xref: /freebsd/contrib/unbound/validator/val_neg.c (revision a7623790fb345e6dc986dfd31df0ace115e6f2e4)
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 
65 int 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 
73 int 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 
86 struct 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 
103 size_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 */
113 static void
114 neg_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 */
122 static void
123 neg_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 
133 void 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  */
147 static 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  */
163 static 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  */
179 static 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  */
197 static 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 
226 void 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  */
272 static 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 
280 struct 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  */
304 static 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  */
324 static 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  */
350 static 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  */
370 static 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  */
410 static 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  */
449 static 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  */
482 static 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 
518 void 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 
530 struct 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 */
566 static 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 */
577 static 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  */
602 static 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  */
629 static 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  */
673 static 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 
748 void 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 */
852 static 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 
878 void 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  */
948 static 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 
968 void val_neg_addreferral(struct val_neg_cache* neg, struct reply_info* rep,
969 	uint8_t* zone_name)
970 {
971 	size_t i, need;
972 	uint8_t* signer;
973 	size_t signer_len;
974 	uint16_t dclass;
975 	struct val_neg_zone* zone;
976 	/* no SOA in this message, find RRSIG over NSEC's signer name.
977 	 * note the NSEC records are maybe not validated yet */
978 	signer = reply_nsec_signer(rep, &signer_len, &dclass);
979 	if(!signer)
980 		return;
981 	if(!dname_subdomain_c(signer, zone_name)) {
982 		/* the signer is not in the bailiwick, throw it out */
983 		return;
984 	}
985 
986 	log_nametypeclass(VERB_ALGO, "negcache insert referral ",
987 		signer, LDNS_RR_TYPE_NS, dclass);
988 
989 	/* ask for enough space to store all of it */
990 	need = calc_data_need(rep) + calc_zone_need(signer, signer_len);
991 	lock_basic_lock(&neg->lock);
992 	neg_make_space(neg, need);
993 
994 	/* find or create the zone entry */
995 	zone = neg_find_zone(neg, signer, signer_len, dclass);
996 	if(!zone) {
997 		if(!(zone = neg_create_zone(neg, signer, signer_len,
998 			dclass))) {
999 			lock_basic_unlock(&neg->lock);
1000 			log_err("out of memory adding negative zone");
1001 			return;
1002 		}
1003 	}
1004 	val_neg_zone_take_inuse(zone);
1005 
1006 	/* insert the NSECs */
1007 	for(i=rep->an_numrrsets; i< rep->an_numrrsets+rep->ns_numrrsets; i++){
1008 		if(ntohs(rep->rrsets[i]->rk.type) != LDNS_RR_TYPE_NSEC &&
1009 			ntohs(rep->rrsets[i]->rk.type) != LDNS_RR_TYPE_NSEC3)
1010 			continue;
1011 		if(!dname_subdomain_c(rep->rrsets[i]->rk.dname,
1012 			zone->name)) continue;
1013 		/* insert NSEC into this zone's tree */
1014 		neg_insert_data(neg, zone, rep->rrsets[i]);
1015 	}
1016 	if(zone->tree.count == 0) {
1017 		/* remove empty zone if inserts failed */
1018 		neg_delete_zone(neg, zone);
1019 	}
1020 	lock_basic_unlock(&neg->lock);
1021 }
1022 
1023 /**
1024  * Check that an NSEC3 rrset does not have a type set.
1025  * None of the nsec3s in a hash-collision are allowed to have the type.
1026  * (since we do not know which one is the nsec3 looked at, flags, ..., we
1027  * ignore the cached item and let it bypass negative caching).
1028  * @param k: the nsec3 rrset to check.
1029  * @param t: type to check
1030  * @return true if no RRs have the type.
1031  */
1032 static int nsec3_no_type(struct ub_packed_rrset_key* k, uint16_t t)
1033 {
1034 	int count = (int)((struct packed_rrset_data*)k->entry.data)->count;
1035 	int i;
1036 	for(i=0; i<count; i++)
1037 		if(nsec3_has_type(k, i, t))
1038 			return 0;
1039 	return 1;
1040 }
1041 
1042 /**
1043  * See if rrset exists in rrset cache.
