: our query state
276  * @param id: module id
277  * @param rcode: error code (DNS errcode).
278  * @return: 0 for use by caller, to make notation easy, like:
279  * 	return error_response(..).
280  */
281 static int
error_response(struct module_qstate * qstate,int id,int rcode)282 error_response(struct module_qstate* qstate, int id, int rcode)
283 {
284 	verbose(VERB_QUERY, "return error response %s",
285 		sldns_lookup_by_id(sldns_rcodes, rcode)?
286 		sldns_lookup_by_id(sldns_rcodes, rcode)->name:"??");
287 	qstate->return_rcode = rcode;
288 	qstate->return_msg = NULL;
289 	qstate->ext_state[id] = module_finished;
290 	return 0;
291 }
292 
293 /**
294  * Return an error to the client and cache the error code in the
295  * message cache (so per qname, qtype, qclass).
296  * @param qstate: our query state
297  * @param id: module id
298  * @param rcode: error code (DNS errcode).
299  * @return: 0 for use by caller, to make notation easy, like:
300  * 	return error_response(..).
301  */
302 static int
error_response_cache(struct module_qstate * qstate,int id,int rcode)303 error_response_cache(struct module_qstate* qstate, int id, int rcode)
304 {
305 	struct reply_info err;
306 	struct msgreply_entry* msg;
307 	if(qstate->no_cache_store) {
308 		return error_response(qstate, id, rcode);
309 	}
310 	if(qstate->prefetch_leeway > NORR_TTL) {
311 		verbose(VERB_ALGO, "error response for prefetch in cache");
312 		/* attempt to adjust the cache entry prefetch */
313 		if(dns_cache_prefetch_adjust(qstate->env, &qstate->qinfo,
314 			NORR_TTL, qstate->query_flags))
315 			return error_response(qstate, id, rcode);
316 		/* if that fails (not in cache), fall through to store err */
317 	}
318 	if((msg=msg_cache_lookup(qstate->env,
319 		qstate->qinfo.qname, qstate->qinfo.qname_len,
320 		qstate->qinfo.qtype, qstate->qinfo.qclass,
321 		qstate->query_flags, 0,
322 		qstate->env->cfg->serve_expired_ttl_reset)) != NULL) {
323 		struct reply_info* rep = (struct reply_info*)msg->entry.data;
324 		if(qstate->env->cfg->serve_expired &&
325 			qstate->env->cfg->serve_expired_ttl_reset && rep &&
326 			*qstate->env->now + qstate->env->cfg->serve_expired_ttl
327 			> rep->serve_expired_ttl) {
328 			verbose(VERB_ALGO, "reset serve-expired-ttl for "
329 				"response in cache");
330 			rep->serve_expired_ttl = *qstate->env->now +
331 				qstate->env->cfg->serve_expired_ttl;
332 		}
333 		if(rep && (FLAGS_GET_RCODE(rep->flags) ==
334 			LDNS_RCODE_NOERROR ||
335 			FLAGS_GET_RCODE(rep->flags) ==
336 			LDNS_RCODE_NXDOMAIN ||
337 			FLAGS_GET_RCODE(rep->flags) ==
338 			LDNS_RCODE_YXDOMAIN) &&
339 			(qstate->env->cfg->serve_expired ||
340 			*qstate->env->now <= rep->ttl)) {
341 			/* we have a good entry, don't overwrite */
342 			lock_rw_unlock(&msg->entry.lock);
343 			return error_response(qstate, id, rcode);
344 		}
345 		lock_rw_unlock(&msg->entry.lock);
346 		/* nothing interesting is cached (already error response or
347 		 * expired good record when we don't serve expired), so this
348 		 * servfail cache entry is useful (stops waste of time on this
349 		 * servfail NORR_TTL) */
350 	}
351 	/* store in cache */
352 	memset(&err, 0, sizeof(err));
353 	err.flags = (uint16_t)(BIT_QR | BIT_RA);
354 	FLAGS_SET_RCODE(err.flags, rcode);
355 	err.qdcount = 1;
356 	err.ttl = NORR_TTL;
357 	err.prefetch_ttl = PREFETCH_TTL_CALC(err.ttl);
358 	err.serve_expired_ttl = NORR_TTL;
359 	/* do not waste time trying to validate this servfail */
360 	err.security = sec_status_indeterminate;
361 	verbose(VERB_ALGO, "store error response in message cache");
362 	iter_dns_store(qstate->env, &qstate->qinfo, &err, 0, 0, 0, NULL,
363 		qstate->query_flags, qstate->qstarttime);
364 	return error_response(qstate, id, rcode);
365 }
366 
367 /** check if prepend item is duplicate item */
368 static int
prepend_is_duplicate(struct ub_packed_rrset_key ** sets,size_t to,struct ub_packed_rrset_key * dup)369 prepend_is_duplicate(struct ub_packed_rrset_key** sets, size_t to,
370 	struct ub_packed_rrset_key* dup)
371 {
372 	size_t i;
373 	for(i=0; i<to; i++) {
374 		if(sets[i]->rk.type == dup->rk.type &&
375 			sets[i]->rk.rrset_class == dup->rk.rrset_class &&
376 			sets[i]->rk.dname_len == dup->rk.dname_len &&
377 			query_dname_compare(sets[i]->rk.dname, dup->rk.dname)
378 			== 0)
379 			return 1;
380 	}
381 	return 0;
382 }
383 
384 /** prepend the prepend list in the answer and authority section of dns_msg */
385 static int
iter_prepend(struct iter_qstate * iq,struct dns_msg * msg,struct regional * region)386 iter_prepend(struct iter_qstate* iq, struct dns_msg* msg,
387 	struct regional* region)
388 {
389 	struct iter_prep_list* p;
390 	struct ub_packed_rrset_key** sets;
391 	size_t num_an = 0, num_ns = 0;;
392 	for(p = iq->an_prepend_list; p; p = p->next)
393 		num_an++;
394 	for(p = iq->ns_prepend_list; p; p = p->next)
395 		num_ns++;
396 	if(num_an + num_ns == 0)
397 		return 1;
398 	verbose(VERB_ALGO, "prepending %d rrsets", (int)num_an + (int)num_ns);
399 	if(num_an > RR_COUNT_MAX || num_ns > RR_COUNT_MAX ||
400 		msg->rep->rrset_count > RR_COUNT_MAX) return 0; /* overflow */
401 	sets = regional_alloc(region, (num_an+num_ns+msg->rep->rrset_count) *
402 		sizeof(struct ub_packed_rrset_key*));
403 	if(!sets)
404 		return 0;
405 	/* ANSWER section */
406 	num_an = 0;
407 	for(p = iq->an_prepend_list; p; p = p->next) {
408 		sets[num_an++] = p->rrset;
409 		if(ub_packed_rrset_ttl(p->rrset) < msg->rep->ttl)
410 			msg->rep->ttl = ub_packed_rrset_ttl(p->rrset);
411 	}
412 	memcpy(sets+num_an, msg->rep->rrsets, msg->rep->an_numrrsets *
413 		sizeof(struct ub_packed_rrset_key*));
414 	/* AUTH section */
415 	num_ns = 0;
416 	for(p = iq->ns_prepend_list; p; p = p->next) {
417 		if(prepend_is_duplicate(sets+msg->rep->an_numrrsets+num_an,
418 			num_ns, p->rrset) || prepend_is_duplicate(
419 			msg->rep->rrsets+msg->rep->an_numrrsets,
420 			msg->rep->ns_numrrsets, p->rrset))
421 			continue;
422 		sets[msg->rep->an_numrrsets + num_an + num_ns++] = p->rrset;
423 		if(ub_packed_rrset_ttl(p->rrset) < msg->rep->ttl)
424 			msg->rep->ttl = ub_packed_rrset_ttl(p->rrset);
425 	}
426 	memcpy(sets + num_an + msg->rep->an_numrrsets + num_ns,
427 		msg->rep->rrsets + msg->rep->an_numrrsets,
428 		(msg->rep->ns_numrrsets + msg->rep->ar_numrrsets) *
429 		sizeof(struct ub_packed_rrset_key*));
430 
431 	/* NXDOMAIN rcode can stay if we prepended DNAME/CNAMEs, because
432 	 * this is what recursors should give. */
433 	msg->rep->rrset_count += num_an + num_ns;
434 	msg->rep->an_numrrsets += num_an;
435 	msg->rep->ns_numrrsets += num_ns;
436 	msg->rep->rrsets = sets;
437 	return 1;
438 }
439 
440 /**
441  * Find rrset in ANSWER prepend list.
442  * to avoid duplicate DNAMEs when a DNAME is traversed twice.
443  * @param iq: iterator query state.
444  * @param rrset: rrset to add.
445  * @return false if not found
446  */
447 static int
iter_find_rrset_in_prepend_answer(struct iter_qstate * iq,struct ub_packed_rrset_key * rrset)448 iter_find_rrset_in_prepend_answer(struct iter_qstate* iq,
449 	struct ub_packed_rrset_key* rrset)
450 {
451 	struct iter_prep_list* p = iq->an_prepend_list;
452 	while(p) {
453 		if(ub_rrset_compare(p->rrset, rrset) == 0 &&
454 			rrsetdata_equal((struct packed_rrset_data*)p->rrset
455 			->entry.data, (struct packed_rrset_data*)rrset
456 			->entry.data))
457 			return 1;
458 		p = p->next;
459 	}
460 	return 0;
461 }
462 
463 /**
464  * Add rrset to ANSWER prepend list
465  * @param qstate: query state.
466  * @param iq: iterator query state.
467  * @param rrset: rrset to add.
468  * @return false on failure (malloc).
469  */
470 static int
iter_add_prepend_answer(struct module_qstate * qstate,struct iter_qstate * iq,struct ub_packed_rrset_key * rrset)471 iter_add_prepend_answer(struct module_qstate* qstate, struct iter_qstate* iq,
472 	struct ub_packed_rrset_key* rrset)
473 {
474 	struct iter_prep_list* p = (struct iter_prep_list*)regional_alloc(
475 		qstate->region, sizeof(struct iter_prep_list));
476 	if(!p)
477 		return 0;
478 	p->rrset = rrset;
479 	p->next = NULL;
480 	/* add at end */
481 	if(iq->an_prepend_last)
482 		iq->an_prepend_last->next = p;
483 	else	iq->an_prepend_list = p;
484 	iq->an_prepend_last = p;
485 	return 1;
486 }
487 
488 /**
489  * Add rrset to AUTHORITY prepend list
490  * @param qstate: query state.
491  * @param iq: iterator query state.
492  * @param rrset: rrset to add.
493  * @return false on failure (malloc).
494  */
495 static int
iter_add_prepend_auth(struct module_qstate * qstate,struct iter_qstate * iq,struct ub_packed_rrset_key * rrset)496 iter_add_prepend_auth(struct module_qstate* qstate, struct iter_qstate* iq,
497 	struct ub_packed_rrset_key* rrset)
498 {
499 	struct iter_prep_list* p = (struct iter_prep_list*)regional_alloc(
500 		qstate->region, sizeof(struct iter_prep_list));
501 	if(!p)
502 		return 0;
503 	p->rrset = rrset;
504 	p->next = NULL;
505 	/* add at end */
506 	if(iq->ns_prepend_last)
507 		iq->ns_prepend_last->next = p;
508 	else	iq->ns_prepend_list = p;
509 	iq->ns_prepend_last = p;
510 	return 1;
511 }
512 
513 /**
514  * Given a CNAME response (defined as a response containing a CNAME or DNAME
515  * that does not answer the request), process the response, modifying the
516  * state as necessary. This follows the CNAME/DNAME chain and returns the
517  * final query name.
518  *
519  * sets the new query name, after following the CNAME/DNAME chain.
520  * @param qstate: query state.
521  * @param iq: iterator query state.
522  * @param msg: the response.
523  * @param mname: returned target new query name.
524  * @param mname_len: length of mname.
525  * @return false on (malloc) error.
526  */
527 static int
handle_cname_response(struct module_qstate * qstate,struct iter_qstate * iq,struct dns_msg * msg,uint8_t ** mname,size_t * mname_len)528 handle_cname_response(struct module_qstate* qstate, struct iter_qstate* iq,
529         struct dns_msg* msg, uint8_t** mname, size_t* mname_len)
530 {
531 	size_t i;
532 	/* Start with the (current) qname. */
533 	*mname = iq->qchase.qname;
534 	*mname_len = iq->qchase.qname_len;
535 
536 	/* Iterate over the ANSWER rrsets in order, looking for CNAMEs and
537 	 * DNAMES. */
538 	for(i=0; i<msg->rep->an_numrrsets; i++) {
539 		struct ub_packed_rrset_key* r = msg->rep->rrsets[i];
540 		/* If there is a (relevant) DNAME, add it to the list.
541 		 * We always expect there to be CNAME that was generated
542 		 * by this DNAME following, so we don't process the DNAME
543 		 * directly.  */
544 		if(ntohs(r->rk.type) == LDNS_RR_TYPE_DNAME &&
545 			dname_strict_subdomain_c(*mname, r->rk.dname) &&
546 			!iter_find_rrset_in_prepend_answer(iq, r)) {
547 			if(!iter_add_prepend_answer(qstate, iq, r))
548 				return 0;
549 			continue;
550 		}
551 
552 		if(ntohs(r->rk.type) == LDNS_RR_TYPE_CNAME &&
553 			query_dname_compare(*mname, r->rk.dname) == 0 &&
554 			!iter_find_rrset_in_prepend_answer(iq, r)) {
555 			/* Add this relevant CNAME rrset to the prepend list.*/
556 			if(!iter_add_prepend_answer(qstate, iq, r))
557 				return 0;
558 			get_cname_target(r, mname, mname_len);
559 		}
560 
561 		/* Other rrsets in the section are ignored. */
562 	}
563 	/* add authority rrsets to authority prepend, for wildcarded CNAMEs */
564 	for(i=msg->rep->an_numrrsets; i<msg->rep->an_numrrsets +
565 		msg->rep->ns_numrrsets; i++) {
566 		struct ub_packed_rrset_key* r = msg->rep->rrsets[i];
567 		/* only add NSEC/NSEC3, as they may be needed for validation */
568 		if(ntohs(r->rk.type) == LDNS_RR_TYPE_NSEC ||
569 			ntohs(r->rk.type) == LDNS_RR_TYPE_NSEC3) {
570 			if(!iter_add_prepend_auth(qstate, iq, r))
571 				return 0;
572 		}
573 	}
574 	return 1;
575 }
576 
577 /** fill fail address for later recovery */
578 static void
fill_fail_addr(struct iter_qstate * iq,struct sockaddr_storage * addr,socklen_t addrlen)579 fill_fail_addr(struct iter_qstate* iq, struct sockaddr_storage* addr,
580 	socklen_t addrlen)
581 {
582 	if(addrlen == 0) {
583 		iq->fail_addr_type = 0;
584 		return;
585 	}
586 	if(((struct sockaddr_in*)addr)->sin_family == AF_INET) {
587 		iq->fail_addr_type = 4;
588 		memcpy(&iq->fail_addr.in,
589 			&((struct sockaddr_in*)addr)->sin_addr,
590 			sizeof(iq->fail_addr.in));
591 	}
592 #ifdef AF_INET6
593 	else if(((struct sockaddr_in*)addr)->sin_family == AF_INET6) {
594 		iq->fail_addr_type = 6;
595 		memcpy(&iq->fail_addr.in6,
596 			&((struct sockaddr_in6*)addr)->sin6_addr,
597 			sizeof(iq->fail_addr.in6));
598 	}
599 #endif
600 	else {
601 		iq->fail_addr_type = 0;
602 	}
603 }
604 
605 /** print fail addr to string */
606 static void
print_fail_addr(struct iter_qstate * iq,char * buf,size_t len)607 print_fail_addr(struct iter_qstate* iq, char* buf, size_t len)
608 {
609 	if(iq->fail_addr_type == 4) {
610 		if(inet_ntop(AF_INET, &iq->fail_addr.in, buf,
611 			(socklen_t)len) == 0)
612 			(void)strlcpy(buf, "(inet_ntop error)", len);
613 	}
614 #ifdef AF_INET6
615 	else if(iq->fail_addr_type == 6) {
616 		if(inet_ntop(AF_INET6, &iq->fail_addr.in6, buf,
617 			(socklen_t)len) == 0)
618 			(void)strlcpy(buf, "(inet_ntop error)", len);
619 	}
620 #endif
621 	else
622 		(void)strlcpy(buf, "", len);
623 }
624 
625 /** add response specific error information for log servfail */
626 static void
errinf_reply(struct module_qstate * qstate,struct iter_qstate * iq)627 errinf_reply(struct module_qstate* qstate, struct iter_qstate* iq)
628 {
629 	if(qstate->env->cfg->val_log_level < 2 && !qstate->env->cfg->log_servfail)
630 		return;
631 	if((qstate->reply && qstate->reply->remote_addrlen != 0) ||
632 		(iq->fail_addr_type != 0)) {
633 		char from[256], frm[512];
634 		if(qstate->reply && qstate->reply->remote_addrlen != 0)
635 			addr_to_str(&qstate->reply->remote_addr,
636 				qstate->reply->remote_addrlen, from,
637 				sizeof(from));
638 		else
639 			print_fail_addr(iq, from, sizeof(from));
640 		snprintf(frm, sizeof(frm), "from %s", from);
641 		errinf(qstate, frm);
642 	}
643 	if(iq->scrub_failures || iq->parse_failures) {
644 		if(iq->scrub_failures)
645 			errinf(qstate, "upstream response failed scrub");
646 		if(iq->parse_failures)
647 			errinf(qstate, "could not parse upstream response");
648 	} else if(iq->response == NULL && iq->timeout_count != 0) {
649 		errinf(qstate, "upstream server timeout");
650 	} else if(iq->response == NULL) {
651 		errinf(qstate, "no server to query");
652 		if(iq->dp) {
653 			if(iq->dp->target_list == NULL)
654 				errinf(qstate, "no addresses for nameservers");
655 			else	errinf(qstate, "nameserver addresses not usable");
656 			if(iq->dp->nslist == NULL)
657 				errinf(qstate, "have no nameserver names");
658 			if(iq->dp->bogus)
659 				errinf(qstate, "NS record was dnssec bogus");
660 		}
661 	}
662 	if(iq->response && iq->response->rep) {
663 		if(FLAGS_GET_RCODE(iq->response->rep->flags) != 0) {
664 			char rcode[256], rc[32];
665 			(void)sldns_wire2str_rcode_buf(
666 				FLAGS_GET_RCODE(iq->response->rep->flags),
667 				rc, sizeof(rc));
668 			snprintf(rcode, sizeof(rcode), "got %s", rc);
669 			errinf(qstate, rcode);
670 		} else {
671 			/* rcode NOERROR */
672 			if(iq->response->rep->an_numrrsets == 0) {
673 				errinf(qstate, "nodata answer");
674 			}
675 		}
676 	}
677 }
678 
679 /** see if last resort is possible - does config allow queries to parent */
680 static int
can_have_last_resort(struct module_env * env,uint8_t * nm,size_t nmlen,uint16_t qclass,struct delegpt ** retdp)681 can_have_last_resort(struct module_env* env, uint8_t* nm, size_t nmlen,
682 	uint16_t qclass, struct delegpt** retdp)
683 {
684 	struct delegpt* fwddp;
685 	struct iter_hints_stub* stub;
686 	int labs = dname_count_labels(nm);
687 	/* do not process a last resort (the parent side) if a stub
688 	 * or forward is configured, because we do not want to go 'above'
689 	 * the configured servers */
690 	if(!dname_is_root(nm) && (stub = (struct iter_hints_stub*)
691 		name_tree_find(&env->hints->tree, nm, nmlen, labs, qclass)) &&
692 		/* has_parent side is turned off for stub_first, where we
693 		 * are allowed to go to the parent */
694 		stub->dp->has_parent_side_NS) {
695 		if(retdp) *retdp = stub->dp;
696 		return 0;
697 	}
698 	if((fwddp = forwards_find(env->fwds, nm, qclass)) &&
699 		/* has_parent_side is turned off for forward_first, where
700 		 * we are allowed to go to the parent */
701 		fwddp->has_parent_side_NS) {
702 		if(retdp) *retdp = fwddp;
703 		return 0;
704 	}
705 	return 1;
706 }
707 
708 /** see if target name is caps-for-id whitelisted */
709 static int
is_caps_whitelisted(struct iter_env * ie,struct iter_qstate * iq)710 is_caps_whitelisted(struct iter_env* ie, struct iter_qstate* iq)
711 {
712 	if(!ie->caps_white) return 0; /* no whitelist, or no capsforid */
713 	return name_tree_lookup(ie->caps_white, iq->qchase.qname,
714 		iq->qchase.qname_len, dname_count_labels(iq->qchase.qname),
715 		iq->qchase.qclass) != NULL;
716 }
717 
718 /**
719  * Create target count structure for this query. This is always explicitly
720  * created for the parent query.
721  */
722 static void
target_count_create(struct iter_qstate * iq)723 target_count_create(struct iter_qstate* iq)
724 {
725 	if(!iq->target_count) {
726 		iq->target_count = (int*)calloc(TARGET_COUNT_MAX, sizeof(int));
727 		/* if calloc fails we simply do not track this number */
728 		if(iq->target_count) {
729 			iq->target_count[TARGET_COUNT_REF] = 1;
730 			iq->nxns_dp = (uint8_t**)calloc(1, sizeof(uint8_t*));
731 		}
732 	}
733 }
734 
735 static void
target_count_increase(struct iter_qstate * iq,int num)736 target_count_increase(struct iter_qstate* iq, int num)
737 {
738 	target_count_create(iq);
739 	if(iq->target_count)
740 		iq->target_count[TARGET_COUNT_QUERIES] += num;
741 	iq->dp_target_count++;
742 }
743 
744 static void
target_count_increase_nx(struct iter_qstate * iq,int num)745 target_count_increase_nx(struct iter_qstate* iq, int num)
746 {
747 	target_count_create(iq);
748 	if(iq->target_count)
749 		iq->target_count[TARGET_COUNT_NX] += num;
750 }
751 
752 /**
753  * Generate a subrequest.
754  * Generate a local request event. Local events are tied to this module, and
755  * have a corresponding (first tier) event that is waiting for this event to
756  * resolve to continue.
757  *
758  * @param qname The query name for this request.
759  * @param qnamelen length of qname
760  * @param qtype The query type for this request.
761  * @param qclass The query class for this request.
762  * @param qstate The event that is generating this event.
763  * @param id: module id.
764  * @param iq: The iterator state that is generating this event.
765  * @param initial_state The initial response state (normally this
766  *          is QUERY_RESP_STATE, unless it is known that the request won't
767  *          need iterative processing
768  * @param finalstate The final state for the response to this request.
769  * @param subq_ret: if newly allocated, the subquerystate, or NULL if it does
770  * 	not need initialisation.
771  * @param v: if true, validation is done on the subquery.
772  * @param detached: true if this qstate should not attach to the subquery
773  * @return false on error (malloc).
