iter_scrub.c revision 356345
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 */
62static void
63remove_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 */
90static int
91has_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 */
110static int
111get_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 */
150static void
151mark_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 */
186static int
187parse_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 */
211static int
212synth_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	if(qnamelen <= dname_rrset->dname_len)
221		return 0;
222	if(qnamelen == 0)
223		return 0;
224	log_assert(qnamelen > dname_rrset->dname_len);
225	/* DNAME from com. to net. with qname example.com. -> example.net. */
226	/* so: \3com\0 to \3net\0 and qname \7example\3com\0 */
227	*aliaslen = qnamelen + dtarglen - dname_rrset->dname_len;
228	if(*aliaslen > LDNS_MAX_DOMAINLEN)
229		return 0; /* should have been RCODE YXDOMAIN */
230	/* decompress dnames into buffer, we know it fits */
231	dname_pkt_copy(pkt, alias, qname);
232	dname_pkt_copy(pkt, alias+(qnamelen-dname_rrset->dname_len), dtarg);
233	return 1;
234}
235
236/** synthesize a CNAME rrset */
237static struct rrset_parse*
238synth_cname_rrset(uint8_t** sname, size_t* snamelen, uint8_t* alias,
239	size_t aliaslen, struct regional* region, struct msg_parse* msg,
240	struct rrset_parse* rrset, struct rrset_parse* prev,
241	struct rrset_parse* nx, sldns_buffer* pkt)
242{
243	struct rrset_parse* cn = (struct rrset_parse*)regional_alloc(region,
244		sizeof(struct rrset_parse));
245	if(!cn)
246		return NULL;
247	memset(cn, 0, sizeof(*cn));
248	cn->rr_first = (struct rr_parse*)regional_alloc(region,
249		sizeof(struct rr_parse));
250	if(!cn->rr_first)
251		return NULL;
252	cn->rr_last = cn->rr_first;
253	/* CNAME from sname to alias */
254	cn->dname = (uint8_t*)regional_alloc(region, *snamelen);
255	if(!cn->dname)
256		return NULL;
257	dname_pkt_copy(pkt, cn->dname, *sname);
258	cn->dname_len = *snamelen;
259	cn->type = LDNS_RR_TYPE_CNAME;
260	cn->section = rrset->section;
261	cn->rrset_class = rrset->rrset_class;
262	cn->rr_count = 1;
263	cn->size = sizeof(uint16_t) + aliaslen;
264	cn->hash=pkt_hash_rrset(pkt, cn->dname, cn->type, cn->rrset_class, 0);
265	/* allocate TTL + rdatalen + uncompressed dname */
266	memset(cn->rr_first, 0, sizeof(struct rr_parse));
267	cn->rr_first->outside_packet = 1;
268	cn->rr_first->ttl_data = (uint8_t*)regional_alloc(region,
269		sizeof(uint32_t)+sizeof(uint16_t)+aliaslen);
270	if(!cn->rr_first->ttl_data)
271		return NULL;
272	sldns_write_uint32(cn->rr_first->ttl_data, 0); /* TTL = 0 */
273	sldns_write_uint16(cn->rr_first->ttl_data+4, aliaslen);
274	memmove(cn->rr_first->ttl_data+6, alias, aliaslen);
275	cn->rr_first->size = sizeof(uint16_t)+aliaslen;
276
277	/* link it in */
278	cn->rrset_all_next = nx;
279	if(prev)
280		prev->rrset_all_next = cn;
281	else	msg->rrset_first = cn;
282	if(nx == NULL)
283		msg->rrset_last = cn;
284	msg->rrset_count ++;
285	msg->an_rrsets++;
286	/* it is not inserted in the msg hashtable. */
287
288	*sname = cn->rr_first->ttl_data + sizeof(uint32_t)+sizeof(uint16_t);
289	*snamelen = aliaslen;
290	return cn;
291}
292
293/** check if DNAME applies to a name */
294static int
295pkt_strict_sub(sldns_buffer* pkt, uint8_t* sname, uint8_t* dr)
296{
297	uint8_t buf1[LDNS_MAX_DOMAINLEN+1];
298	uint8_t buf2[LDNS_MAX_DOMAINLEN+1];
299	/* decompress names */
300	dname_pkt_copy(pkt, buf1, sname);
301	dname_pkt_copy(pkt, buf2, dr);
302	return dname_strict_subdomain_c(buf1, buf2);
303}
304
305/** check subdomain with decompression */
306static int
307pkt_sub(sldns_buffer* pkt, uint8_t* comprname, uint8_t* zone)
308{
309	uint8_t buf[LDNS_MAX_DOMAINLEN+1];
310	dname_pkt_copy(pkt, buf, comprname);
311	return dname_subdomain_c(buf, zone);
312}
313
314/** check subdomain with decompression, compressed is parent */
315static int
316sub_of_pkt(sldns_buffer* pkt, uint8_t* zone, uint8_t* comprname)
317{
318	uint8_t buf[LDNS_MAX_DOMAINLEN+1];
319	dname_pkt_copy(pkt, buf, comprname);
320	return dname_subdomain_c(zone, buf);
321}
322
323/** Check if there are SOA records in the authority section (negative) */
324static int
325soa_in_auth(struct msg_parse* msg)
326{
327	struct rrset_parse* rrset;
328	for(rrset = msg->rrset_first; rrset; rrset = rrset->rrset_all_next)
329		if(rrset->type == LDNS_RR_TYPE_SOA &&
330			rrset->section == LDNS_SECTION_AUTHORITY)
331			return 1;
332	return 0;
333}
334
335/**
336 * This routine normalizes a response. This includes removing "irrelevant"
337 * records from the answer and additional sections and (re)synthesizing
338 * CNAMEs from DNAMEs, if present.
