xref: /freebsd/contrib/unbound/services/cache/dns.c (revision 7a789145f88a6aceacc59029a0cafe7de7aeefea)
1 /*
2  * services/cache/dns.c - Cache services for DNS using msg and rrset caches.
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 the DNS cache.
40  */
41 #include "config.h"
42 #include "iterator/iter_delegpt.h"
43 #include "iterator/iter_utils.h"
44 #include "validator/val_nsec.h"
45 #include "validator/val_utils.h"
46 #include "iterator/iter_utils.h"
47 #include "services/cache/dns.h"
48 #include "services/cache/rrset.h"
49 #include "util/data/msgparse.h"
50 #include "util/data/msgreply.h"
51 #include "util/data/packed_rrset.h"
52 #include "util/data/dname.h"
53 #include "util/module.h"
54 #include "util/net_help.h"
55 #include "util/regional.h"
56 #include "util/config_file.h"
57 #include "sldns/sbuffer.h"
58 
59 /** store rrsets in the rrset cache.
60  * @param env: module environment with caches.
61  * @param rep: contains list of rrsets to store.
62  * @param now: current time.
63  * @param leeway: during prefetch how much leeway to update TTLs.
64  * 	This makes rrsets expire sooner so they get updated with a new full
65  * 	TTL.
66  * 	Child side type NS does get this but TTL checks are done using the time
67  * 	the query was created rather than the time the answer was received.
68  * @param pside: if from parentside discovered NS, so that its NS is okay
69  * 	in a prefetch situation to be updated (without becoming sticky).
70  * @param qrep: update rrsets here if cache is better
71  * @param region: for qrep allocs.
72  * @param qstarttime: time when delegations were looked up, this is perhaps
73  *	earlier than the time in now. The time is used to determine if RRsets
74  *	of type NS have expired, so that they can only be updated using
75  *	lookups of delegation points that did not use them, since they had
76  *	expired then.
77  */
78 static void
store_rrsets(struct module_env * env,struct reply_info * rep,time_t now,time_t leeway,int pside,struct reply_info * qrep,struct regional * region,time_t qstarttime)79 store_rrsets(struct module_env* env, struct reply_info* rep, time_t now,
80 	time_t leeway, int pside, struct reply_info* qrep,
81 	struct regional* region, time_t qstarttime)
82 {
83 	size_t i;
84 	time_t ttl, min_ttl = rep->ttl;
85 	/* see if rrset already exists in cache, if not insert it. */
86 	for(i=0; i<rep->rrset_count; i++) {
87 		rep->ref[i].key = rep->rrsets[i];
88 		rep->ref[i].id = rep->rrsets[i]->id;
89 		/* update ref if it was in the cache */
90 		switch(rrset_cache_update(env->rrset_cache, &rep->ref[i],
91 				env->alloc, ((ntohs(rep->ref[i].key->rk.type)==
92 				LDNS_RR_TYPE_NS && !pside)?qstarttime:now) + leeway)) {
93 		case 0: /* ref unchanged, item inserted */
94 			break;
95 		case 2: /* ref updated, cache is superior */
96 			if(region) {
97 				struct ub_packed_rrset_key* ck;
98 				lock_rw_rdlock(&rep->ref[i].key->entry.lock);
99 				/* if deleted rrset, do not copy it */
100 				if(rep->ref[i].key->id == 0 ||
101 					rep->ref[i].id != rep->ref[i].key->id)
102 					ck = NULL;
103 				else 	ck = packed_rrset_copy_region(
104 					rep->ref[i].key, region,
105 					((ntohs(rep->ref[i].key->rk.type)==
106 					LDNS_RR_TYPE_NS && !pside)?qstarttime:now));
107 				lock_rw_unlock(&rep->ref[i].key->entry.lock);
108 				if(ck) {
109 					/* use cached copy if memory allows */
110 					qrep->rrsets[i] = ck;
111 					ttl = ((struct packed_rrset_data*)
112 					    ck->entry.data)->ttl;
113 					if(ttl < qrep->ttl) {
114 						qrep->ttl = ttl;
115 						qrep->prefetch_ttl = PREFETCH_TTL_CALC(qrep->ttl);
116 						qrep->serve_expired_ttl = qrep->ttl + SERVE_EXPIRED_TTL;
117 					}
118 				}
119 			}
120 			/* no break: also copy key item */
121 			/* the line below is matched by gcc regex and silences
122 			 * the fallthrough warning */
123 			ATTR_FALLTHROUGH
124 			/* fallthrough */
125 		case 1: /* ref updated, item inserted */
126 			rep->rrsets[i] = rep->ref[i].key;
127 			/* ref was updated; make sure the message ttl is
128 			 * updated to the minimum of the current rrsets. */
129 			lock_rw_rdlock(&rep->ref[i].key->entry.lock);
130 			/* if deleted, skip ttl update. */
131 			if(rep->ref[i].key->id != 0 &&
132 				rep->ref[i].id == rep->ref[i].key->id) {
133 				ttl = ((struct packed_rrset_data*)
134 				    rep->rrsets[i]->entry.data)->ttl;
135 				if(ttl < min_ttl) min_ttl = ttl;
136 			}
137 			lock_rw_unlock(&rep->ref[i].key->entry.lock);
138 		}
139 	}
140 	if(min_ttl < rep->ttl) {
141 		rep->ttl = min_ttl;
142 		rep->prefetch_ttl = PREFETCH_TTL_CALC(rep->ttl);
143 		rep->serve_expired_ttl = rep->ttl + SERVE_EXPIRED_TTL;
144 	}
145 }
146 
147 /** delete message from message cache */
148 void
msg_cache_remove(struct module_env * env,uint8_t * qname,size_t qnamelen,uint16_t qtype,uint16_t qclass,uint16_t flags)149 msg_cache_remove(struct module_env* env, uint8_t* qname, size_t qnamelen,
150 	uint16_t qtype, uint16_t qclass, uint16_t flags)
151 {
152 	struct query_info k;
153 	hashvalue_type h;
154 
155 	k.qname = qname;
156 	k.qname_len = qnamelen;
157 	k.qtype = qtype;
158 	k.qclass = qclass;
159 	k.local_alias = NULL;
160 	h = query_info_hash(&k, flags);
161 	slabhash_remove(env->msg_cache, h, &k);
162 }
163 
164 void
dns_cache_store_msg(struct module_env * env,struct query_info * qinfo,hashvalue_type hash,struct reply_info * rep,time_t leeway,int pside,struct reply_info * qrep,uint32_t flags,struct regional * region,time_t qstarttime)165 dns_cache_store_msg(struct module_env* env, struct query_info* qinfo,
166 	hashvalue_type hash, struct reply_info* rep, time_t leeway, int pside,
167 	struct reply_info* qrep, uint32_t flags, struct regional* region,
168 	time_t qstarttime)
169 {
170 	struct msgreply_entry* e;
171 	time_t ttl = rep->ttl;
172 	size_t i;
173 
174 	/* store RRsets */
175 	for(i=0; i<rep->rrset_count; i++) {
176 		rep->ref[i].key = rep->rrsets[i];
177 		rep->ref[i].id = rep->rrsets[i]->id;
178 	}
179 
180 	/* there was a reply_info_sortref(rep) here but it seems to be
181 	 * unnecessary, because the cache gets locked per rrset. */
182 	if((flags & DNSCACHE_STORE_EXPIRED_MSG_CACHEDB)) {
183 		reply_info_absolute_ttls(rep, *env->now, *env->now - ttl);
184 	} else	reply_info_set_ttls(rep, *env->now);
185 	store_rrsets(env, rep, *env->now, leeway, pside, qrep, region,
186 		qstarttime);
187 	if(ttl == 0) {
188 		/* we do not store the message, but we did store the RRs,
189 		 * which could be useful for delegation information */
190 		verbose(VERB_ALGO, "TTL 0: dropped msg from cache");
191 		reply_info_delete(rep, NULL);
192 		/* if the message is in the cache, remove that msg,
193 		 * so that the TTL 0 response can be returned for future
194 		 * responses (i.e. don't get answered from
195 		 * cache, but instead go to recursion to get this TTL0
196 		 * response).