774  */
775 static int
generate_sub_request(uint8_t * qname,size_t qnamelen,uint16_t qtype,uint16_t qclass,struct module_qstate * qstate,int id,struct iter_qstate * iq,enum iter_state initial_state,enum iter_state finalstate,struct module_qstate ** subq_ret,int v,int detached)776 generate_sub_request(uint8_t* qname, size_t qnamelen, uint16_t qtype,
777 	uint16_t qclass, struct module_qstate* qstate, int id,
778 	struct iter_qstate* iq, enum iter_state initial_state,
779 	enum iter_state finalstate, struct module_qstate** subq_ret, int v,
780 	int detached)
781 {
782 	struct module_qstate* subq = NULL;
783 	struct iter_qstate* subiq = NULL;
784 	uint16_t qflags = 0; /* OPCODE QUERY, no flags */
785 	struct query_info qinf;
786 	int prime = (finalstate == PRIME_RESP_STATE)?1:0;
787 	int valrec = 0;
788 	qinf.qname = qname;
789 	qinf.qname_len = qnamelen;
790 	qinf.qtype = qtype;
791 	qinf.qclass = qclass;
792 	qinf.local_alias = NULL;
793 
794 	/* RD should be set only when sending the query back through the INIT
795 	 * state. */
796 	if(initial_state == INIT_REQUEST_STATE)
797 		qflags |= BIT_RD;
798 	/* We set the CD flag so we can send this through the "head" of
799 	 * the resolution chain, which might have a validator. We are
800 	 * uninterested in validating things not on the direct resolution
801 	 * path.  */
802 	if(!v) {
803 		qflags |= BIT_CD;
804 		valrec = 1;
805 	}
806 
807 	if(detached) {
808 		struct mesh_state* sub = NULL;
809 		fptr_ok(fptr_whitelist_modenv_add_sub(
810 			qstate->env->add_sub));
811 		if(!(*qstate->env->add_sub)(qstate, &qinf,
812 			qflags, prime, valrec, &subq, &sub)){
813 			return 0;
814 		}
815 	}
816 	else {
817 		/* attach subquery, lookup existing or make a new one */
818 		fptr_ok(fptr_whitelist_modenv_attach_sub(
819 			qstate->env->attach_sub));
820 		if(!(*qstate->env->attach_sub)(qstate, &qinf, qflags, prime,
821 			valrec, &subq)) {
822 			return 0;
823 		}
824 	}
825 	*subq_ret = subq;
826 	if(subq) {
827 		/* initialise the new subquery */
828 		subq->curmod = id;
829 		subq->ext_state[id] = module_state_initial;
830 		subq->minfo[id] = regional_alloc(subq->region,
831 			sizeof(struct iter_qstate));
832 		if(!subq->minfo[id]) {
833 			log_err("init subq: out of memory");
834 			fptr_ok(fptr_whitelist_modenv_kill_sub(
835 				qstate->env->kill_sub));
836 			(*qstate->env->kill_sub)(subq);
837 			return 0;
838 		}
839 		subiq = (struct iter_qstate*)subq->minfo[id];
840 		memset(subiq, 0, sizeof(*subiq));
841 		subiq->num_target_queries = 0;
842 		target_count_create(iq);
843 		subiq->target_count = iq->target_count;
844 		if(iq->target_count) {
845 			iq->target_count[TARGET_COUNT_REF] ++; /* extra reference */
846 			subiq->nxns_dp = iq->nxns_dp;
847 		}
848 		subiq->dp_target_count = 0;
849 		subiq->num_current_queries = 0;
850 		subiq->depth = iq->depth+1;
851 		outbound_list_init(&subiq->outlist);
852 		subiq->state = initial_state;
853 		subiq->final_state = finalstate;
854 		subiq->qchase = subq->qinfo;
855 		subiq->chase_flags = subq->query_flags;
856 		subiq->refetch_glue = 0;
857 		if(qstate->env->cfg->qname_minimisation)
858 			subiq->minimisation_state = INIT_MINIMISE_STATE;
859 		else
860 			subiq->minimisation_state = DONOT_MINIMISE_STATE;
861 		memset(&subiq->qinfo_out, 0, sizeof(struct query_info));
862 	}
863 	return 1;
864 }
865 
866 /**
867  * Generate and send a root priming request.
868  * @param qstate: the qtstate that triggered the need to prime.
869  * @param iq: iterator query state.
870  * @param id: module id.
871  * @param qclass: the class to prime.
872  * @return 0 on failure
873  */
874 static int
prime_root(struct module_qstate * qstate,struct iter_qstate * iq,int id,uint16_t qclass)875 prime_root(struct module_qstate* qstate, struct iter_qstate* iq, int id,
876 	uint16_t qclass)
877 {
878 	struct delegpt* dp;
879 	struct module_qstate* subq;
880 	verbose(VERB_DETAIL, "priming . %s NS",
881 		sldns_lookup_by_id(sldns_rr_classes, (int)qclass)?
882 		sldns_lookup_by_id(sldns_rr_classes, (int)qclass)->name:"??");
883 	dp = hints_lookup_root(qstate->env->hints, qclass);
884 	if(!dp) {
885 		verbose(VERB_ALGO, "Cannot prime due to lack of hints");
886 		return 0;
887 	}
888 	/* Priming requests start at the QUERYTARGETS state, skipping
889 	 * the normal INIT state logic (which would cause an infloop). */
890 	if(!generate_sub_request((uint8_t*)"\000", 1, LDNS_RR_TYPE_NS,
891 		qclass, qstate, id, iq, QUERYTARGETS_STATE, PRIME_RESP_STATE,
892 		&subq, 0, 0)) {
893 		verbose(VERB_ALGO, "could not prime root");
894 		return 0;
895 	}
896 	if(subq) {
897 		struct iter_qstate* subiq =
898 			(struct iter_qstate*)subq->minfo[id];
899 		/* Set the initial delegation point to the hint.
900 		 * copy dp, it is now part of the root prime query.
901 		 * dp was part of in the fixed hints structure. */
902 		subiq->dp = delegpt_copy(dp, subq->region);
903 		if(!subiq->dp) {
904 			log_err("out of memory priming root, copydp");
905 			fptr_ok(fptr_whitelist_modenv_kill_sub(
906 				qstate->env->kill_sub));
907 			(*qstate->env->kill_sub)(subq);
908 			return 0;
909 		}
910 		/* there should not be any target queries. */
911 		subiq->num_target_queries = 0;
912 		subiq->dnssec_expected = iter_indicates_dnssec(
913 			qstate->env, subiq->dp, NULL, subq->qinfo.qclass);
914 	}
915 
916 	/* this module stops, our submodule starts, and does the query. */
917 	qstate->ext_state[id] = module_wait_subquery;
918 	return 1;
919 }
920 
921 /**
922  * Generate and process a stub priming request. This method tests for the
923  * need to prime a stub zone, so it is safe to call for every request.
924  *
925  * @param qstate: the qtstate that triggered the need to prime.
926  * @param iq: iterator query state.
927  * @param id: module id.
928  * @param qname: request name.
929  * @param qclass: request class.
930  * @return true if a priming subrequest was made, false if not. The will only
931  *         issue a priming request if it detects an unprimed stub.
932  *         Uses value of 2 to signal during stub-prime in root-prime situation
933  *         that a noprime-stub is available and resolution can continue.
934  */
935 static int
prime_stub(struct module_qstate * qstate,struct iter_qstate * iq,int id,uint8_t * qname,uint16_t qclass)936 prime_stub(struct module_qstate* qstate, struct iter_qstate* iq, int id,
937 	uint8_t* qname, uint16_t qclass)
938 {
939 	/* Lookup the stub hint. This will return null if the stub doesn't
940 	 * need to be re-primed. */
941 	struct iter_hints_stub* stub;
942 	struct delegpt* stub_dp;
943 	struct module_qstate* subq;
944 
945 	if(!qname) return 0;
946 	stub = hints_lookup_stub(qstate->env->hints, qname, qclass, iq->dp);
947 	/* The stub (if there is one) does not need priming. */
948 	if(!stub)
949 		return 0;
950 	stub_dp = stub->dp;
951 	/* if we have an auth_zone dp, and stub is equal, don't prime stub
952 	 * yet, unless we want to fallback and avoid the auth_zone */
953 	if(!iq->auth_zone_avoid && iq->dp && iq->dp->auth_dp &&
954 		query_dname_compare(iq->dp->name, stub_dp->name) == 0)
955 		return 0;
956 
957 	/* is it a noprime stub (always use) */
958 	if(stub->noprime) {
959 		int r = 0;
960 		if(iq->dp == NULL) r = 2;
961 		/* copy the dp out of the fixed hints structure, so that
962 		 * it can be changed when servicing this query */
963 		iq->dp = delegpt_copy(stub_dp, qstate->region);
964 		if(!iq->dp) {
965 			log_err("out of memory priming stub");
966 			errinf(qstate, "malloc failure, priming stub");
967 			(void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
968 			return 1; /* return 1 to make module stop, with error */
969 		}
970 		log_nametypeclass(VERB_DETAIL, "use stub", stub_dp->name,
971 			LDNS_RR_TYPE_NS, qclass);
972 		return r;
973 	}
974 
975 	/* Otherwise, we need to (re)prime the stub. */
976 	log_nametypeclass(VERB_DETAIL, "priming stub", stub_dp->name,
977 		LDNS_RR_TYPE_NS, qclass);
978 
979 	/* Stub priming events start at the QUERYTARGETS state to avoid the
980 	 * redundant INIT state processing. */
981 	if(!generate_sub_request(stub_dp->name, stub_dp->namelen,
982 		LDNS_RR_TYPE_NS, qclass, qstate, id, iq,
983 		QUERYTARGETS_STATE, PRIME_RESP_STATE, &subq, 0, 0)) {
984 		verbose(VERB_ALGO, "could not prime stub");
985 		errinf(qstate, "could not generate lookup for stub prime");
986 		(void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
987 		return 1; /* return 1 to make module stop, with error */
988 	}
989 	if(subq) {
990 		struct iter_qstate* subiq =
991 			(struct iter_qstate*)subq->minfo[id];
992 
993 		/* Set the initial delegation point to the hint. */
994 		/* make copy to avoid use of stub dp by different qs/threads */
995 		subiq->dp = delegpt_copy(stub_dp, subq->region);
996 		if(!subiq->dp) {
997 			log_err("out of memory priming stub, copydp");
998 			fptr_ok(fptr_whitelist_modenv_kill_sub(
999 				qstate->env->kill_sub));
1000 			(*qstate->env->kill_sub)(subq);
1001 			errinf(qstate, "malloc failure, in stub prime");
1002 			(void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1003 			return 1; /* return 1 to make module stop, with error */
1004 		}
1005 		/* there should not be any target queries -- although there
1006 		 * wouldn't be anyway, since stub hints never have
1007 		 * missing targets. */
1008 		subiq->num_target_queries = 0;
1009 		subiq->wait_priming_stub = 1;
1010 		subiq->dnssec_expected = iter_indicates_dnssec(
1011 			qstate->env, subiq->dp, NULL, subq->qinfo.qclass);
1012 	}
1013 
1014 	/* this module stops, our submodule starts, and does the query. */
1015 	qstate->ext_state[id] = module_wait_subquery;
1016 	return 1;
1017 }
1018 
1019 /**
1020  * Generate a delegation point for an auth zone (unless cached dp is better)
1021  * false on alloc failure.
1022  */
1023 static int
auth_zone_delegpt(struct module_qstate * qstate,struct iter_qstate * iq,uint8_t * delname,size_t delnamelen)1024 auth_zone_delegpt(struct module_qstate* qstate, struct iter_qstate* iq,
1025 	uint8_t* delname, size_t delnamelen)
1026 {
1027 	struct auth_zone* z;
1028 	if(iq->auth_zone_avoid)
1029 		return 1;
1030 	if(!delname) {
1031 		delname = iq->qchase.qname;
1032 		delnamelen = iq->qchase.qname_len;
1033 	}
1034 	lock_rw_rdlock(&qstate->env->auth_zones->lock);
1035 	z = auth_zones_find_zone(qstate->env->auth_zones, delname, delnamelen,
1036 		qstate->qinfo.qclass);
1037 	if(!z) {
1038 		lock_rw_unlock(&qstate->env->auth_zones->lock);
1039 		return 1;
1040 	}
1041 	lock_rw_rdlock(&z->lock);
1042 	lock_rw_unlock(&qstate->env->auth_zones->lock);
1043 	if(z->for_upstream) {
1044 		if(iq->dp && query_dname_compare(z->name, iq->dp->name) == 0
1045 			&& iq->dp->auth_dp && qstate->blacklist &&
1046 			z->fallback_enabled) {
1047 			/* cache is blacklisted and fallback, and we
1048 			 * already have an auth_zone dp */
1049 			if(verbosity>=VERB_ALGO) {
1050 				char buf[255+1];
1051 				dname_str(z->name, buf);
1052 				verbose(VERB_ALGO, "auth_zone %s "
1053 				  "fallback because cache blacklisted",
1054 				  buf);
1055 			}
1056 			lock_rw_unlock(&z->lock);
1057 			iq->dp = NULL;
1058 			return 1;
1059 		}
1060 		if(iq->dp==NULL || dname_subdomain_c(z->name, iq->dp->name)) {
1061 			struct delegpt* dp;
1062 			if(qstate->blacklist && z->fallback_enabled) {
1063 				/* cache is blacklisted because of a DNSSEC
1064 				 * validation failure, and the zone allows
1065 				 * fallback to the internet, query there. */
1066 				if(verbosity>=VERB_ALGO) {
1067 					char buf[255+1];
1068 					dname_str(z->name, buf);
1069 					verbose(VERB_ALGO, "auth_zone %s "
1070 					  "fallback because cache blacklisted",
1071 					  buf);
1072 				}
1073 				lock_rw_unlock(&z->lock);
1074 				return 1;
1075 			}
1076 			dp = (struct delegpt*)regional_alloc_zero(
1077 				qstate->region, sizeof(*dp));
1078 			if(!dp) {
1079 				log_err("alloc failure");
1080 				if(z->fallback_enabled) {
1081 					lock_rw_unlock(&z->lock);
1082 					return 1; /* just fallback */
1083 				}
1084 				lock_rw_unlock(&z->lock);
1085 				errinf(qstate, "malloc failure");
1086 				return 0;
1087 			}
1088 			dp->name = regional_alloc_init(qstate->region,
1089 				z->name, z->namelen);
1090 			if(!dp->name) {
1091 				log_err("alloc failure");
1092 				if(z->fallback_enabled) {
1093 					lock_rw_unlock(&z->lock);
1094 					return 1; /* just fallback */
1095 				}
1096 				lock_rw_unlock(&z->lock);
1097 				errinf(qstate, "malloc failure");
1098 				return 0;
1099 			}
1100 			dp->namelen = z->namelen;
1101 			dp->namelabs = z->namelabs;
1102 			dp->auth_dp = 1;
1103 			iq->dp = dp;
1104 		}
1105 	}
1106 
1107 	lock_rw_unlock(&z->lock);
1108 	return 1;
1109 }
1110 
1111 /**
1112  * Generate A and AAAA checks for glue that is in-zone for the referral
1113  * we just got to obtain authoritative information on the addresses.
1114  *
1115  * @param qstate: the qtstate that triggered the need to prime.
1116  * @param iq: iterator query state.
1117  * @param id: module id.
1118  */
1119 static void
generate_a_aaaa_check(struct module_qstate * qstate,struct iter_qstate * iq,int id)1120 generate_a_aaaa_check(struct module_qstate* qstate, struct iter_qstate* iq,
1121 	int id)
1122 {
1123 	struct iter_env* ie = (struct iter_env*)qstate->env->modinfo[id];
1124 	struct module_qstate* subq;
1125 	size_t i;
1126 	struct reply_info* rep = iq->response->rep;
1127 	struct ub_packed_rrset_key* s;
1128 	log_assert(iq->dp);
1129 
1130 	if(iq->depth == ie->max_dependency_depth)
1131 		return;
1132 	/* walk through additional, and check if in-zone,
1133 	 * only relevant A, AAAA are left after scrub anyway */
1134 	for(i=rep->an_numrrsets+rep->ns_numrrsets; i<rep->rrset_count; i++) {
1135 		s = rep->rrsets[i];
1136 		/* check *ALL* addresses that are transmitted in additional*/
1137 		/* is it an address ? */
1138 		if( !(ntohs(s->rk.type)==LDNS_RR_TYPE_A ||
1139 			ntohs(s->rk.type)==LDNS_RR_TYPE_AAAA)) {
1140 			continue;
1141 		}
1142 		/* is this query the same as the A/AAAA check for it */
1143 		if(qstate->qinfo.qtype == ntohs(s->rk.type) &&
1144 			qstate->qinfo.qclass == ntohs(s->rk.rrset_class) &&
1145 			query_dname_compare(qstate->qinfo.qname,
1146 				s->rk.dname)==0 &&
1147 			(qstate->query_flags&BIT_RD) &&
1148 			!(qstate->query_flags&BIT_CD))
1149 			continue;
1150 
1151 		/* generate subrequest for it */
1152 		log_nametypeclass(VERB_ALGO, "schedule addr fetch",
1153 			s->rk.dname, ntohs(s->rk.type),
1154 			ntohs(s->rk.rrset_class));
1155 		if(!generate_sub_request(s->rk.dname, s->rk.dname_len,
1156 			ntohs(s->rk.type), ntohs(s->rk.rrset_class),
1157 			qstate, id, iq,
1158 			INIT_REQUEST_STATE, FINISHED_STATE, &subq, 1, 0)) {
1159 			verbose(VERB_ALGO, "could not generate addr check");
1160 			return;
1161 		}
1162 		/* ignore subq - not need for more init */
1163 	}
1164 }
1165 
1166 /**
1167  * Generate a NS check request to obtain authoritative information
1168  * on an NS rrset.
1169  *
1170  * @param qstate: the qstate that triggered the need to prime.
1171  * @param iq: iterator query state.
1172  * @param id: module id.
1173  */
1174 static void
generate_ns_check(struct module_qstate * qstate,struct iter_qstate * iq,int id)1175 generate_ns_check(struct module_qstate* qstate, struct iter_qstate* iq, int id)
1176 {
1177 	struct iter_env* ie = (struct iter_env*)qstate->env->modinfo[id];
1178 	struct module_qstate* subq;
1179 	log_assert(iq->dp);
1180 
1181 	if(iq->depth == ie->max_dependency_depth)
1182 		return;
1183 	if(!can_have_last_resort(qstate->env, iq->dp->name, iq->dp->namelen,
1184 		iq->qchase.qclass, NULL))
1185 		return;
1186 	/* is this query the same as the nscheck? */
1187 	if(qstate->qinfo.qtype == LDNS_RR_TYPE_NS &&
1188 		query_dname_compare(iq->dp->name, qstate->qinfo.qname)==0 &&
1189 		(qstate->query_flags&BIT_RD) && !(qstate->query_flags&BIT_CD)){
1190 		/* spawn off A, AAAA queries for in-zone glue to check */
1191 		generate_a_aaaa_check(qstate, iq, id);
1192 		return;
1193 	}
1194 	/* no need to get the NS record for DS, it is above the zonecut */
1195 	if(qstate->qinfo.qtype == LDNS_RR_TYPE_DS)
1196 		return;
1197 
1198 	log_nametypeclass(VERB_ALGO, "schedule ns fetch",
1199 		iq->dp->name, LDNS_RR_TYPE_NS, iq->qchase.qclass);
1200 	if(!generate_sub_request(iq->dp->name, iq->dp->namelen,
1201 		LDNS_RR_TYPE_NS, iq->qchase.qclass, qstate, id, iq,
1202 		INIT_REQUEST_STATE, FINISHED_STATE, &subq, 1, 0)) {
1203 		verbose(VERB_ALGO, "could not generate ns check");
1204 		return;
1205 	}
1206 	if(subq) {
1207 		struct iter_qstate* subiq =
1208 			(struct iter_qstate*)subq->minfo[id];
1209 
1210 		/* make copy to avoid use of stub dp by different qs/threads */
1211 		/* refetch glue to start higher up the tree */
1212 		subiq->refetch_glue = 1;
1213 		subiq->dp = delegpt_copy(iq->dp, subq->region);
1214 		if(!subiq->dp) {
1215 			log_err("out of memory generating ns check, copydp");
1216 			fptr_ok(fptr_whitelist_modenv_kill_sub(
1217 				qstate->env->kill_sub));
1218 			(*qstate->env->kill_sub)(subq);
1219 			return;
1220 		}
1221 	}
1222 }
1223 
1224 /**
1225  * Generate a DNSKEY prefetch query to get the DNSKEY for the DS record we
1226  * just got in a referral (where we have dnssec_expected, thus have trust
1227  * anchors above it).  Note that right after calling this routine the
1228  * iterator detached subqueries (because of following the referral), and thus
1229  * the DNSKEY query becomes detached, its return stored in the cache for
1230  * later lookup by the validator.  This cache lookup by the validator avoids
1231  * the roundtrip incurred by the DNSKEY query.  The DNSKEY query is now
1232  * performed at about the same time the original query is sent to the domain,
1233  * thus the two answers are likely to be returned at about the same time,
1234  * saving a roundtrip from the validated lookup.
1235  *
1236  * @param qstate: the qtstate that triggered the need to prime.
1237  * @param iq: iterator query state.
1238  * @param id: module id.
1239  */
1240 static void
generate_dnskey_prefetch(struct module_qstate * qstate,struct iter_qstate * iq,int id)1241 generate_dnskey_prefetch(struct module_qstate* qstate,
1242 	struct iter_qstate* iq, int id)
1243 {
1244 	struct module_qstate* subq;
1245 	log_assert(iq->dp);
1246 
1247 	/* is this query the same as the prefetch? */
1248 	if(qstate->qinfo.qtype == LDNS_RR_TYPE_DNSKEY &&
1249 		query_dname_compare(iq->dp->name, qstate->qinfo.qname)==0 &&
1250 		(qstate->query_flags&BIT_RD) && !(qstate->query_flags&BIT_CD)){
1251 		return;
1252 	}
1253 	/* we do not generate this prefetch when the query list is full,
1254 	 * the query is fetched, if needed, when the validator wants it.
1255 	 * At that time the validator waits for it, after spawning it.
1256 	 * This means there is one state that uses cpu and a socket, the
1257 	 * spawned while this one waits, and not several at the same time,
1258 	 * if we had created the lookup here. And this helps to keep
1259 	 * the total load down, but the query still succeeds to resolve. */
1260 	if(mesh_jostle_exceeded(qstate->env->mesh))
1261 		return;
1262 
1263 	/* if the DNSKEY is in the cache this lookup will stop quickly */
1264 	log_nametypeclass(VERB_ALGO, "schedule dnskey prefetch",
1265 		iq->dp->name, LDNS_RR_TYPE_DNSKEY, iq->qchase.qclass);
1266 	if(!generate_sub_request(iq->dp->name, iq->dp->namelen,
1267 		LDNS_RR_TYPE_DNSKEY, iq->qchase.qclass, qstate, id, iq,
1268 		INIT_REQUEST_STATE, FINISHED_STATE, &subq, 0, 0)) {
1269 		/* we'll be slower, but it'll work */
1270 		verbose(VERB_ALGO, "could not generate dnskey prefetch");
1271 		return;
1272 	}
1273 	if(subq) {
1274 		struct iter_qstate* subiq =
1275 			(struct iter_qstate*)subq->minfo[id];
1276 		/* this qstate has the right delegation for the dnskey lookup*/
1277 		/* make copy to avoid use of stub dp by different qs/threads */
1278 		subiq->dp = delegpt_copy(iq->dp, subq->region);
1279 		/* if !subiq->dp, it'll start from the cache, no problem */
1280 	}
1281 }
1282 
1283 /**
1284  * See if the query needs forwarding.
1285  *
1286  * @param qstate: query state.
1287  * @param iq: iterator query state.
1288  * @return true if the request is forwarded, false if not.
1289  * 	If returns true but, iq->dp is NULL then a malloc failure occurred.
1290  */
1291 static int
forward_request(struct module_qstate * qstate,struct iter_qstate * iq)1292 forward_request(struct module_qstate* qstate, struct iter_qstate* iq)
1293 {
1294 	struct delegpt* dp;
1295 	uint8_t* delname = iq->qchase.qname;
1296 	size_t delnamelen = iq->qchase.qname_len;
1297 	if(iq->refetch_glue && iq->dp) {
1298 		delname = iq->dp->name;
1299 		delnamelen = iq->dp->namelen;
1300 	}
1301 	/* strip one label off of DS query to lookup higher for it */
1302 	if( (iq->qchase.qtype == LDNS_RR_TYPE_DS || iq->refetch_glue)
1303 		&& !dname_is_root(iq->qchase.qname))
1304 		dname_remove_label(&delname, &delnamelen);
1305 	dp = forwards_lookup(qstate->env->fwds, delname, iq->qchase.qclass);
1306 	if(!dp)
1307 		return 0;
1308 	/* send recursion desired to forward addr */
1309 	iq->chase_flags |= BIT_RD;
1310 	iq->dp = delegpt_copy(dp, qstate->region);
1311 	/* iq->dp checked by caller */
1312 	verbose(VERB_ALGO, "forwarding request");
1313 	return 1;
1314 }
1315 
1316 /**
1317  * Process the initial part of the request handling. This state roughly
1318  * corresponds to resolver algorithms steps 1 (find answer in cache) and 2
1319  * (find the best servers to ask).
1320  *
1321  * Note that all requests start here, and query restarts revisit this state.
1322  *
1323  * This state either generates: 1) a response, from cache or error, 2) a
1324  * priming event, or 3) forwards the request to the next state (init2,
1325  * generally).
1326  *
1327  * @param qstate: query state.
1328  * @param iq: iterator query state.
1329  * @param ie: iterator shared global environment.
1330  * @param id: module id.
1331  * @return true if the event needs more request processing immediately,
1332  *         false if not.
