xref: /freebsd/contrib/unbound/services/outside_network.c (revision 2f834a0b41079f9be4dc33ff877d50a5fba869d4)
1 /*
2  * services/outside_network.c - implement sending of queries and wait answer.
3  *
4  * Copyright (c) 2007, NLnet Labs. All rights reserved.
5  *
6  * This software is open source.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  *
12  * Redistributions of source code must retain the above copyright notice,
13  * this list of conditions and the following disclaimer.
14  *
15  * Redistributions in binary form must reproduce the above copyright notice,
16  * this list of conditions and the following disclaimer in the documentation
17  * and/or other materials provided with the distribution.
18  *
19  * Neither the name of the NLNET LABS nor the names of its contributors may
20  * be used to endorse or promote products derived from this software without
21  * specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
26  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
27  * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
28  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
29  * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
30  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
31  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
32  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34  */
35 
36 /**
37  * \file
38  *
39  * This file has functions to send queries to authoritative servers and
40  * wait for the pending answer events.
41  */
42 #include "config.h"
43 #include <ctype.h>
44 #ifdef HAVE_SYS_TYPES_H
45 #  include <sys/types.h>
46 #endif
47 #include <sys/time.h>
48 #include "services/outside_network.h"
49 #include "services/listen_dnsport.h"
50 #include "services/cache/infra.h"
51 #include "util/data/msgparse.h"
52 #include "util/data/msgreply.h"
53 #include "util/data/msgencode.h"
54 #include "util/data/dname.h"
55 #include "util/netevent.h"
56 #include "util/log.h"
57 #include "util/net_help.h"
58 #include "util/random.h"
59 #include "util/fptr_wlist.h"
60 #include "ldns/sbuffer.h"
61 #ifdef HAVE_OPENSSL_SSL_H
62 #include <openssl/ssl.h>
63 #endif
64 
65 #ifdef HAVE_NETDB_H
66 #include <netdb.h>
67 #endif
68 #include <fcntl.h>
69 
70 /** number of times to retry making a random ID that is unique. */
71 #define MAX_ID_RETRY 1000
72 /** number of times to retry finding interface, port that can be opened. */
73 #define MAX_PORT_RETRY 10000
74 /** number of retries on outgoing UDP queries */
75 #define OUTBOUND_UDP_RETRY 1
76 
77 /** initiate TCP transaction for serviced query */
78 static void serviced_tcp_initiate(struct outside_network* outnet,
79 	struct serviced_query* sq, sldns_buffer* buff);
80 /** with a fd available, randomize and send UDP */
81 static int randomize_and_send_udp(struct outside_network* outnet,
82 	struct pending* pend, sldns_buffer* packet, int timeout);
83 
84 int
85 pending_cmp(const void* key1, const void* key2)
86 {
87 	struct pending *p1 = (struct pending*)key1;
88 	struct pending *p2 = (struct pending*)key2;
89 	if(p1->id < p2->id)
90 		return -1;
91 	if(p1->id > p2->id)
92 		return 1;
93 	log_assert(p1->id == p2->id);
94 	return sockaddr_cmp(&p1->addr, p1->addrlen, &p2->addr, p2->addrlen);
95 }
96 
97 int
98 serviced_cmp(const void* key1, const void* key2)
99 {
100 	struct serviced_query* q1 = (struct serviced_query*)key1;
101 	struct serviced_query* q2 = (struct serviced_query*)key2;
102 	int r;
103 	if(q1->qbuflen < q2->qbuflen)
104 		return -1;
105 	if(q1->qbuflen > q2->qbuflen)
106 		return 1;
107 	log_assert(q1->qbuflen == q2->qbuflen);
108 	log_assert(q1->qbuflen >= 15 /* 10 header, root, type, class */);
109 	/* alternate casing of qname is still the same query */
110 	if((r = memcmp(q1->qbuf, q2->qbuf, 10)) != 0)
111 		return r;
112 	if((r = memcmp(q1->qbuf+q1->qbuflen-4, q2->qbuf+q2->qbuflen-4, 4)) != 0)
113 		return r;
114 	if(q1->dnssec != q2->dnssec) {
115 		if(q1->dnssec < q2->dnssec)
116 			return -1;
117 		return 1;
118 	}
119 	if((r = query_dname_compare(q1->qbuf+10, q2->qbuf+10)) != 0)
120 		return r;
121 	return sockaddr_cmp(&q1->addr, q1->addrlen, &q2->addr, q2->addrlen);
122 }
123 
124 /** delete waiting_tcp entry. Does not unlink from waiting list.
125  * @param w: to delete.
126  */
127 static void
128 waiting_tcp_delete(struct waiting_tcp* w)
129 {
130 	if(!w) return;
131 	if(w->timer)
132 		comm_timer_delete(w->timer);
133 	free(w);
134 }
135 
136 /**
137  * Pick random outgoing-interface of that family, and bind it.
138  * port set to 0 so OS picks a port number for us.
139  * if it is the ANY address, do not bind.
140  * @param w: tcp structure with destination address.
141  * @param s: socket fd.
142  * @return false on error, socket closed.
143  */
144 static int
145 pick_outgoing_tcp(struct waiting_tcp* w, int s)
146 {
147 	struct port_if* pi = NULL;
148 	int num;
149 #ifdef INET6
150 	if(addr_is_ip6(&w->addr, w->addrlen))
151 		num = w->outnet->num_ip6;
152 	else
153 #endif
154 		num = w->outnet->num_ip4;
155 	if(num == 0) {
156 		log_err("no TCP outgoing interfaces of family");
157 		log_addr(VERB_OPS, "for addr", &w->addr, w->addrlen);
158 #ifndef USE_WINSOCK
159 		close(s);
160 #else
161 		closesocket(s);
162 #endif
163 		return 0;
164 	}
165 #ifdef INET6
166 	if(addr_is_ip6(&w->addr, w->addrlen))
167 		pi = &w->outnet->ip6_ifs[ub_random_max(w->outnet->rnd, num)];
168 	else
169 #endif
170 		pi = &w->outnet->ip4_ifs[ub_random_max(w->outnet->rnd, num)];
171 	log_assert(pi);
172 	if(addr_is_any(&pi->addr, pi->addrlen)) {
173 		/* binding to the ANY interface is for listening sockets */
174 		return 1;
175 	}
176 	/* set port to 0 */
177 	if(addr_is_ip6(&pi->addr, pi->addrlen))
178 		((struct sockaddr_in6*)&pi->addr)->sin6_port = 0;
179 	else	((struct sockaddr_in*)&pi->addr)->sin_port = 0;
180 	if(bind(s, (struct sockaddr*)&pi->addr, pi->addrlen) != 0) {
181 #ifndef USE_WINSOCK
182 		log_err("outgoing tcp: bind: %s", strerror(errno));
183 		close(s);
184 #else
185 		log_err("outgoing tcp: bind: %s",
186 			wsa_strerror(WSAGetLastError()));
187 		closesocket(s);
188 #endif
189 		return 0;
190 	}
191 	log_addr(VERB_ALGO, "tcp bound to src", &pi->addr, pi->addrlen);
192 	return 1;
193 }
194 
195 /** use next free buffer to service a tcp query */
196 static int
197 outnet_tcp_take_into_use(struct waiting_tcp* w, uint8_t* pkt, size_t pkt_len)
198 {
199 	struct pending_tcp* pend = w->outnet->tcp_free;
200 	int s;
201 	log_assert(pend);
202 	log_assert(pkt);
203 	log_assert(w->addrlen > 0);
204 	/* open socket */
205 #ifdef INET6
206 	if(addr_is_ip6(&w->addr, w->addrlen))
207 		s = socket(PF_INET6, SOCK_STREAM, IPPROTO_TCP);
208 	else
209 #endif
210 		s = socket(PF_INET, SOCK_STREAM, IPPROTO_TCP);
211 	if(s == -1) {
212 #ifndef USE_WINSOCK
213 		log_err("outgoing tcp: socket: %s", strerror(errno));
214 #else
215 		log_err("outgoing tcp: socket: %s",
216 			wsa_strerror(WSAGetLastError()));
217 #endif
218 		log_addr(0, "failed address", &w->addr, w->addrlen);
219 		return 0;
220 	}
221 	if(!pick_outgoing_tcp(w, s))
222 		return 0;
223 
224 	fd_set_nonblock(s);
225 	if(connect(s, (struct sockaddr*)&w->addr, w->addrlen) == -1) {
226 #ifndef USE_WINSOCK
227 #ifdef EINPROGRESS
228 		if(errno != EINPROGRESS) {
229 #else
230 		if(1) {
231 #endif
232 			if(tcp_connect_errno_needs_log(
233 				(struct sockaddr*)&w->addr, w->addrlen))
234 				log_err("outgoing tcp: connect: %s",
235 					strerror(errno));
236 			close(s);
237 #else /* USE_WINSOCK */
238 		if(WSAGetLastError() != WSAEINPROGRESS &&
239 			WSAGetLastError() != WSAEWOULDBLOCK) {
240 			closesocket(s);
241 #endif
242 			log_addr(0, "failed address", &w->addr, w->addrlen);
243 			return 0;
244 		}
245 	}
246 	if(w->outnet->sslctx && w->ssl_upstream) {
247 		pend->c->ssl = outgoing_ssl_fd(w->outnet->sslctx, s);
248 		if(!pend->c->ssl) {
249 			pend->c->fd = s;
250 			comm_point_close(pend->c);
251 			return 0;
252 		}
253 #ifdef USE_WINSOCK
254 		comm_point_tcp_win_bio_cb(pend->c, pend->c->ssl);
255 #endif
256 		pend->c->ssl_shake_state = comm_ssl_shake_write;
257 	}
258 	w->pkt = NULL;
259 	w->next_waiting = (void*)pend;
260 	pend->id = LDNS_ID_WIRE(pkt);
261 	w->outnet->tcp_free = pend->next_free;
262 	pend->next_free = NULL;
263 	pend->query = w;
264 	pend->c->repinfo.addrlen = w->addrlen;
265 	memcpy(&pend->c->repinfo.addr, &w->addr, w->addrlen);
266 	sldns_buffer_clear(pend->c->buffer);
267 	sldns_buffer_write(pend->c->buffer, pkt, pkt_len);