197 		 * Possible messages that could be in the cache:
198 		 * - SERVFAIL
199 		 * - NXDOMAIN
200 		 * - NODATA
201 		 * - an older record that is expired
202 		 * - an older record that did not yet expire */
203 		msg_cache_remove(env, qinfo->qname, qinfo->qname_len,
204 			qinfo->qtype, qinfo->qclass, flags);
205 		return;
206 	}
207 
208 	/* store msg in the cache */
209 	reply_info_sortref(rep);
210 	if(!(e = query_info_entrysetup(qinfo, rep, hash))) {
211 		log_err("store_msg: malloc failed");
212 		reply_info_delete(rep, NULL);
213 		return;
214 	}
215 	slabhash_insert(env->msg_cache, hash, &e->entry, rep, env->alloc);
216 }
217 
218 /** find closest NS or DNAME and returns the rrset (locked) */
219 static struct ub_packed_rrset_key*
find_closest_of_type(struct module_env * env,uint8_t * qname,size_t qnamelen,uint16_t qclass,time_t now,uint16_t searchtype,int stripfront,int noexpiredabove,uint8_t * expiretop,size_t expiretoplen)220 find_closest_of_type(struct module_env* env, uint8_t* qname, size_t qnamelen,
221 	uint16_t qclass, time_t now, uint16_t searchtype, int stripfront,
222 	int noexpiredabove, uint8_t* expiretop, size_t expiretoplen)
223 {
224 	struct ub_packed_rrset_key *rrset;
225 	uint8_t lablen;
226 
227 	if(stripfront) {
228 		/* strip off so that DNAMEs have strict subdomain match */
229 		lablen = *qname;
230 		qname += lablen + 1;
231 		qnamelen -= lablen + 1;
232 	}
233 
234 	/* snip off front part of qname until the type is found */
235 	while(qnamelen > 0) {
236 		rrset = rrset_cache_lookup(env->rrset_cache, qname,
237 			qnamelen, searchtype, qclass, 0, now, 0);
238 		if(!rrset && searchtype == LDNS_RR_TYPE_DNAME)
239 			/* If not found, for type DNAME, try 0TTL stored,
240 			 * for its grace period. */
241 			rrset = rrset_cache_lookup(env->rrset_cache, qname,
242 				qnamelen, searchtype, qclass,
243 				PACKED_RRSET_UPSTREAM_0TTL, now, 0);
244 		if(rrset) {
245 			uint8_t* origqname = qname;
246 			size_t origqnamelen = qnamelen;
247 			if(!noexpiredabove)
248 				return rrset;
249 			/* if expiretop set, do not look above it, but
250 			 * qname is equal, so the just found result is also
251 			 * the nonexpired above part. */
252 			if(expiretop && qnamelen == expiretoplen &&
253 				query_dname_compare(qname, expiretop)==0)
254 				return rrset;
255 			/* check for expiry, but we have to let go of the rrset
256 			 * for the lock ordering */
257 			lock_rw_unlock(&rrset->entry.lock);
258 			/* the rrset_cache_expired_above function always takes
259 			 * off one label (if qnamelen>0) and returns the final
260 			 * qname where it searched, so we can continue from
261 			 * there turning the O N*N search into O N. */
262 			if(!rrset_cache_expired_above(env->rrset_cache, &qname,
263 				&qnamelen, searchtype, qclass, now, expiretop,
264 				expiretoplen)) {
265 				/* we want to return rrset, but it may be
266 				 * gone from cache, if so, just loop like
267 				 * it was not in the cache in the first place.
268 				 */
269 				if((rrset = rrset_cache_lookup(env->
270 					rrset_cache, origqname, origqnamelen,
271 					searchtype, qclass, 0, now, 0))) {
272 					return rrset;
273 				}
274 			}
275 			log_nametypeclass(VERB_ALGO, "ignoring rrset because expired rrsets exist above it", origqname, searchtype, qclass);
276 			continue;
277 		}
278 
279 		/* snip off front label */
280 		lablen = *qname;
281 		if(lablen == 0)
282 			break;
283 		qname += lablen + 1;
284 		qnamelen -= lablen + 1;
285 	}
286 	return NULL;
287 }
288 
289 /** add addr to additional section */
290 static void
addr_to_additional(struct ub_packed_rrset_key * rrset,struct regional * region,struct dns_msg * msg,time_t now)291 addr_to_additional(struct ub_packed_rrset_key* rrset, struct regional* region,
292 	struct dns_msg* msg, time_t now)
293 {
294 	if((msg->rep->rrsets[msg->rep->rrset_count] =
295 		packed_rrset_copy_region(rrset, region, now))) {
296 		struct packed_rrset_data* d = rrset->entry.data;
297 		msg->rep->ar_numrrsets++;
298 		msg->rep->rrset_count++;
299 		UPDATE_TTL_FROM_RRSET(msg->rep->ttl, d->ttl);
300 	}
301 }
302 
303 /** lookup message in message cache */
304 struct msgreply_entry*
msg_cache_lookup(struct module_env * env,uint8_t * qname,size_t qnamelen,uint16_t qtype,uint16_t qclass,uint16_t flags,time_t now,int wr)305 msg_cache_lookup(struct module_env* env, uint8_t* qname, size_t qnamelen,
306 	uint16_t qtype, uint16_t qclass, uint16_t flags, time_t now, int wr)
307 {
308 	struct lruhash_entry* e;
309 	struct query_info k;
310 	hashvalue_type h;
311 
312 	k.qname = qname;
313 	k.qname_len = qnamelen;
314 	k.qtype = qtype;
315 	k.qclass = qclass;
316 	k.local_alias = NULL;
317 	h = query_info_hash(&k, flags);
318 	e = slabhash_lookup(env->msg_cache, h, &k, wr);
319 
320 	if(!e) return NULL;
321 	if( now > ((struct reply_info*)e->data)->ttl ) {
322 		lock_rw_unlock(&e->lock);
323 		return NULL;
324 	}
325 	return (struct msgreply_entry*)e->key;
326 }
327 
328 /** find and add A and AAAA records for nameservers in delegpt */
329 static int
find_add_addrs(struct module_env * env,uint16_t qclass,struct regional * region,struct delegpt * dp,time_t now,struct dns_msg ** msg)330 find_add_addrs(struct module_env* env, uint16_t qclass,
331 	struct regional* region, struct delegpt* dp, time_t now,