1333  */
1334 static int
processInitRequest(struct module_qstate * qstate,struct iter_qstate * iq,struct iter_env * ie,int id)1335 processInitRequest(struct module_qstate* qstate, struct iter_qstate* iq,
1336 	struct iter_env* ie, int id)
1337 {
1338 	uint8_t* delname, *dpname=NULL;
1339 	size_t delnamelen, dpnamelen=0;
1340 	struct dns_msg* msg = NULL;
1341 
1342 	log_query_info(VERB_DETAIL, "resolving", &qstate->qinfo);
1343 	/* check effort */
1344 
1345 	/* We enforce a maximum number of query restarts. This is primarily a
1346 	 * cheap way to prevent CNAME loops. */
1347 	if(iq->query_restart_count > ie->max_query_restarts) {
1348 		verbose(VERB_QUERY, "request has exceeded the maximum number"
1349 			" of query restarts with %d", iq->query_restart_count);
1350 		errinf(qstate, "request has exceeded the maximum number "
1351 			"restarts (eg. indirections)");
1352 		if(iq->qchase.qname)
1353 			errinf_dname(qstate, "stop at", iq->qchase.qname);
1354 		return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1355 	}
1356 
1357 	/* We enforce a maximum recursion/dependency depth -- in general,
1358 	 * this is unnecessary for dependency loops (although it will
1359 	 * catch those), but it provides a sensible limit to the amount
1360 	 * of work required to answer a given query. */
1361 	verbose(VERB_ALGO, "request has dependency depth of %d", iq->depth);
1362 	if(iq->depth > ie->max_dependency_depth) {
1363 		verbose(VERB_QUERY, "request has exceeded the maximum "
1364 			"dependency depth with depth of %d", iq->depth);
1365 		errinf(qstate, "request has exceeded the maximum dependency "
1366 			"depth (eg. nameserver lookup recursion)");
1367 		return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1368 	}
1369 
1370 	/* If the request is qclass=ANY, setup to generate each class */
1371 	if(qstate->qinfo.qclass == LDNS_RR_CLASS_ANY) {
1372 		iq->qchase.qclass = 0;
1373 		return next_state(iq, COLLECT_CLASS_STATE);
1374 	}
1375 
1376 	/*
1377 	 * If we are restricted by a forward-zone or a stub-zone, we
1378 	 * can't re-fetch glue for this delegation point.
1379 	 * we won’t try to re-fetch glue if the iq->dp is null.
1380 	 */
1381 	if (iq->refetch_glue &&
1382 	        iq->dp &&
1383 	        !can_have_last_resort(qstate->env, iq->dp->name,
1384 	             iq->dp->namelen, iq->qchase.qclass, NULL)) {
1385 	    iq->refetch_glue = 0;
1386 	}
1387 
1388 	/* Resolver Algorithm Step 1 -- Look for the answer in local data. */
1389 
1390 	/* This either results in a query restart (CNAME cache response), a
1391 	 * terminating response (ANSWER), or a cache miss (null). */
1392 
1393 	if (iter_stub_fwd_no_cache(qstate, &iq->qchase, &dpname, &dpnamelen)) {
1394 		/* Asked to not query cache. */
1395 		verbose(VERB_ALGO, "no-cache set, going to the network");
1396 		qstate->no_cache_lookup = 1;
1397 		qstate->no_cache_store = 1;
1398 		msg = NULL;
1399 	} else if(qstate->blacklist) {
1400 		/* if cache, or anything else, was blacklisted then
1401 		 * getting older results from cache is a bad idea, no cache */
1402 		verbose(VERB_ALGO, "cache blacklisted, going to the network");
1403 		msg = NULL;
1404 	} else if(!qstate->no_cache_lookup) {
1405 		msg = dns_cache_lookup(qstate->env, iq->qchase.qname,
1406 			iq->qchase.qname_len, iq->qchase.qtype,
1407 			iq->qchase.qclass, qstate->query_flags,
1408 			qstate->region, qstate->env->scratch, 0, dpname,
1409 			dpnamelen);
1410 		if(!msg && qstate->env->neg_cache &&
1411 			iter_qname_indicates_dnssec(qstate->env, &iq->qchase)) {
1412 			/* lookup in negative cache; may result in
1413 			 * NOERROR/NODATA or NXDOMAIN answers that need validation */
1414 			msg = val_neg_getmsg(qstate->env->neg_cache, &iq->qchase,
1415 				qstate->region, qstate->env->rrset_cache,
1416 				qstate->env->scratch_buffer,
1417 				*qstate->env->now, 1/*add SOA*/, NULL,
1418 				qstate->env->cfg);
1419 		}
1420 		/* item taken from cache does not match our query name, thus
1421 		 * security needs to be re-examined later */
1422 		if(msg && query_dname_compare(qstate->qinfo.qname,
1423 			iq->qchase.qname) != 0)
1424 			msg->rep->security = sec_status_unchecked;
1425 	}
1426 	if(msg) {
1427 		/* handle positive cache response */
1428 		enum response_type type = response_type_from_cache(msg,
1429 			&iq->qchase);
1430 		if(verbosity >= VERB_ALGO) {
1431 			log_dns_msg("msg from cache lookup", &msg->qinfo,
1432 				msg->rep);
1433 			verbose(VERB_ALGO, "msg ttl is %d, prefetch ttl %d",
1434 				(int)msg->rep->ttl,
1435 				(int)msg->rep->prefetch_ttl);
1436 		}
1437 
1438 		if(type == RESPONSE_TYPE_CNAME) {
1439 			uint8_t* sname = 0;
1440 			size_t slen = 0;
1441 			verbose(VERB_ALGO, "returning CNAME response from "
1442 				"cache");
1443 			if(!handle_cname_response(qstate, iq, msg,
1444 				&sname, &slen)) {
1445 				errinf(qstate, "failed to prepend CNAME "
1446 					"components, malloc failure");
1447 				return error_response(qstate, id,
1448 					LDNS_RCODE_SERVFAIL);
1449 			}
1450 			iq->qchase.qname = sname;
1451 			iq->qchase.qname_len = slen;
1452 			if(qstate->env->auth_zones) {
1453 				/* apply rpz qname triggers after cname */
1454 				struct dns_msg* forged_response =
1455 					rpz_callback_from_iterator_cname(qstate, iq);
1456 				while(forged_response && reply_find_rrset_section_an(
1457 					forged_response->rep, iq->qchase.qname,
1458 					iq->qchase.qname_len, LDNS_RR_TYPE_CNAME,
1459 					iq->qchase.qclass)) {
1460 					/* another cname to follow */
1461 					if(!handle_cname_response(qstate, iq, forged_response,
1462 						&sname, &slen)) {
1463 						errinf(qstate, "malloc failure, CNAME info");
1464 						return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1465 					}
1466 					iq->qchase.qname = sname;
1467 					iq->qchase.qname_len = slen;
1468 					forged_response =
1469 						rpz_callback_from_iterator_cname(qstate, iq);
1470 				}
1471 				if(forged_response != NULL) {
1472 					qstate->ext_state[id] = module_finished;
1473 					qstate->return_rcode = LDNS_RCODE_NOERROR;
1474 					qstate->return_msg = forged_response;
1475 					iq->response = forged_response;
1476 					next_state(iq, FINISHED_STATE);
1477 					if(!iter_prepend(iq, qstate->return_msg, qstate->region)) {
1478 						log_err("rpz: after cached cname, prepend rrsets: out of memory");
1479 						return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1480 					}
1481 					qstate->return_msg->qinfo = qstate->qinfo;
1482 					return 0;
1483 				}
1484 			}
1485 			/* This *is* a query restart, even if it is a cheap
1486 			 * one. */
1487 			iq->dp = NULL;
1488 			iq->refetch_glue = 0;
1489 			iq->query_restart_count++;
1490 			iq->sent_count = 0;
1491 			iq->dp_target_count = 0;
1492 			sock_list_insert(&qstate->reply_origin, NULL, 0, qstate->region);
1493 			if(qstate->env->cfg->qname_minimisation)
1494 				iq->minimisation_state = INIT_MINIMISE_STATE;
1495 			return next_state(iq, INIT_REQUEST_STATE);
1496 		}
1497 
1498 		/* if from cache, NULL, else insert 'cache IP' len=0 */
1499 		if(qstate->reply_origin)
1500 			sock_list_insert(&qstate->reply_origin, NULL, 0, qstate->region);
1501 		if(FLAGS_GET_RCODE(msg->rep->flags) == LDNS_RCODE_SERVFAIL)
1502 			errinf(qstate, "SERVFAIL in cache");
1503 		/* it is an answer, response, to final state */
1504 		verbose(VERB_ALGO, "returning answer from cache.");
1505 		iq->response = msg;
1506 		return final_state(iq);
1507 	}
1508 
1509 	/* attempt to forward the request */
1510 	if(forward_request(qstate, iq))
1511 	{
1512 		if(!iq->dp) {
1513 			log_err("alloc failure for forward dp");
1514 			errinf(qstate, "malloc failure for forward zone");
1515 			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1516 		}
1517 		if((qstate->query_flags&BIT_RD)==0) {
1518 			/* If the server accepts RD=0 queries and forwards
1519 			 * with RD=1, then if the server is listed as an NS
1520 			 * entry, it starts query loops. Stop that loop by
1521 			 * disallowing the query. The RD=0 was previously used
1522 			 * to check the cache with allow_snoop. For stubs,
1523 			 * the iterator pass would have primed the stub and
1524 			 * then cached information can be used for further
1525 			 * queries. */
1526 			verbose(VERB_ALGO, "cannot forward RD=0 query, to stop query loops");
1527 			errinf(qstate, "cannot forward RD=0 query");
1528 			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1529 		}
1530 		iq->refetch_glue = 0;
1531 		iq->minimisation_state = DONOT_MINIMISE_STATE;
1532 		/* the request has been forwarded.
1533 		 * forwarded requests need to be immediately sent to the
1534 		 * next state, QUERYTARGETS. */
1535 		return next_state(iq, QUERYTARGETS_STATE);
1536 	}
1537 
1538 	/* Resolver Algorithm Step 2 -- find the "best" servers. */
1539 
1540 	/* first, adjust for DS queries. To avoid the grandparent problem,
1541 	 * we just look for the closest set of server to the parent of qname.
1542 	 * When re-fetching glue we also need to ask the parent.
1543 	 */
1544 	if(iq->refetch_glue) {
1545 		if(!iq->dp) {
1546 			log_err("internal or malloc fail: no dp for refetch");
1547 			errinf(qstate, "malloc failure, for delegation info");
1548 			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1549 		}
1550 		delname = iq->dp->name;
1551 		delnamelen = iq->dp->namelen;
1552 	} else {
1553 		delname = iq->qchase.qname;
1554 		delnamelen = iq->qchase.qname_len;
1555 	}
1556 	if(iq->qchase.qtype == LDNS_RR_TYPE_DS || iq->refetch_glue ||
1557 	   (iq->qchase.qtype == LDNS_RR_TYPE_NS && qstate->prefetch_leeway
1558 	   && can_have_last_resort(qstate->env, delname, delnamelen, iq->qchase.qclass, NULL))) {
1559 		/* remove first label from delname, root goes to hints,
1560 		 * but only to fetch glue, not for qtype=DS. */
1561 		/* also when prefetching an NS record, fetch it again from
1562 		 * its parent, just as if it expired, so that you do not
1563 		 * get stuck on an older nameserver that gives old NSrecords */
1564 		if(dname_is_root(delname) && (iq->refetch_glue ||
1565 			(iq->qchase.qtype == LDNS_RR_TYPE_NS &&
1566 			qstate->prefetch_leeway)))
1567 			delname = NULL; /* go to root priming */
1568 		else 	dname_remove_label(&delname, &delnamelen);
1569 	}
1570 	/* delname is the name to lookup a delegation for. If NULL rootprime */
1571 	while(1) {
1572 
1573 		/* Lookup the delegation in the cache. If null, then the
1574 		 * cache needs to be primed for the qclass. */
1575 		if(delname)
1576 		     iq->dp = dns_cache_find_delegation(qstate->env, delname,
1577 			delnamelen, iq->qchase.qtype, iq->qchase.qclass,
1578 			qstate->region, &iq->deleg_msg,
1579 			*qstate->env->now+qstate->prefetch_leeway, 1,
1580 			dpname, dpnamelen);
1581 		else iq->dp = NULL;
1582 
1583 		/* If the cache has returned nothing, then we have a
1584 		 * root priming situation. */
1585 		if(iq->dp == NULL) {
1586 			int r;
1587 			/* if under auth zone, no prime needed */
1588 			if(!auth_zone_delegpt(qstate, iq, delname, delnamelen))
1589 				return error_response(qstate, id,
1590 					LDNS_RCODE_SERVFAIL);
1591 			if(iq->dp) /* use auth zone dp */
1592 				return next_state(iq, INIT_REQUEST_2_STATE);
1593 			/* if there is a stub, then no root prime needed */
1594 			r = prime_stub(qstate, iq, id, delname,
1595 				iq->qchase.qclass);
1596 			if(r == 2)
1597 				break; /* got noprime-stub-zone, continue */
1598 			else if(r)
1599 				return 0; /* stub prime request made */
1600 			if(forwards_lookup_root(qstate->env->fwds,
1601 				iq->qchase.qclass)) {
1602 				/* forward zone root, no root prime needed */
1603 				/* fill in some dp - safety belt */
1604 				iq->dp = hints_lookup_root(qstate->env->hints,
1605 					iq->qchase.qclass);
1606 				if(!iq->dp) {
1607 					log_err("internal error: no hints dp");
1608 					errinf(qstate, "no hints for this class");
1609 					return error_response(qstate, id,
1610 						LDNS_RCODE_SERVFAIL);
1611 				}
1612 				iq->dp = delegpt_copy(iq->dp, qstate->region);
1613 				if(!iq->dp) {
1614 					log_err("out of memory in safety belt");
1615 					errinf(qstate, "malloc failure, in safety belt");
1616 					return error_response(qstate, id,
1617 						LDNS_RCODE_SERVFAIL);
1618 				}
1619 				return next_state(iq, INIT_REQUEST_2_STATE);
1620 			}
1621 			/* Note that the result of this will set a new
1622 			 * DelegationPoint based on the result of priming. */
1623 			if(!prime_root(qstate, iq, id, iq->qchase.qclass))
1624 				return error_response(qstate, id,
1625 					LDNS_RCODE_REFUSED);
1626 
1627 			/* priming creates and sends a subordinate query, with
1628 			 * this query as the parent. So further processing for
1629 			 * this event will stop until reactivated by the
1630 			 * results of priming. */
1631 			return 0;
1632 		}
1633 		if(!iq->ratelimit_ok && qstate->prefetch_leeway)
1634 			iq->ratelimit_ok = 1; /* allow prefetches, this keeps
1635 			otherwise valid data in the cache */
1636 
1637 		/* see if this dp not useless.
1638 		 * It is useless if:
1639 		 *	o all NS items are required glue.
1640 		 *	  or the query is for NS item that is required glue.
1641 		 *	o no addresses are provided.
1642 		 *	o RD qflag is on.
1643 		 * Instead, go up one level, and try to get even further
1644 		 * If the root was useless, use safety belt information.
1645 		 * Only check cache returns, because replies for servers
1646 		 * could be useless but lead to loops (bumping into the
1647 		 * same server reply) if useless-checked.
1648 		 */
1649 		if(iter_dp_is_useless(&qstate->qinfo, qstate->query_flags,
1650 			iq->dp, ie->supports_ipv4, ie->supports_ipv6,
1651 			ie->use_nat64)) {
1652 			struct delegpt* retdp = NULL;
1653 			if(!can_have_last_resort(qstate->env, iq->dp->name, iq->dp->namelen, iq->qchase.qclass, &retdp)) {
1654 				if(retdp) {
1655 					verbose(VERB_QUERY, "cache has stub "
1656 						"or fwd but no addresses, "
1657 						"fallback to config");
1658 					iq->dp = delegpt_copy(retdp,
1659 						qstate->region);
1660 					if(!iq->dp) {
1661 						log_err("out of memory in "
1662 							"stub/fwd fallback");
1663 						errinf(qstate, "malloc failure, for fallback to config");
1664 						return error_response(qstate,
1665 						    id, LDNS_RCODE_SERVFAIL);
1666 					}
1667 					break;
1668 				}
1669 				verbose(VERB_ALGO, "useless dp "
1670 					"but cannot go up, servfail");
1671 				delegpt_log(VERB_ALGO, iq->dp);
1672 				errinf(qstate, "no useful nameservers, "
1673 					"and cannot go up");
1674 				errinf_dname(qstate, "for zone", iq->dp->name);
1675 				return error_response(qstate, id,
1676 					LDNS_RCODE_SERVFAIL);
1677 			}
1678 			if(dname_is_root(iq->dp->name)) {
1679 				/* use safety belt */
1680 				verbose(VERB_QUERY, "Cache has root NS but "
1681 				"no addresses. Fallback to the safety belt.");
1682 				iq->dp = hints_lookup_root(qstate->env->hints,
1683 					iq->qchase.qclass);
1684 				/* note deleg_msg is from previous lookup,
1685 				 * but RD is on, so it is not used */
1686 				if(!iq->dp) {
1687 					log_err("internal error: no hints dp");
1688 					return error_response(qstate, id,
1689 						LDNS_RCODE_REFUSED);
1690 				}
1691 				iq->dp = delegpt_copy(iq->dp, qstate->region);
1692 				if(!iq->dp) {
1693 					log_err("out of memory in safety belt");
1694 					errinf(qstate, "malloc failure, in safety belt, for root");
1695 					return error_response(qstate, id,
1696 						LDNS_RCODE_SERVFAIL);
1697 				}
1698 				break;
1699 			} else {
1700 				verbose(VERB_ALGO,
1701 					"cache delegation was useless:");
1702 				delegpt_log(VERB_ALGO, iq->dp);
1703 				/* go up */
1704 				delname = iq->dp->name;
1705 				delnamelen = iq->dp->namelen;
1706 				dname_remove_label(&delname, &delnamelen);
1707 			}
1708 		} else break;
1709 	}
1710 
1711 	verbose(VERB_ALGO, "cache delegation returns delegpt");
1712 	delegpt_log(VERB_ALGO, iq->dp);
1713 
1714 	/* Otherwise, set the current delegation point and move on to the
1715 	 * next state. */
1716 	return next_state(iq, INIT_REQUEST_2_STATE);
1717 }
1718 
1719 /**
1720  * Process the second part of the initial request handling. This state
1721  * basically exists so that queries that generate root priming events have
1722  * the same init processing as ones that do not. Request events that reach
1723  * this state must have a valid currentDelegationPoint set.
1724  *
1725  * This part is primarily handling stub zone priming. Events that reach this
1726  * state must have a current delegation point.
1727  *
1728  * @param qstate: query state.
1729  * @param iq: iterator query state.
1730  * @param id: module id.
1731  * @return true if the event needs more request processing immediately,
1732  *         false if not.
1733  */
1734 static int
processInitRequest2(struct module_qstate * qstate,struct iter_qstate * iq,int id)1735 processInitRequest2(struct module_qstate* qstate, struct iter_qstate* iq,
1736 	int id)
1737 {
1738 	uint8_t* delname;
1739 	size_t delnamelen;
1740 	log_query_info(VERB_QUERY, "resolving (init part 2): ",
1741 		&qstate->qinfo);
1742 
1743 	delname = iq->qchase.qname;
1744 	delnamelen = iq->qchase.qname_len;
1745 	if(iq->refetch_glue) {
1746 		struct iter_hints_stub* stub;
1747 		if(!iq->dp) {
1748 			log_err("internal or malloc fail: no dp for refetch");
1749 			errinf(qstate, "malloc failure, no delegation info");
1750 			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1751 		}
1752 		/* Do not send queries above stub, do not set delname to dp if
1753 		 * this is above stub without stub-first. */
1754 		stub = hints_lookup_stub(
1755 			qstate->env->hints, iq->qchase.qname, iq->qchase.qclass,
1756 			iq->dp);
1757 		if(!stub || !stub->dp->has_parent_side_NS ||
1758 			dname_subdomain_c(iq->dp->name, stub->dp->name)) {
1759 			delname = iq->dp->name;
1760 			delnamelen = iq->dp->namelen;
1761 		}
1762 	}
1763 	if(iq->qchase.qtype == LDNS_RR_TYPE_DS || iq->refetch_glue) {
1764 		if(!dname_is_root(delname))
1765 			dname_remove_label(&delname, &delnamelen);
1766 		iq->refetch_glue = 0; /* if CNAME causes restart, no refetch */
1767 	}
1768 
1769 	/* see if we have an auth zone to answer from, improves dp from cache
1770 	 * (if any dp from cache) with auth zone dp, if that is lower */
1771 	if(!auth_zone_delegpt(qstate, iq, delname, delnamelen))
1772 		return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
1773 
1774 	/* Check to see if we need to prime a stub zone. */
1775 	if(prime_stub(qstate, iq, id, delname, iq->qchase.qclass)) {
1776 		/* A priming sub request was made */
1777 		return 0;
1778 	}
1779 
1780 	/* most events just get forwarded to the next state. */
1781 	return next_state(iq, INIT_REQUEST_3_STATE);
1782 }
1783 
1784 /**
1785  * Process the third part of the initial request handling. This state exists
1786  * as a separate state so that queries that generate stub priming events
1787  * will get the tail end of the init process but not repeat the stub priming
1788  * check.
1789  *
1790  * @param qstate: query state.
1791  * @param iq: iterator query state.
1792  * @param id: module id.
1793  * @return true, advancing the event to the QUERYTARGETS_STATE.
1794  */
1795 static int
processInitRequest3(struct module_qstate * qstate,struct iter_qstate * iq,int id)1796 processInitRequest3(struct module_qstate* qstate, struct iter_qstate* iq,
1797 	int id)
1798 {
1799 	log_query_info(VERB_QUERY, "resolving (init part 3): ",
1800 		&qstate->qinfo);
1801 	/* if the cache reply dp equals a validation anchor or msg has DS,
1802 	 * then DNSSEC RRSIGs are expected in the reply */
1803 	iq->dnssec_expected = iter_indicates_dnssec(qstate->env, iq->dp,
1804 		iq->deleg_msg, iq->qchase.qclass);
1805 
1806 	/* If the RD flag wasn't set, then we just finish with the
1807 	 * cached referral as the response. */
1808 	if(!(qstate->query_flags & BIT_RD) && iq->deleg_msg) {
1809 		iq->response = iq->deleg_msg;
1810 		if(verbosity >= VERB_ALGO && iq->response)
1811 			log_dns_msg("no RD requested, using delegation msg",
1812 				&iq->response->qinfo, iq->response->rep);
1813 		if(qstate->reply_origin)
1814 			sock_list_insert(&qstate->reply_origin, NULL, 0, qstate->region);
1815 		return final_state(iq);
1816 	}
1817 	/* After this point, unset the RD flag -- this query is going to
1818 	 * be sent to an auth. server. */
1819 	iq->chase_flags &= ~BIT_RD;
1820 
1821 	/* if dnssec expected, fetch key for the trust-anchor or cached-DS */
1822 	if(iq->dnssec_expected && qstate->env->cfg->prefetch_key &&
1823 		!(qstate->query_flags&BIT_CD)) {
1824 		generate_dnskey_prefetch(qstate, iq, id);
1825 		fptr_ok(fptr_whitelist_modenv_detach_subs(
1826 			qstate->env->detach_subs));
1827 		(*qstate->env->detach_subs)(qstate);
1828 	}
1829 
1830 	/* Jump to the next state. */
1831 	return next_state(iq, QUERYTARGETS_STATE);
1832 }
1833 
1834 /**
1835  * Given a basic query, generate a parent-side "target" query.
1836  * These are subordinate queries for missing delegation point target addresses,
1837  * for which only the parent of the delegation provides correct IP addresses.
1838  *
1839  * @param qstate: query state.
1840  * @param iq: iterator query state.
1841  * @param id: module id.
1842  * @param name: target qname.
1843  * @param namelen: target qname length.
1844  * @param qtype: target qtype (either A or AAAA).
1845  * @param qclass: target qclass.
1846  * @return true on success, false on failure.
