xref: /freebsd/contrib/unbound/iterator/iter_scrub.c (revision 389e4940069316fe667ffa263fa7d6390d0a960f)
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
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 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 #include "sldns/sbuffer.h"
57 
58 /** RRset flag used during scrubbing. The RRset is OK. */
59 #define RRSET_SCRUB_OK	0x80
60 
61 /** remove rrset, update loop variables */
62 static void
63 remove_rrset(const char* str, sldns_buffer* pkt, struct msg_parse* msg,
64 	struct rrset_parse* prev, struct rrset_parse** rrset)
65 {
66 	if(verbosity >= VERB_QUERY && str
67 		&& (*rrset)->dname_len <= LDNS_MAX_DOMAINLEN) {
68 		uint8_t buf[LDNS_MAX_DOMAINLEN+1];
69 		dname_pkt_copy(pkt, buf, (*rrset)->dname);
70 		log_nametypeclass(VERB_QUERY, str, buf,
71 			(*rrset)->type, ntohs((*rrset)->rrset_class));
72 	}
73 	if(prev)
74 		prev->rrset_all_next = (*rrset)->rrset_all_next;
75 	else	msg->rrset_first = (*rrset)->rrset_all_next;
76 	if(msg->rrset_last == *rrset)
77 		msg->rrset_last = prev;
78 	msg->rrset_count --;
79 	switch((*rrset)->section) {
80 		case LDNS_SECTION_ANSWER: msg->an_rrsets--; break;
81 		case LDNS_SECTION_AUTHORITY: msg->ns_rrsets--; break;
82 		case LDNS_SECTION_ADDITIONAL: msg->ar_rrsets--; break;
83 		default: log_assert(0);
84 	}
85 	msgparse_bucket_remove(msg, *rrset);
86 	*rrset = (*rrset)->rrset_all_next;
87 }
88 
89 /** return true if rr type has additional names in it */
90 static int
91 has_additional(uint16_t t)
92 {
93 	switch(t) {
94 		case LDNS_RR_TYPE_MB:
95 		case LDNS_RR_TYPE_MD:
96 		case LDNS_RR_TYPE_MF:
97 		case LDNS_RR_TYPE_NS:
98 		case LDNS_RR_TYPE_MX:
99 		case LDNS_RR_TYPE_KX:
100 		case LDNS_RR_TYPE_SRV:
101 			return 1;
102 		case LDNS_RR_TYPE_NAPTR:
103 			/* TODO: NAPTR not supported, glue stripped off */
104 			return 0;
105 	}
106 	return 0;
107 }
108 
109 /** get additional name from rrset RR, return false if no name present */
110 static int
111 get_additional_name(struct rrset_parse* rrset, struct rr_parse* rr,
112 	uint8_t** nm, size_t* nmlen, sldns_buffer* pkt)
113 {
114 	size_t offset = 0;
115 	size_t len, oldpos;
116 	switch(rrset->type) {
117 		case LDNS_RR_TYPE_MB:
118 		case LDNS_RR_TYPE_MD:
119 		case LDNS_RR_TYPE_MF:
120 		case LDNS_RR_TYPE_NS:
121 			offset = 0;
122 			break;
123 		case LDNS_RR_TYPE_MX:
124 		case LDNS_RR_TYPE_KX:
125 			offset = 2;
126 			break;
127 		case LDNS_RR_TYPE_SRV:
128 			offset = 6;
129 			break;
130 		case LDNS_RR_TYPE_NAPTR:
131 			/* TODO: NAPTR not supported, glue stripped off */
132 			return 0;
133 		default:
134 			return 0;
135 	}
136 	len = sldns_read_uint16(rr->ttl_data+sizeof(uint32_t));
137 	if(len < offset+1)
138 		return 0; /* rdata field too small */
139 	*nm = rr->ttl_data+sizeof(uint32_t)+sizeof(uint16_t)+offset;
140 	oldpos = sldns_buffer_position(pkt);
141 	sldns_buffer_set_position(pkt, (size_t)(*nm - sldns_buffer_begin(pkt)));
142 	*nmlen = pkt_dname_len(pkt);
143 	sldns_buffer_set_position(pkt, oldpos);
144 	if(*nmlen == 0)
145 		return 0;
146 	return 1;
147 }
148 
149 /** Place mark on rrsets in additional section they are OK */