1044  * If it does, the bit is checked, and if not expired, it is returned
1045  * allocated in region.
1046  * @param rrset_cache: rrset cache
1047  * @param qname: to lookup rrset name
1048  * @param qname_len: length of qname.
1049  * @param qtype: type of rrset to lookup, host order
1050  * @param qclass: class of rrset to lookup, host order
1051  * @param flags: flags for rrset to lookup
1052  * @param region: where to alloc result
1053  * @param checkbit: if true, a bit in the nsec typemap is checked for absence.
1054  * @param checktype: which bit to check
1055  * @param now: to check ttl against
1056  * @return rrset or NULL
1057  */
1058 static struct ub_packed_rrset_key*
1059 grab_nsec(struct rrset_cache* rrset_cache, uint8_t* qname, size_t qname_len,
1060 	uint16_t qtype, uint16_t qclass, uint32_t flags,
1061 	struct regional* region, int checkbit, uint16_t checktype,
1062 	time_t now)
1063 {
1064 	struct ub_packed_rrset_key* r, *k = rrset_cache_lookup(rrset_cache,
1065 		qname, qname_len, qtype, qclass, flags, now, 0);
1066 	struct packed_rrset_data* d;
1067 	if(!k) return NULL;
1068 	d = (struct packed_rrset_data*)k->entry.data;
1069 	if(d->ttl < now) {
1070 		lock_rw_unlock(&k->entry.lock);
1071 		return NULL;
1072 	}
1073 	/* only secure or unchecked records that have signatures. */
1074 	if( ! ( d->security == sec_status_secure ||
1075 		(d->security == sec_status_unchecked &&
1076 		d->rrsig_count > 0) ) ) {
1077 		lock_rw_unlock(&k->entry.lock);
1078 		return NULL;
1079 	}
1080 	/* check if checktype is absent */
1081 	if(checkbit && (
1082 		(qtype == LDNS_RR_TYPE_NSEC && nsec_has_type(k, checktype)) ||
1083 		(qtype == LDNS_RR_TYPE_NSEC3 && !nsec3_no_type(k, checktype))
1084 		)) {
1085 		lock_rw_unlock(&k->entry.lock);
1086 		return NULL;
1087 	}
1088 	/* looks OK! copy to region and return it */
1089 	r = packed_rrset_copy_region(k, region, now);
1090 	/* if it failed, we return the NULL */
1091 	lock_rw_unlock(&k->entry.lock);
1092 	return r;
1093 }
1094 
1095 /**
1096  * Get best NSEC record for qname. Might be matching, covering or totally
1097  * useless.
1098  * @param neg_cache: neg cache
1099  * @param qname: to lookup rrset name
1100  * @param qname_len: length of qname.
1101  * @param qclass: class of rrset to lookup, host order
1102  * @param rrset_cache: rrset cache
1103  * @param now: to check ttl against
1104  * @param region: where to alloc result
1105  * @return rrset or NULL
1106  */
1107 static struct ub_packed_rrset_key*
1108 neg_find_nsec(struct val_neg_cache* neg_cache, uint8_t* qname, size_t qname_len,
1109 	uint16_t qclass, struct rrset_cache* rrset_cache, time_t now,
1110 	struct regional* region)
1111 {
1112 	int labs;
1113 	uint32_t flags;
1114 	struct val_neg_zone* zone;
1115 	struct val_neg_data* data;
1116 	struct ub_packed_rrset_key* nsec;
1117 
1118 	labs = dname_count_labels(qname);
1119 	lock_basic_lock(&neg_cache->lock);
1120 	zone = neg_closest_zone_parent(neg_cache, qname, qname_len, labs,
1121 		qclass);
1122 	while(zone && !zone->in_use)
1123 		zone = zone->parent;
1124 	if(!zone) {
1125 		lock_basic_unlock(&neg_cache->lock);
1126 		return NULL;
1127 	}
1128 
1129 	/* NSEC only for now */
1130 	if(zone->nsec3_hash) {
1131 		lock_basic_unlock(&neg_cache->lock);
1132 		return NULL;
1133 	}