1847  */
1848 static int
generate_parentside_target_query(struct module_qstate * qstate,struct iter_qstate * iq,int id,uint8_t * name,size_t namelen,uint16_t qtype,uint16_t qclass)1849 generate_parentside_target_query(struct module_qstate* qstate,
1850 	struct iter_qstate* iq, int id, uint8_t* name, size_t namelen,
1851 	uint16_t qtype, uint16_t qclass)
1852 {
1853 	struct module_qstate* subq;
1854 	if(!generate_sub_request(name, namelen, qtype, qclass, qstate,
1855 		id, iq, INIT_REQUEST_STATE, FINISHED_STATE, &subq, 0, 0))
1856 		return 0;
1857 	if(subq) {
1858 		struct iter_qstate* subiq =
1859 			(struct iter_qstate*)subq->minfo[id];
1860 		/* blacklist the cache - we want to fetch parent stuff */
1861 		sock_list_insert(&subq->blacklist, NULL, 0, subq->region);
1862 		subiq->query_for_pside_glue = 1;
1863 		if(dname_subdomain_c(name, iq->dp->name)) {
1864 			subiq->dp = delegpt_copy(iq->dp, subq->region);
1865 			subiq->dnssec_expected = iter_indicates_dnssec(
1866 				qstate->env, subiq->dp, NULL,
1867 				subq->qinfo.qclass);
1868 			subiq->refetch_glue = 1;
1869 		} else {
1870 			subiq->dp = dns_cache_find_delegation(qstate->env,
1871 				name, namelen, qtype, qclass, subq->region,
1872 				&subiq->deleg_msg,
1873 				*qstate->env->now+subq->prefetch_leeway,
1874 				1, NULL, 0);
1875 			/* if no dp, then it's from root, refetch unneeded */
1876 			if(subiq->dp) {
1877 				subiq->dnssec_expected = iter_indicates_dnssec(
1878 					qstate->env, subiq->dp, NULL,
1879 					subq->qinfo.qclass);
1880 				subiq->refetch_glue = 1;
1881 			}
1882 		}
1883 	}
1884 	log_nametypeclass(VERB_QUERY, "new pside target", name, qtype, qclass);
1885 	return 1;
1886 }
1887 
1888 /**
1889  * Given a basic query, generate a "target" query. These are subordinate
1890  * queries for missing delegation point target addresses.
1891  *
1892  * @param qstate: query state.
1893  * @param iq: iterator query state.
1894  * @param id: module id.
1895  * @param name: target qname.
1896  * @param namelen: target qname length.
1897  * @param qtype: target qtype (either A or AAAA).
1898  * @param qclass: target qclass.
1899  * @return true on success, false on failure.
1900  */
1901 static int
generate_target_query(struct module_qstate * qstate,struct iter_qstate * iq,int id,uint8_t * name,size_t namelen,uint16_t qtype,uint16_t qclass)1902 generate_target_query(struct module_qstate* qstate, struct iter_qstate* iq,
1903         int id, uint8_t* name, size_t namelen, uint16_t qtype, uint16_t qclass)
1904 {
1905 	struct module_qstate* subq;
1906 	if(!generate_sub_request(name, namelen, qtype, qclass, qstate,
1907 		id, iq, INIT_REQUEST_STATE, FINISHED_STATE, &subq, 0, 0))
1908 		return 0;
1909 	log_nametypeclass(VERB_QUERY, "new target", name, qtype, qclass);
1910 	return 1;
1911 }
1912 
1913 /**
1914  * Given an event at a certain state, generate zero or more target queries
1915  * for it's current delegation point.
1916  *
1917  * @param qstate: query state.
1918  * @param iq: iterator query state.
1919  * @param ie: iterator shared global environment.
1920  * @param id: module id.
1921  * @param maxtargets: The maximum number of targets to query for.
1922  *	if it is negative, there is no maximum number of targets.
1923  * @param num: returns the number of queries generated and processed,
1924  *	which may be zero if there were no missing targets.
1925  * @return false on error.
1926  */
1927 static int
query_for_targets(struct module_qstate * qstate,struct iter_qstate * iq,struct iter_env * ie,int id,int maxtargets,int * num)1928 query_for_targets(struct module_qstate* qstate, struct iter_qstate* iq,
1929         struct iter_env* ie, int id, int maxtargets, int* num)
1930 {
1931 	int query_count = 0;
1932 	struct delegpt_ns* ns;
1933 	int missing;
1934 	int toget = 0;
1935 
1936 	iter_mark_cycle_targets(qstate, iq->dp);
1937 	missing = (int)delegpt_count_missing_targets(iq->dp, NULL);
1938 	log_assert(maxtargets != 0); /* that would not be useful */
1939 
1940 	/* Generate target requests. Basically, any missing targets
1941 	 * are queried for here, regardless if it is necessary to do
1942 	 * so to continue processing. */
1943 	if(maxtargets < 0 || maxtargets > missing)
1944 		toget = missing;
1945 	else	toget = maxtargets;
1946 	if(toget == 0) {
1947 		*num = 0;
1948 		return 1;
1949 	}
1950 
1951 	/* now that we are sure that a target query is going to be made,
1952 	 * check the limits. */
1953 	if(iq->depth == ie->max_dependency_depth)
1954 		return 0;
1955 	if(iq->depth > 0 && iq->target_count &&
1956 		iq->target_count[TARGET_COUNT_QUERIES] > MAX_TARGET_COUNT) {
1957 		char s[LDNS_MAX_DOMAINLEN+1];
1958 		dname_str(qstate->qinfo.qname, s);
1959 		verbose(VERB_QUERY, "request %s has exceeded the maximum "
1960 			"number of glue fetches %d", s,
1961 			iq->target_count[TARGET_COUNT_QUERIES]);
1962 		return 0;
1963 	}
1964 	if(iq->dp_target_count > MAX_DP_TARGET_COUNT) {
1965 		char s[LDNS_MAX_DOMAINLEN+1];
1966 		dname_str(qstate->qinfo.qname, s);
1967 		verbose(VERB_QUERY, "request %s has exceeded the maximum "
1968 			"number of glue fetches %d to a single delegation point",
1969 			s, iq->dp_target_count);
1970 		return 0;
1971 	}
1972 
1973 	/* select 'toget' items from the total of 'missing' items */
1974 	log_assert(toget <= missing);
1975 
1976 	/* loop over missing targets */
1977 	for(ns = iq->dp->nslist; ns; ns = ns->next) {
1978 		if(ns->resolved)
1979 			continue;
1980 
1981 		/* randomly select this item with probability toget/missing */
1982 		if(!iter_ns_probability(qstate->env->rnd, toget, missing)) {
1983 			/* do not select this one, next; select toget number
1984 			 * of items from a list one less in size */
1985 			missing --;
1986 			continue;
1987 		}
1988 
1989 		if(ie->supports_ipv6 &&
1990 			((ns->lame && !ns->done_pside6) ||
1991 			(!ns->lame && !ns->got6))) {
1992 			/* Send the AAAA request. */
1993 			if(!generate_target_query(qstate, iq, id,
1994 				ns->name, ns->namelen,
1995 				LDNS_RR_TYPE_AAAA, iq->qchase.qclass)) {
1996 				*num = query_count;
1997 				if(query_count > 0)
1998 					qstate->ext_state[id] = module_wait_subquery;
1999 				return 0;
2000 			}
2001 			query_count++;
2002 			/* If the mesh query list is full, exit the loop here.
2003 			 * This makes the routine spawn one query at a time,
2004 			 * and this means there is no query state load
2005 			 * increase, because the spawned state uses cpu and a
2006 			 * socket while this state waits for that spawned
2007 			 * state. Next time we can look up further targets */
2008 			if(mesh_jostle_exceeded(qstate->env->mesh))
2009 				break;
2010 		}
2011 		/* Send the A request. */
2012 		if((ie->supports_ipv4 || ie->use_nat64) &&
2013 			((ns->lame && !ns->done_pside4) ||
2014 			(!ns->lame && !ns->got4))) {
2015 			if(!generate_target_query(qstate, iq, id,
2016 				ns->name, ns->namelen,
2017 				LDNS_RR_TYPE_A, iq->qchase.qclass)) {
2018 				*num = query_count;
2019 				if(query_count > 0)
2020 					qstate->ext_state[id] = module_wait_subquery;
2021 				return 0;
2022 			}
2023 			query_count++;
2024 			/* If the mesh query list is full, exit the loop. */
2025 			if(mesh_jostle_exceeded(qstate->env->mesh))
2026 				break;
2027 		}
2028 
2029 		/* mark this target as in progress. */
2030 		ns->resolved = 1;
2031 		missing--;
2032 		toget--;
2033 		if(toget == 0)
2034 			break;
2035 	}
2036 	*num = query_count;
2037 	if(query_count > 0)
2038 		qstate->ext_state[id] = module_wait_subquery;
2039 
2040 	return 1;
2041 }
2042 
2043 /**
2044  * Called by processQueryTargets when it would like extra targets to query
2045  * but it seems to be out of options.  At last resort some less appealing
2046  * options are explored.  If there are no more options, the result is SERVFAIL
2047  *
2048  * @param qstate: query state.
2049  * @param iq: iterator query state.
2050  * @param ie: iterator shared global environment.
2051  * @param id: module id.
2052  * @return true if the event requires more request processing immediately,
2053  *         false if not.
2054  */
2055 static int
processLastResort(struct module_qstate * qstate,struct iter_qstate * iq,struct iter_env * ie,int id)2056 processLastResort(struct module_qstate* qstate, struct iter_qstate* iq,
2057 	struct iter_env* ie, int id)
2058 {
2059 	struct delegpt_ns* ns;
2060 	int query_count = 0;
2061 	verbose(VERB_ALGO, "No more query targets, attempting last resort");
2062 	log_assert(iq->dp);
2063 
2064 	if(!can_have_last_resort(qstate->env, iq->dp->name, iq->dp->namelen,
2065 		iq->qchase.qclass, NULL)) {
2066 		/* fail -- no more targets, no more hope of targets, no hope
2067 		 * of a response. */
2068 		errinf(qstate, "all the configured stub or forward servers failed,");
2069 		errinf_dname(qstate, "at zone", iq->dp->name);
2070 		errinf_reply(qstate, iq);
2071 		verbose(VERB_QUERY, "configured stub or forward servers failed -- returning SERVFAIL");
2072 		return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
2073 	}
2074 	if(!iq->dp->has_parent_side_NS && dname_is_root(iq->dp->name)) {
2075 		struct delegpt* p = hints_lookup_root(qstate->env->hints,
2076 			iq->qchase.qclass);
2077 		if(p) {
2078 			struct delegpt_addr* a;
2079 			iq->chase_flags &= ~BIT_RD; /* go to authorities */
2080 			for(ns = p->nslist; ns; ns=ns->next) {
2081 				(void)delegpt_add_ns(iq->dp, qstate->region,
2082 					ns->name, ns->lame, ns->tls_auth_name,
2083 					ns->port);
2084 			}
2085 			for(a = p->target_list; a; a=a->next_target) {
2086 				(void)delegpt_add_addr(iq->dp, qstate->region,
2087 					&a->addr, a->addrlen, a->bogus,
2088 					a->lame, a->tls_auth_name, -1, NULL);
2089 			}
2090 		}
2091 		iq->dp->has_parent_side_NS = 1;
2092 	} else if(!iq->dp->has_parent_side_NS) {
2093 		if(!iter_lookup_parent_NS_from_cache(qstate->env, iq->dp,
2094 			qstate->region, &qstate->qinfo)
2095 			|| !iq->dp->has_parent_side_NS) {
2096 			/* if: malloc failure in lookup go up to try */
2097 			/* if: no parent NS in cache - go up one level */
2098 			verbose(VERB_ALGO, "try to grab parent NS");
2099 			iq->store_parent_NS = iq->dp;
2100 			iq->chase_flags &= ~BIT_RD; /* go to authorities */
2101 			iq->deleg_msg = NULL;
2102 			iq->refetch_glue = 1;
2103 			iq->query_restart_count++;
2104 			iq->sent_count = 0;
2105 			iq->dp_target_count = 0;
2106 			if(qstate->env->cfg->qname_minimisation)
2107 				iq->minimisation_state = INIT_MINIMISE_STATE;
2108 			return next_state(iq, INIT_REQUEST_STATE);
2109 		}
2110 	}
2111 	/* see if that makes new names available */
2112 	if(!cache_fill_missing(qstate->env, iq->qchase.qclass,
2113 		qstate->region, iq->dp))
2114 		log_err("out of memory in cache_fill_missing");
2115 	if(iq->dp->usable_list) {
2116 		verbose(VERB_ALGO, "try parent-side-name, w. glue from cache");
2117 		return next_state(iq, QUERYTARGETS_STATE);
2118 	}
2119 	/* try to fill out parent glue from cache */
2120 	if(iter_lookup_parent_glue_from_cache(qstate->env, iq->dp,
2121 		qstate->region, &qstate->qinfo)) {
2122 		/* got parent stuff from cache, see if we can continue */
2123 		verbose(VERB_ALGO, "try parent-side glue from cache");
2124 		return next_state(iq, QUERYTARGETS_STATE);
2125 	}
2126 	/* query for an extra name added by the parent-NS record */
2127 	if(delegpt_count_missing_targets(iq->dp, NULL) > 0) {
2128 		int qs = 0;
2129 		verbose(VERB_ALGO, "try parent-side target name");
2130 		if(!query_for_targets(qstate, iq, ie, id, 1, &qs)) {
2131 			errinf(qstate, "could not fetch nameserver");
2132 			errinf_dname(qstate, "at zone", iq->dp->name);
2133 			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2134 		}
2135 		iq->num_target_queries += qs;
2136 		target_count_increase(iq, qs);
2137 		if(qs != 0) {
2138 			qstate->ext_state[id] = module_wait_subquery;
2139 			return 0; /* and wait for them */
2140 		}
2141 	}
2142 	if(iq->depth == ie->max_dependency_depth) {
2143 		verbose(VERB_QUERY, "maxdepth and need more nameservers, fail");
2144 		errinf(qstate, "cannot fetch more nameservers because at max dependency depth");
2145 		return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
2146 	}
2147 	if(iq->depth > 0 && iq->target_count &&
2148 		iq->target_count[TARGET_COUNT_QUERIES] > MAX_TARGET_COUNT) {
2149 		char s[LDNS_MAX_DOMAINLEN+1];
2150 		dname_str(qstate->qinfo.qname, s);
2151 		verbose(VERB_QUERY, "request %s has exceeded the maximum "
2152 			"number of glue fetches %d", s,
2153 			iq->target_count[TARGET_COUNT_QUERIES]);
2154 		errinf(qstate, "exceeded the maximum number of glue fetches");
2155 		return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
2156 	}
2157 	/* mark cycle targets for parent-side lookups */
2158 	iter_mark_pside_cycle_targets(qstate, iq->dp);
2159 	/* see if we can issue queries to get nameserver addresses */
2160 	/* this lookup is not randomized, but sequential. */
2161 	for(ns = iq->dp->nslist; ns; ns = ns->next) {
2162 		/* if this nameserver is at a delegation point, but that
2163 		 * delegation point is a stub and we cannot go higher, skip*/
2164 		if( ((ie->supports_ipv6 && !ns->done_pside6) ||
2165 		    ((ie->supports_ipv4 || ie->use_nat64) && !ns->done_pside4)) &&
2166 		    !can_have_last_resort(qstate->env, ns->name, ns->namelen,
2167 			iq->qchase.qclass, NULL)) {
2168 			log_nametypeclass(VERB_ALGO, "cannot pside lookup ns "
2169 				"because it is also a stub/forward,",
2170 				ns->name, LDNS_RR_TYPE_NS, iq->qchase.qclass);
2171 			if(ie->supports_ipv6) ns->done_pside6 = 1;
2172 			if(ie->supports_ipv4 || ie->use_nat64) ns->done_pside4 = 1;
2173 			continue;
2174 		}
2175 		/* query for parent-side A and AAAA for nameservers */
2176 		if(ie->supports_ipv6 && !ns->done_pside6) {
2177 			/* Send the AAAA request. */
2178 			if(!generate_parentside_target_query(qstate, iq, id,
2179 				ns->name, ns->namelen,
2180 				LDNS_RR_TYPE_AAAA, iq->qchase.qclass)) {
2181 				errinf_dname(qstate, "could not generate nameserver AAAA lookup for", ns->name);
2182 				return error_response(qstate, id,
2183 					LDNS_RCODE_SERVFAIL);
2184 			}
2185 			ns->done_pside6 = 1;
2186 			query_count++;
2187 			if(mesh_jostle_exceeded(qstate->env->mesh)) {
2188 				/* Wait for the lookup; do not spawn multiple
2189 				 * lookups at a time. */
2190 				verbose(VERB_ALGO, "try parent-side glue lookup");
2191 				iq->num_target_queries += query_count;
2192 				target_count_increase(iq, query_count);
2193 				qstate->ext_state[id] = module_wait_subquery;
2194 				return 0;
2195 			}
2196 		}
2197 		if((ie->supports_ipv4 || ie->use_nat64) && !ns->done_pside4) {
2198 			/* Send the A request. */
2199 			if(!generate_parentside_target_query(qstate, iq, id,
2200 				ns->name, ns->namelen,
2201 				LDNS_RR_TYPE_A, iq->qchase.qclass)) {
2202 				errinf_dname(qstate, "could not generate nameserver A lookup for", ns->name);
2203 				return error_response(qstate, id,
2204 					LDNS_RCODE_SERVFAIL);
2205 			}
2206 			ns->done_pside4 = 1;
2207 			query_count++;
2208 		}
2209 		if(query_count != 0) { /* suspend to await results */
2210 			verbose(VERB_ALGO, "try parent-side glue lookup");
2211 			iq->num_target_queries += query_count;
2212 			target_count_increase(iq, query_count);
2213 			qstate->ext_state[id] = module_wait_subquery;
2214 			return 0;
2215 		}
2216 	}
2217 
2218 	/* if this was a parent-side glue query itself, then store that
2219 	 * failure in cache. */
2220 	if(!qstate->no_cache_store && iq->query_for_pside_glue
2221 		&& !iq->pside_glue)
2222 			iter_store_parentside_neg(qstate->env, &qstate->qinfo,
2223 				iq->deleg_msg?iq->deleg_msg->rep:
2224 				(iq->response?iq->response->rep:NULL));
2225 
2226 	errinf(qstate, "all servers for this domain failed,");
2227 	errinf_dname(qstate, "at zone", iq->dp->name);
2228 	errinf_reply(qstate, iq);
2229 	verbose(VERB_QUERY, "out of query targets -- returning SERVFAIL");
2230 	/* fail -- no more targets, no more hope of targets, no hope
2231 	 * of a response. */
2232 	return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
2233 }
2234 
2235 /**
2236  * Try to find the NS record set that will resolve a qtype DS query. Due
2237  * to grandparent/grandchild reasons we did not get a proper lookup right
2238  * away.  We need to create type NS queries until we get the right parent
2239  * for this lookup.  We remove labels from the query to find the right point.
2240  * If we end up at the old dp name, then there is no solution.
2241  *
2242  * @param qstate: query state.
2243  * @param iq: iterator query state.
2244  * @param id: module id.
2245  * @return true if the event requires more immediate processing, false if
2246  *         not. This is generally only true when forwarding the request to
2247  *         the final state (i.e., on answer).
2248  */
2249 static int
processDSNSFind(struct module_qstate * qstate,struct iter_qstate * iq,int id)2250 processDSNSFind(struct module_qstate* qstate, struct iter_qstate* iq, int id)
2251 {
2252 	struct module_qstate* subq = NULL;
2253 	verbose(VERB_ALGO, "processDSNSFind");
2254 
2255 	if(!iq->dsns_point) {
2256 		/* initialize */
2257 		iq->dsns_point = iq->qchase.qname;
2258 		iq->dsns_point_len = iq->qchase.qname_len;
2259 	}
2260 	/* robustcheck for internal error: we are not underneath the dp */
2261 	if(!dname_subdomain_c(iq->dsns_point, iq->dp->name)) {
2262 		errinf_dname(qstate, "for DS query parent-child nameserver search the query is not under the zone", iq->dp->name);
2263 		return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
2264 	}
2265 
2266 	/* go up one (more) step, until we hit the dp, if so, end */
2267 	dname_remove_label(&iq->dsns_point, &iq->dsns_point_len);
2268 	if(query_dname_compare(iq->dsns_point, iq->dp->name) == 0) {
2269 		/* there was no inbetween nameserver, use the old delegation
2270 		 * point again.  And this time, because dsns_point is nonNULL
2271 		 * we are going to accept the (bad) result */
2272 		iq->state = QUERYTARGETS_STATE;
2273 		return 1;
2274 	}
2275 	iq->state = DSNS_FIND_STATE;
2276 
2277 	/* spawn NS lookup (validation not needed, this is for DS lookup) */
2278 	log_nametypeclass(VERB_ALGO, "fetch nameservers",
2279 		iq->dsns_point, LDNS_RR_TYPE_NS, iq->qchase.qclass);
2280 	if(!generate_sub_request(iq->dsns_point, iq->dsns_point_len,
2281 		LDNS_RR_TYPE_NS, iq->qchase.qclass, qstate, id, iq,
2282 		INIT_REQUEST_STATE, FINISHED_STATE, &subq, 0, 0)) {
2283 		errinf_dname(qstate, "for DS query parent-child nameserver search, could not generate NS lookup for", iq->dsns_point);
2284 		return error_response_cache(qstate, id, LDNS_RCODE_SERVFAIL);
2285 	}
2286 
2287 	return 0;
2288 }
2289 
2290 /**
2291  * Check if we wait responses for sent queries and update the iterator's
2292  * external state.
2293  */
2294 static void
check_waiting_queries(struct iter_qstate * iq,struct module_qstate * qstate,int id)2295 check_waiting_queries(struct iter_qstate* iq, struct module_qstate* qstate,
2296 	int id)
2297 {
2298 	if(iq->num_target_queries>0 && iq->num_current_queries>0) {
2299 		verbose(VERB_ALGO, "waiting for %d targets to "
2300 			"resolve or %d outstanding queries to "
2301 			"respond", iq->num_target_queries,
2302 			iq->num_current_queries);
2303 		qstate->ext_state[id] = module_wait_reply;
2304 	} else if(iq->num_target_queries>0) {
2305 		verbose(VERB_ALGO, "waiting for %d targets to "
2306 			"resolve", iq->num_target_queries);
2307 		qstate->ext_state[id] = module_wait_subquery;
2308 	} else {
2309 		verbose(VERB_ALGO, "waiting for %d "
2310 			"outstanding queries to respond",
2311 			iq->num_current_queries);
2312 		qstate->ext_state[id] = module_wait_reply;
2313 	}
2314 }
2315 
2316 /**
2317  * This is the request event state where the request will be sent to one of
2318  * its current query targets. This state also handles issuing target lookup
2319  * queries for missing target IP addresses. Queries typically iterate on
2320  * this state, both when they are just trying different targets for a given
2321  * delegation point, and when they change delegation points. This state
2322  * roughly corresponds to RFC 1034 algorithm steps 3 and 4.
2323  *
2324  * @param qstate: query state.
2325  * @param iq: iterator query state.
2326  * @param ie: iterator shared global environment.
2327  * @param id: module id.
2328  * @return true if the event requires more request processing immediately,
2329  *         false if not. This state only returns true when it is generating
2330  *         a SERVFAIL response because the query has hit a dead end.
2331  */
2332 static int
processQueryTargets(struct module_qstate * qstate,struct iter_qstate * iq,struct iter_env * ie,int id)2333 processQueryTargets(struct module_qstate* qstate, struct iter_qstate* iq,
2334 	struct iter_env* ie, int id)
2335 {
2336 	int tf_policy;
2337 	struct delegpt_addr* target;
2338 	struct outbound_entry* outq;
2339 	struct sockaddr_storage real_addr;
2340 	socklen_t real_addrlen;
2341 	int auth_fallback = 0;
2342 	uint8_t* qout_orig = NULL;
2343 	size_t qout_orig_len = 0;
2344 	int sq_check_ratelimit = 1;
2345 	int sq_was_ratelimited = 0;
2346 	int can_do_promisc = 0;
2347 
2348 	/* NOTE: a request will encounter this state for each target it
2349 	 * needs to send a query to. That is, at least one per referral,
2350 	 * more if some targets timeout or return throwaway answers. */
2351 
2352 	log_query_info(VERB_QUERY, "processQueryTargets:", &qstate->qinfo);
2353 	verbose(VERB_ALGO, "processQueryTargets: targetqueries %d, "
2354 		"currentqueries %d sentcount %d", iq->num_target_queries,
2355 		iq->num_current_queries, iq->sent_count);
2356 
2357 	/* Make sure that we haven't run away */
2358 	if(iq->referral_count > MAX_REFERRAL_COUNT) {
2359 		verbose(VERB_QUERY, "request has exceeded the maximum "
2360 			"number of referrrals with %d", iq->referral_count);
2361 		errinf(qstate, "exceeded the maximum of referrals");
2362 		return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2363 	}
2364 	if(iq->sent_count > ie->max_sent_count) {
2365 		verbose(VERB_QUERY, "request has exceeded the maximum "
2366 			"number of sends with %d", iq->sent_count);
2367 		errinf(qstate, "exceeded the maximum number of sends");
2368 		return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2369 	}
2370 
2371 	/* Check if we reached MAX_TARGET_NX limit without a fallback activation. */
2372 	if(iq->target_count && !*iq->nxns_dp &&
2373 		iq->target_count[TARGET_COUNT_NX] > MAX_TARGET_NX) {
2374 		struct delegpt_ns* ns;
2375 		/* If we can wait for resolution, do so. */
2376 		if(iq->num_target_queries>0 || iq->num_current_queries>0) {
2377 			check_waiting_queries(iq, qstate, id);
2378 			return 0;
2379 		}
2380 		verbose(VERB_ALGO, "request has exceeded the maximum "
2381 			"number of nxdomain nameserver lookups (%d) with %d",
2382 			MAX_TARGET_NX, iq->target_count[TARGET_COUNT_NX]);
2383 		/* Check for dp because we require one below */
2384 		if(!iq->dp) {
2385 			verbose(VERB_QUERY, "Failed to get a delegation, "
2386 				"giving up");
2387 			errinf(qstate, "failed to get a delegation (eg. prime "
2388 				"failure)");
2389 			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2390 		}
2391 		/* We reached the limit but we already have parent side
2392 		 * information; stop resolution */
2393 		if(iq->dp->has_parent_side_NS) {
2394 			verbose(VERB_ALGO, "parent-side information is "
2395 				"already present for the delegation point, no "
2396 				"fallback possible");
2397 			errinf(qstate, "exceeded the maximum nameserver nxdomains");
2398 			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2399 		}
2400 		verbose(VERB_ALGO, "initiating parent-side fallback for "
2401 			"nxdomain nameserver lookups");
2402 		/* Mark all the current NSes as resolved to allow for parent
2403 		 * fallback */
2404 		for(ns=iq->dp->nslist; ns; ns=ns->next) {
2405 			ns->resolved = 1;
2406 		}
2407 		/* Note the delegation point that triggered the NXNS fallback;
2408 		 * no reason for shared queries to keep trying there.
