]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/blob - contrib/unbound/services/outside_network.c
unbound: Vendor import 1.18.0
[FreeBSD/FreeBSD.git] / contrib / unbound / services / outside_network.c
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 "iterator/iterator.h"
52 #include "util/data/msgparse.h"
53 #include "util/data/msgreply.h"
54 #include "util/data/msgencode.h"
55 #include "util/data/dname.h"
56 #include "util/netevent.h"
57 #include "util/log.h"
58 #include "util/net_help.h"
59 #include "util/random.h"
60 #include "util/fptr_wlist.h"
61 #include "util/edns.h"
62 #include "sldns/sbuffer.h"
63 #include "dnstap/dnstap.h"
64 #ifdef HAVE_OPENSSL_SSL_H
65 #include <openssl/ssl.h>
66 #endif
67 #ifdef HAVE_X509_VERIFY_PARAM_SET1_HOST
68 #include <openssl/x509v3.h>
69 #endif
70
71 #ifdef HAVE_NETDB_H
72 #include <netdb.h>
73 #endif
74 #include <fcntl.h>
75
76 /** number of times to retry making a random ID that is unique. */
77 #define MAX_ID_RETRY 1000
78 /** number of times to retry finding interface, port that can be opened. */
79 #define MAX_PORT_RETRY 10000
80 /** number of retries on outgoing UDP queries */
81 #define OUTBOUND_UDP_RETRY 1
82
83 /** initiate TCP transaction for serviced query */
84 static void serviced_tcp_initiate(struct serviced_query* sq, sldns_buffer* buff);
85 /** with a fd available, randomize and send UDP */
86 static int randomize_and_send_udp(struct pending* pend, sldns_buffer* packet,
87         int timeout);
88
89 /** select a DNS ID for a TCP stream */
90 static uint16_t tcp_select_id(struct outside_network* outnet,
91         struct reuse_tcp* reuse);
92
93 /** Perform serviced query UDP sending operation */
94 static int serviced_udp_send(struct serviced_query* sq, sldns_buffer* buff);
95
96 /** Send serviced query over TCP return false on initial failure */
97 static int serviced_tcp_send(struct serviced_query* sq, sldns_buffer* buff);
98
99 /** call the callbacks for a serviced query */
100 static void serviced_callbacks(struct serviced_query* sq, int error,
101         struct comm_point* c, struct comm_reply* rep);
102
103 int 
104 pending_cmp(const void* key1, const void* key2)
105 {
106         struct pending *p1 = (struct pending*)key1;
107         struct pending *p2 = (struct pending*)key2;
108         if(p1->id < p2->id)
109                 return -1;
110         if(p1->id > p2->id)
111                 return 1;
112         log_assert(p1->id == p2->id);
113         return sockaddr_cmp(&p1->addr, p1->addrlen, &p2->addr, p2->addrlen);
114 }
115
116 int 
117 serviced_cmp(const void* key1, const void* key2)
118 {
119         struct serviced_query* q1 = (struct serviced_query*)key1;
120         struct serviced_query* q2 = (struct serviced_query*)key2;
121         int r;
122         if(q1->qbuflen < q2->qbuflen)
123                 return -1;
124         if(q1->qbuflen > q2->qbuflen)
125                 return 1;
126         log_assert(q1->qbuflen == q2->qbuflen);
127         log_assert(q1->qbuflen >= 15 /* 10 header, root, type, class */);
128         /* alternate casing of qname is still the same query */
129         if((r = memcmp(q1->qbuf, q2->qbuf, 10)) != 0)
130                 return r;
131         if((r = memcmp(q1->qbuf+q1->qbuflen-4, q2->qbuf+q2->qbuflen-4, 4)) != 0)
132                 return r;
133         if(q1->dnssec != q2->dnssec) {
134                 if(q1->dnssec < q2->dnssec)
135                         return -1;
136                 return 1;
137         }
138         if((r = query_dname_compare(q1->qbuf+10, q2->qbuf+10)) != 0)
139                 return r;
140         if((r = edns_opt_list_compare(q1->opt_list, q2->opt_list)) != 0)
141                 return r;
142         return sockaddr_cmp(&q1->addr, q1->addrlen, &q2->addr, q2->addrlen);
143 }
144
145 /** compare if the reuse element has the same address, port and same ssl-is
146  * used-for-it characteristic */
147 static int
148 reuse_cmp_addrportssl(const void* key1, const void* key2)
149 {
150         struct reuse_tcp* r1 = (struct reuse_tcp*)key1;
151         struct reuse_tcp* r2 = (struct reuse_tcp*)key2;
152         int r;
153         /* compare address and port */
154         r = sockaddr_cmp(&r1->addr, r1->addrlen, &r2->addr, r2->addrlen);
155         if(r != 0)
156                 return r;
157
158         /* compare if SSL-enabled */
159         if(r1->is_ssl && !r2->is_ssl)
160                 return 1;
161         if(!r1->is_ssl && r2->is_ssl)
162                 return -1;
163         return 0;
164 }
165
166 int
167 reuse_cmp(const void* key1, const void* key2)
168 {
169         int r;
170         r = reuse_cmp_addrportssl(key1, key2);
171         if(r != 0)
172                 return r;
173
174         /* compare ptr value */
175         if(key1 < key2) return -1;
176         if(key1 > key2) return 1;
177         return 0;
178 }
179
180 int reuse_id_cmp(const void* key1, const void* key2)
181 {
182         struct waiting_tcp* w1 = (struct waiting_tcp*)key1;
183         struct waiting_tcp* w2 = (struct waiting_tcp*)key2;
184         if(w1->id < w2->id)
185                 return -1;
186         if(w1->id > w2->id)
187                 return 1;
188         return 0;
189 }
190
191 /** delete waiting_tcp entry. Does not unlink from waiting list. 
192  * @param w: to delete.
193  */
194 static void
195 waiting_tcp_delete(struct waiting_tcp* w)
196 {
197         if(!w) return;
198         if(w->timer)
199                 comm_timer_delete(w->timer);
200         free(w);
201 }
202
203 /** 
204  * Pick random outgoing-interface of that family, and bind it.
205  * port set to 0 so OS picks a port number for us.
206  * if it is the ANY address, do not bind.
207  * @param pend: pending tcp structure, for storing the local address choice.
208  * @param w: tcp structure with destination address.
209  * @param s: socket fd.
210  * @return false on error, socket closed.
211  */
212 static int
213 pick_outgoing_tcp(struct pending_tcp* pend, struct waiting_tcp* w, int s)
214 {
215         struct port_if* pi = NULL;
216         int num;
217         pend->pi = NULL;
218 #ifdef INET6
219         if(addr_is_ip6(&w->addr, w->addrlen))
220                 num = w->outnet->num_ip6;
221         else
222 #endif
223                 num = w->outnet->num_ip4;
224         if(num == 0) {
225                 log_err("no TCP outgoing interfaces of family");
226                 log_addr(VERB_OPS, "for addr", &w->addr, w->addrlen);
227                 sock_close(s);
228                 return 0;
229         }
230 #ifdef INET6
231         if(addr_is_ip6(&w->addr, w->addrlen))
232                 pi = &w->outnet->ip6_ifs[ub_random_max(w->outnet->rnd, num)];
233         else
234 #endif
235                 pi = &w->outnet->ip4_ifs[ub_random_max(w->outnet->rnd, num)];
236         log_assert(pi);
237         pend->pi = pi;
238         if(addr_is_any(&pi->addr, pi->addrlen)) {
239                 /* binding to the ANY interface is for listening sockets */
240                 return 1;
241         }
242         /* set port to 0 */
243         if(addr_is_ip6(&pi->addr, pi->addrlen))
244                 ((struct sockaddr_in6*)&pi->addr)->sin6_port = 0;
245         else    ((struct sockaddr_in*)&pi->addr)->sin_port = 0;
246         if(bind(s, (struct sockaddr*)&pi->addr, pi->addrlen) != 0) {
247 #ifndef USE_WINSOCK
248 #ifdef EADDRNOTAVAIL
249                 if(!(verbosity < 4 && errno == EADDRNOTAVAIL))
250 #endif
251 #else /* USE_WINSOCK */
252                 if(!(verbosity < 4 && WSAGetLastError() == WSAEADDRNOTAVAIL))
253 #endif
254                     log_err("outgoing tcp: bind: %s", sock_strerror(errno));
255                 sock_close(s);
256                 return 0;
257         }
258         log_addr(VERB_ALGO, "tcp bound to src", &pi->addr, pi->addrlen);
259         return 1;
260 }
261
262 /** get TCP file descriptor for address, returns -1 on failure,
263  * tcp_mss is 0 or maxseg size to set for TCP packets. */
264 int
265 outnet_get_tcp_fd(struct sockaddr_storage* addr, socklen_t addrlen, int tcp_mss, int dscp)
266 {
267         int s;
268         int af;
269         char* err;
270 #if defined(SO_REUSEADDR) || defined(IP_BIND_ADDRESS_NO_PORT)
271         int on = 1;
272 #endif
273 #ifdef INET6
274         if(addr_is_ip6(addr, addrlen)){
275                 s = socket(PF_INET6, SOCK_STREAM, IPPROTO_TCP);
276                 af = AF_INET6;
277         } else {
278 #else
279         {
280 #endif
281                 af = AF_INET;
282                 s = socket(PF_INET, SOCK_STREAM, IPPROTO_TCP);
283         }
284         if(s == -1) {
285                 log_err_addr("outgoing tcp: socket", sock_strerror(errno),
286                         addr, addrlen);
287                 return -1;
288         }
289
290 #ifdef SO_REUSEADDR
291         if(setsockopt(s, SOL_SOCKET, SO_REUSEADDR, (void*)&on,
292                 (socklen_t)sizeof(on)) < 0) {
293                 verbose(VERB_ALGO, "outgoing tcp:"
294                         " setsockopt(.. SO_REUSEADDR ..) failed");
295         }
296 #endif
297
298         err = set_ip_dscp(s, af, dscp);
299         if(err != NULL) {
300                 verbose(VERB_ALGO, "outgoing tcp:"
301                         "error setting IP DiffServ codepoint on socket");
302         }
303
304         if(tcp_mss > 0) {
305 #if defined(IPPROTO_TCP) && defined(TCP_MAXSEG)
306                 if(setsockopt(s, IPPROTO_TCP, TCP_MAXSEG,
307                         (void*)&tcp_mss, (socklen_t)sizeof(tcp_mss)) < 0) {
308                         verbose(VERB_ALGO, "outgoing tcp:"
309                                 " setsockopt(.. TCP_MAXSEG ..) failed");
310                 }
311 #else
312                 verbose(VERB_ALGO, "outgoing tcp:"
313                         " setsockopt(TCP_MAXSEG) unsupported");
314 #endif /* defined(IPPROTO_TCP) && defined(TCP_MAXSEG) */
315         }
316 #ifdef IP_BIND_ADDRESS_NO_PORT
317         if(setsockopt(s, IPPROTO_IP, IP_BIND_ADDRESS_NO_PORT, (void*)&on,
318                 (socklen_t)sizeof(on)) < 0) {
319                 verbose(VERB_ALGO, "outgoing tcp:"
320                         " setsockopt(.. IP_BIND_ADDRESS_NO_PORT ..) failed");
321         }
322 #endif /* IP_BIND_ADDRESS_NO_PORT */
323         return s;
324 }
325
326 /** connect tcp connection to addr, 0 on failure */
327 int
328 outnet_tcp_connect(int s, struct sockaddr_storage* addr, socklen_t addrlen)
329 {
330         if(connect(s, (struct sockaddr*)addr, addrlen) == -1) {
331 #ifndef USE_WINSOCK
332 #ifdef EINPROGRESS
333                 if(errno != EINPROGRESS) {
334 #endif
335                         if(tcp_connect_errno_needs_log(
336                                 (struct sockaddr*)addr, addrlen))
337                                 log_err_addr("outgoing tcp: connect",
338                                         strerror(errno), addr, addrlen);
339                         close(s);
340                         return 0;
341 #ifdef EINPROGRESS
342                 }
343 #endif
344 #else /* USE_WINSOCK */
345                 if(WSAGetLastError() != WSAEINPROGRESS &&
346                         WSAGetLastError() != WSAEWOULDBLOCK) {
347                         closesocket(s);
348                         return 0;
349                 }
350 #endif
351         }
352         return 1;
353 }
354
355 /** log reuse item addr and ptr with message */
356 static void
357 log_reuse_tcp(enum verbosity_value v, const char* msg, struct reuse_tcp* reuse)
358 {
359         uint16_t port;
360         char addrbuf[128];
361         if(verbosity < v) return;
362         if(!reuse || !reuse->pending || !reuse->pending->c)
363                 return;
364         addr_to_str(&reuse->addr, reuse->addrlen, addrbuf, sizeof(addrbuf));
365         port = ntohs(((struct sockaddr_in*)&reuse->addr)->sin_port);
366         verbose(v, "%s %s#%u fd %d", msg, addrbuf, (unsigned)port,
367                 reuse->pending->c->fd);
368 }
369
370 /** pop the first element from the writewait list */
371 struct waiting_tcp*
372 reuse_write_wait_pop(struct reuse_tcp* reuse)
373 {
374         struct waiting_tcp* w = reuse->write_wait_first;
375         if(!w)
376                 return NULL;
377         log_assert(w->write_wait_queued);
378         log_assert(!w->write_wait_prev);
379         reuse->write_wait_first = w->write_wait_next;
380         if(w->write_wait_next)
381                 w->write_wait_next->write_wait_prev = NULL;
382         else    reuse->write_wait_last = NULL;
383         w->write_wait_queued = 0;
384         w->write_wait_next = NULL;
385         w->write_wait_prev = NULL;
386         return w;
387 }
388
389 /** remove the element from the writewait list */
390 void
391 reuse_write_wait_remove(struct reuse_tcp* reuse, struct waiting_tcp* w)
392 {
393         log_assert(w);
394         log_assert(w->write_wait_queued);
395         if(!w)
396                 return;
397         if(!w->write_wait_queued)
398                 return;
399         if(w->write_wait_prev)
400                 w->write_wait_prev->write_wait_next = w->write_wait_next;
401         else    reuse->write_wait_first = w->write_wait_next;
402         log_assert(!w->write_wait_prev ||
403                 w->write_wait_prev->write_wait_next != w->write_wait_prev);
404         if(w->write_wait_next)
405                 w->write_wait_next->write_wait_prev = w->write_wait_prev;
406         else    reuse->write_wait_last = w->write_wait_prev;
407         log_assert(!w->write_wait_next
408                 || w->write_wait_next->write_wait_prev != w->write_wait_next);
409         w->write_wait_queued = 0;
410         w->write_wait_next = NULL;
411         w->write_wait_prev = NULL;
412 }
413
414 /** push the element after the last on the writewait list */
415 void
416 reuse_write_wait_push_back(struct reuse_tcp* reuse, struct waiting_tcp* w)
417 {
418         if(!w) return;
419         log_assert(!w->write_wait_queued);
420         if(reuse->write_wait_last) {
421                 reuse->write_wait_last->write_wait_next = w;
422                 log_assert(reuse->write_wait_last->write_wait_next !=
423                         reuse->write_wait_last);
424                 w->write_wait_prev = reuse->write_wait_last;
425         } else {
426                 reuse->write_wait_first = w;
427                 w->write_wait_prev = NULL;
428         }
429         w->write_wait_next = NULL;
430         reuse->write_wait_last = w;
431         w->write_wait_queued = 1;
432 }
433
434 /** insert element in tree by id */
435 void
436 reuse_tree_by_id_insert(struct reuse_tcp* reuse, struct waiting_tcp* w)
437 {
438 #ifdef UNBOUND_DEBUG
439         rbnode_type* added;
440 #endif
441         log_assert(w->id_node.key == NULL);
442         w->id_node.key = w;
443 #ifdef UNBOUND_DEBUG
444         added =
445 #else
446         (void)
447 #endif
448         rbtree_insert(&reuse->tree_by_id, &w->id_node);
449         log_assert(added);  /* should have been added */
450 }
451
452 /** find element in tree by id */
453 struct waiting_tcp*
454 reuse_tcp_by_id_find(struct reuse_tcp* reuse, uint16_t id)
455 {
456         struct waiting_tcp key_w;
457         rbnode_type* n;
458         memset(&key_w, 0, sizeof(key_w));
459         key_w.id_node.key = &key_w;
460         key_w.id = id;
461         n = rbtree_search(&reuse->tree_by_id, &key_w);
462         if(!n) return NULL;
463         return (struct waiting_tcp*)n->key;
464 }
465
466 /** return ID value of rbnode in tree_by_id */
467 static uint16_t
468 tree_by_id_get_id(rbnode_type* node)
469 {
470         struct waiting_tcp* w = (struct waiting_tcp*)node->key;
471         return w->id;
472 }
473
474 /** insert into reuse tcp tree and LRU, false on failure (duplicate) */
475 int
476 reuse_tcp_insert(struct outside_network* outnet, struct pending_tcp* pend_tcp)
477 {
478         log_reuse_tcp(VERB_CLIENT, "reuse_tcp_insert", &pend_tcp->reuse);
479         if(pend_tcp->reuse.item_on_lru_list) {
480                 if(!pend_tcp->reuse.node.key)
481                         log_err("internal error: reuse_tcp_insert: "
482                                 "in lru list without key");
483                 return 1;
484         }
485         pend_tcp->reuse.node.key = &pend_tcp->reuse;
486         pend_tcp->reuse.pending = pend_tcp;
487         if(!rbtree_insert(&outnet->tcp_reuse, &pend_tcp->reuse.node)) {
488                 /* We are not in the LRU list but we are already in the
489                  * tcp_reuse tree, strange.
