]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/blob - contrib/unbound/edns-subnet/subnetmod.c
Fix multiple vulnerabilities in unbound.
[FreeBSD/FreeBSD.git] / contrib / unbound / edns-subnet / subnetmod.c
1 /*
2  * edns-subnet/subnetmod.c - edns subnet module. Must be called before validator
3  * and iterator.
4  *
5  * Copyright (c) 2013, NLnet Labs. All rights reserved.
6  *
7  * This software is open source.
8  * 
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 
13  * Redistributions of source code must retain the above copyright notice,
14  * this list of conditions and the following disclaimer.
15  * 
16  * Redistributions in binary form must reproduce the above copyright notice,
17  * this list of conditions and the following disclaimer in the documentation
18  * and/or other materials provided with the distribution.
19  * 
20  * Neither the name of the NLNET LABS nor the names of its contributors may
21  * be used to endorse or promote products derived from this software without
22  * specific prior written permission.
23  * 
24  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
25  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
26  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
27  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
28  * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
29  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
30  * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
31  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
32  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
33  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
34  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35  */
36  /**
37  * \file
38  * subnet module for unbound.
39  */
40
41 #include "config.h"
42
43 #ifdef CLIENT_SUBNET /* keeps splint happy */
44
45 #include "edns-subnet/subnetmod.h"
46 #include "edns-subnet/edns-subnet.h"
47 #include "edns-subnet/addrtree.h"
48 #include "edns-subnet/subnet-whitelist.h"
49
50 #include "services/mesh.h"
51 #include "services/cache/dns.h"
52 #include "util/module.h"
53 #include "util/regional.h"
54 #include "util/storage/slabhash.h"
55 #include "util/config_file.h"
56 #include "util/data/msgreply.h"
57 #include "sldns/sbuffer.h"
58 #include "iterator/iter_utils.h"
59
60 /** externally called */
61 void 
62 subnet_data_delete(void *d, void *ATTR_UNUSED(arg))
63 {
64         struct subnet_msg_cache_data *r;
65         r = (struct subnet_msg_cache_data*)d;
66         addrtree_delete(r->tree4);
67         addrtree_delete(r->tree6);
68         free(r);
69 }
70
71 /** externally called */
72 size_t 
73 msg_cache_sizefunc(void *k, void *d)
74 {
75         struct msgreply_entry *q = (struct msgreply_entry*)k;
76         struct subnet_msg_cache_data *r = (struct subnet_msg_cache_data*)d;
77         size_t s = sizeof(struct msgreply_entry) 
78                 + sizeof(struct subnet_msg_cache_data)
79                 + q->key.qname_len + lock_get_mem(&q->entry.lock);
80         s += addrtree_size(r->tree4);
81         s += addrtree_size(r->tree6);
82         return s;
83 }
84
85 /** new query for ecs module */
86 static int
87 subnet_new_qstate(struct module_qstate *qstate, int id)
88 {
89         struct subnet_qstate *sq = (struct subnet_qstate*)regional_alloc(
90                 qstate->region, sizeof(struct subnet_qstate));
91         if(!sq) 
92                 return 0;
93         qstate->minfo[id] = sq;
94         memset(sq, 0, sizeof(*sq));
95         sq->started_no_cache_store = qstate->no_cache_store;
96         return 1;
97 }
98
99 /** Add ecs struct to edns list, after parsing it to wire format. */
100 static void
101 ecs_opt_list_append(struct ecs_data* ecs, struct edns_option** list,
102         struct module_qstate *qstate)
103 {
104         size_t sn_octs, sn_octs_remainder;
105         sldns_buffer* buf = qstate->env->scratch_buffer;
106
107         if(ecs->subnet_validdata) {
108                 log_assert(ecs->subnet_addr_fam == EDNSSUBNET_ADDRFAM_IP4 || 
109                         ecs->subnet_addr_fam == EDNSSUBNET_ADDRFAM_IP6);
110                 log_assert(ecs->subnet_addr_fam != EDNSSUBNET_ADDRFAM_IP4 || 
111                         ecs->subnet_source_mask <=  INET_SIZE*8);
112                 log_assert(ecs->subnet_addr_fam != EDNSSUBNET_ADDRFAM_IP6 || 
113                         ecs->subnet_source_mask <= INET6_SIZE*8);
114
115                 sn_octs = ecs->subnet_source_mask / 8;
116                 sn_octs_remainder =
117                         (size_t)((ecs->subnet_source_mask % 8)>0?1:0);
118                 
119                 log_assert(sn_octs + sn_octs_remainder <= INET6_SIZE);