2409 		 * This also marks the fallback activation. */
2410 		*iq->nxns_dp = malloc(iq->dp->namelen);
2411 		if(!*iq->nxns_dp) {
2412 			verbose(VERB_ALGO, "out of memory while initiating "
2413 				"fallback");
2414 			errinf(qstate, "exceeded the maximum nameserver "
2415 				"nxdomains (malloc)");
2416 			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2417 		}
2418 		memcpy(*iq->nxns_dp, iq->dp->name, iq->dp->namelen);
2419 	} else if(iq->target_count && *iq->nxns_dp) {
2420 		/* Handle the NXNS fallback case. */
2421 		/* If we can wait for resolution, do so. */
2422 		if(iq->num_target_queries>0 || iq->num_current_queries>0) {
2423 			check_waiting_queries(iq, qstate, id);
2424 			return 0;
2425 		}
2426 		/* Check for dp because we require one below */
2427 		if(!iq->dp) {
2428 			verbose(VERB_QUERY, "Failed to get a delegation, "
2429 				"giving up");
2430 			errinf(qstate, "failed to get a delegation (eg. prime "
2431 				"failure)");
2432 			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2433 		}
2434 
2435 		if(iq->target_count[TARGET_COUNT_NX] > MAX_TARGET_NX_FALLBACK) {
2436 			verbose(VERB_ALGO, "request has exceeded the maximum "
2437 				"number of fallback nxdomain nameserver "
2438 				"lookups (%d) with %d", MAX_TARGET_NX_FALLBACK,
2439 				iq->target_count[TARGET_COUNT_NX]);
2440 			errinf(qstate, "exceeded the maximum nameserver nxdomains");
2441 			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2442 		}
2443 
2444 		if(!iq->dp->has_parent_side_NS) {
2445 			struct delegpt_ns* ns;
2446 			if(!dname_canonical_compare(*iq->nxns_dp, iq->dp->name)) {
2447 				verbose(VERB_ALGO, "this delegation point "
2448 					"initiated the fallback, marking the "
2449 					"nslist as resolved");
2450 				for(ns=iq->dp->nslist; ns; ns=ns->next) {
2451 					ns->resolved = 1;
2452 				}
2453 			}
2454 		}
2455 	}
2456 
2457 	/* Make sure we have a delegation point, otherwise priming failed
2458 	 * or another failure occurred */
2459 	if(!iq->dp) {
2460 		verbose(VERB_QUERY, "Failed to get a delegation, giving up");
2461 		errinf(qstate, "failed to get a delegation (eg. prime failure)");
2462 		return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2463 	}
2464 	if(!ie->supports_ipv6)
2465 		delegpt_no_ipv6(iq->dp);
2466 	if(!ie->supports_ipv4 && !ie->use_nat64)
2467 		delegpt_no_ipv4(iq->dp);
2468 	delegpt_log(VERB_ALGO, iq->dp);
2469 
2470 	if(iq->num_current_queries>0) {
2471 		/* already busy answering a query, this restart is because
2472 		 * more delegpt addrs became available, wait for existing
2473 		 * query. */
2474 		verbose(VERB_ALGO, "woke up, but wait for outstanding query");
2475 		qstate->ext_state[id] = module_wait_reply;
2476 		return 0;
2477 	}
2478 
2479 	if(iq->minimisation_state == INIT_MINIMISE_STATE
2480 		&& !(iq->chase_flags & BIT_RD)) {
2481 		/* (Re)set qinfo_out to (new) delegation point, except when
2482 		 * qinfo_out is already a subdomain of dp. This happens when
2483 		 * increasing by more than one label at once (QNAMEs with more
2484 		 * than MAX_MINIMISE_COUNT labels). */
2485 		if(!(iq->qinfo_out.qname_len
2486 			&& dname_subdomain_c(iq->qchase.qname,
2487 				iq->qinfo_out.qname)
2488 			&& dname_subdomain_c(iq->qinfo_out.qname,
2489 				iq->dp->name))) {
2490 			iq->qinfo_out.qname = iq->dp->name;
2491 			iq->qinfo_out.qname_len = iq->dp->namelen;
2492 			iq->qinfo_out.qtype = LDNS_RR_TYPE_A;
2493 			iq->qinfo_out.qclass = iq->qchase.qclass;
2494 			iq->qinfo_out.local_alias = NULL;
2495 			iq->minimise_count = 0;
2496 		}
2497 
2498 		iq->minimisation_state = MINIMISE_STATE;
2499 	}
2500 	if(iq->minimisation_state == MINIMISE_STATE) {
2501 		int qchaselabs = dname_count_labels(iq->qchase.qname);
2502 		int labdiff = qchaselabs -
2503 			dname_count_labels(iq->qinfo_out.qname);
2504 
2505 		qout_orig = iq->qinfo_out.qname;
2506 		qout_orig_len = iq->qinfo_out.qname_len;
2507 		iq->qinfo_out.qname = iq->qchase.qname;
2508 		iq->qinfo_out.qname_len = iq->qchase.qname_len;
2509 		iq->minimise_count++;
2510 		iq->timeout_count = 0;
2511 
2512 		iter_dec_attempts(iq->dp, 1, ie->outbound_msg_retry);
2513 
2514 		/* Limit number of iterations for QNAMEs with more
2515 		 * than MAX_MINIMISE_COUNT labels. Send first MINIMISE_ONE_LAB
2516 		 * labels of QNAME always individually.
2517 		 */
2518 		if(qchaselabs > MAX_MINIMISE_COUNT && labdiff > 1 &&
2519 			iq->minimise_count > MINIMISE_ONE_LAB) {
2520 			if(iq->minimise_count < MAX_MINIMISE_COUNT) {
2521 				int multilabs = qchaselabs - 1 -
2522 					MINIMISE_ONE_LAB;
2523 				int extralabs = multilabs /
2524 					MINIMISE_MULTIPLE_LABS;
2525 
2526 				if (MAX_MINIMISE_COUNT - iq->minimise_count >=
2527 					multilabs % MINIMISE_MULTIPLE_LABS)
2528 					/* Default behaviour is to add 1 label
2529 					 * every iteration. Therefore, decrement
2530 					 * the extralabs by 1 */
2531 					extralabs--;
2532 				if (extralabs < labdiff)
2533 					labdiff -= extralabs;
2534 				else
2535 					labdiff = 1;
2536 			}
2537 			/* Last minimised iteration, send all labels with
2538 			 * QTYPE=NS */
2539 			else
2540 				labdiff = 1;
2541 		}
2542 
2543 		if(labdiff > 1) {
2544 			verbose(VERB_QUERY, "removing %d labels", labdiff-1);
2545 			dname_remove_labels(&iq->qinfo_out.qname,
2546 				&iq->qinfo_out.qname_len,
2547 				labdiff-1);
2548 		}
2549 		if(labdiff < 1 || (labdiff < 2
2550 			&& (iq->qchase.qtype == LDNS_RR_TYPE_DS
2551 			|| iq->qchase.qtype == LDNS_RR_TYPE_A)))
2552 			/* Stop minimising this query, resolve "as usual" */
2553 			iq->minimisation_state = DONOT_MINIMISE_STATE;
2554 		else if(!qstate->no_cache_lookup) {
2555 			struct dns_msg* msg = dns_cache_lookup(qstate->env,
2556 				iq->qinfo_out.qname, iq->qinfo_out.qname_len,
2557 				iq->qinfo_out.qtype, iq->qinfo_out.qclass,
2558 				qstate->query_flags, qstate->region,
2559 				qstate->env->scratch, 0, iq->dp->name,
2560 				iq->dp->namelen);
2561 			if(msg && FLAGS_GET_RCODE(msg->rep->flags) ==
2562 				LDNS_RCODE_NOERROR)
2563 				/* no need to send query if it is already
2564 				 * cached as NOERROR */
2565 				return 1;
2566 			if(msg && FLAGS_GET_RCODE(msg->rep->flags) ==
2567 				LDNS_RCODE_NXDOMAIN &&
2568 				qstate->env->need_to_validate &&
2569 				qstate->env->cfg->harden_below_nxdomain) {
2570 				if(msg->rep->security == sec_status_secure) {
2571 					iq->response = msg;
2572 					return final_state(iq);
2573 				}
2574 				if(msg->rep->security == sec_status_unchecked) {
2575 					struct module_qstate* subq = NULL;
2576 					if(!generate_sub_request(
2577 						iq->qinfo_out.qname,
2578 						iq->qinfo_out.qname_len,
2579 						iq->qinfo_out.qtype,
2580 						iq->qinfo_out.qclass,
2581 						qstate, id, iq,
2582 						INIT_REQUEST_STATE,
2583 						FINISHED_STATE, &subq, 1, 1))
2584 						verbose(VERB_ALGO,
2585 						"could not validate NXDOMAIN "
2586 						"response");
2587 				}
2588 			}
2589 			if(msg && FLAGS_GET_RCODE(msg->rep->flags) ==
2590 				LDNS_RCODE_NXDOMAIN) {
2591 				/* return and add a label in the next
2592 				 * minimisation iteration.
2593 				 */
2594 				return 1;
2595 			}
2596 		}
2597 	}
2598 	if(iq->minimisation_state == SKIP_MINIMISE_STATE) {
2599 		if(iq->timeout_count < MAX_MINIMISE_TIMEOUT_COUNT)
2600 			/* Do not increment qname, continue incrementing next
2601 			 * iteration */
2602 			iq->minimisation_state = MINIMISE_STATE;
2603 		else if(!qstate->env->cfg->qname_minimisation_strict)
2604 			/* Too many time-outs detected for this QNAME and QTYPE.
2605 			 * We give up, disable QNAME minimisation. */
2606 			iq->minimisation_state = DONOT_MINIMISE_STATE;
2607 	}
2608 	if(iq->minimisation_state == DONOT_MINIMISE_STATE)
2609 		iq->qinfo_out = iq->qchase;
2610 
2611 	/* now find an answer to this query */
2612 	/* see if authority zones have an answer */
2613 	/* now we know the dp, we can check the auth zone for locally hosted
2614 	 * contents */
2615 	if(!iq->auth_zone_avoid && qstate->blacklist) {
2616 		if(auth_zones_can_fallback(qstate->env->auth_zones,
2617 			iq->dp->name, iq->dp->namelen, iq->qinfo_out.qclass)) {
2618 			/* if cache is blacklisted and this zone allows us
2619 			 * to fallback to the internet, then do so, and
2620 			 * fetch results from the internet servers */
2621 			iq->auth_zone_avoid = 1;
2622 		}
2623 	}
2624 	if(iq->auth_zone_avoid) {
2625 		iq->auth_zone_avoid = 0;
2626 		auth_fallback = 1;
2627 	} else if(auth_zones_lookup(qstate->env->auth_zones, &iq->qinfo_out,
2628 		qstate->region, &iq->response, &auth_fallback, iq->dp->name,
2629 		iq->dp->namelen)) {
2630 		/* use this as a response to be processed by the iterator */
2631 		if(verbosity >= VERB_ALGO) {
2632 			log_dns_msg("msg from auth zone",
2633 				&iq->response->qinfo, iq->response->rep);
2634 		}
2635 		if((iq->chase_flags&BIT_RD) && !(iq->response->rep->flags&BIT_AA)) {
2636 			verbose(VERB_ALGO, "forwarder, ignoring referral from auth zone");
2637 		} else {
2638 			lock_rw_wrlock(&qstate->env->auth_zones->lock);
2639 			qstate->env->auth_zones->num_query_up++;
2640 			lock_rw_unlock(&qstate->env->auth_zones->lock);
2641 			iq->num_current_queries++;
2642 			iq->chase_to_rd = 0;
2643 			iq->dnssec_lame_query = 0;
2644 			iq->auth_zone_response = 1;
2645 			return next_state(iq, QUERY_RESP_STATE);
2646 		}
2647 	}
2648 	iq->auth_zone_response = 0;
2649 	if(auth_fallback == 0) {
2650 		/* like we got servfail from the auth zone lookup, and
2651 		 * no internet fallback */
2652 		verbose(VERB_ALGO, "auth zone lookup failed, no fallback,"
2653 			" servfail");
2654 		errinf(qstate, "auth zone lookup failed, fallback is off");
2655 		return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2656 	}
2657 	if(iq->dp->auth_dp) {
2658 		/* we wanted to fallback, but had no delegpt, only the
2659 		 * auth zone generated delegpt, create an actual one */
2660 		iq->auth_zone_avoid = 1;
2661 		return next_state(iq, INIT_REQUEST_STATE);
2662 	}
2663 	/* but mostly, fallback==1 (like, when no such auth zone exists)
2664 	 * and we continue with lookups */
2665 
2666 	tf_policy = 0;
2667 	/* < not <=, because although the array is large enough for <=, the
2668 	 * generated query will immediately be discarded due to depth and
2669 	 * that servfail is cached, which is not good as opportunism goes. */
2670 	if(iq->depth < ie->max_dependency_depth
2671 		&& iq->num_target_queries == 0
2672 		&& (!iq->target_count || iq->target_count[TARGET_COUNT_NX]==0)
2673 		&& iq->sent_count < TARGET_FETCH_STOP) {
2674 		can_do_promisc = 1;
2675 	}
2676 	/* if the mesh query list is full, then do not waste cpu and sockets to
2677 	 * fetch promiscuous targets. They can be looked up when needed. */
2678 	if(can_do_promisc && !mesh_jostle_exceeded(qstate->env->mesh)) {
2679 		tf_policy = ie->target_fetch_policy[iq->depth];
2680 	}
2681 
2682 	/* if in 0x20 fallback get as many targets as possible */
2683 	if(iq->caps_fallback) {
2684 		int extra = 0;
2685 		size_t naddr, nres, navail;
2686 		if(!query_for_targets(qstate, iq, ie, id, -1, &extra)) {
2687 			errinf(qstate, "could not fetch nameservers for 0x20 fallback");
2688 			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2689 		}
2690 		iq->num_target_queries += extra;
2691 		target_count_increase(iq, extra);
2692 		if(iq->num_target_queries > 0) {
2693 			/* wait to get all targets, we want to try em */
2694 			verbose(VERB_ALGO, "wait for all targets for fallback");
2695 			qstate->ext_state[id] = module_wait_reply;
2696 			/* undo qname minimise step because we'll get back here
2697 			 * to do it again */
2698 			if(qout_orig && iq->minimise_count > 0) {
2699 				iq->minimise_count--;
2700 				iq->qinfo_out.qname = qout_orig;
2701 				iq->qinfo_out.qname_len = qout_orig_len;
2702 			}
2703 			return 0;
2704 		}
2705 		/* did we do enough fallback queries already? */
2706 		delegpt_count_addr(iq->dp, &naddr, &nres, &navail);
2707 		/* the current caps_server is the number of fallbacks sent.
2708 		 * the original query is one that matched too, so we have
2709 		 * caps_server+1 number of matching queries now */
2710 		if(iq->caps_server+1 >= naddr*3 ||
2711 			iq->caps_server*2+2 >= (size_t)ie->max_sent_count) {
2712 			/* *2 on sentcount check because ipv6 may fail */
2713 			/* we're done, process the response */
2714 			verbose(VERB_ALGO, "0x20 fallback had %d responses "
2715 				"match for %d wanted, done.",
2716 				(int)iq->caps_server+1, (int)naddr*3);
2717 			iq->response = iq->caps_response;
2718 			iq->caps_fallback = 0;
2719 			iter_dec_attempts(iq->dp, 3, ie->outbound_msg_retry); /* space for fallback */
2720 			iq->num_current_queries++; /* RespState decrements it*/
2721 			iq->referral_count++; /* make sure we don't loop */
2722 			iq->sent_count = 0;
2723 			iq->dp_target_count = 0;
2724 			iq->state = QUERY_RESP_STATE;
2725 			return 1;
2726 		}
2727 		verbose(VERB_ALGO, "0x20 fallback number %d",
2728 			(int)iq->caps_server);
2729 
2730 	/* if there is a policy to fetch missing targets
2731 	 * opportunistically, do it. we rely on the fact that once a
2732 	 * query (or queries) for a missing name have been issued,
2733 	 * they will not show up again. */
2734 	} else if(tf_policy != 0) {
2735 		int extra = 0;
2736 		verbose(VERB_ALGO, "attempt to get extra %d targets",
2737 			tf_policy);
2738 		(void)query_for_targets(qstate, iq, ie, id, tf_policy, &extra);
2739 		/* errors ignored, these targets are not strictly necessary for
2740 		 * this result, we do not have to reply with SERVFAIL */
2741 		iq->num_target_queries += extra;
2742 		target_count_increase(iq, extra);
2743 	}
2744 
2745 	/* Add the current set of unused targets to our queue. */
2746 	delegpt_add_unused_targets(iq->dp);
2747 
2748 	if(qstate->env->auth_zones) {
2749 		/* apply rpz triggers at query time */
2750 		struct dns_msg* forged_response = rpz_callback_from_iterator_module(qstate, iq);
2751 		if(forged_response != NULL) {
2752 			qstate->ext_state[id] = module_finished;
2753 			qstate->return_rcode = LDNS_RCODE_NOERROR;
2754 			qstate->return_msg = forged_response;
2755 			iq->response = forged_response;
2756 			next_state(iq, FINISHED_STATE);
2757 			if(!iter_prepend(iq, qstate->return_msg, qstate->region)) {
2758 				log_err("rpz: prepend rrsets: out of memory");
2759 				return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2760 			}
2761 			return 0;
2762 		}
2763 	}
2764 
2765 	/* Select the next usable target, filtering out unsuitable targets. */
2766 	target = iter_server_selection(ie, qstate->env, iq->dp,
2767 		iq->dp->name, iq->dp->namelen, iq->qchase.qtype,
2768 		&iq->dnssec_lame_query, &iq->chase_to_rd,
2769 		iq->num_target_queries, qstate->blacklist,
2770 		qstate->prefetch_leeway);
2771 
2772 	/* If no usable target was selected... */
2773 	if(!target) {
2774 		/* Here we distinguish between three states: generate a new
2775 		 * target query, just wait, or quit (with a SERVFAIL).
2776 		 * We have the following information: number of active
2777 		 * target queries, number of active current queries,
2778 		 * the presence of missing targets at this delegation
2779 		 * point, and the given query target policy. */
2780 
2781 		/* Check for the wait condition. If this is true, then
2782 		 * an action must be taken. */
2783 		if(iq->num_target_queries==0 && iq->num_current_queries==0) {
2784 			/* If there is nothing to wait for, then we need
2785 			 * to distinguish between generating (a) new target
2786 			 * query, or failing. */
2787 			if(delegpt_count_missing_targets(iq->dp, NULL) > 0) {
2788 				int qs = 0;
2789 				verbose(VERB_ALGO, "querying for next "
2790 					"missing target");
2791 				if(!query_for_targets(qstate, iq, ie, id,
2792 					1, &qs)) {
2793 					errinf(qstate, "could not fetch nameserver");
2794 					errinf_dname(qstate, "at zone", iq->dp->name);
2795 					return error_response(qstate, id,
2796 						LDNS_RCODE_SERVFAIL);
2797 				}
2798 				if(qs == 0 &&
2799 				   delegpt_count_missing_targets(iq->dp, NULL) == 0){
2800 					/* it looked like there were missing
2801 					 * targets, but they did not turn up.
2802 					 * Try the bad choices again (if any),
2803 					 * when we get back here missing==0,
2804 					 * so this is not a loop. */
2805 					return 1;
2806 				}
2807 				iq->num_target_queries += qs;
2808 				target_count_increase(iq, qs);
2809 			}
2810 			/* Since a target query might have been made, we
2811 			 * need to check again. */
2812 			if(iq->num_target_queries == 0) {
2813 				/* if in capsforid fallback, instead of last
2814 				 * resort, we agree with the current reply
2815 				 * we have (if any) (our count of addrs bad)*/
2816 				if(iq->caps_fallback && iq->caps_reply) {
2817 					/* we're done, process the response */
2818 					verbose(VERB_ALGO, "0x20 fallback had %d responses, "
2819 						"but no more servers except "
2820 						"last resort, done.",
2821 						(int)iq->caps_server+1);
2822 					iq->response = iq->caps_response;
2823 					iq->caps_fallback = 0;
2824 					iter_dec_attempts(iq->dp, 3, ie->outbound_msg_retry); /* space for fallback */
2825 					iq->num_current_queries++; /* RespState decrements it*/
2826 					iq->referral_count++; /* make sure we don't loop */
2827 					iq->sent_count = 0;
2828 					iq->dp_target_count = 0;
2829 					iq->state = QUERY_RESP_STATE;
2830 					return 1;
2831 				}
2832 				return processLastResort(qstate, iq, ie, id);
2833 			}
2834 		}
2835 
2836 		/* otherwise, we have no current targets, so submerge
2837 		 * until one of the target or direct queries return. */
2838 		verbose(VERB_ALGO, "no current targets");
2839 		check_waiting_queries(iq, qstate, id);
2840 		/* undo qname minimise step because we'll get back here
2841 		 * to do it again */
2842 		if(qout_orig && iq->minimise_count > 0) {
2843 			iq->minimise_count--;
2844 			iq->qinfo_out.qname = qout_orig;
2845 			iq->qinfo_out.qname_len = qout_orig_len;
2846 		}
2847 		return 0;
2848 	}
2849 
2850 	/* We have a target. We could have created promiscuous target
2851 	 * queries but we are currently under pressure (mesh_jostle_exceeded).