490                  * Continue to add ourselves to the LRU list. */
491                 log_err("internal error: reuse_tcp_insert: in lru list but "
492                         "not in the tree");
493         }
494         /* insert into LRU, first is newest */
495         pend_tcp->reuse.lru_prev = NULL;
496         if(outnet->tcp_reuse_first) {
497                 pend_tcp->reuse.lru_next = outnet->tcp_reuse_first;
498                 log_assert(pend_tcp->reuse.lru_next != &pend_tcp->reuse);
499                 outnet->tcp_reuse_first->lru_prev = &pend_tcp->reuse;
500                 log_assert(outnet->tcp_reuse_first->lru_prev !=
501                         outnet->tcp_reuse_first);
502         } else {
503                 pend_tcp->reuse.lru_next = NULL;
504                 outnet->tcp_reuse_last = &pend_tcp->reuse;
505         }
506         outnet->tcp_reuse_first = &pend_tcp->reuse;
507         pend_tcp->reuse.item_on_lru_list = 1;
508         log_assert((!outnet->tcp_reuse_first && !outnet->tcp_reuse_last) ||
509                 (outnet->tcp_reuse_first && outnet->tcp_reuse_last));
510         log_assert(outnet->tcp_reuse_first != outnet->tcp_reuse_first->lru_next &&
511                 outnet->tcp_reuse_first != outnet->tcp_reuse_first->lru_prev);
512         log_assert(outnet->tcp_reuse_last != outnet->tcp_reuse_last->lru_next &&
513                 outnet->tcp_reuse_last != outnet->tcp_reuse_last->lru_prev);
514         return 1;
515 }
516
517 /** find reuse tcp stream to destination for query, or NULL if none */
518 static struct reuse_tcp*
519 reuse_tcp_find(struct outside_network* outnet, struct sockaddr_storage* addr,
520         socklen_t addrlen, int use_ssl)
521 {
522         struct waiting_tcp key_w;
523         struct pending_tcp key_p;
524         struct comm_point c;
525         rbnode_type* result = NULL, *prev;
526         verbose(VERB_CLIENT, "reuse_tcp_find");
527         memset(&key_w, 0, sizeof(key_w));
528         memset(&key_p, 0, sizeof(key_p));
529         memset(&c, 0, sizeof(c));
530         key_p.query = &key_w;
531         key_p.c = &c;
532         key_p.reuse.pending = &key_p;
533         key_p.reuse.node.key = &key_p.reuse;
534         if(use_ssl)
535                 key_p.reuse.is_ssl = 1;
536         if(addrlen > (socklen_t)sizeof(key_p.reuse.addr))
537                 return NULL;
538         memmove(&key_p.reuse.addr, addr, addrlen);
539         key_p.reuse.addrlen = addrlen;
540
541         verbose(VERB_CLIENT, "reuse_tcp_find: num reuse streams %u",
542                 (unsigned)outnet->tcp_reuse.count);
543         if(outnet->tcp_reuse.root == NULL ||
544                 outnet->tcp_reuse.root == RBTREE_NULL)
545                 return NULL;
546         if(rbtree_find_less_equal(&outnet->tcp_reuse, &key_p.reuse,
547                 &result)) {
548                 /* exact match */
549                 /* but the key is on stack, and ptr is compared, impossible */
550                 log_assert(&key_p.reuse != (struct reuse_tcp*)result);
551                 log_assert(&key_p != ((struct reuse_tcp*)result)->pending);
552         }
553         /* not found, return null */
554
555         /* It is possible that we search for something before the first element
556          * in the tree. Replace a null pointer with the first element.
557          */
558         if (!result) {
559                 verbose(VERB_CLIENT, "reuse_tcp_find: taking first");
560                 result = rbtree_first(&outnet->tcp_reuse);
561         }
562
563         if(!result || result == RBTREE_NULL)
564                 return NULL;
565
566         /* It is possible that we got the previous address, but that the
567          * address we are looking for is in the tree. If the address we got
568          * is less than the address we are looking, then take the next entry.
569          */
570         if (reuse_cmp_addrportssl(result->key, &key_p.reuse) < 0) {
571                 verbose(VERB_CLIENT, "reuse_tcp_find: key too low");
572                 result = rbtree_next(result);
573         }
574
575         verbose(VERB_CLIENT, "reuse_tcp_find check inexact match");
576         /* inexact match, find one of possibly several connections to the
577          * same destination address, with the correct port, ssl, and
578          * also less than max number of open queries, or else, fail to open
579          * a new one */
580         /* rewind to start of sequence of same address,port,ssl */
581         prev = rbtree_previous(result);
582         while(prev && prev != RBTREE_NULL &&
583                 reuse_cmp_addrportssl(prev->key, &key_p.reuse) == 0) {
584                 result = prev;
585                 prev = rbtree_previous(result);
586         }
587
588         /* loop to find first one that has correct characteristics */
589         while(result && result != RBTREE_NULL &&
590                 reuse_cmp_addrportssl(result->key, &key_p.reuse) == 0) {
591                 if(((struct reuse_tcp*)result)->tree_by_id.count <
592                         outnet->max_reuse_tcp_queries) {
593                         /* same address, port, ssl-yes-or-no, and has
594                          * space for another query */
595                         return (struct reuse_tcp*)result;
596                 }
597                 result = rbtree_next(result);
598         }
599         return NULL;
600 }
601
602 /** use the buffer to setup writing the query */
603 static void
604 outnet_tcp_take_query_setup(int s, struct pending_tcp* pend,
605         struct waiting_tcp* w)
606 {
607         struct timeval tv;
608         verbose(VERB_CLIENT, "outnet_tcp_take_query_setup: setup packet to write "
609                 "len %d timeout %d msec",
610                 (int)w->pkt_len, w->timeout);
611         pend->c->tcp_write_pkt = w->pkt;
612         pend->c->tcp_write_pkt_len = w->pkt_len;
613         pend->c->tcp_write_and_read = 1;
614         pend->c->tcp_write_byte_count = 0;
615         pend->c->tcp_is_reading = 0;
616         comm_point_start_listening(pend->c, s, -1);
617         /* set timer on the waiting_tcp entry, this is the write timeout
618          * for the written packet.  The timer on pend->c is the timer
619          * for when there is no written packet and we have readtimeouts */
620 #ifndef S_SPLINT_S
621         tv.tv_sec = w->timeout/1000;
622         tv.tv_usec = (w->timeout%1000)*1000;
623 #endif
624         /* if the waiting_tcp was previously waiting for a buffer in the
625          * outside_network.tcpwaitlist, then the timer is reset now that
626          * we start writing it */
627         comm_timer_set(w->timer, &tv);
628 }
629
630 /** use next free buffer to service a tcp query */
631 static int
632 outnet_tcp_take_into_use(struct waiting_tcp* w)
633 {
634         struct pending_tcp* pend = w->outnet->tcp_free;
635         int s;
636         log_assert(pend);
637         log_assert(w->pkt);
638         log_assert(w->pkt_len > 0);
639         log_assert(w->addrlen > 0);
640         pend->c->tcp_do_toggle_rw = 0;
641         pend->c->tcp_do_close = 0;
642
643         /* Consistency check, if we have ssl_upstream but no sslctx, then
644          * log an error and return failure.
645          */
646         if (w->ssl_upstream && !w->outnet->sslctx) {
647                 log_err("SSL upstream requested but no SSL context");
648                 return 0;
649         }
650
651         /* open socket */
652         s = outnet_get_tcp_fd(&w->addr, w->addrlen, w->outnet->tcp_mss, w->outnet->ip_dscp);
653
654         if(s == -1)
655                 return 0;
656
657         if(!pick_outgoing_tcp(pend, w, s))
658                 return 0;
659
660         fd_set_nonblock(s);
661 #ifdef USE_OSX_MSG_FASTOPEN
662         /* API for fast open is different here. We use a connectx() function and 
663            then writes can happen as normal even using SSL.*/
664         /* connectx requires that the len be set in the sockaddr struct*/
665         struct sockaddr_in *addr_in = (struct sockaddr_in *)&w->addr;
666         addr_in->sin_len = w->addrlen;
667         sa_endpoints_t endpoints;
668         endpoints.sae_srcif = 0;
669         endpoints.sae_srcaddr = NULL;
670         endpoints.sae_srcaddrlen = 0;
671         endpoints.sae_dstaddr = (struct sockaddr *)&w->addr;
672         endpoints.sae_dstaddrlen = w->addrlen;
673         if (connectx(s, &endpoints, SAE_ASSOCID_ANY,  
674                      CONNECT_DATA_IDEMPOTENT | CONNECT_RESUME_ON_READ_WRITE,
675                      NULL, 0, NULL, NULL) == -1) {
676                 /* if fails, failover to connect for OSX 10.10 */
677 #ifdef EINPROGRESS
678                 if(errno != EINPROGRESS) {
679 #else
680                 if(1) {
681 #endif
682                         if(connect(s, (struct sockaddr*)&w->addr, w->addrlen) == -1) {
683 #else /* USE_OSX_MSG_FASTOPEN*/
684 #ifdef USE_MSG_FASTOPEN
685         pend->c->tcp_do_fastopen = 1;
686         /* Only do TFO for TCP in which case no connect() is required here.
687            Don't combine client TFO with SSL, since OpenSSL can't 
688            currently support doing a handshake on fd that already isn't connected*/
689         if (w->outnet->sslctx && w->ssl_upstream) {
690                 if(connect(s, (struct sockaddr*)&w->addr, w->addrlen) == -1) {
691 #else /* USE_MSG_FASTOPEN*/
692         if(connect(s, (struct sockaddr*)&w->addr, w->addrlen) == -1) {
693 #endif /* USE_MSG_FASTOPEN*/
694 #endif /* USE_OSX_MSG_FASTOPEN*/
695 #ifndef USE_WINSOCK
696 #ifdef EINPROGRESS
697                 if(errno != EINPROGRESS) {
698 #else
699                 if(1) {
700 #endif
701                         if(tcp_connect_errno_needs_log(
702                                 (struct sockaddr*)&w->addr, w->addrlen))
703                                 log_err_addr("outgoing tcp: connect",
704                                         strerror(errno), &w->addr, w->addrlen);
705                         close(s);
706 #else /* USE_WINSOCK */
707                 if(WSAGetLastError() != WSAEINPROGRESS &&
708                         WSAGetLastError() != WSAEWOULDBLOCK) {
709                         closesocket(s);
710 #endif
711                         return 0;
712                 }
713         }
714 #ifdef USE_MSG_FASTOPEN
715         }
716 #endif /* USE_MSG_FASTOPEN */
717 #ifdef USE_OSX_MSG_FASTOPEN
718                 }
719         }
720 #endif /* USE_OSX_MSG_FASTOPEN */
721         if(w->outnet->sslctx && w->ssl_upstream) {
722                 pend->c->ssl = outgoing_ssl_fd(w->outnet->sslctx, s);
723                 if(!pend->c->ssl) {
724                         pend->c->fd = s;
725                         comm_point_close(pend->c);
726                         return 0;
727                 }
728                 verbose(VERB_ALGO, "the query is using TLS encryption, for %s",
729                         (w->tls_auth_name?w->tls_auth_name:"an unauthenticated connection"));
730 #ifdef USE_WINSOCK
731                 comm_point_tcp_win_bio_cb(pend->c, pend->c->ssl);
732 #endif
733                 pend->c->ssl_shake_state = comm_ssl_shake_write;
734                 if(!set_auth_name_on_ssl(pend->c->ssl, w->tls_auth_name,
735                         w->outnet->tls_use_sni)) {
736                         pend->c->fd = s;
737 #ifdef HAVE_SSL
738                         SSL_free(pend->c->ssl);
739 #endif
740                         pend->c->ssl = NULL;
741                         comm_point_close(pend->c);
742                         return 0;
743                 }
744         }
745         w->next_waiting = (void*)pend;
746         w->outnet->num_tcp_outgoing++;
747         w->outnet->tcp_free = pend->next_free;
748         pend->next_free = NULL;
749         pend->query = w;
750         pend->reuse.outnet = w->outnet;
751         pend->c->repinfo.remote_addrlen = w->addrlen;
752         pend->c->tcp_more_read_again = &pend->reuse.cp_more_read_again;
753         pend->c->tcp_more_write_again = &pend->reuse.cp_more_write_again;
754         pend->reuse.cp_more_read_again = 0;
755         pend->reuse.cp_more_write_again = 0;
756         memcpy(&pend->c->repinfo.remote_addr, &w->addr, w->addrlen);
757         pend->reuse.pending = pend;
758
759         /* Remove from tree in case the is_ssl will be different and causes the
760          * identity of the reuse_tcp to change; could result in nodes not being
761          * deleted from the tree (because the new identity does not match the
762          * previous node) but their ->key would be changed to NULL. */
763         if(pend->reuse.node.key)
764                 reuse_tcp_remove_tree_list(w->outnet, &pend->reuse);
765
766         if(pend->c->ssl)
767                 pend->reuse.is_ssl = 1;
768         else    pend->reuse.is_ssl = 0;
769         /* insert in reuse by address tree if not already inserted there */
770         (void)reuse_tcp_insert(w->outnet, pend);
771         reuse_tree_by_id_insert(&pend->reuse, w);
772         outnet_tcp_take_query_setup(s, pend, w);
773         return 1;
774 }
775
776 /** Touch the lru of a reuse_tcp element, it is in use.