332 	struct dns_msg** msg)
333 {
334 	struct delegpt_ns* ns;
335 	struct msgreply_entry* neg;
336 	struct ub_packed_rrset_key* akey;
337 	for(ns = dp->nslist; ns; ns = ns->next) {
338 		akey = rrset_cache_lookup(env->rrset_cache, ns->name,
339 			ns->namelen, LDNS_RR_TYPE_A, qclass, 0, now, 0);
340 		if(akey) {
341 			if(!delegpt_add_rrset_A(dp, region, akey, 0, NULL)) {
342 				lock_rw_unlock(&akey->entry.lock);
343 				return 0;
344 			}
345 			if(msg)
346 				addr_to_additional(akey, region, *msg, now);
347 			lock_rw_unlock(&akey->entry.lock);
348 		} else {
349 			/* BIT_CD on false because delegpt lookup does
350 			 * not use dns64 translation */
351 			neg = msg_cache_lookup(env, ns->name, ns->namelen,
352 				LDNS_RR_TYPE_A, qclass, 0, now, 0);
353 			if(neg) {
354 				delegpt_add_neg_msg(dp, neg);
355 				lock_rw_unlock(&neg->entry.lock);
356 			}
357 		}
358 		akey = rrset_cache_lookup(env->rrset_cache, ns->name,
359 			ns->namelen, LDNS_RR_TYPE_AAAA, qclass, 0, now, 0);
360 		if(akey) {
361 			if(!delegpt_add_rrset_AAAA(dp, region, akey, 0, NULL)) {
362 				lock_rw_unlock(&akey->entry.lock);
363 				return 0;
364 			}
365 			if(msg)
366 				addr_to_additional(akey, region, *msg, now);
367 			lock_rw_unlock(&akey->entry.lock);
368 		} else {
369 			/* BIT_CD on false because delegpt lookup does
370 			 * not use dns64 translation */
371 			neg = msg_cache_lookup(env, ns->name, ns->namelen,
372 				LDNS_RR_TYPE_AAAA, qclass, 0, now, 0);
373 			/* Because recursion for lookup uses BIT_CD, check
374 			 * for that so it stops the recursion lookup, if a
375 			 * negative answer is cached. Because the cache uses
376 			 * the CD flag for type AAAA. */
377 			if(!neg)
378 				neg = msg_cache_lookup(env, ns->name, ns->namelen,
379 					LDNS_RR_TYPE_AAAA, qclass, BIT_CD, now, 0);
380 			if(neg) {
381 				delegpt_add_neg_msg(dp, neg);
382 				lock_rw_unlock(&neg->entry.lock);
383 			}
384 		}
385 	}
386 	return 1;
387 }
388 
389 /** find and add A and AAAA records for missing nameservers in delegpt */
390 int
cache_fill_missing(struct module_env * env,uint16_t qclass,struct regional * region,struct delegpt * dp,uint32_t flags)391 cache_fill_missing(struct module_env* env, uint16_t qclass,
392 	struct regional* region, struct delegpt* dp, uint32_t flags)
393 {
394 	struct delegpt_ns* ns;
395 	struct msgreply_entry* neg;
396 	struct ub_packed_rrset_key* akey;
397 	time_t now = *env->now;
398 	for(ns = dp->nslist; ns; ns = ns->next) {
399 		if(ns->cache_lookup_count > ITERATOR_NAME_CACHELOOKUP_MAX)
400 			continue;
401 		ns->cache_lookup_count++;
402 		akey = rrset_cache_lookup(env->rrset_cache, ns->name,
403 			ns->namelen, LDNS_RR_TYPE_A, qclass, flags, now, 0);
404 		if(akey) {
405 			if(!delegpt_add_rrset_A(dp, region, akey, ns->lame,
406 				NULL)) {
407 				lock_rw_unlock(&akey->entry.lock);
408 				return 0;
409 			}
410 			log_nametypeclass(VERB_ALGO, "found in cache",
411 				ns->name, LDNS_RR_TYPE_A, qclass);
412 			lock_rw_unlock(&akey->entry.lock);
413 		} else {
414 			/* BIT_CD on false because delegpt lookup does
415 			 * not use dns64 translation */
416 			neg = msg_cache_lookup(env, ns->name, ns->namelen,
417 				LDNS_RR_TYPE_A, qclass, 0, now, 0);
418 			if(neg) {
419 				delegpt_add_neg_msg(dp, neg);
420 				lock_rw_unlock(&neg->entry.lock);
421 			}
422 		}
423 		akey = rrset_cache_lookup(env->rrset_cache, ns->name,
424 			ns->namelen, LDNS_RR_TYPE_AAAA, qclass, flags, now, 0);
425 		if(akey) {
426 			if(!delegpt_add_rrset_AAAA(dp, region, akey, ns->lame,
427 				NULL)) {
428 				lock_rw_unlock(&akey->entry.lock);
429 				return 0;
430 			}
431 			log_nametypeclass(VERB_ALGO, "found in cache",
432 				ns->name, LDNS_RR_TYPE_AAAA, qclass);
433 			lock_rw_unlock(&akey->entry.lock);
434 		} else {
435 			/* BIT_CD on false because delegpt lookup does
436 			 * not use dns64 translation */
437 			neg = msg_cache_lookup(env, ns->name, ns->namelen,
438 				LDNS_RR_TYPE_AAAA, qclass, 0, now, 0);
439 			/* Because recursion for lookup uses BIT_CD, check
440 			 * for that so it stops the recursion lookup, if a
441 			 * negative answer is cached. Because the cache uses
442 			 * the CD flag for type AAAA. */
443 			if(!neg)
444 				neg = msg_cache_lookup(env, ns->name, ns->namelen,
445 					LDNS_RR_TYPE_AAAA, qclass, BIT_CD, now, 0);
446 			if(neg) {
447 				delegpt_add_neg_msg(dp, neg);
448 				lock_rw_unlock(&neg->entry.lock);
449 			}
450 		}
451 	}
452 	return 1;
453 }
454 
455 /** find and add DS or NSEC to delegation msg */
456 static void
find_add_ds(struct module_env * env,struct regional * region,struct dns_msg * msg,struct delegpt * dp,time_t now)457 find_add_ds(struct module_env* env, struct regional* region,
458 	struct dns_msg* msg, struct delegpt* dp, time_t now)
459 {
460 	/* Lookup the DS or NSEC at the delegation point. */
461 	struct ub_packed_rrset_key* rrset = rrset_cache_lookup(
462 		env->rrset_cache, dp->name, dp->namelen, LDNS_RR_TYPE_DS,
463 		msg->qinfo.qclass, 0, now, 0);
464 	if(!rrset) {
465 		/* NOTE: this won't work for alternate NSEC schemes
466 		 *	(opt-in, NSEC3) */
467 		rrset = rrset_cache_lookup(env->rrset_cache, dp->name,
468 			dp->namelen, LDNS_RR_TYPE_NSEC, msg->qinfo.qclass,
469 			0, now, 0);
470 		/* Note: the PACKED_RRSET_NSEC_AT_APEX flag is not used.
