contrib/unbound/services/cache/dns.c

   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 "validator/val_nsec.h"
  44 #include "validator/val_utils.h"
  45 #include "services/cache/dns.h"
  46 #include "services/cache/rrset.h"
  47 #include "util/data/msgreply.h"
  48 #include "util/data/packed_rrset.h"
  49 #include "util/data/dname.h"
  50 #include "util/module.h"
  51 #include "util/net_help.h"
  52 #include "util/regional.h"
  53 #include "util/config_file.h"
  54 #include "sldns/sbuffer.h"
  55
  56 /** store rrsets in the rrset cache.
  57  * @param env: module environment with caches.
  58  * @param rep: contains list of rrsets to store.
  59  * @param now: current time.
  60  * @param leeway: during prefetch how much leeway to update TTLs.
  61  *      This makes rrsets (other than type NS) timeout sooner so they get
  62  *      updated with a new full TTL.
  63  *      Type NS does not get this, because it must not be refreshed from the
  64  *      child domain, but keep counting down properly.
  65  * @param pside: if from parentside discovered NS, so that its NS is okay
  66  *      in a prefetch situation to be updated (without becoming sticky).
  67  * @param qrep: update rrsets here if cache is better
  68  * @param region: for qrep allocs.
  69  */
  70 static void
  71 store_rrsets(struct module_env* env, struct reply_info* rep, time_t now,
  72         time_t leeway, int pside, struct reply_info* qrep,
  73         struct regional* region)
  74 {
  75         size_t i;
  76         /* see if rrset already exists in cache, if not insert it. */
  77         for(i=0; i<rep->rrset_count; i++) {
  78                 rep->ref[i].key = rep->rrsets[i];
  79                 rep->ref[i].id = rep->rrsets[i]->id;
  80                 /* update ref if it was in the cache */
  81                 switch(rrset_cache_update(env->rrset_cache, &rep->ref[i],
  82                         env->alloc, now + ((ntohs(rep->ref[i].key->rk.type)==
  83                         LDNS_RR_TYPE_NS && !pside)?0:leeway))) {
  84                 case 0: /* ref unchanged, item inserted */
  85                         break;
  86                 case 2: /* ref updated, cache is superior */
  87                         if(region) {
  88                                 struct ub_packed_rrset_key* ck;
  89                                 lock_rw_rdlock(&rep->ref[i].key->entry.lock);
  90                                 /* if deleted rrset, do not copy it */
  91                                 if(rep->ref[i].key->id == 0)
  92                                         ck = NULL;
  93                                 else    ck = packed_rrset_copy_region(
  94                                         rep->ref[i].key, region, now);
  95                                 lock_rw_unlock(&rep->ref[i].key->entry.lock);
  96                                 if(ck) {
  97                                         /* use cached copy if memory allows */
  98                                         qrep->rrsets[i] = ck;
  99                                 }
 100                         }
 101                         /* no break: also copy key item */
 102                         /* the line below is matched by gcc regex and silences
 103                          * the fallthrough warning */
 104                         /* fallthrough */
 105                 case 1: /* ref updated, item inserted */
 106                         rep->rrsets[i] = rep->ref[i].key;
 107                 }
 108         }
 109 }
 110
 111 void
 112 dns_cache_store_msg(struct module_env* env, struct query_info* qinfo,
 113         hashvalue_type hash, struct reply_info* rep, time_t leeway, int pside,
 114         struct reply_info* qrep, uint32_t flags, struct regional* region)
 115 {
 116         struct msgreply_entry* e;
 117         time_t ttl = rep->ttl;
 118         size_t i;
 119
 120         /* store RRsets */
 121         for(i=0; i<rep->rrset_count; i++) {
 122                 rep->ref[i].key = rep->rrsets[i];
 123                 rep->ref[i].id = rep->rrsets[i]->id;
 124         }
 125
 126         /* there was a reply_info_sortref(rep) here but it seems to be
 127          * unnecessary, because the cache gets locked per rrset. */
 128         reply_info_set_ttls(rep, *env->now);
 129         store_rrsets(env, rep, *env->now, leeway, pside, qrep, region);
 130         if(ttl == 0 && !(flags & DNSCACHE_STORE_ZEROTTL)) {
 131                 /* we do not store the message, but we did store the RRs,
 132                  * which could be useful for delegation information */
 133                 verbose(VERB_ALGO, "TTL 0: dropped msg from cache");
 134                 free(rep);
 135                 return;
 136         }
 137
 138         /* store msg in the cache */
 139         reply_info_sortref(rep);
 140         if(!(e = query_info_entrysetup(qinfo, rep, hash))) {
 141                 log_err("store_msg: malloc failed");
 142                 return;
 143         }
 144         slabhash_insert(env->msg_cache, hash, &e->entry, rep, env->alloc);
 145 }
 146
 147 /** find closest NS or DNAME and returns the rrset (locked) */
 148 static struct ub_packed_rrset_key*
 149 find_closest_of_type(struct module_env* env, uint8_t* qname, size_t qnamelen,
 150         uint16_t qclass, time_t now, uint16_t searchtype, int stripfront)
 151 {
 152         struct ub_packed_rrset_key *rrset;
