xref: /freebsd/contrib/unbound/iterator/iter_scrub.c (revision 3b8f08459569bf0faa21473e5cec2491e95c9349)
1 /*
2  * iterator/iter_scrub.c - scrubbing, normalization, sanitization of DNS msgs.
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 LIMITED
25  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
26  * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE
27  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
30  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
31  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
32  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
33  * POSSIBILITY OF SUCH DAMAGE.
34  */
35 
36 /**
37  * \file
38  *
39  * This file has routine(s) for cleaning up incoming DNS messages from
40  * possible useless or malicious junk in it.
41  */
42 #include "config.h"
43 #include "iterator/iter_scrub.h"
44 #include "iterator/iterator.h"
45 #include "iterator/iter_priv.h"
46 #include "services/cache/rrset.h"
47 #include "util/log.h"
48 #include "util/net_help.h"
49 #include "util/regional.h"
50 #include "util/config_file.h"
51 #include "util/module.h"
52 #include "util/data/msgparse.h"
53 #include "util/data/dname.h"
54 #include "util/data/msgreply.h"
55 #include "util/alloc.h"
56 
57 /** RRset flag used during scrubbing. The RRset is OK. */
58 #define RRSET_SCRUB_OK	0x80
59 
60 /** remove rrset, update loop variables */
61 static void
62 remove_rrset(const char* str, ldns_buffer* pkt, struct msg_parse* msg,
63 	struct rrset_parse* prev, struct rrset_parse** rrset)
64 {
65 	if(verbosity >= VERB_QUERY
66 		&& (*rrset)->dname_len <= LDNS_MAX_DOMAINLEN) {
67 		uint8_t buf[LDNS_MAX_DOMAINLEN+1];
68 		dname_pkt_copy(pkt, buf, (*rrset)->dname);
69 		log_nametypeclass(VERB_QUERY, str, buf,
70 			(*rrset)->type, ntohs((*rrset)->rrset_class));
71 	}
72 	if(prev)
73 		prev->rrset_all_next = (*rrset)->rrset_all_next;
74 	else	msg->rrset_first = (*rrset)->rrset_all_next;
75 	if(msg->rrset_last == *rrset)
76 		msg->rrset_last = prev;
77 	msg->rrset_count --;
78 	switch((*rrset)->section) {
79 		case LDNS_SECTION_ANSWER: msg->an_rrsets--; break;
80 		case LDNS_SECTION_AUTHORITY: msg->ns_rrsets--; break;
81 		case LDNS_SECTION_ADDITIONAL: msg->ar_rrsets--; break;
82 		default: log_assert(0);
83 	}
84 	msgparse_bucket_remove(msg, *rrset);
85 	*rrset = (*rrset)->rrset_all_next;
86 }
87 
88 /** return true if rr type has additional names in it */
89 static int
90 has_additional(uint16_t t)
91 {
92 	switch(t) {
93 		case LDNS_RR_TYPE_MB:
94 		case LDNS_RR_TYPE_MD:
95 		case LDNS_RR_TYPE_MF:
96 		case LDNS_RR_TYPE_NS:
97 		case LDNS_RR_TYPE_MX:
98 		case LDNS_RR_TYPE_KX:
99 		case LDNS_RR_TYPE_SRV:
100 			return 1;
101 		case LDNS_RR_TYPE_NAPTR:
102 			/* TODO: NAPTR not supported, glue stripped off */
103 			return 0;
104 	}
105 	return 0;
106 }
107 
108 /** get additional name from rrset RR, return false if no name present */
109 static int
110 get_additional_name(struct rrset_parse* rrset, struct rr_parse* rr,
111 	uint8_t** nm, size_t* nmlen, ldns_buffer* pkt)
112 {
113 	size_t offset = 0;
114 	size_t len, oldpos;
115 	switch(rrset->type) {
116 		case LDNS_RR_TYPE_MB:
