val_neg.c revision 356345
1/*
2 * validator/val_neg.c - validator aggressive negative caching functions.
3 *
4 * Copyright (c) 2008, 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 contains helper functions for the validator module.
40 * The functions help with aggressive negative caching.
41 * This creates new denials of existence, and proofs for absence of types
42 * from cached NSEC records.
43 */
44#include "config.h"
45#ifdef HAVE_OPENSSL_SSL_H
46#include "openssl/ssl.h"
47#define NSEC3_SHA_LEN SHA_DIGEST_LENGTH
48#else
49#define NSEC3_SHA_LEN 20
50#endif
51#include "validator/val_neg.h"
52#include "validator/val_nsec.h"
53#include "validator/val_nsec3.h"
54#include "validator/val_utils.h"
55#include "util/data/dname.h"
56#include "util/data/msgreply.h"
57#include "util/log.h"
58#include "util/net_help.h"
59#include "util/config_file.h"
60#include "services/cache/rrset.h"
61#include "services/cache/dns.h"
62#include "sldns/rrdef.h"
63#include "sldns/sbuffer.h"
64
65int val_neg_data_compare(const void* a, const void* b)
66{
67	struct val_neg_data* x = (struct val_neg_data*)a;
68	struct val_neg_data* y = (struct val_neg_data*)b;
69	int m;
70	return dname_canon_lab_cmp(x->name, x->labs, y->name, y->labs, &m);
71}
72
73int val_neg_zone_compare(const void* a, const void* b)
74{
75	struct val_neg_zone* x = (struct val_neg_zone*)a;
76	struct val_neg_zone* y = (struct val_neg_zone*)b;
77	int m;
78	if(x->dclass != y->dclass) {
79		if(x->dclass < y->dclass)
80			return -1;
81		return 1;
82	}
83	return dname_canon_lab_cmp(x->name, x->labs, y->name, y->labs, &m);
84}
85
86struct val_neg_cache* val_neg_create(struct config_file* cfg, size_t maxiter)
87{
88	struct val_neg_cache* neg = (struct val_neg_cache*)calloc(1,
89		sizeof(*neg));
90	if(!neg) {
91		log_err("Could not create neg cache: out of memory");
92		return NULL;
93	}
94	neg->nsec3_max_iter = maxiter;
95	neg->max = 1024*1024; /* 1 M is thousands of entries */
96	if(cfg) neg->max = cfg->neg_cache_size;
97	rbtree_init(&neg->tree, &val_neg_zone_compare);
98	lock_basic_init(&neg->lock);
99	lock_protect(&neg->lock, neg, sizeof(*neg));
100	return neg;
101}
102
103size_t val_neg_get_mem(struct val_neg_cache* neg)
104{
105	size_t result;
106	lock_basic_lock(&neg->lock);
107	result = sizeof(*neg) + neg->use;
108	lock_basic_unlock(&neg->lock);
109	return result;
110}
111
112/** clear datas on cache deletion */
113static void
114neg_clear_datas(rbnode_type* n, void* ATTR_UNUSED(arg))
115{
116	struct val_neg_data* d = (struct val_neg_data*)n;
117	free(d->name);
118	free(d);
119}
120
121/** clear zones on cache deletion */
122static void
123neg_clear_zones(rbnode_type* n, void* ATTR_UNUSED(arg))
124{
125	struct val_neg_zone* z = (struct val_neg_zone*)n;
126	/* delete all the rrset entries in the tree */
127	traverse_postorder(&z->tree, &neg_clear_datas, NULL);
128	free(z->nsec3_salt);
129	free(z->name);
130	free(z);
131}
132
133void neg_cache_delete(struct val_neg_cache* neg)
134{
135	if(!neg) return;
136	lock_basic_destroy(&neg->lock);
137	/* delete all the zones in the tree */
138	traverse_postorder(&neg->tree, &neg_clear_zones, NULL);
139	free(neg);
140}
141
142/**
143 * Put data element at the front of the LRU list.
144 * @param neg: negative cache with LRU start and end.
145 * @param data: this data is fronted.
146 */
147static void neg_lru_front(struct val_neg_cache* neg,
148	struct val_neg_data* data)
149{
150	data->prev = NULL;
151	data->next = neg->first;
152	if(!neg->first)
153		neg->last = data;
154	else	neg->first->prev = data;
155	neg->first = data;
156}
157
158/**
159 * Remove data element from LRU list.
160 * @param neg: negative cache with LRU start and end.
161 * @param data: this data is removed from the list.
162 */
163static void neg_lru_remove(struct val_neg_cache* neg,
164	struct val_neg_data* data)
165{
166	if(data->prev)
167		data->prev->next = data->next;
168	else	neg->first = data->next;
169	if(data->next)
170		data->next->prev = data->prev;
171	else	neg->last = data->prev;
172}
173
174/**
175 * Touch LRU for data element, put it at the start of the LRU list.
176 * @param neg: negative cache with LRU start and end.
177 * @param data: this data is used.
178 */
179static void neg_lru_touch(struct val_neg_cache* neg,
180	struct val_neg_data* data)
181{
182	if(data == neg->first)
183		return; /* nothing to do */
184	/* remove from current lru position */
185	neg_lru_remove(neg, data);
186	/* add at front */
187	neg_lru_front(neg, data);
188}
189
190/**
191 * Delete a zone element from the negative cache.
192 * May delete other zone elements to keep tree coherent, or
193 * only mark the element as 'not in use'.
194 * @param neg: negative cache.
195 * @param z: zone element to delete.
196 */
197static void neg_delete_zone(struct val_neg_cache* neg, struct val_neg_zone* z)
198{
199	struct val_neg_zone* p, *np;
200	if(!z) return;
201	log_assert(z->in_use);
202	log_assert(z->count > 0);
203	z->in_use = 0;
204
205	/* go up the tree and reduce counts */
206	p = z;
207	while(p) {
208		log_assert(p->count > 0);
209		p->count --;
210		p = p->parent;
211	}
212
213	/* remove zones with zero count */
214	p = z;
215	while(p && p->count == 0) {
216		np = p->parent;
217		(void)rbtree_delete(&neg->tree, &p->node);
218		neg->use -= p->len + sizeof(*p);
219		free(p->nsec3_salt);
220		free(p->name);
221		free(p);
222		p = np;
223	}
224}
225
226void neg_delete_data(struct val_neg_cache* neg, struct val_neg_data* el)
227{
228	struct val_neg_zone* z;
229	struct val_neg_data* p, *np;
230	if(!el) return;
231	z = el->zone;
232	log_assert(el->in_use);
233	log_assert(el->count > 0);
234	el->in_use = 0;
235
236	/* remove it from the lru list */
237	neg_lru_remove(neg, el);
238	log_assert(neg->first != el && neg->last != el);
239
240	/* go up the tree and reduce counts */
241	p = el;
242	while(p) {
243		log_assert(p->count > 0);
244		p->count --;
245		p = p->parent;
246	}
247
248	/* delete 0 count items from tree */
249	p = el;
250	while(p && p->count == 0) {
251		np = p->parent;
252		(void)rbtree_delete(&z->tree, &p->node);
253		neg->use -= p->len + sizeof(*p);
254		free(p->name);
255		free(p);
256		p = np;
257	}
258
259	/* check if the zone is now unused */
260	if(z->tree.count == 0) {
261		neg_delete_zone(neg, z);
262	}
263}
264
265/**
266 * Create more space in negative cache
267 * The oldest elements are deleted until enough space is present.
268 * Empty zones are deleted.
269 * @param neg: negative cache.
270 * @param need: how many bytes are needed.