268 	sldns_buffer_flip(pend->c->buffer);
269 	pend->c->tcp_is_reading = 0;
270 	pend->c->tcp_byte_count = 0;
271 	comm_point_start_listening(pend->c, s, -1);
272 	return 1;
273 }
274 
275 /** see if buffers can be used to service TCP queries */
276 static void
277 use_free_buffer(struct outside_network* outnet)
278 {
279 	struct waiting_tcp* w;
280 	while(outnet->tcp_free && outnet->tcp_wait_first
281 		&& !outnet->want_to_quit) {
282 		w = outnet->tcp_wait_first;
283 		outnet->tcp_wait_first = w->next_waiting;
284 		if(outnet->tcp_wait_last == w)
285 			outnet->tcp_wait_last = NULL;
286 		if(!outnet_tcp_take_into_use(w, w->pkt, w->pkt_len)) {
287 			comm_point_callback_t* cb = w->cb;
288 			void* cb_arg = w->cb_arg;
289 			waiting_tcp_delete(w);
290 			fptr_ok(fptr_whitelist_pending_tcp(cb));
291 			(void)(*cb)(NULL, cb_arg, NETEVENT_CLOSED, NULL);
292 		}
293 	}
294 }
295 
296 /** decomission a tcp buffer, closes commpoint and frees waiting_tcp entry */
297 static void
298 decomission_pending_tcp(struct outside_network* outnet,
299 	struct pending_tcp* pend)
300 {
301 	if(pend->c->ssl) {
302 #ifdef HAVE_SSL
303 		SSL_shutdown(pend->c->ssl);
304 		SSL_free(pend->c->ssl);
305 		pend->c->ssl = NULL;
306 #endif
307 	}
308 	comm_point_close(pend->c);
309 	pend->next_free = outnet->tcp_free;
310 	outnet->tcp_free = pend;
311 	waiting_tcp_delete(pend->query);
312 	pend->query = NULL;
313 	use_free_buffer(outnet);
314 }
315 
316 int
317 outnet_tcp_cb(struct comm_point* c, void* arg, int error,
318 	struct comm_reply *reply_info)
319 {
320 	struct pending_tcp* pend = (struct pending_tcp*)arg;
321 	struct outside_network* outnet = pend->query->outnet;
322 	verbose(VERB_ALGO, "outnettcp cb");
323 	if(error != NETEVENT_NOERROR) {
324 		verbose(VERB_QUERY, "outnettcp got tcp error %d", error);
325 		/* pass error below and exit */
326 	} else {
327 		/* check ID */
328 		if(sldns_buffer_limit(c->buffer) < sizeof(uint16_t) ||
329 			LDNS_ID_WIRE(sldns_buffer_begin(c->buffer))!=pend->id) {
330 			log_addr(VERB_QUERY,
331 				"outnettcp: bad ID in reply, from:",
332 				&pend->query->addr, pend->query->addrlen);
333 			error = NETEVENT_CLOSED;
334 		}
335 	}
336 	fptr_ok(fptr_whitelist_pending_tcp(pend->query->cb));
337 	(void)(*pend->query->cb)(c, pend->query->cb_arg, error, reply_info);
338 	decomission_pending_tcp(outnet, pend);
339 	return 0;
340 }
341 
342 /** lower use count on pc, see if it can be closed */
343 static void
344 portcomm_loweruse(struct outside_network* outnet, struct port_comm* pc)
345 {
346 	struct port_if* pif;
347 	pc->num_outstanding--;
348 	if(pc->num_outstanding > 0) {
349 		return;
350 	}
351 	/* close it and replace in unused list */
352 	verbose(VERB_ALGO, "close of port %d", pc->number);
353 	comm_point_close(pc->cp);
354 	pif = pc->pif;
355 	log_assert(pif->inuse > 0);
356 	pif->avail_ports[pif->avail_total - pif->inuse] = pc->number;
357 	pif->inuse--;
358 	pif->out[pc->index] = pif->out[pif->inuse];
359 	pif->out[pc->index]->index = pc->index;
360 	pc->next = outnet->unused_fds;
361 	outnet->unused_fds = pc;
362 }
363 
364 /** try to send waiting UDP queries */
365 static void
366 outnet_send_wait_udp(struct outside_network* outnet)
367 {
368 	struct pending* pend;
369 	/* process waiting queries */
370 	while(outnet->udp_wait_first && outnet->unused_fds
371 		&& !outnet->want_to_quit) {
372 		pend = outnet->udp_wait_first;
373 		outnet->udp_wait_first = pend->next_waiting;
374 		if(!pend->next_waiting) outnet->udp_wait_last = NULL;
375 		sldns_buffer_clear(outnet->udp_buff);
376 		sldns_buffer_write(outnet->udp_buff, pend->pkt, pend->pkt_len);
377 		sldns_buffer_flip(outnet->udp_buff);
378 		free(pend->pkt); /* freeing now makes get_mem correct */
379 		pend->pkt = NULL;
380 		pend->pkt_len = 0;
381 		if(!randomize_and_send_udp(outnet, pend, outnet->udp_buff,
382 			pend->timeout)) {
383 			/* callback error on pending */
384 			if(pend->cb) {
385 				fptr_ok(fptr_whitelist_pending_udp(pend->cb));
386 				(void)(*pend->cb)(outnet->unused_fds->cp, pend->cb_arg,
387 					NETEVENT_CLOSED, NULL);
388 			}
389 			pending_delete(outnet, pend);
390 		}
391 	}
392 }
393 
394 int
395 outnet_udp_cb(struct comm_point* c, void* arg, int error,
396 	struct comm_reply *reply_info)
397 {
398 	struct outside_network* outnet = (struct outside_network*)arg;
399 	struct pending key;
400 	struct pending* p;
401 	verbose(VERB_ALGO, "answer cb");
402 
403 	if(error != NETEVENT_NOERROR) {
404 		verbose(VERB_QUERY, "outnetudp got udp error %d", error);
405 		return 0;
406 	}
407 	if(sldns_buffer_limit(c->buffer) < LDNS_HEADER_SIZE) {
408 		verbose(VERB_QUERY, "outnetudp udp too short");
409 		return 0;
410 	}
411 	log_assert(reply_info);
412 
413 	/* setup lookup key */
414 	key.id = (unsigned)LDNS_ID_WIRE(sldns_buffer_begin(c->buffer));
415 	memcpy(&key.addr, &reply_info->addr, reply_info->addrlen);
416 	key.addrlen = reply_info->addrlen;
417 	verbose(VERB_ALGO, "Incoming reply id = %4.4x", key.id);
418 	log_addr(VERB_ALGO, "Incoming reply addr =",
419 		&reply_info->addr, reply_info->addrlen);
420 
421 	/* find it, see if this thing is a valid query response */
422 	verbose(VERB_ALGO, "lookup size is %d entries", (int)outnet->pending->count);
423 	p = (struct pending*)rbtree_search(outnet->pending, &key);
424 	if(!p) {
425 		verbose(VERB_QUERY, "received unwanted or unsolicited udp reply dropped.");
426 		log_buf(VERB_ALGO, "dropped message", c->buffer);
427 		outnet->unwanted_replies++;
428 		if(outnet->unwanted_threshold && ++outnet->unwanted_total
429 			>= outnet->unwanted_threshold) {
430 			log_warn("unwanted reply total reached threshold (%u)"
431 				" you may be under attack."
432 				" defensive action: clearing the cache",
433 				(unsigned)outnet->unwanted_threshold);
434 			fptr_ok(fptr_whitelist_alloc_cleanup(
435 				outnet->unwanted_action));
436 			(*outnet->unwanted_action)(outnet->unwanted_param);
437 			outnet->unwanted_total = 0;
438 		}
439 		return 0;
440 	}
441 
442 	verbose(VERB_ALGO, "received udp reply.");
443 	log_buf(VERB_ALGO, "udp message", c->buffer);
444 	if(p->pc->cp != c) {
445 		verbose(VERB_QUERY, "received reply id,addr on wrong port. "
446 			"dropped.");
447 		outnet->unwanted_replies++;
448 		if(outnet->unwanted_threshold && ++outnet->unwanted_total
449 			>= outnet->unwanted_threshold) {
450 			log_warn("unwanted reply total reached threshold (%u)"
451 				" you may be under attack."
452 				" defensive action: clearing the cache",
453 				(unsigned)outnet->unwanted_threshold);
454 			fptr_ok(fptr_whitelist_alloc_cleanup(
455 				outnet->unwanted_action));
456 			(*outnet->unwanted_action)(outnet->unwanted_param);
457 			outnet->unwanted_total = 0;
458 		}
459 		return 0;
460 	}
461 	comm_timer_disable(p->timer);
462 	verbose(VERB_ALGO, "outnet handle udp reply");
463 	/* delete from tree first in case callback creates a retry */
464 	(void)rbtree_delete(outnet->pending, p->node.key);
465 	if(p->cb) {
466 		fptr_ok(fptr_whitelist_pending_udp(p->cb));
467 		(void)(*p->cb)(p->pc->cp, p->cb_arg, NETEVENT_NOERROR, reply_info);
468 	}
469 	portcomm_loweruse(outnet, p->pc);
470 	pending_delete(NULL, p);
471 	outnet_send_wait_udp(outnet);
472 	return 0;
473 }
474 
475 /** calculate number of ip4 and ip6 interfaces*/
476 static void
477 calc_num46(char** ifs, int num_ifs, int do_ip4, int do_ip6,
478 	int* num_ip4, int* num_ip6)
479 {
480 	int i;
481 	*num_ip4 = 0;
482 	*num_ip6 = 0;
483 	if(num_ifs <= 0) {
484 		if(do_ip4)
485 			*num_ip4 = 1;
486 		if(do_ip6)
487 			*num_ip6 = 1;
488 		return;
489 	}
490 	for(i=0; i<num_ifs; i++)
491 	{
492 		if(str_is_ip6(ifs[i])) {
493 			if(do_ip6)
494 				(*num_ip6)++;
495 		} else {
496 			if(do_ip4)
497 				(*num_ip4)++;
498 		}
499 	}
500 
501 }
502 
503 void
504 pending_udp_timer_delay_cb(void* arg)
505 {
506 	struct pending* p = (struct pending*)arg;
507 	struct outside_network* outnet = p->outnet;
508 	verbose(VERB_ALGO, "timeout udp with delay");
509 	portcomm_loweruse(outnet, p->pc);
510 	pending_delete(outnet, p);
511 	outnet_send_wait_udp(outnet);
512 }
513 
514 void
515 pending_udp_timer_cb(void *arg)
516 {
517 	struct pending* p = (struct pending*)arg;
518 	struct outside_network* outnet = p->outnet;
519 	/* it timed out */
520 	verbose(VERB_ALGO, "timeout udp");
521 	if(p->cb) {
522 		fptr_ok(fptr_whitelist_pending_udp(p->cb));
523 		(void)(*p->cb)(p->pc->cp, p->cb_arg, NETEVENT_TIMEOUT, NULL);
524 	}
525 	/* if delayclose, keep port open for a longer time.