 153         uint8_t lablen;
 154
 155         if(stripfront) {
 156                 /* strip off so that DNAMEs have strict subdomain match */
 157                 lablen = *qname;
 158                 qname += lablen + 1;
 159                 qnamelen -= lablen + 1;
 160         }
 161
 162         /* snip off front part of qname until the type is found */
 163         while(qnamelen > 0) {
 164                 if((rrset = rrset_cache_lookup(env->rrset_cache, qname,
 165                         qnamelen, searchtype, qclass, 0, now, 0)))
 166                         return rrset;
 167
 168                 /* snip off front label */
 169                 lablen = *qname;
 170                 qname += lablen + 1;
 171                 qnamelen -= lablen + 1;
 172         }
 173         return NULL;
 174 }
 175
 176 /** add addr to additional section */
 177 static void
 178 addr_to_additional(struct ub_packed_rrset_key* rrset, struct regional* region,
 179         struct dns_msg* msg, time_t now)
 180 {
 181         if((msg->rep->rrsets[msg->rep->rrset_count] =
 182                 packed_rrset_copy_region(rrset, region, now))) {
 183                 msg->rep->ar_numrrsets++;
 184                 msg->rep->rrset_count++;
 185         }
 186 }
 187
 188 /** lookup message in message cache */
 189 struct msgreply_entry*
 190 msg_cache_lookup(struct module_env* env, uint8_t* qname, size_t qnamelen,
 191         uint16_t qtype, uint16_t qclass, uint16_t flags, time_t now, int wr)
 192 {
 193         struct lruhash_entry* e;
 194         struct query_info k;
 195         hashvalue_type h;
 196
 197         k.qname = qname;
 198         k.qname_len = qnamelen;
 199         k.qtype = qtype;
 200         k.qclass = qclass;
 201         k.local_alias = NULL;
 202         h = query_info_hash(&k, flags);
 203         e = slabhash_lookup(env->msg_cache, h, &k, wr);
 204
 205         if(!e) return NULL;
 206         if( now > ((struct reply_info*)e->data)->ttl ) {
 207                 lock_rw_unlock(&e->lock);
 208                 return NULL;
 209         }
 210         return (struct msgreply_entry*)e->key;
 211 }
 212
 213 /** find and add A and AAAA records for nameservers in delegpt */
 214 static int
 215 find_add_addrs(struct module_env* env, uint16_t qclass,
 216         struct regional* region, struct delegpt* dp, time_t now,
 217         struct dns_msg** msg)
 218 {
 219         struct delegpt_ns* ns;
 220         struct msgreply_entry* neg;
 221         struct ub_packed_rrset_key* akey;
 222         for(ns = dp->nslist; ns; ns = ns->next) {
 223                 akey = rrset_cache_lookup(env->rrset_cache, ns->name,
 224                         ns->namelen, LDNS_RR_TYPE_A, qclass, 0, now, 0);
 225                 if(akey) {
 226                         if(!delegpt_add_rrset_A(dp, region, akey, 0)) {
 227                                 lock_rw_unlock(&akey->entry.lock);
 228                                 return 0;
 229                         }
 230                         if(msg)
 231                                 addr_to_additional(akey, region, *msg, now);
 232                         lock_rw_unlock(&akey->entry.lock);
 233                 } else {
 234                         /* BIT_CD on false because delegpt lookup does
 235                          * not use dns64 translation */
 236                         neg = msg_cache_lookup(env, ns->name, ns->namelen,
 237                                 LDNS_RR_TYPE_A, qclass, 0, now, 0);
 238                         if(neg) {
 239                                 delegpt_add_neg_msg(dp, neg);
 240                                 lock_rw_unlock(&neg->entry.lock);
 241                         }
 242                 }
 243                 akey = rrset_cache_lookup(env->rrset_cache, ns->name,
 244                         ns->namelen, LDNS_RR_TYPE_AAAA, qclass, 0, now, 0);
 245                 if(akey) {
 246                         if(!delegpt_add_rrset_AAAA(dp, region, akey, 0)) {
 247                                 lock_rw_unlock(&akey->entry.lock);
 248                                 return 0;
 249                         }
 250                         if(msg)
 251                                 addr_to_additional(akey, region, *msg, now);
 252                         lock_rw_unlock(&akey->entry.lock);
 253                 } else {
 254                         /* BIT_CD on false because delegpt lookup does
 255                          * not use dns64 translation */
 256                         neg = msg_cache_lookup(env, ns->name, ns->namelen,
 257                                 LDNS_RR_TYPE_AAAA, qclass, 0, now, 0);
 258                         if(neg) {
 259                                 delegpt_add_neg_msg(dp, neg);
 260                                 lock_rw_unlock(&neg->entry.lock);
 261                         }
 262                 }
 263         }
 264         return 1;
 265 }
 266
 267 /** find and add A and AAAA records for missing nameservers in delegpt */
 268 int
 269 cache_fill_missing(struct module_env* env, uint16_t qclass,
 270         struct regional* region, struct delegpt* dp)
 271 {
 272         struct delegpt_ns* ns;
 273         struct msgreply_entry* neg;
 274         struct ub_packed_rrset_key* akey;
 275         time_t now = *env->now;
 276         for(ns = dp->nslist; ns; ns = ns->next) {
 277                 akey = rrset_cache_lookup(env->rrset_cache, ns->name,
