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, sldns_buffer* pkt)
189{
190	size_t oldpos, dlen;
191	if(rrset->rr_count != 1) {
192		struct rr_parse* sig;
193		verbose(VERB_ALGO, "Found CNAME rrset with "
194			"size > 1: %u", (unsigned)rrset->rr_count);
195		/* use the first CNAME! */
196		rrset->rr_count = 1;
197		rrset->size = rrset->rr_first->size;
198		for(sig=rrset->rrsig_first; sig; sig=sig->next)
199			rrset->size += sig->size;
200		rrset->rr_last = rrset->rr_first;
201		rrset->rr_first->next = NULL;
202	}
203	if(rrset->rr_first->size < sizeof(uint16_t)+1)
204		return 0; /* CNAME rdata too small */
205	*sname = rrset->rr_first->ttl_data + sizeof(uint32_t)
206		+ sizeof(uint16_t); /* skip ttl, rdatalen */
207	*snamelen = rrset->rr_first->size - sizeof(uint16_t);
208
209	if(rrset->rr_first->outside_packet) {
210		if(!dname_valid(*sname, *snamelen))
211			return 0;
212		return 1;
213	}
214	oldpos = sldns_buffer_position(pkt);
215	sldns_buffer_set_position(pkt, (size_t)(*sname - sldns_buffer_begin(pkt)));
216	dlen = pkt_dname_len(pkt);
217	sldns_buffer_set_position(pkt, oldpos);
218	if(dlen == 0)
219		return 0; /* parse fail on the rdata name */
220	*snamelen = dlen;
221	return 1;
222}
223
224/** Synthesize CNAME from DNAME, false if too long */
225static int
226synth_cname(uint8_t* qname, size_t qnamelen, struct rrset_parse* dname_rrset,
227	uint8_t* alias, size_t* aliaslen, sldns_buffer* pkt)
228{
229	/* we already know that sname is a strict subdomain of DNAME owner */
230	uint8_t* dtarg = NULL;
231	size_t dtarglen;
232	if(!parse_get_cname_target(dname_rrset, &dtarg, &dtarglen, pkt))
233		return 0;
234	if(qnamelen <= dname_rrset->dname_len)
235		return 0;
236	if(qnamelen == 0)
237		return 0;
238	log_assert(qnamelen > dname_rrset->dname_len);
239	/* DNAME from com. to net. with qname example.com. -> example.net. */
240	/* so: \3com\0 to \3net\0 and qname \7example\3com\0 */
241	*aliaslen = qnamelen + dtarglen - dname_rrset->dname_len;
242	if(*aliaslen > LDNS_MAX_DOMAINLEN)
243		return 0; /* should have been RCODE YXDOMAIN */
244	/* decompress dnames into buffer, we know it fits */
245	dname_pkt_copy(pkt, alias, qname);
246	dname_pkt_copy(pkt, alias+(qnamelen-dname_rrset->dname_len), dtarg);
247	return 1;
248}
249
250/** synthesize a CNAME rrset */
251static struct rrset_parse*
252synth_cname_rrset(uint8_t** sname, size_t* snamelen, uint8_t* alias,
253	size_t aliaslen, struct regional* region, struct msg_parse* msg,
254	struct rrset_parse* rrset, struct rrset_parse* prev,
255	struct rrset_parse* nx, sldns_buffer* pkt)
256{
257	struct rrset_parse* cn = (struct rrset_parse*)regional_alloc(region,
258		sizeof(struct rrset_parse));
259	if(!cn)
260		return NULL;
261	memset(cn, 0, sizeof(*cn));
262	cn->rr_first = (struct rr_parse*)regional_alloc(region,
263		sizeof(struct rr_parse));
264	if(!cn->rr_first)
265		return NULL;
266	cn->rr_last = cn->rr_first;
267	/* CNAME from sname to alias */
268	cn->dname = (uint8_t*)regional_alloc(region, *snamelen);