117 		case LDNS_RR_TYPE_MD:
118 		case LDNS_RR_TYPE_MF:
119 		case LDNS_RR_TYPE_NS:
120 			offset = 0;
121 			break;
122 		case LDNS_RR_TYPE_MX:
123 		case LDNS_RR_TYPE_KX:
124 			offset = 2;
125 			break;
126 		case LDNS_RR_TYPE_SRV:
127 			offset = 6;
128 			break;
129 		case LDNS_RR_TYPE_NAPTR:
130 			/* TODO: NAPTR not supported, glue stripped off */
131 			return 0;
132 		default:
133 			return 0;
134 	}
135 	len = ldns_read_uint16(rr->ttl_data+sizeof(uint32_t));
136 	if(len < offset+1)
137 		return 0; /* rdata field too small */
138 	*nm = rr->ttl_data+sizeof(uint32_t)+sizeof(uint16_t)+offset;
139 	oldpos = ldns_buffer_position(pkt);
140 	ldns_buffer_set_position(pkt, (size_t)(*nm - ldns_buffer_begin(pkt)));
141 	*nmlen = pkt_dname_len(pkt);
142 	ldns_buffer_set_position(pkt, oldpos);
143 	if(*nmlen == 0)
144 		return 0;
145 	return 1;
146 }
147 
148 /** Place mark on rrsets in additional section they are OK */
149 static void
150 mark_additional_rrset(ldns_buffer* pkt, struct msg_parse* msg,
151 	struct rrset_parse* rrset)
152 {
153 	/* Mark A and AAAA for NS as appropriate additional section info. */
154 	uint8_t* nm = NULL;
155 	size_t nmlen = 0;
156 	struct rr_parse* rr;
157 
158 	if(!has_additional(rrset->type))
159 		return;
160 	for(rr = rrset->rr_first; rr; rr = rr->next) {
161 		if(get_additional_name(rrset, rr, &nm, &nmlen, pkt)) {
162 			/* mark A */
163 			hashvalue_t h = pkt_hash_rrset(pkt, nm, LDNS_RR_TYPE_A,
164 				rrset->rrset_class, 0);
165 			struct rrset_parse* r = msgparse_hashtable_lookup(
166 				msg, pkt, h, 0, nm, nmlen,
167 				LDNS_RR_TYPE_A, rrset->rrset_class);
168 			if(r && r->section == LDNS_SECTION_ADDITIONAL) {
169 				r->flags |= RRSET_SCRUB_OK;
170 			}
171 
172 			/* mark AAAA */
173 			h = pkt_hash_rrset(pkt, nm, LDNS_RR_TYPE_AAAA,
174 				rrset->rrset_class, 0);
175 			r = msgparse_hashtable_lookup(msg, pkt, h, 0, nm,
176 				nmlen, LDNS_RR_TYPE_AAAA, rrset->rrset_class);
177 			if(r && r->section == LDNS_SECTION_ADDITIONAL) {
178 				r->flags |= RRSET_SCRUB_OK;
179 			}
180 		}
181 	}
182 }
183 
184 /** Get target name of a CNAME */
185 static int
186 parse_get_cname_target(struct rrset_parse* rrset, uint8_t** sname,
187 	size_t* snamelen)
188 {
189 	if(rrset->rr_count != 1) {
190 		struct rr_parse* sig;
191 		verbose(VERB_ALGO, "Found CNAME rrset with "
192 			"size > 1: %u", (unsigned)rrset->rr_count);
193 		/* use the first CNAME! */
194 		rrset->rr_count = 1;
195 		rrset->size = rrset->rr_first->size;
196 		for(sig=rrset->rrsig_first; sig; sig=sig->next)
197 			rrset->size += sig->size;
198 		rrset->rr_last = rrset->rr_first;
199 		rrset->rr_first->next = NULL;
200 	}
201 	if(rrset->rr_first->size < sizeof(uint16_t)+1)
202 		return 0; /* CNAME rdata too small */
203 	*sname = rrset->rr_first->ttl_data + sizeof(uint32_t)
204 		+ sizeof(uint16_t); /* skip ttl, rdatalen */
205 	*snamelen = rrset->rr_first->size - sizeof(uint16_t);
206 	return 1;
207 }
208 
209 /** Synthesize CNAME from DNAME, false if too long */
210 static int
211 synth_cname(uint8_t* qname, size_t qnamelen, struct rrset_parse* dname_rrset,
212 	uint8_t* alias, size_t* aliaslen, ldns_buffer* pkt)