271 */
272static void neg_make_space(struct val_neg_cache* neg, size_t need)
273{
274	/* delete elements until enough space or its empty */
275	while(neg->last && neg->max < neg->use + need) {
276		neg_delete_data(neg, neg->last);
277	}
278}
279
280struct val_neg_zone* neg_find_zone(struct val_neg_cache* neg,
281	uint8_t* nm, size_t len, uint16_t dclass)
282{
283	struct val_neg_zone lookfor;
284	struct val_neg_zone* result;
285	lookfor.node.key = &lookfor;
286	lookfor.name = nm;
287	lookfor.len = len;
288	lookfor.labs = dname_count_labels(lookfor.name);
289	lookfor.dclass = dclass;
290
291	result = (struct val_neg_zone*)
292		rbtree_search(&neg->tree, lookfor.node.key);
293	return result;
294}
295
296/**
297 * Find the given data
298 * @param zone: negative zone
299 * @param nm: what to look for.
300 * @param len: length of nm
301 * @param labs: labels in nm
302 * @return data or NULL if not found.
303 */
304static struct val_neg_data* neg_find_data(struct val_neg_zone* zone,
305	uint8_t* nm, size_t len, int labs)
306{
307	struct val_neg_data lookfor;
308	struct val_neg_data* result;
309	lookfor.node.key = &lookfor;
310	lookfor.name = nm;
311	lookfor.len = len;
312	lookfor.labs = labs;
313
314	result = (struct val_neg_data*)
315		rbtree_search(&zone->tree, lookfor.node.key);
316	return result;
317}
318
319/**
320 * Calculate space needed for the data and all its parents
321 * @param rep: NSEC entries.
322 * @return size.
323 */
324static size_t calc_data_need(struct reply_info* rep)
325{
326	uint8_t* d;
327	size_t i, len, res = 0;
328
329	for(i=rep->an_numrrsets; i<rep->an_numrrsets+rep->ns_numrrsets; i++) {
330		if(ntohs(rep->rrsets[i]->rk.type) == LDNS_RR_TYPE_NSEC) {
331			d = rep->rrsets[i]->rk.dname;
332			len = rep->rrsets[i]->rk.dname_len;
333			res = sizeof(struct val_neg_data) + len;
334			while(!dname_is_root(d)) {
335				log_assert(len > 1); /* not root label */
336				dname_remove_label(&d, &len);
337				res += sizeof(struct val_neg_data) + len;
338			}
339		}
340	}
341	return res;
342}
343
344/**
345 * Calculate space needed for zone and all its parents
346 * @param d: name of zone
347 * @param len: length of name
348 * @return size.
349 */
350static size_t calc_zone_need(uint8_t* d, size_t len)
351{
352	size_t res = sizeof(struct val_neg_zone) + len;
353	while(!dname_is_root(d)) {
354		log_assert(len > 1); /* not root label */
355		dname_remove_label(&d, &len);
356		res += sizeof(struct val_neg_zone) + len;
357	}
358	return res;
359}
360
361/**
362 * Find closest existing parent zone of the given name.
363 * @param neg: negative cache.
364 * @param nm: name to look for
365 * @param nm_len: length of nm
366 * @param labs: labelcount of nm.
367 * @param qclass: class.
368 * @return the zone or NULL if none found.
369 */
370static struct val_neg_zone* neg_closest_zone_parent(struct val_neg_cache* neg,
371	uint8_t* nm, size_t nm_len, int labs, uint16_t qclass)
372{
373	struct val_neg_zone key;
374	struct val_neg_zone* result;
375	rbnode_type* res = NULL;
376	key.node.key = &key;
377	key.name = nm;
378	key.len = nm_len;
379	key.labs = labs;
380	key.dclass = qclass;
381	if(rbtree_find_less_equal(&neg->tree, &key, &res)) {
382		/* exact match */
383		result = (struct val_neg_zone*)res;
384	} else {
385		/* smaller element (or no element) */
386		int m;
387		result = (struct val_neg_zone*)res;
388		if(!result || result->dclass != qclass)
389			return NULL;
390		/* count number of labels matched */
391		(void)dname_lab_cmp(result->name, result->labs, key.name,
392			key.labs, &m);
393		while(result) { /* go up until qname is subdomain of stub */
394			if(result->labs <= m)
395				break;
396			result = result->parent;
397		}
398	}
399	return result;
400}
401
402/**
403 * Find closest existing parent data for the given name.
404 * @param zone: to look in.
405 * @param nm: name to look for
406 * @param nm_len: length of nm
407 * @param labs: labelcount of nm.
408 * @return the data or NULL if none found.
409 */
410static struct val_neg_data* neg_closest_data_parent(
411	struct val_neg_zone* zone, uint8_t* nm, size_t nm_len, int labs)
412{
413	struct val_neg_data key;
414	struct val_neg_data* result;
415	rbnode_type* res = NULL;
416	key.node.key = &key;
417	key.name = nm;
418	key.len = nm_len;
419	key.labs = labs;
420	if(rbtree_find_less_equal(&zone->tree, &key, &res)) {
421		/* exact match */
422		result = (struct val_neg_data*)res;
423	} else {
424		/* smaller element (or no element) */
425		int m;
426		result = (struct val_neg_data*)res;
427		if(!result)
428			return NULL;
429		/* count number of labels matched */
430		(void)dname_lab_cmp(result->name, result->labs, key.name,
431			key.labs, &m);
432		while(result) { /* go up until qname is subdomain of stub */
433			if(result->labs <= m)
434				break;
435			result = result->parent;
436		}
437	}
438	return result;
439}
440
441/**
442 * Create a single zone node
443 * @param nm: name for zone (copied)
444 * @param nm_len: length of name
445 * @param labs: labels in name.
446 * @param dclass: class of zone, host order.
447 * @return new zone or NULL on failure
448 */
449static struct val_neg_zone* neg_setup_zone_node(
450	uint8_t* nm, size_t nm_len, int labs, uint16_t dclass)
451{
452	struct val_neg_zone* zone =
453		(struct val_neg_zone*)calloc(1, sizeof(*zone));
454	if(!zone) {
455		return NULL;
456	}
457	zone->node.key = zone;
458	zone->name = memdup(nm, nm_len);
459	if(!zone->name) {
460		free(zone);
461		return NULL;
462	}
463	zone->len = nm_len;
464	zone->labs = labs;
465	zone->dclass = dclass;
466
467	rbtree_init(&zone->tree, &val_neg_data_compare);
468	return zone;
469}
470
471/**
472 * Create a linked list of parent zones, starting at longname ending on
473 * the parent (can be NULL, creates to the root).
474 * @param nm: name for lowest in chain
475 * @param nm_len: length of name
476 * @param labs: labels in name.
477 * @param dclass: class of zone.
478 * @param parent: NULL for to root, else so it fits under here.
479 * @return zone; a chain of zones and their parents up to the parent.