269	if(!cn->dname)
270		return NULL;
271	dname_pkt_copy(pkt, cn->dname, *sname);
272	cn->dname_len = *snamelen;
273	cn->type = LDNS_RR_TYPE_CNAME;
274	cn->section = rrset->section;
275	cn->rrset_class = rrset->rrset_class;
276	cn->rr_count = 1;
277	cn->size = sizeof(uint16_t) + aliaslen;
278	cn->hash=pkt_hash_rrset(pkt, cn->dname, cn->type, cn->rrset_class, 0);
279	/* allocate TTL + rdatalen + uncompressed dname */
280	memset(cn->rr_first, 0, sizeof(struct rr_parse));
281	cn->rr_first->outside_packet = 1;
282	cn->rr_first->ttl_data = (uint8_t*)regional_alloc(region,
283		sizeof(uint32_t)+sizeof(uint16_t)+aliaslen);
284	if(!cn->rr_first->ttl_data)
285		return NULL;
286	sldns_write_uint32(cn->rr_first->ttl_data, 0); /* TTL = 0 */
287	sldns_write_uint16(cn->rr_first->ttl_data+4, aliaslen);
288	memmove(cn->rr_first->ttl_data+6, alias, aliaslen);
289	cn->rr_first->size = sizeof(uint16_t)+aliaslen;
290
291	/* link it in */
292	cn->rrset_all_next = nx;
293	if(prev)
294		prev->rrset_all_next = cn;
295	else	msg->rrset_first = cn;
296	if(nx == NULL)
297		msg->rrset_last = cn;
298	msg->rrset_count ++;
299	msg->an_rrsets++;
300	/* it is not inserted in the msg hashtable. */
301
302	*sname = cn->rr_first->ttl_data + sizeof(uint32_t)+sizeof(uint16_t);
303	*snamelen = aliaslen;
304	return cn;
305}
306
307/** check if DNAME applies to a name */
308static int
309pkt_strict_sub(sldns_buffer* pkt, uint8_t* sname, uint8_t* dr)
310{
311	uint8_t buf1[LDNS_MAX_DOMAINLEN+1];
312	uint8_t buf2[LDNS_MAX_DOMAINLEN+1];
313	/* decompress names */
314	dname_pkt_copy(pkt, buf1, sname);
315	dname_pkt_copy(pkt, buf2, dr);
316	return dname_strict_subdomain_c(buf1, buf2);
317}
318
319/** check subdomain with decompression */
320static int
321pkt_sub(sldns_buffer* pkt, uint8_t* comprname, uint8_t* zone)
322{
323	uint8_t buf[LDNS_MAX_DOMAINLEN+1];
324	dname_pkt_copy(pkt, buf, comprname);
325	return dname_subdomain_c(buf, zone);
326}
327
328/** check subdomain with decompression, compressed is parent */
329static int
330sub_of_pkt(sldns_buffer* pkt, uint8_t* zone, uint8_t* comprname)
331{
332	uint8_t buf[LDNS_MAX_DOMAINLEN+1];
333	dname_pkt_copy(pkt, buf, comprname);
334	return dname_subdomain_c(zone, buf);
335}
336
337/** Check if there are SOA records in the authority section (negative) */
338static int
339soa_in_auth(struct msg_parse* msg)
340{
341	struct rrset_parse* rrset;
342	for(rrset = msg->rrset_first; rrset; rrset = rrset->rrset_all_next)
343		if(rrset->type == LDNS_RR_TYPE_SOA &&
344			rrset->section == LDNS_SECTION_AUTHORITY)
345			return 1;
346	return 0;
347}
348
349/**
350 * This routine normalizes a response. This includes removing "irrelevant"
351 * records from the answer and additional sections and (re)synthesizing
352 * CNAMEs from DNAMEs, if present.
353 *
354 * @param pkt: packet.
355 * @param msg: msg to normalize.
356 * @param qinfo: original query.
357 * @param region: where to allocate synthesized CNAMEs.
358 * @return 0 on error.