213 {
214 	/* we already know that sname is a strict subdomain of DNAME owner */
215 	uint8_t* dtarg = NULL;
216 	size_t dtarglen;
217 	if(!parse_get_cname_target(dname_rrset, &dtarg, &dtarglen))
218 		return 0;
219 	log_assert(qnamelen > dname_rrset->dname_len);
220 	/* DNAME from com. to net. with qname example.com. -> example.net. */
221 	/* so: \3com\0 to \3net\0 and qname \7example\3com\0 */
222 	*aliaslen = qnamelen + dtarglen - dname_rrset->dname_len;
223 	if(*aliaslen > LDNS_MAX_DOMAINLEN)
224 		return 0; /* should have been RCODE YXDOMAIN */
225 	/* decompress dnames into buffer, we know it fits */
226 	dname_pkt_copy(pkt, alias, qname);
227 	dname_pkt_copy(pkt, alias+(qnamelen-dname_rrset->dname_len), dtarg);
228 	return 1;
229 }
230 
231 /** synthesize a CNAME rrset */
232 static struct rrset_parse*
233 synth_cname_rrset(uint8_t** sname, size_t* snamelen, uint8_t* alias,
234 	size_t aliaslen, struct regional* region, struct msg_parse* msg,
235 	struct rrset_parse* rrset, struct rrset_parse* prev,
236 	struct rrset_parse* nx, ldns_buffer* pkt)
237 {
238 	struct rrset_parse* cn = (struct rrset_parse*)regional_alloc(region,
239 		sizeof(struct rrset_parse));
240 	if(!cn)
241 		return NULL;
242 	memset(cn, 0, sizeof(*cn));
243 	cn->rr_first = (struct rr_parse*)regional_alloc(region,
244 		sizeof(struct rr_parse));
245 	if(!cn->rr_first)
246 		return NULL;
247 	cn->rr_last = cn->rr_first;
248 	/* CNAME from sname to alias */
249 	cn->dname = (uint8_t*)regional_alloc(region, *snamelen);
250 	if(!cn->dname)
251 		return NULL;
252 	dname_pkt_copy(pkt, cn->dname, *sname);
253 	cn->dname_len = *snamelen;
254 	cn->type = LDNS_RR_TYPE_CNAME;
255 	cn->section = rrset->section;
256 	cn->rrset_class = rrset->rrset_class;
257 	cn->rr_count = 1;
258 	cn->size = sizeof(uint16_t) + aliaslen;
259 	cn->hash=pkt_hash_rrset(pkt, cn->dname, cn->type, cn->rrset_class, 0);
260 	/* allocate TTL + rdatalen + uncompressed dname */
261 	memset(cn->rr_first, 0, sizeof(struct rr_parse));
262 	cn->rr_first->outside_packet = 1;
263 	cn->rr_first->ttl_data = (uint8_t*)regional_alloc(region,
264 		sizeof(uint32_t)+sizeof(uint16_t)+aliaslen);
265 	if(!cn->rr_first->ttl_data)
266 		return NULL;
267 	ldns_write_uint32(cn->rr_first->ttl_data, 0); /* TTL = 0 */
268 	ldns_write_uint16(cn->rr_first->ttl_data+4, aliaslen);
269 	memmove(cn->rr_first->ttl_data+6, alias, aliaslen);
270 	cn->rr_first->size = sizeof(uint16_t)+aliaslen;
271 
272 	/* link it in */
273 	cn->rrset_all_next = nx;
274 	if(prev)
275 		prev->rrset_all_next = cn;
276 	else	msg->rrset_first = cn;
277 	if(nx == NULL)
278 		msg->rrset_last = cn;
279 	msg->rrset_count ++;
280 	msg->an_rrsets++;
281 	/* it is not inserted in the msg hashtable. */
282 
283 	*sname = cn->rr_first->ttl_data + sizeof(uint32_t)+sizeof(uint16_t);
284 	*snamelen = aliaslen;
285 	return cn;
286 }
287 
288 /** check if DNAME applies to a name */
289 static int
290 pkt_strict_sub(ldns_buffer* pkt, uint8_t* sname, uint8_t* dr)
291 {
292 	uint8_t buf1[LDNS_MAX_DOMAINLEN+1];
293 	uint8_t buf2[LDNS_MAX_DOMAINLEN+1];
294 	/* decompress names */
295 	dname_pkt_copy(pkt, buf1, sname);