480 *  	or NULL on malloc failure
481 */
482static struct val_neg_zone* neg_zone_chain(
483	uint8_t* nm, size_t nm_len, int labs, uint16_t dclass,
484	struct val_neg_zone* parent)
485{
486	int i;
487	int tolabs = parent?parent->labs:0;
488	struct val_neg_zone* zone, *prev = NULL, *first = NULL;
489
490	/* create the new subtree, i is labelcount of current creation */
491	/* this creates a 'first' to z->parent=NULL list of zones */
492	for(i=labs; i!=tolabs; i--) {
493		/* create new item */
494		zone = neg_setup_zone_node(nm, nm_len, i, dclass);
495		if(!zone) {
496			/* need to delete other allocations in this routine!*/
497			struct val_neg_zone* p=first, *np;
498			while(p) {
499				np = p->parent;
500				free(p->name);
501				free(p);
502				p = np;
503			}
504			return NULL;
505		}
506		if(i == labs) {
507			first = zone;
508		} else {
509			prev->parent = zone;
510		}
511		/* prepare for next name */
512		prev = zone;
513		dname_remove_label(&nm, &nm_len);
514	}
515	return first;
516}
517
518void val_neg_zone_take_inuse(struct val_neg_zone* zone)
519{
520	if(!zone->in_use) {
521		struct val_neg_zone* p;
522		zone->in_use = 1;
523		/* increase usage count of all parents */
524		for(p=zone; p; p = p->parent) {
525			p->count++;
526		}
527	}
528}
529
530struct val_neg_zone* neg_create_zone(struct val_neg_cache* neg,
531	uint8_t* nm, size_t nm_len, uint16_t dclass)
532{
533	struct val_neg_zone* zone;
534	struct val_neg_zone* parent;
535	struct val_neg_zone* p, *np;
536	int labs = dname_count_labels(nm);
537
538	/* find closest enclosing parent zone that (still) exists */
539	parent = neg_closest_zone_parent(neg, nm, nm_len, labs, dclass);
540	if(parent && query_dname_compare(parent->name, nm) == 0)
541		return parent; /* already exists, weird */
542	/* if parent exists, it is in use */
543	log_assert(!parent || parent->count > 0);
544	zone = neg_zone_chain(nm, nm_len, labs, dclass, parent);
545	if(!zone) {
546		return NULL;
547	}
548
549	/* insert the list of zones into the tree */
550	p = zone;
551	while(p) {
552		np = p->parent;
553		/* mem use */
554		neg->use += sizeof(struct val_neg_zone) + p->len;
555		/* insert in tree */
556		(void)rbtree_insert(&neg->tree, &p->node);
557		/* last one needs proper parent pointer */
558		if(np == NULL)
559			p->parent = parent;
560		p = np;
561	}
562	return zone;
563}
564
565/** find zone name of message, returns the SOA record */
566static struct ub_packed_rrset_key* reply_find_soa(struct reply_info* rep)
567{
568	size_t i;
569	for(i=rep->an_numrrsets; i< rep->an_numrrsets+rep->ns_numrrsets; i++){
570		if(ntohs(rep->rrsets[i]->rk.type) == LDNS_RR_TYPE_SOA)
571			return rep->rrsets[i];
572	}
573	return NULL;
574}
575
576/** see if the reply has NSEC records worthy of caching */
577static int reply_has_nsec(struct reply_info* rep)
578{
579	size_t i;
580	struct packed_rrset_data* d;
581	if(rep->security != sec_status_secure)
582		return 0;
583	for(i=rep->an_numrrsets; i< rep->an_numrrsets+rep->ns_numrrsets; i++){
584		if(ntohs(rep->rrsets[i]->rk.type) == LDNS_RR_TYPE_NSEC) {
585			d = (struct packed_rrset_data*)rep->rrsets[i]->
586				entry.data;
587			if(d->security == sec_status_secure)
588				return 1;
589		}
590	}
591	return 0;
592}
593
594
595/**
596 * Create single node of data element.
597 * @param nm: name (copied)
598 * @param nm_len: length of name
599 * @param labs: labels in name.
600 * @return element with name nm, or NULL malloc failure.
601 */
602static struct val_neg_data* neg_setup_data_node(
603	uint8_t* nm, size_t nm_len, int labs)
604{
605	struct val_neg_data* el;
606	el = (struct val_neg_data*)calloc(1, sizeof(*el));
607	if(!el) {
608		return NULL;
609	}
610	el->node.key = el;
611	el->name = memdup(nm, nm_len);
612	if(!el->name) {
613		free(el);
614		return NULL;
615	}
616	el->len = nm_len;
617	el->labs = labs;
618	return el;
619}
620
621/**
622 * Create chain of data element and parents
623 * @param nm: name
624 * @param nm_len: length of name
625 * @param labs: labels in name.
626 * @param parent: up to where to make, if NULL up to root label.
627 * @return lowest element with name nm, or NULL malloc failure.
628 */
629static struct val_neg_data* neg_data_chain(
630	uint8_t* nm, size_t nm_len, int labs, struct val_neg_data* parent)
631{
632	int i;
633	int tolabs = parent?parent->labs:0;
634	struct val_neg_data* el, *first = NULL, *prev = NULL;
635
636	/* create the new subtree, i is labelcount of current creation */
637	/* this creates a 'first' to z->parent=NULL list of zones */
638	for(i=labs; i!=tolabs; i--) {
639		/* create new item */
640		el = neg_setup_data_node(nm, nm_len, i);
641		if(!el) {
642			/* need to delete other allocations in this routine!*/
643			struct val_neg_data* p = first, *np;
644			while(p) {
645				np = p->parent;
646				free(p->name);
647				free(p);
648				p = np;
649			}
650			return NULL;
651		}
652		if(i == labs) {
653			first = el;
654		} else {
655			prev->parent = el;
656		}
657
658		/* prepare for next name */
659		prev = el;
660		dname_remove_label(&nm, &nm_len);
661	}
662	return first;
663}
664
665/**
666 * Remove NSEC records between start and end points.
667 * By walking the tree, the tree is sorted canonically.
668 * @param neg: negative cache.
669 * @param zone: the zone
670 * @param el: element to start walking at.
671 * @param nsec: the nsec record with the end point
672 */
673static void wipeout(struct val_neg_cache* neg, struct val_neg_zone* zone,
674	struct val_neg_data* el, struct ub_packed_rrset_key* nsec)
675{
676	struct packed_rrset_data* d = (struct packed_rrset_data*)nsec->
677		entry.data;
678	uint8_t* end;
679	size_t end_len;
680	int end_labs, m;
681	rbnode_type* walk, *next;
682	struct val_neg_data* cur;
683	uint8_t buf[257];
684	/* get endpoint */
685	if(!d || d->count == 0 || d->rr_len[0] < 2+1)
686		return;
687	if(ntohs(nsec->rk.type) == LDNS_RR_TYPE_NSEC) {
688		end = d->rr_data[0]+2;
689		end_len = dname_valid(end, d->rr_len[0]-2);
690		end_labs = dname_count_labels(end);
691	} else {
692		/* NSEC3 */
693		if(!nsec3_get_nextowner_b32(nsec, 0, buf, sizeof(buf)))
694			return;
695		end = buf;
696		end_labs = dname_count_size_labels(end, &end_len);
697	}
698
699	/* sanity check, both owner and end must be below the zone apex */
700	if(!dname_subdomain_c(el->name, zone->name) ||
701		!dname_subdomain_c(end, zone->name))
702		return;
703
704	/* detect end of zone NSEC ; wipe until the end of zone */
705	if(query_dname_compare(end, zone->name) == 0) {
706		end = NULL;
707	}
708
709	walk = rbtree_next(&el->node);
710	while(walk && walk != RBTREE_NULL) {
711		cur = (struct val_neg_data*)walk;
712		/* sanity check: must be larger than start */
713		if(dname_canon_lab_cmp(cur->name, cur->labs,
714			el->name, el->labs, &m) <= 0) {
715			/* r == 0 skip original record. */
716			/* r < 0  too small! */
717			walk = rbtree_next(walk);
718			continue;
719		}
720		/* stop at endpoint, also data at empty nonterminals must be
721		 * removed (no NSECs there) so everything between
722		 * start and end */
723		if(end && dname_canon_lab_cmp(cur->name, cur->labs,
724			end, end_labs, &m) >= 0) {
725			break;
726		}
727		/* this element has to be deleted, but we cannot do it
728		 * now, because we are walking the tree still ... */
729		/* get the next element: */
730		next = rbtree_next(walk);
731		/* now delete the original element, this may trigger
732		 * rbtree rebalances, but really, the next element is
733		 * the one we need.
734		 * But it may trigger delete of other data and the
735		 * entire zone. However, if that happens, this is done
736		 * by deleting the *parents* of the element for deletion,
737		 * and maybe also the entire zone if it is empty.