526 	 * But if the udpwaitlist exists, then we are struggling to
527 	 * keep up with demand for sockets, so do not wait, but service
528 	 * the customer (customer service more important than portICMPs) */
529 	if(outnet->delayclose && !outnet->udp_wait_first) {
530 		p->cb = NULL;
531 		p->timer->callback = &pending_udp_timer_delay_cb;
532 		comm_timer_set(p->timer, &outnet->delay_tv);
533 		return;
534 	}
535 	portcomm_loweruse(outnet, p->pc);
536 	pending_delete(outnet, p);
537 	outnet_send_wait_udp(outnet);
538 }
539 
540 /** create pending_tcp buffers */
541 static int
542 create_pending_tcp(struct outside_network* outnet, size_t bufsize)
543 {
544 	size_t i;
545 	if(outnet->num_tcp == 0)
546 		return 1; /* no tcp needed, nothing to do */
547 	if(!(outnet->tcp_conns = (struct pending_tcp **)calloc(
548 			outnet->num_tcp, sizeof(struct pending_tcp*))))
549 		return 0;
550 	for(i=0; i<outnet->num_tcp; i++) {
551 		if(!(outnet->tcp_conns[i] = (struct pending_tcp*)calloc(1,
552 			sizeof(struct pending_tcp))))
553 			return 0;
554 		outnet->tcp_conns[i]->next_free = outnet->tcp_free;
555 		outnet->tcp_free = outnet->tcp_conns[i];
556 		outnet->tcp_conns[i]->c = comm_point_create_tcp_out(
557 			outnet->base, bufsize, outnet_tcp_cb,
558 			outnet->tcp_conns[i]);
559 		if(!outnet->tcp_conns[i]->c)
560 			return 0;
561 	}
562 	return 1;
563 }
564 
565 /** setup an outgoing interface, ready address */
566 static int setup_if(struct port_if* pif, const char* addrstr,
567 	int* avail, int numavail, size_t numfd)
568 {
569 	pif->avail_total = numavail;
570 	pif->avail_ports = (int*)memdup(avail, (size_t)numavail*sizeof(int));
571 	if(!pif->avail_ports)
572 		return 0;
573 	if(!ipstrtoaddr(addrstr, UNBOUND_DNS_PORT, &pif->addr, &pif->addrlen))
574 		return 0;
575 	pif->maxout = (int)numfd;
576 	pif->inuse = 0;
577 	pif->out = (struct port_comm**)calloc(numfd,
578 		sizeof(struct port_comm*));
579 	if(!pif->out)
580 		return 0;
581 	return 1;
582 }
583 
584 struct outside_network*
585 outside_network_create(struct comm_base *base, size_t bufsize,
586 	size_t num_ports, char** ifs, int num_ifs, int do_ip4,
587 	int do_ip6, size_t num_tcp, struct infra_cache* infra,
588 	struct ub_randstate* rnd, int use_caps_for_id, int* availports,
589 	int numavailports, size_t unwanted_threshold,
590 	void (*unwanted_action)(void*), void* unwanted_param, int do_udp,
591 	void* sslctx, int delayclose)
592 {
593 	struct outside_network* outnet = (struct outside_network*)
594 		calloc(1, sizeof(struct outside_network));
595 	size_t k;
596 	if(!outnet) {
597 		log_err("malloc failed");
598 		return NULL;
599 	}
600 	comm_base_timept(base, &outnet->now_secs, &outnet->now_tv);
601 	outnet->base = base;
602 	outnet->num_tcp = num_tcp;
603 	outnet->infra = infra;
604 	outnet->rnd = rnd;
605 	outnet->sslctx = sslctx;
606 	outnet->svcd_overhead = 0;
607 	outnet->want_to_quit = 0;
608 	outnet->unwanted_threshold = unwanted_threshold;
609 	outnet->unwanted_action = unwanted_action;
610 	outnet->unwanted_param = unwanted_param;
611 	outnet->use_caps_for_id = use_caps_for_id;
612 	outnet->do_udp = do_udp;
613 #ifndef S_SPLINT_S
614 	if(delayclose) {
615 		outnet->delayclose = 1;
616 		outnet->delay_tv.tv_sec = delayclose/1000;
617 		outnet->delay_tv.tv_usec = (delayclose%1000)*1000;
618 	}
619 #endif
620 	if(numavailports == 0) {
621 		log_err("no outgoing ports available");
622 		outside_network_delete(outnet);
623 		return NULL;
624 	}
625 #ifndef INET6
626 	do_ip6 = 0;
627 #endif
628 	calc_num46(ifs, num_ifs, do_ip4, do_ip6,
629 		&outnet->num_ip4, &outnet->num_ip6);
630 	if(outnet->num_ip4 != 0) {
631 		if(!(outnet->ip4_ifs = (struct port_if*)calloc(
632 			(size_t)outnet->num_ip4, sizeof(struct port_if)))) {
633 			log_err("malloc failed");
634 			outside_network_delete(outnet);
635 			return NULL;
636 		}
637 	}
638 	if(outnet->num_ip6 != 0) {
639 		if(!(outnet->ip6_ifs = (struct port_if*)calloc(
640 			(size_t)outnet->num_ip6, sizeof(struct port_if)))) {
641 			log_err("malloc failed");
642 			outside_network_delete(outnet);
643 			return NULL;
644 		}
645 	}
646 	if(	!(outnet->udp_buff = sldns_buffer_new(bufsize)) ||
647 		!(outnet->pending = rbtree_create(pending_cmp)) ||
648 		!(outnet->serviced = rbtree_create(serviced_cmp)) ||
649 		!create_pending_tcp(outnet, bufsize)) {
650 		log_err("malloc failed");
651 		outside_network_delete(outnet);
652 		return NULL;
653 	}
654 
655 	/* allocate commpoints */
656 	for(k=0; k<num_ports; k++) {
657 		struct port_comm* pc;
658 		pc = (struct port_comm*)calloc(1, sizeof(*pc));
659 		if(!pc) {
660 			log_err("malloc failed");
661 			outside_network_delete(outnet);
662 			return NULL;
663 		}
664 		pc->cp = comm_point_create_udp(outnet->base, -1,
665 			outnet->udp_buff, outnet_udp_cb, outnet);
666 		if(!pc->cp) {
667 			log_err("malloc failed");
668 			free(pc);
669 			outside_network_delete(outnet);
670 			return NULL;
671 		}
672 		pc->next = outnet->unused_fds;
673 		outnet->unused_fds = pc;
674 	}
675 
676 	/* allocate interfaces */
677 	if(num_ifs == 0) {
678 		if(do_ip4 && !setup_if(&outnet->ip4_ifs[0], "0.0.0.0",
679 			availports, numavailports, num_ports)) {
680 			log_err("malloc failed");
681 			outside_network_delete(outnet);
682 			return NULL;
683 		}
684 		if(do_ip6 && !setup_if(&outnet->ip6_ifs[0], "::",
685 			availports, numavailports, num_ports)) {
686 			log_err("malloc failed");
687 			outside_network_delete(outnet);
688 			return NULL;
689 		}
690 	} else {
691 		size_t done_4 = 0, done_6 = 0;
692 		int i;
693 		for(i=0; i<num_ifs; i++) {
694 			if(str_is_ip6(ifs[i]) && do_ip6) {
695 				if(!setup_if(&outnet->ip6_ifs[done_6], ifs[i],
696 					availports, numavailports, num_ports)){
697 					log_err("malloc failed");
698 					outside_network_delete(outnet);
699 					return NULL;
700 				}
701 				done_6++;
702 			}
703 			if(!str_is_ip6(ifs[i]) && do_ip4) {
704 				if(!setup_if(&outnet->ip4_ifs[done_4], ifs[i],
705 					availports, numavailports, num_ports)){
706 					log_err("malloc failed");
707 					outside_network_delete(outnet);
708 					return NULL;
709 				}
710 				done_4++;
711 			}
712 		}
713 	}
714 	return outnet;
715 }
716 
717 /** helper pending delete */
718 static void
719 pending_node_del(rbnode_t* node, void* arg)
720 {
721 	struct pending* pend = (struct pending*)node;
722 	struct outside_network* outnet = (struct outside_network*)arg;
723 	pending_delete(outnet, pend);
724 }
725 
726 /** helper serviced delete */
727 static void
728 serviced_node_del(rbnode_t* node, void* ATTR_UNUSED(arg))
729 {
730 	struct serviced_query* sq = (struct serviced_query*)node;
731 	struct service_callback* p = sq->cblist, *np;
732 	free(sq->qbuf);
733 	free(sq->zone);
734 	while(p) {
735 		np = p->next;
736 		free(p);
737 		p = np;
738 	}
739 	free(sq);
740 }
741 
742 void
743 outside_network_quit_prepare(struct outside_network* outnet)
744 {
745 	if(!outnet)
746 		return;
747 	/* prevent queued items from being sent */
748 	outnet->want_to_quit = 1;
749 }
750 
751 void
752 outside_network_delete(struct outside_network* outnet)
753 {
754 	if(!outnet)
755 		return;
756 	outnet->want_to_quit = 1;
757 	/* check every element, since we can be called on malloc error */
758 	if(outnet->pending) {
759 		/* free pending elements, but do no unlink from tree. */
760 		traverse_postorder(outnet->pending, pending_node_del, NULL);
761 		free(outnet->pending);
762 	}
763 	if(outnet->serviced) {
764 		traverse_postorder(outnet->serviced, serviced_node_del, NULL);
765 		free(outnet->serviced);
766 	}
767 	if(outnet->udp_buff)
768 		sldns_buffer_free(outnet->udp_buff);
769 	if(outnet->unused_fds) {
770 		struct port_comm* p = outnet->unused_fds, *np;
771 		while(p) {
772 			np = p->next;
773 			comm_point_delete(p->cp);
774 			free(p);
775 			p = np;
776 		}
777 		outnet->unused_fds = NULL;
778 	}
779 	if(outnet->ip4_ifs) {
780 		int i, k;
781 		for(i=0; i<outnet->num_ip4; i++) {
782 			for(k=0; k<outnet->ip4_ifs[i].inuse; k++) {
783 				struct port_comm* pc = outnet->ip4_ifs[i].
784 					out[k];
785 				comm_point_delete(pc->cp);
786 				free(pc);
787 			}
788 			free(outnet->ip4_ifs[i].avail_ports);
789 			free(outnet->ip4_ifs[i].out);
790 		}
791 		free(outnet->ip4_ifs);
792 	}
793 	if(outnet->ip6_ifs) {
794 		int i, k;
795 		for(i=0; i<outnet->num_ip6; i++) {
796 			for(k=0; k<outnet->ip6_ifs[i].inuse; k++) {
797 				struct port_comm* pc = outnet->ip6_ifs[i].
798 					out[k];
799 				comm_point_delete(pc->cp);
800 				free(pc);
801 			}
802 			free(outnet->ip6_ifs[i].avail_ports);
803 			free(outnet->ip6_ifs[i].out);
804 		}
805 		free(outnet->ip6_ifs);
806 	}
807 	if(outnet->tcp_conns) {
808 		size_t i;
809 		for(i=0; i<outnet->num_tcp; i++)
810 			if(outnet->tcp_conns[i]) {
811 				comm_point_delete(outnet->tcp_conns[i]->c);
812 				waiting_tcp_delete(outnet->tcp_conns[i]->query);
813 				free(outnet->tcp_conns[i]);
814 			}
815 		free(outnet->tcp_conns);
816 	}
817 	if(outnet->tcp_wait_first) {
818 		struct waiting_tcp* p = outnet->tcp_wait_first, *np;
819 		while(p) {
820 			np = p->next_waiting;
821 			waiting_tcp_delete(p);
822 			p = np;
823 		}
824 	}
825 	if(outnet->udp_wait_first) {
826 		struct pending* p = outnet->udp_wait_first, *np;
827 		while(p) {
828 			np = p->next_waiting;
829 			pending_delete(NULL, p);
830 			p = np;
831 		}
832 	}
833 	free(outnet);
834 }
835 
836 void
837 pending_delete(struct outside_network* outnet, struct pending* p)
838 {
839 	if(!p)
840 		return;
841 	if(outnet && outnet->udp_wait_first &&
842 		(p->next_waiting || p == outnet->udp_wait_last) ) {
843 		/* delete from waiting list, if it is in the waiting list */
844 		struct pending* prev = NULL, *x = outnet->udp_wait_first;
845 		while(x && x != p) {
846 			prev = x;
847 			x = x->next_waiting;
848 		}
849 		if(x) {
850 			log_assert(x == p);
851 			if(prev)
852 				prev->next_waiting = p->next_waiting;
853 			else	outnet->udp_wait_first = p->next_waiting;
854 			if(outnet->udp_wait_last == p)
855 				outnet->udp_wait_last = prev;
856 		}
857 	}
858 	if(outnet) {
859 		(void)rbtree_delete(outnet->pending, p->node.key);
860 	}
861 	if(p->timer)
862 		comm_timer_delete(p->timer);
863 	free(p->pkt);
864 	free(p);
865 }
866 
867 /**
868  * Try to open a UDP socket for outgoing communication.