120                 
121                 sldns_buffer_clear(buf);
122                 sldns_buffer_write_u16(buf, ecs->subnet_addr_fam);
123                 sldns_buffer_write_u8(buf, ecs->subnet_source_mask);
124                 sldns_buffer_write_u8(buf, ecs->subnet_scope_mask);
125                 sldns_buffer_write(buf, ecs->subnet_addr, sn_octs);
126                 if(sn_octs_remainder)
127                         sldns_buffer_write_u8(buf, ecs->subnet_addr[sn_octs] & 
128                                 ~(0xFF >> (ecs->subnet_source_mask % 8)));
129                 sldns_buffer_flip(buf);
130
131                 edns_opt_list_append(list,
132                                 qstate->env->cfg->client_subnet_opcode,
133                                 sn_octs + sn_octs_remainder + 4,
134                                 sldns_buffer_begin(buf), qstate->region);
135         }
136 }
137
138 int ecs_whitelist_check(struct query_info* qinfo,
139         uint16_t ATTR_UNUSED(flags), struct module_qstate* qstate,
140         struct sockaddr_storage* addr, socklen_t addrlen,
141         uint8_t* ATTR_UNUSED(zone), size_t ATTR_UNUSED(zonelen),
142         struct regional* ATTR_UNUSED(region), int id, void* ATTR_UNUSED(cbargs))
143 {
144         struct subnet_qstate *sq;
145         struct subnet_env *sn_env;
146         
147         if(!(sq=(struct subnet_qstate*)qstate->minfo[id]))
148                 return 1;
149         sn_env = (struct subnet_env*)qstate->env->modinfo[id];
150
151         /* Cache by default, might be disabled after parsing EDNS option
152          * received from nameserver. */
153         if(!iter_stub_fwd_no_cache(qstate, &qstate->qinfo)) {
154                 qstate->no_cache_store = 0;
155         }
156
157         if(sq->ecs_server_out.subnet_validdata && ((sq->subnet_downstream &&
158                 qstate->env->cfg->client_subnet_always_forward) ||
159                 ecs_is_whitelisted(sn_env->whitelist, 
160                 addr, addrlen, qinfo->qname, qinfo->qname_len,
161                 qinfo->qclass))) {
162                 /* Address on whitelist or client query contains ECS option, we
163                  * want to sent out ECS. Only add option if it is not already
164                  * set. */
165                 if(!(sq->subnet_sent)) {
166                         ecs_opt_list_append(&sq->ecs_server_out,
167                                 &qstate->edns_opts_back_out, qstate);
168                         sq->subnet_sent = 1;
169                 }
170         }
171         else if(sq->subnet_sent) {
172                 /* Outgoing ECS option is set, but we don't want to sent it to
173                  * this address, remove option. */
174                 edns_opt_list_remove(&qstate->edns_opts_back_out,
175                         qstate->env->cfg->client_subnet_opcode);
176                 sq->subnet_sent = 0;
177         }
178         return 1;
179 }
180
181
182 void
183 subnet_markdel(void* key)
184 {
185         struct msgreply_entry *e = (struct msgreply_entry*)key;
186         e->key.qtype = 0;
187         e->key.qclass = 0;
188 }
189
190 int
191 subnetmod_init(struct module_env *env, int id)
192 {
193         struct subnet_env *sn_env = (struct subnet_env*)calloc(1,
194                 sizeof(struct subnet_env));
195         if(!sn_env) {
196                 log_err("malloc failure");
197                 return 0;
198         }
199         alloc_init(&sn_env->alloc, NULL, 0);
200         env->modinfo[id] = (void*)sn_env;
201         /* Copy msg_cache settings */
202         sn_env->subnet_msg_cache = slabhash_create(env->cfg->msg_cache_slabs,
203                 HASH_DEFAULT_STARTARRAY, env->cfg->msg_cache_size,
204                 msg_cache_sizefunc, query_info_compare, query_entry_delete,
205                 subnet_data_delete, NULL);
206         slabhash_setmarkdel(sn_env->subnet_msg_cache, &subnet_markdel);
207         if(!sn_env->subnet_msg_cache) {
208                 log_err("subnet: could not create cache");
209                 free(sn_env);
210                 env->modinfo[id] = NULL;
211                 return 0;
212         }
213         /* whitelist for edns subnet capable servers */
214         sn_env->whitelist = ecs_whitelist_create();
215         if(!sn_env->whitelist ||
216                 !ecs_whitelist_apply_cfg(sn_env->whitelist, env->cfg)) {
217                 log_err("subnet: could not create ECS whitelist");
218                 slabhash_delete(sn_env->subnet_msg_cache);
219                 free(sn_env);
220                 env->modinfo[id] = NULL;
221                 return 0;
222         }