471 		 * since this is a referral, we need the NSEC at the parent
472 		 * side of the zone cut, not the NSEC at apex side. */
473 		if(rrset && nsec_has_type(rrset, LDNS_RR_TYPE_DS)) {
474 			lock_rw_unlock(&rrset->entry.lock);
475 			rrset = NULL; /* discard wrong NSEC */
476 		}
477 	}
478 	if(rrset) {
479 		/* add it to auth section. This is the second rrset. */
480 		if((msg->rep->rrsets[msg->rep->rrset_count] =
481 			packed_rrset_copy_region(rrset, region, now))) {
482 			struct packed_rrset_data* d = rrset->entry.data;
483 			msg->rep->ns_numrrsets++;
484 			msg->rep->rrset_count++;
485 			UPDATE_TTL_FROM_RRSET(msg->rep->ttl, d->ttl);
486 		}
487 		lock_rw_unlock(&rrset->entry.lock);
488 	}
489 }
490 
491 struct dns_msg*
dns_msg_create(uint8_t * qname,size_t qnamelen,uint16_t qtype,uint16_t qclass,struct regional * region,size_t capacity)492 dns_msg_create(uint8_t* qname, size_t qnamelen, uint16_t qtype,
493 	uint16_t qclass, struct regional* region, size_t capacity)
494 {
495 	struct dns_msg* msg = (struct dns_msg*)regional_alloc(region,
496 		sizeof(struct dns_msg));
497 	if(!msg)
498 		return NULL;
499 	msg->qinfo.qname = regional_alloc_init(region, qname, qnamelen);
500 	if(!msg->qinfo.qname)
501 		return NULL;
502 	msg->qinfo.qname_len = qnamelen;
503 	msg->qinfo.qtype = qtype;
504 	msg->qinfo.qclass = qclass;
505 	msg->qinfo.local_alias = NULL;
506 	/* non-packed reply_info, because it needs to grow the array */
507 	msg->rep = (struct reply_info*)regional_alloc_zero(region,
508 		sizeof(struct reply_info)-sizeof(struct rrset_ref));
509 	if(!msg->rep)
510 		return NULL;
511 	if(capacity > RR_COUNT_MAX)
512 		return NULL; /* integer overflow protection */
513 	msg->rep->flags = BIT_QR; /* with QR, no AA */
514 	msg->rep->qdcount = 1;
515 	msg->rep->ttl = MAX_TTL; /* will be updated (brought down) while we add
516 				  * rrsets to the message */
517 	msg->rep->reason_bogus = LDNS_EDE_NONE;
518 	msg->rep->rrsets = (struct ub_packed_rrset_key**)
519 		regional_alloc(region,
520 		capacity*sizeof(struct ub_packed_rrset_key*));
521 	if(!msg->rep->rrsets)
522 		return NULL;
523 	return msg;
524 }
525 
526 int
dns_msg_authadd(struct dns_msg * msg,struct regional * region,struct ub_packed_rrset_key * rrset,time_t now)527 dns_msg_authadd(struct dns_msg* msg, struct regional* region,
528 	struct ub_packed_rrset_key* rrset, time_t now)
529 {
530 	struct packed_rrset_data* d = rrset->entry.data;
531 	if(!(msg->rep->rrsets[msg->rep->rrset_count++] =
532 		packed_rrset_copy_region(rrset, region, now)))
533 		return 0;
534 	msg->rep->ns_numrrsets++;
535 	UPDATE_TTL_FROM_RRSET(msg->rep->ttl, d->ttl);
536 	return 1;
537 }
538 
539 int
dns_msg_ansadd(struct dns_msg * msg,struct regional * region,struct ub_packed_rrset_key * rrset,time_t now)540 dns_msg_ansadd(struct dns_msg* msg, struct regional* region,
541 	struct ub_packed_rrset_key* rrset, time_t now)
542 {
543 	struct packed_rrset_data* d = rrset->entry.data;
544 	if(!(msg->rep->rrsets[msg->rep->rrset_count++] =
545 		packed_rrset_copy_region(rrset, region, now)))
546 		return 0;
547 	msg->rep->an_numrrsets++;
548 	UPDATE_TTL_FROM_RRSET(msg->rep->ttl, d->ttl);
549 	return 1;
550 }
551 
552 struct delegpt*
dns_cache_find_delegation(struct module_env * env,uint8_t * qname,size_t qnamelen,uint16_t qtype,uint16_t qclass,struct regional * region,struct dns_msg ** msg,time_t now,int noexpiredabove,uint8_t * expiretop,size_t expiretoplen)553 dns_cache_find_delegation(struct module_env* env, uint8_t* qname,
554 	size_t qnamelen, uint16_t qtype, uint16_t qclass,
555 	struct regional* region, struct dns_msg** msg, time_t now,
556 	int noexpiredabove, uint8_t* expiretop, size_t expiretoplen)
557 {
558 	/* try to find closest NS rrset */
559 	struct ub_packed_rrset_key* nskey;
560 	struct packed_rrset_data* nsdata;
561 	struct delegpt* dp;
562 
563 	nskey = find_closest_of_type(env, qname, qnamelen, qclass, now,
564 		LDNS_RR_TYPE_NS, 0, noexpiredabove, expiretop, expiretoplen);
565 	if(!nskey) /* hope the caller has hints to prime or something */
566 		return NULL;
567 	nsdata = (struct packed_rrset_data*)nskey->entry.data;
568 	/* got the NS key, create delegation point */
569 	dp = delegpt_create(region);
570 	if(!dp || !delegpt_set_name(dp, region, nskey->rk.dname)) {
571 		lock_rw_unlock(&nskey->entry.lock);
572 		log_err("find_delegation: out of memory");
573 		return NULL;
574 	}
575 	/* create referral message */
576 	if(msg) {
577 		/* allocate the array to as much as we could need:
578 		 *	NS rrset + DS/NSEC rrset +
579 		 *	A rrset for every NS RR
580 		 *	AAAA rrset for every NS RR
581 		 */
582 		*msg = dns_msg_create(qname, qnamelen, qtype, qclass, region,
583 			2 + nsdata->count*2);
584 		if(!*msg || !dns_msg_authadd(*msg, region, nskey, now)) {
585 			lock_rw_unlock(&nskey->entry.lock);
586 			log_err("find_delegation: out of memory");
587 			return NULL;
588 		}
589 	}
590 	if(!delegpt_rrset_add_ns(dp, region, nskey, 0,
591 		deleg_port_number(env))) {
592 		lock_rw_unlock(&nskey->entry.lock);
593 		log_err("find_delegation: addns out of memory");
594 		return NULL;
595 	}
596 	lock_rw_unlock(&nskey->entry.lock); /* first unlock before next lookup*/
597 	/* find and add DS/NSEC (if any) */
598 	if(msg)
599 		find_add_ds(env, region, *msg, dp, now);
600 	/* find and add A entries */
601 	if(!find_add_addrs(env, qclass, region, dp, now, msg))
602 		log_err("find_delegation: addrs out of memory");
603 	return dp;
604 }
605 
606 /** allocate dns_msg from query_info and reply_info */
607 static struct dns_msg*
gen_dns_msg(struct regional * region,struct query_info * q,size_t num)608 gen_dns_msg(struct regional* region, struct query_info* q, size_t num)
609 {
610 	struct dns_msg* msg = (struct dns_msg*)regional_alloc(region,
611 		sizeof(struct dns_msg));
612 	if(!msg)
613 		return NULL;
614 	memcpy(&msg->qinfo, q, sizeof(struct query_info));
615 	msg->qinfo.qname = regional_alloc_init(region, q->qname, q->qname_len);
616 	if(!msg->qinfo.qname)