339 *
340 * @param pkt: packet.
341 * @param msg: msg to normalize.
342 * @param qinfo: original query.
343 * @param region: where to allocate synthesized CNAMEs.
344 * @return 0 on error.
345 */
346static int
347scrub_normalize(sldns_buffer* pkt, struct msg_parse* msg,
348	struct query_info* qinfo, struct regional* region)
349{
350	uint8_t* sname = qinfo->qname;
351	size_t snamelen = qinfo->qname_len;
352	struct rrset_parse* rrset, *prev, *nsset=NULL;
353
354	if(FLAGS_GET_RCODE(msg->flags) != LDNS_RCODE_NOERROR &&
355		FLAGS_GET_RCODE(msg->flags) != LDNS_RCODE_NXDOMAIN)
356		return 1;
357
358	/* For the ANSWER section, remove all "irrelevant" records and add
359	 * synthesized CNAMEs from DNAMEs
360	 * This will strip out-of-order CNAMEs as well. */
361
362	/* walk through the parse packet rrset list, keep track of previous
363	 * for insert and delete ease, and examine every RRset */
364	prev = NULL;
365	rrset = msg->rrset_first;
366	while(rrset && rrset->section == LDNS_SECTION_ANSWER) {
367		if(rrset->type == LDNS_RR_TYPE_DNAME &&
368			pkt_strict_sub(pkt, sname, rrset->dname)) {
369			/* check if next rrset is correct CNAME. else,
370			 * synthesize a CNAME */
371			struct rrset_parse* nx = rrset->rrset_all_next;
372			uint8_t alias[LDNS_MAX_DOMAINLEN+1];
373			size_t aliaslen = 0;
374			if(rrset->rr_count != 1) {
375				verbose(VERB_ALGO, "Found DNAME rrset with "
376					"size > 1: %u",
377					(unsigned)rrset->rr_count);
378				return 0;
379			}
380			if(!synth_cname(sname, snamelen, rrset, alias,
381				&aliaslen, pkt)) {
382				verbose(VERB_ALGO, "synthesized CNAME "
383					"too long");
384				return 0;
385			}
386			if(nx && nx->type == LDNS_RR_TYPE_CNAME &&
387			   dname_pkt_compare(pkt, sname, nx->dname) == 0) {
388				/* check next cname */
389				uint8_t* t = NULL;
390				size_t tlen = 0;
391				if(!parse_get_cname_target(nx, &t, &tlen))
392					return 0;
393				if(dname_pkt_compare(pkt, alias, t) == 0) {
394					/* it's OK and better capitalized */
395					prev = rrset;
396					rrset = nx;
397					continue;
398				}
399				/* synth ourselves */
400			}
401			/* synth a CNAME rrset */
402			prev = synth_cname_rrset(&sname, &snamelen, alias,
403				aliaslen, region, msg, rrset, rrset, nx, pkt);
404			if(!prev) {
405				log_err("out of memory synthesizing CNAME");
406				return 0;
407			}
408			/* FIXME: resolve the conflict between synthesized
409			 * CNAME ttls and the cache. */
410			rrset = nx;
411			continue;
412
413		}
414
415		/* The only records in the ANSWER section not allowed to */
416		if(dname_pkt_compare(pkt, sname, rrset->dname) != 0) {
417			remove_rrset("normalize: removing irrelevant RRset:",
418				pkt, msg, prev, &rrset);
419			continue;
420		}
421
422		/* Follow the CNAME chain. */
423		if(rrset->type == LDNS_RR_TYPE_CNAME) {
424			struct rrset_parse* nx = rrset->rrset_all_next;
425			uint8_t* oldsname = sname;
426			/* see if the next one is a DNAME, if so, swap them */
427			if(nx && nx->section == LDNS_SECTION_ANSWER &&
428				nx->type == LDNS_RR_TYPE_DNAME &&
429				nx->rr_count == 1 &&
430				pkt_strict_sub(pkt, sname, nx->dname)) {
431				/* there is a DNAME after this CNAME, it
432				 * is in the ANSWER section, and the DNAME
433				 * applies to the name we cover */
434				/* check if the alias of the DNAME equals