223
224         verbose(VERB_QUERY, "subnet: option registered (%d)",
225                 env->cfg->client_subnet_opcode);
226         /* Create new mesh state for all queries. */
227         env->unique_mesh = 1;
228         if(!edns_register_option(env->cfg->client_subnet_opcode,
229                 env->cfg->client_subnet_always_forward /* bypass cache */,
230                 0 /* no aggregation */, env)) {
231                 log_err("subnet: could not register opcode");
232                 ecs_whitelist_delete(sn_env->whitelist);
233                 slabhash_delete(sn_env->subnet_msg_cache);
234                 free(sn_env);
235                 env->modinfo[id] = NULL;
236                 return 0;
237         }
238         inplace_cb_register((void*)ecs_whitelist_check, inplace_cb_query, NULL,
239                 env, id);
240         inplace_cb_register((void*)ecs_edns_back_parsed,
241                 inplace_cb_edns_back_parsed, NULL, env, id);
242         inplace_cb_register((void*)ecs_query_response,
243                 inplace_cb_query_response, NULL, env, id);
244         lock_rw_init(&sn_env->biglock);
245         return 1;
246 }
247
248 void
249 subnetmod_deinit(struct module_env *env, int id)
250 {
251         struct subnet_env *sn_env;
252         if(!env || !env->modinfo[id])
253                 return;
254         sn_env = (struct subnet_env*)env->modinfo[id];
255         lock_rw_destroy(&sn_env->biglock);
256         inplace_cb_delete(env, inplace_cb_edns_back_parsed, id);
257         inplace_cb_delete(env, inplace_cb_query, id);
258         inplace_cb_delete(env, inplace_cb_query_response, id);
259         ecs_whitelist_delete(sn_env->whitelist);
260         slabhash_delete(sn_env->subnet_msg_cache);
261         alloc_clear(&sn_env->alloc);
262         free(sn_env);
263         env->modinfo[id] = NULL;
264 }
265
266 /** Tells client that upstream has no/improper support */
267 static void
268 cp_edns_bad_response(struct ecs_data *target, struct ecs_data *source)
269 {
270         target->subnet_scope_mask  = 0;
271         target->subnet_source_mask = source->subnet_source_mask;
272         target->subnet_addr_fam    = source->subnet_addr_fam;
273         memcpy(target->subnet_addr, source->subnet_addr, INET6_SIZE);
274         target->subnet_validdata = 1;
275 }
276
277 static void
278 delfunc(void *envptr, void *elemptr) {
279         struct reply_info *elem = (struct reply_info *)elemptr;
280         struct subnet_env *env = (struct subnet_env *)envptr;
281         reply_info_parsedelete(elem, &env->alloc);
282 }
283
284 static size_t
285 sizefunc(void *elemptr) {
286         struct reply_info *elem  = (struct reply_info *)elemptr;
287         return sizeof (struct reply_info) - sizeof (struct rrset_ref)
288                 + elem->rrset_count * sizeof (struct rrset_ref)
289                 + elem->rrset_count * sizeof (struct ub_packed_rrset_key *);
290 }
291
292 /**
293  * Select tree from cache entry based on edns data.
294  * If for address family not present it will create a new one.
295  * NULL on failure to create. */
296 static struct addrtree* 
297 get_tree(struct subnet_msg_cache_data *data, struct ecs_data *edns, 
298         struct subnet_env *env, struct config_file* cfg)
299 {
300         struct addrtree *tree;
301         if (edns->subnet_addr_fam == EDNSSUBNET_ADDRFAM_IP4) {
302                 if (!data->tree4)
303                         data->tree4 = addrtree_create(
304                                 cfg->max_client_subnet_ipv4, &delfunc,
305                                 &sizefunc, env, cfg->max_ecs_tree_size_ipv4);
306                 tree = data->tree4;
307         } else {
308                 if (!data->tree6)
309                         data->tree6 = addrtree_create(
310                                 cfg->max_client_subnet_ipv6, &delfunc,
311                                 &sizefunc, env, cfg->max_ecs_tree_size_ipv6);
312                 tree = data->tree6;
313         }
314         return tree;
315 }
316
317 static void
318 update_cache(struct module_qstate *qstate, int id)
319 {
320         struct msgreply_entry *mrep_entry;
321         struct addrtree *tree;
322         struct reply_info *rep;
323         struct query_info qinf;
324         struct subnet_env *sne = qstate->env->modinfo[id];
325         struct subnet_qstate *sq = (struct subnet_qstate*)qstate->minfo[id];
326         struct slabhash *subnet_msg_cache = sne->subnet_msg_cache;
327         struct ecs_data *edns = &sq->ecs_client_in;
328         size_t i;
329
330         /* We already calculated hash upon lookup */
331         hashvalue_type h = qstate->minfo[id] ? 