617 		return NULL;
618 	/* allocate replyinfo struct and rrset key array separately */
619 	msg->rep = (struct reply_info*)regional_alloc(region,
620 		sizeof(struct reply_info) - sizeof(struct rrset_ref));
621 	if(!msg->rep)
622 		return NULL;
623 	msg->rep->ttl = MAX_TTL;
624 	msg->rep->reason_bogus = LDNS_EDE_NONE;
625 	msg->rep->reason_bogus_str = NULL;
626 	if(num > RR_COUNT_MAX)
627 		return NULL; /* integer overflow protection */
628 	msg->rep->rrsets = (struct ub_packed_rrset_key**)
629 		regional_alloc(region,
630 		num * sizeof(struct ub_packed_rrset_key*));
631 	if(!msg->rep->rrsets)
632 		return NULL;
633 	return msg;
634 }
635 
636 struct dns_msg*
tomsg(struct module_env * env,struct query_info * q,struct reply_info * r,struct regional * region,time_t now,int allow_expired,struct regional * scratch)637 tomsg(struct module_env* env, struct query_info* q, struct reply_info* r,
638 	struct regional* region, time_t now, int allow_expired,
639 	struct regional* scratch)
640 {
641 	struct dns_msg* msg;
642 	size_t i;
643 	int is_expired = 0;
644 	time_t now_control = now;
645 	if(TTL_IS_EXPIRED(r->ttl, now)) {
646 		/* Check if we are allowed to serve expired */
647 		if(!allow_expired || !reply_info_can_answer_expired(r, now))
648 			return NULL;
649 		/* Change the current time so we can pass the below TTL checks
650 		 * when serving expired data. */
651 		now_control = 0;
652 		is_expired = 1;
653 	}
654 
655 	msg = gen_dns_msg(region, q, r->rrset_count);
656 	if(!msg) return NULL;
657 	msg->rep->flags = r->flags;
658 	msg->rep->qdcount = r->qdcount;
659 	msg->rep->security = r->security;
660 	msg->rep->an_numrrsets = r->an_numrrsets;
661 	msg->rep->ns_numrrsets = r->ns_numrrsets;
662 	msg->rep->ar_numrrsets = r->ar_numrrsets;
663 	msg->rep->rrset_count = r->rrset_count;
664 	msg->rep->authoritative = r->authoritative;
665 	msg->rep->reason_bogus = r->reason_bogus;
666 	if(r->reason_bogus_str) {
667 		msg->rep->reason_bogus_str = regional_strdup(region, r->reason_bogus_str);
668 	}
669 
670 	if(!rrset_array_lock(r->ref, r->rrset_count, now_control)) {
671 		return NULL;
672 	}
673 	if(r->an_numrrsets > 0 && (r->rrsets[0]->rk.type == htons(
674 		LDNS_RR_TYPE_CNAME) || r->rrsets[0]->rk.type == htons(
675 		LDNS_RR_TYPE_DNAME)) && !reply_check_cname_chain(q, r)) {
676 		/* cname chain is now invalid, reconstruct msg */
677 		rrset_array_unlock(r->ref, r->rrset_count);
678 		return NULL;
679 	}
680 	if(r->security == sec_status_secure && !reply_all_rrsets_secure(r)) {
681 		/* message rrsets have changed status, revalidate */
682 		rrset_array_unlock(r->ref, r->rrset_count);
683 		return NULL;
684 	}
685 	for(i=0; i<msg->rep->rrset_count; i++) {
686 		struct packed_rrset_data* d;
687 		msg->rep->rrsets[i] = packed_rrset_copy_region(r->rrsets[i],
688 			region, now);
689 		if(!msg->rep->rrsets[i]) {
690 			rrset_array_unlock(r->ref, r->rrset_count);
691 			return NULL;
692 		}
693 		d = msg->rep->rrsets[i]->entry.data;
694 		UPDATE_TTL_FROM_RRSET(msg->rep->ttl, d->ttl);
695 	}
696 	if(msg->rep->rrset_count < 1) {
697 		msg->rep->ttl = is_expired
698 			?SERVE_EXPIRED_REPLY_TTL
699 			:r->ttl - now;
700 		if(r->prefetch_ttl > now)
701 			msg->rep->prefetch_ttl = r->prefetch_ttl - now;
702 		else
703 			msg->rep->prefetch_ttl = PREFETCH_TTL_CALC(msg->rep->ttl);
704 	} else {
705 		/* msg->rep->ttl has been updated through the RRSets above */
706 		msg->rep->prefetch_ttl = PREFETCH_TTL_CALC(msg->rep->ttl);
707 	}
708 	msg->rep->serve_expired_ttl = msg->rep->ttl + SERVE_EXPIRED_TTL;
709 	msg->rep->serve_expired_norec_ttl = 0;
710 	if(env)
711 		rrset_array_unlock_touch(env->rrset_cache, scratch, r->ref,
712 		r->rrset_count);
713 	else
714 		rrset_array_unlock(r->ref, r->rrset_count);
715 	return msg;
716 }
717 
718 struct dns_msg*
dns_msg_deepcopy_region(struct dns_msg * origin,struct regional * region)719 dns_msg_deepcopy_region(struct dns_msg* origin, struct regional* region)
720 {
721 	size_t i;
722 	struct ub_packed_rrset_key** saved_rrsets;
723 	struct dns_msg* res = NULL;
724 	size_t rep_alloc_size = sizeof(struct reply_info)
725 		- sizeof(struct rrset_ref);  /* this is the size of res->rep
726 						allocated in gen_dns_msg() */
727 	res = gen_dns_msg(region, &origin->qinfo, origin->rep->rrset_count);
728 	if(!res) return NULL;
729 	saved_rrsets = res->rep->rrsets; /* save rrsets alloc by gen_dns_msg */
730 	memcpy(res->rep, origin->rep, rep_alloc_size);
731 	res->rep->rrsets = saved_rrsets;
732 	if(origin->rep->reason_bogus_str) {
733 		res->rep->reason_bogus_str = regional_strdup(region,
734 			origin->rep->reason_bogus_str);
735 	}
736 	for(i=0; i<res->rep->rrset_count; i++) {
737 		res->rep->rrsets[i] = packed_rrset_copy_region(
738 			origin->rep->rrsets[i], region, 0);
739 		if(!res->rep->rrsets[i]) {
740 			return NULL;
741 		}
742 	}
743 	return res;
744 }
745 
746 /** synthesize RRset-only response from cached RRset item */
747 static struct dns_msg*
rrset_msg(struct ub_packed_rrset_key * rrset,struct regional * region,time_t now,struct query_info * q)748 rrset_msg(struct ub_packed_rrset_key* rrset, struct regional* region,
749 	time_t now, struct query_info* q)
750 {
751 	struct dns_msg* msg;
752 	struct packed_rrset_data* d = (struct packed_rrset_data*)
753 		rrset->entry.data;
754 	if(TTL_IS_EXPIRED(d->ttl, now))
755 		return NULL;
756 	msg = gen_dns_msg(region, q, 1); /* only the CNAME (or other) RRset */
757 	if(!msg)
758 		return NULL;
759 	msg->rep->flags = BIT_QR; /* reply, no AA, no error */
760         msg->rep->authoritative = 0; /* reply stored in cache can't be authoritative */
761 	msg->rep->qdcount = 1;
762 	msg->rep->ttl = d->ttl - now;
763 	msg->rep->prefetch_ttl = PREFETCH_TTL_CALC(msg->rep->ttl);
764 	msg->rep->serve_expired_ttl = msg->rep->ttl + SERVE_EXPIRED_TTL;
765 	msg->rep->serve_expired_norec_ttl = 0;
766 	msg->rep->security = sec_status_unchecked;
767 	msg->rep->an_numrrsets = 1;
768 	msg->rep->ns_numrrsets = 0;
769 	msg->rep->ar_numrrsets = 0;