435				 * this CNAME */
436				uint8_t alias[LDNS_MAX_DOMAINLEN+1];
437				size_t aliaslen = 0;
438				uint8_t* t = NULL;
439				size_t tlen = 0;
440				if(synth_cname(sname, snamelen, nx, alias,
441					&aliaslen, pkt) &&
442					parse_get_cname_target(rrset, &t, &tlen) &&
443			   		dname_pkt_compare(pkt, alias, t) == 0) {
444					/* the synthesized CNAME equals the
445					 * current CNAME.  This CNAME is the
446					 * one that the DNAME creates, and this
447					 * CNAME is better capitalised */
448					verbose(VERB_ALGO, "normalize: re-order of DNAME and its CNAME");
449					if(prev) prev->rrset_all_next = nx;
450					else msg->rrset_first = nx;
451					if(nx->rrset_all_next == NULL)
452						msg->rrset_last = rrset;
453					rrset->rrset_all_next =
454						nx->rrset_all_next;
455					nx->rrset_all_next = rrset;
456					/* prev = nx; unused, enable if there
457					 * is other rrset removal code after
458					 * this */
459				}
460			}
461
462			/* move to next name in CNAME chain */
463			if(!parse_get_cname_target(rrset, &sname, &snamelen))
464				return 0;
465			prev = rrset;
466			rrset = rrset->rrset_all_next;
467			/* in CNAME ANY response, can have data after CNAME */
468			if(qinfo->qtype == LDNS_RR_TYPE_ANY) {
469				while(rrset && rrset->section ==
470					LDNS_SECTION_ANSWER &&
471					dname_pkt_compare(pkt, oldsname,
472					rrset->dname) == 0) {
473					prev = rrset;
474					rrset = rrset->rrset_all_next;
475				}
476			}
477			continue;
478		}
479
480		/* Otherwise, make sure that the RRset matches the qtype. */
481		if(qinfo->qtype != LDNS_RR_TYPE_ANY &&
482			qinfo->qtype != rrset->type) {
483			remove_rrset("normalize: removing irrelevant RRset:",
484				pkt, msg, prev, &rrset);
485			continue;
486		}
487
488		/* Mark the additional names from relevant rrset as OK. */
489		/* only for RRsets that match the query name, other ones
490		 * will be removed by sanitize, so no additional for them */
491		if(dname_pkt_compare(pkt, qinfo->qname, rrset->dname) == 0)
492			mark_additional_rrset(pkt, msg, rrset);
493
494		prev = rrset;
495		rrset = rrset->rrset_all_next;
496	}
497
498	/* Mark additional names from AUTHORITY */
499	while(rrset && rrset->section == LDNS_SECTION_AUTHORITY) {
500		if(rrset->type==LDNS_RR_TYPE_DNAME ||
501			rrset->type==LDNS_RR_TYPE_CNAME ||
502			rrset->type==LDNS_RR_TYPE_A ||
503			rrset->type==LDNS_RR_TYPE_AAAA) {
504			remove_rrset("normalize: removing irrelevant "
505				"RRset:", pkt, msg, prev, &rrset);
506			continue;
507		}
508		/* only one NS set allowed in authority section */
509		if(rrset->type==LDNS_RR_TYPE_NS) {
510			/* NS set must be pertinent to the query */
511			if(!sub_of_pkt(pkt, qinfo->qname, rrset->dname)) {
512				remove_rrset("normalize: removing irrelevant "
513					"RRset:", pkt, msg, prev, &rrset);
514				continue;
515			}
516			/* we don't want NS sets for NXDOMAIN answers,
517			 * because they could contain poisonous contents,
518			 * from. eg. fragmentation attacks, inserted after
519			 * long RRSIGs in the packet get to the packet
520			 * border and such */
521			/* also for NODATA answers */
522			if(FLAGS_GET_RCODE(msg->flags) == LDNS_RCODE_NXDOMAIN ||
523			   (FLAGS_GET_RCODE(msg->flags) == LDNS_RCODE_NOERROR
524			    && soa_in_auth(msg) && msg->an_rrsets == 0)) {
525				remove_rrset("normalize: removing irrelevant "