332                 ((struct subnet_qstate*)qstate->minfo[id])->qinfo_hash : 
333                 query_info_hash(&qstate->qinfo, qstate->query_flags);
334         /* Step 1, general qinfo lookup */
335         struct lruhash_entry *lru_entry = slabhash_lookup(subnet_msg_cache, h,
336                 &qstate->qinfo, 1);
337         int need_to_insert = (lru_entry == NULL);
338         if (!lru_entry) {
339                 void* data = calloc(1,
340                         sizeof(struct subnet_msg_cache_data));
341                 if(!data) {
342                         log_err("malloc failed");
343                         return;
344                 }
345                 qinf = qstate->qinfo;
346                 qinf.qname = memdup(qstate->qinfo.qname,
347                         qstate->qinfo.qname_len);
348                 if(!qinf.qname) {
349                         free(data);
350                         log_err("memdup failed");
351                         return;
352                 }
353                 mrep_entry = query_info_entrysetup(&qinf, data, h);
354                 free(qinf.qname); /* if qname 'consumed', it is set to NULL */
355                 if (!mrep_entry) {
356                         free(data);
357                         log_err("query_info_entrysetup failed");
358                         return;
359                 }
360                 lru_entry = &mrep_entry->entry;
361                 lock_rw_wrlock(&lru_entry->lock);
362         }
363         /* lru_entry->lock is locked regardless of how we got here,
364          * either from the slabhash_lookup, or above in the new allocated */
365         /* Step 2, find the correct tree */
366         if (!(tree = get_tree(lru_entry->data, edns, sne, qstate->env->cfg))) {
367                 lock_rw_unlock(&lru_entry->lock);
368                 log_err("Subnet cache insertion failed");
369                 return;
370         }
371         lock_quick_lock(&sne->alloc.lock);
372         rep = reply_info_copy(qstate->return_msg->rep, &sne->alloc, NULL);
373         lock_quick_unlock(&sne->alloc.lock);
374         if (!rep) {
375                 lock_rw_unlock(&lru_entry->lock);
376                 log_err("Subnet cache insertion failed");
377                 return;
378         }
379         
380         /* store RRsets */
381         for(i=0; i<rep->rrset_count; i++) {
382                 rep->ref[i].key = rep->rrsets[i];
383                 rep->ref[i].id = rep->rrsets[i]->id;
384         }
385         reply_info_set_ttls(rep, *qstate->env->now);
386         rep->flags |= (BIT_RA | BIT_QR); /* fix flags to be sensible for */
387         rep->flags &= ~(BIT_AA | BIT_CD);/* a reply based on the cache   */
388         addrtree_insert(tree, (addrkey_t*)edns->subnet_addr, 
389                 edns->subnet_source_mask, 
390                 sq->ecs_server_in.subnet_scope_mask, rep,
391                 rep->ttl, *qstate->env->now);
392
393         lock_rw_unlock(&lru_entry->lock);
394         if (need_to_insert) {
395                 slabhash_insert(subnet_msg_cache, h, lru_entry, lru_entry->data,
396                         NULL);
397         }
398 }
399
400 /** Lookup in cache and reply true iff reply is sent. */
401 static int
402 lookup_and_reply(struct module_qstate *qstate, int id, struct subnet_qstate *sq)
403 {
404         struct lruhash_entry *e;
405         struct module_env *env = qstate->env;
406         struct subnet_env *sne = (struct subnet_env*)env->modinfo[id];
407         hashvalue_type h = query_info_hash(&qstate->qinfo, qstate->query_flags);
408         struct subnet_msg_cache_data *data;
409         struct ecs_data *ecs = &sq->ecs_client_in;
410         struct addrtree *tree;
411         struct addrnode *node;
412         uint8_t scope;
413
414         memset(&sq->ecs_client_out, 0, sizeof(sq->ecs_client_out));
415
416         if (sq) sq->qinfo_hash = h; /* Might be useful on cache miss */
417         e = slabhash_lookup(sne->subnet_msg_cache, h, &qstate->qinfo, 1);
418         if (!e) return 0; /* qinfo not in cache */
419         data = e->data;
420         tree = (ecs->subnet_addr_fam == EDNSSUBNET_ADDRFAM_IP4)?