777  * This moves it to the front of the list, where it is not likely to
778  * be closed.  Items at the back of the list are closed to make space. */
779 void
780 reuse_tcp_lru_touch(struct outside_network* outnet, struct reuse_tcp* reuse)
781 {
782         if(!reuse->item_on_lru_list) {
783                 log_err("internal error: we need to touch the lru_list but item not in list");
784                 return; /* not on the list, no lru to modify */
785         }
786         log_assert(reuse->lru_prev ||
787                 (!reuse->lru_prev && outnet->tcp_reuse_first == reuse));
788         if(!reuse->lru_prev)
789                 return; /* already first in the list */
790         /* remove at current position */
791         /* since it is not first, there is a previous element */
792         reuse->lru_prev->lru_next = reuse->lru_next;
793         log_assert(reuse->lru_prev->lru_next != reuse->lru_prev);
794         if(reuse->lru_next)
795                 reuse->lru_next->lru_prev = reuse->lru_prev;
796         else    outnet->tcp_reuse_last = reuse->lru_prev;
797         log_assert(!reuse->lru_next || reuse->lru_next->lru_prev != reuse->lru_next);
798         log_assert(outnet->tcp_reuse_last != outnet->tcp_reuse_last->lru_next &&
799                 outnet->tcp_reuse_last != outnet->tcp_reuse_last->lru_prev);
800         /* insert at the front */
801         reuse->lru_prev = NULL;
802         reuse->lru_next = outnet->tcp_reuse_first;
803         if(outnet->tcp_reuse_first) {
804                 outnet->tcp_reuse_first->lru_prev = reuse;
805         }
806         log_assert(reuse->lru_next != reuse);
807         /* since it is not first, it is not the only element and
808          * lru_next is thus not NULL and thus reuse is now not the last in
809          * the list, so outnet->tcp_reuse_last does not need to be modified */
810         outnet->tcp_reuse_first = reuse;
811         log_assert(outnet->tcp_reuse_first != outnet->tcp_reuse_first->lru_next &&
812                 outnet->tcp_reuse_first != outnet->tcp_reuse_first->lru_prev);
813         log_assert((!outnet->tcp_reuse_first && !outnet->tcp_reuse_last) ||
814                 (outnet->tcp_reuse_first && outnet->tcp_reuse_last));
815 }
816
817 /** Snip the last reuse_tcp element off of the LRU list */
818 struct reuse_tcp*
819 reuse_tcp_lru_snip(struct outside_network* outnet)
820 {
821         struct reuse_tcp* reuse = outnet->tcp_reuse_last;
822         if(!reuse) return NULL;
823         /* snip off of LRU */
824         log_assert(reuse->lru_next == NULL);
825         if(reuse->lru_prev) {
826                 outnet->tcp_reuse_last = reuse->lru_prev;
827                 reuse->lru_prev->lru_next = NULL;
828         } else {
829                 outnet->tcp_reuse_last = NULL;
830                 outnet->tcp_reuse_first = NULL;
831         }
832         log_assert((!outnet->tcp_reuse_first && !outnet->tcp_reuse_last) ||
833                 (outnet->tcp_reuse_first && outnet->tcp_reuse_last));
834         reuse->item_on_lru_list = 0;
835         reuse->lru_next = NULL;
836         reuse->lru_prev = NULL;
837         return reuse;
838 }
839
840 /** remove waiting tcp from the outnet waiting list */
841 void
842 outnet_waiting_tcp_list_remove(struct outside_network* outnet, struct waiting_tcp* w)
843 {
844         struct waiting_tcp* p = outnet->tcp_wait_first, *prev = NULL;
845         w->on_tcp_waiting_list = 0;
846         while(p) {
847                 if(p == w) {
848                         /* remove w */
849                         if(prev)
850                                 prev->next_waiting = w->next_waiting;
851                         else    outnet->tcp_wait_first = w->next_waiting;
852                         if(outnet->tcp_wait_last == w)
853                                 outnet->tcp_wait_last = prev;
854                         w->next_waiting = NULL;
855                         return;
856                 }
857                 prev = p;
858                 p = p->next_waiting;
859         }
860         /* outnet_waiting_tcp_list_remove is currently called only with items
861          * that are already in the waiting list. */
862         log_assert(0);
863 }
864
865 /** pop the first waiting tcp from the outnet waiting list */
866 struct waiting_tcp*
867 outnet_waiting_tcp_list_pop(struct outside_network* outnet)
868 {
869         struct waiting_tcp* w = outnet->tcp_wait_first;
870         if(!outnet->tcp_wait_first) return NULL;
871         log_assert(w->on_tcp_waiting_list);
872         outnet->tcp_wait_first = w->next_waiting;
873         if(outnet->tcp_wait_last == w)
874                 outnet->tcp_wait_last = NULL;
875         w->on_tcp_waiting_list = 0;
876         w->next_waiting = NULL;
877         return w;
878 }
879
880 /** add waiting_tcp element to the outnet tcp waiting list */
881 void
882 outnet_waiting_tcp_list_add(struct outside_network* outnet,
883         struct waiting_tcp* w, int set_timer)
884 {
885         struct timeval tv;
886         log_assert(!w->on_tcp_waiting_list);
887         if(w->on_tcp_waiting_list)
888                 return;
889         w->next_waiting = NULL;
890         if(outnet->tcp_wait_last)
891                 outnet->tcp_wait_last->next_waiting = w;
892         else    outnet->tcp_wait_first = w;
893         outnet->tcp_wait_last = w;
894         w->on_tcp_waiting_list = 1;
895         if(set_timer) {
896 #ifndef S_SPLINT_S
897                 tv.tv_sec = w->timeout/1000;
898                 tv.tv_usec = (w->timeout%1000)*1000;
899 #endif
900                 comm_timer_set(w->timer, &tv);
901         }
902 }
903
904 /** add waiting_tcp element as first to the outnet tcp waiting list */
905 void
906 outnet_waiting_tcp_list_add_first(struct outside_network* outnet,
907         struct waiting_tcp* w, int reset_timer)
908 {
909         struct timeval tv;
910         log_assert(!w->on_tcp_waiting_list);
911         if(w->on_tcp_waiting_list)
912                 return;
913         w->next_waiting = outnet->tcp_wait_first;
914         log_assert(w->next_waiting != w);
915         if(!outnet->tcp_wait_last)
916                 outnet->tcp_wait_last = w;
917         outnet->tcp_wait_first = w;
918         w->on_tcp_waiting_list = 1;
919         if(reset_timer) {
920 #ifndef S_SPLINT_S
921                 tv.tv_sec = w->timeout/1000;
922                 tv.tv_usec = (w->timeout%1000)*1000;
923 #endif
924                 comm_timer_set(w->timer, &tv);
925         }
926         log_assert(
927                 (!outnet->tcp_reuse_first && !outnet->tcp_reuse_last) ||
928                 (outnet->tcp_reuse_first && outnet->tcp_reuse_last));
929 }
930
931 /** call callback on waiting_tcp, if not NULL */
932 static void
933 waiting_tcp_callback(struct waiting_tcp* w, struct comm_point* c, int error,
934         struct comm_reply* reply_info)
935 {
936         if(w && w->cb) {
937                 fptr_ok(fptr_whitelist_pending_tcp(w->cb));
938                 (void)(*w->cb)(c, w->cb_arg, error, reply_info);
939         }
940 }
941
942 /** see if buffers can be used to service TCP queries */
943 static void
944 use_free_buffer(struct outside_network* outnet)
945 {
946         struct waiting_tcp* w;
947         while(outnet->tcp_wait_first && !outnet->want_to_quit) {
948 #ifdef USE_DNSTAP
949                 struct pending_tcp* pend_tcp = NULL;
950 #endif
951                 struct reuse_tcp* reuse = NULL;
952                 w = outnet_waiting_tcp_list_pop(outnet);
953                 log_assert(
954                         (!outnet->tcp_reuse_first && !outnet->tcp_reuse_last) ||
955                         (outnet->tcp_reuse_first && outnet->tcp_reuse_last));
956                 reuse = reuse_tcp_find(outnet, &w->addr, w->addrlen,
957                         w->ssl_upstream);
958                 /* re-select an ID when moving to a new TCP buffer */
959                 w->id = tcp_select_id(outnet, reuse);
960                 LDNS_ID_SET(w->pkt, w->id);
961                 if(reuse) {
962                         log_reuse_tcp(VERB_CLIENT, "use free buffer for waiting tcp: "
963                                 "found reuse", reuse);
964 #ifdef USE_DNSTAP
965                         pend_tcp = reuse->pending;
966 #endif
967                         reuse_tcp_lru_touch(outnet, reuse);
968                         comm_timer_disable(w->timer);
969                         w->next_waiting = (void*)reuse->pending;
970                         reuse_tree_by_id_insert(reuse, w);
971                         if(reuse->pending->query) {
972                                 /* on the write wait list */
973                                 reuse_write_wait_push_back(reuse, w);
974                         } else {
975                                 /* write straight away */
976                                 /* stop the timer on read of the fd */
977                                 comm_point_stop_listening(reuse->pending->c);
978                                 reuse->pending->query = w;
979                                 outnet_tcp_take_query_setup(
980                                         reuse->pending->c->fd, reuse->pending,
981                                         w);
982                         }
983                 } else if(outnet->tcp_free) {
984                         struct pending_tcp* pend = w->outnet->tcp_free;
985                         rbtree_init(&pend->reuse.tree_by_id, reuse_id_cmp);
986                         pend->reuse.pending = pend;
987                         memcpy(&pend->reuse.addr, &w->addr, w->addrlen);
988                         pend->reuse.addrlen = w->addrlen;
989                         if(!outnet_tcp_take_into_use(w)) {
990                                 waiting_tcp_callback(w, NULL, NETEVENT_CLOSED,
991                                         NULL);
992                                 waiting_tcp_delete(w);
993 #ifdef USE_DNSTAP
994                                 w = NULL;
995 #endif
996                         }
997 #ifdef USE_DNSTAP
998                         pend_tcp = pend;
999 #endif
1000                 } else {
1001                         /* no reuse and no free buffer, put back at the start */
1002                         outnet_waiting_tcp_list_add_first(outnet, w, 0);
1003                         break;
1004                 }
1005 #ifdef USE_DNSTAP
1006                 if(outnet->dtenv && pend_tcp && w && w->sq &&
1007                         (outnet->dtenv->log_resolver_query_messages ||
1008                         outnet->dtenv->log_forwarder_query_messages)) {
1009                         sldns_buffer tmp;
1010                         sldns_buffer_init_frm_data(&tmp, w->pkt, w->pkt_len);
1011                         dt_msg_send_outside_query(outnet->dtenv, &w->sq->addr,
1012                                 &pend_tcp->pi->addr, comm_tcp, w->sq->zone,
1013                                 w->sq->zonelen, &tmp);
1014                 }
1015 #endif
1016         }
1017 }
1018
1019 /** delete element from tree by id */
1020 static void
1021 reuse_tree_by_id_delete(struct reuse_tcp* reuse, struct waiting_tcp* w)
1022 {
1023 #ifdef UNBOUND_DEBUG
1024         rbnode_type* rem;
1025 #endif
1026         log_assert(w->id_node.key != NULL);
1027 #ifdef UNBOUND_DEBUG
1028         rem =
1029 #else
1030         (void)
1031 #endif
1032         rbtree_delete(&reuse->tree_by_id, w);
1033         log_assert(rem);  /* should have been there */
1034         w->id_node.key = NULL;
1035 }
1036
1037 /** move writewait list to go for another connection. */
1038 static void
1039 reuse_move_writewait_away(struct outside_network* outnet,
1040         struct pending_tcp* pend)
1041 {
1042         /* the writewait list has not been written yet, so if the
1043          * stream was closed, they have not actually been failed, only
1044          * the queries written.  Other queries can get written to another
1045          * stream.  For upstreams that do not support multiple queries
1046          * and answers, the stream can get closed, and then the queries
1047          * can get written on a new socket */
1048         struct waiting_tcp* w;
1049         if(pend->query && pend->query->error_count == 0 &&
1050                 pend->c->tcp_write_pkt == pend->query->pkt &&
1051                 pend->c->tcp_write_pkt_len == pend->query->pkt_len) {
1052                 /* since the current query is not written, it can also
1053                  * move to a free buffer */
1054                 if(verbosity >= VERB_CLIENT && pend->query->pkt_len > 12+2+2 &&
1055                         LDNS_QDCOUNT(pend->query->pkt) > 0 &&
1056                         dname_valid(pend->query->pkt+12, pend->query->pkt_len-12)) {
1057                         char buf[LDNS_MAX_DOMAINLEN+1];
1058                         dname_str(pend->query->pkt+12, buf);
1059                         verbose(VERB_CLIENT, "reuse_move_writewait_away current %s %d bytes were written",
1060                                 buf, (int)pend->c->tcp_write_byte_count);
1061                 }
1062                 pend->c->tcp_write_pkt = NULL;
1063                 pend->c->tcp_write_pkt_len = 0;
1064                 pend->c->tcp_write_and_read = 0;
1065                 pend->reuse.cp_more_read_again = 0;
1066                 pend->reuse.cp_more_write_again = 0;
1067                 pend->c->tcp_is_reading = 1;
1068                 w = pend->query;
1069                 pend->query = NULL;
1070                 /* increase error count, so that if the next socket fails too
1071                  * the server selection is run again with this query failed
1072                  * and it can select a different server (if possible), or
1073                  * fail the query */
1074                 w->error_count ++;
1075                 reuse_tree_by_id_delete(&pend->reuse, w);
1076                 outnet_waiting_tcp_list_add(outnet, w, 1);
1077         }
1078         while((w = reuse_write_wait_pop(&pend->reuse)) != NULL) {
1079                 if(verbosity >= VERB_CLIENT && w->pkt_len > 12+2+2 &&
1080                         LDNS_QDCOUNT(w->pkt) > 0 &&
1081                         dname_valid(w->pkt+12, w->pkt_len-12)) {
1082                         char buf[LDNS_MAX_DOMAINLEN+1];
1083                         dname_str(w->pkt+12, buf);
1084                         verbose(VERB_CLIENT, "reuse_move_writewait_away item %s", buf);
1085                 }
1086                 reuse_tree_by_id_delete(&pend->reuse, w);
1087                 outnet_waiting_tcp_list_add(outnet, w, 1);
1088         }
1089 }
1090
1091 /** remove reused element from tree and lru list */
1092 void
1093 reuse_tcp_remove_tree_list(struct outside_network* outnet,
1094         struct reuse_tcp* reuse)
1095 {
1096         verbose(VERB_CLIENT, "reuse_tcp_remove_tree_list");
1097         if(reuse->node.key) {
1098                 /* delete it from reuse tree */
1099                 if(!rbtree_delete(&outnet->tcp_reuse, reuse)) {
1100                         /* should not be possible, it should be there */
1101                         char buf[256];
1102                         addr_to_str(&reuse->addr, reuse->addrlen, buf,
1103                                 sizeof(buf));
1104                         log_err("reuse tcp delete: node not present, internal error, %s ssl %d lru %d", buf, reuse->is_ssl, reuse->item_on_lru_list);
1105                 }
1106                 reuse->node.key = NULL;
1107                 /* defend against loops on broken tree by zeroing the
1108                  * rbnode structure */
1109                 memset(&reuse->node, 0, sizeof(reuse->node));
1110         }
1111         /* delete from reuse list */
1112         if(reuse->item_on_lru_list) {
1113                 if(reuse->lru_prev) {
1114                         /* assert that members of the lru list are waiting
1115                          * and thus have a pending pointer to the struct */
1116                         log_assert(reuse->lru_prev->pending);
1117                         reuse->lru_prev->lru_next = reuse->lru_next;
1118                         log_assert(reuse->lru_prev->lru_next != reuse->lru_prev);
1119                 } else {
1120                         log_assert(!reuse->lru_next || reuse->lru_next->pending);
1121                         outnet->tcp_reuse_first = reuse->lru_next;
1122                         log_assert(!outnet->tcp_reuse_first ||
1123                                 (outnet->tcp_reuse_first !=
1124                                  outnet->tcp_reuse_first->lru_next &&
1125                                  outnet->tcp_reuse_first !=
1126                                  outnet->tcp_reuse_first->lru_prev));
1127                 }
1128                 if(reuse->lru_next) {
1129                         /* assert that members of the lru list are waiting
1130                          * and thus have a pending pointer to the struct */
1131                         log_assert(reuse->lru_next->pending);
1132                         reuse->lru_next->lru_prev = reuse->lru_prev;
1133                         log_assert(reuse->lru_next->lru_prev != reuse->lru_next);
1134                 } else {
1135                         log_assert(!reuse->lru_prev || reuse->lru_prev->pending);
1136                         outnet->tcp_reuse_last = reuse->lru_prev;
1137                         log_assert(!outnet->tcp_reuse_last ||
1138                                 (outnet->tcp_reuse_last !=
1139                                  outnet->tcp_reuse_last->lru_next &&
1140                                  outnet->tcp_reuse_last !=
1141                                  outnet->tcp_reuse_last->lru_prev));
1142                 }
1143                 log_assert((!outnet->tcp_reuse_first && !outnet->tcp_reuse_last) ||
1144                         (outnet->tcp_reuse_first && outnet->tcp_reuse_last));
1145                 reuse->item_on_lru_list = 0;
1146                 reuse->lru_next = NULL;
1147                 reuse->lru_prev = NULL;
1148         }
1149         reuse->pending = NULL;
1150 }
1151
1152 /** helper function that deletes an element from the tree of readwait
1153  * elements in tcp reuse structure */
1154 static void reuse_del_readwait_elem(rbnode_type* node, void* ATTR_UNUSED(arg))
1155 {
1156         struct waiting_tcp* w = (struct waiting_tcp*)node->key;
1157         waiting_tcp_delete(w);
1158 }
1159
1160 /** delete readwait waiting_tcp elements, deletes the elements in the list */
1161 void reuse_del_readwait(rbtree_type* tree_by_id)
1162 {
1163         if(tree_by_id->root == NULL ||
1164                 tree_by_id->root == RBTREE_NULL)
1165                 return;
1166         traverse_postorder(tree_by_id, &reuse_del_readwait_elem, NULL);
1167         rbtree_init(tree_by_id, reuse_id_cmp);
1168 }
1169
1170 /** decommission a tcp buffer, closes commpoint and frees waiting_tcp entry */
1171 static void
1172 decommission_pending_tcp(struct outside_network* outnet, 
1173         struct pending_tcp* pend)
1174 {
1175         verbose(VERB_CLIENT, "decommission_pending_tcp");
1176         /* A certain code path can lead here twice for the same pending_tcp
1177          * creating a loop in the free pending_tcp list. */
1178         if(outnet->tcp_free != pend) {
1179                 pend->next_free = outnet->tcp_free;
1180                 outnet->tcp_free = pend;
1181         }
1182         if(pend->reuse.node.key) {
1183                 /* needs unlink from the reuse tree to get deleted */
1184                 reuse_tcp_remove_tree_list(outnet, &pend->reuse);
1185         }
1186         /* free SSL structure after remove from outnet tcp reuse tree,
1187          * because the c->ssl null or not is used for sorting in the tree */
1188         if(pend->c->ssl) {
1189 #ifdef HAVE_SSL
1190                 SSL_shutdown(pend->c->ssl);
1191                 SSL_free(pend->c->ssl);
1192                 pend->c->ssl = NULL;
1193 #endif
1194         }
1195         comm_point_close(pend->c);
1196         pend->reuse.cp_more_read_again = 0;
1197         pend->reuse.cp_more_write_again = 0;
1198         /* unlink the query and writewait list, it is part of the tree
1199          * nodes and is deleted */
1200         pend->query = NULL;
1201         pend->reuse.write_wait_first = NULL;
1202         pend->reuse.write_wait_last = NULL;
1203         reuse_del_readwait(&pend->reuse.tree_by_id);
1204 }
1205
1206 /** perform failure callbacks for waiting queries in reuse read rbtree */
1207 static void reuse_cb_readwait_for_failure(rbtree_type* tree_by_id, int err)
1208 {
1209         rbnode_type* node;
1210         if(tree_by_id->root == NULL ||
1211                 tree_by_id->root == RBTREE_NULL)
1212                 return;
1213         node = rbtree_first(tree_by_id);
1214         while(node && node != RBTREE_NULL) {
1215                 struct waiting_tcp* w = (struct waiting_tcp*)node->key;
1216                 waiting_tcp_callback(w, NULL, err, NULL);
1217                 node = rbtree_next(node);
1218         }
1219 }
1220
1221 /** mark the entry for being in the cb_and_decommission stage */
1222 static void mark_for_cb_and_decommission(rbnode_type* node,
1223         void* ATTR_UNUSED(arg))
1224 {
1225         struct waiting_tcp* w = (struct waiting_tcp*)node->key;
1226         /* Mark the waiting_tcp to signal later code (serviced_delete) that
1227          * this item is part of the backed up tree_by_id and will be deleted
1228          * later. */
1229         w->in_cb_and_decommission = 1;
1230         /* Mark the serviced_query for deletion so that later code through
1231          * callbacks (iter_clear .. outnet_serviced_query_stop) won't
1232          * prematurely delete it. */
1233         if(w->cb)
1234                 ((struct serviced_query*)w->cb_arg)->to_be_deleted = 1;
1235 }
1236
1237 /** perform callbacks for failure and also decommission pending tcp.