296 	dname_pkt_copy(pkt, buf2, dr);
297 	return dname_strict_subdomain_c(buf1, buf2);
298 }
299 
300 /** check subdomain with decompression */
301 static int
302 pkt_sub(ldns_buffer* pkt, uint8_t* comprname, uint8_t* zone)
303 {
304 	uint8_t buf[LDNS_MAX_DOMAINLEN+1];
305 	dname_pkt_copy(pkt, buf, comprname);
306 	return dname_subdomain_c(buf, zone);
307 }
308 
309 /** check subdomain with decompression, compressed is parent */
310 static int
311 sub_of_pkt(ldns_buffer* pkt, uint8_t* zone, uint8_t* comprname)
312 {
313 	uint8_t buf[LDNS_MAX_DOMAINLEN+1];
314 	dname_pkt_copy(pkt, buf, comprname);
315 	return dname_subdomain_c(zone, buf);
316 }
317 
318 /**
319  * This routine normalizes a response. This includes removing "irrelevant"
320  * records from the answer and additional sections and (re)synthesizing
321  * CNAMEs from DNAMEs, if present.
322  *
323  * @param pkt: packet.
324  * @param msg: msg to normalize.
325  * @param qinfo: original query.
326  * @param region: where to allocate synthesized CNAMEs.
327  * @return 0 on error.
328  */
329 static int
330 scrub_normalize(ldns_buffer* pkt, struct msg_parse* msg,
331 	struct query_info* qinfo, struct regional* region)
332 {
333 	uint8_t* sname = qinfo->qname;
334 	size_t snamelen = qinfo->qname_len;
335 	struct rrset_parse* rrset, *prev, *nsset=NULL;
336 
337 	if(FLAGS_GET_RCODE(msg->flags) != LDNS_RCODE_NOERROR &&
338 		FLAGS_GET_RCODE(msg->flags) != LDNS_RCODE_NXDOMAIN)
339 		return 1;
340 
341 	/* For the ANSWER section, remove all "irrelevant" records and add
342 	 * synthesized CNAMEs from DNAMEs
343 	 * This will strip out-of-order CNAMEs as well. */
344 
345 	/* walk through the parse packet rrset list, keep track of previous
346 	 * for insert and delete ease, and examine every RRset */
347 	prev = NULL;
348 	rrset = msg->rrset_first;
349 	while(rrset && rrset->section == LDNS_SECTION_ANSWER) {
350 		if(rrset->type == LDNS_RR_TYPE_DNAME &&
351 			pkt_strict_sub(pkt, sname, rrset->dname)) {
352 			/* check if next rrset is correct CNAME. else,
353 			 * synthesize a CNAME */
354 			struct rrset_parse* nx = rrset->rrset_all_next;
355 			uint8_t alias[LDNS_MAX_DOMAINLEN+1];
356 			size_t aliaslen = 0;
357 			if(rrset->rr_count != 1) {
358 				verbose(VERB_ALGO, "Found DNAME rrset with "
359 					"size > 1: %u",
360 					(unsigned)rrset->rr_count);
361 				return 0;
362 			}
363 			if(!synth_cname(sname, snamelen, rrset, alias,
364 				&aliaslen, pkt)) {
365 				verbose(VERB_ALGO, "synthesized CNAME "
366 					"too long");
367 				return 0;
368 			}
369 			if(nx && nx->type == LDNS_RR_TYPE_CNAME &&
370 			   dname_pkt_compare(pkt, sname, nx->dname) == 0) {
371 				/* check next cname */
372 				uint8_t* t = NULL;
373 				size_t tlen = 0;
374 				if(!parse_get_cname_target(rrset, &t, &tlen))
375 					return 0;
376 				if(dname_pkt_compare(pkt, alias, t) == 0) {
377 					/* it's OK and better capitalized */
378 					prev = rrset;
379 					rrset = nx;
380 					continue;
381 				}
382 				/* synth ourselves */
383 			}
384 			/* synth a CNAME rrset */
385 			prev = synth_cname_rrset(&sname, &snamelen, alias,
386 				aliaslen, region, msg, rrset, rrset, nx, pkt);