526					"RRset:", pkt, msg, prev, &rrset);
527				continue;
528			}
529			if(nsset == NULL) {
530				nsset = rrset;
531			} else {
532				remove_rrset("normalize: removing irrelevant "
533					"RRset:", pkt, msg, prev, &rrset);
534				continue;
535			}
536		}
537		/* if this is type DS and we query for type DS we just got
538		 * a referral answer for our type DS query, fix packet */
539		if(rrset->type==LDNS_RR_TYPE_DS &&
540			qinfo->qtype == LDNS_RR_TYPE_DS &&
541			dname_pkt_compare(pkt, qinfo->qname, rrset->dname) == 0) {
542			rrset->section = LDNS_SECTION_ANSWER;
543			msg->ancount = rrset->rr_count + rrset->rrsig_count;
544			msg->nscount = 0;
545			msg->arcount = 0;
546			msg->an_rrsets = 1;
547			msg->ns_rrsets = 0;
548			msg->ar_rrsets = 0;
549			msg->rrset_count = 1;
550			msg->rrset_first = rrset;
551			msg->rrset_last = rrset;
552			rrset->rrset_all_next = NULL;
553			return 1;
554		}
555		mark_additional_rrset(pkt, msg, rrset);
556		prev = rrset;
557		rrset = rrset->rrset_all_next;
558	}
559
560	/* For each record in the additional section, remove it if it is an
561	 * address record and not in the collection of additional names
562	 * found in ANSWER and AUTHORITY. */
563	/* These records have not been marked OK previously */
564	while(rrset && rrset->section == LDNS_SECTION_ADDITIONAL) {
565		/* FIXME: what about other types? */
566		if(rrset->type==LDNS_RR_TYPE_A ||
567			rrset->type==LDNS_RR_TYPE_AAAA)
568		{
569			if((rrset->flags & RRSET_SCRUB_OK)) {
570				/* remove flag to clean up flags variable */
571				rrset->flags &= ~RRSET_SCRUB_OK;
572			} else {
573				remove_rrset("normalize: removing irrelevant "
574					"RRset:", pkt, msg, prev, &rrset);
575				continue;
576			}
577		}
578		if(rrset->type==LDNS_RR_TYPE_DNAME ||
579			rrset->type==LDNS_RR_TYPE_CNAME ||
580			rrset->type==LDNS_RR_TYPE_NS) {
581			remove_rrset("normalize: removing irrelevant "
582				"RRset:", pkt, msg, prev, &rrset);
583			continue;
584		}
585		prev = rrset;
586		rrset = rrset->rrset_all_next;
587	}
588
589	return 1;
590}
591
592/**
593 * Store potential poison in the cache (only if hardening disabled).
594 * The rrset is stored in the cache but removed from the message.
595 * So that it will be used for infrastructure purposes, but not be
596 * returned to the client.
597 * @param pkt: packet
598 * @param msg: message parsed
599 * @param env: environment with cache
600 * @param rrset: to store.
601 */
602static void
603store_rrset(sldns_buffer* pkt, struct msg_parse* msg, struct module_env* env,
604	struct rrset_parse* rrset)
605{
606	struct ub_packed_rrset_key* k;
607	struct packed_rrset_data* d;
608	struct rrset_ref ref;
609	time_t now = *env->now;
610
611	k = alloc_special_obtain(env->alloc);
612	if(!k)
613		return;
614	k->entry.data = NULL;
615	if(!parse_copy_decompress_rrset(pkt, msg, rrset, NULL, k)) {
616		alloc_special_release(env->alloc, k);
617		return;
618	}
619	d = (struct packed_rrset_data*)k->entry.data;
620	packed_rrset_ttl_add(d, now);
621	ref.key = k;
622	ref.id = k->id;
623	/*ignore ret: it was in the cache, ref updated */
624	(void)rrset_cache_update(env->rrset_cache, &ref, env->alloc, now);
625}
626
627/**
628 * Check if right hand name in NSEC is within zone
629 * @param rrset: the NSEC rrset
630 * @param zonename: the zone name.