1238  * the callbacks remove references in sq->pending to the waiting_tcp
1239  * members of the tree_by_id in the pending tcp.  The pending_tcp is
1240  * removed before the callbacks, so that the callbacks do not modify
1241  * the pending_tcp due to its reference in the outside_network reuse tree */
1242 static void reuse_cb_and_decommission(struct outside_network* outnet,
1243         struct pending_tcp* pend, int error)
1244 {
1245         rbtree_type store;
1246         store = pend->reuse.tree_by_id;
1247         pend->query = NULL;
1248         rbtree_init(&pend->reuse.tree_by_id, reuse_id_cmp);
1249         pend->reuse.write_wait_first = NULL;
1250         pend->reuse.write_wait_last = NULL;
1251         decommission_pending_tcp(outnet, pend);
1252         if(store.root != NULL && store.root != RBTREE_NULL) {
1253                 traverse_postorder(&store, &mark_for_cb_and_decommission, NULL);
1254         }
1255         reuse_cb_readwait_for_failure(&store, error);
1256         reuse_del_readwait(&store);
1257 }
1258
1259 /** set timeout on tcp fd and setup read event to catch incoming dns msgs */
1260 static void
1261 reuse_tcp_setup_timeout(struct pending_tcp* pend_tcp, int tcp_reuse_timeout)
1262 {
1263         log_reuse_tcp(VERB_CLIENT, "reuse_tcp_setup_timeout", &pend_tcp->reuse);
1264         comm_point_start_listening(pend_tcp->c, -1, tcp_reuse_timeout);
1265 }
1266
1267 /** set timeout on tcp fd and setup read event to catch incoming dns msgs */
1268 static void
1269 reuse_tcp_setup_read_and_timeout(struct pending_tcp* pend_tcp, int tcp_reuse_timeout)
1270 {
1271         log_reuse_tcp(VERB_CLIENT, "reuse_tcp_setup_readtimeout", &pend_tcp->reuse);
1272         sldns_buffer_clear(pend_tcp->c->buffer);
1273         pend_tcp->c->tcp_is_reading = 1;
1274         pend_tcp->c->tcp_byte_count = 0;
1275         comm_point_stop_listening(pend_tcp->c);
1276         comm_point_start_listening(pend_tcp->c, -1, tcp_reuse_timeout);
1277 }
1278
1279 int 
1280 outnet_tcp_cb(struct comm_point* c, void* arg, int error,
1281         struct comm_reply *reply_info)
1282 {
1283         struct pending_tcp* pend = (struct pending_tcp*)arg;
1284         struct outside_network* outnet = pend->reuse.outnet;
1285         struct waiting_tcp* w = NULL;
1286         log_assert(pend->reuse.item_on_lru_list && pend->reuse.node.key);
1287         verbose(VERB_ALGO, "outnettcp cb");
1288         if(error == NETEVENT_TIMEOUT) {
1289                 if(pend->c->tcp_write_and_read) {
1290                         verbose(VERB_QUERY, "outnettcp got tcp timeout "
1291                                 "for read, ignored because write underway");
1292                         /* if we are writing, ignore readtimer, wait for write timer
1293                          * or write is done */
1294                         return 0;
1295                 } else {
1296                         verbose(VERB_QUERY, "outnettcp got tcp timeout %s",
1297                                 (pend->reuse.tree_by_id.count?"for reading pkt":
1298                                 "for keepalive for reuse"));
1299                 }
1300                 /* must be timeout for reading or keepalive reuse,
1301                  * close it. */
1302                 reuse_tcp_remove_tree_list(outnet, &pend->reuse);
1303         } else if(error == NETEVENT_PKT_WRITTEN) {
1304                 /* the packet we want to write has been written. */
1305                 verbose(VERB_ALGO, "outnet tcp pkt was written event");
1306                 log_assert(c == pend->c);
1307                 log_assert(pend->query->pkt == pend->c->tcp_write_pkt);
1308                 log_assert(pend->query->pkt_len == pend->c->tcp_write_pkt_len);
1309                 pend->c->tcp_write_pkt = NULL;
1310                 pend->c->tcp_write_pkt_len = 0;
1311                 /* the pend.query is already in tree_by_id */
1312                 log_assert(pend->query->id_node.key);
1313                 pend->query = NULL;
1314                 /* setup to write next packet or setup read timeout */
1315                 if(pend->reuse.write_wait_first) {
1316                         verbose(VERB_ALGO, "outnet tcp setup next pkt");
1317                         /* we can write it straight away perhaps, set flag
1318                          * because this callback called after a tcp write
1319                          * succeeded and likely more buffer space is available
1320                          * and we can write some more. */
1321                         pend->reuse.cp_more_write_again = 1;
1322                         pend->query = reuse_write_wait_pop(&pend->reuse);
1323                         comm_point_stop_listening(pend->c);
1324                         outnet_tcp_take_query_setup(pend->c->fd, pend,
1325                                 pend->query);
1326                 } else {
1327                         verbose(VERB_ALGO, "outnet tcp writes done, wait");
1328                         pend->c->tcp_write_and_read = 0;
1329                         pend->reuse.cp_more_read_again = 0;
1330                         pend->reuse.cp_more_write_again = 0;
1331                         pend->c->tcp_is_reading = 1;
1332                         comm_point_stop_listening(pend->c);
1333                         reuse_tcp_setup_timeout(pend, outnet->tcp_reuse_timeout);
1334                 }
1335                 return 0;
1336         } else if(error != NETEVENT_NOERROR) {
1337                 verbose(VERB_QUERY, "outnettcp got tcp error %d", error);
1338                 reuse_move_writewait_away(outnet, pend);
1339                 /* pass error below and exit */
1340         } else {
1341                 /* check ID */
1342                 if(sldns_buffer_limit(c->buffer) < sizeof(uint16_t)) {
1343                         log_addr(VERB_QUERY, 
1344                                 "outnettcp: bad ID in reply, too short, from:",
1345                                 &pend->reuse.addr, pend->reuse.addrlen);
1346                         error = NETEVENT_CLOSED;
1347                 } else {
1348                         uint16_t id = LDNS_ID_WIRE(sldns_buffer_begin(
1349                                 c->buffer));
1350                         /* find the query the reply is for */
1351                         w = reuse_tcp_by_id_find(&pend->reuse, id);
1352                         /* Make sure that the reply we got is at least for a
1353                          * sent query with the same ID; the waiting_tcp that
1354                          * gets a reply is assumed to not be waiting to be
1355                          * sent. */
1356                         if(w && (w->on_tcp_waiting_list || w->write_wait_queued))
1357                                 w = NULL;
1358                 }
1359         }
1360         if(error == NETEVENT_NOERROR && !w) {
1361                 /* no struct waiting found in tree, no reply to call */
1362                 log_addr(VERB_QUERY, "outnettcp: bad ID in reply, from:",
1363                         &pend->reuse.addr, pend->reuse.addrlen);
1364                 error = NETEVENT_CLOSED;
1365         }
1366         if(error == NETEVENT_NOERROR) {
1367                 /* add to reuse tree so it can be reused, if not a failure.
1368                  * This is possible if the state machine wants to make a tcp
1369                  * query again to the same destination. */
1370                 if(outnet->tcp_reuse.count < outnet->tcp_reuse_max) {
1371                         (void)reuse_tcp_insert(outnet, pend);
1372                 }
1373         }
1374         if(w) {
1375                 log_assert(!w->on_tcp_waiting_list);
1376                 log_assert(!w->write_wait_queued);
1377                 reuse_tree_by_id_delete(&pend->reuse, w);
1378                 verbose(VERB_CLIENT, "outnet tcp callback query err %d buflen %d",
1379                         error, (int)sldns_buffer_limit(c->buffer));
1380                 waiting_tcp_callback(w, c, error, reply_info);
1381                 waiting_tcp_delete(w);
1382         }
1383         verbose(VERB_CLIENT, "outnet_tcp_cb reuse after cb");
1384         if(error == NETEVENT_NOERROR && pend->reuse.node.key) {
1385                 verbose(VERB_CLIENT, "outnet_tcp_cb reuse after cb: keep it");
1386                 /* it is in the reuse_tcp tree, with other queries, or
1387                  * on the empty list. do not decommission it */
1388                 /* if there are more outstanding queries, we could try to
1389                  * read again, to see if it is on the input,
1390                  * because this callback called after a successful read
1391                  * and there could be more bytes to read on the input */
1392                 if(pend->reuse.tree_by_id.count != 0)
1393                         pend->reuse.cp_more_read_again = 1;
1394                 reuse_tcp_setup_read_and_timeout(pend, outnet->tcp_reuse_timeout);
1395                 return 0;
1396         }
1397         verbose(VERB_CLIENT, "outnet_tcp_cb reuse after cb: decommission it");
1398         /* no queries on it, no space to keep it. or timeout or closed due
1399          * to error.  Close it */
1400         reuse_cb_and_decommission(outnet, pend, (error==NETEVENT_TIMEOUT?
1401                 NETEVENT_TIMEOUT:NETEVENT_CLOSED));
1402         use_free_buffer(outnet);
1403         return 0;
1404 }
1405
1406 /** lower use count on pc, see if it can be closed */
1407 static void
1408 portcomm_loweruse(struct outside_network* outnet, struct port_comm* pc)
1409 {
1410         struct port_if* pif;
1411         pc->num_outstanding--;
1412         if(pc->num_outstanding > 0) {
1413                 return;
1414         }
1415         /* close it and replace in unused list */
1416         verbose(VERB_ALGO, "close of port %d", pc->number);
1417         comm_point_close(pc->cp);
1418         pif = pc->pif;
1419         log_assert(pif->inuse > 0);
1420 #ifndef DISABLE_EXPLICIT_PORT_RANDOMISATION
1421         pif->avail_ports[pif->avail_total - pif->inuse] = pc->number;
1422 #endif
1423         pif->inuse--;
1424         pif->out[pc->index] = pif->out[pif->inuse];
1425         pif->out[pc->index]->index = pc->index;
1426         pc->next = outnet->unused_fds;
1427         outnet->unused_fds = pc;
1428 }
1429
1430 /** try to send waiting UDP queries */
1431 static void
1432 outnet_send_wait_udp(struct outside_network* outnet)
1433 {
1434         struct pending* pend;
1435         /* process waiting queries */
1436         while(outnet->udp_wait_first && outnet->unused_fds
1437                 && !outnet->want_to_quit) {
1438                 pend = outnet->udp_wait_first;
1439                 outnet->udp_wait_first = pend->next_waiting;
1440                 if(!pend->next_waiting) outnet->udp_wait_last = NULL;
1441                 sldns_buffer_clear(outnet->udp_buff);
1442                 sldns_buffer_write(outnet->udp_buff, pend->pkt, pend->pkt_len);
1443                 sldns_buffer_flip(outnet->udp_buff);
1444                 free(pend->pkt); /* freeing now makes get_mem correct */
1445                 pend->pkt = NULL;
1446                 pend->pkt_len = 0;
1447                 log_assert(!pend->sq->busy);
1448                 pend->sq->busy = 1;
1449                 if(!randomize_and_send_udp(pend, outnet->udp_buff,
1450                         pend->timeout)) {
1451                         /* callback error on pending */
1452                         if(pend->cb) {
1453                                 fptr_ok(fptr_whitelist_pending_udp(pend->cb));
1454                                 (void)(*pend->cb)(outnet->unused_fds->cp, pend->cb_arg, 
1455                                         NETEVENT_CLOSED, NULL);
1456                         }
1457                         pending_delete(outnet, pend);
1458                 } else {
1459                         pend->sq->busy = 0;
1460                 }
1461         }
1462 }
1463
1464 int 
1465 outnet_udp_cb(struct comm_point* c, void* arg, int error,
1466         struct comm_reply *reply_info)
1467 {
1468         struct outside_network* outnet = (struct outside_network*)arg;
1469         struct pending key;
1470         struct pending* p;
1471         verbose(VERB_ALGO, "answer cb");
1472
1473         if(error != NETEVENT_NOERROR) {
1474                 verbose(VERB_QUERY, "outnetudp got udp error %d", error);
1475                 return 0;
1476         }
1477         if(sldns_buffer_limit(c->buffer) < LDNS_HEADER_SIZE) {
1478                 verbose(VERB_QUERY, "outnetudp udp too short");
1479                 return 0;
1480         }
1481         log_assert(reply_info);
1482
1483         /* setup lookup key */
1484         key.id = (unsigned)LDNS_ID_WIRE(sldns_buffer_begin(c->buffer));
1485         memcpy(&key.addr, &reply_info->remote_addr, reply_info->remote_addrlen);
1486         key.addrlen = reply_info->remote_addrlen;
1487         verbose(VERB_ALGO, "Incoming reply id = %4.4x", key.id);
1488         log_addr(VERB_ALGO, "Incoming reply addr =", 
1489                 &reply_info->remote_addr, reply_info->remote_addrlen);
1490
1491         /* find it, see if this thing is a valid query response */
1492         verbose(VERB_ALGO, "lookup size is %d entries", (int)outnet->pending->count);
1493         p = (struct pending*)rbtree_search(outnet->pending, &key);
1494         if(!p) {
1495                 verbose(VERB_QUERY, "received unwanted or unsolicited udp reply dropped.");
1496                 log_buf(VERB_ALGO, "dropped message", c->buffer);
1497                 outnet->unwanted_replies++;
1498                 if(outnet->unwanted_threshold && ++outnet->unwanted_total 
1499                         >= outnet->unwanted_threshold) {
1500                         log_warn("unwanted reply total reached threshold (%u)"
1501                                 " you may be under attack."
1502                                 " defensive action: clearing the cache",
1503                                 (unsigned)outnet->unwanted_threshold);
1504                         fptr_ok(fptr_whitelist_alloc_cleanup(
1505                                 outnet->unwanted_action));
1506                         (*outnet->unwanted_action)(outnet->unwanted_param);
1507                         outnet->unwanted_total = 0;
1508                 }
1509                 return 0;
1510         }
1511
1512         verbose(VERB_ALGO, "received udp reply.");
1513         log_buf(VERB_ALGO, "udp message", c->buffer);
1514         if(p->pc->cp != c) {
1515                 verbose(VERB_QUERY, "received reply id,addr on wrong port. "
1516                         "dropped.");
1517                 outnet->unwanted_replies++;
1518                 if(outnet->unwanted_threshold && ++outnet->unwanted_total 
1519                         >= outnet->unwanted_threshold) {
1520                         log_warn("unwanted reply total reached threshold (%u)"
1521                                 " you may be under attack."
1522                                 " defensive action: clearing the cache",
1523                                 (unsigned)outnet->unwanted_threshold);
1524                         fptr_ok(fptr_whitelist_alloc_cleanup(
1525                                 outnet->unwanted_action));
1526                         (*outnet->unwanted_action)(outnet->unwanted_param);
1527                         outnet->unwanted_total = 0;
1528                 }
1529                 return 0;
1530         }
1531         comm_timer_disable(p->timer);
1532         verbose(VERB_ALGO, "outnet handle udp reply");
1533         /* delete from tree first in case callback creates a retry */
1534         (void)rbtree_delete(outnet->pending, p->node.key);
1535         if(p->cb) {
1536                 fptr_ok(fptr_whitelist_pending_udp(p->cb));
1537                 (void)(*p->cb)(p->pc->cp, p->cb_arg, NETEVENT_NOERROR, reply_info);
1538         }
1539         portcomm_loweruse(outnet, p->pc);
1540         pending_delete(NULL, p);
1541         outnet_send_wait_udp(outnet);
1542         return 0;
1543 }
1544
1545 /** calculate number of ip4 and ip6 interfaces*/
1546 static void 
1547 calc_num46(char** ifs, int num_ifs, int do_ip4, int do_ip6, 
1548         int* num_ip4, int* num_ip6)
1549 {
1550         int i;
1551         *num_ip4 = 0;
1552         *num_ip6 = 0;
1553         if(num_ifs <= 0) {
1554                 if(do_ip4)
1555                         *num_ip4 = 1;
1556                 if(do_ip6)
1557                         *num_ip6 = 1;
1558                 return;
1559         }
1560         for(i=0; i<num_ifs; i++)
1561         {
1562                 if(str_is_ip6(ifs[i])) {
1563                         if(do_ip6)
1564                                 (*num_ip6)++;
1565                 } else {
1566                         if(do_ip4)
1567                                 (*num_ip4)++;
1568                 }
1569         }
1570 }
1571
1572 void
1573 pending_udp_timer_delay_cb(void* arg)
1574 {
1575         struct pending* p = (struct pending*)arg;
1576         struct outside_network* outnet = p->outnet;
1577         verbose(VERB_ALGO, "timeout udp with delay");
1578         portcomm_loweruse(outnet, p->pc);
1579         pending_delete(outnet, p);
1580         outnet_send_wait_udp(outnet);
1581 }
1582
1583 void 
1584 pending_udp_timer_cb(void *arg)
1585 {
1586         struct pending* p = (struct pending*)arg;
1587         struct outside_network* outnet = p->outnet;
1588         /* it timed out */
1589         verbose(VERB_ALGO, "timeout udp");
1590         if(p->cb) {
1591                 fptr_ok(fptr_whitelist_pending_udp(p->cb));
1592                 (void)(*p->cb)(p->pc->cp, p->cb_arg, NETEVENT_TIMEOUT, NULL);
1593         }
1594         /* if delayclose, keep port open for a longer time.