1134 
1135 	/* ignore return value, don't care if it is an exact or smaller match */
1136 	(void)neg_closest_data(zone, qname, qname_len, labs, &data);
1137 	if(!data) {
1138 		lock_basic_unlock(&neg_cache->lock);
1139 		return NULL;
1140 	}
1141 
1142 	/* ENT nodes are not in use, try the previous node. If the previous node
1143 	 * is not in use, we don't have an useful NSEC and give up. */
1144 	if(!data->in_use) {
1145 		data = (struct val_neg_data*)rbtree_previous((rbnode_type*)data);
1146 		if((rbnode_type*)data == RBTREE_NULL || !data->in_use) {
1147 			lock_basic_unlock(&neg_cache->lock);
1148 			return NULL;
1149 		}
1150 	}
1151 
1152 	flags = 0;
1153 	if(query_dname_compare(data->name, zone->name) == 0)
1154 		flags = PACKED_RRSET_NSEC_AT_APEX;
1155 
1156 	nsec = grab_nsec(rrset_cache, data->name, data->len, LDNS_RR_TYPE_NSEC,
1157 		zone->dclass, flags, region, 0, 0, now);
1158 	lock_basic_unlock(&neg_cache->lock);
1159 	return nsec;
1160 }
1161 
1162 /** find nsec3 closest encloser in neg cache */
1163 static struct val_neg_data*
1164 neg_find_nsec3_ce(struct val_neg_zone* zone, uint8_t* qname, size_t qname_len,
1165 		int qlabs, sldns_buffer* buf, uint8_t* hashnc, size_t* nclen)
1166 {
1167 	struct val_neg_data* data;
1168 	uint8_t hashce[NSEC3_SHA_LEN];
1169 	uint8_t b32[257];
1170 	size_t celen, b32len;
1171 
1172 	*nclen = 0;
1173 	while(qlabs > 0) {
1174 		/* hash */
1175 		if(!(celen=nsec3_get_hashed(buf, qname, qname_len,
1176 			zone->nsec3_hash, zone->nsec3_iter, zone->nsec3_salt,
1177 			zone->nsec3_saltlen, hashce, sizeof(hashce))))
1178 			return NULL;
1179 		if(!(b32len=nsec3_hash_to_b32(hashce, celen, zone->name,
1180 			zone->len, b32, sizeof(b32))))
1181 			return NULL;
1182 
1183 		/* lookup (exact match only) */
1184 		data = neg_find_data(zone, b32, b32len, zone->labs+1);
1185 		if(data && data->in_use) {
1186 			/* found ce match! */
1187 			return data;
1188 		}
1189 
1190 		*nclen = celen;
1191 		memmove(hashnc, hashce, celen);
1192 		dname_remove_label(&qname, &qname_len);
1193 		qlabs --;
1194 	}
1195 	return NULL;
1196 }
1197 
1198 /** check nsec3 parameters on nsec3 rrset with current zone values */
1199 static int
1200 neg_params_ok(struct val_neg_zone* zone, struct ub_packed_rrset_key* rrset)
1201 {
1202 	int h;
1203 	uint8_t* s;
1204 	size_t slen, it;
1205 	if(!nsec3_get_params(rrset, 0, &h, &it, &s, &slen))
1206 		return 0;
1207 	return (h == zone->nsec3_hash && it == zone->nsec3_iter &&
1208 		slen == zone->nsec3_saltlen &&
1209 		memcmp(zone->nsec3_salt, s, slen) == 0);
1210 }
1211 
1212 /** get next closer for nsec3 proof */
1213 static struct ub_packed_rrset_key*
1214 neg_nsec3_getnc(struct val_neg_zone* zone, uint8_t* hashnc, size_t nclen,
1215 	struct rrset_cache* rrset_cache, struct regional* region,
1216 	time_t now, uint8_t* b32, size_t maxb32)
1217 {
1218 	struct ub_packed_rrset_key* nc_rrset;
1219 	struct val_neg_data* data;
1220 	size_t b32len;
1221 
1222 	if(!(b32len=nsec3_hash_to_b32(hashnc, nclen, zone->name,
1223 		zone->len, b32, maxb32)))
1224 		return NULL;
1225 	(void)neg_closest_data(zone, b32, b32len, zone->labs+1, &data);
1226 	if(!data && zone->tree.count != 0) {
1227 		/* could be before the first entry ; return the last
1228 		 * entry (possibly the rollover nsec3 at end) */