 278                         ns->namelen, LDNS_RR_TYPE_A, qclass, 0, now, 0);
 279                 if(akey) {
 280                         if(!delegpt_add_rrset_A(dp, region, akey, ns->lame)) {
 281                                 lock_rw_unlock(&akey->entry.lock);
 282                                 return 0;
 283                         }
 284                         log_nametypeclass(VERB_ALGO, "found in cache",
 285                                 ns->name, LDNS_RR_TYPE_A, qclass);
 286                         lock_rw_unlock(&akey->entry.lock);
 287                 } else {
 288                         /* BIT_CD on false because delegpt lookup does
 289                          * not use dns64 translation */
 290                         neg = msg_cache_lookup(env, ns->name, ns->namelen,
 291                                 LDNS_RR_TYPE_A, qclass, 0, now, 0);
 292                         if(neg) {
 293                                 delegpt_add_neg_msg(dp, neg);
 294                                 lock_rw_unlock(&neg->entry.lock);
 295                         }
 296                 }
 297                 akey = rrset_cache_lookup(env->rrset_cache, ns->name,
 298                         ns->namelen, LDNS_RR_TYPE_AAAA, qclass, 0, now, 0);
 299                 if(akey) {
 300                         if(!delegpt_add_rrset_AAAA(dp, region, akey, ns->lame)) {
 301                                 lock_rw_unlock(&akey->entry.lock);
 302                                 return 0;
 303                         }
 304                         log_nametypeclass(VERB_ALGO, "found in cache",
 305                                 ns->name, LDNS_RR_TYPE_AAAA, qclass);
 306                         lock_rw_unlock(&akey->entry.lock);
 307                 } else {
 308                         /* BIT_CD on false because delegpt lookup does
 309                          * not use dns64 translation */
 310                         neg = msg_cache_lookup(env, ns->name, ns->namelen,
 311                                 LDNS_RR_TYPE_AAAA, qclass, 0, now, 0);
 312                         if(neg) {
 313                                 delegpt_add_neg_msg(dp, neg);
 314                                 lock_rw_unlock(&neg->entry.lock);
 315                         }
 316                 }
 317         }
 318         return 1;
 319 }
 320
 321 /** find and add DS or NSEC to delegation msg */
 322 static void
 323 find_add_ds(struct module_env* env, struct regional* region,
 324         struct dns_msg* msg, struct delegpt* dp, time_t now)
 325 {
 326         /* Lookup the DS or NSEC at the delegation point. */
 327         struct ub_packed_rrset_key* rrset = rrset_cache_lookup(
 328                 env->rrset_cache, dp->name, dp->namelen, LDNS_RR_TYPE_DS,
 329                 msg->qinfo.qclass, 0, now, 0);
 330         if(!rrset) {
 331                 /* NOTE: this won't work for alternate NSEC schemes
 332                  *      (opt-in, NSEC3) */
 333                 rrset = rrset_cache_lookup(env->rrset_cache, dp->name,
 334                         dp->namelen, LDNS_RR_TYPE_NSEC, msg->qinfo.qclass,
 335                         0, now, 0);
 336                 /* Note: the PACKED_RRSET_NSEC_AT_APEX flag is not used.
 337                  * since this is a referral, we need the NSEC at the parent
 338                  * side of the zone cut, not the NSEC at apex side. */
 339                 if(rrset && nsec_has_type(rrset, LDNS_RR_TYPE_DS)) {
 340                         lock_rw_unlock(&rrset->entry.lock);
 341                         rrset = NULL; /* discard wrong NSEC */
 342                 }
 343         }
 344         if(rrset) {
 345                 /* add it to auth section. This is the second rrset. */
 346                 if((msg->rep->rrsets[msg->rep->rrset_count] =
 347                         packed_rrset_copy_region(rrset, region, now))) {
 348                         msg->rep->ns_numrrsets++;
 349                         msg->rep->rrset_count++;
 350                 }
 351                 lock_rw_unlock(&rrset->entry.lock);
 352         }
 353 }
 354
 355 struct dns_msg*
 356 dns_msg_create(uint8_t* qname, size_t qnamelen, uint16_t qtype,
 357         uint16_t qclass, struct regional* region, size_t capacity)
 358 {
 359         struct dns_msg* msg = (struct dns_msg*)regional_alloc(region,
 360                 sizeof(struct dns_msg));
 361         if(!msg)
 362                 return NULL;
 363         msg->qinfo.qname = regional_alloc_init(region, qname, qnamelen);
 364         if(!msg->qinfo.qname)
 365                 return NULL;
 366         msg->qinfo.qname_len = qnamelen;
 367         msg->qinfo.qtype = qtype;
 368         msg->qinfo.qclass = qclass;
 369         msg->qinfo.local_alias = NULL;
 370         /* non-packed reply_info, because it needs to grow the array */
 371         msg->rep = (struct reply_info*)regional_alloc_zero(region,
 372                 sizeof(struct reply_info)-sizeof(struct rrset_ref));
 373         if(!msg->rep)
 374                 return NULL;
 375         if(capacity > RR_COUNT_MAX)
 376                 return NULL; /* integer overflow protection */
 377         msg->rep->flags = BIT_QR; /* with QR, no AA */
 378         msg->rep->qdcount = 1;
 379         msg->rep->rrsets = (struct ub_packed_rrset_key**)
 380                 regional_alloc(region,