1595          * But if the udpwaitlist exists, then we are struggling to
1596          * keep up with demand for sockets, so do not wait, but service
1597          * the customer (customer service more important than portICMPs) */
1598         if(outnet->delayclose && !outnet->udp_wait_first) {
1599                 p->cb = NULL;
1600                 p->timer->callback = &pending_udp_timer_delay_cb;
1601                 comm_timer_set(p->timer, &outnet->delay_tv);
1602                 return;
1603         }
1604         portcomm_loweruse(outnet, p->pc);
1605         pending_delete(outnet, p);
1606         outnet_send_wait_udp(outnet);
1607 }
1608
1609 /** create pending_tcp buffers */
1610 static int
1611 create_pending_tcp(struct outside_network* outnet, size_t bufsize)
1612 {
1613         size_t i;
1614         if(outnet->num_tcp == 0)
1615                 return 1; /* no tcp needed, nothing to do */
1616         if(!(outnet->tcp_conns = (struct pending_tcp **)calloc(
1617                         outnet->num_tcp, sizeof(struct pending_tcp*))))
1618                 return 0;
1619         for(i=0; i<outnet->num_tcp; i++) {
1620                 if(!(outnet->tcp_conns[i] = (struct pending_tcp*)calloc(1, 
1621                         sizeof(struct pending_tcp))))
1622                         return 0;
1623                 outnet->tcp_conns[i]->next_free = outnet->tcp_free;
1624                 outnet->tcp_free = outnet->tcp_conns[i];
1625                 outnet->tcp_conns[i]->c = comm_point_create_tcp_out(
1626                         outnet->base, bufsize, outnet_tcp_cb, 
1627                         outnet->tcp_conns[i]);
1628                 if(!outnet->tcp_conns[i]->c)
1629                         return 0;
1630         }
1631         return 1;
1632 }
1633
1634 /** setup an outgoing interface, ready address */
1635 static int setup_if(struct port_if* pif, const char* addrstr, 
1636         int* avail, int numavail, size_t numfd)
1637 {
1638 #ifndef DISABLE_EXPLICIT_PORT_RANDOMISATION
1639         pif->avail_total = numavail;
1640         pif->avail_ports = (int*)memdup(avail, (size_t)numavail*sizeof(int));
1641         if(!pif->avail_ports)
1642                 return 0;
1643 #endif
1644         if(!ipstrtoaddr(addrstr, UNBOUND_DNS_PORT, &pif->addr, &pif->addrlen) &&
1645            !netblockstrtoaddr(addrstr, UNBOUND_DNS_PORT,
1646                               &pif->addr, &pif->addrlen, &pif->pfxlen))
1647                 return 0;
1648         pif->maxout = (int)numfd;
1649         pif->inuse = 0;
1650         pif->out = (struct port_comm**)calloc(numfd, 
1651                 sizeof(struct port_comm*));
1652         if(!pif->out)
1653                 return 0;
1654         return 1;
1655 }
1656
1657 struct outside_network* 
1658 outside_network_create(struct comm_base *base, size_t bufsize, 
1659         size_t num_ports, char** ifs, int num_ifs, int do_ip4, 
1660         int do_ip6, size_t num_tcp, int dscp, struct infra_cache* infra,
1661         struct ub_randstate* rnd, int use_caps_for_id, int* availports, 
1662         int numavailports, size_t unwanted_threshold, int tcp_mss,
1663         void (*unwanted_action)(void*), void* unwanted_param, int do_udp,
1664         void* sslctx, int delayclose, int tls_use_sni, struct dt_env* dtenv,
1665         int udp_connect, int max_reuse_tcp_queries, int tcp_reuse_timeout,
1666         int tcp_auth_query_timeout)
1667 {
1668         struct outside_network* outnet = (struct outside_network*)
1669                 calloc(1, sizeof(struct outside_network));
1670         size_t k;
1671         if(!outnet) {
1672                 log_err("malloc failed");
1673                 return NULL;
1674         }
1675         comm_base_timept(base, &outnet->now_secs, &outnet->now_tv);
1676         outnet->base = base;
1677         outnet->num_tcp = num_tcp;
1678         outnet->max_reuse_tcp_queries = max_reuse_tcp_queries;
1679         outnet->tcp_reuse_timeout= tcp_reuse_timeout;
1680         outnet->tcp_auth_query_timeout = tcp_auth_query_timeout;
1681         outnet->num_tcp_outgoing = 0;
1682         outnet->num_udp_outgoing = 0;
1683         outnet->infra = infra;
1684         outnet->rnd = rnd;
1685         outnet->sslctx = sslctx;
1686         outnet->tls_use_sni = tls_use_sni;
1687 #ifdef USE_DNSTAP
1688         outnet->dtenv = dtenv;
1689 #else
1690         (void)dtenv;
1691 #endif
1692         outnet->svcd_overhead = 0;
1693         outnet->want_to_quit = 0;
1694         outnet->unwanted_threshold = unwanted_threshold;
1695         outnet->unwanted_action = unwanted_action;
1696         outnet->unwanted_param = unwanted_param;
1697         outnet->use_caps_for_id = use_caps_for_id;
1698         outnet->do_udp = do_udp;
1699         outnet->tcp_mss = tcp_mss;
1700         outnet->ip_dscp = dscp;
1701 #ifndef S_SPLINT_S
1702         if(delayclose) {
1703                 outnet->delayclose = 1;
1704                 outnet->delay_tv.tv_sec = delayclose/1000;
1705                 outnet->delay_tv.tv_usec = (delayclose%1000)*1000;
1706         }
1707 #endif
1708         if(udp_connect) {
1709                 outnet->udp_connect = 1;
1710         }
1711         if(numavailports == 0 || num_ports == 0) {
1712                 log_err("no outgoing ports available");
1713                 outside_network_delete(outnet);
1714                 return NULL;
1715         }
1716 #ifndef INET6
1717         do_ip6 = 0;
1718 #endif
1719         calc_num46(ifs, num_ifs, do_ip4, do_ip6, 
1720                 &outnet->num_ip4, &outnet->num_ip6);
1721         if(outnet->num_ip4 != 0) {
1722                 if(!(outnet->ip4_ifs = (struct port_if*)calloc(
1723                         (size_t)outnet->num_ip4, sizeof(struct port_if)))) {
1724                         log_err("malloc failed");
1725                         outside_network_delete(outnet);
1726                         return NULL;
1727                 }
1728         }
1729         if(outnet->num_ip6 != 0) {
1730                 if(!(outnet->ip6_ifs = (struct port_if*)calloc(
1731                         (size_t)outnet->num_ip6, sizeof(struct port_if)))) {
1732                         log_err("malloc failed");
1733                         outside_network_delete(outnet);
1734                         return NULL;
1735                 }
1736         }
1737         if(     !(outnet->udp_buff = sldns_buffer_new(bufsize)) ||
1738                 !(outnet->pending = rbtree_create(pending_cmp)) ||
1739                 !(outnet->serviced = rbtree_create(serviced_cmp)) ||
1740                 !create_pending_tcp(outnet, bufsize)) {
1741                 log_err("malloc failed");
1742                 outside_network_delete(outnet);
1743                 return NULL;
1744         }
1745         rbtree_init(&outnet->tcp_reuse, reuse_cmp);
1746         outnet->tcp_reuse_max = num_tcp;
1747
1748         /* allocate commpoints */
1749         for(k=0; k<num_ports; k++) {
1750                 struct port_comm* pc;
1751                 pc = (struct port_comm*)calloc(1, sizeof(*pc));
1752                 if(!pc) {
1753                         log_err("malloc failed");
1754                         outside_network_delete(outnet);
1755                         return NULL;
1756                 }
1757                 pc->cp = comm_point_create_udp(outnet->base, -1, 
1758                         outnet->udp_buff, 0, outnet_udp_cb, outnet, NULL);
1759                 if(!pc->cp) {
1760                         log_err("malloc failed");
1761                         free(pc);
1762                         outside_network_delete(outnet);
1763                         return NULL;
1764                 }
1765                 pc->next = outnet->unused_fds;
1766                 outnet->unused_fds = pc;
1767         }
1768
1769         /* allocate interfaces */
1770         if(num_ifs == 0) {
1771                 if(do_ip4 && !setup_if(&outnet->ip4_ifs[0], "0.0.0.0", 
1772                         availports, numavailports, num_ports)) {
1773                         log_err("malloc failed");
1774                         outside_network_delete(outnet);
1775                         return NULL;
1776                 }
1777                 if(do_ip6 && !setup_if(&outnet->ip6_ifs[0], "::", 
1778                         availports, numavailports, num_ports)) {
1779                         log_err("malloc failed");
1780                         outside_network_delete(outnet);
1781                         return NULL;
1782                 }
1783         } else {
1784                 size_t done_4 = 0, done_6 = 0;
1785                 int i;
1786                 for(i=0; i<num_ifs; i++) {
1787                         if(str_is_ip6(ifs[i]) && do_ip6) {
1788                                 if(!setup_if(&outnet->ip6_ifs[done_6], ifs[i],
1789                                         availports, numavailports, num_ports)){
1790                                         log_err("malloc failed");
1791                                         outside_network_delete(outnet);
1792                                         return NULL;
1793                                 }
1794                                 done_6++;
1795                         }
1796                         if(!str_is_ip6(ifs[i]) && do_ip4) {
1797                                 if(!setup_if(&outnet->ip4_ifs[done_4], ifs[i],
1798                                         availports, numavailports, num_ports)){
1799                                         log_err("malloc failed");
1800                                         outside_network_delete(outnet);
1801                                         return NULL;
1802                                 }
1803                                 done_4++;
1804                         }
1805                 }
1806         }
1807         return outnet;
1808 }
1809
1810 /** helper pending delete */
1811 static void
1812 pending_node_del(rbnode_type* node, void* arg)
1813 {
1814         struct pending* pend = (struct pending*)node;
1815         struct outside_network* outnet = (struct outside_network*)arg;
1816         pending_delete(outnet, pend);
1817 }
1818
1819 /** helper serviced delete */
1820 static void
1821 serviced_node_del(rbnode_type* node, void* ATTR_UNUSED(arg))
1822 {
1823         struct serviced_query* sq = (struct serviced_query*)node;
1824         alloc_reg_release(sq->alloc, sq->region);
1825         if(sq->timer)
1826                 comm_timer_delete(sq->timer);
1827         free(sq);
1828 }
1829
1830 void 
1831 outside_network_quit_prepare(struct outside_network* outnet)
1832 {
1833         if(!outnet)
1834                 return;
1835         /* prevent queued items from being sent */
1836         outnet->want_to_quit = 1; 
1837 }
1838
1839 void 
1840 outside_network_delete(struct outside_network* outnet)
1841 {
1842         if(!outnet)
1843                 return;
1844         outnet->want_to_quit = 1;
1845         /* check every element, since we can be called on malloc error */
1846         if(outnet->pending) {
1847                 /* free pending elements, but do no unlink from tree. */
1848                 traverse_postorder(outnet->pending, pending_node_del, NULL);
1849                 free(outnet->pending);
1850         }
1851         if(outnet->serviced) {
1852                 traverse_postorder(outnet->serviced, serviced_node_del, NULL);
1853                 free(outnet->serviced);
1854         }
1855         if(outnet->udp_buff)
1856                 sldns_buffer_free(outnet->udp_buff);
1857         if(outnet->unused_fds) {
1858                 struct port_comm* p = outnet->unused_fds, *np;
1859                 while(p) {
1860                         np = p->next;
1861                         comm_point_delete(p->cp);
1862                         free(p);
1863                         p = np;
1864                 }
1865                 outnet->unused_fds = NULL;
1866         }
1867         if(outnet->ip4_ifs) {
1868                 int i, k;
1869                 for(i=0; i<outnet->num_ip4; i++) {
1870                         for(k=0; k<outnet->ip4_ifs[i].inuse; k++) {
1871                                 struct port_comm* pc = outnet->ip4_ifs[i].
1872                                         out[k];
1873                                 comm_point_delete(pc->cp);
1874                                 free(pc);
1875                         }
1876 #ifndef DISABLE_EXPLICIT_PORT_RANDOMISATION
1877                         free(outnet->ip4_ifs[i].avail_ports);
1878 #endif
1879                         free(outnet->ip4_ifs[i].out);
1880                 }
1881                 free(outnet->ip4_ifs);
1882         }
1883         if(outnet->ip6_ifs) {
1884                 int i, k;
1885                 for(i=0; i<outnet->num_ip6; i++) {
1886                         for(k=0; k<outnet->ip6_ifs[i].inuse; k++) {
1887                                 struct port_comm* pc = outnet->ip6_ifs[i].
1888                                         out[k];
1889                                 comm_point_delete(pc->cp);
1890                                 free(pc);
1891                         }
1892 #ifndef DISABLE_EXPLICIT_PORT_RANDOMISATION
1893                         free(outnet->ip6_ifs[i].avail_ports);
1894 #endif
1895                         free(outnet->ip6_ifs[i].out);
1896                 }
1897                 free(outnet->ip6_ifs);
1898         }
1899         if(outnet->tcp_conns) {
1900                 size_t i;
1901                 for(i=0; i<outnet->num_tcp; i++)
1902                         if(outnet->tcp_conns[i]) {
1903                                 struct pending_tcp* pend;
1904                                 pend = outnet->tcp_conns[i];
1905                                 if(pend->reuse.item_on_lru_list) {
1906                                         /* delete waiting_tcp elements that
1907                                          * the tcp conn is working on */
1908                                         decommission_pending_tcp(outnet, pend);
1909                                 }
1910                                 comm_point_delete(outnet->tcp_conns[i]->c);
1911                                 free(outnet->tcp_conns[i]);
1912                                 outnet->tcp_conns[i] = NULL;
1913                         }
1914                 free(outnet->tcp_conns);
1915                 outnet->tcp_conns = NULL;
1916         }
1917         if(outnet->tcp_wait_first) {
1918                 struct waiting_tcp* p = outnet->tcp_wait_first, *np;
1919                 while(p) {
1920                         np = p->next_waiting;
1921                         waiting_tcp_delete(p);
1922                         p = np;
1923                 }
1924         }
1925         /* was allocated in struct pending that was deleted above */
1926         rbtree_init(&outnet->tcp_reuse, reuse_cmp);
1927         outnet->tcp_reuse_first = NULL;
1928         outnet->tcp_reuse_last = NULL;
1929         if(outnet->udp_wait_first) {
1930                 struct pending* p = outnet->udp_wait_first, *np;
1931                 while(p) {
1932                         np = p->next_waiting;
1933                         pending_delete(NULL, p);
1934                         p = np;
1935                 }
1936         }
1937         free(outnet);
1938 }
1939
1940 void 
1941 pending_delete(struct outside_network* outnet, struct pending* p)
1942 {
1943         if(!p)
1944                 return;
1945         if(outnet && outnet->udp_wait_first &&
1946                 (p->next_waiting || p == outnet->udp_wait_last) ) {
1947                 /* delete from waiting list, if it is in the waiting list */
1948                 struct pending* prev = NULL, *x = outnet->udp_wait_first;
1949                 while(x && x != p) {
1950                         prev = x;
1951                         x = x->next_waiting;
1952                 }
1953                 if(x) {
1954                         log_assert(x == p);
1955                         if(prev)
1956                                 prev->next_waiting = p->next_waiting;
1957                         else    outnet->udp_wait_first = p->next_waiting;
1958                         if(outnet->udp_wait_last == p)
1959                                 outnet->udp_wait_last = prev;
1960                 }
1961         }
1962         if(outnet) {
1963                 (void)rbtree_delete(outnet->pending, p->node.key);
1964         }
1965         if(p->timer)
1966                 comm_timer_delete(p->timer);
1967         free(p->pkt);
1968         free(p);
1969 }
1970
1971 static void
1972 sai6_putrandom(struct sockaddr_in6 *sa, int pfxlen, struct ub_randstate *rnd)
1973 {
1974         int i, last;
1975         if(!(pfxlen > 0 && pfxlen < 128))
1976                 return;
1977         for(i = 0; i < (128 - pfxlen) / 8; i++) {
1978                 sa->sin6_addr.s6_addr[15-i] = (uint8_t)ub_random_max(rnd, 256);
1979         }
1980         last = pfxlen & 7;
1981         if(last != 0) {
1982                 sa->sin6_addr.s6_addr[15-i] |=
1983                         ((0xFF >> last) & ub_random_max(rnd, 256));
1984         }
1985 }
1986
1987 /**
1988  * Try to open a UDP socket for outgoing communication.