770 	msg->rep->rrset_count = 1;
771 	msg->rep->reason_bogus = LDNS_EDE_NONE;
772 	msg->rep->rrsets[0] = packed_rrset_copy_region(rrset, region, now);
773 	if(!msg->rep->rrsets[0]) /* copy CNAME */
774 		return NULL;
775 	return msg;
776 }
777 
778 /** synthesize DNAME+CNAME response from cached DNAME item */
779 static struct dns_msg*
synth_dname_msg(struct ub_packed_rrset_key * rrset,struct regional * region,time_t now,struct query_info * q,enum sec_status * sec_status)780 synth_dname_msg(struct ub_packed_rrset_key* rrset, struct regional* region,
781 	time_t now, struct query_info* q, enum sec_status* sec_status)
782 {
783 	struct dns_msg* msg;
784 	struct ub_packed_rrset_key* ck;
785 	struct packed_rrset_data* newd, *d = (struct packed_rrset_data*)
786 		rrset->entry.data;
787 	uint8_t* newname, *dtarg = NULL;
788 	size_t newlen, dtarglen;
789 	time_t rr_ttl;
790 	int graceperiod = 0;
791 	if(TTL_IS_EXPIRED(d->ttl, now)) {
792 		/* Allow TTL=0 DNAME from upstream within grace period */
793 		if(!(rrset->rk.flags & PACKED_RRSET_UPSTREAM_0TTL))
794 			return NULL;
795 		rr_ttl = 0;
796 		/* Since PACKED_RRSET_UPSTREAM_0TTL set the flag that
797 		 * the grace period has been applied, this stops the rrset
798 		 * from getting stored back into the cache with a bigger TTL.*/
799 		graceperiod = 1;
800 	} else {
801 		rr_ttl = d->ttl - now;
802 	}
803 	/* only allow validated (with DNSSEC) DNAMEs used from cache
804 	 * for insecure DNAMEs, query again. */
805 	*sec_status = d->security;
806 	/* return sec status, so the status of the CNAME can be checked
807 	 * by the calling routine. */
808 	msg = gen_dns_msg(region, q, 2); /* DNAME + CNAME RRset */
809 	if(!msg)
810 		return NULL;
811 	msg->rep->flags = BIT_QR; /* reply, no AA, no error */
812         msg->rep->authoritative = 0; /* reply stored in cache can't be authoritative */
813 	msg->rep->qdcount = 1;
814 	msg->rep->ttl = rr_ttl;
815 	msg->rep->prefetch_ttl = PREFETCH_TTL_CALC(msg->rep->ttl);
816 	msg->rep->serve_expired_ttl = msg->rep->ttl + SERVE_EXPIRED_TTL;
817 	msg->rep->serve_expired_norec_ttl = 0;
818 	msg->rep->security = sec_status_unchecked;
819 	msg->rep->an_numrrsets = 1;
820 	msg->rep->ns_numrrsets = 0;
821 	msg->rep->ar_numrrsets = 0;
822 	msg->rep->rrset_count = 1;
823 	msg->rep->reason_bogus = LDNS_EDE_NONE;
824 	msg->rep->rrsets[0] = packed_rrset_copy_region(rrset, region, now);
825 	if(!msg->rep->rrsets[0]) /* copy DNAME */
826 		return NULL;
827 	if(graceperiod)
828 		msg->rep->rrsets[0]->rk.flags |= PACKED_RRSET_0TTL_GRACE;
829 	/* synth CNAME rrset */
830 	get_cname_target(rrset, &dtarg, &dtarglen);
831 	if(!dtarg)
832 		return NULL;
833 	newlen = q->qname_len + dtarglen - rrset->rk.dname_len;
834 	if(newlen > LDNS_MAX_DOMAINLEN) {
835 		msg->rep->flags |= LDNS_RCODE_YXDOMAIN;
836 		return msg;
837 	}
838 	newname = (uint8_t*)regional_alloc(region, newlen);
839 	if(!newname)
840 		return NULL;
841 	/* new name is concatenation of qname front (without DNAME owner)
842 	 * and DNAME target name */
843 	memcpy(newname, q->qname, q->qname_len-rrset->rk.dname_len);
844 	memmove(newname+(q->qname_len-rrset->rk.dname_len), dtarg, dtarglen);
845 	/* create rest of CNAME rrset */
846 	ck = (struct ub_packed_rrset_key*)regional_alloc(region,
847 		sizeof(struct ub_packed_rrset_key));
848 	if(!ck)
849 		return NULL;
850 	memset(&ck->entry, 0, sizeof(ck->entry));
851 	msg->rep->rrsets[1] = ck;
852 	ck->entry.key = ck;
853 	ck->rk.type = htons(LDNS_RR_TYPE_CNAME);
854 	ck->rk.rrset_class = rrset->rk.rrset_class;
855 	ck->rk.flags = 0;
856 	ck->rk.dname = regional_alloc_init(region, q->qname, q->qname_len);
857 	if(!ck->rk.dname)
858 		return NULL;
859 	ck->rk.dname_len = q->qname_len;
860 	ck->entry.hash = rrset_key_hash(&ck->rk);
861 	newd = (struct packed_rrset_data*)regional_alloc_zero(region,
862 		sizeof(struct packed_rrset_data) + sizeof(size_t) +
863 		sizeof(uint8_t*) + sizeof(time_t) + sizeof(uint16_t)
864 		+ newlen);
865 	if(!newd)
866 		return NULL;
867 	ck->entry.data = newd;
868 	newd->ttl = rr_ttl; /* RFC6672: synth CNAME TTL == DNAME TTL */
869 	newd->count = 1;
870 	newd->rrsig_count = 0;
871 	newd->trust = rrset_trust_ans_noAA;
872 	newd->rr_len = (size_t*)((uint8_t*)newd +
873 		sizeof(struct packed_rrset_data));
874 	newd->rr_len[0] = newlen + sizeof(uint16_t);
875 	packed_rrset_ptr_fixup(newd);
876 	newd->rr_ttl[0] = newd->ttl;
877 	msg->rep->ttl = newd->ttl;
878 	msg->rep->prefetch_ttl = PREFETCH_TTL_CALC(newd->ttl);
879 	msg->rep->serve_expired_ttl = newd->ttl + SERVE_EXPIRED_TTL;
880 	sldns_write_uint16(newd->rr_data[0], newlen);
881 	memmove(newd->rr_data[0] + sizeof(uint16_t), newname, newlen);
882 	msg->rep->an_numrrsets ++;
883 	msg->rep->rrset_count ++;
884 	return msg;
885 }
886 
887 /** Fill TYPE_ANY response with some data from cache */
888 static struct dns_msg*
fill_any(struct module_env * env,uint8_t * qname,size_t qnamelen,uint16_t qtype,uint16_t qclass,struct regional * region)889 fill_any(struct module_env* env,
890 	uint8_t* qname, size_t qnamelen, uint16_t qtype, uint16_t qclass,
891 	struct regional* region)
892 {
893 	time_t now = *env->now;
894 	struct dns_msg* msg = NULL;
895 	uint16_t lookup[] = {LDNS_RR_TYPE_A, LDNS_RR_TYPE_AAAA,
896 		LDNS_RR_TYPE_MX, LDNS_RR_TYPE_SOA, LDNS_RR_TYPE_NS,
897 		LDNS_RR_TYPE_DNAME, 0};
898 	int i, num=6; /* number of RR types to look up */
899 	log_assert(lookup[num] == 0);
900 
901 	if(env->cfg->deny_any) {
902 		/* return empty message */
903 		msg = dns_msg_create(qname, qnamelen, qtype, qclass,
904 			region, 0);
905 		if(!msg) {
906 			return NULL;
907 		}
908 		/* set NOTIMPL for RFC 8482 */
909 		msg->rep->flags |= LDNS_RCODE_NOTIMPL;
910 		msg->rep->security = sec_status_indeterminate;
911 		msg->rep->ttl = 1; /* empty NOTIMPL response will never be
912 				    * updated with rrsets, set TTL to 1 */
913 		return msg;
914 	}
915 
916 	for(i=0; i<num; i++) {
917 		/* look up this RR for inclusion in type ANY response */
918 		struct ub_packed_rrset_key* rrset = rrset_cache_lookup(