 381                 capacity*sizeof(struct ub_packed_rrset_key*));
 382         if(!msg->rep->rrsets)
 383                 return NULL;
 384         return msg;
 385 }
 386
 387 int
 388 dns_msg_authadd(struct dns_msg* msg, struct regional* region,
 389         struct ub_packed_rrset_key* rrset, time_t now)
 390 {
 391         if(!(msg->rep->rrsets[msg->rep->rrset_count++] =
 392                 packed_rrset_copy_region(rrset, region, now)))
 393                 return 0;
 394         msg->rep->ns_numrrsets++;
 395         return 1;
 396 }
 397
 398 int
 399 dns_msg_ansadd(struct dns_msg* msg, struct regional* region,
 400         struct ub_packed_rrset_key* rrset, time_t now)
 401 {
 402         if(!(msg->rep->rrsets[msg->rep->rrset_count++] =
 403                 packed_rrset_copy_region(rrset, region, now)))
 404                 return 0;
 405         msg->rep->an_numrrsets++;
 406         return 1;
 407 }
 408
 409 struct delegpt*
 410 dns_cache_find_delegation(struct module_env* env, uint8_t* qname,
 411         size_t qnamelen, uint16_t qtype, uint16_t qclass,
 412         struct regional* region, struct dns_msg** msg, time_t now)
 413 {
 414         /* try to find closest NS rrset */
 415         struct ub_packed_rrset_key* nskey;
 416         struct packed_rrset_data* nsdata;
 417         struct delegpt* dp;
 418
 419         nskey = find_closest_of_type(env, qname, qnamelen, qclass, now,
 420                 LDNS_RR_TYPE_NS, 0);
 421         if(!nskey) /* hope the caller has hints to prime or something */
 422                 return NULL;
 423         nsdata = (struct packed_rrset_data*)nskey->entry.data;
 424         /* got the NS key, create delegation point */
 425         dp = delegpt_create(region);
 426         if(!dp || !delegpt_set_name(dp, region, nskey->rk.dname)) {
 427                 lock_rw_unlock(&nskey->entry.lock);
 428                 log_err("find_delegation: out of memory");
 429                 return NULL;
 430         }
 431         /* create referral message */
 432         if(msg) {
 433                 /* allocate the array to as much as we could need:
 434                  *      NS rrset + DS/NSEC rrset +
 435                  *      A rrset for every NS RR
 436                  *      AAAA rrset for every NS RR
 437                  */
 438                 *msg = dns_msg_create(qname, qnamelen, qtype, qclass, region,
 439                         2 + nsdata->count*2);
 440                 if(!*msg || !dns_msg_authadd(*msg, region, nskey, now)) {
 441                         lock_rw_unlock(&nskey->entry.lock);
 442                         log_err("find_delegation: out of memory");
 443                         return NULL;
 444                 }
 445         }
 446         if(!delegpt_rrset_add_ns(dp, region, nskey, 0))
 447                 log_err("find_delegation: addns out of memory");
 448         lock_rw_unlock(&nskey->entry.lock); /* first unlock before next lookup*/
 449         /* find and add DS/NSEC (if any) */
 450         if(msg)
 451                 find_add_ds(env, region, *msg, dp, now);
 452         /* find and add A entries */
 453         if(!find_add_addrs(env, qclass, region, dp, now, msg))
 454                 log_err("find_delegation: addrs out of memory");
 455         return dp;
 456 }
 457
 458 /** allocate dns_msg from query_info and reply_info */
 459 static struct dns_msg*
 460 gen_dns_msg(struct regional* region, struct query_info* q, size_t num)
 461 {
 462         struct dns_msg* msg = (struct dns_msg*)regional_alloc(region,
 463                 sizeof(struct dns_msg));
 464         if(!msg)
 465                 return NULL;
 466         memcpy(&msg->qinfo, q, sizeof(struct query_info));
 467         msg->qinfo.qname = regional_alloc_init(region, q->qname, q->qname_len);
 468         if(!msg->qinfo.qname)
 469                 return NULL;
 470         /* allocate replyinfo struct and rrset key array separately */
 471         msg->rep = (struct reply_info*)regional_alloc(region,
 472                 sizeof(struct reply_info) - sizeof(struct rrset_ref));
 473         if(!msg->rep)
 474                 return NULL;
 475         if(num > RR_COUNT_MAX)
 476                 return NULL; /* integer overflow protection */
 477         msg->rep->rrsets = (struct ub_packed_rrset_key**)
 478                 regional_alloc(region,
 479                 num * sizeof(struct ub_packed_rrset_key*));
 480         if(!msg->rep->rrsets)
 481                 return NULL;
 482         return msg;
 483 }
 484
 485 struct dns_msg*
 486 tomsg(struct module_env* env, struct query_info* q, struct reply_info* r,
 487         struct regional* region, time_t now, struct regional* scratch)
 488 {
 489         struct dns_msg* msg;
 490         size_t i;
 491         if(now > r->ttl)
 492                 return NULL;
 493         msg = gen_dns_msg(region, q, r->rrset_count);
 494         if(!msg)
 495                 return NULL;
 496         msg->rep->flags = r->flags;
 497         msg->rep->qdcount = r->qdcount;
 498         msg->rep->ttl = r->ttl - now;
 499         if(r->prefetch_ttl > now)
 500                 msg->rep->prefetch_ttl = r->prefetch_ttl - now;
 501         else    msg->rep->prefetch_ttl = PREFETCH_TTL_CALC(msg->rep->ttl);
 502         msg->rep->security = r->security;
 503         msg->rep->an_numrrsets = r->an_numrrsets;
 504         msg->rep->ns_numrrsets = r->ns_numrrsets;