631 * @return true if BAD.
632 */
633static int sanitize_nsec_is_overreach(struct rrset_parse* rrset,
634	uint8_t* zonename)
635{
636	struct rr_parse* rr;
637	uint8_t* rhs;
638	size_t len;
639	log_assert(rrset->type == LDNS_RR_TYPE_NSEC);
640	for(rr = rrset->rr_first; rr; rr = rr->next) {
641		rhs = rr->ttl_data+4+2;
642		len = sldns_read_uint16(rr->ttl_data+4);
643		if(!dname_valid(rhs, len)) {
644			/* malformed domain name in rdata */
645			return 1;
646		}
647		if(!dname_subdomain_c(rhs, zonename)) {
648			/* overreaching */
649			return 1;
650		}
651	}
652	/* all NSEC RRs OK */
653	return 0;
654}
655
656/**
657 * Given a response event, remove suspect RRsets from the response.
658 * "Suspect" rrsets are potentially poison. Note that this routine expects
659 * the response to be in a "normalized" state -- that is, all "irrelevant"
660 * RRsets have already been removed, CNAMEs are in order, etc.
661 *
662 * @param pkt: packet.
663 * @param msg: msg to normalize.
664 * @param qinfo: the question originally asked.
665 * @param zonename: name of server zone.
666 * @param env: module environment with config and cache.
667 * @param ie: iterator environment with private address data.
668 * @return 0 on error.
669 */
670static int
671scrub_sanitize(sldns_buffer* pkt, struct msg_parse* msg,
672	struct query_info* qinfo, uint8_t* zonename, struct module_env* env,
673	struct iter_env* ie)
674{
675	int del_addi = 0; /* if additional-holding rrsets are deleted, we
676		do not trust the normalized additional-A-AAAA any more */
677	struct rrset_parse* rrset, *prev;
678	prev = NULL;
679	rrset = msg->rrset_first;
680
681	/* the first DNAME is allowed to stay. It needs checking before
682	 * it can be used from the cache. After normalization, an initial
683	 * DNAME will have a correctly synthesized CNAME after it. */
684	if(rrset && rrset->type == LDNS_RR_TYPE_DNAME &&
685		rrset->section == LDNS_SECTION_ANSWER &&
686		pkt_strict_sub(pkt, qinfo->qname, rrset->dname) &&
687		pkt_sub(pkt, rrset->dname, zonename)) {
688		prev = rrset; /* DNAME allowed to stay in answer section */
689		rrset = rrset->rrset_all_next;
690	}
691
692	/* remove all records from the answer section that are
693	 * not the same domain name as the query domain name.
694	 * The answer section should contain rrsets with the same name
695	 * as the question. For DNAMEs a CNAME has been synthesized.
696	 * Wildcards have the query name in answer section.
697	 * ANY queries get query name in answer section.
698	 * Remainders of CNAME chains are cut off and resolved by iterator. */
699	while(rrset && rrset->section == LDNS_SECTION_ANSWER) {
700		if(dname_pkt_compare(pkt, qinfo->qname, rrset->dname) != 0) {
701			if(has_additional(rrset->type)) del_addi = 1;
702			remove_rrset("sanitize: removing extraneous answer "
703				"RRset:", pkt, msg, prev, &rrset);
704			continue;
705		}
706		prev = rrset;
707		rrset = rrset->rrset_all_next;
708	}
709
710	/* At this point, we brutally remove ALL rrsets that aren't
711	 * children of the originating zone. The idea here is that,
712	 * as far as we know, the server that we contacted is ONLY
713	 * authoritative for the originating zone. It, of course, MAY
714	 * be authoritative for any other zones, and of course, MAY
715	 * NOT be authoritative for some subdomains of the originating
716	 * zone. */
717	prev = NULL;
718	rrset = msg->rrset_first;
719	while(rrset) {
720
721		/* remove private addresses */
722		if( (rrset->type == LDNS_RR_TYPE_A ||
723			rrset->type == LDNS_RR_TYPE_AAAA)) {
724
725			/* do not set servfail since this leads to too
726			 * many drops of other people using rfc1918 space */
727			/* also do not remove entire rrset, unless all records
728			 * in it are bad */
729			if(priv_rrset_bad(ie->priv, pkt, rrset)) {
730				remove_rrset(NULL, pkt, msg, prev, &rrset);
731				continue;
732			}
733		}
734
735		/* skip DNAME records -- they will always be followed by a
736		 * synthesized CNAME, which will be relevant.