1229 		data = (struct val_neg_data*)rbtree_last(&zone->tree);
1230 	}
1231 	while(data && !data->in_use)
1232 		data = data->parent;
1233 	if(!data)
1234 		return NULL;
1235 	/* got a data element in tree, grab it */
1236 	nc_rrset = grab_nsec(rrset_cache, data->name, data->len,
1237 		LDNS_RR_TYPE_NSEC3, zone->dclass, 0, region, 0, 0, now);
1238 	if(!nc_rrset)
1239 		return NULL;
1240 	if(!neg_params_ok(zone, nc_rrset))
1241 		return NULL;
1242 	return nc_rrset;
1243 }
1244 
1245 /** neg cache nsec3 proof procedure*/
1246 static struct dns_msg*
1247 neg_nsec3_proof_ds(struct val_neg_zone* zone, uint8_t* qname, size_t qname_len,
1248 		int qlabs, sldns_buffer* buf, struct rrset_cache* rrset_cache,
1249 		struct regional* region, time_t now, uint8_t* topname)
1250 {
1251 	struct dns_msg* msg;
1252 	struct val_neg_data* data;
1253 	uint8_t hashnc[NSEC3_SHA_LEN];
1254 	size_t nclen;
1255 	struct ub_packed_rrset_key* ce_rrset, *nc_rrset;
1256 	struct nsec3_cached_hash c;
1257 	uint8_t nc_b32[257];
1258 
1259 	/* for NSEC3 ; determine the closest encloser for which we
1260 	 * can find an exact match. Remember the hashed lower name,
1261 	 * since that is the one we need a closest match for.
1262 	 * If we find a match straight away, then it becomes NODATA.
1263 	 * Otherwise, NXDOMAIN or if OPTOUT, an insecure delegation.
1264 	 * Also check that parameters are the same on closest encloser
1265 	 * and on closest match.
1266 	 */
1267 	if(!zone->nsec3_hash)
1268 		return NULL; /* not nsec3 zone */
1269 
1270 	if(!(data=neg_find_nsec3_ce(zone, qname, qname_len, qlabs, buf,
1271 		hashnc, &nclen))) {
1272 		return NULL;
1273 	}
1274 
1275 	/* grab the ce rrset */
1276 	ce_rrset = grab_nsec(rrset_cache, data->name, data->len,
1277 		LDNS_RR_TYPE_NSEC3, zone->dclass, 0, region, 1,
1278 		LDNS_RR_TYPE_DS, now);
1279 	if(!ce_rrset)
1280 		return NULL;
1281 	if(!neg_params_ok(zone, ce_rrset))
1282 		return NULL;
1283 
1284 	if(nclen == 0) {
1285 		/* exact match, just check the type bits */
1286 		/* need: -SOA, -DS, +NS */
1287 		if(nsec3_has_type(ce_rrset, 0, LDNS_RR_TYPE_SOA) ||
1288 			nsec3_has_type(ce_rrset, 0, LDNS_RR_TYPE_DS) ||
1289 			!nsec3_has_type(ce_rrset, 0, LDNS_RR_TYPE_NS))
1290 			return NULL;
1291 		if(!(msg = dns_msg_create(qname, qname_len,
1292 			LDNS_RR_TYPE_DS, zone->dclass, region, 1)))
1293 			return NULL;
1294 		/* TTL reduced in grab_nsec */
1295 		if(!dns_msg_authadd(msg, region, ce_rrset, 0))
1296 			return NULL;
1297 		return msg;
1298 	}
1299 
1300 	/* optout is not allowed without knowing the trust-anchor in use,
1301 	 * otherwise the optout could spoof away that anchor */
1302 	if(!topname)
1303 		return NULL;
1304 
1305 	/* if there is no exact match, it must be in an optout span
1306 	 * (an existing DS implies an NSEC3 must exist) */
1307 	nc_rrset = neg_nsec3_getnc(zone, hashnc, nclen, rrset_cache,
1308 		region, now, nc_b32, sizeof(nc_b32));
1309 	if(!nc_rrset)
1310 		return NULL;
1311 	if(!neg_params_ok(zone, nc_rrset))
1312 		return NULL;
1313 	if(!nsec3_has_optout(nc_rrset, 0))
1314 		return NULL;
1315 	c.hash = hashnc;
1316 	c.hash_len = nclen;
1317 	c.b32 = nc_b32+1;
1318 	c.b32_len = (size_t)nc_b32[0];
1319 	if(nsec3_covers(zone->name, &c, nc_rrset, 0, buf)) {
1320 		/* nc_rrset covers the next closer name.