359 */
360static int
361scrub_normalize(sldns_buffer* pkt, struct msg_parse* msg,
362	struct query_info* qinfo, struct regional* region)
363{
364	uint8_t* sname = qinfo->qname;
365	size_t snamelen = qinfo->qname_len;
366	struct rrset_parse* rrset, *prev, *nsset=NULL;
367
368	if(FLAGS_GET_RCODE(msg->flags) != LDNS_RCODE_NOERROR &&
369		FLAGS_GET_RCODE(msg->flags) != LDNS_RCODE_NXDOMAIN)
370		return 1;
371
372	/* For the ANSWER section, remove all "irrelevant" records and add
373	 * synthesized CNAMEs from DNAMEs
374	 * This will strip out-of-order CNAMEs as well. */
375
376	/* walk through the parse packet rrset list, keep track of previous
377	 * for insert and delete ease, and examine every RRset */
378	prev = NULL;
379	rrset = msg->rrset_first;
380	while(rrset && rrset->section == LDNS_SECTION_ANSWER) {
381		if(rrset->type == LDNS_RR_TYPE_DNAME &&
382			pkt_strict_sub(pkt, sname, rrset->dname)) {
383			/* check if next rrset is correct CNAME. else,
384			 * synthesize a CNAME */
385			struct rrset_parse* nx = rrset->rrset_all_next;
386			uint8_t alias[LDNS_MAX_DOMAINLEN+1];
387			size_t aliaslen = 0;
388			if(rrset->rr_count != 1) {
389				verbose(VERB_ALGO, "Found DNAME rrset with "
390					"size > 1: %u",
391					(unsigned)rrset->rr_count);
392				return 0;
393			}
394			if(!synth_cname(sname, snamelen, rrset, alias,
395				&aliaslen, pkt)) {
396				verbose(VERB_ALGO, "synthesized CNAME "
397					"too long");
398				return 0;
399			}
400			if(nx && nx->type == LDNS_RR_TYPE_CNAME &&
401			   dname_pkt_compare(pkt, sname, nx->dname) == 0) {
402				/* check next cname */
403				uint8_t* t = NULL;
404				size_t tlen = 0;
405				if(!parse_get_cname_target(nx, &t, &tlen, pkt))
406					return 0;
407				if(dname_pkt_compare(pkt, alias, t) == 0) {
408					/* it's OK and better capitalized */
409					prev = rrset;
410					rrset = nx;
411					continue;
412				}
413				/* synth ourselves */
414			}
415			/* synth a CNAME rrset */
416			prev = synth_cname_rrset(&sname, &snamelen, alias,
417				aliaslen, region, msg, rrset, rrset, nx, pkt);
418			if(!prev) {
419				log_err("out of memory synthesizing CNAME");
420				return 0;
421			}
422			/* FIXME: resolve the conflict between synthesized
423			 * CNAME ttls and the cache. */
424			rrset = nx;
425			continue;
426
427		}
428
429		/* The only records in the ANSWER section not allowed to */
430		if(dname_pkt_compare(pkt, sname, rrset->dname) != 0) {
431			remove_rrset("normalize: removing irrelevant RRset:",
432				pkt, msg, prev, &rrset);
433			continue;
434		}
435
436		/* Follow the CNAME chain. */
437		if(rrset->type == LDNS_RR_TYPE_CNAME) {
438			struct rrset_parse* nx = rrset->rrset_all_next;
439			uint8_t* oldsname = sname;
440			/* see if the next one is a DNAME, if so, swap them */
441			if(nx && nx->section == LDNS_SECTION_ANSWER &&
442				nx->type == LDNS_RR_TYPE_DNAME &&
443				nx->rr_count == 1 &&
444				pkt_strict_sub(pkt, sname, nx->dname)) {
445				/* there is a DNAME after this CNAME, it
446				 * is in the ANSWER section, and the DNAME
447				 * applies to the name we cover */
448				/* check if the alias of the DNAME equals
449				 * this CNAME */
450				uint8_t alias[LDNS_MAX_DOMAINLEN+1];
451				size_t aliaslen = 0;
452				uint8_t* t = NULL;
453				size_t tlen = 0;