150 static void
151 mark_additional_rrset(sldns_buffer* pkt, struct msg_parse* msg,
152 	struct rrset_parse* rrset)
153 {
154 	/* Mark A and AAAA for NS as appropriate additional section info. */
155 	uint8_t* nm = NULL;
156 	size_t nmlen = 0;
157 	struct rr_parse* rr;
158 
159 	if(!has_additional(rrset->type))
160 		return;
161 	for(rr = rrset->rr_first; rr; rr = rr->next) {
162 		if(get_additional_name(rrset, rr, &nm, &nmlen, pkt)) {
163 			/* mark A */
164 			hashvalue_type h = pkt_hash_rrset(pkt, nm,
165 				LDNS_RR_TYPE_A, rrset->rrset_class, 0);
166 			struct rrset_parse* r = msgparse_hashtable_lookup(
167 				msg, pkt, h, 0, nm, nmlen,
168 				LDNS_RR_TYPE_A, rrset->rrset_class);
169 			if(r && r->section == LDNS_SECTION_ADDITIONAL) {
170 				r->flags |= RRSET_SCRUB_OK;
171 			}
172 
173 			/* mark AAAA */
174 			h = pkt_hash_rrset(pkt, nm, LDNS_RR_TYPE_AAAA,
175 				rrset->rrset_class, 0);
176 			r = msgparse_hashtable_lookup(msg, pkt, h, 0, nm,
177 				nmlen, LDNS_RR_TYPE_AAAA, rrset->rrset_class);
178 			if(r && r->section == LDNS_SECTION_ADDITIONAL) {
179 				r->flags |= RRSET_SCRUB_OK;
180 			}
181 		}
182 	}
183 }
184 
185 /** Get target name of a CNAME */
186 static int
187 parse_get_cname_target(struct rrset_parse* rrset, uint8_t** sname,
188 	size_t* snamelen)
189 {
190 	if(rrset->rr_count != 1) {
191 		struct rr_parse* sig;
192 		verbose(VERB_ALGO, "Found CNAME rrset with "
193 			"size > 1: %u", (unsigned)rrset->rr_count);
194 		/* use the first CNAME! */
195 		rrset->rr_count = 1;
196 		rrset->size = rrset->rr_first->size;
197 		for(sig=rrset->rrsig_first; sig; sig=sig->next)
198 			rrset->size += sig->size;
199 		rrset->rr_last = rrset->rr_first;
200 		rrset->rr_first->next = NULL;
201 	}
202 	if(rrset->rr_first->size < sizeof(uint16_t)+1)
203 		return 0; /* CNAME rdata too small */
204 	*sname = rrset->rr_first->ttl_data + sizeof(uint32_t)
205 		+ sizeof(uint16_t); /* skip ttl, rdatalen */
206 	*snamelen = rrset->rr_first->size - sizeof(uint16_t);
207 	return 1;
208 }
209 
210 /** Synthesize CNAME from DNAME, false if too long */
211 static int
212 synth_cname(uint8_t* qname, size_t qnamelen, struct rrset_parse* dname_rrset,
213 	uint8_t* alias, size_t* aliaslen, sldns_buffer* pkt)
214 {
215 	/* we already know that sname is a strict subdomain of DNAME owner */
216 	uint8_t* dtarg = NULL;
217 	size_t dtarglen;
218 	if(!parse_get_cname_target(dname_rrset, &dtarg, &dtarglen))
219 		return 0;
220 	log_assert(qnamelen > dname_rrset->dname_len);
221 	/* DNAME from com. to net. with qname example.com. -> example.net. */
222 	/* so: \3com\0 to \3net\0 and qname \7example\3com\0 */
223 	*aliaslen = qnamelen + dtarglen - dname_rrset->dname_len;
224 	if(*aliaslen > LDNS_MAX_DOMAINLEN)
225 		return 0; /* should have been RCODE YXDOMAIN */
226 	/* decompress dnames into buffer, we know it fits */
227 	dname_pkt_copy(pkt, alias, qname);
228 	dname_pkt_copy(pkt, alias+(qnamelen-dname_rrset->dname_len), dtarg);
229 	return 1;
230 }
231 
232 /** synthesize a CNAME rrset */
233 static struct rrset_parse*
234 synth_cname_rrset(uint8_t** sname, size_t* snamelen, uint8_t* alias,