1321 		 * ce_rrset equals a closer encloser.
1322 		 * nc_rrset is optout.
1323 		 * No need to check wildcard for type DS */
1324 		/* capacity=3: ce + nc + soa(if needed) */
1325 		if(!(msg = dns_msg_create(qname, qname_len,
1326 			LDNS_RR_TYPE_DS, zone->dclass, region, 3)))
1327 			return NULL;
1328 		/* now=0 because TTL was reduced in grab_nsec */
1329 		if(!dns_msg_authadd(msg, region, ce_rrset, 0))
1330 			return NULL;
1331 		if(!dns_msg_authadd(msg, region, nc_rrset, 0))
1332 			return NULL;
1333 		return msg;
1334 	}
1335 	return NULL;
1336 }
1337 
1338 /**
1339  * Add SOA record for external responses.
1340  * @param rrset_cache: to look into.
1341  * @param now: current time.
1342  * @param region: where to perform the allocation
1343  * @param msg: current msg with NSEC.
1344  * @param zone: val_neg_zone if we have one.
1345  * @return false on lookup or alloc failure.
1346  */
1347 static int add_soa(struct rrset_cache* rrset_cache, time_t now,
1348 	struct regional* region, struct dns_msg* msg, struct val_neg_zone* zone)
1349 {
1350 	struct ub_packed_rrset_key* soa;
1351 	uint8_t* nm;
1352 	size_t nmlen;
1353 	uint16_t dclass;
1354 	if(zone) {
1355 		nm = zone->name;
1356 		nmlen = zone->len;
1357 		dclass = zone->dclass;
1358 	} else {
1359 		/* Assumes the signer is the zone SOA to add */
1360 		nm = reply_nsec_signer(msg->rep, &nmlen, &dclass);
1361 		if(!nm)
1362 			return 0;
1363 	}
1364 	soa = rrset_cache_lookup(rrset_cache, nm, nmlen, LDNS_RR_TYPE_SOA,
1365 		dclass, PACKED_RRSET_SOA_NEG, now, 0);
1366 	if(!soa)
1367 		return 0;
1368 	if(!dns_msg_authadd(msg, region, soa, now)) {
1369 		lock_rw_unlock(&soa->entry.lock);
1370 		return 0;
1371 	}
1372 	lock_rw_unlock(&soa->entry.lock);
1373 	return 1;
1374 }
1375 
1376 struct dns_msg*
1377 val_neg_getmsg(struct val_neg_cache* neg, struct query_info* qinfo,
1378 	struct regional* region, struct rrset_cache* rrset_cache,
1379 	sldns_buffer* buf, time_t now, int addsoa, uint8_t* topname,
1380 	struct config_file* cfg)
1381 {
1382 	struct dns_msg* msg;
1383 	struct ub_packed_rrset_key* nsec; /* qname matching/covering nsec */
1384 	struct ub_packed_rrset_key* wcrr; /* wildcard record or nsec */
1385 	uint8_t* nodata_wc = NULL;
1386 	uint8_t* ce = NULL;
1387 	size_t ce_len;
1388 	uint8_t wc_ce[LDNS_MAX_DOMAINLEN+3];
1389 	struct query_info wc_qinfo;
1390 	struct ub_packed_rrset_key* cache_wc;
1391 	struct packed_rrset_data* wcrr_data;
1392 	int rcode = LDNS_RCODE_NOERROR;
1393 	uint8_t* zname;
1394 	size_t zname_len;
1395 	int zname_labs;
1396 	struct val_neg_zone* zone;
1397 
1398 	/* only for DS queries when aggressive use of NSEC is disabled */
1399 	if(qinfo->qtype != LDNS_RR_TYPE_DS && !cfg->aggressive_nsec)
1400 		return NULL;