 505         msg->rep->ar_numrrsets = r->ar_numrrsets;
 506         msg->rep->rrset_count = r->rrset_count;
 507         msg->rep->authoritative = r->authoritative;
 508         if(!rrset_array_lock(r->ref, r->rrset_count, now))
 509                 return NULL;
 510         if(r->an_numrrsets > 0 && (r->rrsets[0]->rk.type == htons(
 511                 LDNS_RR_TYPE_CNAME) || r->rrsets[0]->rk.type == htons(
 512                 LDNS_RR_TYPE_DNAME)) && !reply_check_cname_chain(q, r)) {
 513                 /* cname chain is now invalid, reconstruct msg */
 514                 rrset_array_unlock(r->ref, r->rrset_count);
 515                 return NULL;
 516         }
 517         if(r->security == sec_status_secure && !reply_all_rrsets_secure(r)) {
 518                 /* message rrsets have changed status, revalidate */
 519                 rrset_array_unlock(r->ref, r->rrset_count);
 520                 return NULL;
 521         }
 522         for(i=0; i<msg->rep->rrset_count; i++) {
 523                 msg->rep->rrsets[i] = packed_rrset_copy_region(r->rrsets[i],
 524                         region, now);
 525                 if(!msg->rep->rrsets[i]) {
 526                         rrset_array_unlock(r->ref, r->rrset_count);
 527                         return NULL;
 528                 }
 529         }
 530         if(env)
 531                 rrset_array_unlock_touch(env->rrset_cache, scratch, r->ref,
 532                 r->rrset_count);
 533         else
 534                 rrset_array_unlock(r->ref, r->rrset_count);
 535         return msg;
 536 }
 537
 538 /** synthesize RRset-only response from cached RRset item */
 539 static struct dns_msg*
 540 rrset_msg(struct ub_packed_rrset_key* rrset, struct regional* region,
 541         time_t now, struct query_info* q)
 542 {
 543         struct dns_msg* msg;
 544         struct packed_rrset_data* d = (struct packed_rrset_data*)
 545                 rrset->entry.data;
 546         if(now > d->ttl)
 547                 return NULL;
 548         msg = gen_dns_msg(region, q, 1); /* only the CNAME (or other) RRset */
 549         if(!msg)
 550                 return NULL;
 551         msg->rep->flags = BIT_QR; /* reply, no AA, no error */
 552         msg->rep->authoritative = 0; /* reply stored in cache can't be authoritative */
 553         msg->rep->qdcount = 1;
 554         msg->rep->ttl = d->ttl - now;
 555         msg->rep->prefetch_ttl = PREFETCH_TTL_CALC(msg->rep->ttl);
 556         msg->rep->security = sec_status_unchecked;
 557         msg->rep->an_numrrsets = 1;
 558         msg->rep->ns_numrrsets = 0;
 559         msg->rep->ar_numrrsets = 0;
 560         msg->rep->rrset_count = 1;
 561         msg->rep->rrsets[0] = packed_rrset_copy_region(rrset, region, now);
 562         if(!msg->rep->rrsets[0]) /* copy CNAME */
 563                 return NULL;
 564         return msg;
 565 }
 566
 567 /** synthesize DNAME+CNAME response from cached DNAME item */
 568 static struct dns_msg*
 569 synth_dname_msg(struct ub_packed_rrset_key* rrset, struct regional* region,
 570         time_t now, struct query_info* q, enum sec_status* sec_status)
 571 {
 572         struct dns_msg* msg;
 573         struct ub_packed_rrset_key* ck;
 574         struct packed_rrset_data* newd, *d = (struct packed_rrset_data*)
 575                 rrset->entry.data;
 576         uint8_t* newname, *dtarg = NULL;
 577         size_t newlen, dtarglen;
 578         if(now > d->ttl)
 579                 return NULL;
 580         /* only allow validated (with DNSSEC) DNAMEs used from cache
 581          * for insecure DNAMEs, query again. */
 582         *sec_status = d->security;
 583         /* return sec status, so the status of the CNAME can be checked
 584          * by the calling routine. */
 585         msg = gen_dns_msg(region, q, 2); /* DNAME + CNAME RRset */
 586         if(!msg)
 587                 return NULL;
 588         msg->rep->flags = BIT_QR; /* reply, no AA, no error */
 589         msg->rep->authoritative = 0; /* reply stored in cache can't be authoritative */
 590         msg->rep->qdcount = 1;
 591         msg->rep->ttl = d->ttl - now;
 592         msg->rep->prefetch_ttl = PREFETCH_TTL_CALC(msg->rep->ttl);
 593         msg->rep->security = sec_status_unchecked;
 594         msg->rep->an_numrrsets = 1;
 595         msg->rep->ns_numrrsets = 0;
 596         msg->rep->ar_numrrsets = 0;
 597         msg->rep->rrset_count = 1;
 598         msg->rep->rrsets[0] = packed_rrset_copy_region(rrset, region, now);
 599         if(!msg->rep->rrsets[0]) /* copy DNAME */
 600                 return NULL;
 601         /* synth CNAME rrset */
 602         get_cname_target(rrset, &dtarg, &dtarglen);
 603         if(!dtarg)
 604                 return NULL;
 605         newlen = q->qname_len + dtarglen - rrset->rk.dname_len;
 606         if(newlen > LDNS_MAX_DOMAINLEN) {
 607                 msg->rep->flags |= LDNS_RCODE_YXDOMAIN;
 608                 return msg;
 609         }
 610         newname = (uint8_t*)regional_alloc(region, newlen);
 611         if(!newname)
 612                 return NULL;
 613         /* new name is concatenation of qname front (without DNAME owner)
 614          * and DNAME target name */
 615         memcpy(newname, q->qname, q->qname_len-rrset->rk.dname_len);
 616         memmove(newname+(q->qname_len-rrset->rk.dname_len), dtarg, dtarglen);
 617         /* create rest of CNAME rrset */