235 	size_t aliaslen, struct regional* region, struct msg_parse* msg,
236 	struct rrset_parse* rrset, struct rrset_parse* prev,
237 	struct rrset_parse* nx, sldns_buffer* pkt)
238 {
239 	struct rrset_parse* cn = (struct rrset_parse*)regional_alloc(region,
240 		sizeof(struct rrset_parse));
241 	if(!cn)
242 		return NULL;
243 	memset(cn, 0, sizeof(*cn));
244 	cn->rr_first = (struct rr_parse*)regional_alloc(region,
245 		sizeof(struct rr_parse));
246 	if(!cn->rr_first)
247 		return NULL;
248 	cn->rr_last = cn->rr_first;
249 	/* CNAME from sname to alias */
250 	cn->dname = (uint8_t*)regional_alloc(region, *snamelen);
251 	if(!cn->dname)
252 		return NULL;
253 	dname_pkt_copy(pkt, cn->dname, *sname);
254 	cn->dname_len = *snamelen;
255 	cn->type = LDNS_RR_TYPE_CNAME;
256 	cn->section = rrset->section;
257 	cn->rrset_class = rrset->rrset_class;
258 	cn->rr_count = 1;
259 	cn->size = sizeof(uint16_t) + aliaslen;
260 	cn->hash=pkt_hash_rrset(pkt, cn->dname, cn->type, cn->rrset_class, 0);
261 	/* allocate TTL + rdatalen + uncompressed dname */
262 	memset(cn->rr_first, 0, sizeof(struct rr_parse));
263 	cn->rr_first->outside_packet = 1;
264 	cn->rr_first->ttl_data = (uint8_t*)regional_alloc(region,
265 		sizeof(uint32_t)+sizeof(uint16_t)+aliaslen);
266 	if(!cn->rr_first->ttl_data)
267 		return NULL;
268 	sldns_write_uint32(cn->rr_first->ttl_data, 0); /* TTL = 0 */
269 	sldns_write_uint16(cn->rr_first->ttl_data+4, aliaslen);
270 	memmove(cn->rr_first->ttl_data+6, alias, aliaslen);
271 	cn->rr_first->size = sizeof(uint16_t)+aliaslen;
272 
273 	/* link it in */
274 	cn->rrset_all_next = nx;
275 	if(prev)
276 		prev->rrset_all_next = cn;
277 	else	msg->rrset_first = cn;
278 	if(nx == NULL)
279 		msg->rrset_last = cn;
280 	msg->rrset_count ++;
281 	msg->an_rrsets++;
282 	/* it is not inserted in the msg hashtable. */
283 
284 	*sname = cn->rr_first->ttl_data + sizeof(uint32_t)+sizeof(uint16_t);
285 	*snamelen = aliaslen;
286 	return cn;
287 }
288 
289 /** check if DNAME applies to a name */
290 static int
291 pkt_strict_sub(sldns_buffer* pkt, uint8_t* sname, uint8_t* dr)
292 {
293 	uint8_t buf1[LDNS_MAX_DOMAINLEN+1];
294 	uint8_t buf2[LDNS_MAX_DOMAINLEN+1];
295 	/* decompress names */
296 	dname_pkt_copy(pkt, buf1, sname);
297 	dname_pkt_copy(pkt, buf2, dr);
298 	return dname_strict_subdomain_c(buf1, buf2);
299 }
300 
301 /** check subdomain with decompression */
302 static int
303 pkt_sub(sldns_buffer* pkt, uint8_t* comprname, uint8_t* zone)
304 {
305 	uint8_t buf[LDNS_MAX_DOMAINLEN+1];
306 	dname_pkt_copy(pkt, buf, comprname);
307 	return dname_subdomain_c(buf, zone);
308 }
309 
310 /** check subdomain with decompression, compressed is parent */
311 static int
312 sub_of_pkt(sldns_buffer* pkt, uint8_t* zone, uint8_t* comprname)
313 {
314 	uint8_t buf[LDNS_MAX_DOMAINLEN+1];
315 	dname_pkt_copy(pkt, buf, comprname);
316 	return dname_subdomain_c(zone, buf);
317 }
318 
319 /**
320  * This routine normalizes a response. This includes removing "irrelevant"
321  * records from the answer and additional sections and (re)synthesizing
322  * CNAMEs from DNAMEs, if present.
323  *
324  * @param pkt: packet.