869  * Sets sockets options as needed.
870  * @param addr: socket address.
871  * @param addrlen: length of address.
872  * @param port: port override for addr.
873  * @param inuse: if -1 is returned, this bool means the port was in use.
874  * @return fd or -1
875  */
876 static int
877 udp_sockport(struct sockaddr_storage* addr, socklen_t addrlen, int port,
878 	int* inuse)
879 {
880 	int fd, noproto;
881 	if(addr_is_ip6(addr, addrlen)) {
882 		struct sockaddr_in6* sa = (struct sockaddr_in6*)addr;
883 		sa->sin6_port = (in_port_t)htons((uint16_t)port);
884 		fd = create_udp_sock(AF_INET6, SOCK_DGRAM,
885 			(struct sockaddr*)addr, addrlen, 1, inuse, &noproto,
886 			0, 0, 0, NULL);
887 	} else {
888 		struct sockaddr_in* sa = (struct sockaddr_in*)addr;
889 		sa->sin_port = (in_port_t)htons((uint16_t)port);
890 		fd = create_udp_sock(AF_INET, SOCK_DGRAM,
891 			(struct sockaddr*)addr, addrlen, 1, inuse, &noproto,
892 			0, 0, 0, NULL);
893 	}
894 	return fd;
895 }
896 
897 /** Select random ID */
898 static int
899 select_id(struct outside_network* outnet, struct pending* pend,
900 	sldns_buffer* packet)
901 {
902 	int id_tries = 0;
903 	pend->id = ((unsigned)ub_random(outnet->rnd)>>8) & 0xffff;
904 	LDNS_ID_SET(sldns_buffer_begin(packet), pend->id);
905 
906 	/* insert in tree */
907 	pend->node.key = pend;
908 	while(!rbtree_insert(outnet->pending, &pend->node)) {
909 		/* change ID to avoid collision */
910 		pend->id = ((unsigned)ub_random(outnet->rnd)>>8) & 0xffff;
911 		LDNS_ID_SET(sldns_buffer_begin(packet), pend->id);
912 		id_tries++;
913 		if(id_tries == MAX_ID_RETRY) {
914 			pend->id=99999; /* non existant ID */
915 			log_err("failed to generate unique ID, drop msg");
916 			return 0;
917 		}
918 	}
919 	verbose(VERB_ALGO, "inserted new pending reply id=%4.4x", pend->id);
920 	return 1;
921 }
922 
923 /** Select random interface and port */
924 static int
925 select_ifport(struct outside_network* outnet, struct pending* pend,
926 	int num_if, struct port_if* ifs)
927 {
928 	int my_if, my_port, fd, portno, inuse, tries=0;
929 	struct port_if* pif;
930 	/* randomly select interface and port */
931 	if(num_if == 0) {
932 		verbose(VERB_QUERY, "Need to send query but have no "
933 			"outgoing interfaces of that family");
934 		return 0;
935 	}
936 	log_assert(outnet->unused_fds);
937 	tries = 0;
938 	while(1) {
939 		my_if = ub_random_max(outnet->rnd, num_if);
940 		pif = &ifs[my_if];
941 		my_port = ub_random_max(outnet->rnd, pif->avail_total);
942 		if(my_port < pif->inuse) {
943 			/* port already open */
944 			pend->pc = pif->out[my_port];
945 			verbose(VERB_ALGO, "using UDP if=%d port=%d",
946 				my_if, pend->pc->number);
947 			break;
948 		}
949 		/* try to open new port, if fails, loop to try again */
950 		log_assert(pif->inuse < pif->maxout);
951 		portno = pif->avail_ports[my_port - pif->inuse];
952 		fd = udp_sockport(&pif->addr, pif->addrlen, portno, &inuse);
953 		if(fd == -1 && !inuse) {
954 			/* nonrecoverable error making socket */
955 			return 0;
956 		}
957 		if(fd != -1) {
958 			verbose(VERB_ALGO, "opened UDP if=%d port=%d",
959 				my_if, portno);
960 			/* grab fd */
961 			pend->pc = outnet->unused_fds;
962 			outnet->unused_fds = pend->pc->next;
963 
964 			/* setup portcomm */
965 			pend->pc->next = NULL;
966 			pend->pc->number = portno;
967 			pend->pc->pif = pif;
968 			pend->pc->index = pif->inuse;
969 			pend->pc->num_outstanding = 0;
970 			comm_point_start_listening(pend->pc->cp, fd, -1);
971 
972 			/* grab port in interface */
973 			pif->out[pif->inuse] = pend->pc;
974 			pif->avail_ports[my_port - pif->inuse] =
975 				pif->avail_ports[pif->avail_total-pif->inuse-1];
976 			pif->inuse++;
977 			break;
978 		}
979 		/* failed, already in use */
980 		verbose(VERB_QUERY, "port %d in use, trying another", portno);
981 		tries++;
982 		if(tries == MAX_PORT_RETRY) {
983 			log_err("failed to find an open port, drop msg");
984 			return 0;
985 		}
986 	}
987 	log_assert(pend->pc);
988 	pend->pc->num_outstanding++;
989 
990 	return 1;
991 }
992 
993 static int
994 randomize_and_send_udp(struct outside_network* outnet, struct pending* pend,
995 	sldns_buffer* packet, int timeout)
996 {
997 	struct timeval tv;
998 
999 	/* select id */
1000 	if(!select_id(outnet, pend, packet)) {
1001 		return 0;
1002 	}
1003 
1004 	/* select src_if, port */
1005 	if(addr_is_ip6(&pend->addr, pend->addrlen)) {
1006 		if(!select_ifport(outnet, pend,
1007 			outnet->num_ip6, outnet->ip6_ifs))
1008 			return 0;
1009 	} else {
1010 		if(!select_ifport(outnet, pend,
1011 			outnet->num_ip4, outnet->ip4_ifs))
1012 			return 0;
1013 	}
1014 	log_assert(pend->pc && pend->pc->cp);
1015 
1016 	/* send it over the commlink */
1017 	if(!comm_point_send_udp_msg(pend->pc->cp, packet,
1018 		(struct sockaddr*)&pend->addr, pend->addrlen)) {
1019 		portcomm_loweruse(outnet, pend->pc);
1020 		return 0;
1021 	}
1022 
1023 	/* system calls to set timeout after sending UDP to make roundtrip
1024 	   smaller. */
1025 #ifndef S_SPLINT_S
1026 	tv.tv_sec = timeout/1000;
1027 	tv.tv_usec = (timeout%1000)*1000;
1028 #endif
1029 	comm_timer_set(pend->timer, &tv);
1030 	return 1;
1031 }
1032 
1033 struct pending*
1034 pending_udp_query(struct outside_network* outnet, sldns_buffer* packet,
1035 	struct sockaddr_storage* addr, socklen_t addrlen, int timeout,
1036 	comm_point_callback_t* cb, void* cb_arg)
1037 {
1038 	struct pending* pend = (struct pending*)calloc(1, sizeof(*pend));
1039 	if(!pend) return NULL;
1040 	pend->outnet = outnet;
1041 	pend->addrlen = addrlen;
1042 	memmove(&pend->addr, addr, addrlen);
1043 	pend->cb = cb;
1044 	pend->cb_arg = cb_arg;
1045 	pend->node.key = pend;
1046 	pend->timer = comm_timer_create(outnet->base, pending_udp_timer_cb,
1047 		pend);
1048 	if(!pend->timer) {
1049 		free(pend);
1050 		return NULL;
1051 	}
1052 
1053 	if(outnet->unused_fds == NULL) {
1054 		/* no unused fd, cannot create a new port (randomly) */
1055 		verbose(VERB_ALGO, "no fds available, udp query waiting");
1056 		pend->timeout = timeout;
1057 		pend->pkt_len = sldns_buffer_limit(packet);
1058 		pend->pkt = (uint8_t*)memdup(sldns_buffer_begin(packet),
1059 			pend->pkt_len);
1060 		if(!pend->pkt) {
1061 			comm_timer_delete(pend->timer);
1062 			free(pend);
1063 			return NULL;
1064 		}
1065 		/* put at end of waiting list */
1066 		if(outnet->udp_wait_last)
1067 			outnet->udp_wait_last->next_waiting = pend;
1068 		else
1069 			outnet->udp_wait_first = pend;
1070 		outnet->udp_wait_last = pend;
1071 		return pend;
1072 	}
1073 	if(!randomize_and_send_udp(outnet, pend, packet, timeout)) {
1074 		pending_delete(outnet, pend);
1075 		return NULL;
1076 	}
1077 	return pend;
1078 }
1079 
1080 void
1081 outnet_tcptimer(void* arg)
1082 {
1083 	struct waiting_tcp* w = (struct waiting_tcp*)arg;
1084 	struct outside_network* outnet = w->outnet;
1085 	comm_point_callback_t* cb;
1086 	void* cb_arg;
1087 	if(w->pkt) {
1088 		/* it is on the waiting list */
1089 		struct waiting_tcp* p=outnet->tcp_wait_first, *prev=NULL;
1090 		while(p) {
1091 			if(p == w) {
1092 				if(prev) prev->next_waiting = w->next_waiting;
1093 				else	outnet->tcp_wait_first=w->next_waiting;
1094 				outnet->tcp_wait_last = prev;
1095 				break;
1096 			}
1097 			prev = p;
1098 			p=p->next_waiting;
1099 		}
1100 	} else {
1101 		/* it was in use */
1102 		struct pending_tcp* pend=(struct pending_tcp*)w->next_waiting;
1103 		comm_point_close(pend->c);
1104 		pend->query = NULL;
1105 		pend->next_free = outnet->tcp_free;
1106 		outnet->tcp_free = pend;
1107 	}
1108 	cb = w->cb;
1109 	cb_arg = w->cb_arg;
1110 	waiting_tcp_delete(w);
1111 	fptr_ok(fptr_whitelist_pending_tcp(cb));
1112 	(void)(*cb)(NULL, cb_arg, NETEVENT_TIMEOUT, NULL);
1113 	use_free_buffer(outnet);
1114 }
1115 
1116 struct waiting_tcp*
1117 pending_tcp_query(struct outside_network* outnet, sldns_buffer* packet,
1118 	struct sockaddr_storage* addr, socklen_t addrlen, int timeout,
1119 	comm_point_callback_t* callback, void* callback_arg, int ssl_upstream)
1120 {
1121 	struct pending_tcp* pend = outnet->tcp_free;
1122 	struct waiting_tcp* w;
1123 	struct timeval tv;
1124 	uint16_t id;
1125 	/* if no buffer is free allocate space to store query */
1126 	w = (struct waiting_tcp*)malloc(sizeof(struct waiting_tcp)
1127 		+ (pend?0:sldns_buffer_limit(packet)));
1128 	if(!w) {