 618         ck = (struct ub_packed_rrset_key*)regional_alloc(region,
 619                 sizeof(struct ub_packed_rrset_key));
 620         if(!ck)
 621                 return NULL;
 622         memset(&ck->entry, 0, sizeof(ck->entry));
 623         msg->rep->rrsets[1] = ck;
 624         ck->entry.key = ck;
 625         ck->rk.type = htons(LDNS_RR_TYPE_CNAME);
 626         ck->rk.rrset_class = rrset->rk.rrset_class;
 627         ck->rk.flags = 0;
 628         ck->rk.dname = regional_alloc_init(region, q->qname, q->qname_len);
 629         if(!ck->rk.dname)
 630                 return NULL;
 631         ck->rk.dname_len = q->qname_len;
 632         ck->entry.hash = rrset_key_hash(&ck->rk);
 633         newd = (struct packed_rrset_data*)regional_alloc_zero(region,
 634                 sizeof(struct packed_rrset_data) + sizeof(size_t) +
 635                 sizeof(uint8_t*) + sizeof(time_t) + sizeof(uint16_t)
 636                 + newlen);
 637         if(!newd)
 638                 return NULL;
 639         ck->entry.data = newd;
 640         newd->ttl = 0; /* 0 for synthesized CNAME TTL */
 641         newd->count = 1;
 642         newd->rrsig_count = 0;
 643         newd->trust = rrset_trust_ans_noAA;
 644         newd->rr_len = (size_t*)((uint8_t*)newd +
 645                 sizeof(struct packed_rrset_data));
 646         newd->rr_len[0] = newlen + sizeof(uint16_t);
 647         packed_rrset_ptr_fixup(newd);
 648         newd->rr_ttl[0] = newd->ttl;
 649         msg->rep->ttl = newd->ttl;
 650         msg->rep->prefetch_ttl = PREFETCH_TTL_CALC(newd->ttl);
 651         sldns_write_uint16(newd->rr_data[0], newlen);
 652         memmove(newd->rr_data[0] + sizeof(uint16_t), newname, newlen);
 653         msg->rep->an_numrrsets ++;
 654         msg->rep->rrset_count ++;
 655         return msg;
 656 }
 657
 658 /** Fill TYPE_ANY response with some data from cache */
 659 static struct dns_msg*
 660 fill_any(struct module_env* env,
 661         uint8_t* qname, size_t qnamelen, uint16_t qtype, uint16_t qclass,
 662         struct regional* region)
 663 {
 664         time_t now = *env->now;
 665         struct dns_msg* msg = NULL;
 666         uint16_t lookup[] = {LDNS_RR_TYPE_A, LDNS_RR_TYPE_AAAA,
 667                 LDNS_RR_TYPE_MX, LDNS_RR_TYPE_SOA, LDNS_RR_TYPE_NS,
 668                 LDNS_RR_TYPE_DNAME, 0};
 669         int i, num=6; /* number of RR types to look up */
 670         log_assert(lookup[num] == 0);
 671
 672         for(i=0; i<num; i++) {
 673                 /* look up this RR for inclusion in type ANY response */
 674                 struct ub_packed_rrset_key* rrset = rrset_cache_lookup(
 675                         env->rrset_cache, qname, qnamelen, lookup[i],
 676                         qclass, 0, now, 0);
 677                 struct packed_rrset_data *d;
 678                 if(!rrset)
 679                         continue;
 680
 681                 /* only if rrset from answer section */
 682                 d = (struct packed_rrset_data*)rrset->entry.data;
 683                 if(d->trust == rrset_trust_add_noAA ||
 684                         d->trust == rrset_trust_auth_noAA ||
 685                         d->trust == rrset_trust_add_AA ||
 686                         d->trust == rrset_trust_auth_AA) {
 687                         lock_rw_unlock(&rrset->entry.lock);
 688                         continue;
 689                 }
 690
 691                 /* create msg if none */
 692                 if(!msg) {
 693                         msg = dns_msg_create(qname, qnamelen, qtype, qclass,
 694                                 region, (size_t)(num-i));
 695                         if(!msg) {
 696                                 lock_rw_unlock(&rrset->entry.lock);
 697                                 return NULL;
 698                         }
 699                 }
 700
 701                 /* add RRset to response */
 702                 if(!dns_msg_ansadd(msg, region, rrset, now)) {
 703                         lock_rw_unlock(&rrset->entry.lock);
 704                         return NULL;
 705                 }
 706                 lock_rw_unlock(&rrset->entry.lock);
 707         }
 708         return msg;
 709 }
 710
 711 struct dns_msg*
 712 dns_cache_lookup(struct module_env* env,
 713         uint8_t* qname, size_t qnamelen, uint16_t qtype, uint16_t qclass,
 714         uint16_t flags, struct regional* region, struct regional* scratch,
 715         int no_partial)
 716 {
 717         struct lruhash_entry* e;
 718         struct query_info k;
 719         hashvalue_type h;
 720         time_t now = *env->now;
 721         struct ub_packed_rrset_key* rrset;
 722
 723         /* lookup first, this has both NXdomains and ANSWER responses */
 724         k.qname = qname;
 725         k.qname_len = qnamelen;
 726         k.qtype = qtype;
 727         k.qclass = qclass;
 728         k.local_alias = NULL;
 729         h = query_info_hash(&k, flags);
 730         e = slabhash_lookup(env->msg_cache, h, &k, 0);
 731         if(e) {
 732                 struct msgreply_entry* key = (struct msgreply_entry*)e->key;
 733                 struct reply_info* data = (struct reply_info*)e->data;
 734                 struct dns_msg* msg = tomsg(env, &key->key, data, region, now,
 735                         scratch);
 736                 if(msg) {
 737                         lock_rw_unlock(&e->lock);