325  * @param msg: msg to normalize.
326  * @param qinfo: original query.
327  * @param region: where to allocate synthesized CNAMEs.
328  * @return 0 on error.
329  */
330 static int
331 scrub_normalize(sldns_buffer* pkt, struct msg_parse* msg,
332 	struct query_info* qinfo, struct regional* region)
333 {
334 	uint8_t* sname = qinfo->qname;
335 	size_t snamelen = qinfo->qname_len;
336 	struct rrset_parse* rrset, *prev, *nsset=NULL;
337 
338 	if(FLAGS_GET_RCODE(msg->flags) != LDNS_RCODE_NOERROR &&
339 		FLAGS_GET_RCODE(msg->flags) != LDNS_RCODE_NXDOMAIN)
340 		return 1;
341 
342 	/* For the ANSWER section, remove all "irrelevant" records and add
343 	 * synthesized CNAMEs from DNAMEs
344 	 * This will strip out-of-order CNAMEs as well. */
345 
346 	/* walk through the parse packet rrset list, keep track of previous
347 	 * for insert and delete ease, and examine every RRset */
348 	prev = NULL;
349 	rrset = msg->rrset_first;
350 	while(rrset && rrset->section == LDNS_SECTION_ANSWER) {
351 		if(rrset->type == LDNS_RR_TYPE_DNAME &&
352 			pkt_strict_sub(pkt, sname, rrset->dname)) {
353 			/* check if next rrset is correct CNAME. else,
354 			 * synthesize a CNAME */
355 			struct rrset_parse* nx = rrset->rrset_all_next;
356 			uint8_t alias[LDNS_MAX_DOMAINLEN+1];
357 			size_t aliaslen = 0;
358 			if(rrset->rr_count != 1) {
359 				verbose(VERB_ALGO, "Found DNAME rrset with "
360 					"size > 1: %u",
361 					(unsigned)rrset->rr_count);
362 				return 0;
363 			}
364 			if(!synth_cname(sname, snamelen, rrset, alias,
365 				&aliaslen, pkt)) {
366 				verbose(VERB_ALGO, "synthesized CNAME "
367 					"too long");
368 				return 0;
369 			}
370 			if(nx && nx->type == LDNS_RR_TYPE_CNAME &&
371 			   dname_pkt_compare(pkt, sname, nx->dname) == 0) {
372 				/* check next cname */
373 				uint8_t* t = NULL;
374 				size_t tlen = 0;
375 				if(!parse_get_cname_target(nx, &t, &tlen))
376 					return 0;
377 				if(dname_pkt_compare(pkt, alias, t) == 0) {
378 					/* it's OK and better capitalized */
379 					prev = rrset;
380 					rrset = nx;
381 					continue;
382 				}
383 				/* synth ourselves */
384 			}
385 			/* synth a CNAME rrset */
386 			prev = synth_cname_rrset(&sname, &snamelen, alias,
387 				aliaslen, region, msg, rrset, rrset, nx, pkt);
388 			if(!prev) {
389 				log_err("out of memory synthesizing CNAME");
390 				return 0;
391 			}
392 			/* FIXME: resolve the conflict between synthesized
393 			 * CNAME ttls and the cache. */
394 			rrset = nx;
395 			continue;
396 
397 		}
398 
399 		/* The only records in the ANSWER section not allowed to */
400 		if(dname_pkt_compare(pkt, sname, rrset->dname) != 0) {
401 			remove_rrset("normalize: removing irrelevant RRset:",
402 				pkt, msg, prev, &rrset);
403 			continue;
404 		}
405 
406 		/* Follow the CNAME chain. */
407 		if(rrset->type == LDNS_RR_TYPE_CNAME) {
408 			struct rrset_parse* nx = rrset->rrset_all_next;
409 			uint8_t* oldsname = sname;
410 			/* see if the next one is a DNAME, if so, swap them */
411 			if(nx && nx->section == LDNS_SECTION_ANSWER &&
412 				nx->type == LDNS_RR_TYPE_DNAME &&
413 				nx->rr_count == 1 &&
414 				pkt_strict_sub(pkt, sname, nx->dname)) {
415 				/* there is a DNAME after this CNAME, it
416 				 * is in the ANSWER section, and the DNAME
417 				 * applies to the name we cover */