1129 		return NULL;
1130 	}
1131 	if(!(w->timer = comm_timer_create(outnet->base, outnet_tcptimer, w))) {
1132 		free(w);
1133 		return NULL;
1134 	}
1135 	w->pkt = NULL;
1136 	w->pkt_len = 0;
1137 	id = ((unsigned)ub_random(outnet->rnd)>>8) & 0xffff;
1138 	LDNS_ID_SET(sldns_buffer_begin(packet), id);
1139 	memcpy(&w->addr, addr, addrlen);
1140 	w->addrlen = addrlen;
1141 	w->outnet = outnet;
1142 	w->cb = callback;
1143 	w->cb_arg = callback_arg;
1144 	w->ssl_upstream = ssl_upstream;
1145 #ifndef S_SPLINT_S
1146 	tv.tv_sec = timeout;
1147 	tv.tv_usec = 0;
1148 #endif
1149 	comm_timer_set(w->timer, &tv);
1150 	if(pend) {
1151 		/* we have a buffer available right now */
1152 		if(!outnet_tcp_take_into_use(w, sldns_buffer_begin(packet),
1153 			sldns_buffer_limit(packet))) {
1154 			waiting_tcp_delete(w);
1155 			return NULL;
1156 		}
1157 	} else {
1158 		/* queue up */
1159 		w->pkt = (uint8_t*)w + sizeof(struct waiting_tcp);
1160 		w->pkt_len = sldns_buffer_limit(packet);
1161 		memmove(w->pkt, sldns_buffer_begin(packet), w->pkt_len);
1162 		w->next_waiting = NULL;
1163 		if(outnet->tcp_wait_last)
1164 			outnet->tcp_wait_last->next_waiting = w;
1165 		else	outnet->tcp_wait_first = w;
1166 		outnet->tcp_wait_last = w;
1167 	}
1168 	return w;
1169 }
1170 
1171 /** create query for serviced queries */
1172 static void
1173 serviced_gen_query(sldns_buffer* buff, uint8_t* qname, size_t qnamelen,
1174 	uint16_t qtype, uint16_t qclass, uint16_t flags)
1175 {
1176 	sldns_buffer_clear(buff);
1177 	/* skip id */
1178 	sldns_buffer_write_u16(buff, flags);
1179 	sldns_buffer_write_u16(buff, 1); /* qdcount */
1180 	sldns_buffer_write_u16(buff, 0); /* ancount */
1181 	sldns_buffer_write_u16(buff, 0); /* nscount */
1182 	sldns_buffer_write_u16(buff, 0); /* arcount */
1183 	sldns_buffer_write(buff, qname, qnamelen);
1184 	sldns_buffer_write_u16(buff, qtype);
1185 	sldns_buffer_write_u16(buff, qclass);
1186 	sldns_buffer_flip(buff);
1187 }
1188 
1189 /** lookup serviced query in serviced query rbtree */
1190 static struct serviced_query*
1191 lookup_serviced(struct outside_network* outnet, sldns_buffer* buff, int dnssec,
1192 	struct sockaddr_storage* addr, socklen_t addrlen)
1193 {
1194 	struct serviced_query key;
1195 	key.node.key = &key;
1196 	key.qbuf = sldns_buffer_begin(buff);
1197 	key.qbuflen = sldns_buffer_limit(buff);
1198 	key.dnssec = dnssec;
1199 	memcpy(&key.addr, addr, addrlen);
1200 	key.addrlen = addrlen;
1201 	key.outnet = outnet;
1202 	return (struct serviced_query*)rbtree_search(outnet->serviced, &key);
1203 }
1204 
1205 /** Create new serviced entry */
1206 static struct serviced_query*
1207 serviced_create(struct outside_network* outnet, sldns_buffer* buff, int dnssec,
1208 	int want_dnssec, int tcp_upstream, int ssl_upstream,
1209 	struct sockaddr_storage* addr, socklen_t addrlen, uint8_t* zone,
1210 	size_t zonelen, int qtype)
1211 {
1212 	struct serviced_query* sq = (struct serviced_query*)malloc(sizeof(*sq));
1213 #ifdef UNBOUND_DEBUG
1214 	rbnode_t* ins;
1215 #endif
1216 	if(!sq)
1217 		return NULL;
1218 	sq->node.key = sq;
1219 	sq->qbuf = memdup(sldns_buffer_begin(buff), sldns_buffer_limit(buff));
1220 	if(!sq->qbuf) {
1221 		free(sq);
1222 		return NULL;
1223 	}
1224 	sq->qbuflen = sldns_buffer_limit(buff);
1225 	sq->zone = memdup(zone, zonelen);
1226 	if(!sq->zone) {
1227 		free(sq->qbuf);
1228 		free(sq);
1229 		return NULL;
1230 	}
1231 	sq->zonelen = zonelen;
1232 	sq->qtype = qtype;
1233 	sq->dnssec = dnssec;
1234 	sq->want_dnssec = want_dnssec;
1235 	sq->tcp_upstream = tcp_upstream;
1236 	sq->ssl_upstream = ssl_upstream;
1237 	memcpy(&sq->addr, addr, addrlen);
1238 	sq->addrlen = addrlen;
1239 	sq->outnet = outnet;
1240 	sq->cblist = NULL;
1241 	sq->pending = NULL;
1242 	sq->status = serviced_initial;
1243 	sq->retry = 0;
1244 	sq->to_be_deleted = 0;
1245 #ifdef UNBOUND_DEBUG
1246 	ins =
1247 #else
1248 	(void)
1249 #endif
1250 	rbtree_insert(outnet->serviced, &sq->node);
1251 	log_assert(ins != NULL); /* must not be already present */
1252 	return sq;
1253 }
1254 
1255 /** remove waiting tcp from the outnet waiting list */
1256 static void
1257 waiting_list_remove(struct outside_network* outnet, struct waiting_tcp* w)
1258 {
1259 	struct waiting_tcp* p = outnet->tcp_wait_first, *prev = NULL;
1260 	while(p) {
1261 		if(p == w) {
1262 			/* remove w */
1263 			if(prev)
1264 				prev->next_waiting = w->next_waiting;
1265 			else	outnet->tcp_wait_first = w->next_waiting;
1266 			if(outnet->tcp_wait_last == w)
1267 				outnet->tcp_wait_last = prev;
1268 			return;
1269 		}
1270 		prev = p;
1271 		p = p->next_waiting;
1272 	}
1273 }
1274 
1275 /** cleanup serviced query entry */
1276 static void
1277 serviced_delete(struct serviced_query* sq)
1278 {
1279 	if(sq->pending) {
1280 		/* clear up the pending query */
1281 		if(sq->status == serviced_query_UDP_EDNS ||
1282 			sq->status == serviced_query_UDP ||
1283 			sq->status == serviced_query_PROBE_EDNS ||
1284 			sq->status == serviced_query_UDP_EDNS_FRAG ||
1285 			sq->status == serviced_query_UDP_EDNS_fallback) {
1286 			struct pending* p = (struct pending*)sq->pending;
1287 			if(p->pc)
1288 				portcomm_loweruse(sq->outnet, p->pc);
1289 			pending_delete(sq->outnet, p);
1290 			/* this call can cause reentrant calls back into the
1291 			 * mesh */
1292 			outnet_send_wait_udp(sq->outnet);
1293 		} else {
1294 			struct waiting_tcp* p = (struct waiting_tcp*)
1295 				sq->pending;
1296 			if(p->pkt == NULL) {
1297 				decomission_pending_tcp(sq->outnet,
1298 					(struct pending_tcp*)p->next_waiting);
1299 			} else {
1300 				waiting_list_remove(sq->outnet, p);
1301 				waiting_tcp_delete(p);
1302 			}
1303 		}
1304 	}
1305 	/* does not delete from tree, caller has to do that */
1306 	serviced_node_del(&sq->node, NULL);
1307 }
1308 
1309 /** perturb a dname capitalization randomly */
1310 static void
1311 serviced_perturb_qname(struct ub_randstate* rnd, uint8_t* qbuf, size_t len)
1312 {
1313 	uint8_t lablen;
1314 	uint8_t* d = qbuf + 10;
1315 	long int random = 0;
1316 	int bits = 0;
1317 	log_assert(len >= 10 + 5 /* offset qname, root, qtype, qclass */);
1318 	lablen = *d++;
1319 	while(lablen) {
1320 		while(lablen--) {
1321 			/* only perturb A-Z, a-z */
1322 			if(isalpha((int)*d)) {
1323 				/* get a random bit */
1324 				if(bits == 0) {
1325 					random = ub_random(rnd);
1326 					bits = 30;
1327 				}
1328 				if(random & 0x1) {
1329 					*d = (uint8_t)toupper((int)*d);
1330 				} else {
1331 					*d = (uint8_t)tolower((int)*d);
1332 				}
1333 				random >>= 1;
1334 				bits--;
1335 			}
1336 			d++;
1337 		}
1338 		lablen = *d++;
1339 	}
1340 	if(verbosity >= VERB_ALGO) {
1341 		char buf[LDNS_MAX_DOMAINLEN+1];
1342 		dname_str(qbuf+10, buf);
1343 		verbose(VERB_ALGO, "qname perturbed to %s", buf);
1344 	}
1345 }
1346 
1347 /** put serviced query into a buffer */
1348 static void
1349 serviced_encode(struct serviced_query* sq, sldns_buffer* buff, int with_edns)
1350 {
1351 	/* if we are using 0x20 bits for ID randomness, perturb them */
1352 	if(sq->outnet->use_caps_for_id) {
1353 		serviced_perturb_qname(sq->outnet->rnd, sq->qbuf, sq->qbuflen);
1354 	}
1355 	/* generate query */
1356 	sldns_buffer_clear(buff);
1357 	sldns_buffer_write_u16(buff, 0); /* id placeholder */
1358 	sldns_buffer_write(buff, sq->qbuf, sq->qbuflen);
1359 	sldns_buffer_flip(buff);
1360 	if(with_edns) {
1361 		/* add edns section */
1362 		struct edns_data edns;
1363 		edns.edns_present = 1;
1364 		edns.ext_rcode = 0;
1365 		edns.edns_version = EDNS_ADVERTISED_VERSION;
1366 		if(sq->status == serviced_query_UDP_EDNS_FRAG) {
1367 			if(addr_is_ip6(&sq->addr, sq->addrlen)) {
1368 				if(EDNS_FRAG_SIZE_IP6 < EDNS_ADVERTISED_SIZE)
1369 					edns.udp_size = EDNS_FRAG_SIZE_IP6;
1370 				else	edns.udp_size = EDNS_ADVERTISED_SIZE;
1371 			} else {
1372 				if(EDNS_FRAG_SIZE_IP4 < EDNS_ADVERTISED_SIZE)
1373 					edns.udp_size = EDNS_FRAG_SIZE_IP4;
1374 				else	edns.udp_size = EDNS_ADVERTISED_SIZE;
1375 			}
1376 		} else {
1377 			edns.udp_size = EDNS_ADVERTISED_SIZE;
1378 		}
1379 		edns.bits = 0;
1380 		if(sq->dnssec & EDNS_DO)
1381 			edns.bits = EDNS_DO;
1382 		if(sq->dnssec & BIT_CD)
1383 			LDNS_CD_SET(sldns_buffer_begin(buff));
1384 		attach_edns_record(buff, &edns);
1385 	}
1386 }
1387 
1388 /**
1389  * Perform serviced query UDP sending operation.