919 			env->rrset_cache, qname, qnamelen, lookup[i],
920 			qclass, 0, now, 0);
921 		struct packed_rrset_data *d;
922 		if(!rrset)
923 			continue;
924 
925 		/* only if rrset from answer section */
926 		d = (struct packed_rrset_data*)rrset->entry.data;
927 		if(d->trust == rrset_trust_add_noAA ||
928 			d->trust == rrset_trust_auth_noAA ||
929 			d->trust == rrset_trust_add_AA ||
930 			d->trust == rrset_trust_auth_AA) {
931 			lock_rw_unlock(&rrset->entry.lock);
932 			continue;
933 		}
934 
935 		/* create msg if none */
936 		if(!msg) {
937 			msg = dns_msg_create(qname, qnamelen, qtype, qclass,
938 				region, (size_t)(num-i));
939 			if(!msg) {
940 				lock_rw_unlock(&rrset->entry.lock);
941 				return NULL;
942 			}
943 		}
944 
945 		/* add RRset to response */
946 		if(!dns_msg_ansadd(msg, region, rrset, now)) {
947 			lock_rw_unlock(&rrset->entry.lock);
948 			return NULL;
949 		}
950 		lock_rw_unlock(&rrset->entry.lock);
951 	}
952 	return msg;
953 }
954 
955 struct dns_msg*
dns_cache_lookup(struct module_env * env,uint8_t * qname,size_t qnamelen,uint16_t qtype,uint16_t qclass,uint16_t flags,struct regional * region,struct regional * scratch,int no_partial,uint8_t * dpname,size_t dpnamelen)956 dns_cache_lookup(struct module_env* env,
957 	uint8_t* qname, size_t qnamelen, uint16_t qtype, uint16_t qclass,
958 	uint16_t flags, struct regional* region, struct regional* scratch,
959 	int no_partial, uint8_t* dpname, size_t dpnamelen)
960 {
961 	struct lruhash_entry* e;
962 	struct query_info k;
963 	hashvalue_type h;
964 	time_t now = *env->now;
965 	struct ub_packed_rrset_key* rrset;
966 
967 	/* lookup first, this has both NXdomains and ANSWER responses */
968 	k.qname = qname;
969 	k.qname_len = qnamelen;
970 	k.qtype = qtype;
971 	k.qclass = qclass;
972 	k.local_alias = NULL;
973 	h = query_info_hash(&k, flags);
974 	e = slabhash_lookup(env->msg_cache, h, &k, 0);
975 	if(e) {
976 		struct msgreply_entry* key = (struct msgreply_entry*)e->key;
977 		struct reply_info* data = (struct reply_info*)e->data;
978 		struct dns_msg* msg = tomsg(env, &key->key, data, region, now, 0,
979 			scratch);
980 		if(msg) {
981 			lock_rw_unlock(&e->lock);
982 			return msg;
983 		}
984 		/* could be msg==NULL; due to TTL or not all rrsets available */
985 		lock_rw_unlock(&e->lock);
986 	}
987 
988 	/* see if a DNAME exists. Checked for first, to enforce that DNAMEs
989 	 * are more important, the CNAME is resynthesized and thus
990 	 * consistent with the DNAME */
991 	if(!no_partial &&
992 		(rrset=find_closest_of_type(env, qname, qnamelen, qclass, now,
993 		LDNS_RR_TYPE_DNAME, 1, 0, NULL, 0))) {
994 		/* synthesize a DNAME+CNAME message based on this */
995 		enum sec_status sec_status = sec_status_unchecked;
996 		struct dns_msg* msg = synth_dname_msg(rrset, region, now, &k,
997 			&sec_status);
998 		if(msg) {
999 			struct ub_packed_rrset_key* cname_rrset;
1000 			lock_rw_unlock(&rrset->entry.lock);
1001 			/* now, after unlocking the DNAME rrset lock,
1002 			 * check the sec_status, and see if we need to look
1003 			 * up the CNAME record associated before it can
1004 			 * be used */
1005 			/* normally, only secure DNAMEs allowed from cache*/
1006 			if(sec_status == sec_status_secure)
1007 				return msg;
1008 			/* but if we have a CNAME cached with this name, then we
1009 			 * have previously already allowed this name to pass.
1010 			 * the next cache lookup is going to fetch that CNAME itself,
1011 			 * but it is better to have the (unsigned)DNAME + CNAME in
1012 			 * that case */
1013 			cname_rrset = rrset_cache_lookup(
1014 				env->rrset_cache, qname, qnamelen,
1015 				LDNS_RR_TYPE_CNAME, qclass, 0, now, 0);
1016 			if(cname_rrset) {
1017 				/* CNAME already synthesized by
1018 				 * synth_dname_msg routine, so we can
1019 				 * straight up return the msg */
1020 				lock_rw_unlock(&cname_rrset->entry.lock);
1021 				return msg;
1022 			}
1023 		} else {
1024 			lock_rw_unlock(&rrset->entry.lock);
1025 		}
1026 	}
1027 
1028 	/* see if we have CNAME for this domain,
1029 	 * but not for DS records (which are part of the parent) */
1030 	if(!no_partial && qtype != LDNS_RR_TYPE_DS &&
1031 	   (rrset=rrset_cache_lookup(env->rrset_cache, qname, qnamelen,
1032 		LDNS_RR_TYPE_CNAME, qclass, 0, now, 0))) {
1033 		uint8_t* wc = NULL;
1034 		size_t wl;
1035 		/* if the rrset is not a wildcard expansion, with wcname */
1036 		/* because, if we return that CNAME rrset on its own, it is
1037 		 * missing the NSEC or NSEC3 proof */
1038 		if(!(val_rrset_wildcard(rrset, &wc, &wl) && wc != NULL)) {
1039 			struct dns_msg* msg = rrset_msg(rrset, region, now, &k);
1040 			if(msg) {
1041 				lock_rw_unlock(&rrset->entry.lock);
1042 				return msg;
1043 			}
1044 		}
1045 		lock_rw_unlock(&rrset->entry.lock);
1046 	}
1047 
1048 	/* construct DS, DNSKEY messages from rrset cache. */
1049 	if((qtype == LDNS_RR_TYPE_DS || qtype == LDNS_RR_TYPE_DNSKEY) &&
1050 		(rrset=rrset_cache_lookup(env->rrset_cache, qname, qnamelen,
1051 		qtype, qclass, 0, now, 0))) {
1052 		/* if the rrset is from the additional section, and the
1053 		 * signatures have fallen off, then do not synthesize a msg
1054 		 * instead, allow a full query for signed results to happen.
1055 		 * Forego all rrset data from additional section, because
1056 		 * some signatures may not be present and cause validation
1057 		 * failure.
1058 		 */
1059 		struct packed_rrset_data *d = (struct packed_rrset_data*)
1060 			rrset->entry.data;
1061 		if(d->trust != rrset_trust_add_noAA &&
1062 			d->trust != rrset_trust_add_AA &&
1063 			(qtype == LDNS_RR_TYPE_DS ||
1064 				(d->trust != rrset_trust_auth_noAA
1065 				&& d->trust != rrset_trust_auth_AA) )) {
1066 			struct dns_msg* msg = rrset_msg(rrset, region, now, &k);
1067 			if(msg) {
1068 				lock_rw_unlock(&rrset->entry.lock);
1069 				return msg;
1070 			}
1071 		}
1072 		lock_rw_unlock(&rrset->entry.lock);
1073 	}
1074 
1075 	/* stop downwards cache search on NXDOMAIN.