738		 * But parents are smaller in canonical compare, thus,
739		 * if a larger element exists, then it is not a parent,
740		 * it cannot get deleted, the zone cannot get empty.
741		 * If the next==NULL, then zone can be empty. */
742		if(cur->in_use)
743			neg_delete_data(neg, cur);
744		walk = next;
745	}
746}
747
748void neg_insert_data(struct val_neg_cache* neg,
749	struct val_neg_zone* zone, struct ub_packed_rrset_key* nsec)
750{
751	struct packed_rrset_data* d;
752	struct val_neg_data* parent;
753	struct val_neg_data* el;
754	uint8_t* nm = nsec->rk.dname;
755	size_t nm_len = nsec->rk.dname_len;
756	int labs = dname_count_labels(nsec->rk.dname);
757
758	d = (struct packed_rrset_data*)nsec->entry.data;
759	if( !(d->security == sec_status_secure ||
760		(d->security == sec_status_unchecked && d->rrsig_count > 0)))
761		return;
762	log_nametypeclass(VERB_ALGO, "negcache rr",
763		nsec->rk.dname, ntohs(nsec->rk.type),
764		ntohs(nsec->rk.rrset_class));
765
766	/* find closest enclosing parent data that (still) exists */
767	parent = neg_closest_data_parent(zone, nm, nm_len, labs);
768	if(parent && query_dname_compare(parent->name, nm) == 0) {
769		/* perfect match already exists */
770		log_assert(parent->count > 0);
771		el = parent;
772	} else {
773		struct val_neg_data* p, *np;
774
775		/* create subtree for perfect match */
776		/* if parent exists, it is in use */
777		log_assert(!parent || parent->count > 0);
778
779		el = neg_data_chain(nm, nm_len, labs, parent);
780		if(!el) {
781			log_err("out of memory inserting NSEC negative cache");
782			return;
783		}
784		el->in_use = 0; /* set on below */
785
786		/* insert the list of zones into the tree */
787		p = el;
788		while(p) {
789			np = p->parent;
790			/* mem use */
791			neg->use += sizeof(struct val_neg_data) + p->len;
792			/* insert in tree */
793			p->zone = zone;
794			(void)rbtree_insert(&zone->tree, &p->node);
795			/* last one needs proper parent pointer */
796			if(np == NULL)
797				p->parent = parent;
798			p = np;
799		}
800	}
801
802	if(!el->in_use) {
803		struct val_neg_data* p;
804
805		el->in_use = 1;
806		/* increase usage count of all parents */
807		for(p=el; p; p = p->parent) {
808			p->count++;
809		}
810
811		neg_lru_front(neg, el);
812	} else {
813		/* in use, bring to front, lru */
814		neg_lru_touch(neg, el);
815	}
816
817	/* if nsec3 store last used parameters */
818	if(ntohs(nsec->rk.type) == LDNS_RR_TYPE_NSEC3) {
819		int h;
820		uint8_t* s;
821		size_t slen, it;
822		if(nsec3_get_params(nsec, 0, &h, &it, &s, &slen) &&
823			it <= neg->nsec3_max_iter &&
824			(h != zone->nsec3_hash || it != zone->nsec3_iter ||
825			slen != zone->nsec3_saltlen ||
826			memcmp(zone->nsec3_salt, s, slen) != 0)) {
827
828			if(slen > 0) {
829				uint8_t* sa = memdup(s, slen);
830				if(sa) {
831					free(zone->nsec3_salt);
832					zone->nsec3_salt = sa;
833					zone->nsec3_saltlen = slen;
834					zone->nsec3_iter = it;
835					zone->nsec3_hash = h;
836				}
837			} else {
838				free(zone->nsec3_salt);
839				zone->nsec3_salt = NULL;
840				zone->nsec3_saltlen = 0;
841				zone->nsec3_iter = it;
842				zone->nsec3_hash = h;
843			}
844		}
845	}
846
847	/* wipe out the cache items between NSEC start and end */
848	wipeout(neg, zone, el, nsec);
849}
850
851/** see if the reply has signed NSEC records and return the signer */
852static uint8_t* reply_nsec_signer(struct reply_info* rep, size_t* signer_len,
853	uint16_t* dclass)
854{
855	size_t i;
856	struct packed_rrset_data* d;
857	uint8_t* s;
858	for(i=rep->an_numrrsets; i< rep->an_numrrsets+rep->ns_numrrsets; i++){
859		if(ntohs(rep->rrsets[i]->rk.type) == LDNS_RR_TYPE_NSEC ||
860			ntohs(rep->rrsets[i]->rk.type) == LDNS_RR_TYPE_NSEC3) {
861			d = (struct packed_rrset_data*)rep->rrsets[i]->
862				entry.data;
863			/* return first signer name of first NSEC */
864			if(d->rrsig_count != 0) {
865				val_find_rrset_signer(rep->rrsets[i],
866					&s, signer_len);
867				if(s && *signer_len) {
868					*dclass = ntohs(rep->rrsets[i]->
869						rk.rrset_class);
870					return s;
871				}
872			}
873		}
874	}
875	return 0;
876}
877
878void val_neg_addreply(struct val_neg_cache* neg, struct reply_info* rep)
879{
880	size_t i, need;
881	struct ub_packed_rrset_key* soa;
882	uint8_t* dname = NULL;
883	size_t dname_len;
884	uint16_t rrset_class;
885	struct val_neg_zone* zone;
886	/* see if secure nsecs inside */
887	if(!reply_has_nsec(rep))
888		return;
889	/* find the zone name in message */
890	if((soa = reply_find_soa(rep))) {
891		dname = soa->rk.dname;
892		dname_len = soa->rk.dname_len;
893		rrset_class = ntohs(soa->rk.rrset_class);
894	}
895	else {
896		/* No SOA in positive (wildcard) answer. Use signer from the
897		 * validated answer RRsets' signature. */
898		if(!(dname = reply_nsec_signer(rep, &dname_len, &rrset_class)))
899			return;
900	}
901
902	log_nametypeclass(VERB_ALGO, "negcache insert for zone",
903		dname, LDNS_RR_TYPE_SOA, rrset_class);
904
905	/* ask for enough space to store all of it */
906	need = calc_data_need(rep) +
907		calc_zone_need(dname, dname_len);
908	lock_basic_lock(&neg->lock);
909	neg_make_space(neg, need);
910
911	/* find or create the zone entry */
912	zone = neg_find_zone(neg, dname, dname_len, rrset_class);
913	if(!zone) {
914		if(!(zone = neg_create_zone(neg, dname, dname_len,
915			rrset_class))) {
916			lock_basic_unlock(&neg->lock);
917			log_err("out of memory adding negative zone");
918			return;
919		}
920	}
921	val_neg_zone_take_inuse(zone);
922
923	/* insert the NSECs */
924	for(i=rep->an_numrrsets; i< rep->an_numrrsets+rep->ns_numrrsets; i++){
925		if(ntohs(rep->rrsets[i]->rk.type) != LDNS_RR_TYPE_NSEC)
926			continue;
927		if(!dname_subdomain_c(rep->rrsets[i]->rk.dname,
928			zone->name)) continue;
929		/* insert NSEC into this zone's tree */
930		neg_insert_data(neg, zone, rep->rrsets[i]);
931	}
932	if(zone->tree.count == 0) {
933		/* remove empty zone if inserts failed */
934		neg_delete_zone(neg, zone);
935	}
936	lock_basic_unlock(&neg->lock);
937}
938
939/**
940 * Lookup closest data record. For NSEC denial.
941 * @param zone: zone to look in
942 * @param qname: name to look for.
943 * @param len: length of name
944 * @param labs: labels in name
945 * @param data: data element, exact or smaller or NULL
946 * @return true if exact match.