387 			if(!prev) {
388 				log_err("out of memory synthesizing CNAME");
389 				return 0;
390 			}
391 			/* FIXME: resolve the conflict between synthesized
392 			 * CNAME ttls and the cache. */
393 			rrset = nx;
394 			continue;
395 
396 		}
397 
398 		/* The only records in the ANSWER section not allowed to */
399 		if(dname_pkt_compare(pkt, sname, rrset->dname) != 0) {
400 			remove_rrset("normalize: removing irrelevant RRset:",
401 				pkt, msg, prev, &rrset);
402 			continue;
403 		}
404 
405 		/* Follow the CNAME chain. */
406 		if(rrset->type == LDNS_RR_TYPE_CNAME) {
407 			uint8_t* oldsname = sname;
408 			if(!parse_get_cname_target(rrset, &sname, &snamelen))
409 				return 0;
410 			prev = rrset;
411 			rrset = rrset->rrset_all_next;
412 			/* in CNAME ANY response, can have data after CNAME */
413 			if(qinfo->qtype == LDNS_RR_TYPE_ANY) {
414 				while(rrset && rrset->section ==
415 					LDNS_SECTION_ANSWER &&
416 					dname_pkt_compare(pkt, oldsname,
417 					rrset->dname) == 0) {
418 					prev = rrset;
419 					rrset = rrset->rrset_all_next;
420 				}
421 			}
422 			continue;
423 		}
424 
425 		/* Otherwise, make sure that the RRset matches the qtype. */
426 		if(qinfo->qtype != LDNS_RR_TYPE_ANY &&
427 			qinfo->qtype != rrset->type) {
428 			remove_rrset("normalize: removing irrelevant RRset:",
429 				pkt, msg, prev, &rrset);
430 			continue;
431 		}
432 
433 		/* Mark the additional names from relevant rrset as OK. */
434 		/* only for RRsets that match the query name, other ones
435 		 * will be removed by sanitize, so no additional for them */
436 		if(dname_pkt_compare(pkt, qinfo->qname, rrset->dname) == 0)
437 			mark_additional_rrset(pkt, msg, rrset);
438 
439 		prev = rrset;
440 		rrset = rrset->rrset_all_next;
441 	}
442 
443 	/* Mark additional names from AUTHORITY */
444 	while(rrset && rrset->section == LDNS_SECTION_AUTHORITY) {
445 		if(rrset->type==LDNS_RR_TYPE_DNAME ||
446 			rrset->type==LDNS_RR_TYPE_CNAME ||
447 			rrset->type==LDNS_RR_TYPE_A ||
448 			rrset->type==LDNS_RR_TYPE_AAAA) {
449 			remove_rrset("normalize: removing irrelevant "
450 				"RRset:", pkt, msg, prev, &rrset);
451 			continue;
452 		}
453 		/* only one NS set allowed in authority section */
454 		if(rrset->type==LDNS_RR_TYPE_NS) {
455 			/* NS set must be pertinent to the query */
456 			if(!sub_of_pkt(pkt, qinfo->qname, rrset->dname)) {
457 				remove_rrset("normalize: removing irrelevant "
458 					"RRset:", pkt, msg, prev, &rrset);
459 				continue;
460 			}
461 			if(nsset == NULL) {
462 				nsset = rrset;
463 			} else {
464 				remove_rrset("normalize: removing irrelevant "
465 					"RRset:", pkt, msg, prev, &rrset);
466 				continue;
467 			}
468 		}
469 		mark_additional_rrset(pkt, msg, rrset);
470 		prev = rrset;
471 		rrset = rrset->rrset_all_next;
472 	}
473 
474 	/* For each record in the additional section, remove it if it is an
475 	 * address record and not in the collection of additional names
476 	 * found in ANSWER and AUTHORITY. */
477 	/* These records have not been marked OK previously */
478 	while(rrset && rrset->section == LDNS_SECTION_ADDITIONAL) {
