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