418 				/* check if the alias of the DNAME equals
419 				 * this CNAME */
420 				uint8_t alias[LDNS_MAX_DOMAINLEN+1];
421 				size_t aliaslen = 0;
422 				uint8_t* t = NULL;
423 				size_t tlen = 0;
424 				if(synth_cname(sname, snamelen, nx, alias,
425 					&aliaslen, pkt) &&
426 					parse_get_cname_target(rrset, &t, &tlen) &&
427 			   		dname_pkt_compare(pkt, alias, t) == 0) {
428 					/* the synthesized CNAME equals the
429 					 * current CNAME.  This CNAME is the
430 					 * one that the DNAME creates, and this
431 					 * CNAME is better capitalised */
432 					verbose(VERB_ALGO, "normalize: re-order of DNAME and its CNAME");
433 					if(prev) prev->rrset_all_next = nx;
434 					else msg->rrset_first = nx;
435 					if(nx->rrset_all_next == NULL)
436 						msg->rrset_last = rrset;
437 					rrset->rrset_all_next =
438 						nx->rrset_all_next;
439 					nx->rrset_all_next = rrset;
440 					prev = nx;
441 				}
442 			}
443 
444 			/* move to next name in CNAME chain */
445 			if(!parse_get_cname_target(rrset, &sname, &snamelen))
446 				return 0;
447 			prev = rrset;
448 			rrset = rrset->rrset_all_next;
449 			/* in CNAME ANY response, can have data after CNAME */
450 			if(qinfo->qtype == LDNS_RR_TYPE_ANY) {
451 				while(rrset && rrset->section ==
452 					LDNS_SECTION_ANSWER &&
453 					dname_pkt_compare(pkt, oldsname,
454 					rrset->dname) == 0) {
455 					prev = rrset;
456 					rrset = rrset->rrset_all_next;
457 				}
458 			}
459 			continue;
460 		}
461 
462 		/* Otherwise, make sure that the RRset matches the qtype. */
463 		if(qinfo->qtype != LDNS_RR_TYPE_ANY &&
464 			qinfo->qtype != rrset->type) {
465 			remove_rrset("normalize: removing irrelevant RRset:",
466 				pkt, msg, prev, &rrset);
467 			continue;
468 		}
469 
470 		/* Mark the additional names from relevant rrset as OK. */
471 		/* only for RRsets that match the query name, other ones
472 		 * will be removed by sanitize, so no additional for them */
473 		if(dname_pkt_compare(pkt, qinfo->qname, rrset->dname) == 0)
474 			mark_additional_rrset(pkt, msg, rrset);
475 
476 		prev = rrset;
477 		rrset = rrset->rrset_all_next;
478 	}
479 
480 	/* Mark additional names from AUTHORITY */
481 	while(rrset && rrset->section == LDNS_SECTION_AUTHORITY) {
482 		if(rrset->type==LDNS_RR_TYPE_DNAME ||
483 			rrset->type==LDNS_RR_TYPE_CNAME ||
484 			rrset->type==LDNS_RR_TYPE_A ||
485 			rrset->type==LDNS_RR_TYPE_AAAA) {
486 			remove_rrset("normalize: removing irrelevant "
487 				"RRset:", pkt, msg, prev, &rrset);
488 			continue;
489 		}
490 		/* only one NS set allowed in authority section */
491 		if(rrset->type==LDNS_RR_TYPE_NS) {
492 			/* NS set must be pertinent to the query */
493 			if(!sub_of_pkt(pkt, qinfo->qname, rrset->dname)) {
494 				remove_rrset("normalize: removing irrelevant "
495 					"RRset:", pkt, msg, prev, &rrset);
496 				continue;
497 			}
498 			if(nsset == NULL) {
499 				nsset = rrset;
500 			} else {
501 				remove_rrset("normalize: removing irrelevant "
502 					"RRset:", pkt, msg, prev, &rrset);
503 				continue;
504 			}
505 		}
506 		/* if this is type DS and we query for type DS we just got
507 		 * a referral answer for our type DS query, fix packet */
508 		if(rrset->type==LDNS_RR_TYPE_DS &&
509 			qinfo->qtype == LDNS_RR_TYPE_DS &&