421                 data->tree4 : data->tree6;
422         if (!tree) { /* qinfo in cache but not for this family */
423                 lock_rw_unlock(&e->lock);
424                 return 0;
425         }
426         node = addrtree_find(tree, (addrkey_t*)ecs->subnet_addr, 
427                 ecs->subnet_source_mask, *env->now);
428         if (!node) { /* plain old cache miss */
429                 lock_rw_unlock(&e->lock);
430                 return 0;
431         }
432
433         qstate->return_msg = tomsg(NULL, &qstate->qinfo,
434                 (struct reply_info *)node->elem, qstate->region, *env->now, 0,
435                 env->scratch);
436         scope = (uint8_t)node->scope;
437         lock_rw_unlock(&e->lock);
438         
439         if (!qstate->return_msg) { /* Failed allocation or expired TTL */
440                 return 0;
441         }
442         
443         if (sq->subnet_downstream) { /* relay to interested client */
444                 sq->ecs_client_out.subnet_scope_mask = scope;
445                 sq->ecs_client_out.subnet_addr_fam = ecs->subnet_addr_fam;
446                 sq->ecs_client_out.subnet_source_mask = ecs->subnet_source_mask;
447                 memcpy(&sq->ecs_client_out.subnet_addr, &ecs->subnet_addr,
448                         INET6_SIZE);
449                 sq->ecs_client_out.subnet_validdata = 1;
450         }
451         return 1;
452 }
453
454 /**
455  * Test first bits of addresses for equality. Caller is responsible
456  * for making sure that both a and b are at least net/8 octets long.
457  * @param a: first address.
458  * @param a: seconds address.
459  * @param net: Number of bits to test.
460  * @return: 1 if equal, 0 otherwise.
461  */
462 static int 
463 common_prefix(uint8_t *a, uint8_t *b, uint8_t net)
464 {
465         size_t n = (size_t)net / 8;
466         return !memcmp(a, b, n) && ((net % 8) == 0 || a[n] == b[n]);
467 }
468
469 static enum module_ext_state
470 eval_response(struct module_qstate *qstate, int id, struct subnet_qstate *sq)
471 {
472         struct subnet_env *sne = qstate->env->modinfo[id];
473
474         struct ecs_data *c_in  = &sq->ecs_client_in; /* rcvd from client */
475         struct ecs_data *c_out = &sq->ecs_client_out;/* will send to client */
476         struct ecs_data *s_in  = &sq->ecs_server_in; /* rcvd from auth */
477         struct ecs_data *s_out = &sq->ecs_server_out;/* sent to auth */
478
479         memset(c_out, 0, sizeof(*c_out));
480
481         if (!qstate->return_msg) {
482                 /* already an answer and its not a message, but retain
483                  * the actual rcode, instead of module_error, so send
484                  * module_finished */
485                 return module_finished;
486         }
487         
488         /* We have not asked for subnet data */
489         if (!sq->subnet_sent) {
490                 if (s_in->subnet_validdata)
491                         verbose(VERB_QUERY, "subnet: received spurious data");
492                 if (sq->subnet_downstream) /* Copy back to client */
493                         cp_edns_bad_response(c_out, c_in);
494                 return module_finished;
495         }
496         
497         /* subnet sent but nothing came back */
498         if (!s_in->subnet_validdata) {
499                 /* The authority indicated no support for edns subnet. As a
500                  * consequence the answer ended up in the regular cache. It
501                  * is still usefull to put it in the edns subnet cache for
502                  * when a client explicitly asks for subnet specific answer. */
503                 verbose(VERB_QUERY, "subnet: Authority indicates no support");
504                 if(!sq->started_no_cache_store) {
505                         lock_rw_wrlock(&sne->biglock);
506                         update_cache(qstate, id);
507                         lock_rw_unlock(&sne->biglock);
508                 }
509                 if (sq->subnet_downstream)
510                         cp_edns_bad_response(c_out, c_in);
511                 return module_finished;
512         }
513         
514         /* Being here means we have asked for and got a subnet specific 
515          * answer. Also, the answer from the authority is not yet cached 
516          * anywhere. */
517         
518         /* can we accept response? */
519         if(s_out->subnet_addr_fam != s_in->subnet_addr_fam ||
520                 s_out->subnet_source_mask != s_in->subnet_source_mask ||
521                 !common_prefix(s_out->subnet_addr, s_in->subnet_addr, 
522                         s_out->subnet_source_mask))
523         {
524                 /* we can not accept, restart query without option */
525                 verbose(VERB_QUERY, "subnet: forged data");
526                 s_out->subnet_validdata = 0;
527                 (void)edns_opt_list_remove(&qstate->edns_opts_back_out,
528                         qstate->env->cfg->client_subnet_opcode);
529                 sq->subnet_sent = 0;
530                 return module_restart_next;
531         }
532
533         lock_rw_wrlock(&sne->biglock);
534         if(!sq->started_no_cache_store) {
535                 update_cache(qstate, id);
536         }
537         sne->num_msg_nocache++;
538         lock_rw_unlock(&sne->biglock);
539         
540         if (sq->subnet_downstream) {
541                 /* Client wants to see the answer, echo option back
542                  * and adjust the scope. */
543                 c_out->subnet_addr_fam = c_in->subnet_addr_fam;
544                 c_out->subnet_source_mask = c_in->subnet_source_mask;