1989  * Sets sockets options as needed.
1990  * @param addr: socket address.
1991  * @param addrlen: length of address.
1992  * @param pfxlen: length of network prefix (for address randomisation).
1993  * @param port: port override for addr.
1994  * @param inuse: if -1 is returned, this bool means the port was in use.
1995  * @param rnd: random state (for address randomisation).
1996  * @param dscp: DSCP to use.
1997  * @return fd or -1
1998  */
1999 static int
2000 udp_sockport(struct sockaddr_storage* addr, socklen_t addrlen, int pfxlen,
2001         int port, int* inuse, struct ub_randstate* rnd, int dscp)
2002 {
2003         int fd, noproto;
2004         if(addr_is_ip6(addr, addrlen)) {
2005                 int freebind = 0;
2006                 struct sockaddr_in6 sa = *(struct sockaddr_in6*)addr;
2007                 sa.sin6_port = (in_port_t)htons((uint16_t)port);
2008                 sa.sin6_flowinfo = 0;
2009                 sa.sin6_scope_id = 0;
2010                 if(pfxlen != 0) {
2011                         freebind = 1;
2012                         sai6_putrandom(&sa, pfxlen, rnd);
2013                 }
2014                 fd = create_udp_sock(AF_INET6, SOCK_DGRAM, 
2015                         (struct sockaddr*)&sa, addrlen, 1, inuse, &noproto,
2016                         0, 0, 0, NULL, 0, freebind, 0, dscp);
2017         } else {
2018                 struct sockaddr_in* sa = (struct sockaddr_in*)addr;
2019                 sa->sin_port = (in_port_t)htons((uint16_t)port);
2020                 fd = create_udp_sock(AF_INET, SOCK_DGRAM, 
2021                         (struct sockaddr*)addr, addrlen, 1, inuse, &noproto,
2022                         0, 0, 0, NULL, 0, 0, 0, dscp);
2023         }
2024         return fd;
2025 }
2026
2027 /** Select random ID */
2028 static int
2029 select_id(struct outside_network* outnet, struct pending* pend,
2030         sldns_buffer* packet)
2031 {
2032         int id_tries = 0;
2033         pend->id = GET_RANDOM_ID(outnet->rnd);
2034         LDNS_ID_SET(sldns_buffer_begin(packet), pend->id);
2035
2036         /* insert in tree */
2037         pend->node.key = pend;
2038         while(!rbtree_insert(outnet->pending, &pend->node)) {
2039                 /* change ID to avoid collision */
2040                 pend->id = GET_RANDOM_ID(outnet->rnd);
2041                 LDNS_ID_SET(sldns_buffer_begin(packet), pend->id);
2042                 id_tries++;
2043                 if(id_tries == MAX_ID_RETRY) {
2044                         pend->id=99999; /* non existent ID */
2045                         log_err("failed to generate unique ID, drop msg");
2046                         return 0;
2047                 }
2048         }
2049         verbose(VERB_ALGO, "inserted new pending reply id=%4.4x", pend->id);
2050         return 1;
2051 }
2052
2053 /** return true is UDP connect error needs to be logged */
2054 static int udp_connect_needs_log(int err)
2055 {
2056         switch(err) {
2057         case ECONNREFUSED:
2058 #  ifdef ENETUNREACH
2059         case ENETUNREACH:
2060 #  endif
2061 #  ifdef EHOSTDOWN
2062         case EHOSTDOWN:
2063 #  endif
2064 #  ifdef EHOSTUNREACH
2065         case EHOSTUNREACH:
2066 #  endif
2067 #  ifdef ENETDOWN
2068         case ENETDOWN:
2069 #  endif
2070 #  ifdef EADDRNOTAVAIL
2071         case EADDRNOTAVAIL:
2072 #  endif
2073         case EPERM:
2074         case EACCES:
2075                 if(verbosity >= VERB_ALGO)
2076                         return 1;
2077                 return 0;
2078         default:
2079                 break;
2080         }
2081         return 1;
2082 }
2083
2084
2085 /** Select random interface and port */
2086 static int
2087 select_ifport(struct outside_network* outnet, struct pending* pend,
2088         int num_if, struct port_if* ifs)
2089 {
2090         int my_if, my_port, fd, portno, inuse, tries=0;
2091         struct port_if* pif;
2092         /* randomly select interface and port */
2093         if(num_if == 0) {
2094                 verbose(VERB_QUERY, "Need to send query but have no "
2095                         "outgoing interfaces of that family");
2096                 return 0;
2097         }
2098         log_assert(outnet->unused_fds);
2099         tries = 0;
2100         while(1) {
2101                 my_if = ub_random_max(outnet->rnd, num_if);
2102                 pif = &ifs[my_if];
2103 #ifndef DISABLE_EXPLICIT_PORT_RANDOMISATION
2104                 if(outnet->udp_connect) {
2105                         /* if we connect() we cannot reuse fds for a port */
2106                         if(pif->inuse >= pif->avail_total) {
2107                                 tries++;
2108                                 if(tries < MAX_PORT_RETRY)
2109                                         continue;
2110                                 log_err("failed to find an open port, drop msg");
2111                                 return 0;
2112                         }
2113                         my_port = pif->inuse + ub_random_max(outnet->rnd,
2114                                 pif->avail_total - pif->inuse);
2115                 } else  {
2116                         my_port = ub_random_max(outnet->rnd, pif->avail_total);
2117                         if(my_port < pif->inuse) {
2118                                 /* port already open */
2119                                 pend->pc = pif->out[my_port];
2120                                 verbose(VERB_ALGO, "using UDP if=%d port=%d",
2121                                         my_if, pend->pc->number);
2122                                 break;
2123                         }
2124                 }
2125                 /* try to open new port, if fails, loop to try again */
2126                 log_assert(pif->inuse < pif->maxout);
2127                 portno = pif->avail_ports[my_port - pif->inuse];
2128 #else
2129                 my_port = portno = 0;
2130 #endif
2131                 fd = udp_sockport(&pif->addr, pif->addrlen, pif->pfxlen,
2132                         portno, &inuse, outnet->rnd, outnet->ip_dscp);
2133                 if(fd == -1 && !inuse) {
2134                         /* nonrecoverable error making socket */
2135                         return 0;
2136                 }
2137                 if(fd != -1) {
2138                         verbose(VERB_ALGO, "opened UDP if=%d port=%d", 
2139                                 my_if, portno);
2140                         if(outnet->udp_connect) {
2141                                 /* connect() to the destination */
2142                                 if(connect(fd, (struct sockaddr*)&pend->addr,
2143                                         pend->addrlen) < 0) {
2144                                         if(udp_connect_needs_log(errno)) {
2145                                                 log_err_addr("udp connect failed",
2146                                                         strerror(errno), &pend->addr,
2147                                                         pend->addrlen);
2148                                         }
2149                                         sock_close(fd);
2150                                         return 0;
2151                                 }
2152                         }
2153                         /* grab fd */
2154                         pend->pc = outnet->unused_fds;
2155                         outnet->unused_fds = pend->pc->next;
2156
2157                         /* setup portcomm */
2158                         pend->pc->next = NULL;
2159                         pend->pc->number = portno;
2160                         pend->pc->pif = pif;
2161                         pend->pc->index = pif->inuse;
2162                         pend->pc->num_outstanding = 0;
2163                         comm_point_start_listening(pend->pc->cp, fd, -1);
2164
2165                         /* grab port in interface */
2166                         pif->out[pif->inuse] = pend->pc;
2167 #ifndef DISABLE_EXPLICIT_PORT_RANDOMISATION
2168                         pif->avail_ports[my_port - pif->inuse] =
2169                                 pif->avail_ports[pif->avail_total-pif->inuse-1];
2170 #endif
2171                         pif->inuse++;
2172                         break;
2173                 }
2174                 /* failed, already in use */
2175                 verbose(VERB_QUERY, "port %d in use, trying another", portno);
2176                 tries++;
2177                 if(tries == MAX_PORT_RETRY) {
2178                         log_err("failed to find an open port, drop msg");
2179                         return 0;
2180                 }
2181         }
2182         log_assert(pend->pc);
2183         pend->pc->num_outstanding++;
2184
2185         return 1;
2186 }
2187
2188 static int
2189 randomize_and_send_udp(struct pending* pend, sldns_buffer* packet, int timeout)
2190 {
2191         struct timeval tv;
2192         struct outside_network* outnet = pend->sq->outnet;
2193
2194         /* select id */
2195         if(!select_id(outnet, pend, packet)) {
2196                 return 0;
2197         }
2198
2199         /* select src_if, port */
2200         if(addr_is_ip6(&pend->addr, pend->addrlen)) {
2201                 if(!select_ifport(outnet, pend, 
2202                         outnet->num_ip6, outnet->ip6_ifs))
2203                         return 0;
2204         } else {
2205                 if(!select_ifport(outnet, pend, 
2206                         outnet->num_ip4, outnet->ip4_ifs))
2207                         return 0;
2208         }
2209         log_assert(pend->pc && pend->pc->cp);
2210
2211         /* send it over the commlink */
2212         if(!comm_point_send_udp_msg(pend->pc->cp, packet,
2213                 (struct sockaddr*)&pend->addr, pend->addrlen, outnet->udp_connect)) {
2214                 portcomm_loweruse(outnet, pend->pc);
2215                 return 0;
2216         }
2217         outnet->num_udp_outgoing++;
2218
2219         /* system calls to set timeout after sending UDP to make roundtrip
2220            smaller. */
2221 #ifndef S_SPLINT_S
2222         tv.tv_sec = timeout/1000;
2223         tv.tv_usec = (timeout%1000)*1000;
2224 #endif
2225         comm_timer_set(pend->timer, &tv);
2226
2227 #ifdef USE_DNSTAP
2228         /*
2229          * sending src (local service)/dst (upstream) addresses over DNSTAP
2230          * There are no chances to get the src (local service) addr if unbound
2231          * is not configured with specific outgoing IP-addresses. So we will
2232          * pass 0.0.0.0 (::) to argument for
2233          * dt_msg_send_outside_query()/dt_msg_send_outside_response() calls.
2234          */
2235         if(outnet->dtenv &&
2236            (outnet->dtenv->log_resolver_query_messages ||
2237                 outnet->dtenv->log_forwarder_query_messages)) {
2238                         log_addr(VERB_ALGO, "from local addr", &pend->pc->pif->addr, pend->pc->pif->addrlen);
2239                         log_addr(VERB_ALGO, "request to upstream", &pend->addr, pend->addrlen);
2240                         dt_msg_send_outside_query(outnet->dtenv, &pend->addr, &pend->pc->pif->addr, comm_udp,
2241                                 pend->sq->zone, pend->sq->zonelen, packet);
2242         }
2243 #endif
2244         return 1;
2245 }
2246
2247 struct pending* 
2248 pending_udp_query(struct serviced_query* sq, struct sldns_buffer* packet,
2249         int timeout, comm_point_callback_type* cb, void* cb_arg)
2250 {
2251         struct pending* pend = (struct pending*)calloc(1, sizeof(*pend));
2252         if(!pend) return NULL;
2253         pend->outnet = sq->outnet;
2254         pend->sq = sq;
2255         pend->addrlen = sq->addrlen;
2256         memmove(&pend->addr, &sq->addr, sq->addrlen);
2257         pend->cb = cb;
2258         pend->cb_arg = cb_arg;
2259         pend->node.key = pend;
2260         pend->timer = comm_timer_create(sq->outnet->base, pending_udp_timer_cb,
2261                 pend);
2262         if(!pend->timer) {
2263                 free(pend);
2264                 return NULL;
2265         }
2266
2267         if(sq->outnet->unused_fds == NULL) {
2268                 /* no unused fd, cannot create a new port (randomly) */
2269                 verbose(VERB_ALGO, "no fds available, udp query waiting");
2270                 pend->timeout = timeout;
2271                 pend->pkt_len = sldns_buffer_limit(packet);
2272                 pend->pkt = (uint8_t*)memdup(sldns_buffer_begin(packet),
2273                         pend->pkt_len);
2274                 if(!pend->pkt) {
2275                         comm_timer_delete(pend->timer);
2276                         free(pend);
2277                         return NULL;
2278                 }
2279                 /* put at end of waiting list */
2280                 if(sq->outnet->udp_wait_last)
2281                         sq->outnet->udp_wait_last->next_waiting = pend;
2282                 else 
2283                         sq->outnet->udp_wait_first = pend;
2284                 sq->outnet->udp_wait_last = pend;
2285                 return pend;
2286         }
2287         log_assert(!sq->busy);
2288         sq->busy = 1;
2289         if(!randomize_and_send_udp(pend, packet, timeout)) {
2290                 pending_delete(sq->outnet, pend);
2291                 return NULL;
2292         }
2293         sq->busy = 0;
2294         return pend;
2295 }
2296
2297 void
2298 outnet_tcptimer(void* arg)
2299 {
2300         struct waiting_tcp* w = (struct waiting_tcp*)arg;
2301         struct outside_network* outnet = w->outnet;
2302         verbose(VERB_CLIENT, "outnet_tcptimer");
2303         if(w->on_tcp_waiting_list) {
2304                 /* it is on the waiting list */
2305                 outnet_waiting_tcp_list_remove(outnet, w);
2306                 waiting_tcp_callback(w, NULL, NETEVENT_TIMEOUT, NULL);
2307                 waiting_tcp_delete(w);
2308         } else {
2309                 /* it was in use */
2310                 struct pending_tcp* pend=(struct pending_tcp*)w->next_waiting;
2311                 reuse_cb_and_decommission(outnet, pend, NETEVENT_TIMEOUT);
2312         }
2313         use_free_buffer(outnet);
2314 }
2315
2316 /** close the oldest reuse_tcp connection to make a fd and struct pend
2317  * available for a new stream connection */
2318 static void
2319 reuse_tcp_close_oldest(struct outside_network* outnet)
2320 {
2321         struct reuse_tcp* reuse;
2322         verbose(VERB_CLIENT, "reuse_tcp_close_oldest");
2323         reuse = reuse_tcp_lru_snip(outnet);
2324         if(!reuse) return;
2325         /* free up */
2326         reuse_cb_and_decommission(outnet, reuse->pending, NETEVENT_CLOSED);
2327 }
2328
2329 static uint16_t
2330 tcp_select_id(struct outside_network* outnet, struct reuse_tcp* reuse)
2331 {
2332         if(reuse)
2333                 return reuse_tcp_select_id(reuse, outnet);
2334         return GET_RANDOM_ID(outnet->rnd);
2335 }
2336
2337 /** find spare ID value for reuse tcp stream.  That is random and also does
2338  * not collide with an existing query ID that is in use or waiting */
2339 uint16_t
2340 reuse_tcp_select_id(struct reuse_tcp* reuse, struct outside_network* outnet)
2341 {
2342         uint16_t id = 0, curid, nextid;
2343         const int try_random = 2000;
2344         int i;
2345         unsigned select, count, space;
2346         rbnode_type* node;
2347
2348         /* make really sure the tree is not empty */
2349         if(reuse->tree_by_id.count == 0) {
2350                 id = GET_RANDOM_ID(outnet->rnd);
2351                 return id;
2352         }
2353
2354         /* try to find random empty spots by picking them */
2355         for(i = 0; i<try_random; i++) {
2356                 id = GET_RANDOM_ID(outnet->rnd);
2357                 if(!reuse_tcp_by_id_find(reuse, id)) {
2358                         return id;
2359                 }
2360         }
2361
2362         /* equally pick a random unused element from the tree that is
2363          * not in use.  Pick a the n-th index of an unused number,
2364          * then loop over the empty spaces in the tree and find it */
2365         log_assert(reuse->tree_by_id.count < 0xffff);
2366         select = ub_random_max(outnet->rnd, 0xffff - reuse->tree_by_id.count);
2367         /* select value now in 0 .. num free - 1 */
2368
2369         count = 0; /* number of free spaces passed by */
2370         node = rbtree_first(&reuse->tree_by_id);
2371         log_assert(node && node != RBTREE_NULL); /* tree not empty */
2372         /* see if select is before first node */
2373         if(select < (unsigned)tree_by_id_get_id(node))
2374                 return select;
2375         count += tree_by_id_get_id(node);
2376         /* perhaps select is between nodes */
2377         while(node && node != RBTREE_NULL) {
2378                 rbnode_type* next = rbtree_next(node);
2379                 if(next && next != RBTREE_NULL) {
2380                         curid = tree_by_id_get_id(node);
2381                         nextid = tree_by_id_get_id(next);
2382                         log_assert(curid < nextid);
2383                         if(curid != 0xffff && curid + 1 < nextid) {
2384                                 /* space between nodes */
2385                                 space = nextid - curid - 1;
2386                                 log_assert(select >= count);
2387                                 if(select < count + space) {
2388                                         /* here it is */
2389                                         return curid + 1 + (select - count);
2390                                 }
2391                                 count += space;
2392                         }
2393                 }
2394                 node = next;
2395         }
2396
2397         /* select is after the last node */
2398         /* count is the number of free positions before the nodes in the
2399          * tree */
2400         node = rbtree_last(&reuse->tree_by_id);
2401         log_assert(node && node != RBTREE_NULL); /* tree not empty */
2402         curid = tree_by_id_get_id(node);
2403         log_assert(count + (0xffff-curid) + reuse->tree_by_id.count == 0xffff);
2404         return curid + 1 + (select - count);
2405 }
2406
2407 struct waiting_tcp*
2408 pending_tcp_query(struct serviced_query* sq, sldns_buffer* packet,
2409         int timeout, comm_point_callback_type* callback, void* callback_arg)
2410 {
2411         struct pending_tcp* pend = sq->outnet->tcp_free;
2412         struct reuse_tcp* reuse = NULL;
2413         struct waiting_tcp* w;
2414
2415         verbose(VERB_CLIENT, "pending_tcp_query");
2416         if(sldns_buffer_limit(packet) < sizeof(uint16_t)) {
2417                 verbose(VERB_ALGO, "pending tcp query with too short buffer < 2");
2418                 return NULL;
2419         }
2420
2421         /* find out if a reused stream to the target exists */
2422         /* if so, take it into use */
2423         reuse = reuse_tcp_find(sq->outnet, &sq->addr, sq->addrlen,
2424                 sq->ssl_upstream);
2425         if(reuse) {
2426                 log_reuse_tcp(VERB_CLIENT, "pending_tcp_query: found reuse", reuse);
2427                 log_assert(reuse->pending);
2428                 pend = reuse->pending;
2429                 reuse_tcp_lru_touch(sq->outnet, reuse);
2430         }
2431
2432         log_assert(!reuse || (reuse && pend));
2433         /* if !pend but we have reuse streams, close a reuse stream
2434          * to be able to open a new one to this target, no use waiting
2435          * to reuse a file descriptor while another query needs to use
2436          * that buffer and file descriptor now. */
2437         if(!pend) {
2438                 reuse_tcp_close_oldest(sq->outnet);
2439                 pend = sq->outnet->tcp_free;
2440                 log_assert(!reuse || (pend == reuse->pending));
2441         }
2442
2443         /* allocate space to store query */
2444         w = (struct waiting_tcp*)malloc(sizeof(struct waiting_tcp) 
2445                 + sldns_buffer_limit(packet));
2446         if(!w) {
2447                 return NULL;
2448         }
2449         if(!(w->timer = comm_timer_create(sq->outnet->base, outnet_tcptimer, w))) {
2450                 free(w);
2451                 return NULL;
2452         }
2453         w->pkt = (uint8_t*)w + sizeof(struct waiting_tcp);