479 		/* FIXME: what about other types? */
480 		if(rrset->type==LDNS_RR_TYPE_A ||
481 			rrset->type==LDNS_RR_TYPE_AAAA)
482 		{
483 			if((rrset->flags & RRSET_SCRUB_OK)) {
484 				/* remove flag to clean up flags variable */
485 				rrset->flags &= ~RRSET_SCRUB_OK;
486 			} else {
487 				remove_rrset("normalize: removing irrelevant "
488 					"RRset:", pkt, msg, prev, &rrset);
489 				continue;
490 			}
491 		}
492 		if(rrset->type==LDNS_RR_TYPE_DNAME ||
493 			rrset->type==LDNS_RR_TYPE_CNAME ||
494 			rrset->type==LDNS_RR_TYPE_NS) {
495 			remove_rrset("normalize: removing irrelevant "
496 				"RRset:", pkt, msg, prev, &rrset);
497 			continue;
498 		}
499 		prev = rrset;
500 		rrset = rrset->rrset_all_next;
501 	}
502 
503 	return 1;
504 }
505 
506 /**
507  * Store potential poison in the cache (only if hardening disabled).
508  * The rrset is stored in the cache but removed from the message.
509  * So that it will be used for infrastructure purposes, but not be
510  * returned to the client.
511  * @param pkt: packet
512  * @param msg: message parsed
513  * @param env: environment with cache
514  * @param rrset: to store.
515  */
516 static void
517 store_rrset(ldns_buffer* pkt, struct msg_parse* msg, struct module_env* env,
518 	struct rrset_parse* rrset)
519 {
520 	struct ub_packed_rrset_key* k;
521 	struct packed_rrset_data* d;
522 	struct rrset_ref ref;
523 	uint32_t now = *env->now;
524 
525 	k = alloc_special_obtain(env->alloc);
526 	if(!k)
527 		return;
528 	k->entry.data = NULL;
529 	if(!parse_copy_decompress_rrset(pkt, msg, rrset, NULL, k)) {
530 		alloc_special_release(env->alloc, k);
531 		return;
532 	}
533 	d = (struct packed_rrset_data*)k->entry.data;
534 	packed_rrset_ttl_add(d, now);
535 	ref.key = k;
536 	ref.id = k->id;
537 	/*ignore ret: it was in the cache, ref updated */
538 	(void)rrset_cache_update(env->rrset_cache, &ref, env->alloc, now);
539 }
540 
541 /** Check if there are SOA records in the authority section (negative) */
542 static int
543 soa_in_auth(struct msg_parse* msg)
544 {
545 	struct rrset_parse* rrset;
546 	for(rrset = msg->rrset_first; rrset; rrset = rrset->rrset_all_next)
547 		if(rrset->type == LDNS_RR_TYPE_SOA &&
548 			rrset->section == LDNS_SECTION_AUTHORITY)
549 			return 1;
550 	return 0;
551 }
552 
553 /**
554  * Check if right hand name in NSEC is within zone
555  * @param rrset: the NSEC rrset
556  * @param zonename: the zone name.
557  * @return true if BAD.
558  */
559 static int sanitize_nsec_is_overreach(struct rrset_parse* rrset,
560 	uint8_t* zonename)
561 {
562 	struct rr_parse* rr;
563 	uint8_t* rhs;
564 	size_t len;
565 	log_assert(rrset->type == LDNS_RR_TYPE_NSEC);
566 	for(rr = rrset->rr_first; rr; rr = rr->next) {
567 		rhs = rr->ttl_data+4+2;
568 		len = ldns_read_uint16(rr->ttl_data+4);
569 		if(!dname_valid(rhs, len)) {
570 			/* malformed domain name in rdata */
571 			return 1;
572 		}
573 		if(!dname_subdomain_c(rhs, zonename)) {
574 			/* overreaching */
575 			return 1;
576 		}
577 	}
578 	/* all NSEC RRs OK */
579 	return 0;
580 }
581 
582 /**
583  * Given a response event, remove suspect RRsets from the response.