545                 memcpy(&c_out->subnet_addr, &c_in->subnet_addr, INET6_SIZE);
546                 c_out->subnet_scope_mask = s_in->subnet_scope_mask;
547                 /* Limit scope returned to client to scope used for caching. */
548                 if(c_out->subnet_addr_fam == EDNSSUBNET_ADDRFAM_IP4) {
549                         if(c_out->subnet_scope_mask >
550                                 qstate->env->cfg->max_client_subnet_ipv4) {
551                                 c_out->subnet_scope_mask =
552                                         qstate->env->cfg->max_client_subnet_ipv4;
553                         }
554                 }
555                 else if(c_out->subnet_scope_mask >
556                                 qstate->env->cfg->max_client_subnet_ipv6) {
557                                 c_out->subnet_scope_mask =
558                                         qstate->env->cfg->max_client_subnet_ipv6;
559                 }
560                 c_out->subnet_validdata = 1;
561         }
562         return module_finished;
563 }
564
565 /** Parse EDNS opt data containing ECS */
566 static int
567 parse_subnet_option(struct edns_option* ecs_option, struct ecs_data* ecs)
568 {
569         memset(ecs, 0, sizeof(*ecs));
570         if (ecs_option->opt_len < 4)
571                 return 0;
572
573         ecs->subnet_addr_fam = sldns_read_uint16(ecs_option->opt_data);
574         ecs->subnet_source_mask = ecs_option->opt_data[2];
575         ecs->subnet_scope_mask = ecs_option->opt_data[3];
576         /* remaining bytes indicate address */
577         
578         /* validate input*/
579         /* option length matches calculated length? */
580         if (ecs_option->opt_len != (size_t)((ecs->subnet_source_mask+7)/8 + 4))
581                 return 0;
582         if (ecs_option->opt_len - 4 > INET6_SIZE || ecs_option->opt_len == 0)
583                 return 0;
584         if (ecs->subnet_addr_fam == EDNSSUBNET_ADDRFAM_IP4) {
585                 if (ecs->subnet_source_mask > 32 || ecs->subnet_scope_mask > 32)
586                         return 0;
587         } else if (ecs->subnet_addr_fam == EDNSSUBNET_ADDRFAM_IP6) {
588                 if (ecs->subnet_source_mask > 128 ||
589                         ecs->subnet_scope_mask > 128)
590                         return 0;
591         } else
592                 return 0;
593         
594         /* valid ECS data, write to ecs_data */
595         if (copy_clear(ecs->subnet_addr, INET6_SIZE, ecs_option->opt_data + 4, 
596                 ecs_option->opt_len - 4, ecs->subnet_source_mask))
597                 return 0;
598         ecs->subnet_validdata = 1;
599         return 1;
600 }
601
602 static void
603 subnet_option_from_ss(struct sockaddr_storage *ss, struct ecs_data* ecs,
604         struct config_file* cfg)
605 {
606         void* sinaddr;
607
608         /* Construct subnet option from original query */
609         if(((struct sockaddr_in*)ss)->sin_family == AF_INET) {
610                 ecs->subnet_source_mask = cfg->max_client_subnet_ipv4;
611                 ecs->subnet_addr_fam = EDNSSUBNET_ADDRFAM_IP4;
612                 sinaddr = &((struct sockaddr_in*)ss)->sin_addr;
613                 if (!copy_clear( ecs->subnet_addr, INET6_SIZE,
614                         (uint8_t *)sinaddr, INET_SIZE,
615                         ecs->subnet_source_mask)) {
616                         ecs->subnet_validdata = 1;
617                 }
618         }
619 #ifdef INET6
620         else {
621                 ecs->subnet_source_mask = cfg->max_client_subnet_ipv6;
622                 ecs->subnet_addr_fam = EDNSSUBNET_ADDRFAM_IP6;
623                 sinaddr = &((struct sockaddr_in6*)ss)->sin6_addr;
624                 if (!copy_clear( ecs->subnet_addr, INET6_SIZE,
625                         (uint8_t *)sinaddr, INET6_SIZE,
626                         ecs->subnet_source_mask)) {
627                         ecs->subnet_validdata = 1;
628                 }
629         }
630 #else
631                         /* We don't know how to handle ip6, just pass */
632 #endif /* INET6 */
633 }
634
635 int
636 ecs_query_response(struct module_qstate* qstate, struct dns_msg* response,
637         int id, void* ATTR_UNUSED(cbargs))
638 {
639         struct subnet_qstate *sq;
640         
641         if(!response || !(sq=(struct subnet_qstate*)qstate->minfo[id]))
642                 return 1;
643
644         if(sq->subnet_sent &&
645                 FLAGS_GET_RCODE(response->rep->flags) == LDNS_RCODE_REFUSED) {
646                 /* REFUSED response to ECS query, remove ECS option. */
647                 edns_opt_list_remove(&qstate->edns_opts_back_out,
648                         qstate->env->cfg->client_subnet_opcode);
649                 sq->subnet_sent = 0;
650                 memset(&sq->ecs_server_out, 0, sizeof(sq->ecs_server_out));
651         }
652         return 1;
653 }
654
655 int
656 ecs_edns_back_parsed(struct module_qstate* qstate, int id,
657         void* ATTR_UNUSED(cbargs))
658 {
659         struct subnet_qstate *sq;
660         struct edns_option* ecs_opt;
661         
662         if(!(sq=(struct subnet_qstate*)qstate->minfo[id]))
663                 return 1;
664         if((ecs_opt = edns_opt_list_find(
665                 qstate->edns_opts_back_in,
666                 qstate->env->cfg->client_subnet_opcode))) {
667                 if(parse_subnet_option(ecs_opt, &sq->ecs_server_in) &&
668                         sq->subnet_sent &&
669                         sq->ecs_server_in.subnet_validdata)
670                         /* Only skip global cache store if we sent an ECS option