2454         w->pkt_len = sldns_buffer_limit(packet);
2455         memmove(w->pkt, sldns_buffer_begin(packet), w->pkt_len);
2456         w->id = tcp_select_id(sq->outnet, reuse);
2457         LDNS_ID_SET(w->pkt, w->id);
2458         memcpy(&w->addr, &sq->addr, sq->addrlen);
2459         w->addrlen = sq->addrlen;
2460         w->outnet = sq->outnet;
2461         w->on_tcp_waiting_list = 0;
2462         w->next_waiting = NULL;
2463         w->cb = callback;
2464         w->cb_arg = callback_arg;
2465         w->ssl_upstream = sq->ssl_upstream;
2466         w->tls_auth_name = sq->tls_auth_name;
2467         w->timeout = timeout;
2468         w->id_node.key = NULL;
2469         w->write_wait_prev = NULL;
2470         w->write_wait_next = NULL;
2471         w->write_wait_queued = 0;
2472         w->error_count = 0;
2473 #ifdef USE_DNSTAP
2474         w->sq = NULL;
2475 #endif
2476         w->in_cb_and_decommission = 0;
2477         if(pend) {
2478                 /* we have a buffer available right now */
2479                 if(reuse) {
2480                         log_assert(reuse == &pend->reuse);
2481                         /* reuse existing fd, write query and continue */
2482                         /* store query in tree by id */
2483                         verbose(VERB_CLIENT, "pending_tcp_query: reuse, store");
2484                         w->next_waiting = (void*)pend;
2485                         reuse_tree_by_id_insert(&pend->reuse, w);
2486                         /* can we write right now? */
2487                         if(pend->query == NULL) {
2488                                 /* write straight away */
2489                                 /* stop the timer on read of the fd */
2490                                 comm_point_stop_listening(pend->c);
2491                                 pend->query = w;
2492                                 outnet_tcp_take_query_setup(pend->c->fd, pend,
2493                                         w);
2494                         } else {
2495                                 /* put it in the waiting list for
2496                                  * this stream */
2497                                 reuse_write_wait_push_back(&pend->reuse, w);
2498                         }
2499                 } else {
2500                         /* create new fd and connect to addr, setup to
2501                          * write query */
2502                         verbose(VERB_CLIENT, "pending_tcp_query: new fd, connect");
2503                         rbtree_init(&pend->reuse.tree_by_id, reuse_id_cmp);
2504                         pend->reuse.pending = pend;
2505                         memcpy(&pend->reuse.addr, &sq->addr, sq->addrlen);
2506                         pend->reuse.addrlen = sq->addrlen;
2507                         if(!outnet_tcp_take_into_use(w)) {
2508                                 waiting_tcp_delete(w);
2509                                 return NULL;
2510                         }
2511                 }
2512 #ifdef USE_DNSTAP
2513                 if(sq->outnet->dtenv &&
2514                    (sq->outnet->dtenv->log_resolver_query_messages ||
2515                     sq->outnet->dtenv->log_forwarder_query_messages)) {
2516                         /* use w->pkt, because it has the ID value */
2517                         sldns_buffer tmp;
2518                         sldns_buffer_init_frm_data(&tmp, w->pkt, w->pkt_len);
2519                         dt_msg_send_outside_query(sq->outnet->dtenv, &sq->addr,
2520                                 &pend->pi->addr, comm_tcp, sq->zone,
2521                                 sq->zonelen, &tmp);
2522                 }
2523 #endif
2524         } else {
2525                 /* queue up */
2526                 /* waiting for a buffer on the outside network buffer wait
2527                  * list */
2528                 verbose(VERB_CLIENT, "pending_tcp_query: queue to wait");
2529 #ifdef USE_DNSTAP
2530                 w->sq = sq;
2531 #endif
2532                 outnet_waiting_tcp_list_add(sq->outnet, w, 1);
2533         }
2534         return w;
2535 }
2536
2537 /** create query for serviced queries */
2538 static void
2539 serviced_gen_query(sldns_buffer* buff, uint8_t* qname, size_t qnamelen, 
2540         uint16_t qtype, uint16_t qclass, uint16_t flags)
2541 {
2542         sldns_buffer_clear(buff);
2543         /* skip id */
2544         sldns_buffer_write_u16(buff, flags);
2545         sldns_buffer_write_u16(buff, 1); /* qdcount */
2546         sldns_buffer_write_u16(buff, 0); /* ancount */
2547         sldns_buffer_write_u16(buff, 0); /* nscount */
2548         sldns_buffer_write_u16(buff, 0); /* arcount */
2549         sldns_buffer_write(buff, qname, qnamelen);
2550         sldns_buffer_write_u16(buff, qtype);
2551         sldns_buffer_write_u16(buff, qclass);
2552         sldns_buffer_flip(buff);
2553 }
2554
2555 /** lookup serviced query in serviced query rbtree */
2556 static struct serviced_query*
2557 lookup_serviced(struct outside_network* outnet, sldns_buffer* buff, int dnssec,
2558         struct sockaddr_storage* addr, socklen_t addrlen,
2559         struct edns_option* opt_list)
2560 {
2561         struct serviced_query key;
2562         key.node.key = &key;
2563         key.qbuf = sldns_buffer_begin(buff);
2564         key.qbuflen = sldns_buffer_limit(buff);
2565         key.dnssec = dnssec;
2566         memcpy(&key.addr, addr, addrlen);
2567         key.addrlen = addrlen;
2568         key.outnet = outnet;
2569         key.opt_list = opt_list;
2570         return (struct serviced_query*)rbtree_search(outnet->serviced, &key);
2571 }
2572
2573 void
2574 serviced_timer_cb(void* arg)
2575 {
2576         struct serviced_query* sq = (struct serviced_query*)arg;
2577         struct outside_network* outnet = sq->outnet;
2578         verbose(VERB_ALGO, "serviced send timer");
2579         /* By the time this cb is called, if we don't have any registered
2580          * callbacks for this serviced_query anymore; do not send. */
2581         if(!sq->cblist)
2582                 goto delete;
2583         /* perform first network action */
2584         if(outnet->do_udp && !(sq->tcp_upstream || sq->ssl_upstream)) {
2585                 if(!serviced_udp_send(sq, outnet->udp_buff))
2586                         goto delete;
2587         } else {
2588                 if(!serviced_tcp_send(sq, outnet->udp_buff))
2589                         goto delete;
2590         }
2591         /* Maybe by this time we don't have callbacks attached anymore. Don't
2592          * proactively try to delete; let it run and maybe another callback
2593          * will get attached by the time we get an answer. */
2594         return;
2595 delete:
2596         serviced_callbacks(sq, NETEVENT_CLOSED, NULL, NULL);
2597 }
2598
2599 /** Create new serviced entry */
2600 static struct serviced_query*
2601 serviced_create(struct outside_network* outnet, sldns_buffer* buff, int dnssec,
2602         int want_dnssec, int nocaps, int tcp_upstream, int ssl_upstream,
2603         char* tls_auth_name, struct sockaddr_storage* addr, socklen_t addrlen,
2604         uint8_t* zone, size_t zonelen, int qtype, struct edns_option* opt_list,
2605         size_t pad_queries_block_size, struct alloc_cache* alloc,
2606         struct regional* region)
2607 {
2608         struct serviced_query* sq = (struct serviced_query*)malloc(sizeof(*sq));
2609         struct timeval t;
2610 #ifdef UNBOUND_DEBUG
2611         rbnode_type* ins;
2612 #endif
2613         if(!sq) {
2614                 alloc_reg_release(alloc, region);
2615                 return NULL;
2616         }
2617         sq->node.key = sq;
2618         sq->alloc = alloc;
2619         sq->region = region;
2620         sq->qbuf = regional_alloc_init(region, sldns_buffer_begin(buff),
2621                 sldns_buffer_limit(buff));
2622         if(!sq->qbuf) {
2623                 alloc_reg_release(alloc, region);
2624                 free(sq);
2625                 return NULL;
2626         }
2627         sq->qbuflen = sldns_buffer_limit(buff);
2628         sq->zone = regional_alloc_init(region, zone, zonelen);
2629         if(!sq->zone) {
2630                 alloc_reg_release(alloc, region);
2631                 free(sq);
2632                 return NULL;
2633         }
2634         sq->zonelen = zonelen;
2635         sq->qtype = qtype;
2636         sq->dnssec = dnssec;
2637         sq->want_dnssec = want_dnssec;
2638         sq->nocaps = nocaps;
2639         sq->tcp_upstream = tcp_upstream;
2640         sq->ssl_upstream = ssl_upstream;
2641         if(tls_auth_name) {
2642                 sq->tls_auth_name = regional_strdup(region, tls_auth_name);
2643                 if(!sq->tls_auth_name) {
2644                         alloc_reg_release(alloc, region);
2645                         free(sq);
2646                         return NULL;
2647                 }
2648         } else {
2649                 sq->tls_auth_name = NULL;
2650         }
2651         memcpy(&sq->addr, addr, addrlen);
2652         sq->addrlen = addrlen;
2653         sq->opt_list = opt_list;
2654         sq->busy = 0;
2655         sq->timer = comm_timer_create(outnet->base, serviced_timer_cb, sq);
2656         if(!sq->timer) {
2657                 alloc_reg_release(alloc, region);
2658                 free(sq);
2659                 return NULL;
2660         }
2661         memset(&t, 0, sizeof(t));
2662         comm_timer_set(sq->timer, &t);
2663         sq->outnet = outnet;
2664         sq->cblist = NULL;
2665         sq->pending = NULL;
2666         sq->status = serviced_initial;
2667         sq->retry = 0;
2668         sq->to_be_deleted = 0;
2669         sq->padding_block_size = pad_queries_block_size;
2670 #ifdef UNBOUND_DEBUG
2671         ins =
2672 #else
2673         (void)
2674 #endif
2675         rbtree_insert(outnet->serviced, &sq->node);
2676         log_assert(ins != NULL); /* must not be already present */
2677         return sq;
2678 }
2679
2680 /** reuse tcp stream, remove serviced query from stream,
2681  * return true if the stream is kept, false if it is to be closed */
2682 static int
2683 reuse_tcp_remove_serviced_keep(struct waiting_tcp* w,
2684         struct serviced_query* sq)
2685 {
2686         struct pending_tcp* pend_tcp = (struct pending_tcp*)w->next_waiting;
2687         verbose(VERB_CLIENT, "reuse_tcp_remove_serviced_keep");
2688         /* remove the callback. let query continue to write to not cancel
2689          * the stream itself.  also keep it as an entry in the tree_by_id,
2690          * in case the answer returns (that we no longer want), but we cannot
2691          * pick the same ID number meanwhile */
2692         w->cb = NULL;
2693         /* see if can be entered in reuse tree
2694          * for that the FD has to be non-1 */
2695         if(pend_tcp->c->fd == -1) {
2696                 verbose(VERB_CLIENT, "reuse_tcp_remove_serviced_keep: -1 fd");
2697                 return 0;
2698         }
2699         /* if in tree and used by other queries */
2700         if(pend_tcp->reuse.node.key) {
2701                 verbose(VERB_CLIENT, "reuse_tcp_remove_serviced_keep: in use by other queries");
2702                 /* do not reset the keepalive timer, for that
2703                  * we'd need traffic, and this is where the serviced is
2704                  * removed due to state machine internal reasons,
2705                  * eg. iterator no longer interested in this query */
2706                 return 1;
2707         }
2708         /* if still open and want to keep it open */
2709         if(pend_tcp->c->fd != -1 && sq->outnet->tcp_reuse.count <
2710                 sq->outnet->tcp_reuse_max) {
2711                 verbose(VERB_CLIENT, "reuse_tcp_remove_serviced_keep: keep open");
2712                 /* set a keepalive timer on it */
2713                 if(!reuse_tcp_insert(sq->outnet, pend_tcp)) {
2714                         return 0;
2715                 }
2716                 reuse_tcp_setup_timeout(pend_tcp, sq->outnet->tcp_reuse_timeout);
2717                 return 1;
2718         }
2719         return 0;
2720 }
2721
2722 /** cleanup serviced query entry */
2723 static void
2724 serviced_delete(struct serviced_query* sq)
2725 {
2726         verbose(VERB_CLIENT, "serviced_delete");
2727         if(sq->pending) {
2728                 /* clear up the pending query */
2729                 if(sq->status == serviced_query_UDP_EDNS ||
2730                         sq->status == serviced_query_UDP ||
2731                         sq->status == serviced_query_UDP_EDNS_FRAG ||
2732                         sq->status == serviced_query_UDP_EDNS_fallback) {
2733                         struct pending* p = (struct pending*)sq->pending;
2734                         verbose(VERB_CLIENT, "serviced_delete: UDP");
2735                         if(p->pc)
2736                                 portcomm_loweruse(sq->outnet, p->pc);
2737                         pending_delete(sq->outnet, p);
2738                         /* this call can cause reentrant calls back into the
2739                          * mesh */
2740                         outnet_send_wait_udp(sq->outnet);
2741                 } else {
2742                         struct waiting_tcp* w = (struct waiting_tcp*)
2743                                 sq->pending;
2744                         verbose(VERB_CLIENT, "serviced_delete: TCP");
2745                         log_assert(!(w->write_wait_queued && w->on_tcp_waiting_list));
2746                         /* if on stream-write-waiting list then
2747                          * remove from waiting list and waiting_tcp_delete */
2748                         if(w->write_wait_queued) {
2749                                 struct pending_tcp* pend =
2750                                         (struct pending_tcp*)w->next_waiting;
2751                                 verbose(VERB_CLIENT, "serviced_delete: writewait");
2752                                 if(!w->in_cb_and_decommission)
2753                                         reuse_tree_by_id_delete(&pend->reuse, w);
2754                                 reuse_write_wait_remove(&pend->reuse, w);
2755                                 if(!w->in_cb_and_decommission)
2756                                         waiting_tcp_delete(w);
2757                         } else if(!w->on_tcp_waiting_list) {
2758                                 struct pending_tcp* pend =
2759                                         (struct pending_tcp*)w->next_waiting;
2760                                 verbose(VERB_CLIENT, "serviced_delete: tcpreusekeep");
2761                                 /* w needs to stay on tree_by_id to not assign
2762                                  * the same ID; remove the callback since its
2763                                  * serviced_query will be gone. */
2764                                 w->cb = NULL;
2765                                 if(!reuse_tcp_remove_serviced_keep(w, sq)) {
2766                                         if(!w->in_cb_and_decommission)
2767                                                 reuse_cb_and_decommission(sq->outnet,
2768                                                         pend, NETEVENT_CLOSED);
2769                                         use_free_buffer(sq->outnet);
2770                                 }
2771                                 sq->pending = NULL;
2772                         } else {
2773                                 verbose(VERB_CLIENT, "serviced_delete: tcpwait");
2774                                 outnet_waiting_tcp_list_remove(sq->outnet, w);
2775                                 if(!w->in_cb_and_decommission)
2776                                         waiting_tcp_delete(w);
2777                         }
2778                 }
2779         }
2780         /* does not delete from tree, caller has to do that */
2781         serviced_node_del(&sq->node, NULL);
2782 }
2783
2784 /** perturb a dname capitalization randomly */
2785 static void
2786 serviced_perturb_qname(struct ub_randstate* rnd, uint8_t* qbuf, size_t len)
2787 {
2788         uint8_t lablen;
2789         uint8_t* d = qbuf + 10;
2790         long int random = 0;
2791         int bits = 0;
2792         log_assert(len >= 10 + 5 /* offset qname, root, qtype, qclass */);
2793         (void)len;
2794         lablen = *d++;
2795         while(lablen) {
2796                 while(lablen--) {
2797                         /* only perturb A-Z, a-z */
2798                         if(isalpha((unsigned char)*d)) {
2799                                 /* get a random bit */  
2800                                 if(bits == 0) {
2801                                         random = ub_random(rnd);
2802                                         bits = 30;
2803                                 }
2804                                 if(random & 0x1) {
2805                                         *d = (uint8_t)toupper((unsigned char)*d);
2806                                 } else {
2807                                         *d = (uint8_t)tolower((unsigned char)*d);
2808                                 }
2809                                 random >>= 1;
2810                                 bits--;
2811                         }
2812                         d++;
2813                 }
2814                 lablen = *d++;
2815         }
2816         if(verbosity >= VERB_ALGO) {
2817                 char buf[LDNS_MAX_DOMAINLEN+1];
2818                 dname_str(qbuf+10, buf);
2819                 verbose(VERB_ALGO, "qname perturbed to %s", buf);
2820         }
2821 }
2822
2823 /** put serviced query into a buffer */
2824 static void
2825 serviced_encode(struct serviced_query* sq, sldns_buffer* buff, int with_edns)
2826 {
2827         /* if we are using 0x20 bits for ID randomness, perturb them */
2828         if(sq->outnet->use_caps_for_id && !sq->nocaps) {
2829                 serviced_perturb_qname(sq->outnet->rnd, sq->qbuf, sq->qbuflen);
2830         }
2831         /* generate query */
2832         sldns_buffer_clear(buff);
2833         sldns_buffer_write_u16(buff, 0); /* id placeholder */
2834         sldns_buffer_write(buff, sq->qbuf, sq->qbuflen);
2835         sldns_buffer_flip(buff);
2836         if(with_edns) {
2837                 /* add edns section */
2838                 struct edns_data edns;
2839                 struct edns_option padding_option;
2840                 edns.edns_present = 1;
2841                 edns.ext_rcode = 0;
2842                 edns.edns_version = EDNS_ADVERTISED_VERSION;
2843                 edns.opt_list_in = NULL;
2844                 edns.opt_list_out = sq->opt_list;
2845                 edns.opt_list_inplace_cb_out = NULL;
2846                 if(sq->status == serviced_query_UDP_EDNS_FRAG) {
2847                         if(addr_is_ip6(&sq->addr, sq->addrlen)) {
2848                                 if(EDNS_FRAG_SIZE_IP6 < EDNS_ADVERTISED_SIZE)
2849                                         edns.udp_size = EDNS_FRAG_SIZE_IP6;
2850                                 else    edns.udp_size = EDNS_ADVERTISED_SIZE;
2851                         } else {
2852                                 if(EDNS_FRAG_SIZE_IP4 < EDNS_ADVERTISED_SIZE)
2853                                         edns.udp_size = EDNS_FRAG_SIZE_IP4;
2854                                 else    edns.udp_size = EDNS_ADVERTISED_SIZE;
2855                         }
2856                 } else {
2857                         edns.udp_size = EDNS_ADVERTISED_SIZE;
2858                 }
2859                 edns.bits = 0;
2860                 if(sq->dnssec & EDNS_DO)
2861                         edns.bits = EDNS_DO;
2862                 if(sq->dnssec & BIT_CD)
2863                         LDNS_CD_SET(sldns_buffer_begin(buff));
2864                 if (sq->ssl_upstream && sq->padding_block_size) {
2865                         padding_option.opt_code = LDNS_EDNS_PADDING;
2866                         padding_option.opt_len = 0;
2867                         padding_option.opt_data = NULL;
2868                         padding_option.next = edns.opt_list_out;
2869                         edns.opt_list_out = &padding_option;
2870                         edns.padding_block_size = sq->padding_block_size;
2871                 }
2872                 attach_edns_record(buff, &edns);
2873         }
2874 }
2875
2876 /**
2877  * Perform serviced query UDP sending operation.