1390  * Sends UDP with EDNS, unless infra host marked non EDNS.
1391  * @param sq: query to send.
1392  * @param buff: buffer scratch space.
1393  * @return 0 on error.
1394  */
1395 static int
1396 serviced_udp_send(struct serviced_query* sq, sldns_buffer* buff)
1397 {
1398 	int rtt, vs;
1399 	uint8_t edns_lame_known;
1400 	time_t now = *sq->outnet->now_secs;
1401 
1402 	if(!infra_host(sq->outnet->infra, &sq->addr, sq->addrlen, sq->zone,
1403 		sq->zonelen, now, &vs, &edns_lame_known, &rtt))
1404 		return 0;
1405 	sq->last_rtt = rtt;
1406 	verbose(VERB_ALGO, "EDNS lookup known=%d vs=%d", edns_lame_known, vs);
1407 	if(sq->status == serviced_initial) {
1408 		if(edns_lame_known == 0 && rtt > 5000 && rtt < 10001) {
1409 			/* perform EDNS lame probe - check if server is
1410 			 * EDNS lame (EDNS queries to it are dropped) */
1411 			verbose(VERB_ALGO, "serviced query: send probe to see "
1412 				" if use of EDNS causes timeouts");
1413 			/* even 700 msec may be too small */
1414 			rtt = 1000;
1415 			sq->status = serviced_query_PROBE_EDNS;
1416 		} else if(vs != -1) {
1417 			sq->status = serviced_query_UDP_EDNS;
1418 		} else {
1419 			sq->status = serviced_query_UDP;
1420 		}
1421 	}
1422 	serviced_encode(sq, buff, (sq->status == serviced_query_UDP_EDNS) ||
1423 		(sq->status == serviced_query_UDP_EDNS_FRAG));
1424 	sq->last_sent_time = *sq->outnet->now_tv;
1425 	sq->edns_lame_known = (int)edns_lame_known;
1426 	verbose(VERB_ALGO, "serviced query UDP timeout=%d msec", rtt);
1427 	sq->pending = pending_udp_query(sq->outnet, buff, &sq->addr,
1428 		sq->addrlen, rtt, serviced_udp_callback, sq);
1429 	if(!sq->pending)
1430 		return 0;
1431 	return 1;
1432 }
1433 
1434 /** check that perturbed qname is identical */
1435 static int
1436 serviced_check_qname(sldns_buffer* pkt, uint8_t* qbuf, size_t qbuflen)
1437 {
1438 	uint8_t* d1 = sldns_buffer_at(pkt, 12);
1439 	uint8_t* d2 = qbuf+10;
1440 	uint8_t len1, len2;
1441 	int count = 0;
1442 	log_assert(qbuflen >= 15 /* 10 header, root, type, class */);
1443 	len1 = *d1++;
1444 	len2 = *d2++;
1445 	if(sldns_buffer_limit(pkt) < 12+1+4) /* packet too small for qname */
1446 		return 0;
1447 	while(len1 != 0 || len2 != 0) {
1448 		if(LABEL_IS_PTR(len1)) {
1449 			d1 = sldns_buffer_at(pkt, PTR_OFFSET(len1, *d1));
1450 			if(d1 >= sldns_buffer_at(pkt, sldns_buffer_limit(pkt)))
1451 				return 0;
1452 			len1 = *d1++;
1453 			if(count++ > MAX_COMPRESS_PTRS)
1454 				return 0;
1455 			continue;
1456 		}
1457 		if(d2 > qbuf+qbuflen)
1458 			return 0;
1459 		if(len1 != len2)
1460 			return 0;
1461 		if(len1 > LDNS_MAX_LABELLEN)
1462 			return 0;
1463 		log_assert(len1 <= LDNS_MAX_LABELLEN);
1464 		log_assert(len2 <= LDNS_MAX_LABELLEN);
1465 		log_assert(len1 == len2 && len1 != 0);
1466 		/* compare the labels - bitwise identical */
1467 		if(memcmp(d1, d2, len1) != 0)
1468 			return 0;
1469 		d1 += len1;
1470 		d2 += len2;
1471 		len1 = *d1++;
1472 		len2 = *d2++;
1473 	}
1474 	return 1;
1475 }
1476 
1477 /** call the callbacks for a serviced query */
1478 static void
1479 serviced_callbacks(struct serviced_query* sq, int error, struct comm_point* c,
1480 	struct comm_reply* rep)
1481 {
1482 	struct service_callback* p;
1483 	int dobackup = (sq->cblist && sq->cblist->next); /* >1 cb*/
1484 	uint8_t *backup_p = NULL;
1485 	size_t backlen = 0;
1486 #ifdef UNBOUND_DEBUG
1487 	rbnode_t* rem =
1488 #else
1489 	(void)
1490 #endif
1491 	/* remove from tree, and schedule for deletion, so that callbacks
1492 	 * can safely deregister themselves and even create new serviced
1493 	 * queries that are identical to this one. */
1494 	rbtree_delete(sq->outnet->serviced, sq);
1495 	log_assert(rem); /* should have been present */
1496 	sq->to_be_deleted = 1;
1497 	verbose(VERB_ALGO, "svcd callbacks start");
1498 	if(sq->outnet->use_caps_for_id && error == NETEVENT_NOERROR && c) {
1499 		/* noerror and nxdomain must have a qname in reply */
1500 		if(sldns_buffer_read_u16_at(c->buffer, 4) == 0 &&
1501 			(LDNS_RCODE_WIRE(sldns_buffer_begin(c->buffer))
1502 				== LDNS_RCODE_NOERROR ||
1503 			 LDNS_RCODE_WIRE(sldns_buffer_begin(c->buffer))
1504 				== LDNS_RCODE_NXDOMAIN)) {
1505 			verbose(VERB_DETAIL, "no qname in reply to check 0x20ID");
1506 			log_addr(VERB_DETAIL, "from server",
1507 				&sq->addr, sq->addrlen);
1508 			log_buf(VERB_DETAIL, "for packet", c->buffer);
1509 			error = NETEVENT_CLOSED;
1510 			c = NULL;
1511 		} else if(sldns_buffer_read_u16_at(c->buffer, 4) > 0 &&
1512 			!serviced_check_qname(c->buffer, sq->qbuf,
1513 			sq->qbuflen)) {
1514 			verbose(VERB_DETAIL, "wrong 0x20-ID in reply qname");
1515 			log_addr(VERB_DETAIL, "from server",
1516 				&sq->addr, sq->addrlen);
1517 			log_buf(VERB_DETAIL, "for packet", c->buffer);
1518 			error = NETEVENT_CAPSFAIL;
1519 			/* and cleanup too */
1520 			pkt_dname_tolower(c->buffer,
1521 				sldns_buffer_at(c->buffer, 12));
1522 		} else {
1523 			verbose(VERB_ALGO, "good 0x20-ID in reply qname");
1524 			/* cleanup caps, prettier cache contents. */
1525 			pkt_dname_tolower(c->buffer,
1526 				sldns_buffer_at(c->buffer, 12));
1527 		}
1528 	}
1529 	if(dobackup && c) {
1530 		/* make a backup of the query, since the querystate processing
1531 		 * may send outgoing queries that overwrite the buffer.
1532 		 * use secondary buffer to store the query.