947 */
948static int neg_closest_data(struct val_neg_zone* zone,
949	uint8_t* qname, size_t len, int labs, struct val_neg_data** data)
950{
951	struct val_neg_data key;
952	rbnode_type* r;
953	key.node.key = &key;
954	key.name = qname;
955	key.len = len;
956	key.labs = labs;
957	if(rbtree_find_less_equal(&zone->tree, &key, &r)) {
958		/* exact match */
959		*data = (struct val_neg_data*)r;
960		return 1;
961	} else {
962		/* smaller match */
963		*data = (struct val_neg_data*)r;
964		return 0;
965	}
966}
967
968int val_neg_dlvlookup(struct val_neg_cache* neg, uint8_t* qname, size_t len,
969        uint16_t qclass, struct rrset_cache* rrset_cache, time_t now)
970{
971	/* lookup closest zone */
972	struct val_neg_zone* zone;
973	struct val_neg_data* data;
974	int labs;
975	struct ub_packed_rrset_key* nsec;
976	struct packed_rrset_data* d;
977	uint32_t flags;
978	uint8_t* wc;
979	struct query_info qinfo;
980	if(!neg) return 0;
981
982	log_nametypeclass(VERB_ALGO, "negcache dlvlookup", qname,
983		LDNS_RR_TYPE_DLV, qclass);
984
985	labs = dname_count_labels(qname);
986	lock_basic_lock(&neg->lock);
987	zone = neg_closest_zone_parent(neg, qname, len, labs, qclass);
988	while(zone && !zone->in_use)
989		zone = zone->parent;
990	if(!zone) {
991		lock_basic_unlock(&neg->lock);
992		return 0;
993	}
994	log_nametypeclass(VERB_ALGO, "negcache zone", zone->name, 0,
995		zone->dclass);
996
997	/* DLV is defined to use NSEC only */
998	if(zone->nsec3_hash) {
999		lock_basic_unlock(&neg->lock);
1000		return 0;
1001	}
1002
1003	/* lookup closest data record */
1004	(void)neg_closest_data(zone, qname, len, labs, &data);
1005	while(data && !data->in_use)
1006		data = data->parent;
1007	if(!data) {
1008		lock_basic_unlock(&neg->lock);
1009		return 0;
1010	}
1011	log_nametypeclass(VERB_ALGO, "negcache rr", data->name,
1012		LDNS_RR_TYPE_NSEC, zone->dclass);
1013
1014	/* lookup rrset in rrset cache */
1015	flags = 0;
1016	if(query_dname_compare(data->name, zone->name) == 0)
1017		flags = PACKED_RRSET_NSEC_AT_APEX;
1018	nsec = rrset_cache_lookup(rrset_cache, data->name, data->len,
1019		LDNS_RR_TYPE_NSEC, zone->dclass, flags, now, 0);
1020
1021	/* check if secure and TTL ok */
1022	if(!nsec) {
1023		lock_basic_unlock(&neg->lock);
1024		return 0;
1025	}
1026	d = (struct packed_rrset_data*)nsec->entry.data;
1027	if(!d || now > d->ttl) {
1028		lock_rw_unlock(&nsec->entry.lock);
1029		/* delete data record if expired */
1030		neg_delete_data(neg, data);
1031		lock_basic_unlock(&neg->lock);
1032		return 0;
1033	}
1034	if(d->security != sec_status_secure) {
1035		lock_rw_unlock(&nsec->entry.lock);
1036		neg_delete_data(neg, data);
1037		lock_basic_unlock(&neg->lock);
1038		return 0;
1039	}
1040	verbose(VERB_ALGO, "negcache got secure rrset");
1041
1042	/* check NSEC security */
1043	/* check if NSEC proves no DLV type exists */
1044	/* check if NSEC proves NXDOMAIN for qname */
1045	qinfo.qname = qname;
1046	qinfo.qtype = LDNS_RR_TYPE_DLV;
1047	qinfo.qclass = qclass;
1048	qinfo.local_alias = NULL;
1049	if(!nsec_proves_nodata(nsec, &qinfo, &wc) &&
1050		!val_nsec_proves_name_error(nsec, qname)) {
1051		/* the NSEC is not a denial for the DLV */
1052		lock_rw_unlock(&nsec->entry.lock);
1053		lock_basic_unlock(&neg->lock);
1054		verbose(VERB_ALGO, "negcache not proven");
1055		return 0;
1056	}
1057	/* so the NSEC was a NODATA proof, or NXDOMAIN proof. */
1058
1059	/* no need to check for wildcard NSEC; no wildcards in DLV repos */
1060	/* no need to lookup SOA record for client; no response message */
1061
1062	lock_rw_unlock(&nsec->entry.lock);
1063	/* if OK touch the LRU for neg_data element */
1064	neg_lru_touch(neg, data);
1065	lock_basic_unlock(&neg->lock);
1066	verbose(VERB_ALGO, "negcache DLV denial proven");
1067	return 1;
1068}
1069
1070void val_neg_addreferral(struct val_neg_cache* neg, struct reply_info* rep,
1071	uint8_t* zone_name)
1072{
1073	size_t i, need;
1074	uint8_t* signer;
1075	size_t signer_len;
1076	uint16_t dclass;
1077	struct val_neg_zone* zone;
1078	/* no SOA in this message, find RRSIG over NSEC's signer name.
1079	 * note the NSEC records are maybe not validated yet */
1080	signer = reply_nsec_signer(rep, &signer_len, &dclass);
1081	if(!signer)
1082		return;
1083	if(!dname_subdomain_c(signer, zone_name)) {
1084		/* the signer is not in the bailiwick, throw it out */
1085		return;
1086	}
1087
1088	log_nametypeclass(VERB_ALGO, "negcache insert referral ",
1089		signer, LDNS_RR_TYPE_NS, dclass);
1090
1091	/* ask for enough space to store all of it */
1092	need = calc_data_need(rep) + calc_zone_need(signer, signer_len);
1093	lock_basic_lock(&neg->lock);
1094	neg_make_space(neg, need);
1095
1096	/* find or create the zone entry */
1097	zone = neg_find_zone(neg, signer, signer_len, dclass);
1098	if(!zone) {
1099		if(!(zone = neg_create_zone(neg, signer, signer_len,
1100			dclass))) {
1101			lock_basic_unlock(&neg->lock);
1102			log_err("out of memory adding negative zone");
1103			return;
1104		}
1105	}
1106	val_neg_zone_take_inuse(zone);
1107
1108	/* insert the NSECs */
1109	for(i=rep->an_numrrsets; i< rep->an_numrrsets+rep->ns_numrrsets; i++){
1110		if(ntohs(rep->rrsets[i]->rk.type) != LDNS_RR_TYPE_NSEC &&
1111			ntohs(rep->rrsets[i]->rk.type) != LDNS_RR_TYPE_NSEC3)
1112			continue;
1113		if(!dname_subdomain_c(rep->rrsets[i]->rk.dname,
1114			zone->name)) continue;
1115		/* insert NSEC into this zone's tree */
1116		neg_insert_data(neg, zone, rep->rrsets[i]);
1117	}
1118	if(zone->tree.count == 0) {
1119		/* remove empty zone if inserts failed */
1120		neg_delete_zone(neg, zone);
1121	}
1122	lock_basic_unlock(&neg->lock);
1123}
1124
1125/**
1126 * Check that an NSEC3 rrset does not have a type set.
1127 * None of the nsec3s in a hash-collision are allowed to have the type.
1128 * (since we do not know which one is the nsec3 looked at, flags, ..., we
1129 * ignore the cached item and let it bypass negative caching).
1130 * @param k: the nsec3 rrset to check.
1131 * @param t: type to check
1132 * @return true if no RRs have the type.