1076 	 * Empty nonterminals are NOERROR, so an NXDOMAIN for foo
1077 	 * means bla.foo also does not exist.  The DNSSEC proofs are
1078 	 * the same.  We search upwards for NXDOMAINs. */
1079 	if(env->cfg->harden_below_nxdomain) {
1080 		while(!dname_is_root(k.qname)) {
1081 			if(dpname && dpnamelen
1082 				&& !dname_strict_subdomain_c(k.qname, dpname))
1083 				break; /* no synth nxdomain above the stub */
1084 			dname_remove_label(&k.qname, &k.qname_len);
1085 			h = query_info_hash(&k, flags);
1086 			e = slabhash_lookup(env->msg_cache, h, &k, 0);
1087 			if(!e && k.qtype != LDNS_RR_TYPE_A &&
1088 				env->cfg->qname_minimisation) {
1089 				k.qtype = LDNS_RR_TYPE_A;
1090 				h = query_info_hash(&k, flags);
1091 				e = slabhash_lookup(env->msg_cache, h, &k, 0);
1092 			}
1093 			if(e) {
1094 				struct reply_info* data = (struct reply_info*)e->data;
1095 				struct dns_msg* msg;
1096 				if(FLAGS_GET_RCODE(data->flags) == LDNS_RCODE_NXDOMAIN
1097 					&& data->security == sec_status_secure
1098 					&& (data->an_numrrsets == 0 ||
1099 						ntohs(data->rrsets[0]->rk.type) != LDNS_RR_TYPE_CNAME)
1100 					&& (msg=tomsg(env, &k, data, region, now, 0, scratch))) {
1101 					lock_rw_unlock(&e->lock);
1102 					msg->qinfo.qname=qname;
1103 					msg->qinfo.qname_len=qnamelen;
1104 					/* check that DNSSEC really works out */
1105 					msg->rep->security = sec_status_unchecked;
1106 					iter_scrub_nxdomain(msg);
1107 					return msg;
1108 				}
1109 				lock_rw_unlock(&e->lock);
1110 			}
1111 			k.qtype = qtype;
1112 		}
1113 	}
1114 
1115 	/* fill common RR types for ANY response to avoid requery */
1116 	if(qtype == LDNS_RR_TYPE_ANY) {
1117 		return fill_any(env, qname, qnamelen, qtype, qclass, region);
1118 	}
1119 
1120 	return NULL;
1121 }
1122 
1123 int
dns_cache_store(struct module_env * env,struct query_info * msgqinf,struct reply_info * msgrep,int is_referral,time_t leeway,int pside,struct regional * region,uint32_t flags,time_t qstarttime,int is_valrec)1124 dns_cache_store(struct module_env* env, struct query_info* msgqinf,
1125 	struct reply_info* msgrep, int is_referral, time_t leeway, int pside,
1126 	struct regional* region, uint32_t flags, time_t qstarttime,
1127 	int is_valrec)
1128 {
1129 	struct reply_info* rep = NULL;
1130 	if(SERVE_EXPIRED) {
1131 		/* We are serving expired records. Before caching, check if a
1132 		 * useful expired record exists. */
1133 		struct msgreply_entry* e = msg_cache_lookup(env,
1134 			msgqinf->qname, msgqinf->qname_len, msgqinf->qtype,
1135 			msgqinf->qclass, flags, 0, 1);
1136 		if(e) {
1137 			struct reply_info* cached = e->entry.data;
1138 			if(TTL_IS_EXPIRED(cached->ttl, *env->now)
1139 				&& reply_info_could_use_expired(cached, *env->now)
1140 				/* If we are validating make sure only
1141 				 * validating modules can update such messages.
1142 				 * In that case don't cache it and let a
1143 				 * subsequent module handle the caching. For
1144 				 * example, the iterator should not replace an
1145 				 * expired secure answer with a fresh unchecked
1146 				 * one and let the validator manage caching. */
1147 				&& cached->security != sec_status_bogus
1148 				&& (env->need_to_validate &&
1149 				msgrep->security == sec_status_unchecked)
1150 				/* Exceptions to that rule are:
1151 				 * o recursions that don't need validation but
1152 				 *   need to update the cache for coherence
1153 				 *   (delegation information while iterating,
1154 				 *   DNSKEY and DS lookups from validator)
1155 				 * o explicit RRSIG queries that are not
1156 				 *   validated. */
1157 				&& !is_valrec
1158 				&& msgqinf->qtype != LDNS_RR_TYPE_RRSIG) {
1159 				if((int)FLAGS_GET_RCODE(msgrep->flags) !=
1160 					LDNS_RCODE_NOERROR &&
1161 					(int)FLAGS_GET_RCODE(msgrep->flags) !=
1162 					LDNS_RCODE_NXDOMAIN) {
1163 					/* The current response has an
1164 					 * erroneous rcode. Adjust norec time
1165 					 * so that additional lookups are not
1166 					 * performed for some time. */
1167 					verbose(VERB_ALGO, "set "
1168 						"serve-expired-norec-ttl for "
1169 						"response in cache");
1170 					cached->serve_expired_norec_ttl =
1171 						NORR_TTL + *env->now;
1172 					if(env->cfg->serve_expired_ttl_reset &&
1173 					    cached->serve_expired_ttl
1174 					    < *env->now +
1175 					    env->cfg->serve_expired_ttl) {
1176 						/* Reset serve-expired-ttl for
1177 						 * valid response in cache. */
1178 						verbose(VERB_ALGO, "reset "
1179 							"serve-expired-ttl "
1180 							"for response in cache");
1181 						cached->serve_expired_ttl =
1182 						    *env->now +
1183 						    env->cfg->serve_expired_ttl;
1184 					}
1185 				}
1186 				verbose(VERB_ALGO, "a validated expired entry "
1187 					"could be overwritten, skip caching "
1188 					"the new message at this stage");
1189 				lock_rw_unlock(&e->entry.lock);
1190 				return 1;
1191 			}
1192 			lock_rw_unlock(&e->entry.lock);
1193 		}
1194 	}
1195 	/* alloc, malloc properly (not in region, like msg is) */
1196 	rep = reply_info_copy(msgrep, env->alloc, NULL);
1197 	if(!rep)
1198 		return 0;
1199 	/* ttl must be relative ;i.e. 0..86400 not  time(0)+86400.
1200 	 * the env->now is added to message and RRsets in this routine. */
1201 	/* the leeway is used to invalidate other rrsets earlier */
1202 	if(is_referral) {
1203 		/* store rrsets */
1204 		struct rrset_ref ref;
1205 		size_t i;
1206 		for(i=0; i<rep->rrset_count; i++) {
1207 			packed_rrset_ttl_add((struct packed_rrset_data*)
1208 				rep->rrsets[i]->entry.data, *env->now);
1209 			ref.key = rep->rrsets[i];
1210 			ref.id = rep->rrsets[i]->id;
1211 			/*ignore ret: it was in the cache, ref updated */
1212 			/* no leeway for typeNS */
1213 			(void)rrset_cache_update(env->rrset_cache, &ref,
1214 				env->alloc,
1215 				((ntohs(ref.key->rk.type)==LDNS_RR_TYPE_NS
1216 				 && !pside) ? qstarttime:*env->now + leeway));
1217 		}
1218 		reply_info_delete(rep, NULL);
1219 		return 1;
1220 	} else {
1221 		/* store msg, and rrsets */
1222 		struct query_info qinf;
1223 		hashvalue_type h;
1224 
1225 		qinf = *msgqinf;
1226 		qinf.qname = memdup(msgqinf->qname, msgqinf->qname_len);
1227 		if(!qinf.qname) {
1228 			reply_info_parsedelete(rep, env->alloc);
1229 			return 0;
1230 		}
1231 		/* fixup flags to be sensible for a reply based on the cache */
1232 		/* this module means that RA is available. It is an answer QR.
1233 		 * Not AA from cache. Not CD in cache (depends on client bit). */
1234 		rep->flags |= (BIT_RA | BIT_QR);
1235 		rep->flags &= ~(BIT_AA | BIT_CD);
1236 		h = query_info_hash(&qinf, (uint16_t)flags);
1237 		dns_cache_store_msg(env, &qinf, h, rep, leeway, pside, msgrep,
1238 			flags, region, qstarttime);
1239 		/* qname is used inside query_info_entrysetup, and set to
1240 		 * NULL. If it has not been used, free it. free(0) is safe. */
1241 		free(qinf.qname);
1242 	}
1243 	return 1;
1244 }
1245 
1246 int
dns_cache_prefetch_adjust(struct module_env * env,struct query_info * qinfo,time_t adjust,uint16_t flags)1247 dns_cache_prefetch_adjust(struct module_env* env, struct query_info* qinfo,
1248         time_t adjust, uint16_t flags)
1249 {
1250 	struct msgreply_entry* msg;
1251 	msg = msg_cache_lookup(env, qinfo->qname, qinfo->qname_len,
1252 		qinfo->qtype, qinfo->qclass, flags, *env->now, 1);
1253 	if(msg) {
1254 		struct reply_info* rep = (struct reply_info*)msg->entry.data;
1255 		if(rep) {
1256 			rep->prefetch_ttl += adjust;
1257 			lock_rw_unlock(&msg->entry.lock);
1258 			return 1;
1259 		}
1260 		lock_rw_unlock(&msg->entry.lock);
1261 	}
1262 	return 0;
1263 }
1264