584  * "Suspect" rrsets are potentially poison. Note that this routine expects
585  * the response to be in a "normalized" state -- that is, all "irrelevant"
586  * RRsets have already been removed, CNAMEs are in order, etc.
587  *
588  * @param pkt: packet.
589  * @param msg: msg to normalize.
590  * @param qinfo: the question originally asked.
591  * @param zonename: name of server zone.
592  * @param env: module environment with config and cache.
593  * @param ie: iterator environment with private address data.
594  * @return 0 on error.
595  */
596 static int
597 scrub_sanitize(ldns_buffer* pkt, struct msg_parse* msg,
598 	struct query_info* qinfo, uint8_t* zonename, struct module_env* env,
599 	struct iter_env* ie)
600 {
601 	int del_addi = 0; /* if additional-holding rrsets are deleted, we
602 		do not trust the normalized additional-A-AAAA any more */
603 	struct rrset_parse* rrset, *prev;
604 	prev = NULL;
605 	rrset = msg->rrset_first;
606 
607 	/* the first DNAME is allowed to stay. It needs checking before
608 	 * it can be used from the cache. After normalization, an initial
609 	 * DNAME will have a correctly synthesized CNAME after it. */
610 	if(rrset && rrset->type == LDNS_RR_TYPE_DNAME &&
611 		rrset->section == LDNS_SECTION_ANSWER &&
612 		pkt_strict_sub(pkt, qinfo->qname, rrset->dname) &&
613 		pkt_sub(pkt, rrset->dname, zonename)) {
614 		prev = rrset; /* DNAME allowed to stay in answer section */
615 		rrset = rrset->rrset_all_next;
616 	}
617 
618 	/* remove all records from the answer section that are
619 	 * not the same domain name as the query domain name.
620 	 * The answer section should contain rrsets with the same name
621 	 * as the question. For DNAMEs a CNAME has been synthesized.
622 	 * Wildcards have the query name in answer section.
623 	 * ANY queries get query name in answer section.
624 	 * Remainders of CNAME chains are cut off and resolved by iterator. */
625 	while(rrset && rrset->section == LDNS_SECTION_ANSWER) {
626 		if(dname_pkt_compare(pkt, qinfo->qname, rrset->dname) != 0) {
627 			if(has_additional(rrset->type)) del_addi = 1;
628 			remove_rrset("sanitize: removing extraneous answer "
629 				"RRset:", pkt, msg, prev, &rrset);
630 			continue;
631 		}
632 		prev = rrset;
633 		rrset = rrset->rrset_all_next;
634 	}
635 
636 	/* At this point, we brutally remove ALL rrsets that aren't
637 	 * children of the originating zone. The idea here is that,
638 	 * as far as we know, the server that we contacted is ONLY
639 	 * authoritative for the originating zone. It, of course, MAY
640 	 * be authoriative for any other zones, and of course, MAY
641 	 * NOT be authoritative for some subdomains of the originating
642 	 * zone. */
643 	prev = NULL;
644 	rrset = msg->rrset_first;
645 	while(rrset) {
646 
647 		/* remove private addresses */
648 		if( (rrset->type == LDNS_RR_TYPE_A ||
649 			rrset->type == LDNS_RR_TYPE_AAAA) &&
650 			priv_rrset_bad(ie->priv, pkt, rrset)) {
651 
652 			/* do not set servfail since this leads to too
653 			 * many drops of other people using rfc1918 space */
654 			remove_rrset("sanitize: removing public name with "
655 				"private address", pkt, msg, prev, &rrset);
656 			continue;
657 		}
658 
659 		/* skip DNAME records -- they will always be followed by a
660 		 * synthesized CNAME, which will be relevant.