 738                         return msg;
 739                 }
 740                 /* could be msg==NULL; due to TTL or not all rrsets available */
 741                 lock_rw_unlock(&e->lock);
 742         }
 743
 744         /* see if a DNAME exists. Checked for first, to enforce that DNAMEs
 745          * are more important, the CNAME is resynthesized and thus
 746          * consistent with the DNAME */
 747         if(!no_partial &&
 748                 (rrset=find_closest_of_type(env, qname, qnamelen, qclass, now,
 749                 LDNS_RR_TYPE_DNAME, 1))) {
 750                 /* synthesize a DNAME+CNAME message based on this */
 751                 enum sec_status sec_status = sec_status_unchecked;
 752                 struct dns_msg* msg = synth_dname_msg(rrset, region, now, &k,
 753                         &sec_status);
 754                 if(msg) {
 755                         struct ub_packed_rrset_key* cname_rrset;
 756                         lock_rw_unlock(&rrset->entry.lock);
 757                         /* now, after unlocking the DNAME rrset lock,
 758                          * check the sec_status, and see if we need to look
 759                          * up the CNAME record associated before it can
 760                          * be used */
 761                         /* normally, only secure DNAMEs allowed from cache*/
 762                         if(sec_status == sec_status_secure)
 763                                 return msg;
 764                         /* but if we have a CNAME cached with this name, then we
 765                          * have previously already allowed this name to pass.
 766                          * the next cache lookup is going to fetch that CNAME itself,
 767                          * but it is better to have the (unsigned)DNAME + CNAME in
 768                          * that case */
 769                         cname_rrset = rrset_cache_lookup(
 770                                 env->rrset_cache, qname, qnamelen,
 771                                 LDNS_RR_TYPE_CNAME, qclass, 0, now, 0);
 772                         if(cname_rrset) {
 773                                 /* CNAME already synthesized by
 774                                  * synth_dname_msg routine, so we can
 775                                  * straight up return the msg */
 776                                 lock_rw_unlock(&cname_rrset->entry.lock);
 777                                 return msg;
 778                         }
 779                 } else {
 780                         lock_rw_unlock(&rrset->entry.lock);
 781                 }
 782         }
 783
 784         /* see if we have CNAME for this domain,
 785          * but not for DS records (which are part of the parent) */
 786         if(!no_partial && qtype != LDNS_RR_TYPE_DS &&
 787            (rrset=rrset_cache_lookup(env->rrset_cache, qname, qnamelen,
 788                 LDNS_RR_TYPE_CNAME, qclass, 0, now, 0))) {
 789                 uint8_t* wc = NULL;
 790                 size_t wl;
 791                 /* if the rrset is not a wildcard expansion, with wcname */
 792                 /* because, if we return that CNAME rrset on its own, it is
 793                  * missing the NSEC or NSEC3 proof */
 794                 if(!(val_rrset_wildcard(rrset, &wc, &wl) && wc != NULL)) {
 795                         struct dns_msg* msg = rrset_msg(rrset, region, now, &k);
 796                         if(msg) {
 797                                 lock_rw_unlock(&rrset->entry.lock);
 798                                 return msg;
 799                         }
 800                 }
 801                 lock_rw_unlock(&rrset->entry.lock);
 802         }
 803
 804         /* construct DS, DNSKEY, DLV messages from rrset cache. */
 805         if((qtype == LDNS_RR_TYPE_DS || qtype == LDNS_RR_TYPE_DNSKEY ||
 806                 qtype == LDNS_RR_TYPE_DLV) &&
 807                 (rrset=rrset_cache_lookup(env->rrset_cache, qname, qnamelen,
 808                 qtype, qclass, 0, now, 0))) {
 809                 /* if the rrset is from the additional section, and the
 810                  * signatures have fallen off, then do not synthesize a msg
 811                  * instead, allow a full query for signed results to happen.
 812                  * Forego all rrset data from additional section, because
 813                  * some signatures may not be present and cause validation
 814                  * failure.
 815                  */
 816                 struct packed_rrset_data *d = (struct packed_rrset_data*)
 817                         rrset->entry.data;
 818                 if(d->trust != rrset_trust_add_noAA &&
 819                         d->trust != rrset_trust_add_AA &&
 820                         (qtype == LDNS_RR_TYPE_DS ||
 821                                 (d->trust != rrset_trust_auth_noAA
 822                                 && d->trust != rrset_trust_auth_AA) )) {
 823                         struct dns_msg* msg = rrset_msg(rrset, region, now, &k);
 824                         if(msg) {
 825                                 lock_rw_unlock(&rrset->entry.lock);
 826                                 return msg;
 827                         }
 828                 }
 829                 lock_rw_unlock(&rrset->entry.lock);
 830         }
 831
 832         /* stop downwards cache search on NXDOMAIN.