1533 		 * This is a data copy, but faster than packet to server */
1534 		backlen = sldns_buffer_limit(c->buffer);
1535 		backup_p = memdup(sldns_buffer_begin(c->buffer), backlen);
1536 		if(!backup_p) {
1537 			log_err("malloc failure in serviced query callbacks");
1538 			error = NETEVENT_CLOSED;
1539 			c = NULL;
1540 		}
1541 		sq->outnet->svcd_overhead = backlen;
1542 	}
1543 	/* test the actual sq->cblist, because the next elem could be deleted*/
1544 	while((p=sq->cblist) != NULL) {
1545 		sq->cblist = p->next; /* remove this element */
1546 		if(dobackup && c) {
1547 			sldns_buffer_clear(c->buffer);
1548 			sldns_buffer_write(c->buffer, backup_p, backlen);
1549 			sldns_buffer_flip(c->buffer);
1550 		}
1551 		fptr_ok(fptr_whitelist_serviced_query(p->cb));
1552 		(void)(*p->cb)(c, p->cb_arg, error, rep);
1553 		free(p);
1554 	}
1555 	if(backup_p) {
1556 		free(backup_p);
1557 		sq->outnet->svcd_overhead = 0;
1558 	}
1559 	verbose(VERB_ALGO, "svcd callbacks end");
1560 	log_assert(sq->cblist == NULL);
1561 	serviced_delete(sq);
1562 }
1563 
1564 int
1565 serviced_tcp_callback(struct comm_point* c, void* arg, int error,
1566         struct comm_reply* rep)
1567 {
1568 	struct serviced_query* sq = (struct serviced_query*)arg;
1569 	struct comm_reply r2;
1570 	sq->pending = NULL; /* removed after this callback */
1571 	if(error != NETEVENT_NOERROR)
1572 		log_addr(VERB_QUERY, "tcp error for address",
1573 			&sq->addr, sq->addrlen);
1574 	if(error==NETEVENT_NOERROR)
1575 		infra_update_tcp_works(sq->outnet->infra, &sq->addr,
1576 			sq->addrlen, sq->zone, sq->zonelen);
1577 	if(error==NETEVENT_NOERROR && sq->status == serviced_query_TCP_EDNS &&
1578 		(LDNS_RCODE_WIRE(sldns_buffer_begin(c->buffer)) ==
1579 		LDNS_RCODE_FORMERR || LDNS_RCODE_WIRE(sldns_buffer_begin(
1580 		c->buffer)) == LDNS_RCODE_NOTIMPL) ) {
1581 		/* attempt to fallback to nonEDNS */
1582 		sq->status = serviced_query_TCP_EDNS_fallback;
1583 		serviced_tcp_initiate(sq->outnet, sq, c->buffer);
1584 		return 0;
1585 	} else if(error==NETEVENT_NOERROR &&
1586 		sq->status == serviced_query_TCP_EDNS_fallback &&
1587 			(LDNS_RCODE_WIRE(sldns_buffer_begin(c->buffer)) ==
1588 			LDNS_RCODE_NOERROR || LDNS_RCODE_WIRE(
1589 			sldns_buffer_begin(c->buffer)) == LDNS_RCODE_NXDOMAIN
1590 			|| LDNS_RCODE_WIRE(sldns_buffer_begin(c->buffer))
1591 			== LDNS_RCODE_YXDOMAIN)) {
1592 		/* the fallback produced a result that looks promising, note
1593 		 * that this server should be approached without EDNS */
1594 		/* only store noEDNS in cache if domain is noDNSSEC */
1595 		if(!sq->want_dnssec)
1596 		  if(!infra_edns_update(sq->outnet->infra, &sq->addr,
1597 			sq->addrlen, sq->zone, sq->zonelen, -1,
1598 			*sq->outnet->now_secs))
1599 			log_err("Out of memory caching no edns for host");
1600 		sq->status = serviced_query_TCP;
1601 	}
1602 	if(sq->tcp_upstream || sq->ssl_upstream) {
1603 	    struct timeval now = *sq->outnet->now_tv;
1604 	    if(now.tv_sec > sq->last_sent_time.tv_sec ||
1605 		(now.tv_sec == sq->last_sent_time.tv_sec &&
1606 		now.tv_usec > sq->last_sent_time.tv_usec)) {
1607 		/* convert from microseconds to milliseconds */
1608 		int roundtime = ((int)(now.tv_sec - sq->last_sent_time.tv_sec))*1000
1609 		  + ((int)now.tv_usec - (int)sq->last_sent_time.tv_usec)/1000;
1610 		verbose(VERB_ALGO, "measured TCP-time at %d msec", roundtime);
1611 		log_assert(roundtime >= 0);
1612 		/* only store if less then AUTH_TIMEOUT seconds, it could be
1613 		 * huge due to system-hibernated and we woke up */
1614 		if(roundtime < TCP_AUTH_QUERY_TIMEOUT*1000) {
1615 		    if(!infra_rtt_update(sq->outnet->infra, &sq->addr,
1616 			sq->addrlen, sq->zone, sq->zonelen, sq->qtype,
1617 			roundtime, sq->last_rtt, (time_t)now.tv_sec))
1618 			log_err("out of memory noting rtt.");
1619 		}
1620 	    }
1621 	}
1622 	/* insert address into reply info */
1623 	if(!rep) {
1624 		/* create one if there isn't (on errors) */
1625 		rep = &r2;
1626 		r2.c = c;
1627 	}
1628 	memcpy(&rep->addr, &sq->addr, sq->addrlen);
1629 	rep->addrlen = sq->addrlen;
1630 	serviced_callbacks(sq, error, c, rep);
1631 	return 0;
1632 }
1633 
1634 static void
1635 serviced_tcp_initiate(struct outside_network* outnet,
1636 	struct serviced_query* sq, sldns_buffer* buff)
1637 {
1638 	verbose(VERB_ALGO, "initiate TCP query %s",
1639 		sq->status==serviced_query_TCP_EDNS?"EDNS":"");
1640 	serviced_encode(sq, buff, sq->status == serviced_query_TCP_EDNS);
1641 	sq->last_sent_time = *sq->outnet->now_tv;
1642 	sq->pending = pending_tcp_query(outnet, buff, &sq->addr,
1643 		sq->addrlen, TCP_AUTH_QUERY_TIMEOUT, serviced_tcp_callback,
1644 		sq, sq->ssl_upstream);
1645 	if(!sq->pending) {
1646 		/* delete from tree so that a retry by above layer does not
1647 		 * clash with this entry */
1648 		log_err("serviced_tcp_initiate: failed to send tcp query");
1649 		serviced_callbacks(sq, NETEVENT_CLOSED, NULL, NULL);
1650 	}
1651 }
1652 
1653 /** Send serviced query over TCP return false on initial failure */
1654 static int
1655 serviced_tcp_send(struct serviced_query* sq, sldns_buffer* buff)
1656 {
1657 	int vs, rtt;
1658 	uint8_t edns_lame_known;
1659 	if(!infra_host(sq->outnet->infra, &sq->addr, sq->addrlen, sq->zone,
1660 		sq->zonelen, *sq->outnet->now_secs, &vs, &edns_lame_known,
1661 		&rtt))
1662 		return 0;
1663 	if(vs != -1)
1664 		sq->status = serviced_query_TCP_EDNS;
1665 	else 	sq->status = serviced_query_TCP;
1666 	serviced_encode(sq, buff, sq->status == serviced_query_TCP_EDNS);
1667 	sq->last_sent_time = *sq->outnet->now_tv;
1668 	sq->pending = pending_tcp_query(sq->outnet, buff, &sq->addr,
1669 		sq->addrlen, TCP_AUTH_QUERY_TIMEOUT, serviced_tcp_callback,
1670 		sq, sq->ssl_upstream);
1671 	return sq->pending != NULL;
1672 }
1673 
1674 int
1675 serviced_udp_callback(struct comm_point* c, void* arg, int error,
1676         struct comm_reply* rep)
1677 {
1678 	struct serviced_query* sq = (struct serviced_query*)arg;
1679 	struct outside_network* outnet = sq->outnet;
1680 	struct timeval now = *sq->outnet->now_tv;
1681 	int fallback_tcp = 0;
1682 
1683 	sq->pending = NULL; /* removed after callback */
1684 	if(error == NETEVENT_TIMEOUT) {
1685 		int rto = 0;
1686 		if(sq->status == serviced_query_PROBE_EDNS) {
1687 			/* non-EDNS probe failed; we do not know its status,
1688 			 * keep trying with EDNS, timeout may not be caused
1689 			 * by EDNS. */
1690 			sq->status = serviced_query_UDP_EDNS;
1691 		}
1692 		if(sq->status == serviced_query_UDP_EDNS && sq->last_rtt < 5000) {
1693 			/* fallback to 1480/1280 */
1694 			sq->status = serviced_query_UDP_EDNS_FRAG;
1695 			log_name_addr(VERB_ALGO, "try edns1xx0", sq->qbuf+10,
1696 				&sq->addr, sq->addrlen);
1697 			if(!serviced_udp_send(sq, c->buffer)) {
1698 				serviced_callbacks(sq, NETEVENT_CLOSED, c, rep);
1699 			}
1700 			return 0;
1701 		}
1702 		if(sq->status == serviced_query_UDP_EDNS_FRAG) {
1703 			/* fragmentation size did not fix it */
1704 			sq->status = serviced_query_UDP_EDNS;
1705 		}
1706 		sq->retry++;
1707 		if(!(rto=infra_rtt_update(outnet->infra, &sq->addr, sq->addrlen,
1708 			sq->zone, sq->zonelen, sq->qtype, -1, sq->last_rtt,
1709 			(time_t)now.tv_sec)))
1710 			log_err("out of memory in UDP exponential backoff");
1711 		if(sq->retry < OUTBOUND_UDP_RETRY) {
1712 			log_name_addr(VERB_ALGO, "retry query", sq->qbuf+10,
1713 				&sq->addr, sq->addrlen);
1714 			if(!serviced_udp_send(sq, c->buffer)) {
1715 				serviced_callbacks(sq, NETEVENT_CLOSED, c, rep);
1716 			}
1717 			return 0;
1718 		}
1719 		if(rto >= RTT_MAX_TIMEOUT) {
1720 			fallback_tcp = 1;
1721 			/* UDP does not work, fallback to TCP below */
1722 		} else {
1723 			serviced_callbacks(sq, NETEVENT_TIMEOUT, c, rep);
1724 			return 0;
1725 		}
1726 	} else if(error != NETEVENT_NOERROR) {
1727 		/* udp returns error (due to no ID or interface available) */
1728 		serviced_callbacks(sq, error, c, rep);
1729 		return 0;
1730 	}
1731 	if(!fallback_tcp) {
1732 	    if( (sq->status == serviced_query_UDP_EDNS
1733 	        ||sq->status == serviced_query_UDP_EDNS_FRAG)
1734 		&& (LDNS_RCODE_WIRE(sldns_buffer_begin(c->buffer))
1735 			== LDNS_RCODE_FORMERR || LDNS_RCODE_WIRE(
1736 			sldns_buffer_begin(c->buffer)) == LDNS_RCODE_NOTIMPL)) {
1737 		/* try to get an answer by falling back without EDNS */
1738 		verbose(VERB_ALGO, "serviced query: attempt without EDNS");
1739 		sq->status = serviced_query_UDP_EDNS_fallback;
1740 		sq->retry = 0;
1741 		if(!serviced_udp_send(sq, c->buffer)) {
1742 			serviced_callbacks(sq, NETEVENT_CLOSED, c, rep);
1743 		}
1744 		return 0;
1745 	    } else if(sq->status == serviced_query_PROBE_EDNS) {
1746 		/* probe without EDNS succeeds, so we conclude that this
1747 		 * host likely has EDNS packets dropped */
1748 		log_addr(VERB_DETAIL, "timeouts, concluded that connection to "
1749 			"host drops EDNS packets", &sq->addr, sq->addrlen);
1750 		/* only store noEDNS in cache if domain is noDNSSEC */
1751 		if(!sq->want_dnssec)
1752 		  if(!infra_edns_update(outnet->infra, &sq->addr, sq->addrlen,
1753 			sq->zone, sq->zonelen, -1, (time_t)now.tv_sec)) {
1754 			log_err("Out of memory caching no edns for host");
1755 		  }
1756 		sq->status = serviced_query_UDP;
1757 	    } else if(sq->status == serviced_query_UDP_EDNS &&
1758 		!sq->edns_lame_known) {
1759 		/* now we know that edns queries received answers store that */