510 			dname_pkt_compare(pkt, qinfo->qname, rrset->dname) == 0) {
511 			rrset->section = LDNS_SECTION_ANSWER;
512 			msg->ancount = rrset->rr_count + rrset->rrsig_count;
513 			msg->nscount = 0;
514 			msg->arcount = 0;
515 			msg->an_rrsets = 1;
516 			msg->ns_rrsets = 0;
517 			msg->ar_rrsets = 0;
518 			msg->rrset_count = 1;
519 			msg->rrset_first = rrset;
520 			msg->rrset_last = rrset;
521 			rrset->rrset_all_next = NULL;
522 			return 1;
523 		}
524 		mark_additional_rrset(pkt, msg, rrset);
525 		prev = rrset;
526 		rrset = rrset->rrset_all_next;
527 	}
528 
529 	/* For each record in the additional section, remove it if it is an
530 	 * address record and not in the collection of additional names
531 	 * found in ANSWER and AUTHORITY. */
532 	/* These records have not been marked OK previously */
533 	while(rrset && rrset->section == LDNS_SECTION_ADDITIONAL) {
534 		/* FIXME: what about other types? */
535 		if(rrset->type==LDNS_RR_TYPE_A ||
536 			rrset->type==LDNS_RR_TYPE_AAAA)
537 		{
538 			if((rrset->flags & RRSET_SCRUB_OK)) {
539 				/* remove flag to clean up flags variable */
540 				rrset->flags &= ~RRSET_SCRUB_OK;
541 			} else {
542 				remove_rrset("normalize: removing irrelevant "
543 					"RRset:", pkt, msg, prev, &rrset);
544 				continue;
545 			}
546 		}
547 		if(rrset->type==LDNS_RR_TYPE_DNAME ||
548 			rrset->type==LDNS_RR_TYPE_CNAME ||
549 			rrset->type==LDNS_RR_TYPE_NS) {
550 			remove_rrset("normalize: removing irrelevant "
551 				"RRset:", pkt, msg, prev, &rrset);
552 			continue;
553 		}
554 		prev = rrset;
555 		rrset = rrset->rrset_all_next;
556 	}
557 
558 	return 1;
559 }
560 
561 /**
562  * Store potential poison in the cache (only if hardening disabled).
563  * The rrset is stored in the cache but removed from the message.
564  * So that it will be used for infrastructure purposes, but not be
565  * returned to the client.
566  * @param pkt: packet
567  * @param msg: message parsed
568  * @param env: environment with cache
569  * @param rrset: to store.
570  */
571 static void
572 store_rrset(sldns_buffer* pkt, struct msg_parse* msg, struct module_env* env,
573 	struct rrset_parse* rrset)
574 {
575 	struct ub_packed_rrset_key* k;
576 	struct packed_rrset_data* d;
577 	struct rrset_ref ref;
578 	time_t now = *env->now;
579 
580 	k = alloc_special_obtain(env->alloc);
581 	if(!k)
582 		return;
583 	k->entry.data = NULL;
584 	if(!parse_copy_decompress_rrset(pkt, msg, rrset, NULL, k)) {
585 		alloc_special_release(env->alloc, k);
586 		return;
587 	}
588 	d = (struct packed_rrset_data*)k->entry.data;
589 	packed_rrset_ttl_add(d, now);
590 	ref.key = k;
591 	ref.id = k->id;
592 	/*ignore ret: it was in the cache, ref updated */
593 	(void)rrset_cache_update(env->rrset_cache, &ref, env->alloc, now);
594 }
595 
596 /** Check if there are SOA records in the authority section (negative) */
597 static int
598 soa_in_auth(struct msg_parse* msg)
599 {
600 	struct rrset_parse* rrset;
601 	for(rrset = msg->rrset_first; rrset; rrset = rrset->rrset_all_next)
602 		if(rrset->type == LDNS_RR_TYPE_SOA &&
603 			rrset->section == LDNS_SECTION_AUTHORITY)
604 			return 1;
605 	return 0;
606 }
607 
608 /**
609  * Check if right hand name in NSEC is within zone
610  * @param rrset: the NSEC rrset
611  * @param zonename: the zone name.
612  * @return true if BAD.