454				if(synth_cname(sname, snamelen, nx, alias,
455					&aliaslen, pkt) &&
456					parse_get_cname_target(rrset, &t, &tlen, pkt) &&
457			   		dname_pkt_compare(pkt, alias, t) == 0) {
458					/* the synthesized CNAME equals the
459					 * current CNAME.  This CNAME is the
460					 * one that the DNAME creates, and this
461					 * CNAME is better capitalised */
462					verbose(VERB_ALGO, "normalize: re-order of DNAME and its CNAME");
463					if(prev) prev->rrset_all_next = nx;
464					else msg->rrset_first = nx;
465					if(nx->rrset_all_next == NULL)
466						msg->rrset_last = rrset;
467					rrset->rrset_all_next =
468						nx->rrset_all_next;
469					nx->rrset_all_next = rrset;
470					/* prev = nx; unused, enable if there
471					 * is other rrset removal code after
472					 * this */
473				}
474			}
475
476			/* move to next name in CNAME chain */
477			if(!parse_get_cname_target(rrset, &sname, &snamelen, pkt))
478				return 0;
479			prev = rrset;
480			rrset = rrset->rrset_all_next;
481			/* in CNAME ANY response, can have data after CNAME */
482			if(qinfo->qtype == LDNS_RR_TYPE_ANY) {
483				while(rrset && rrset->section ==
484					LDNS_SECTION_ANSWER &&
485					dname_pkt_compare(pkt, oldsname,
486					rrset->dname) == 0) {
487					prev = rrset;
488					rrset = rrset->rrset_all_next;
489				}
490			}
491			continue;
492		}
493
494		/* Otherwise, make sure that the RRset matches the qtype. */
495		if(qinfo->qtype != LDNS_RR_TYPE_ANY &&
496			qinfo->qtype != rrset->type) {
497			remove_rrset("normalize: removing irrelevant RRset:",
498				pkt, msg, prev, &rrset);
499			continue;
500		}
501
502		/* Mark the additional names from relevant rrset as OK. */
503		/* only for RRsets that match the query name, other ones
504		 * will be removed by sanitize, so no additional for them */
505		if(dname_pkt_compare(pkt, qinfo->qname, rrset->dname) == 0)
506			mark_additional_rrset(pkt, msg, rrset);
507
508		prev = rrset;
509		rrset = rrset->rrset_all_next;
510	}
511
512	/* Mark additional names from AUTHORITY */
513	while(rrset && rrset->section == LDNS_SECTION_AUTHORITY) {
514		if(rrset->type==LDNS_RR_TYPE_DNAME ||
515			rrset->type==LDNS_RR_TYPE_CNAME ||
516			rrset->type==LDNS_RR_TYPE_A ||
517			rrset->type==LDNS_RR_TYPE_AAAA) {
518			remove_rrset("normalize: removing irrelevant "
519				"RRset:", pkt, msg, prev, &rrset);
520			continue;
521		}
522		/* only one NS set allowed in authority section */
523		if(rrset->type==LDNS_RR_TYPE_NS) {
524			/* NS set must be pertinent to the query */
525			if(!sub_of_pkt(pkt, qinfo->qname, rrset->dname)) {
526				remove_rrset("normalize: removing irrelevant "
527					"RRset:", pkt, msg, prev, &rrset);
528				continue;
529			}
530			/* we don't want NS sets for NXDOMAIN answers,
531			 * because they could contain poisonous contents,
532			 * from. eg. fragmentation attacks, inserted after
533			 * long RRSIGs in the packet get to the packet
534			 * border and such */
535			/* also for NODATA answers */
536			if(FLAGS_GET_RCODE(msg->flags) == LDNS_RCODE_NXDOMAIN ||
537			   (FLAGS_GET_RCODE(msg->flags) == LDNS_RCODE_NOERROR
538			    && soa_in_auth(msg) && msg->an_rrsets == 0)) {
539				remove_rrset("normalize: removing irrelevant "
540					"RRset:", pkt, msg, prev, &rrset);
541				continue;