1760 		log_addr(VERB_ALGO, "serviced query: EDNS works for",
1761 			&sq->addr, sq->addrlen);
1762 		if(!infra_edns_update(outnet->infra, &sq->addr, sq->addrlen,
1763 			sq->zone, sq->zonelen, 0, (time_t)now.tv_sec)) {
1764 			log_err("Out of memory caching edns works");
1765 		}
1766 		sq->edns_lame_known = 1;
1767 	    } else if(sq->status == serviced_query_UDP_EDNS_fallback &&
1768 		!sq->edns_lame_known && (LDNS_RCODE_WIRE(
1769 		sldns_buffer_begin(c->buffer)) == LDNS_RCODE_NOERROR ||
1770 		LDNS_RCODE_WIRE(sldns_buffer_begin(c->buffer)) ==
1771 		LDNS_RCODE_NXDOMAIN || LDNS_RCODE_WIRE(sldns_buffer_begin(
1772 		c->buffer)) == LDNS_RCODE_YXDOMAIN)) {
1773 		/* the fallback produced a result that looks promising, note
1774 		 * that this server should be approached without EDNS */
1775 		/* only store noEDNS in cache if domain is noDNSSEC */
1776 		if(!sq->want_dnssec) {
1777 		  log_addr(VERB_ALGO, "serviced query: EDNS fails for",
1778 			&sq->addr, sq->addrlen);
1779 		  if(!infra_edns_update(outnet->infra, &sq->addr, sq->addrlen,
1780 			sq->zone, sq->zonelen, -1, (time_t)now.tv_sec)) {
1781 			log_err("Out of memory caching no edns for host");
1782 		  }
1783 		} else {
1784 		  log_addr(VERB_ALGO, "serviced query: EDNS fails, but "
1785 		  	"not stored because need DNSSEC for", &sq->addr,
1786 			sq->addrlen);
1787 		}
1788 		sq->status = serviced_query_UDP;
1789 	    }
1790 	    if(now.tv_sec > sq->last_sent_time.tv_sec ||
1791 		(now.tv_sec == sq->last_sent_time.tv_sec &&
1792 		now.tv_usec > sq->last_sent_time.tv_usec)) {
1793 		/* convert from microseconds to milliseconds */
1794 		int roundtime = ((int)(now.tv_sec - sq->last_sent_time.tv_sec))*1000
1795 		  + ((int)now.tv_usec - (int)sq->last_sent_time.tv_usec)/1000;
1796 		verbose(VERB_ALGO, "measured roundtrip at %d msec", roundtime);
1797 		log_assert(roundtime >= 0);
1798 		/* in case the system hibernated, do not enter a huge value,
1799 		 * above this value gives trouble with server selection */
1800 		if(roundtime < 60000) {
1801 		    if(!infra_rtt_update(outnet->infra, &sq->addr, sq->addrlen,
1802 			sq->zone, sq->zonelen, sq->qtype, roundtime,
1803 			sq->last_rtt, (time_t)now.tv_sec))
1804 			log_err("out of memory noting rtt.");
1805 		}
1806 	    }
1807 	} /* end of if_!fallback_tcp */
1808 	/* perform TC flag check and TCP fallback after updating our
1809 	 * cache entries for EDNS status and RTT times */
1810 	if(LDNS_TC_WIRE(sldns_buffer_begin(c->buffer)) || fallback_tcp) {
1811 		/* fallback to TCP */
1812 		/* this discards partial UDP contents */
1813 		if(sq->status == serviced_query_UDP_EDNS ||
1814 			sq->status == serviced_query_UDP_EDNS_FRAG ||
1815 			sq->status == serviced_query_UDP_EDNS_fallback)
1816 			/* if we have unfinished EDNS_fallback, start again */
1817 			sq->status = serviced_query_TCP_EDNS;
1818 		else	sq->status = serviced_query_TCP;
1819 		serviced_tcp_initiate(outnet, sq, c->buffer);
1820 		return 0;
1821 	}
1822 	/* yay! an answer */
1823 	serviced_callbacks(sq, error, c, rep);
1824 	return 0;
1825 }
1826 
1827 struct serviced_query*
1828 outnet_serviced_query(struct outside_network* outnet,
1829 	uint8_t* qname, size_t qnamelen, uint16_t qtype, uint16_t qclass,
1830 	uint16_t flags, int dnssec, int want_dnssec, int tcp_upstream,
1831 	int ssl_upstream, struct sockaddr_storage* addr, socklen_t addrlen,
1832 	uint8_t* zone, size_t zonelen, comm_point_callback_t* callback,
1833 	void* callback_arg, sldns_buffer* buff)
1834 {
1835 	struct serviced_query* sq;
1836 	struct service_callback* cb;
1837 	serviced_gen_query(buff, qname, qnamelen, qtype, qclass, flags);
1838 	sq = lookup_serviced(outnet, buff, dnssec, addr, addrlen);
1839 	/* duplicate entries are included in the callback list, because
1840 	 * there is a counterpart registration by our caller that needs to
1841 	 * be doubly-removed (with callbacks perhaps). */
1842 	if(!(cb = (struct service_callback*)malloc(sizeof(*cb))))
1843 		return NULL;
1844 	if(!sq) {
1845 		/* make new serviced query entry */
1846 		sq = serviced_create(outnet, buff, dnssec, want_dnssec,
1847 			tcp_upstream, ssl_upstream, addr, addrlen, zone,
1848 			zonelen, (int)qtype);
1849 		if(!sq) {
1850 			free(cb);
1851 			return NULL;
1852 		}
1853 		/* perform first network action */
1854 		if(outnet->do_udp && !(tcp_upstream || ssl_upstream)) {
1855 			if(!serviced_udp_send(sq, buff)) {
1856 				(void)rbtree_delete(outnet->serviced, sq);
1857 				free(sq->qbuf);
1858 				free(sq->zone);
1859 				free(sq);
1860 				free(cb);
1861 				return NULL;
1862 			}
1863 		} else {
1864 			if(!serviced_tcp_send(sq, buff)) {
1865 				(void)rbtree_delete(outnet->serviced, sq);
1866 				free(sq->qbuf);
1867 				free(sq->zone);
1868 				free(sq);
1869 				free(cb);
1870 				return NULL;
1871 			}
1872 		}
1873 	}
1874 	/* add callback to list of callbacks */
1875 	cb->cb = callback;
1876 	cb->cb_arg = callback_arg;
1877 	cb->next = sq->cblist;
1878 	sq->cblist = cb;
1879 	return sq;
1880 }
1881 
1882 /** remove callback from list */
1883 static void
1884 callback_list_remove(struct serviced_query* sq, void* cb_arg)
1885 {
1886 	struct service_callback** pp = &sq->cblist;
1887 	while(*pp) {
1888 		if((*pp)->cb_arg == cb_arg) {
1889 			struct service_callback* del = *pp;
1890 			*pp = del->next;
1891 			free(del);
1892 			return;
1893 		}
1894 		pp = &(*pp)->next;
1895 	}
1896 }
1897 
1898 void outnet_serviced_query_stop(struct serviced_query* sq, void* cb_arg)
1899 {
1900 	if(!sq)
1901 		return;
1902 	callback_list_remove(sq, cb_arg);
1903 	/* if callbacks() routine scheduled deletion, let it do that */
1904 	if(!sq->cblist && !sq->to_be_deleted) {
1905 #ifdef UNBOUND_DEBUG
1906 		rbnode_t* rem =
1907 #else
1908 		(void)
1909 #endif
1910 		rbtree_delete(sq->outnet->serviced, sq);
1911 		log_assert(rem); /* should be present */
1912 		serviced_delete(sq);
1913 	}
1914 }
1915 
1916 /** get memory used by waiting tcp entry (in use or not) */
1917 static size_t
1918 waiting_tcp_get_mem(struct waiting_tcp* w)
1919 {
1920 	size_t s;
1921 	if(!w) return 0;
1922 	s = sizeof(*w) + w->pkt_len;
1923 	if(w->timer)
1924 		s += comm_timer_get_mem(w->timer);
1925 	return s;
1926 }
1927 
1928 /** get memory used by port if */
1929 static size_t
1930 if_get_mem(struct port_if* pif)
1931 {
1932 	size_t s;
1933 	int i;
1934 	s = sizeof(*pif) + sizeof(int)*pif->avail_total +
1935 		sizeof(struct port_comm*)*pif->maxout;
1936 	for(i=0; i<pif->inuse; i++)
1937 		s += sizeof(*pif->out[i]) +
1938 			comm_point_get_mem(pif->out[i]->cp);
1939 	return s;
1940 }
1941 
1942 /** get memory used by waiting udp */
1943 static size_t
1944 waiting_udp_get_mem(struct pending* w)
1945 {
1946 	size_t s;
1947 	s = sizeof(*w) + comm_timer_get_mem(w->timer) + w->pkt_len;
1948 	return s;
1949 }
1950 
1951 size_t outnet_get_mem(struct outside_network* outnet)
1952 {
1953 	size_t i;
1954 	int k;
1955 	struct waiting_tcp* w;
1956 	struct pending* u;
1957 	struct serviced_query* sq;
1958 	struct service_callback* sb;
1959 	struct port_comm* pc;
1960 	size_t s = sizeof(*outnet) + sizeof(*outnet->base) +
1961 		sizeof(*outnet->udp_buff) +
1962 		sldns_buffer_capacity(outnet->udp_buff);
1963 	/* second buffer is not ours */
1964 	for(pc = outnet->unused_fds; pc; pc = pc->next) {
1965 		s += sizeof(*pc) + comm_point_get_mem(pc->cp);
1966 	}
1967 	for(k=0; k<outnet->num_ip4; k++)
1968 		s += if_get_mem(&outnet->ip4_ifs[k]);
1969 	for(k=0; k<outnet->num_ip6; k++)
1970 		s += if_get_mem(&outnet->ip6_ifs[k]);
1971 	for(u=outnet->udp_wait_first; u; u=u->next_waiting)
1972 		s += waiting_udp_get_mem(u);
1973 
1974 	s += sizeof(struct pending_tcp*)*outnet->num_tcp;
1975 	for(i=0; i<outnet->num_tcp; i++) {
1976 		s += sizeof(struct pending_tcp);
1977 		s += comm_point_get_mem(outnet->tcp_conns[i]->c);
1978 		if(outnet->tcp_conns[i]->query)
1979 			s += waiting_tcp_get_mem(outnet->tcp_conns[i]->query);
1980 	}
1981 	for(w=outnet->tcp_wait_first; w; w = w->next_waiting)
1982 		s += waiting_tcp_get_mem(w);
1983 	s += sizeof(*outnet->pending);
1984 	s += (sizeof(struct pending) + comm_timer_get_mem(NULL)) *
1985 		outnet->pending->count;
1986 	s += sizeof(*outnet->serviced);
1987 	s += outnet->svcd_overhead;
1988 	RBTREE_FOR(sq, struct serviced_query*, outnet->serviced) {
1989 		s += sizeof(*sq) + sq->qbuflen;
1990 		for(sb = sq->cblist; sb; sb = sb->next)
1991 			s += sizeof(*sb);
1992 	}
1993 	return s;
1994 }
1995 
1996 size_t
1997 serviced_get_mem(struct serviced_query* sq)
1998 {
1999 	struct service_callback* sb;
2000 	size_t s;
2001 	s = sizeof(*sq) + sq->qbuflen;
2002 	for(sb = sq->cblist; sb; sb = sb->next)
2003 		s += sizeof(*sb);
2004 	if(sq->status == serviced_query_UDP_EDNS ||
2005 		sq->status == serviced_query_UDP ||
2006 		sq->status == serviced_query_PROBE_EDNS ||
2007 		sq->status == serviced_query_UDP_EDNS_FRAG ||
2008 		sq->status == serviced_query_UDP_EDNS_fallback) {
2009 		s += sizeof(struct pending);
2010 		s += comm_timer_get_mem(NULL);
2011 	} else {
2012 		/* does not have size of the pkt pointer */
2013 		/* always has a timer except on malloc failures */
2014 
2015 		/* these sizes are part of the main outside network mem */
2016 		/*
2017 		s += sizeof(struct waiting_tcp);
2018 		s += comm_timer_get_mem(NULL);
2019 		*/
2020 	}
2021 	return s;
2022 }
2023 
2024