671                          * and received one back. Answers from non-whitelisted
672                          * servers will end up in global cache. Answers for
673                          * queries with 0 source will not (unless nameserver
674                          * does not support ECS). */
675                         qstate->no_cache_store = 1;
676         }
677
678         return 1;
679 }
680
681 void
682 subnetmod_operate(struct module_qstate *qstate, enum module_ev event, 
683         int id, struct outbound_entry* outbound)
684 {
685         struct subnet_env *sne = qstate->env->modinfo[id];
686         struct subnet_qstate *sq = (struct subnet_qstate*)qstate->minfo[id];
687         
688         verbose(VERB_QUERY, "subnet[module %d] operate: extstate:%s "
689                 "event:%s", id, strextstate(qstate->ext_state[id]), 
690                 strmodulevent(event));
691         log_query_info(VERB_QUERY, "subnet operate: query", &qstate->qinfo);
692
693         if((event == module_event_new || event == module_event_pass) &&
694                 sq == NULL) {
695                 struct edns_option* ecs_opt;
696                 if(!subnet_new_qstate(qstate, id)) {
697                         qstate->return_msg = NULL;
698                         qstate->ext_state[id] = module_finished;
699                         return;
700                 }
701
702                 sq = (struct subnet_qstate*)qstate->minfo[id];
703
704                 if((ecs_opt = edns_opt_list_find(
705                         qstate->edns_opts_front_in,
706                         qstate->env->cfg->client_subnet_opcode))) {
707                         if(!parse_subnet_option(ecs_opt, &sq->ecs_client_in)) {
708                                 /* Wrongly formatted ECS option. RFC mandates to
709                                  * return FORMERROR. */
710                                 qstate->return_rcode = LDNS_RCODE_FORMERR;
711                                 qstate->ext_state[id] = module_finished;
712                                 return;
713                         }
714                         sq->subnet_downstream = 1;
715                 }
716                 else if(qstate->mesh_info->reply_list) {
717                         subnet_option_from_ss(
718                                 &qstate->mesh_info->reply_list->query_reply.addr,
719                                 &sq->ecs_client_in, qstate->env->cfg);
720                 }
721                 
722                 if(sq->ecs_client_in.subnet_validdata == 0) {
723                         /* No clients are interested in result or we could not
724                          * parse it, we don't do client subnet */
725                         sq->ecs_server_out.subnet_validdata = 0;
726                         verbose(VERB_ALGO, "subnet: pass to next module");
727                         qstate->ext_state[id] = module_wait_module;
728                         return;
729                 }
730
731                 /* Limit to minimum allowed source mask */
732                 if(sq->ecs_client_in.subnet_source_mask != 0 && (
733                         (sq->ecs_client_in.subnet_addr_fam == EDNSSUBNET_ADDRFAM_IP4 &&
734                          sq->ecs_client_in.subnet_source_mask < qstate->env->cfg->min_client_subnet_ipv4) ||
735                         (sq->ecs_client_in.subnet_addr_fam == EDNSSUBNET_ADDRFAM_IP6 &&
736                          sq->ecs_client_in.subnet_source_mask < qstate->env->cfg->min_client_subnet_ipv6))) {
737                                 qstate->return_rcode = LDNS_RCODE_REFUSED;
738                                 qstate->ext_state[id] = module_finished;
739                                 return;
740                 }
741
742                 lock_rw_wrlock(&sne->biglock);
743                 if (lookup_and_reply(qstate, id, sq)) {
744                         sne->num_msg_cache++;
745                         lock_rw_unlock(&sne->biglock);
746                         verbose(VERB_QUERY, "subnet: answered from cache");
747                         qstate->ext_state[id] = module_finished;
748
749                         ecs_opt_list_append(&sq->ecs_client_out,
750                                 &qstate->edns_opts_front_out, qstate);
751                         return;
752                 }
753                 lock_rw_unlock(&sne->biglock);
754                 
755                 sq->ecs_server_out.subnet_addr_fam =
756                         sq->ecs_client_in.subnet_addr_fam;
757                 sq->ecs_server_out.subnet_source_mask =
758                         sq->ecs_client_in.subnet_source_mask;
759                 /* Limit source prefix to configured maximum */
760                 if(sq->ecs_server_out.subnet_addr_fam == EDNSSUBNET_ADDRFAM_IP4 
761                         && sq->ecs_server_out.subnet_source_mask >
762                         qstate->env->cfg->max_client_subnet_ipv4)
763                         sq->ecs_server_out.subnet_source_mask =
764                                 qstate->env->cfg->max_client_subnet_ipv4;
765                 else if(sq->ecs_server_out.subnet_addr_fam == EDNSSUBNET_ADDRFAM_IP6 
766                         && sq->ecs_server_out.subnet_source_mask >
767                         qstate->env->cfg->max_client_subnet_ipv6)
768                         sq->ecs_server_out.subnet_source_mask =
769                                 qstate->env->cfg->max_client_subnet_ipv6;
770                 /* Safe to copy completely, even if the source is limited by the
771                  * configuration. ecs_opt_list_append() will limit the address.