542			}
543			if(nsset == NULL) {
544				nsset = rrset;
545			} else {
546				remove_rrset("normalize: removing irrelevant "
547					"RRset:", pkt, msg, prev, &rrset);
548				continue;
549			}
550		}
551		/* if this is type DS and we query for type DS we just got
552		 * a referral answer for our type DS query, fix packet */
553		if(rrset->type==LDNS_RR_TYPE_DS &&
554			qinfo->qtype == LDNS_RR_TYPE_DS &&
555			dname_pkt_compare(pkt, qinfo->qname, rrset->dname) == 0) {
556			rrset->section = LDNS_SECTION_ANSWER;
557			msg->ancount = rrset->rr_count + rrset->rrsig_count;
558			msg->nscount = 0;
559			msg->arcount = 0;
560			msg->an_rrsets = 1;
561			msg->ns_rrsets = 0;
562			msg->ar_rrsets = 0;
563			msg->rrset_count = 1;
564			msg->rrset_first = rrset;
565			msg->rrset_last = rrset;
566			rrset->rrset_all_next = NULL;
567			return 1;
568		}
569		mark_additional_rrset(pkt, msg, rrset);
570		prev = rrset;
571		rrset = rrset->rrset_all_next;
572	}
573
574	/* For each record in the additional section, remove it if it is an
575	 * address record and not in the collection of additional names
576	 * found in ANSWER and AUTHORITY. */
577	/* These records have not been marked OK previously */
578	while(rrset && rrset->section == LDNS_SECTION_ADDITIONAL) {
579		/* FIXME: what about other types? */
580		if(rrset->type==LDNS_RR_TYPE_A ||
581			rrset->type==LDNS_RR_TYPE_AAAA)
582		{
583			if((rrset->flags & RRSET_SCRUB_OK)) {
584				/* remove flag to clean up flags variable */
585				rrset->flags &= ~RRSET_SCRUB_OK;
586			} else {
587				remove_rrset("normalize: removing irrelevant "
588					"RRset:", pkt, msg, prev, &rrset);
589				continue;
590			}
591		}
592		if(rrset->type==LDNS_RR_TYPE_DNAME ||
593			rrset->type==LDNS_RR_TYPE_CNAME ||
594			rrset->type==LDNS_RR_TYPE_NS) {
595			remove_rrset("normalize: removing irrelevant "
596				"RRset:", pkt, msg, prev, &rrset);
597			continue;
598		}
599		prev = rrset;
600		rrset = rrset->rrset_all_next;
601	}
602
603	return 1;
604}
605
606/**
607 * Store potential poison in the cache (only if hardening disabled).
608 * The rrset is stored in the cache but removed from the message.
609 * So that it will be used for infrastructure purposes, but not be
610 * returned to the client.
611 * @param pkt: packet
612 * @param msg: message parsed
613 * @param env: environment with cache
614 * @param rrset: to store.
615 */
616static void
617store_rrset(sldns_buffer* pkt, struct msg_parse* msg, struct module_env* env,
618	struct rrset_parse* rrset)
619{
620	struct ub_packed_rrset_key* k;
621	struct packed_rrset_data* d;
622	struct rrset_ref ref;
623	time_t now = *env->now;
624
625	k = alloc_special_obtain(env->alloc);
626	if(!k)
627		return;
628	k->entry.data = NULL;
629	if(!parse_copy_decompress_rrset(pkt, msg, rrset, NULL, k)) {
630		alloc_special_release(env->alloc, k);
631		return;
632	}
633	d = (struct packed_rrset_data*)k->entry.data;
634	packed_rrset_ttl_add(d, now);
635	ref.key = k;
636	ref.id = k->id;
637	/*ignore ret: it was in the cache, ref updated */
638	(void)rrset_cache_update(env->rrset_cache, &ref, env->alloc, now);
639}
640
641/**
642 * Check if right hand name in NSEC is within zone
643 * @param rrset: the NSEC rrset
644 * @param zonename: the zone name.
645 * @return true if BAD.