1133 */
1134static int nsec3_no_type(struct ub_packed_rrset_key* k, uint16_t t)
1135{
1136	int count = (int)((struct packed_rrset_data*)k->entry.data)->count;
1137	int i;
1138	for(i=0; i<count; i++)
1139		if(nsec3_has_type(k, i, t))
1140			return 0;
1141	return 1;
1142}
1143
1144/**
1145 * See if rrset exists in rrset cache.
1146 * If it does, the bit is checked, and if not expired, it is returned
1147 * allocated in region.
1148 * @param rrset_cache: rrset cache
1149 * @param qname: to lookup rrset name
1150 * @param qname_len: length of qname.
1151 * @param qtype: type of rrset to lookup, host order
1152 * @param qclass: class of rrset to lookup, host order
1153 * @param flags: flags for rrset to lookup
1154 * @param region: where to alloc result
1155 * @param checkbit: if true, a bit in the nsec typemap is checked for absence.
1156 * @param checktype: which bit to check
1157 * @param now: to check ttl against
1158 * @return rrset or NULL
1159 */
1160static struct ub_packed_rrset_key*
1161grab_nsec(struct rrset_cache* rrset_cache, uint8_t* qname, size_t qname_len,
1162	uint16_t qtype, uint16_t qclass, uint32_t flags,
1163	struct regional* region, int checkbit, uint16_t checktype,
1164	time_t now)
1165{
1166	struct ub_packed_rrset_key* r, *k = rrset_cache_lookup(rrset_cache,
1167		qname, qname_len, qtype, qclass, flags, now, 0);
1168	struct packed_rrset_data* d;
1169	if(!k) return NULL;
1170	d = (struct packed_rrset_data*)k->entry.data;
1171	if(d->ttl < now) {
1172		lock_rw_unlock(&k->entry.lock);
1173		return NULL;
1174	}
1175	/* only secure or unchecked records that have signatures. */
1176	if( ! ( d->security == sec_status_secure ||
1177		(d->security == sec_status_unchecked &&
1178		d->rrsig_count > 0) ) ) {
1179		lock_rw_unlock(&k->entry.lock);
1180		return NULL;
1181	}
1182	/* check if checktype is absent */
1183	if(checkbit && (
1184		(qtype == LDNS_RR_TYPE_NSEC && nsec_has_type(k, checktype)) ||
1185		(qtype == LDNS_RR_TYPE_NSEC3 && !nsec3_no_type(k, checktype))
1186		)) {
1187		lock_rw_unlock(&k->entry.lock);
1188		return NULL;
1189	}
1190	/* looks OK! copy to region and return it */
1191	r = packed_rrset_copy_region(k, region, now);
1192	/* if it failed, we return the NULL */
1193	lock_rw_unlock(&k->entry.lock);
1194	return r;
1195}
1196
1197/**
1198 * Get best NSEC record for qname. Might be matching, covering or totally
1199 * useless.
1200 * @param neg_cache: neg cache
1201 * @param qname: to lookup rrset name
1202 * @param qname_len: length of qname.
1203 * @param qclass: class of rrset to lookup, host order
1204 * @param rrset_cache: rrset cache
1205 * @param now: to check ttl against
1206 * @param region: where to alloc result
1207 * @return rrset or NULL
1208 */
1209static struct ub_packed_rrset_key*
1210neg_find_nsec(struct val_neg_cache* neg_cache, uint8_t* qname, size_t qname_len,
1211	uint16_t qclass, struct rrset_cache* rrset_cache, time_t now,
1212	struct regional* region)
1213{
1214	int labs;
1215	uint32_t flags;
1216	struct val_neg_zone* zone;
1217	struct val_neg_data* data;
1218	struct ub_packed_rrset_key* nsec;
1219
1220	labs = dname_count_labels(qname);
1221	lock_basic_lock(&neg_cache->lock);
1222	zone = neg_closest_zone_parent(neg_cache, qname, qname_len, labs,
1223		qclass);
1224	while(zone && !zone->in_use)
1225		zone = zone->parent;
1226	if(!zone) {
1227		lock_basic_unlock(&neg_cache->lock);
1228		return NULL;
1229	}
1230
1231	/* NSEC only for now */
1232	if(zone->nsec3_hash) {
1233		lock_basic_unlock(&neg_cache->lock);
1234		return NULL;
1235	}
1236
1237	/* ignore return value, don't care if it is an exact or smaller match */
1238	(void)neg_closest_data(zone, qname, qname_len, labs, &data);
1239	if(!data) {
1240		lock_basic_unlock(&neg_cache->lock);
1241		return NULL;
1242	}
1243
1244	/* ENT nodes are not in use, try the previous node. If the previous node
1245	 * is not in use, we don't have an useful NSEC and give up. */
1246	if(!data->in_use) {
1247		data = (struct val_neg_data*)rbtree_previous((rbnode_type*)data);
1248		if((rbnode_type*)data == RBTREE_NULL || !data->in_use) {
1249			lock_basic_unlock(&neg_cache->lock);
1250			return NULL;
1251		}
1252	}
1253
1254	flags = 0;
1255	if(query_dname_compare(data->name, zone->name) == 0)
1256		flags = PACKED_RRSET_NSEC_AT_APEX;
1257
1258	nsec = grab_nsec(rrset_cache, data->name, data->len, LDNS_RR_TYPE_NSEC,
1259		zone->dclass, flags, region, 0, 0, now);
1260	lock_basic_unlock(&neg_cache->lock);
1261	return nsec;
1262}
1263
1264/** find nsec3 closest encloser in neg cache */
1265static struct val_neg_data*
1266neg_find_nsec3_ce(struct val_neg_zone* zone, uint8_t* qname, size_t qname_len,
1267		int qlabs, sldns_buffer* buf, uint8_t* hashnc, size_t* nclen)
1268{
1269	struct val_neg_data* data;
1270	uint8_t hashce[NSEC3_SHA_LEN];
1271	uint8_t b32[257];
1272	size_t celen, b32len;
1273
1274	*nclen = 0;
1275	while(qlabs > 0) {
1276		/* hash */
1277		if(!(celen=nsec3_get_hashed(buf, qname, qname_len,
1278			zone->nsec3_hash, zone->nsec3_iter, zone->nsec3_salt,
1279			zone->nsec3_saltlen, hashce, sizeof(hashce))))
1280			return NULL;
1281		if(!(b32len=nsec3_hash_to_b32(hashce, celen, zone->name,
1282			zone->len, b32, sizeof(b32))))
1283			return NULL;
1284
1285		/* lookup (exact match only) */
1286		data = neg_find_data(zone, b32, b32len, zone->labs+1);
1287		if(data && data->in_use) {
1288			/* found ce match! */
1289			return data;
1290		}
1291
1292		*nclen = celen;
1293		memmove(hashnc, hashce, celen);
1294		dname_remove_label(&qname, &qname_len);
1295		qlabs --;
1296	}
1297	return NULL;
1298}
1299
1300/** check nsec3 parameters on nsec3 rrset with current zone values */
1301static int
1302neg_params_ok(struct val_neg_zone* zone, struct ub_packed_rrset_key* rrset)
1303{
1304	int h;
1305	uint8_t* s;
1306	size_t slen, it;
1307	if(!nsec3_get_params(rrset, 0, &h, &it, &s, &slen))
1308		return 0;
1309	return (h == zone->nsec3_hash && it == zone->nsec3_iter &&
1310		slen == zone->nsec3_saltlen &&
1311		memcmp(zone->nsec3_salt, s, slen) == 0);
1312}
1313
1314/** get next closer for nsec3 proof */
1315static struct ub_packed_rrset_key*
1316neg_nsec3_getnc(struct val_neg_zone* zone, uint8_t* hashnc, size_t nclen,
1317	struct rrset_cache* rrset_cache, struct regional* region,
1318	time_t now, uint8_t* b32, size_t maxb32)
1319{
1320	struct ub_packed_rrset_key* nc_rrset;
1321	struct val_neg_data* data;
1322	size_t b32len;
1323
1324	if(!(b32len=nsec3_hash_to_b32(hashnc, nclen, zone->name,
1325		zone->len, b32, maxb32)))
1326		return NULL;
1327	(void)neg_closest_data(zone, b32, b32len, zone->labs+1, &data);
1328	if(!data && zone->tree.count != 0) {
1329		/* could be before the first entry ; return the last
1330		 * entry (possibly the rollover nsec3 at end) */
1331		data = (struct val_neg_data*)rbtree_last(&zone->tree);
1332	}
1333	while(data && !data->in_use)
1334		data = data->parent;
1335	if(!data)
1336		return NULL;
1337	/* got a data element in tree, grab it */
1338	nc_rrset = grab_nsec(rrset_cache, data->name, data->len,
1339		LDNS_RR_TYPE_NSEC3, zone->dclass, 0, region, 0, 0, now);
1340	if(!nc_rrset)
1341		return NULL;
1342	if(!neg_params_ok(zone, nc_rrset))
1343		return NULL;
1344	return nc_rrset;
1345}
1346
1347/** neg cache nsec3 proof procedure*/
1348static struct dns_msg*
1349neg_nsec3_proof_ds(struct val_neg_zone* zone, uint8_t* qname, size_t qname_len,
1350		int qlabs, sldns_buffer* buf, struct rrset_cache* rrset_cache,
1351		struct regional* region, time_t now, uint8_t* topname)
1352{
1353	struct dns_msg* msg;
1354	struct val_neg_data* data;
1355	uint8_t hashnc[NSEC3_SHA_LEN];
1356	size_t nclen;
1357	struct ub_packed_rrset_key* ce_rrset, *nc_rrset;
1358	struct nsec3_cached_hash c;
1359	uint8_t nc_b32[257];
1360
1361	/* for NSEC3 ; determine the closest encloser for which we
1362	 * can find an exact match. Remember the hashed lower name,
1363	 * since that is the one we need a closest match for.