772                  * */
773                 memcpy(&sq->ecs_server_out.subnet_addr,
774                         sq->ecs_client_in.subnet_addr, INET6_SIZE);
775                 sq->ecs_server_out.subnet_scope_mask = 0;
776                 sq->ecs_server_out.subnet_validdata = 1;
777                 if(sq->ecs_server_out.subnet_source_mask != 0 &&
778                         qstate->env->cfg->client_subnet_always_forward &&
779                         sq->subnet_downstream)
780                         /* ECS specific data required, do not look at the global
781                          * cache in other modules. */
782                         qstate->no_cache_lookup = 1;
783                 
784                 /* pass request to next module */
785                 verbose(VERB_ALGO,
786                         "subnet: not found in cache. pass to next module");
787                 qstate->ext_state[id] = module_wait_module;
788                 return;
789         }
790         /* Query handed back by next module, we have a 'final' answer */
791         if(sq && event == module_event_moddone) {
792                 qstate->ext_state[id] = eval_response(qstate, id, sq);
793                 if(qstate->ext_state[id] == module_finished &&
794                         qstate->return_msg) {
795                         ecs_opt_list_append(&sq->ecs_client_out,
796                                 &qstate->edns_opts_front_out, qstate);
797                 }
798                 qstate->no_cache_store = sq->started_no_cache_store;
799                 return;
800         }
801         if(sq && outbound) {
802                 return;
803         }
804         /* We are being revisited */
805         if(event == module_event_pass || event == module_event_new) {
806                 /* Just pass it on, we already did the work */
807                 verbose(VERB_ALGO, "subnet: pass to next module");
808                 qstate->ext_state[id] = module_wait_module;
809                 return;
810         }
811         if(!sq && (event == module_event_moddone)) {
812                 /* during priming, module done but we never started */
813                 qstate->ext_state[id] = module_finished;
814                 return;
815         }
816         log_err("subnet: bad event %s", strmodulevent(event));
817         qstate->ext_state[id] = module_error;
818         return;
819 }
820
821 void
822 subnetmod_clear(struct module_qstate *ATTR_UNUSED(qstate),
823         int ATTR_UNUSED(id))
824 {
825         /* qstate has no data outside region */
826 }
827
828 void
829 subnetmod_inform_super(struct module_qstate *ATTR_UNUSED(qstate),
830         int ATTR_UNUSED(id), struct module_qstate *ATTR_UNUSED(super))
831 {
832         /* Not used */
833 }
834
835 size_t
836 subnetmod_get_mem(struct module_env *env, int id)
837 {
838         struct subnet_env *sn_env = env->modinfo[id];
839         if (!sn_env) return 0;
840         return sizeof(*sn_env) + 
841                 slabhash_get_mem(sn_env->subnet_msg_cache) +
842                 ecs_whitelist_get_mem(sn_env->whitelist);
843 }
844
845 /**
846  * The module function block 
847  */
848 static struct module_func_block subnetmod_block = {
849         "subnet", &subnetmod_init, &subnetmod_deinit, &subnetmod_operate,
850         &subnetmod_inform_super, &subnetmod_clear, &subnetmod_get_mem
851 };
852
853 struct module_func_block*
854 subnetmod_get_funcblock(void)
855 {
856         return &subnetmod_block;
857 }
858
859 /** Wrappers for static functions to unit test */
860 size_t
861 unittest_wrapper_subnetmod_sizefunc(void *elemptr)
862 {
863         return sizefunc(elemptr);
864 }
865
866 #endif  /* CLIENT_SUBNET */