661 		 * FIXME: should this do something differently with DNAME
662 		 * rrsets NOT in Section.ANSWER? */
663 		/* But since DNAME records are also subdomains of the zone,
664 		 * same check can be used */
665 
666 		if(!pkt_sub(pkt, rrset->dname, zonename)) {
667 			if(msg->an_rrsets == 0 &&
668 				rrset->type == LDNS_RR_TYPE_NS &&
669 				rrset->section == LDNS_SECTION_AUTHORITY &&
670 				FLAGS_GET_RCODE(msg->flags) ==
671 				LDNS_RCODE_NOERROR && !soa_in_auth(msg) &&
672 				sub_of_pkt(pkt, zonename, rrset->dname)) {
673 				/* noerror, nodata and this NS rrset is above
674 				 * the zone. This is LAME!
675 				 * Leave in the NS for lame classification. */
676 				/* remove everything from the additional
677 				 * (we dont want its glue that was approved
678 				 * during the normalize action) */
679 				del_addi = 1;
680 			} else if(!env->cfg->harden_glue) {
681 				/* store in cache! Since it is relevant
682 				 * (from normalize) it will be picked up
683 				 * from the cache to be used later */
684 				store_rrset(pkt, msg, env, rrset);
685 				remove_rrset("sanitize: storing potential "
686 				"poison RRset:", pkt, msg, prev, &rrset);
687 				continue;
688 			} else {
689 				if(has_additional(rrset->type)) del_addi = 1;
690 				remove_rrset("sanitize: removing potential "
691 				"poison RRset:", pkt, msg, prev, &rrset);
692 				continue;
693 			}
694 		}
695 		if(del_addi && rrset->section == LDNS_SECTION_ADDITIONAL) {
696 			remove_rrset("sanitize: removing potential "
697 			"poison reference RRset:", pkt, msg, prev, &rrset);
698 			continue;
699 		}
700 		/* check if right hand side of NSEC is within zone */
701 		if(rrset->type == LDNS_RR_TYPE_NSEC &&
702 			sanitize_nsec_is_overreach(rrset, zonename)) {
703 			remove_rrset("sanitize: removing overreaching NSEC "
704 				"RRset:", pkt, msg, prev, &rrset);
705 			continue;
706 		}
707 		prev = rrset;
708 		rrset = rrset->rrset_all_next;
709 	}
710 	return 1;
711 }
712 
713 int
714 scrub_message(ldns_buffer* pkt, struct msg_parse* msg,
715 	struct query_info* qinfo, uint8_t* zonename, struct regional* region,
716 	struct module_env* env, struct iter_env* ie)
717 {
718 	/* basic sanity checks */
719 	log_nametypeclass(VERB_ALGO, "scrub for", zonename, LDNS_RR_TYPE_NS,
720 		qinfo->qclass);
721 	if(msg->qdcount > 1)
722 		return 0;
723 	if( !(msg->flags&BIT_QR) )
724 		return 0;
725 	msg->flags &= ~(BIT_AD|BIT_Z); /* force off bit AD and Z */
726 
727 	/* make sure that a query is echoed back when NOERROR or NXDOMAIN */
728 	/* this is not required for basic operation but is a forgery
729 	 * resistance (security) feature */
730 	if((FLAGS_GET_RCODE(msg->flags) == LDNS_RCODE_NOERROR ||
731 		FLAGS_GET_RCODE(msg->flags) == LDNS_RCODE_NXDOMAIN) &&
732 		msg->qdcount == 0)
733 		return 0;
734 
735 	/* if a query is echoed back, make sure it is correct. Otherwise,
736 	 * this may be not a reply to our query. */
737 	if(msg->qdcount == 1) {
738 		if(dname_pkt_compare(pkt, msg->qname, qinfo->qname) != 0)
739 			return 0;
740 		if(msg->qtype != qinfo->qtype || msg->qclass != qinfo->qclass)
741 			return 0;
742 	}
743 
744 	/* normalize the response, this cleans up the additional.  */
745 	if(!scrub_normalize(pkt, msg, qinfo, region))
746 		return 0;
747 	/* delete all out-of-zone information */
748 	if(!scrub_sanitize(pkt, msg, qinfo, zonename, env, ie))
749 		return 0;
750 	return 1;
751 }
752