1364	 * If we find a match straight away, then it becomes NODATA.
1365	 * Otherwise, NXDOMAIN or if OPTOUT, an insecure delegation.
1366	 * Also check that parameters are the same on closest encloser
1367	 * and on closest match.
1368	 */
1369	if(!zone->nsec3_hash)
1370		return NULL; /* not nsec3 zone */
1371
1372	if(!(data=neg_find_nsec3_ce(zone, qname, qname_len, qlabs, buf,
1373		hashnc, &nclen))) {
1374		return NULL;
1375	}
1376
1377	/* grab the ce rrset */
1378	ce_rrset = grab_nsec(rrset_cache, data->name, data->len,
1379		LDNS_RR_TYPE_NSEC3, zone->dclass, 0, region, 1,
1380		LDNS_RR_TYPE_DS, now);
1381	if(!ce_rrset)
1382		return NULL;
1383	if(!neg_params_ok(zone, ce_rrset))
1384		return NULL;
1385
1386	if(nclen == 0) {
1387		/* exact match, just check the type bits */
1388		/* need: -SOA, -DS, +NS */
1389		if(nsec3_has_type(ce_rrset, 0, LDNS_RR_TYPE_SOA) ||
1390			nsec3_has_type(ce_rrset, 0, LDNS_RR_TYPE_DS) ||
1391			!nsec3_has_type(ce_rrset, 0, LDNS_RR_TYPE_NS))
1392			return NULL;
1393		if(!(msg = dns_msg_create(qname, qname_len,
1394			LDNS_RR_TYPE_DS, zone->dclass, region, 1)))
1395			return NULL;
1396		/* TTL reduced in grab_nsec */
1397		if(!dns_msg_authadd(msg, region, ce_rrset, 0))
1398			return NULL;
1399		return msg;
1400	}
1401
1402	/* optout is not allowed without knowing the trust-anchor in use,
1403	 * otherwise the optout could spoof away that anchor */
1404	if(!topname)
1405		return NULL;
1406
1407	/* if there is no exact match, it must be in an optout span
1408	 * (an existing DS implies an NSEC3 must exist) */
1409	nc_rrset = neg_nsec3_getnc(zone, hashnc, nclen, rrset_cache,
1410		region, now, nc_b32, sizeof(nc_b32));
1411	if(!nc_rrset)
1412		return NULL;
1413	if(!neg_params_ok(zone, nc_rrset))
1414		return NULL;
1415	if(!nsec3_has_optout(nc_rrset, 0))
1416		return NULL;
1417	c.hash = hashnc;
1418	c.hash_len = nclen;
1419	c.b32 = nc_b32+1;
1420	c.b32_len = (size_t)nc_b32[0];
1421	if(nsec3_covers(zone->name, &c, nc_rrset, 0, buf)) {
1422		/* nc_rrset covers the next closer name.
1423		 * ce_rrset equals a closer encloser.
1424		 * nc_rrset is optout.
1425		 * No need to check wildcard for type DS */
1426		/* capacity=3: ce + nc + soa(if needed) */
1427		if(!(msg = dns_msg_create(qname, qname_len,
1428			LDNS_RR_TYPE_DS, zone->dclass, region, 3)))
1429			return NULL;
1430		/* now=0 because TTL was reduced in grab_nsec */
1431		if(!dns_msg_authadd(msg, region, ce_rrset, 0))
1432			return NULL;
1433		if(!dns_msg_authadd(msg, region, nc_rrset, 0))
1434			return NULL;
1435		return msg;
1436	}
1437	return NULL;
1438}
1439
1440/**
1441 * Add SOA record for external responses.
1442 * @param rrset_cache: to look into.
1443 * @param now: current time.
1444 * @param region: where to perform the allocation
1445 * @param msg: current msg with NSEC.
1446 * @param zone: val_neg_zone if we have one.
1447 * @return false on lookup or alloc failure.