646 */
647static int sanitize_nsec_is_overreach(struct rrset_parse* rrset,
648	uint8_t* zonename)
649{
650	struct rr_parse* rr;
651	uint8_t* rhs;
652	size_t len;
653	log_assert(rrset->type == LDNS_RR_TYPE_NSEC);
654	for(rr = rrset->rr_first; rr; rr = rr->next) {
655		rhs = rr->ttl_data+4+2;
656		len = sldns_read_uint16(rr->ttl_data+4);
657		if(!dname_valid(rhs, len)) {
658			/* malformed domain name in rdata */
659			return 1;
660		}
661		if(!dname_subdomain_c(rhs, zonename)) {
662			/* overreaching */
663			return 1;
664		}
665	}
666	/* all NSEC RRs OK */
667	return 0;
668}
669
670/**
671 * Given a response event, remove suspect RRsets from the response.
672 * "Suspect" rrsets are potentially poison. Note that this routine expects
673 * the response to be in a "normalized" state -- that is, all "irrelevant"
674 * RRsets have already been removed, CNAMEs are in order, etc.
675 *
676 * @param pkt: packet.
677 * @param msg: msg to normalize.
678 * @param qinfo: the question originally asked.
679 * @param zonename: name of server zone.
680 * @param env: module environment with config and cache.
681 * @param ie: iterator environment with private address data.
682 * @return 0 on error.
683 */
684static int
685scrub_sanitize(sldns_buffer* pkt, struct msg_parse* msg,
686	struct query_info* qinfo, uint8_t* zonename, struct module_env* env,
687	struct iter_env* ie)
688{
689	int del_addi = 0; /* if additional-holding rrsets are deleted, we
690		do not trust the normalized additional-A-AAAA any more */
691	struct rrset_parse* rrset, *prev;
692	prev = NULL;
693	rrset = msg->rrset_first;
694
695	/* the first DNAME is allowed to stay. It needs checking before
696	 * it can be used from the cache. After normalization, an initial
697	 * DNAME will have a correctly synthesized CNAME after it. */
698	if(rrset && rrset->type == LDNS_RR_TYPE_DNAME &&
699		rrset->section == LDNS_SECTION_ANSWER &&
700		pkt_strict_sub(pkt, qinfo->qname, rrset->dname) &&
701		pkt_sub(pkt, rrset->dname, zonename)) {
702		prev = rrset; /* DNAME allowed to stay in answer section */
703		rrset = rrset->rrset_all_next;
704	}
705
706	/* remove all records from the answer section that are
707	 * not the same domain name as the query domain name.
708	 * The answer section should contain rrsets with the same name
709	 * as the question. For DNAMEs a CNAME has been synthesized.
710	 * Wildcards have the query name in answer section.
711	 * ANY queries get query name in answer section.
712	 * Remainders of CNAME chains are cut off and resolved by iterator. */
713	while(rrset && rrset->section == LDNS_SECTION_ANSWER) {
714		if(dname_pkt_compare(pkt, qinfo->qname, rrset->dname) != 0) {
715			if(has_additional(rrset->type)) del_addi = 1;
716			remove_rrset("sanitize: removing extraneous answer "
717				"RRset:", pkt, msg, prev, &rrset);
718			continue;
719		}
720		prev = rrset;
721		rrset = rrset->rrset_all_next;
722	}
723
724	/* At this point, we brutally remove ALL rrsets that aren't
725	 * children of the originating zone. The idea here is that,
726	 * as far as we know, the server that we contacted is ONLY
727	 * authoritative for the originating zone. It, of course, MAY
728	 * be authoritative for any other zones, and of course, MAY
729	 * NOT be authoritative for some subdomains of the originating
730	 * zone. */
731	prev = NULL;
732	rrset = msg->rrset_first;
733	while(rrset) {
734
735		/* remove private addresses */
736		if( (rrset->type == LDNS_RR_TYPE_A ||
737			rrset->type == LDNS_RR_TYPE_AAAA)) {
738
739			/* do not set servfail since this leads to too
740			 * many drops of other people using rfc1918 space */
741			/* also do not remove entire rrset, unless all records
742			 * in it are bad */
743			if(priv_rrset_bad(ie->priv, pkt, rrset)) {
744				remove_rrset(NULL, pkt, msg, prev, &rrset);
745				continue;
746			}
747		}
748
749		/* skip DNAME records -- they will always be followed by a
750		 * synthesized CNAME, which will be relevant.