1401 	log_assert(!topname || dname_subdomain_c(qinfo->qname, topname));
1402 
1403 	/* Get best available NSEC for qname */
1404 	nsec = neg_find_nsec(neg, qinfo->qname, qinfo->qname_len, qinfo->qclass,
1405 		rrset_cache, now, region);
1406 
1407 	/* Matching NSEC, use to generate No Data answer. Not creating answers
1408 	 * yet for No Data proven using wildcard. */
1409 	if(nsec && nsec_proves_nodata(nsec, qinfo, &nodata_wc) && !nodata_wc) {
1410 		if(!(msg = dns_msg_create(qinfo->qname, qinfo->qname_len,
1411 			qinfo->qtype, qinfo->qclass, region, 2)))
1412 			return NULL;
1413 		if(!dns_msg_authadd(msg, region, nsec, 0))
1414 			return NULL;
1415 		if(addsoa && !add_soa(rrset_cache, now, region, msg, NULL))
1416 			return NULL;
1417 
1418 		lock_basic_lock(&neg->lock);
1419 		neg->num_neg_cache_noerror++;
1420 		lock_basic_unlock(&neg->lock);
1421 		return msg;
1422 	} else if(nsec && val_nsec_proves_name_error(nsec, qinfo->qname)) {
1423 		if(!(msg = dns_msg_create(qinfo->qname, qinfo->qname_len,
1424 			qinfo->qtype, qinfo->qclass, region, 3)))
1425 			return NULL;
1426 		if(!(ce = nsec_closest_encloser(qinfo->qname, nsec)))
1427 			return NULL;
1428 		dname_count_size_labels(ce, &ce_len);
1429 
1430 		/* No extra extra NSEC required if both nameerror qname and
1431 		 * nodata *.ce. are proven already. */
1432 		if(!nodata_wc || query_dname_compare(nodata_wc, ce) != 0) {
1433 			/* Qname proven non existing, get wildcard record for
1434 			 * QTYPE or NSEC covering or matching wildcard. */
1435 
1436 			/* Num labels in ce is always smaller than in qname,
1437 			 * therefore adding the wildcard label cannot overflow
1438 			 * buffer. */
1439 			wc_ce[0] = 1;
1440 			wc_ce[1] = (uint8_t)'*';
1441 			memmove(wc_ce+2, ce, ce_len);
1442 			wc_qinfo.qname = wc_ce;
1443 			wc_qinfo.qname_len = ce_len + 2;
1444 			wc_qinfo.qtype = qinfo->qtype;
1445 
1446 
1447 			if((cache_wc = rrset_cache_lookup(rrset_cache, wc_qinfo.qname,
1448 				wc_qinfo.qname_len, wc_qinfo.qtype,
1449 				qinfo->qclass, 0/*flags*/, now, 0/*read only*/))) {
1450 				/* Synthesize wildcard answer */
1451 				wcrr_data = (struct packed_rrset_data*)cache_wc->entry.data;
1452 				if(!(wcrr_data->security == sec_status_secure ||
1453 					(wcrr_data->security == sec_status_unchecked &&
1454 					wcrr_data->rrsig_count > 0))) {
1455 					lock_rw_unlock(&cache_wc->entry.lock);
1456 					return NULL;
1457 				}
1458 				if(!(wcrr = packed_rrset_copy_region(cache_wc,
1459 					region, now))) {
1460 					lock_rw_unlock(&cache_wc->entry.lock);
1461 					return NULL;
1462 				};
1463 				lock_rw_unlock(&cache_wc->entry.lock);
1464 				wcrr->rk.dname = qinfo->qname;
1465 				wcrr->rk.dname_len = qinfo->qname_len;
1466 				if(!dns_msg_ansadd(msg, region, wcrr, 0))
1467 					return NULL;