2852 	 * If we are configured to allow promiscuous target queries and haven't
2853 	 * gone out to the network for a target query for this delegation, then
2854 	 * it is possible to slip in a promiscuous one with a 1/10 chance. */
2855 	if(can_do_promisc && tf_policy == 0 && iq->depth == 0
2856 		&& iq->depth < ie->max_dependency_depth
2857 		&& ie->target_fetch_policy[iq->depth] != 0
2858 		&& iq->dp_target_count == 0
2859 		&& !ub_random_max(qstate->env->rnd, 10)) {
2860 		int extra = 0;
2861 		verbose(VERB_ALGO, "available target exists in cache but "
2862 			"attempt to get extra 1 target");
2863 		(void)query_for_targets(qstate, iq, ie, id, 1, &extra);
2864 		/* errors ignored, these targets are not strictly necessary for
2865 		* this result, we do not have to reply with SERVFAIL */
2866 		if(extra > 0) {
2867 			iq->num_target_queries += extra;
2868 			target_count_increase(iq, extra);
2869 			check_waiting_queries(iq, qstate, id);
2870 			/* undo qname minimise step because we'll get back here
2871 			 * to do it again */
2872 			if(qout_orig && iq->minimise_count > 0) {
2873 				iq->minimise_count--;
2874 				iq->qinfo_out.qname = qout_orig;
2875 				iq->qinfo_out.qname_len = qout_orig_len;
2876 			}
2877 			return 0;
2878 		}
2879 	}
2880 
2881 	/* Do not check ratelimit for forwarding queries or if we already got a
2882 	 * pass. */
2883 	sq_check_ratelimit = (!(iq->chase_flags & BIT_RD) && !iq->ratelimit_ok);
2884 	/* We have a valid target. */
2885 	if(verbosity >= VERB_QUERY) {
2886 		log_query_info(VERB_QUERY, "sending query:", &iq->qinfo_out);
2887 		log_name_addr(VERB_QUERY, "sending to target:", iq->dp->name,
2888 			&target->addr, target->addrlen);
2889 		verbose(VERB_ALGO, "dnssec status: %s%s",
2890 			iq->dnssec_expected?"expected": "not expected",
2891 			iq->dnssec_lame_query?" but lame_query anyway": "");
2892 	}
2893 
2894 	real_addr = target->addr;
2895 	real_addrlen = target->addrlen;
2896 
2897 	if(ie->use_nat64 && target->addr.ss_family == AF_INET) {
2898 		addr_to_nat64(&target->addr, &ie->nat64_prefix_addr,
2899 			ie->nat64_prefix_addrlen, ie->nat64_prefix_net,
2900 			&real_addr, &real_addrlen);
2901 		log_name_addr(VERB_QUERY, "applied NAT64:",
2902 			iq->dp->name, &real_addr, real_addrlen);
2903 	}
2904 
2905 	fptr_ok(fptr_whitelist_modenv_send_query(qstate->env->send_query));
2906 	outq = (*qstate->env->send_query)(&iq->qinfo_out,
2907 		iq->chase_flags | (iq->chase_to_rd?BIT_RD:0),
2908 		/* unset CD if to forwarder(RD set) and not dnssec retry
2909 		 * (blacklist nonempty) and no trust-anchors are configured
2910 		 * above the qname or on the first attempt when dnssec is on */
2911 		(qstate->env->cfg->disable_edns_do?0:EDNS_DO)|
2912 		((iq->chase_to_rd||(iq->chase_flags&BIT_RD)!=0)&&
2913 		!qstate->blacklist&&(!iter_qname_indicates_dnssec(qstate->env,
2914 		&iq->qinfo_out)||target->attempts==1)?0:BIT_CD),
2915 		iq->dnssec_expected, iq->caps_fallback || is_caps_whitelisted(
2916 		ie, iq), sq_check_ratelimit, &real_addr, real_addrlen,
2917 		iq->dp->name, iq->dp->namelen,
2918 		(iq->dp->tcp_upstream || qstate->env->cfg->tcp_upstream),
2919 		(iq->dp->ssl_upstream || qstate->env->cfg->ssl_upstream),
2920 		target->tls_auth_name, qstate, &sq_was_ratelimited);
2921 	if(!outq) {
2922 		if(sq_was_ratelimited) {
2923 			lock_basic_lock(&ie->queries_ratelimit_lock);
2924 			ie->num_queries_ratelimited++;
2925 			lock_basic_unlock(&ie->queries_ratelimit_lock);
2926 			verbose(VERB_ALGO, "query exceeded ratelimits");
2927 			qstate->was_ratelimited = 1;
2928 			errinf_dname(qstate, "exceeded ratelimit for zone",
2929 				iq->dp->name);
2930 			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
2931 		}
2932 		log_addr(VERB_QUERY, "error sending query to auth server",
2933 			&real_addr, real_addrlen);
2934 		if(qstate->env->cfg->qname_minimisation)
2935 			iq->minimisation_state = SKIP_MINIMISE_STATE;
2936 		return next_state(iq, QUERYTARGETS_STATE);
2937 	}
2938 	outbound_list_insert(&iq->outlist, outq);
2939 	iq->num_current_queries++;
2940 	iq->sent_count++;
2941 	qstate->ext_state[id] = module_wait_reply;
2942 
2943 	return 0;
2944 }
2945 
2946 /** find NS rrset in given list */
2947 static struct ub_packed_rrset_key*
find_NS(struct reply_info * rep,size_t from,size_t to)2948 find_NS(struct reply_info* rep, size_t from, size_t to)
2949 {
2950 	size_t i;
2951 	for(i=from; i<to; i++) {
2952 		if(ntohs(rep->rrsets[i]->rk.type) == LDNS_RR_TYPE_NS)
2953 			return rep->rrsets[i];
2954 	}
2955 	return NULL;
2956 }
2957 
2958 
2959 /**
2960  * Process the query response. All queries end up at this state first. This
2961  * process generally consists of analyzing the response and routing the
2962  * event to the next state (either bouncing it back to a request state, or
2963  * terminating the processing for this event).
2964  *
2965  * @param qstate: query state.
2966  * @param iq: iterator query state.
2967  * @param ie: iterator shared global environment.
2968  * @param id: module id.
2969  * @return true if the event requires more immediate processing, false if
2970  *         not. This is generally only true when forwarding the request to
2971  *         the final state (i.e., on answer).
2972  */
2973 static int
processQueryResponse(struct module_qstate * qstate,struct iter_qstate * iq,struct iter_env * ie,int id)2974 processQueryResponse(struct module_qstate* qstate, struct iter_qstate* iq,
2975 	struct iter_env* ie, int id)
2976 {
2977 	int dnsseclame = 0, origtypecname = 0, orig_empty_nodata_found;
2978 	enum response_type type;
2979 
2980 	iq->num_current_queries--;
2981 
2982 	if(!inplace_cb_query_response_call(qstate->env, qstate, iq->response))
2983 		log_err("unable to call query_response callback");
2984 
2985 	if(iq->response == NULL) {
2986 		/* Don't increment qname when QNAME minimisation is enabled */
2987 		if(qstate->env->cfg->qname_minimisation) {
2988 			iq->minimisation_state = SKIP_MINIMISE_STATE;
2989 		}
2990 		iq->timeout_count++;
2991 		iq->chase_to_rd = 0;
2992 		iq->dnssec_lame_query = 0;
2993 		verbose(VERB_ALGO, "query response was timeout");
2994 		return next_state(iq, QUERYTARGETS_STATE);
2995 	}
2996 	iq->timeout_count = 0;
2997 	orig_empty_nodata_found = iq->empty_nodata_found;
2998 	type = response_type_from_server(
2999 		(int)((iq->chase_flags&BIT_RD) || iq->chase_to_rd),
3000 		iq->response, &iq->qinfo_out, iq->dp, &iq->empty_nodata_found);
3001 	iq->chase_to_rd = 0;
3002 	/* remove TC flag, if this is erroneously set by TCP upstream */
3003 	iq->response->rep->flags &= ~BIT_TC;
3004 	if(orig_empty_nodata_found != iq->empty_nodata_found &&
3005 		iq->empty_nodata_found < EMPTY_NODATA_RETRY_COUNT) {
3006 		/* try to search at another server */
3007 		if(qstate->reply) {
3008 			struct delegpt_addr* a = delegpt_find_addr(
3009 				iq->dp, &qstate->reply->remote_addr,
3010 				qstate->reply->remote_addrlen);
3011 			/* make selection disprefer it */
3012 			if(a) a->lame = 1;
3013 		}
3014 		return next_state(iq, QUERYTARGETS_STATE);
3015 	}
3016 	if(type == RESPONSE_TYPE_REFERRAL && (iq->chase_flags&BIT_RD) &&
3017 		!iq->auth_zone_response) {
3018 		/* When forwarding (RD bit is set), we handle referrals
3019 		 * differently. No queries should be sent elsewhere */
3020 		type = RESPONSE_TYPE_ANSWER;
3021 	}
3022 	if(!qstate->env->cfg->disable_dnssec_lame_check && iq->dnssec_expected
3023                 && !iq->dnssec_lame_query &&
3024 		!(iq->chase_flags&BIT_RD)
3025 		&& iq->sent_count < DNSSEC_LAME_DETECT_COUNT
3026 		&& type != RESPONSE_TYPE_LAME
3027 		&& type != RESPONSE_TYPE_REC_LAME
3028 		&& type != RESPONSE_TYPE_THROWAWAY
3029 		&& type != RESPONSE_TYPE_UNTYPED) {
3030 		/* a possible answer, see if it is missing DNSSEC */
3031 		/* but not when forwarding, so we dont mark fwder lame */
3032 		if(!iter_msg_has_dnssec(iq->response)) {
3033 			/* Mark this address as dnsseclame in this dp,
3034 			 * because that will make serverselection disprefer
3035 			 * it, but also, once it is the only final option,
3036 			 * use dnssec-lame-bypass if it needs to query there.*/
3037 			if(qstate->reply) {
3038 				struct delegpt_addr* a = delegpt_find_addr(
3039 					iq->dp, &qstate->reply->remote_addr,
3040 					qstate->reply->remote_addrlen);
3041 				if(a) a->dnsseclame = 1;
3042 			}
3043 			/* test the answer is from the zone we expected,
3044 		 	 * otherwise, (due to parent,child on same server), we
3045 		 	 * might mark the server,zone lame inappropriately */
3046 			if(!iter_msg_from_zone(iq->response, iq->dp, type,
3047 				iq->qchase.qclass))
3048 				qstate->reply = NULL;
3049 			type = RESPONSE_TYPE_LAME;
3050 			dnsseclame = 1;
3051 		}
3052 	} else iq->dnssec_lame_query = 0;
3053 	/* see if referral brings us close to the target */
3054 	if(type == RESPONSE_TYPE_REFERRAL) {
3055 		struct ub_packed_rrset_key* ns = find_NS(
3056 			iq->response->rep, iq->response->rep->an_numrrsets,
3057 			iq->response->rep->an_numrrsets
3058 			+ iq->response->rep->ns_numrrsets);
3059 		if(!ns) ns = find_NS(iq->response->rep, 0,
3060 				iq->response->rep->an_numrrsets);
3061 		if(!ns || !dname_strict_subdomain_c(ns->rk.dname, iq->dp->name)
3062 			|| !dname_subdomain_c(iq->qchase.qname, ns->rk.dname)){
3063 			verbose(VERB_ALGO, "bad referral, throwaway");
3064 			type = RESPONSE_TYPE_THROWAWAY;
3065 		} else
3066 			iter_scrub_ds(iq->response, ns, iq->dp->name);
3067 	} else iter_scrub_ds(iq->response, NULL, NULL);
3068 	if(type == RESPONSE_TYPE_THROWAWAY &&
3069 		FLAGS_GET_RCODE(iq->response->rep->flags) == LDNS_RCODE_YXDOMAIN) {
3070 		/* YXDOMAIN is a permanent error, no need to retry */
3071 		type = RESPONSE_TYPE_ANSWER;
3072 	}
3073 	if(type == RESPONSE_TYPE_CNAME)
3074 		origtypecname = 1;
3075 	if(type == RESPONSE_TYPE_CNAME && iq->response->rep->an_numrrsets >= 1
3076 		&& ntohs(iq->response->rep->rrsets[0]->rk.type) == LDNS_RR_TYPE_DNAME) {
3077 		uint8_t* sname = NULL;
3078 		size_t snamelen = 0;
3079 		get_cname_target(iq->response->rep->rrsets[0], &sname,
3080 			&snamelen);
3081 		if(snamelen && dname_subdomain_c(sname, iq->response->rep->rrsets[0]->rk.dname)) {
3082 			/* DNAME to a subdomain loop; do not recurse */
3083 			type = RESPONSE_TYPE_ANSWER;
3084 		}
3085 	} else if(type == RESPONSE_TYPE_CNAME &&
3086 		iq->qchase.qtype == LDNS_RR_TYPE_CNAME &&
3087 		iq->minimisation_state == MINIMISE_STATE &&
3088 		query_dname_compare(iq->qchase.qname, iq->qinfo_out.qname) == 0) {
3089 		/* The minimised query for full QTYPE and hidden QTYPE can be
3090 		 * classified as CNAME response type, even when the original
3091 		 * QTYPE=CNAME. This should be treated as answer response type.
3092 		 */
3093 		type = RESPONSE_TYPE_ANSWER;
3094 	}
3095 
3096 	/* handle each of the type cases */
3097 	if(type == RESPONSE_TYPE_ANSWER) {
3098 		/* ANSWER type responses terminate the query algorithm,
3099 		 * so they sent on their */
3100 		if(verbosity >= VERB_DETAIL) {
3101 			verbose(VERB_DETAIL, "query response was %s",
3102 				FLAGS_GET_RCODE(iq->response->rep->flags)
3103 				==LDNS_RCODE_NXDOMAIN?"NXDOMAIN ANSWER":
3104 				(iq->response->rep->an_numrrsets?"ANSWER":
3105 				"nodata ANSWER"));
3106 		}
3107 		/* if qtype is DS, check we have the right level of answer,
3108 		 * like grandchild answer but we need the middle, reject it */
3109 		if(iq->qchase.qtype == LDNS_RR_TYPE_DS && !iq->dsns_point
3110 			&& !(iq->chase_flags&BIT_RD)
3111 			&& iter_ds_toolow(iq->response, iq->dp)
3112 			&& iter_dp_cangodown(&iq->qchase, iq->dp)) {
3113 			/* close down outstanding requests to be discarded */
3114 			outbound_list_clear(&iq->outlist);
3115 			iq->num_current_queries = 0;
3116 			fptr_ok(fptr_whitelist_modenv_detach_subs(
3117 				qstate->env->detach_subs));
3118 			(*qstate->env->detach_subs)(qstate);
3119 			iq->num_target_queries = 0;
3120 			return processDSNSFind(qstate, iq, id);
3121 		}
3122 		if(!qstate->no_cache_store)
3123 			iter_dns_store(qstate->env, &iq->response->qinfo,
3124 				iq->response->rep,
3125 				iq->qchase.qtype != iq->response->qinfo.qtype,
3126 				qstate->prefetch_leeway,
3127 				iq->dp&&iq->dp->has_parent_side_NS,
3128 				qstate->region, qstate->query_flags,
3129 				qstate->qstarttime);
3130 		/* close down outstanding requests to be discarded */
3131 		outbound_list_clear(&iq->outlist);
3132 		iq->num_current_queries = 0;
3133 		fptr_ok(fptr_whitelist_modenv_detach_subs(
3134 			qstate->env->detach_subs));
3135 		(*qstate->env->detach_subs)(qstate);
3136 		iq->num_target_queries = 0;
3137 		if(qstate->reply)
3138 			sock_list_insert(&qstate->reply_origin,
3139 				&qstate->reply->remote_addr,
3140 				qstate->reply->remote_addrlen, qstate->region);
3141 		if(iq->minimisation_state != DONOT_MINIMISE_STATE
3142 			&& !(iq->chase_flags & BIT_RD)) {
3143 			if(FLAGS_GET_RCODE(iq->response->rep->flags) !=
3144 				LDNS_RCODE_NOERROR) {
3145 				if(qstate->env->cfg->qname_minimisation_strict) {
3146 					if(FLAGS_GET_RCODE(iq->response->rep->flags) ==
3147 						LDNS_RCODE_NXDOMAIN) {
3148 						iter_scrub_nxdomain(iq->response);
3149 						return final_state(iq);
3150 					}
3151 					return error_response(qstate, id,
3152 						LDNS_RCODE_SERVFAIL);
3153 				}
3154 				/* Best effort qname-minimisation.
3155 				 * Stop minimising and send full query when
3156 				 * RCODE is not NOERROR. */
3157 				iq->minimisation_state = DONOT_MINIMISE_STATE;
3158 			}
3159 			if(FLAGS_GET_RCODE(iq->response->rep->flags) ==
3160 				LDNS_RCODE_NXDOMAIN && !origtypecname) {
3161 				/* Stop resolving when NXDOMAIN is DNSSEC
3162 				 * signed. Based on assumption that nameservers
3163 				 * serving signed zones do not return NXDOMAIN
3164 				 * for empty-non-terminals. */
3165 				/* If this response is actually a CNAME type,
3166 				 * the nxdomain rcode may not be for the qname,
3167 				 * and so it is not the final response. */
3168 				if(iq->dnssec_expected)
3169 					return final_state(iq);
3170 				/* Make subrequest to validate intermediate
3171 				 * NXDOMAIN if harden-below-nxdomain is
3172 				 * enabled. */
3173 				if(qstate->env->cfg->harden_below_nxdomain &&
3174 					qstate->env->need_to_validate) {
3175 					struct module_qstate* subq = NULL;
3176 					log_query_info(VERB_QUERY,
3177 						"schedule NXDOMAIN validation:",
3178 						&iq->response->qinfo);
3179 					if(!generate_sub_request(
3180 						iq->response->qinfo.qname,
3181 						iq->response->qinfo.qname_len,
3182 						iq->response->qinfo.qtype,
3183 						iq->response->qinfo.qclass,
3184 						qstate, id, iq,
3185 						INIT_REQUEST_STATE,
3186 						FINISHED_STATE, &subq, 1, 1))
3187 						verbose(VERB_ALGO,
3188 						"could not validate NXDOMAIN "
3189 						"response");
3190 				}
3191 			}
3192 			return next_state(iq, QUERYTARGETS_STATE);
3193 		}
3194 		return final_state(iq);
3195 	} else if(type == RESPONSE_TYPE_REFERRAL) {
3196 		/* REFERRAL type responses get a reset of the
3197 		 * delegation point, and back to the QUERYTARGETS_STATE. */
3198 		verbose(VERB_DETAIL, "query response was REFERRAL");
3199 
3200 		/* if hardened, only store referral if we asked for it */
3201 		if(!qstate->no_cache_store &&
3202 		(!qstate->env->cfg->harden_referral_path ||
3203 		    (  qstate->qinfo.qtype == LDNS_RR_TYPE_NS
3204 			&& (qstate->query_flags&BIT_RD)
3205 			&& !(qstate->query_flags&BIT_CD)
3206 			   /* we know that all other NS rrsets are scrubbed
3207 			    * away, thus on referral only one is left.
3208 			    * see if that equals the query name... */
3209 			&& ( /* auth section, but sometimes in answer section*/
3210 			  reply_find_rrset_section_ns(iq->response->rep,
3211 				iq->qchase.qname, iq->qchase.qname_len,
3212 				LDNS_RR_TYPE_NS, iq->qchase.qclass)
3213 			  || reply_find_rrset_section_an(iq->response->rep,
3214 				iq->qchase.qname, iq->qchase.qname_len,
3215 				LDNS_RR_TYPE_NS, iq->qchase.qclass)
3216 			  )
3217 		    ))) {
3218 			/* Store the referral under the current query */
3219 			/* no prefetch-leeway, since its not the answer */
3220 			iter_dns_store(qstate->env, &iq->response->qinfo,
3221 				iq->response->rep, 1, 0, 0, NULL, 0,
3222 				qstate->qstarttime);
3223 			if(iq->store_parent_NS)
3224 				iter_store_parentside_NS(qstate->env,
3225 					iq->response->rep);
3226 			if(qstate->env->neg_cache)
3227 				val_neg_addreferral(qstate->env->neg_cache,
3228 					iq->response->rep, iq->dp->name);
3229 		}
3230 		/* store parent-side-in-zone-glue, if directly queried for */
3231 		if(!qstate->no_cache_store && iq->query_for_pside_glue
3232 			&& !iq->pside_glue) {
3233 				iq->pside_glue = reply_find_rrset(iq->response->rep,
3234 					iq->qchase.qname, iq->qchase.qname_len,
3235 					iq->qchase.qtype, iq->qchase.qclass);
3236 				if(iq->pside_glue) {
3237 					log_rrset_key(VERB_ALGO, "found parent-side "
3238 						"glue", iq->pside_glue);
3239 					iter_store_parentside_rrset(qstate->env,
3240 						iq->pside_glue);
3241 				}
3242 		}
3243 
3244 		/* Reset the event state, setting the current delegation
3245 		 * point to the referral. */
3246 		iq->deleg_msg = iq->response;
3247 		iq->dp = delegpt_from_message(iq->response, qstate->region);
3248 		if (qstate->env->cfg->qname_minimisation)
3249 			iq->minimisation_state = INIT_MINIMISE_STATE;
3250 		if(!iq->dp) {
3251 			errinf(qstate, "malloc failure, for delegation point");
3252 			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3253 		}
3254 		if(!cache_fill_missing(qstate->env, iq->qchase.qclass,
3255 			qstate->region, iq->dp)) {
3256 			errinf(qstate, "malloc failure, copy extra info into delegation point");
3257 			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3258 		}
3259 		if(iq->store_parent_NS && query_dname_compare(iq->dp->name,
3260 			iq->store_parent_NS->name) == 0)
3261 			iter_merge_retry_counts(iq->dp, iq->store_parent_NS,
3262 				ie->outbound_msg_retry);
3263 		delegpt_log(VERB_ALGO, iq->dp);
3264 		/* Count this as a referral. */
3265 		iq->referral_count++;
3266 		iq->sent_count = 0;
3267 		iq->dp_target_count = 0;
3268 		/* see if the next dp is a trust anchor, or a DS was sent
3269 		 * along, indicating dnssec is expected for next zone */
3270 		iq->dnssec_expected = iter_indicates_dnssec(qstate->env,
3271 			iq->dp, iq->response, iq->qchase.qclass);
3272 		/* if dnssec, validating then also fetch the key for the DS */
3273 		if(iq->dnssec_expected && qstate->env->cfg->prefetch_key &&
3274 			!(qstate->query_flags&BIT_CD))
3275 			generate_dnskey_prefetch(qstate, iq, id);
3276 
3277 		/* spawn off NS and addr to auth servers for the NS we just
3278 		 * got in the referral. This gets authoritative answer
3279 		 * (answer section trust level) rrset.
3280 		 * right after, we detach the subs, answer goes to cache. */
3281 		if(qstate->env->cfg->harden_referral_path)
3282 			generate_ns_check(qstate, iq, id);
3283 
3284 		/* stop current outstanding queries.
3285 		 * FIXME: should the outstanding queries be waited for and
3286 		 * handled? Say by a subquery that inherits the outbound_entry.
3287 		 */
3288 		outbound_list_clear(&iq->outlist);
3289 		iq->num_current_queries = 0;
3290 		fptr_ok(fptr_whitelist_modenv_detach_subs(
3291 			qstate->env->detach_subs));
3292 		(*qstate->env->detach_subs)(qstate);
3293 		iq->num_target_queries = 0;
3294 		iq->response = NULL;
3295 		iq->fail_addr_type = 0;
3296 		verbose(VERB_ALGO, "cleared outbound list for next round");
3297 		return next_state(iq, QUERYTARGETS_STATE);
3298 	} else if(type == RESPONSE_TYPE_CNAME) {
3299 		uint8_t* sname = NULL;
3300 		size_t snamelen = 0;
3301 		/* CNAME type responses get a query restart (i.e., get a
3302 		 * reset of the query state and go back to INIT_REQUEST_STATE).
3303 		 */
3304 		verbose(VERB_DETAIL, "query response was CNAME");
3305 		if(verbosity >= VERB_ALGO)
3306 			log_dns_msg("cname msg", &iq->response->qinfo,
3307 				iq->response->rep);
3308 		/* if qtype is DS, check we have the right level of answer,
3309 		 * like grandchild answer but we need the middle, reject it */
3310 		if(iq->qchase.qtype == LDNS_RR_TYPE_DS && !iq->dsns_point
3311 			&& !(iq->chase_flags&BIT_RD)
3312 			&& iter_ds_toolow(iq->response, iq->dp)
3313 			&& iter_dp_cangodown(&iq->qchase, iq->dp)) {
3314 			outbound_list_clear(&iq->outlist);
3315 			iq->num_current_queries = 0;
3316 			fptr_ok(fptr_whitelist_modenv_detach_subs(
3317 				qstate->env->detach_subs));
3318 			(*qstate->env->detach_subs)(qstate);
3319 			iq->num_target_queries = 0;
3320 			return processDSNSFind(qstate, iq, id);
3321 		}
3322 		/* Process the CNAME response. */
3323 		if(!handle_cname_response(qstate, iq, iq->response,
3324 			&sname, &snamelen)) {
3325 			errinf(qstate, "malloc failure, CNAME info");
3326 			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3327 		}
3328 		/* cache the CNAME response under the current query */
3329 		/* NOTE : set referral=1, so that rrsets get stored but not
3330 		 * the partial query answer (CNAME only). */
3331 		/* prefetchleeway applied because this updates answer parts */
3332 		if(!qstate->no_cache_store)
3333 			iter_dns_store(qstate->env, &iq->response->qinfo,
3334 				iq->response->rep, 1, qstate->prefetch_leeway,
3335 				iq->dp&&iq->dp->has_parent_side_NS, NULL,
3336 				qstate->query_flags, qstate->qstarttime);
3337 		/* set the current request's qname to the new value. */
3338 		iq->qchase.qname = sname;
3339 		iq->qchase.qname_len = snamelen;
3340 		if(qstate->env->auth_zones) {
3341 			/* apply rpz qname triggers after cname */
3342 			struct dns_msg* forged_response =
3343 				rpz_callback_from_iterator_cname(qstate, iq);
3344 			while(forged_response && reply_find_rrset_section_an(
3345 				forged_response->rep, iq->qchase.qname,
3346 				iq->qchase.qname_len, LDNS_RR_TYPE_CNAME,
3347 				iq->qchase.qclass)) {
3348 				/* another cname to follow */
3349 				if(!handle_cname_response(qstate, iq, forged_response,
3350 					&sname, &snamelen)) {
3351 					errinf(qstate, "malloc failure, CNAME info");
3352 					return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3353 				}
3354 				iq->qchase.qname = sname;
3355 				iq->qchase.qname_len = snamelen;
3356 				forged_response =
3357 					rpz_callback_from_iterator_cname(qstate, iq);
3358 			}
3359 			if(forged_response != NULL) {
3360 				qstate->ext_state[id] = module_finished;
3361 				qstate->return_rcode = LDNS_RCODE_NOERROR;
3362 				qstate->return_msg = forged_response;
3363 				iq->response = forged_response;
3364 				next_state(iq, FINISHED_STATE);
3365 				if(!iter_prepend(iq, qstate->return_msg, qstate->region)) {
3366 					log_err("rpz: after cname, prepend rrsets: out of memory");
3367 					return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3368 				}
3369 				qstate->return_msg->qinfo = qstate->qinfo;
3370 				return 0;
3371 			}
3372 		}
3373 		/* Clear the query state, since this is a query restart. */
3374 		iq->deleg_msg = NULL;
3375 		iq->dp = NULL;
3376 		iq->dsns_point = NULL;
3377 		iq->auth_zone_response = 0;
3378 		iq->sent_count = 0;
3379 		iq->dp_target_count = 0;
3380 		if(iq->minimisation_state != MINIMISE_STATE)
3381 			/* Only count as query restart when it is not an extra
3382 			 * query as result of qname minimisation. */
3383 			iq->query_restart_count++;
3384 		if(qstate->env->cfg->qname_minimisation)
3385 			iq->minimisation_state = INIT_MINIMISE_STATE;
3386 
3387 		/* stop current outstanding queries.