 833          * Empty nonterminals are NOERROR, so an NXDOMAIN for foo
 834          * means bla.foo also does not exist.  The DNSSEC proofs are
 835          * the same.  We search upwards for NXDOMAINs. */
 836         if(env->cfg->harden_below_nxdomain)
 837             while(!dname_is_root(k.qname)) {
 838                 dname_remove_label(&k.qname, &k.qname_len);
 839                 h = query_info_hash(&k, flags);
 840                 e = slabhash_lookup(env->msg_cache, h, &k, 0);
 841                 if(!e && k.qtype != LDNS_RR_TYPE_A &&
 842                         env->cfg->qname_minimisation) {
 843                         k.qtype = LDNS_RR_TYPE_A;
 844                         h = query_info_hash(&k, flags);
 845                         e = slabhash_lookup(env->msg_cache, h, &k, 0);
 846                 }
 847                 if(e) {
 848                         struct reply_info* data = (struct reply_info*)e->data;
 849                         struct dns_msg* msg;
 850                         if(FLAGS_GET_RCODE(data->flags) == LDNS_RCODE_NXDOMAIN
 851                           && data->security == sec_status_secure
 852                           && (msg=tomsg(env, &k, data, region, now, scratch))){
 853                                 lock_rw_unlock(&e->lock);
 854                                 msg->qinfo.qname=qname;
 855                                 msg->qinfo.qname_len=qnamelen;
 856                                 /* check that DNSSEC really works out */
 857                                 msg->rep->security = sec_status_unchecked;
 858                                 return msg;
 859                         }
 860                         lock_rw_unlock(&e->lock);
 861                 }
 862                 k.qtype = qtype;
 863             }
 864
 865         /* fill common RR types for ANY response to avoid requery */
 866         if(qtype == LDNS_RR_TYPE_ANY) {
 867                 return fill_any(env, qname, qnamelen, qtype, qclass, region);
 868         }
 869
 870         return NULL;
 871 }
 872
 873 int
 874 dns_cache_store(struct module_env* env, struct query_info* msgqinf,
 875         struct reply_info* msgrep, int is_referral, time_t leeway, int pside,
 876         struct regional* region, uint32_t flags)
 877 {
 878         struct reply_info* rep = NULL;
 879         /* alloc, malloc properly (not in region, like msg is) */
 880         rep = reply_info_copy(msgrep, env->alloc, NULL);
 881         if(!rep)
 882                 return 0;
 883         /* ttl must be relative ;i.e. 0..86400 not  time(0)+86400.
 884          * the env->now is added to message and RRsets in this routine. */
 885         /* the leeway is used to invalidate other rrsets earlier */
 886
 887         if(is_referral) {
 888                 /* store rrsets */
 889                 struct rrset_ref ref;
 890                 size_t i;
 891                 for(i=0; i<rep->rrset_count; i++) {
 892                         packed_rrset_ttl_add((struct packed_rrset_data*)
 893                                 rep->rrsets[i]->entry.data, *env->now);
 894                         ref.key = rep->rrsets[i];
 895                         ref.id = rep->rrsets[i]->id;
 896                         /*ignore ret: it was in the cache, ref updated */
 897                         /* no leeway for typeNS */
 898                         (void)rrset_cache_update(env->rrset_cache, &ref,
 899                                 env->alloc, *env->now +
 900                                 ((ntohs(ref.key->rk.type)==LDNS_RR_TYPE_NS
 901                                  && !pside) ? 0:leeway));
 902                 }
 903                 free(rep);
 904                 return 1;
 905         } else {
 906                 /* store msg, and rrsets */
 907                 struct query_info qinf;
 908                 hashvalue_type h;
 909
 910                 qinf = *msgqinf;
 911                 qinf.qname = memdup(msgqinf->qname, msgqinf->qname_len);
 912                 if(!qinf.qname) {
 913                         reply_info_parsedelete(rep, env->alloc);
 914                         return 0;
 915                 }
 916                 /* fixup flags to be sensible for a reply based on the cache */
 917                 /* this module means that RA is available. It is an answer QR.
 918                  * Not AA from cache. Not CD in cache (depends on client bit). */
 919                 rep->flags |= (BIT_RA | BIT_QR);
 920                 rep->flags &= ~(BIT_AA | BIT_CD);
 921                 h = query_info_hash(&qinf, (uint16_t)flags);
 922                 dns_cache_store_msg(env, &qinf, h, rep, leeway, pside, msgrep,
 923                         flags, region);
 924                 /* qname is used inside query_info_entrysetup, and set to
 925                  * NULL. If it has not been used, free it. free(0) is safe. */
 926                 free(qinf.qname);
 927         }
 928         return 1;
 929 }
 930
 931 int
 932 dns_cache_prefetch_adjust(struct module_env* env, struct query_info* qinfo,
 933         time_t adjust, uint16_t flags)
 934 {
 935         struct msgreply_entry* msg;
 936         msg = msg_cache_lookup(env, qinfo->qname, qinfo->qname_len,
 937                 qinfo->qtype, qinfo->qclass, flags, *env->now, 1);
 938         if(msg) {
 939                 struct reply_info* rep = (struct reply_info*)msg->entry.data;
 940                 if(rep) {
 941                         rep->prefetch_ttl += adjust;
 942                         lock_rw_unlock(&msg->entry.lock);
 943                         return 1;
 944                 }
 945                 lock_rw_unlock(&msg->entry.lock);
 946         }
 947         return 0;
 948 }