1468 				/* No SOA needed for wildcard synthesised
1469 				 * answer. */
1470 				addsoa = 0;
1471 			} else {
1472 				/* Get wildcard NSEC for possible non existence
1473 				 * proof */
1474 				if(!(wcrr = neg_find_nsec(neg, wc_qinfo.qname,
1475 					wc_qinfo.qname_len, qinfo->qclass,
1476 					rrset_cache, now, region)))
1477 					return NULL;
1478 
1479 				nodata_wc = NULL;
1480 				if(val_nsec_proves_name_error(wcrr, wc_ce))
1481 					rcode = LDNS_RCODE_NXDOMAIN;
1482 				else if(!nsec_proves_nodata(wcrr, &wc_qinfo,
1483 					&nodata_wc) || nodata_wc)
1484 					/* &nodata_wc shouldn't be set, wc_qinfo
1485 					 * already contains wildcard domain. */
1486 					/* NSEC doesn't prove anything for
1487 					 * wildcard. */
1488 					return NULL;
1489 				if(query_dname_compare(wcrr->rk.dname,
1490 					nsec->rk.dname) != 0)
1491 					if(!dns_msg_authadd(msg, region, wcrr, 0))
1492 						return NULL;
1493 			}
1494 		}
1495 
1496 		if(!dns_msg_authadd(msg, region, nsec, 0))
1497 			return NULL;
1498 		if(addsoa && !add_soa(rrset_cache, now, region, msg, NULL))
1499 			return NULL;
1500 
1501 		/* Increment statistic counters */
1502 		lock_basic_lock(&neg->lock);
1503 		if(rcode == LDNS_RCODE_NOERROR)
1504 			neg->num_neg_cache_noerror++;
1505 		else if(rcode == LDNS_RCODE_NXDOMAIN)
1506 			neg->num_neg_cache_nxdomain++;
1507 		lock_basic_unlock(&neg->lock);
1508 
1509 		FLAGS_SET_RCODE(msg->rep->flags, rcode);
1510 		return msg;
1511 	}
1512 
1513 	/* No aggressive use of NSEC3 for now, only proceed for DS types. */
1514 	if(qinfo->qtype != LDNS_RR_TYPE_DS){
1515 		return NULL;
1516 	}
1517 	/* check NSEC3 neg cache for type DS */
1518 	/* need to look one zone higher for DS type */
1519 	zname = qinfo->qname;
1520 	zname_len = qinfo->qname_len;
1521 	dname_remove_label(&zname, &zname_len);
1522 	zname_labs = dname_count_labels(zname);
1523 
1524 	/* lookup closest zone */
1525 	lock_basic_lock(&neg->lock);
1526 	zone = neg_closest_zone_parent(neg, zname, zname_len, zname_labs,
1527 		qinfo->qclass);
1528 	while(zone && !zone->in_use)
1529 		zone = zone->parent;
1530 	/* check that the zone is not too high up so that we do not pick data
1531 	 * out of a zone that is above the last-seen key (or trust-anchor). */
1532 	if(zone && topname) {
1533 		if(!dname_subdomain_c(zone->name, topname))
1534 			zone = NULL;
1535 	}
1536 	if(!zone) {
1537 		lock_basic_unlock(&neg->lock);
1538 		return NULL;
1539 	}
1540 
1541 	msg = neg_nsec3_proof_ds(zone, qinfo->qname, qinfo->qname_len,
1542 		zname_labs+1, buf, rrset_cache, region, now, topname);
1543 	if(msg && addsoa && !add_soa(rrset_cache, now, region, msg, zone)) {
1544 		lock_basic_unlock(&neg->lock);
1545 		return NULL;
1546 	}
1547 	lock_basic_unlock(&neg->lock);
1548 	return msg;
1549 }
1550