1448 */
1449static int add_soa(struct rrset_cache* rrset_cache, time_t now,
1450	struct regional* region, struct dns_msg* msg, struct val_neg_zone* zone)
1451{
1452	struct ub_packed_rrset_key* soa;
1453	uint8_t* nm;
1454	size_t nmlen;
1455	uint16_t dclass;
1456	if(zone) {
1457		nm = zone->name;
1458		nmlen = zone->len;
1459		dclass = zone->dclass;
1460	} else {
1461		/* Assumes the signer is the zone SOA to add */
1462		nm = reply_nsec_signer(msg->rep, &nmlen, &dclass);
1463		if(!nm)
1464			return 0;
1465	}
1466	soa = rrset_cache_lookup(rrset_cache, nm, nmlen, LDNS_RR_TYPE_SOA,
1467		dclass, PACKED_RRSET_SOA_NEG, now, 0);
1468	if(!soa)
1469		return 0;
1470	if(!dns_msg_authadd(msg, region, soa, now)) {
1471		lock_rw_unlock(&soa->entry.lock);
1472		return 0;
1473	}
1474	lock_rw_unlock(&soa->entry.lock);
1475	return 1;
1476}
1477
1478struct dns_msg*
1479val_neg_getmsg(struct val_neg_cache* neg, struct query_info* qinfo,
1480	struct regional* region, struct rrset_cache* rrset_cache,
1481	sldns_buffer* buf, time_t now, int addsoa, uint8_t* topname,
1482	struct config_file* cfg)
1483{
1484	struct dns_msg* msg;
1485	struct ub_packed_rrset_key* nsec; /* qname matching/covering nsec */
1486	struct ub_packed_rrset_key* wcrr; /* wildcard record or nsec */
1487	uint8_t* nodata_wc = NULL;
1488	uint8_t* ce = NULL;
1489	size_t ce_len;
1490	uint8_t wc_ce[LDNS_MAX_DOMAINLEN+3];
1491	struct query_info wc_qinfo;
1492	struct ub_packed_rrset_key* cache_wc;
1493	struct packed_rrset_data* wcrr_data;
1494	int rcode = LDNS_RCODE_NOERROR;
1495	uint8_t* zname;
1496	size_t zname_len;
1497	int zname_labs;
1498	struct val_neg_zone* zone;
1499
1500	/* only for DS queries when aggressive use of NSEC is disabled */
1501	if(qinfo->qtype != LDNS_RR_TYPE_DS && !cfg->aggressive_nsec)
1502		return NULL;
1503	log_assert(!topname || dname_subdomain_c(qinfo->qname, topname));
1504
1505	/* Get best available NSEC for qname */
1506	nsec = neg_find_nsec(neg, qinfo->qname, qinfo->qname_len, qinfo->qclass,
1507		rrset_cache, now, region);
1508
1509	/* Matching NSEC, use to generate No Data answer. Not creating answers
1510	 * yet for No Data proven using wildcard. */
1511	if(nsec && nsec_proves_nodata(nsec, qinfo, &nodata_wc) && !nodata_wc) {
1512		if(!(msg = dns_msg_create(qinfo->qname, qinfo->qname_len,
1513			qinfo->qtype, qinfo->qclass, region, 2)))
1514			return NULL;
1515		if(!dns_msg_authadd(msg, region, nsec, 0))
1516			return NULL;
1517		if(addsoa && !add_soa(rrset_cache, now, region, msg, NULL))
1518			return NULL;
1519
1520		lock_basic_lock(&neg->lock);
1521		neg->num_neg_cache_noerror++;
1522		lock_basic_unlock(&neg->lock);
1523		return msg;
1524	} else if(nsec && val_nsec_proves_name_error(nsec, qinfo->qname)) {
1525		if(!(msg = dns_msg_create(qinfo->qname, qinfo->qname_len,
1526			qinfo->qtype, qinfo->qclass, region, 3)))
1527			return NULL;
1528		if(!(ce = nsec_closest_encloser(qinfo->qname, nsec)))
1529			return NULL;
1530		dname_count_size_labels(ce, &ce_len);
1531
1532		/* No extra extra NSEC required if both nameerror qname and
1533		 * nodata *.ce. are proven already. */
1534		if(!nodata_wc || query_dname_compare(nodata_wc, ce) != 0) {
1535			/* Qname proven non existing, get wildcard record for
1536			 * QTYPE or NSEC covering or matching wildcard. */
1537
1538			/* Num labels in ce is always smaller than in qname,
1539			 * therefore adding the wildcard label cannot overflow
1540			 * buffer. */
1541			wc_ce[0] = 1;
1542			wc_ce[1] = (uint8_t)'*';
1543			memmove(wc_ce+2, ce, ce_len);
1544			wc_qinfo.qname = wc_ce;
1545			wc_qinfo.qname_len = ce_len + 2;
1546			wc_qinfo.qtype = qinfo->qtype;
1547
1548
1549			if((cache_wc = rrset_cache_lookup(rrset_cache, wc_qinfo.qname,
1550				wc_qinfo.qname_len, wc_qinfo.qtype,
1551				qinfo->qclass, 0/*flags*/, now, 0/*read only*/))) {
1552				/* Synthesize wildcard answer */
1553				wcrr_data = (struct packed_rrset_data*)cache_wc->entry.data;
1554				if(!(wcrr_data->security == sec_status_secure ||
1555					(wcrr_data->security == sec_status_unchecked &&
1556					wcrr_data->rrsig_count > 0))) {
1557					lock_rw_unlock(&cache_wc->entry.lock);
1558					return NULL;
1559				}
1560				if(!(wcrr = packed_rrset_copy_region(cache_wc,
1561					region, now))) {
1562					lock_rw_unlock(&cache_wc->entry.lock);
1563					return NULL;
1564				};
1565				lock_rw_unlock(&cache_wc->entry.lock);
1566				wcrr->rk.dname = qinfo->qname;
1567				wcrr->rk.dname_len = qinfo->qname_len;
1568				if(!dns_msg_ansadd(msg, region, wcrr, 0))
1569					return NULL;
1570				/* No SOA needed for wildcard synthesised
1571				 * answer. */
1572				addsoa = 0;
1573			} else {
1574				/* Get wildcard NSEC for possible non existence
1575				 * proof */
1576				if(!(wcrr = neg_find_nsec(neg, wc_qinfo.qname,
1577					wc_qinfo.qname_len, qinfo->qclass,
1578					rrset_cache, now, region)))
1579					return NULL;
1580
1581				nodata_wc = NULL;
1582				if(val_nsec_proves_name_error(wcrr, wc_ce))
1583					rcode = LDNS_RCODE_NXDOMAIN;
1584				else if(!nsec_proves_nodata(wcrr, &wc_qinfo,
1585					&nodata_wc) || nodata_wc)
1586					/* &nodata_wc shouldn't be set, wc_qinfo
1587					 * already contains wildcard domain. */
1588					/* NSEC doesn't prove anything for
1589					 * wildcard. */
1590					return NULL;
1591				if(query_dname_compare(wcrr->rk.dname,
1592					nsec->rk.dname) != 0)
1593					if(!dns_msg_authadd(msg, region, wcrr, 0))
1594						return NULL;
1595			}
1596		}
1597
1598		if(!dns_msg_authadd(msg, region, nsec, 0))
1599			return NULL;
1600		if(addsoa && !add_soa(rrset_cache, now, region, msg, NULL))
1601			return NULL;
1602
1603		/* Increment statistic counters */
1604		lock_basic_lock(&neg->lock);
1605		if(rcode == LDNS_RCODE_NOERROR)
1606			neg->num_neg_cache_noerror++;
1607		else if(rcode == LDNS_RCODE_NXDOMAIN)
1608			neg->num_neg_cache_nxdomain++;
1609		lock_basic_unlock(&neg->lock);
1610
1611		FLAGS_SET_RCODE(msg->rep->flags, rcode);
1612		return msg;
1613	}
1614
1615	/* No aggressive use of NSEC3 for now, only proceed for DS types. */
1616	if(qinfo->qtype != LDNS_RR_TYPE_DS){
1617		return NULL;
1618	}
1619	/* check NSEC3 neg cache for type DS */
1620	/* need to look one zone higher for DS type */
1621	zname = qinfo->qname;
1622	zname_len = qinfo->qname_len;
1623	dname_remove_label(&zname, &zname_len);
1624	zname_labs = dname_count_labels(zname);
1625
1626	/* lookup closest zone */
1627	lock_basic_lock(&neg->lock);
1628	zone = neg_closest_zone_parent(neg, zname, zname_len, zname_labs,
1629		qinfo->qclass);
1630	while(zone && !zone->in_use)
1631		zone = zone->parent;
1632	/* check that the zone is not too high up so that we do not pick data
1633	 * out of a zone that is above the last-seen key (or trust-anchor). */
1634	if(zone && topname) {
1635		if(!dname_subdomain_c(zone->name, topname))
1636			zone = NULL;
1637	}
1638	if(!zone) {
1639		lock_basic_unlock(&neg->lock);
1640		return NULL;
1641	}
1642
1643	msg = neg_nsec3_proof_ds(zone, qinfo->qname, qinfo->qname_len,
1644		zname_labs+1, buf, rrset_cache, region, now, topname);
1645	if(msg && addsoa && !add_soa(rrset_cache, now, region, msg, zone)) {
1646		lock_basic_unlock(&neg->lock);
1647		return NULL;
1648	}
1649	lock_basic_unlock(&neg->lock);
1650	return msg;
1651}
1652