3388 		 * FIXME: should the outstanding queries be waited for and
3389 		 * handled? Say by a subquery that inherits the outbound_entry.
3390 		 */
3391 		outbound_list_clear(&iq->outlist);
3392 		iq->num_current_queries = 0;
3393 		fptr_ok(fptr_whitelist_modenv_detach_subs(
3394 			qstate->env->detach_subs));
3395 		(*qstate->env->detach_subs)(qstate);
3396 		iq->num_target_queries = 0;
3397 		if(qstate->reply)
3398 			sock_list_insert(&qstate->reply_origin,
3399 				&qstate->reply->remote_addr,
3400 				qstate->reply->remote_addrlen, qstate->region);
3401 		verbose(VERB_ALGO, "cleared outbound list for query restart");
3402 		/* go to INIT_REQUEST_STATE for new qname. */
3403 		return next_state(iq, INIT_REQUEST_STATE);
3404 	} else if(type == RESPONSE_TYPE_LAME) {
3405 		/* Cache the LAMEness. */
3406 		verbose(VERB_DETAIL, "query response was %sLAME",
3407 			dnsseclame?"DNSSEC ":"");
3408 		if(!dname_subdomain_c(iq->qchase.qname, iq->dp->name)) {
3409 			log_err("mark lame: mismatch in qname and dpname");
3410 			/* throwaway this reply below */
3411 		} else if(qstate->reply) {
3412 			/* need addr for lameness cache, but we may have
3413 			 * gotten this from cache, so test to be sure */
3414 			if(!infra_set_lame(qstate->env->infra_cache,
3415 				&qstate->reply->remote_addr,
3416 				qstate->reply->remote_addrlen,
3417 				iq->dp->name, iq->dp->namelen,
3418 				*qstate->env->now, dnsseclame, 0,
3419 				iq->qchase.qtype))
3420 				log_err("mark host lame: out of memory");
3421 		}
3422 	} else if(type == RESPONSE_TYPE_REC_LAME) {
3423 		/* Cache the LAMEness. */
3424 		verbose(VERB_DETAIL, "query response REC_LAME: "
3425 			"recursive but not authoritative server");
3426 		if(!dname_subdomain_c(iq->qchase.qname, iq->dp->name)) {
3427 			log_err("mark rec_lame: mismatch in qname and dpname");
3428 			/* throwaway this reply below */
3429 		} else if(qstate->reply) {
3430 			/* need addr for lameness cache, but we may have
3431 			 * gotten this from cache, so test to be sure */
3432 			verbose(VERB_DETAIL, "mark as REC_LAME");
3433 			if(!infra_set_lame(qstate->env->infra_cache,
3434 				&qstate->reply->remote_addr,
3435 				qstate->reply->remote_addrlen,
3436 				iq->dp->name, iq->dp->namelen,
3437 				*qstate->env->now, 0, 1, iq->qchase.qtype))
3438 				log_err("mark host lame: out of memory");
3439 		}
3440 	} else if(type == RESPONSE_TYPE_THROWAWAY) {
3441 		/* LAME and THROWAWAY responses are handled the same way.
3442 		 * In this case, the event is just sent directly back to
3443 		 * the QUERYTARGETS_STATE without resetting anything,
3444 		 * because, clearly, the next target must be tried. */
3445 		verbose(VERB_DETAIL, "query response was THROWAWAY");
3446 	} else {
3447 		log_warn("A query response came back with an unknown type: %d",
3448 			(int)type);
3449 	}
3450 
3451 	/* LAME, THROWAWAY and "unknown" all end up here.
3452 	 * Recycle to the QUERYTARGETS state to hopefully try a
3453 	 * different target. */
3454 	if (qstate->env->cfg->qname_minimisation &&
3455 		!qstate->env->cfg->qname_minimisation_strict)
3456 		iq->minimisation_state = DONOT_MINIMISE_STATE;
3457 	if(iq->auth_zone_response) {
3458 		/* can we fallback? */
3459 		iq->auth_zone_response = 0;
3460 		if(!auth_zones_can_fallback(qstate->env->auth_zones,
3461 			iq->dp->name, iq->dp->namelen, qstate->qinfo.qclass)) {
3462 			verbose(VERB_ALGO, "auth zone response bad, and no"
3463 				" fallback possible, servfail");
3464 			errinf_dname(qstate, "response is bad, no fallback, "
3465 				"for auth zone", iq->dp->name);
3466 			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3467 		}
3468 		verbose(VERB_ALGO, "auth zone response was bad, "
3469 			"fallback enabled");
3470 		iq->auth_zone_avoid = 1;
3471 		if(iq->dp->auth_dp) {
3472 			/* we are using a dp for the auth zone, with no
3473 			 * nameservers, get one first */
3474 			iq->dp = NULL;
3475 			return next_state(iq, INIT_REQUEST_STATE);
3476 		}
3477 	}
3478 	return next_state(iq, QUERYTARGETS_STATE);
3479 }
3480 
3481 /**
3482  * Return priming query results to interested super querystates.
3483  *
3484  * Sets the delegation point and delegation message (not nonRD queries).
3485  * This is a callback from walk_supers.
3486  *
3487  * @param qstate: priming query state that finished.
3488  * @param id: module id.
3489  * @param forq: the qstate for which priming has been done.
3490  */
3491 static void
prime_supers(struct module_qstate * qstate,int id,struct module_qstate * forq)3492 prime_supers(struct module_qstate* qstate, int id, struct module_qstate* forq)
3493 {
3494 	struct iter_qstate* foriq = (struct iter_qstate*)forq->minfo[id];
3495 	struct delegpt* dp = NULL;
3496 
3497 	log_assert(qstate->is_priming || foriq->wait_priming_stub);
3498 	log_assert(qstate->return_rcode == LDNS_RCODE_NOERROR);
3499 	/* Convert our response to a delegation point */
3500 	dp = delegpt_from_message(qstate->return_msg, forq->region);
3501 	if(!dp) {
3502 		/* if there is no convertible delegation point, then
3503 		 * the ANSWER type was (presumably) a negative answer. */
3504 		verbose(VERB_ALGO, "prime response was not a positive "
3505 			"ANSWER; failing");
3506 		foriq->dp = NULL;
3507 		foriq->state = QUERYTARGETS_STATE;
3508 		return;
3509 	}
3510 
3511 	log_query_info(VERB_DETAIL, "priming successful for", &qstate->qinfo);
3512 	delegpt_log(VERB_ALGO, dp);
3513 	foriq->dp = dp;
3514 	foriq->deleg_msg = dns_copy_msg(qstate->return_msg, forq->region);
3515 	if(!foriq->deleg_msg) {
3516 		log_err("copy prime response: out of memory");
3517 		foriq->dp = NULL;
3518 		foriq->state = QUERYTARGETS_STATE;
3519 		return;
3520 	}
3521 
3522 	/* root priming responses go to init stage 2, priming stub
3523 	 * responses to to stage 3. */
3524 	if(foriq->wait_priming_stub) {
3525 		foriq->state = INIT_REQUEST_3_STATE;
3526 		foriq->wait_priming_stub = 0;
3527 	} else	foriq->state = INIT_REQUEST_2_STATE;
3528 	/* because we are finished, the parent will be reactivated */
3529 }
3530 
3531 /**
3532  * This handles the response to a priming query. This is used to handle both
3533  * root and stub priming responses. This is basically the equivalent of the
3534  * QUERY_RESP_STATE, but will not handle CNAME responses and will treat
3535  * REFERRALs as ANSWERS. It will also update and reactivate the originating
3536  * event.
3537  *
3538  * @param qstate: query state.
3539  * @param id: module id.
3540  * @return true if the event needs more immediate processing, false if not.
3541  *         This state always returns false.
3542  */
3543 static int
processPrimeResponse(struct module_qstate * qstate,int id)3544 processPrimeResponse(struct module_qstate* qstate, int id)
3545 {
3546 	struct iter_qstate* iq = (struct iter_qstate*)qstate->minfo[id];
3547 	enum response_type type;
3548 	iq->response->rep->flags &= ~(BIT_RD|BIT_RA); /* ignore rec-lame */
3549 	type = response_type_from_server(
3550 		(int)((iq->chase_flags&BIT_RD) || iq->chase_to_rd),
3551 		iq->response, &iq->qchase, iq->dp, NULL);
3552 	if(type == RESPONSE_TYPE_ANSWER) {
3553 		qstate->return_rcode = LDNS_RCODE_NOERROR;
3554 		qstate->return_msg = iq->response;
3555 	} else {
3556 		errinf(qstate, "prime response did not get an answer");
3557 		errinf_dname(qstate, "for", qstate->qinfo.qname);
3558 		qstate->return_rcode = LDNS_RCODE_SERVFAIL;
3559 		qstate->return_msg = NULL;
3560 	}
3561 
3562 	/* validate the root or stub after priming (if enabled).
3563 	 * This is the same query as the prime query, but with validation.
3564 	 * Now that we are primed, the additional queries that validation
3565 	 * may need can be resolved. */
3566 	if(qstate->env->cfg->harden_referral_path) {
3567 		struct module_qstate* subq = NULL;
3568 		log_nametypeclass(VERB_ALGO, "schedule prime validation",
3569 			qstate->qinfo.qname, qstate->qinfo.qtype,
3570 			qstate->qinfo.qclass);
3571 		if(!generate_sub_request(qstate->qinfo.qname,
3572 			qstate->qinfo.qname_len, qstate->qinfo.qtype,
3573 			qstate->qinfo.qclass, qstate, id, iq,
3574 			INIT_REQUEST_STATE, FINISHED_STATE, &subq, 1, 0)) {
3575 			verbose(VERB_ALGO, "could not generate prime check");
3576 		}
3577 		generate_a_aaaa_check(qstate, iq, id);
3578 	}
3579 
3580 	/* This event is finished. */
3581 	qstate->ext_state[id] = module_finished;
3582 	return 0;
3583 }
3584 
3585 /**
3586  * Do final processing on responses to target queries. Events reach this
3587  * state after the iterative resolution algorithm terminates. This state is
3588  * responsible for reactivating the original event, and housekeeping related
3589  * to received target responses (caching, updating the current delegation
3590  * point, etc).
3591  * Callback from walk_supers for every super state that is interested in
3592  * the results from this query.
3593  *
3594  * @param qstate: query state.
3595  * @param id: module id.
3596  * @param forq: super query state.
3597  */
3598 static void
processTargetResponse(struct module_qstate * qstate,int id,struct module_qstate * forq)3599 processTargetResponse(struct module_qstate* qstate, int id,
3600 	struct module_qstate* forq)
3601 {
3602 	struct iter_env* ie = (struct iter_env*)qstate->env->modinfo[id];
3603 	struct iter_qstate* iq = (struct iter_qstate*)qstate->minfo[id];
3604 	struct iter_qstate* foriq = (struct iter_qstate*)forq->minfo[id];
3605 	struct ub_packed_rrset_key* rrset;
3606 	struct delegpt_ns* dpns;
3607 	log_assert(qstate->return_rcode == LDNS_RCODE_NOERROR);
3608 
3609 	foriq->state = QUERYTARGETS_STATE;
3610 	log_query_info(VERB_ALGO, "processTargetResponse", &qstate->qinfo);
3611 	log_query_info(VERB_ALGO, "processTargetResponse super", &forq->qinfo);
3612 
3613 	/* Tell the originating event that this target query has finished
3614 	 * (regardless if it succeeded or not). */
3615 	foriq->num_target_queries--;
3616 
3617 	/* check to see if parent event is still interested (in orig name).  */
3618 	if(!foriq->dp) {
3619 		verbose(VERB_ALGO, "subq: parent not interested, was reset");
3620 		return; /* not interested anymore */
3621 	}
3622 	dpns = delegpt_find_ns(foriq->dp, qstate->qinfo.qname,
3623 			qstate->qinfo.qname_len);
3624 	if(!dpns) {
3625 		/* If not interested, just stop processing this event */
3626 		verbose(VERB_ALGO, "subq: parent not interested anymore");
3627 		/* could be because parent was jostled out of the cache,
3628 		   and a new identical query arrived, that does not want it*/
3629 		return;
3630 	}
3631 
3632 	/* if iq->query_for_pside_glue then add the pside_glue (marked lame) */
3633 	if(iq->pside_glue) {
3634 		/* if the pside_glue is NULL, then it could not be found,
3635 		 * the done_pside is already set when created and a cache
3636 		 * entry created in processFinished so nothing to do here */
3637 		log_rrset_key(VERB_ALGO, "add parentside glue to dp",
3638 			iq->pside_glue);
3639 		if(!delegpt_add_rrset(foriq->dp, forq->region,
3640 			iq->pside_glue, 1, NULL))
3641 			log_err("out of memory adding pside glue");
3642 	}
3643 
3644 	/* This response is relevant to the current query, so we
3645 	 * add (attempt to add, anyway) this target(s) and reactivate
3646 	 * the original event.
3647 	 * NOTE: we could only look for the AnswerRRset if the
3648 	 * response type was ANSWER. */
3649 	rrset = reply_find_answer_rrset(&iq->qchase, qstate->return_msg->rep);
3650 	if(rrset) {
3651 		int additions = 0;
3652 		/* if CNAMEs have been followed - add new NS to delegpt. */
3653 		/* BTW. RFC 1918 says NS should not have got CNAMEs. Robust. */
3654 		if(!delegpt_find_ns(foriq->dp, rrset->rk.dname,
3655 			rrset->rk.dname_len)) {
3656 			/* if dpns->lame then set newcname ns lame too */
3657 			if(!delegpt_add_ns(foriq->dp, forq->region,
3658 				rrset->rk.dname, dpns->lame, dpns->tls_auth_name,
3659 				dpns->port))
3660 				log_err("out of memory adding cnamed-ns");
3661 		}
3662 		/* if dpns->lame then set the address(es) lame too */
3663 		if(!delegpt_add_rrset(foriq->dp, forq->region, rrset,
3664 			dpns->lame, &additions))
3665 			log_err("out of memory adding targets");
3666 		if(!additions) {
3667 			/* no new addresses, increase the nxns counter, like
3668 			 * this could be a list of wildcards with no new
3669 			 * addresses */
3670 			target_count_increase_nx(foriq, 1);
3671 		}
3672 		verbose(VERB_ALGO, "added target response");
3673 		delegpt_log(VERB_ALGO, foriq->dp);
3674 	} else {
3675 		verbose(VERB_ALGO, "iterator TargetResponse failed");
3676 		delegpt_mark_neg(dpns, qstate->qinfo.qtype);
3677 		if((dpns->got4 == 2 || (!ie->supports_ipv4 && !ie->use_nat64)) &&
3678 			(dpns->got6 == 2 || !ie->supports_ipv6)) {
3679 			dpns->resolved = 1; /* fail the target */
3680 			/* do not count cached answers */
3681 			if(qstate->reply_origin && qstate->reply_origin->len != 0) {
3682 				target_count_increase_nx(foriq, 1);
3683 			}
3684 		}
3685 	}
3686 }
3687 
3688 /**
3689  * Process response for DS NS Find queries, that attempt to find the delegation
3690  * point where we ask the DS query from.
3691  *
3692  * @param qstate: query state.
3693  * @param id: module id.
3694  * @param forq: super query state.
3695  */
3696 static void
processDSNSResponse(struct module_qstate * qstate,int id,struct module_qstate * forq)3697 processDSNSResponse(struct module_qstate* qstate, int id,
3698 	struct module_qstate* forq)
3699 {
3700 	struct iter_qstate* foriq = (struct iter_qstate*)forq->minfo[id];
3701 
3702 	/* if the finished (iq->response) query has no NS set: continue
3703 	 * up to look for the right dp; nothing to change, do DPNSstate */
3704 	if(qstate->return_rcode != LDNS_RCODE_NOERROR)
3705 		return; /* seek further */
3706 	/* find the NS RRset (without allowing CNAMEs) */
3707 	if(!reply_find_rrset(qstate->return_msg->rep, qstate->qinfo.qname,
3708 		qstate->qinfo.qname_len, LDNS_RR_TYPE_NS,
3709 		qstate->qinfo.qclass)){
3710 		return; /* seek further */
3711 	}
3712 
3713 	/* else, store as DP and continue at querytargets */
3714 	foriq->state = QUERYTARGETS_STATE;
3715 	foriq->dp = delegpt_from_message(qstate->return_msg, forq->region);
3716 	if(!foriq->dp) {
3717 		log_err("out of memory in dsns dp alloc");
3718 		errinf(qstate, "malloc failure, in DS search");
3719 		return; /* dp==NULL in QUERYTARGETS makes SERVFAIL */
3720 	}
3721 	/* success, go query the querytargets in the new dp (and go down) */
3722 }
3723 
3724 /**
3725  * Process response for qclass=ANY queries for a particular class.
3726  * Append to result or error-exit.
3727  *
3728  * @param qstate: query state.
3729  * @param id: module id.
3730  * @param forq: super query state.
3731  */
3732 static void
processClassResponse(struct module_qstate * qstate,int id,struct module_qstate * forq)3733 processClassResponse(struct module_qstate* qstate, int id,
3734 	struct module_qstate* forq)
3735 {
3736 	struct iter_qstate* foriq = (struct iter_qstate*)forq->minfo[id];
3737 	struct dns_msg* from = qstate->return_msg;
3738 	log_query_info(VERB_ALGO, "processClassResponse", &qstate->qinfo);
3739 	log_query_info(VERB_ALGO, "processClassResponse super", &forq->qinfo);
3740 	if(qstate->return_rcode != LDNS_RCODE_NOERROR) {
3741 		/* cause servfail for qclass ANY query */
3742 		foriq->response = NULL;
3743 		foriq->state = FINISHED_STATE;
3744 		return;
3745 	}
3746 	/* append result */
3747 	if(!foriq->response) {
3748 		/* allocate the response: copy RCODE, sec_state */
3749 		foriq->response = dns_copy_msg(from, forq->region);
3750 		if(!foriq->response) {
3751 			log_err("malloc failed for qclass ANY response");
3752 			foriq->state = FINISHED_STATE;
3753 			return;
3754 		}
3755 		foriq->response->qinfo.qclass = forq->qinfo.qclass;
3756 		/* qclass ANY does not receive the AA flag on replies */
3757 		foriq->response->rep->authoritative = 0;
3758 	} else {
3759 		struct dns_msg* to = foriq->response;
3760 		/* add _from_ this response _to_ existing collection */
3761 		/* if there are records, copy RCODE */
3762 		/* lower sec_state if this message is lower */
3763 		if(from->rep->rrset_count != 0) {
3764 			size_t n = from->rep->rrset_count+to->rep->rrset_count;
3765 			struct ub_packed_rrset_key** dest, **d;
3766 			/* copy appropriate rcode */
3767 			to->rep->flags = from->rep->flags;
3768 			/* copy rrsets */
3769 			if(from->rep->rrset_count > RR_COUNT_MAX ||
3770 				to->rep->rrset_count > RR_COUNT_MAX) {
3771 				log_err("malloc failed (too many rrsets) in collect ANY");
3772 				foriq->state = FINISHED_STATE;
3773 				return; /* integer overflow protection */
3774 			}
3775 			dest = regional_alloc(forq->region, sizeof(dest[0])*n);
3776 			if(!dest) {
3777 				log_err("malloc failed in collect ANY");
3778 				foriq->state = FINISHED_STATE;
3779 				return;
3780 			}
3781 			d = dest;
3782 			/* copy AN */
3783 			memcpy(dest, to->rep->rrsets, to->rep->an_numrrsets
3784 				* sizeof(dest[0]));
3785 			dest += to->rep->an_numrrsets;
3786 			memcpy(dest, from->rep->rrsets, from->rep->an_numrrsets
3787 				* sizeof(dest[0]));
3788 			dest += from->rep->an_numrrsets;
3789 			/* copy NS */
3790 			memcpy(dest, to->rep->rrsets+to->rep->an_numrrsets,
3791 				to->rep->ns_numrrsets * sizeof(dest[0]));
3792 			dest += to->rep->ns_numrrsets;
3793 			memcpy(dest, from->rep->rrsets+from->rep->an_numrrsets,
3794 				from->rep->ns_numrrsets * sizeof(dest[0]));
3795 			dest += from->rep->ns_numrrsets;
3796 			/* copy AR */
3797 			memcpy(dest, to->rep->rrsets+to->rep->an_numrrsets+
3798 				to->rep->ns_numrrsets,
3799 				to->rep->ar_numrrsets * sizeof(dest[0]));
3800 			dest += to->rep->ar_numrrsets;
3801 			memcpy(dest, from->rep->rrsets+from->rep->an_numrrsets+
3802 				from->rep->ns_numrrsets,
3803 				from->rep->ar_numrrsets * sizeof(dest[0]));
3804 			/* update counts */
3805 			to->rep->rrsets = d;
3806 			to->rep->an_numrrsets += from->rep->an_numrrsets;
3807 			to->rep->ns_numrrsets += from->rep->ns_numrrsets;
3808 			to->rep->ar_numrrsets += from->rep->ar_numrrsets;
3809 			to->rep->rrset_count = n;
3810 		}
3811 		if(from->rep->security < to->rep->security) /* lowest sec */
3812 			to->rep->security = from->rep->security;
3813 		if(from->rep->qdcount != 0) /* insert qd if appropriate */
3814 			to->rep->qdcount = from->rep->qdcount;
3815 		if(from->rep->ttl < to->rep->ttl) /* use smallest TTL */
3816 			to->rep->ttl = from->rep->ttl;
3817 		if(from->rep->prefetch_ttl < to->rep->prefetch_ttl)
3818 			to->rep->prefetch_ttl = from->rep->prefetch_ttl;
3819 		if(from->rep->serve_expired_ttl < to->rep->serve_expired_ttl)
3820 			to->rep->serve_expired_ttl = from->rep->serve_expired_ttl;
3821 	}
3822 	/* are we done? */
3823 	foriq->num_current_queries --;
3824 	if(foriq->num_current_queries == 0)
3825 		foriq->state = FINISHED_STATE;
3826 }
3827 
3828 /**
3829  * Collect class ANY responses and make them into one response.  This
3830  * state is started and it creates queries for all classes (that have
3831  * root hints).  The answers are then collected.
3832  *
3833  * @param qstate: query state.
3834  * @param id: module id.
3835  * @return true if the event needs more immediate processing, false if not.
3836  */
3837 static int
processCollectClass(struct module_qstate * qstate,int id)3838 processCollectClass(struct module_qstate* qstate, int id)
3839 {
3840 	struct iter_qstate* iq = (struct iter_qstate*)qstate->minfo[id];
3841 	struct module_qstate* subq;
3842 	/* If qchase.qclass == 0 then send out queries for all classes.