613  */
614 static int sanitize_nsec_is_overreach(struct rrset_parse* rrset,
615 	uint8_t* zonename)
616 {
617 	struct rr_parse* rr;
618 	uint8_t* rhs;
619 	size_t len;
620 	log_assert(rrset->type == LDNS_RR_TYPE_NSEC);
621 	for(rr = rrset->rr_first; rr; rr = rr->next) {
622 		rhs = rr->ttl_data+4+2;
623 		len = sldns_read_uint16(rr->ttl_data+4);
624 		if(!dname_valid(rhs, len)) {
625 			/* malformed domain name in rdata */
626 			return 1;
627 		}
628 		if(!dname_subdomain_c(rhs, zonename)) {
629 			/* overreaching */
630 			return 1;
631 		}
632 	}
633 	/* all NSEC RRs OK */
634 	return 0;
635 }
636 
637 /**
638  * Given a response event, remove suspect RRsets from the response.
639  * "Suspect" rrsets are potentially poison. Note that this routine expects
640  * the response to be in a "normalized" state -- that is, all "irrelevant"
641  * RRsets have already been removed, CNAMEs are in order, etc.
642  *
643  * @param pkt: packet.
644  * @param msg: msg to normalize.
645  * @param qinfo: the question originally asked.
646  * @param zonename: name of server zone.
647  * @param env: module environment with config and cache.
648  * @param ie: iterator environment with private address data.
649  * @return 0 on error.
650  */
651 static int
652 scrub_sanitize(sldns_buffer* pkt, struct msg_parse* msg,
653 	struct query_info* qinfo, uint8_t* zonename, struct module_env* env,
654 	struct iter_env* ie)
655 {
656 	int del_addi = 0; /* if additional-holding rrsets are deleted, we
657 		do not trust the normalized additional-A-AAAA any more */
658 	struct rrset_parse* rrset, *prev;
659 	prev = NULL;
660 	rrset = msg->rrset_first;
661 
662 	/* the first DNAME is allowed to stay. It needs checking before
663 	 * it can be used from the cache. After normalization, an initial
664 	 * DNAME will have a correctly synthesized CNAME after it. */
665 	if(rrset && rrset->type == LDNS_RR_TYPE_DNAME &&
666 		rrset->section == LDNS_SECTION_ANSWER &&
667 		pkt_strict_sub(pkt, qinfo->qname, rrset->dname) &&
668 		pkt_sub(pkt, rrset->dname, zonename)) {
669 		prev = rrset; /* DNAME allowed to stay in answer section */
670 		rrset = rrset->rrset_all_next;
671 	}
672 
673 	/* remove all records from the answer section that are
674 	 * not the same domain name as the query domain name.
675 	 * The answer section should contain rrsets with the same name
676 	 * as the question. For DNAMEs a CNAME has been synthesized.
677 	 * Wildcards have the query name in answer section.
678 	 * ANY queries get query name in answer section.
679 	 * Remainders of CNAME chains are cut off and resolved by iterator. */
680 	while(rrset && rrset->section == LDNS_SECTION_ANSWER) {
681 		if(dname_pkt_compare(pkt, qinfo->qname, rrset->dname) != 0) {
682 			if(has_additional(rrset->type)) del_addi = 1;
683 			remove_rrset("sanitize: removing extraneous answer "
684 				"RRset:", pkt, msg, prev, &rrset);
685 			continue;
686 		}
687 		prev = rrset;
688 		rrset = rrset->rrset_all_next;
689 	}
690 
691 	/* At this point, we brutally remove ALL rrsets that aren't
692 	 * children of the originating zone. The idea here is that,
693 	 * as far as we know, the server that we contacted is ONLY
694 	 * authoritative for the originating zone. It, of course, MAY
695 	 * be authoritative for any other zones, and of course, MAY
696 	 * NOT be authoritative for some subdomains of the originating
697 	 * zone. */
698 	prev = NULL;
699 	rrset = msg->rrset_first;
700 	while(rrset) {
701 
702 		/* remove private addresses */
703 		if( (rrset->type == LDNS_RR_TYPE_A ||
704 			rrset->type == LDNS_RR_TYPE_AAAA)) {
705 
706 			/* do not set servfail since this leads to too
707 			 * many drops of other people using rfc1918 space */
708 			/* also do not remove entire rrset, unless all records
709 			 * in it are bad */
710 			if(priv_rrset_bad(ie->priv, pkt, rrset)) {
711 				remove_rrset(NULL, pkt, msg, prev, &rrset);
712 				continue;
713 			}
714 		}
715 
716 		/* skip DNAME records -- they will always be followed by a
717 		 * synthesized CNAME, which will be relevant.