751		 * FIXME: should this do something differently with DNAME
752		 * rrsets NOT in Section.ANSWER? */
753		/* But since DNAME records are also subdomains of the zone,
754		 * same check can be used */
755
756		if(!pkt_sub(pkt, rrset->dname, zonename)) {
757			if(msg->an_rrsets == 0 &&
758				rrset->type == LDNS_RR_TYPE_NS &&
759				rrset->section == LDNS_SECTION_AUTHORITY &&
760				FLAGS_GET_RCODE(msg->flags) ==
761				LDNS_RCODE_NOERROR && !soa_in_auth(msg) &&
762				sub_of_pkt(pkt, zonename, rrset->dname)) {
763				/* noerror, nodata and this NS rrset is above
764				 * the zone. This is LAME!
765				 * Leave in the NS for lame classification. */
766				/* remove everything from the additional
767				 * (we dont want its glue that was approved
768				 * during the normalize action) */
769				del_addi = 1;
770			} else if(!env->cfg->harden_glue && (
771				rrset->type == LDNS_RR_TYPE_A ||
772				rrset->type == LDNS_RR_TYPE_AAAA)) {
773				/* store in cache! Since it is relevant
774				 * (from normalize) it will be picked up
775				 * from the cache to be used later */
776				store_rrset(pkt, msg, env, rrset);
777				remove_rrset("sanitize: storing potential "
778				"poison RRset:", pkt, msg, prev, &rrset);
779				continue;
780			} else {
781				if(has_additional(rrset->type)) del_addi = 1;
782				remove_rrset("sanitize: removing potential "
783				"poison RRset:", pkt, msg, prev, &rrset);
784				continue;
785			}
786		}
787		if(del_addi && rrset->section == LDNS_SECTION_ADDITIONAL) {
788			remove_rrset("sanitize: removing potential "
789			"poison reference RRset:", pkt, msg, prev, &rrset);
790			continue;
791		}
792		/* check if right hand side of NSEC is within zone */
793		if(rrset->type == LDNS_RR_TYPE_NSEC &&
794			sanitize_nsec_is_overreach(rrset, zonename)) {
795			remove_rrset("sanitize: removing overreaching NSEC "
796				"RRset:", pkt, msg, prev, &rrset);
797			continue;
798		}
799		prev = rrset;
800		rrset = rrset->rrset_all_next;
801	}
802	return 1;
803}
804
805int
806scrub_message(sldns_buffer* pkt, struct msg_parse* msg,
807	struct query_info* qinfo, uint8_t* zonename, struct regional* region,
808	struct module_env* env, struct iter_env* ie)
809{
810	/* basic sanity checks */
811	log_nametypeclass(VERB_ALGO, "scrub for", zonename, LDNS_RR_TYPE_NS,
812		qinfo->qclass);
813	if(msg->qdcount > 1)
814		return 0;
815	if( !(msg->flags&BIT_QR) )
816		return 0;
817	msg->flags &= ~(BIT_AD|BIT_Z); /* force off bit AD and Z */
818
819	/* make sure that a query is echoed back when NOERROR or NXDOMAIN */
820	/* this is not required for basic operation but is a forgery
821	 * resistance (security) feature */
822	if((FLAGS_GET_RCODE(msg->flags) == LDNS_RCODE_NOERROR ||
823		FLAGS_GET_RCODE(msg->flags) == LDNS_RCODE_NXDOMAIN) &&
824		msg->qdcount == 0)
825		return 0;
826
827	/* if a query is echoed back, make sure it is correct. Otherwise,
828	 * this may be not a reply to our query. */
829	if(msg->qdcount == 1) {
830		if(dname_pkt_compare(pkt, msg->qname, qinfo->qname) != 0)
831			return 0;
832		if(msg->qtype != qinfo->qtype || msg->qclass != qinfo->qclass)
833			return 0;
834	}
835
836	/* normalize the response, this cleans up the additional.  */
837	if(!scrub_normalize(pkt, msg, qinfo, region))
838		return 0;
839	/* delete all out-of-zone information */
840	if(!scrub_sanitize(pkt, msg, qinfo, zonename, env, ie))
841		return 0;
842	return 1;
843}
844