3843 	 * Otherwise, do nothing (wait for all answers to arrive and the
3844 	 * processClassResponse to put them together, and that moves us
3845 	 * towards the Finished state when done. */
3846 	if(iq->qchase.qclass == 0) {
3847 		uint16_t c = 0;
3848 		iq->qchase.qclass = LDNS_RR_CLASS_ANY;
3849 		while(iter_get_next_root(qstate->env->hints,
3850 			qstate->env->fwds, &c)) {
3851 			/* generate query for this class */
3852 			log_nametypeclass(VERB_ALGO, "spawn collect query",
3853 				qstate->qinfo.qname, qstate->qinfo.qtype, c);
3854 			if(!generate_sub_request(qstate->qinfo.qname,
3855 				qstate->qinfo.qname_len, qstate->qinfo.qtype,
3856 				c, qstate, id, iq, INIT_REQUEST_STATE,
3857 				FINISHED_STATE, &subq,
3858 				(int)!(qstate->query_flags&BIT_CD), 0)) {
3859 				errinf(qstate, "could not generate class ANY"
3860 					" lookup query");
3861 				return error_response(qstate, id,
3862 					LDNS_RCODE_SERVFAIL);
3863 			}
3864 			/* ignore subq, no special init required */
3865 			iq->num_current_queries ++;
3866 			if(c == 0xffff)
3867 				break;
3868 			else c++;
3869 		}
3870 		/* if no roots are configured at all, return */
3871 		if(iq->num_current_queries == 0) {
3872 			verbose(VERB_ALGO, "No root hints or fwds, giving up "
3873 				"on qclass ANY");
3874 			return error_response(qstate, id, LDNS_RCODE_REFUSED);
3875 		}
3876 		/* return false, wait for queries to return */
3877 	}
3878 	/* if woke up here because of an answer, wait for more answers */
3879 	return 0;
3880 }
3881 
3882 /**
3883  * This handles the final state for first-tier responses (i.e., responses to
3884  * externally generated queries).
3885  *
3886  * @param qstate: query state.
3887  * @param iq: iterator query state.
3888  * @param id: module id.
3889  * @return true if the event needs more processing, false if not. Since this
3890  *         is the final state for an event, it always returns false.
3891  */
3892 static int
processFinished(struct module_qstate * qstate,struct iter_qstate * iq,int id)3893 processFinished(struct module_qstate* qstate, struct iter_qstate* iq,
3894 	int id)
3895 {
3896 	log_query_info(VERB_QUERY, "finishing processing for",
3897 		&qstate->qinfo);
3898 
3899 	/* store negative cache element for parent side glue. */
3900 	if(!qstate->no_cache_store && iq->query_for_pside_glue
3901 		&& !iq->pside_glue)
3902 			iter_store_parentside_neg(qstate->env, &qstate->qinfo,
3903 				iq->deleg_msg?iq->deleg_msg->rep:
3904 				(iq->response?iq->response->rep:NULL));
3905 	if(!iq->response) {
3906 		verbose(VERB_ALGO, "No response is set, servfail");
3907 		errinf(qstate, "(no response found at query finish)");
3908 		return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3909 	}
3910 
3911 	/* Make sure that the RA flag is set (since the presence of
3912 	 * this module means that recursion is available) */
3913 	iq->response->rep->flags |= BIT_RA;
3914 
3915 	/* Clear the AA flag */
3916 	/* FIXME: does this action go here or in some other module? */
3917 	iq->response->rep->flags &= ~BIT_AA;
3918 
3919 	/* make sure QR flag is on */
3920 	iq->response->rep->flags |= BIT_QR;
3921 
3922 	/* explicitly set the EDE string to NULL */
3923 	iq->response->rep->reason_bogus_str = NULL;
3924 	if((qstate->env->cfg->val_log_level >= 2 ||
3925 		qstate->env->cfg->log_servfail) && qstate->errinf &&
3926 		!qstate->env->cfg->val_log_squelch) {
3927 		char* err_str = errinf_to_str_misc(qstate);
3928 		if(err_str) {
3929 			size_t err_str_len = strlen(err_str);
3930 			verbose(VERB_ALGO, "iterator EDE: %s", err_str);
3931 			/* allocate space and store the error
3932 			 * string */
3933 			iq->response->rep->reason_bogus_str = regional_alloc(
3934 				qstate->region,
3935 				sizeof(char) * (err_str_len+1));
3936 			memcpy(iq->response->rep->reason_bogus_str,
3937 				err_str, err_str_len+1);
3938 		}
3939 		free(err_str);
3940 	}
3941 
3942 	/* we have finished processing this query */
3943 	qstate->ext_state[id] = module_finished;
3944 
3945 	/* TODO:  we are using a private TTL, trim the response. */
3946 	/* if (mPrivateTTL > 0){IterUtils.setPrivateTTL(resp, mPrivateTTL); } */
3947 
3948 	/* prepend any items we have accumulated */
3949 	if(iq->an_prepend_list || iq->ns_prepend_list) {
3950 		if(!iter_prepend(iq, iq->response, qstate->region)) {
3951 			log_err("prepend rrsets: out of memory");
3952 			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
3953 		}
3954 		/* reset the query name back */
3955 		iq->response->qinfo = qstate->qinfo;
3956 		/* the security state depends on the combination */
3957 		iq->response->rep->security = sec_status_unchecked;
3958 		/* store message with the finished prepended items,
3959 		 * but only if we did recursion. The nonrecursion referral
3960 		 * from cache does not need to be stored in the msg cache. */
3961 		if(!qstate->no_cache_store && qstate->query_flags&BIT_RD) {
3962 			iter_dns_store(qstate->env, &qstate->qinfo,
3963 				iq->response->rep, 0, qstate->prefetch_leeway,
3964 				iq->dp&&iq->dp->has_parent_side_NS,
3965 				qstate->region, qstate->query_flags,
3966 				qstate->qstarttime);
3967 		}
3968 	}
3969 	qstate->return_rcode = LDNS_RCODE_NOERROR;
3970 	qstate->return_msg = iq->response;
3971 	return 0;
3972 }
3973 
3974 /*
3975  * Return priming query results to interested super querystates.
3976  *
3977  * Sets the delegation point and delegation message (not nonRD queries).
3978  * This is a callback from walk_supers.
3979  *
3980  * @param qstate: query state that finished.
3981  * @param id: module id.
3982  * @param super: the qstate to inform.
3983  */
3984 void
iter_inform_super(struct module_qstate * qstate,int id,struct module_qstate * super)3985 iter_inform_super(struct module_qstate* qstate, int id,
3986 	struct module_qstate* super)
3987 {
3988 	if(!qstate->is_priming && super->qinfo.qclass == LDNS_RR_CLASS_ANY)
3989 		processClassResponse(qstate, id, super);
3990 	else if(super->qinfo.qtype == LDNS_RR_TYPE_DS && ((struct iter_qstate*)
3991 		super->minfo[id])->state == DSNS_FIND_STATE)
3992 		processDSNSResponse(qstate, id, super);
3993 	else if(qstate->return_rcode != LDNS_RCODE_NOERROR)
3994 		error_supers(qstate, id, super);
3995 	else if(qstate->is_priming)
3996 		prime_supers(qstate, id, super);
3997 	else	processTargetResponse(qstate, id, super);
3998 }
3999 
4000 /**
4001  * Handle iterator state.
4002  * Handle events. This is the real processing loop for events, responsible
4003  * for moving events through the various states. If a processing method
4004  * returns true, then it will be advanced to the next state. If false, then
4005  * processing will stop.
4006  *
4007  * @param qstate: query state.
4008  * @param ie: iterator shared global environment.
4009  * @param iq: iterator query state.
4010  * @param id: module id.
4011  */
4012 static void
iter_handle(struct module_qstate * qstate,struct iter_qstate * iq,struct iter_env * ie,int id)4013 iter_handle(struct module_qstate* qstate, struct iter_qstate* iq,
4014 	struct iter_env* ie, int id)
4015 {
4016 	int cont = 1;
4017 	while(cont) {
4018 		verbose(VERB_ALGO, "iter_handle processing q with state %s",
4019 			iter_state_to_string(iq->state));
4020 		switch(iq->state) {
4021 			case INIT_REQUEST_STATE:
4022 				cont = processInitRequest(qstate, iq, ie, id);
4023 				break;
4024 			case INIT_REQUEST_2_STATE:
4025 				cont = processInitRequest2(qstate, iq, id);
4026 				break;
4027 			case INIT_REQUEST_3_STATE:
4028 				cont = processInitRequest3(qstate, iq, id);
4029 				break;
4030 			case QUERYTARGETS_STATE:
4031 				cont = processQueryTargets(qstate, iq, ie, id);
4032 				break;
4033 			case QUERY_RESP_STATE:
4034 				cont = processQueryResponse(qstate, iq, ie, id);
4035 				break;
4036 			case PRIME_RESP_STATE:
4037 				cont = processPrimeResponse(qstate, id);
4038 				break;
4039 			case COLLECT_CLASS_STATE:
4040 				cont = processCollectClass(qstate, id);
4041 				break;
4042 			case DSNS_FIND_STATE:
4043 				cont = processDSNSFind(qstate, iq, id);
4044 				break;
4045 			case FINISHED_STATE:
4046 				cont = processFinished(qstate, iq, id);
4047 				break;
4048 			default:
4049 				log_warn("iterator: invalid state: %d",
4050 					iq->state);
4051 				cont = 0;
4052 				break;
4053 		}
4054 	}
4055 }
4056 
4057 /**
4058  * This is the primary entry point for processing request events. Note that
4059  * this method should only be used by external modules.
4060  * @param qstate: query state.
4061  * @param ie: iterator shared global environment.
4062  * @param iq: iterator query state.
4063  * @param id: module id.
4064  */
4065 static void
process_request(struct module_qstate * qstate,struct iter_qstate * iq,struct iter_env * ie,int id)4066 process_request(struct module_qstate* qstate, struct iter_qstate* iq,
4067 	struct iter_env* ie, int id)
4068 {
4069 	/* external requests start in the INIT state, and finish using the
4070 	 * FINISHED state. */
4071 	iq->state = INIT_REQUEST_STATE;
4072 	iq->final_state = FINISHED_STATE;
4073 	verbose(VERB_ALGO, "process_request: new external request event");
4074 	iter_handle(qstate, iq, ie, id);
4075 }
4076 
4077 /** process authoritative server reply */
4078 static void
process_response(struct module_qstate * qstate,struct iter_qstate * iq,struct iter_env * ie,int id,struct outbound_entry * outbound,enum module_ev event)4079 process_response(struct module_qstate* qstate, struct iter_qstate* iq,
4080 	struct iter_env* ie, int id, struct outbound_entry* outbound,
4081 	enum module_ev event)
4082 {
4083 	struct msg_parse* prs;
4084 	struct edns_data edns;
4085 	sldns_buffer* pkt;
4086 
4087 	verbose(VERB_ALGO, "process_response: new external response event");
4088 	iq->response = NULL;
4089 	iq->state = QUERY_RESP_STATE;
4090 	if(event == module_event_noreply || event == module_event_error) {
4091 		if(event == module_event_noreply && iq->timeout_count >= 3 &&
4092 			qstate->env->cfg->use_caps_bits_for_id &&
4093 			!iq->caps_fallback && !is_caps_whitelisted(ie, iq)) {
4094 			/* start fallback */
4095 			iq->caps_fallback = 1;
4096 			iq->caps_server = 0;
4097 			iq->caps_reply = NULL;
4098 			iq->caps_response = NULL;
4099 			iq->caps_minimisation_state = DONOT_MINIMISE_STATE;
4100 			iq->state = QUERYTARGETS_STATE;
4101 			iq->num_current_queries--;
4102 			/* need fresh attempts for the 0x20 fallback, if
4103 			 * that was the cause for the failure */
4104 			iter_dec_attempts(iq->dp, 3, ie->outbound_msg_retry);
4105 			verbose(VERB_DETAIL, "Capsforid: timeouts, starting fallback");
4106 			goto handle_it;
4107 		}
4108 		goto handle_it;
4109 	}
4110 	if( (event != module_event_reply && event != module_event_capsfail)
4111 		|| !qstate->reply) {
4112 		log_err("Bad event combined with response");
4113 		outbound_list_remove(&iq->outlist, outbound);
4114 		errinf(qstate, "module iterator received wrong internal event with a response message");
4115 		(void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
4116 		return;
4117 	}
4118 
4119 	/* parse message */
4120 	fill_fail_addr(iq, &qstate->reply->remote_addr,
4121 		qstate->reply->remote_addrlen);
4122 	prs = (struct msg_parse*)regional_alloc(qstate->env->scratch,
4123 		sizeof(struct msg_parse));
4124 	if(!prs) {
4125 		log_err("out of memory on incoming message");
4126 		/* like packet got dropped */
4127 		goto handle_it;
4128 	}
4129 	memset(prs, 0, sizeof(*prs));
4130 	memset(&edns, 0, sizeof(edns));
4131 	pkt = qstate->reply->c->buffer;
4132 	sldns_buffer_set_position(pkt, 0);
4133 	if(parse_packet(pkt, prs, qstate->env->scratch) != LDNS_RCODE_NOERROR) {
4134 		verbose(VERB_ALGO, "parse error on reply packet");
4135 		iq->parse_failures++;
4136 		goto handle_it;
4137 	}
4138 	/* edns is not examined, but removed from message to help cache */
4139 	if(parse_extract_edns_from_response_msg(prs, &edns, qstate->env->scratch) !=
4140 		LDNS_RCODE_NOERROR) {
4141 		iq->parse_failures++;
4142 		goto handle_it;
4143 	}
4144 
4145 	/* Copy the edns options we may got from the back end */
4146 	if(edns.opt_list_in) {
4147 		qstate->edns_opts_back_in = edns_opt_copy_region(edns.opt_list_in,
4148 			qstate->region);
4149 		if(!qstate->edns_opts_back_in) {
4150 			log_err("out of memory on incoming message");
4151 			/* like packet got dropped */
4152 			goto handle_it;
4153 		}
4154 	}
4155 	if(!inplace_cb_edns_back_parsed_call(qstate->env, qstate)) {
4156 		log_err("unable to call edns_back_parsed callback");
4157 		goto handle_it;
4158 	}
4159 
4160 	/* remove CD-bit, we asked for in case we handle validation ourself */
4161 	prs->flags &= ~BIT_CD;
4162 
4163 	/* normalize and sanitize: easy to delete items from linked lists */
4164 	if(!scrub_message(pkt, prs, &iq->qinfo_out, iq->dp->name,
4165 		qstate->env->scratch, qstate->env, qstate, ie)) {
4166 		/* if 0x20 enabled, start fallback, but we have no message */
4167 		if(event == module_event_capsfail && !iq->caps_fallback) {
4168 			iq->caps_fallback = 1;
4169 			iq->caps_server = 0;
4170 			iq->caps_reply = NULL;
4171 			iq->caps_response = NULL;
4172 			iq->caps_minimisation_state = DONOT_MINIMISE_STATE;
4173 			iq->state = QUERYTARGETS_STATE;
4174 			iq->num_current_queries--;
4175 			verbose(VERB_DETAIL, "Capsforid: scrub failed, starting fallback with no response");
4176 		}
4177 		iq->scrub_failures++;
4178 		goto handle_it;
4179 	}
4180 
4181 	/* allocate response dns_msg in region */
4182 	iq->response = dns_alloc_msg(pkt, prs, qstate->region);
4183 	if(!iq->response)
4184 		goto handle_it;
4185 	log_query_info(VERB_DETAIL, "response for", &qstate->qinfo);
4186 	log_name_addr(VERB_DETAIL, "reply from", iq->dp->name,
4187 		&qstate->reply->remote_addr, qstate->reply->remote_addrlen);
4188 	if(verbosity >= VERB_ALGO)
4189 		log_dns_msg("incoming scrubbed packet:", &iq->response->qinfo,
4190 			iq->response->rep);
4191 
4192 	if(event == module_event_capsfail || iq->caps_fallback) {
4193 		if(qstate->env->cfg->qname_minimisation &&
4194 			iq->minimisation_state != DONOT_MINIMISE_STATE) {
4195 			/* Skip QNAME minimisation for next query, since that
4196 			 * one has to match the current query. */
4197 			iq->minimisation_state = SKIP_MINIMISE_STATE;
4198 		}
4199 		/* for fallback we care about main answer, not additionals */
4200 		/* removing that makes comparison more likely to succeed */
4201 		caps_strip_reply(iq->response->rep);
4202 
4203 		if(iq->caps_fallback &&
4204 			iq->caps_minimisation_state != iq->minimisation_state) {
4205 			/* QNAME minimisation state has changed, restart caps
4206 			 * fallback. */
4207 			iq->caps_fallback = 0;
4208 		}
4209 
4210 		if(!iq->caps_fallback) {
4211 			/* start fallback */
4212 			iq->caps_fallback = 1;
4213 			iq->caps_server = 0;
4214 			iq->caps_reply = iq->response->rep;
4215 			iq->caps_response = iq->response;
4216 			iq->caps_minimisation_state = iq->minimisation_state;
4217 			iq->state = QUERYTARGETS_STATE;
4218 			iq->num_current_queries--;
4219 			verbose(VERB_DETAIL, "Capsforid: starting fallback");
4220 			goto handle_it;
4221 		} else {
4222 			/* check if reply is the same, otherwise, fail */
4223 			if(!iq->caps_reply) {
4224 				iq->caps_reply = iq->response->rep;
4225 				iq->caps_response = iq->response;
4226 				iq->caps_server = -1; /*become zero at ++,
4227 				so that we start the full set of trials */
4228 			} else if(caps_failed_rcode(iq->caps_reply) &&
4229 				!caps_failed_rcode(iq->response->rep)) {
4230 				/* prefer to upgrade to non-SERVFAIL */
4231 				iq->caps_reply = iq->response->rep;
4232 				iq->caps_response = iq->response;
4233 			} else if(!caps_failed_rcode(iq->caps_reply) &&
4234 				caps_failed_rcode(iq->response->rep)) {
4235 				/* if we have non-SERVFAIL as answer then
4236 				 * we can ignore SERVFAILs for the equality
4237 				 * comparison */
4238 				/* no instructions here, skip other else */
4239 			} else if(caps_failed_rcode(iq->caps_reply) &&
4240 				caps_failed_rcode(iq->response->rep)) {
4241 				/* failure is same as other failure in fallbk*/
4242 				/* no instructions here, skip other else */
4243 			} else if(!reply_equal(iq->response->rep, iq->caps_reply,
4244 				qstate->env->scratch)) {
4245 				verbose(VERB_DETAIL, "Capsforid fallback: "
4246 					"getting different replies, failed");
4247 				outbound_list_remove(&iq->outlist, outbound);
4248 				errinf(qstate, "0x20 failed, then got different replies in fallback");
4249 				(void)error_response(qstate, id,
4250 					LDNS_RCODE_SERVFAIL);
4251 				return;
4252 			}
4253 			/* continue the fallback procedure at next server */
4254 			iq->caps_server++;
4255 			iq->state = QUERYTARGETS_STATE;
4256 			iq->num_current_queries--;
4257 			verbose(VERB_DETAIL, "Capsforid: reply is equal. "
4258 				"go to next fallback");
4259 			goto handle_it;
4260 		}
4261 	}
4262 	iq->caps_fallback = 0; /* if we were in fallback, 0x20 is OK now */
4263 
4264 handle_it:
4265 	outbound_list_remove(&iq->outlist, outbound);
4266 	iter_handle(qstate, iq, ie, id);
4267 }
4268 
4269 void
iter_operate(struct module_qstate * qstate,enum module_ev event,int id,struct outbound_entry * outbound)4270 iter_operate(struct module_qstate* qstate, enum module_ev event, int id,
4271 	struct outbound_entry* outbound)
4272 {
4273 	struct iter_env* ie = (struct iter_env*)qstate->env->modinfo[id];
4274 	struct iter_qstate* iq = (struct iter_qstate*)qstate->minfo[id];
4275 	verbose(VERB_QUERY, "iterator[module %d] operate: extstate:%s event:%s",
4276 		id, strextstate(qstate->ext_state[id]), strmodulevent(event));
4277 	if(iq) log_query_info(VERB_QUERY, "iterator operate: query",
4278 		&qstate->qinfo);
4279 	if(iq && qstate->qinfo.qname != iq->qchase.qname)
4280 		log_query_info(VERB_QUERY, "iterator operate: chased to",
4281 			&iq->qchase);
4282 
4283 	/* perform iterator state machine */
4284 	if((event == module_event_new || event == module_event_pass) &&
4285 		iq == NULL) {
4286 		if(!iter_new(qstate, id)) {
4287 			errinf(qstate, "malloc failure, new iterator module allocation");
4288 			(void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
4289 			return;
4290 		}
4291 		iq = (struct iter_qstate*)qstate->minfo[id];
4292 		process_request(qstate, iq, ie, id);
4293 		return;
4294 	}
4295 	if(iq && event == module_event_pass) {
4296 		iter_handle(qstate, iq, ie, id);
4297 		return;
4298 	}
4299 	if(iq && outbound) {
4300 		process_response(qstate, iq, ie, id, outbound, event);
4301 		return;
4302 	}
4303 	if(event == module_event_error) {
4304 		verbose(VERB_ALGO, "got called with event error, giving up");
4305 		errinf(qstate, "iterator module got the error event");
4306 		(void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
4307 		return;
4308 	}
4309 
4310 	log_err("bad event for iterator");
4311 	errinf(qstate, "iterator module received wrong event");
4312 	(void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
4313 }
4314 
4315 void
iter_clear(struct module_qstate * qstate,int id)4316 iter_clear(struct module_qstate* qstate, int id)
4317 {
4318 	struct iter_qstate* iq;
4319 	if(!qstate)
4320 		return;
4321 	iq = (struct iter_qstate*)qstate->minfo[id];
4322 	if(iq) {
4323 		outbound_list_clear(&iq->outlist);
4324 		if(iq->target_count && --iq->target_count[TARGET_COUNT_REF] == 0) {
4325 			free(iq->target_count);
4326 			if(*iq->nxns_dp) free(*iq->nxns_dp);
4327 			free(iq->nxns_dp);
4328 		}
4329 		iq->num_current_queries = 0;
4330 	}
4331 	qstate->minfo[id] = NULL;
4332 }
4333 
4334 size_t
iter_get_mem(struct module_env * env,int id)4335 iter_get_mem(struct module_env* env, int id)
4336 {
4337 	struct iter_env* ie = (struct iter_env*)env->modinfo[id];
4338 	if(!ie)
4339 		return 0;
4340 	return sizeof(*ie) + sizeof(int)*((size_t)ie->max_dependency_depth+1)
4341 		+ donotq_get_mem(ie->donotq) + priv_get_mem(ie->priv);
4342 }
4343 
4344 /**
4345  * The iterator function block
4346  */
4347 static struct module_func_block iter_block = {
4348 	"iterator",
4349 	&iter_init, &iter_deinit, &iter_operate, &iter_inform_super,
4350 	&iter_clear, &iter_get_mem
4351 };
4352 
4353 struct module_func_block*
iter_get_funcblock(void)4354 iter_get_funcblock(void)
4355 {
4356 	return &iter_block;
4357 }
4358 
4359 const char*
iter_state_to_string(enum iter_state state)4360 iter_state_to_string(enum iter_state state)
4361 {
4362 	switch (state)
4363 	{
4364 	case INIT_REQUEST_STATE :
4365 		return "INIT REQUEST STATE";
4366 	case INIT_REQUEST_2_STATE :
4367 		return "INIT REQUEST STATE (stage 2)";
4368 	case INIT_REQUEST_3_STATE:
4369 		return "INIT REQUEST STATE (stage 3)";
4370 	case QUERYTARGETS_STATE :
4371 		return "QUERY TARGETS STATE";
4372 	case PRIME_RESP_STATE :
4373 		return "PRIME RESPONSE STATE";
4374 	case COLLECT_CLASS_STATE :
4375 		return "COLLECT CLASS STATE";
4376 	case DSNS_FIND_STATE :
4377 		return "DSNS FIND STATE";
4378 	case QUERY_RESP_STATE :
4379 		return "QUERY RESPONSE STATE";
4380 	case FINISHED_STATE :
4381 		return "FINISHED RESPONSE STATE";
4382 	default :
4383 		return "UNKNOWN ITER STATE";
4384 	}
4385 }
4386 
4387 int
iter_state_is_responsestate(enum iter_state s)4388 iter_state_is_responsestate(enum iter_state s)
4389 {
4390 	switch(s) {
4391 		case INIT_REQUEST_STATE :
4392 		case INIT_REQUEST_2_STATE :
4393 		case INIT_REQUEST_3_STATE :
4394 		case QUERYTARGETS_STATE :
4395 		case COLLECT_CLASS_STATE :
4396 			return 0;
4397 		default:
4398 			break;
4399 	}
4400 	return 1;
4401 }
4402