718 		 * FIXME: should this do something differently with DNAME
719 		 * rrsets NOT in Section.ANSWER? */
720 		/* But since DNAME records are also subdomains of the zone,
721 		 * same check can be used */
722 
723 		if(!pkt_sub(pkt, rrset->dname, zonename)) {
724 			if(msg->an_rrsets == 0 &&
725 				rrset->type == LDNS_RR_TYPE_NS &&
726 				rrset->section == LDNS_SECTION_AUTHORITY &&
727 				FLAGS_GET_RCODE(msg->flags) ==
728 				LDNS_RCODE_NOERROR && !soa_in_auth(msg) &&
729 				sub_of_pkt(pkt, zonename, rrset->dname)) {
730 				/* noerror, nodata and this NS rrset is above
731 				 * the zone. This is LAME!
732 				 * Leave in the NS for lame classification. */
733 				/* remove everything from the additional
734 				 * (we dont want its glue that was approved
735 				 * during the normalize action) */
736 				del_addi = 1;
737 			} else if(!env->cfg->harden_glue && (
738 				rrset->type == LDNS_RR_TYPE_A ||
739 				rrset->type == LDNS_RR_TYPE_AAAA)) {
740 				/* store in cache! Since it is relevant
741 				 * (from normalize) it will be picked up
742 				 * from the cache to be used later */
743 				store_rrset(pkt, msg, env, rrset);
744 				remove_rrset("sanitize: storing potential "
745 				"poison RRset:", pkt, msg, prev, &rrset);
746 				continue;
747 			} else {
748 				if(has_additional(rrset->type)) del_addi = 1;
749 				remove_rrset("sanitize: removing potential "
750 				"poison RRset:", pkt, msg, prev, &rrset);
751 				continue;
752 			}
753 		}
754 		if(del_addi && rrset->section == LDNS_SECTION_ADDITIONAL) {
755 			remove_rrset("sanitize: removing potential "
756 			"poison reference RRset:", pkt, msg, prev, &rrset);
757 			continue;
758 		}
759 		/* check if right hand side of NSEC is within zone */
760 		if(rrset->type == LDNS_RR_TYPE_NSEC &&
761 			sanitize_nsec_is_overreach(rrset, zonename)) {
762 			remove_rrset("sanitize: removing overreaching NSEC "
763 				"RRset:", pkt, msg, prev, &rrset);
764 			continue;
765 		}
766 		prev = rrset;
767 		rrset = rrset->rrset_all_next;
768 	}
769 	return 1;
770 }
771 
772 int
773 scrub_message(sldns_buffer* pkt, struct msg_parse* msg,
774 	struct query_info* qinfo, uint8_t* zonename, struct regional* region,
775 	struct module_env* env, struct iter_env* ie)
776 {
777 	/* basic sanity checks */
778 	log_nametypeclass(VERB_ALGO, "scrub for", zonename, LDNS_RR_TYPE_NS,
779 		qinfo->qclass);
780 	if(msg->qdcount > 1)
781 		return 0;
782 	if( !(msg->flags&BIT_QR) )
783 		return 0;
784 	msg->flags &= ~(BIT_AD|BIT_Z); /* force off bit AD and Z */
785 
786 	/* make sure that a query is echoed back when NOERROR or NXDOMAIN */
787 	/* this is not required for basic operation but is a forgery
788 	 * resistance (security) feature */
789 	if((FLAGS_GET_RCODE(msg->flags) == LDNS_RCODE_NOERROR ||
790 		FLAGS_GET_RCODE(msg->flags) == LDNS_RCODE_NXDOMAIN) &&
791 		msg->qdcount == 0)
792 		return 0;
793 
794 	/* if a query is echoed back, make sure it is correct. Otherwise,
795 	 * this may be not a reply to our query. */
796 	if(msg->qdcount == 1) {
797 		if(dname_pkt_compare(pkt, msg->qname, qinfo->qname) != 0)
798 			return 0;
799 		if(msg->qtype != qinfo->qtype || msg->qclass != qinfo->qclass)
800 			return 0;
801 	}
802 
803 	/* normalize the response, this cleans up the additional.  */
804 	if(!scrub_normalize(pkt, msg, qinfo, region))
805 		return 0;
806 	/* delete all out-of-zone information */
807 	if(!scrub_sanitize(pkt, msg, qinfo, zonename, env, ie))
808 		return 0;
809 	return 1;
810 }
811