737		 * FIXME: should this do something differently with DNAME
738		 * rrsets NOT in Section.ANSWER? */
739		/* But since DNAME records are also subdomains of the zone,
740		 * same check can be used */
741
742		if(!pkt_sub(pkt, rrset->dname, zonename)) {
743			if(msg->an_rrsets == 0 &&
744				rrset->type == LDNS_RR_TYPE_NS &&
745				rrset->section == LDNS_SECTION_AUTHORITY &&
746				FLAGS_GET_RCODE(msg->flags) ==
747				LDNS_RCODE_NOERROR && !soa_in_auth(msg) &&
748				sub_of_pkt(pkt, zonename, rrset->dname)) {
749				/* noerror, nodata and this NS rrset is above
750				 * the zone. This is LAME!
751				 * Leave in the NS for lame classification. */
752				/* remove everything from the additional
753				 * (we dont want its glue that was approved
754				 * during the normalize action) */
755				del_addi = 1;
756			} else if(!env->cfg->harden_glue && (
757				rrset->type == LDNS_RR_TYPE_A ||
758				rrset->type == LDNS_RR_TYPE_AAAA)) {
759				/* store in cache! Since it is relevant
760				 * (from normalize) it will be picked up
761				 * from the cache to be used later */
762				store_rrset(pkt, msg, env, rrset);
763				remove_rrset("sanitize: storing potential "
764				"poison RRset:", pkt, msg, prev, &rrset);
765				continue;
766			} else {
767				if(has_additional(rrset->type)) del_addi = 1;
768				remove_rrset("sanitize: removing potential "
769				"poison RRset:", pkt, msg, prev, &rrset);
770				continue;
771			}
772		}
773		if(del_addi && rrset->section == LDNS_SECTION_ADDITIONAL) {
774			remove_rrset("sanitize: removing potential "
775			"poison reference RRset:", pkt, msg, prev, &rrset);
776			continue;
777		}
778		/* check if right hand side of NSEC is within zone */
779		if(rrset->type == LDNS_RR_TYPE_NSEC &&
780			sanitize_nsec_is_overreach(rrset, zonename)) {
781			remove_rrset("sanitize: removing overreaching NSEC "
782				"RRset:", pkt, msg, prev, &rrset);
783			continue;
784		}
785		prev = rrset;
786		rrset = rrset->rrset_all_next;
787	}
788	return 1;
789}
790
791int
792scrub_message(sldns_buffer* pkt, struct msg_parse* msg,
793	struct query_info* qinfo, uint8_t* zonename, struct regional* region,
794	struct module_env* env, struct iter_env* ie)
795{
796	/* basic sanity checks */
797	log_nametypeclass(VERB_ALGO, "scrub for", zonename, LDNS_RR_TYPE_NS,
798		qinfo->qclass);
799	if(msg->qdcount > 1)
800		return 0;
801	if( !(msg->flags&BIT_QR) )
802		return 0;
803	msg->flags &= ~(BIT_AD|BIT_Z); /* force off bit AD and Z */
804
805	/* make sure that a query is echoed back when NOERROR or NXDOMAIN */
806	/* this is not required for basic operation but is a forgery
807	 * resistance (security) feature */
808	if((FLAGS_GET_RCODE(msg->flags) == LDNS_RCODE_NOERROR ||
809		FLAGS_GET_RCODE(msg->flags) == LDNS_RCODE_NXDOMAIN) &&
810		msg->qdcount == 0)
811		return 0;
812
813	/* if a query is echoed back, make sure it is correct. Otherwise,
814	 * this may be not a reply to our query. */
815	if(msg->qdcount == 1) {
816		if(dname_pkt_compare(pkt, msg->qname, qinfo->qname) != 0)
817			return 0;
818		if(msg->qtype != qinfo->qtype || msg->qclass != qinfo->qclass)
819			return 0;
820	}
821
822	/* normalize the response, this cleans up the additional.  */
823	if(!scrub_normalize(pkt, msg, qinfo, region))
824		return 0;
825	/* delete all out-of-zone information */
826	if(!scrub_sanitize(pkt, msg, qinfo, zonename, env, ie))
827		return 0;
828	return 1;
829}
830