2878  * Sends UDP with EDNS, unless infra host marked non EDNS.
2879  * @param sq: query to send.
2880  * @param buff: buffer scratch space.
2881  * @return 0 on error.
2882  */
2883 static int
2884 serviced_udp_send(struct serviced_query* sq, sldns_buffer* buff)
2885 {
2886         int rtt, vs;
2887         uint8_t edns_lame_known;
2888         time_t now = *sq->outnet->now_secs;
2889
2890         if(!infra_host(sq->outnet->infra, &sq->addr, sq->addrlen, sq->zone,
2891                 sq->zonelen, now, &vs, &edns_lame_known, &rtt))
2892                 return 0;
2893         sq->last_rtt = rtt;
2894         verbose(VERB_ALGO, "EDNS lookup known=%d vs=%d", edns_lame_known, vs);
2895         if(sq->status == serviced_initial) {
2896                 if(vs != -1) {
2897                         sq->status = serviced_query_UDP_EDNS;
2898                 } else {        
2899                         sq->status = serviced_query_UDP; 
2900                 }
2901         }
2902         serviced_encode(sq, buff, (sq->status == serviced_query_UDP_EDNS) ||
2903                 (sq->status == serviced_query_UDP_EDNS_FRAG));
2904         sq->last_sent_time = *sq->outnet->now_tv;
2905         sq->edns_lame_known = (int)edns_lame_known;
2906         verbose(VERB_ALGO, "serviced query UDP timeout=%d msec", rtt);
2907         sq->pending = pending_udp_query(sq, buff, rtt,
2908                 serviced_udp_callback, sq);
2909         if(!sq->pending)
2910                 return 0;
2911         return 1;
2912 }
2913
2914 /** check that perturbed qname is identical */
2915 static int
2916 serviced_check_qname(sldns_buffer* pkt, uint8_t* qbuf, size_t qbuflen)
2917 {
2918         uint8_t* d1 = sldns_buffer_begin(pkt)+12;
2919         uint8_t* d2 = qbuf+10;
2920         uint8_t len1, len2;
2921         int count = 0;
2922         if(sldns_buffer_limit(pkt) < 12+1+4) /* packet too small for qname */
2923                 return 0;
2924         log_assert(qbuflen >= 15 /* 10 header, root, type, class */);
2925         len1 = *d1++;
2926         len2 = *d2++;
2927         while(len1 != 0 || len2 != 0) {
2928                 if(LABEL_IS_PTR(len1)) {
2929                         /* check if we can read *d1 with compression ptr rest */
2930                         if(d1 >= sldns_buffer_at(pkt, sldns_buffer_limit(pkt)))
2931                                 return 0;
2932                         d1 = sldns_buffer_begin(pkt)+PTR_OFFSET(len1, *d1);
2933                         /* check if we can read the destination *d1 */
2934                         if(d1 >= sldns_buffer_at(pkt, sldns_buffer_limit(pkt)))
2935                                 return 0;
2936                         len1 = *d1++;
2937                         if(count++ > MAX_COMPRESS_PTRS)
2938                                 return 0;
2939                         continue;
2940                 }
2941                 if(d2 > qbuf+qbuflen)
2942                         return 0;
2943                 if(len1 != len2)
2944                         return 0;
2945                 if(len1 > LDNS_MAX_LABELLEN)
2946                         return 0;
2947                 /* check len1 + 1(next length) are okay to read */
2948                 if(d1+len1 >= sldns_buffer_at(pkt, sldns_buffer_limit(pkt)))
2949                         return 0;
2950                 log_assert(len1 <= LDNS_MAX_LABELLEN);
2951                 log_assert(len2 <= LDNS_MAX_LABELLEN);
2952                 log_assert(len1 == len2 && len1 != 0);
2953                 /* compare the labels - bitwise identical */
2954                 if(memcmp(d1, d2, len1) != 0)
2955                         return 0;
2956                 d1 += len1;
2957                 d2 += len2;
2958                 len1 = *d1++;
2959                 len2 = *d2++;
2960         }
2961         return 1;
2962 }
2963
2964 /** call the callbacks for a serviced query */
2965 static void
2966 serviced_callbacks(struct serviced_query* sq, int error, struct comm_point* c,
2967         struct comm_reply* rep)
2968 {
2969         struct service_callback* p;
2970         int dobackup = (sq->cblist && sq->cblist->next); /* >1 cb*/
2971         uint8_t *backup_p = NULL;
2972         size_t backlen = 0;
2973 #ifdef UNBOUND_DEBUG
2974         rbnode_type* rem =
2975 #else
2976         (void)
2977 #endif
2978         /* remove from tree, and schedule for deletion, so that callbacks
2979          * can safely deregister themselves and even create new serviced
2980          * queries that are identical to this one. */
2981         rbtree_delete(sq->outnet->serviced, sq);
2982         log_assert(rem); /* should have been present */
2983         sq->to_be_deleted = 1; 
2984         verbose(VERB_ALGO, "svcd callbacks start");
2985         if(sq->outnet->use_caps_for_id && error == NETEVENT_NOERROR && c &&
2986                 !sq->nocaps && sq->qtype != LDNS_RR_TYPE_PTR) {
2987                 /* for type PTR do not check perturbed name in answer,
2988                  * compatibility with cisco dns guard boxes that mess up
2989                  * reverse queries 0x20 contents */
2990                 /* noerror and nxdomain must have a qname in reply */
2991                 if(sldns_buffer_read_u16_at(c->buffer, 4) == 0 &&
2992                         (LDNS_RCODE_WIRE(sldns_buffer_begin(c->buffer))
2993                                 == LDNS_RCODE_NOERROR || 
2994                          LDNS_RCODE_WIRE(sldns_buffer_begin(c->buffer))
2995                                 == LDNS_RCODE_NXDOMAIN)) {
2996                         verbose(VERB_DETAIL, "no qname in reply to check 0x20ID");
2997                         log_addr(VERB_DETAIL, "from server", 
2998                                 &sq->addr, sq->addrlen);
2999                         log_buf(VERB_DETAIL, "for packet", c->buffer);
3000                         error = NETEVENT_CLOSED;
3001                         c = NULL;
3002                 } else if(sldns_buffer_read_u16_at(c->buffer, 4) > 0 &&
3003                         !serviced_check_qname(c->buffer, sq->qbuf, 
3004                         sq->qbuflen)) {
3005                         verbose(VERB_DETAIL, "wrong 0x20-ID in reply qname");
3006                         log_addr(VERB_DETAIL, "from server", 
3007                                 &sq->addr, sq->addrlen);
3008                         log_buf(VERB_DETAIL, "for packet", c->buffer);
3009                         error = NETEVENT_CAPSFAIL;
3010                         /* and cleanup too */
3011                         pkt_dname_tolower(c->buffer, 
3012                                 sldns_buffer_at(c->buffer, 12));
3013                 } else {
3014                         verbose(VERB_ALGO, "good 0x20-ID in reply qname");
3015                         /* cleanup caps, prettier cache contents. */
3016                         pkt_dname_tolower(c->buffer, 
3017                                 sldns_buffer_at(c->buffer, 12));
3018                 }
3019         }
3020         if(dobackup && c) {
3021                 /* make a backup of the query, since the querystate processing
3022                  * may send outgoing queries that overwrite the buffer.
3023                  * use secondary buffer to store the query.
3024                  * This is a data copy, but faster than packet to server */
3025                 backlen = sldns_buffer_limit(c->buffer);
3026                 backup_p = regional_alloc_init(sq->region,
3027                         sldns_buffer_begin(c->buffer), backlen);
3028                 if(!backup_p) {
3029                         log_err("malloc failure in serviced query callbacks");
3030                         error = NETEVENT_CLOSED;
3031                         c = NULL;
3032                 }
3033                 sq->outnet->svcd_overhead = backlen;
3034         }
3035         /* test the actual sq->cblist, because the next elem could be deleted*/
3036         while((p=sq->cblist) != NULL) {
3037                 sq->cblist = p->next; /* remove this element */
3038                 if(dobackup && c) {
3039                         sldns_buffer_clear(c->buffer);
3040                         sldns_buffer_write(c->buffer, backup_p, backlen);
3041                         sldns_buffer_flip(c->buffer);
3042                 }
3043                 fptr_ok(fptr_whitelist_serviced_query(p->cb));
3044                 (void)(*p->cb)(c, p->cb_arg, error, rep);
3045         }
3046         if(backup_p) {
3047                 sq->outnet->svcd_overhead = 0;
3048         }
3049         verbose(VERB_ALGO, "svcd callbacks end");
3050         log_assert(sq->cblist == NULL);
3051         serviced_delete(sq);
3052 }
3053
3054 int 
3055 serviced_tcp_callback(struct comm_point* c, void* arg, int error,
3056         struct comm_reply* rep)
3057 {
3058         struct serviced_query* sq = (struct serviced_query*)arg;
3059         struct comm_reply r2;
3060 #ifdef USE_DNSTAP
3061         struct waiting_tcp* w = (struct waiting_tcp*)sq->pending;
3062         struct pending_tcp* pend_tcp = NULL;
3063         struct port_if* pi = NULL;
3064         if(w && !w->on_tcp_waiting_list && w->next_waiting) {
3065                 pend_tcp = (struct pending_tcp*)w->next_waiting;
3066                 pi = pend_tcp->pi;
3067         }
3068 #endif
3069         sq->pending = NULL; /* removed after this callback */
3070         if(error != NETEVENT_NOERROR)
3071                 log_addr(VERB_QUERY, "tcp error for address", 
3072                         &sq->addr, sq->addrlen);
3073         if(error==NETEVENT_NOERROR)
3074                 infra_update_tcp_works(sq->outnet->infra, &sq->addr,
3075                         sq->addrlen, sq->zone, sq->zonelen);
3076 #ifdef USE_DNSTAP
3077         /*
3078          * sending src (local service)/dst (upstream) addresses over DNSTAP
3079          */
3080         if(error==NETEVENT_NOERROR && pi && sq->outnet->dtenv &&
3081            (sq->outnet->dtenv->log_resolver_response_messages ||
3082             sq->outnet->dtenv->log_forwarder_response_messages)) {
3083                 log_addr(VERB_ALGO, "response from upstream", &sq->addr, sq->addrlen);
3084                 log_addr(VERB_ALGO, "to local addr", &pi->addr, pi->addrlen);
3085                 dt_msg_send_outside_response(sq->outnet->dtenv, &sq->addr,
3086                         &pi->addr, c->type, sq->zone, sq->zonelen, sq->qbuf,
3087                         sq->qbuflen, &sq->last_sent_time, sq->outnet->now_tv,
3088                         c->buffer);
3089         }
3090 #endif
3091         if(error==NETEVENT_NOERROR && sq->status == serviced_query_TCP_EDNS &&
3092                 (LDNS_RCODE_WIRE(sldns_buffer_begin(c->buffer)) == 
3093                 LDNS_RCODE_FORMERR || LDNS_RCODE_WIRE(sldns_buffer_begin(
3094                 c->buffer)) == LDNS_RCODE_NOTIMPL) ) {
3095                 /* attempt to fallback to nonEDNS */
3096                 sq->status = serviced_query_TCP_EDNS_fallback;
3097                 serviced_tcp_initiate(sq, c->buffer);
3098                 return 0;
3099         } else if(error==NETEVENT_NOERROR && 
3100                 sq->status == serviced_query_TCP_EDNS_fallback &&
3101                         (LDNS_RCODE_WIRE(sldns_buffer_begin(c->buffer)) == 
3102                         LDNS_RCODE_NOERROR || LDNS_RCODE_WIRE(
3103                         sldns_buffer_begin(c->buffer)) == LDNS_RCODE_NXDOMAIN 
3104                         || LDNS_RCODE_WIRE(sldns_buffer_begin(c->buffer)) 
3105                         == LDNS_RCODE_YXDOMAIN)) {
3106                 /* the fallback produced a result that looks promising, note
3107                  * that this server should be approached without EDNS */
3108                 /* only store noEDNS in cache if domain is noDNSSEC */
3109                 if(!sq->want_dnssec)
3110                   if(!infra_edns_update(sq->outnet->infra, &sq->addr, 
3111                         sq->addrlen, sq->zone, sq->zonelen, -1,
3112                         *sq->outnet->now_secs))
3113                         log_err("Out of memory caching no edns for host");
3114                 sq->status = serviced_query_TCP;
3115         }
3116         if(sq->tcp_upstream || sq->ssl_upstream) {
3117             struct timeval now = *sq->outnet->now_tv;
3118             if(error!=NETEVENT_NOERROR) {
3119                 if(!infra_rtt_update(sq->outnet->infra, &sq->addr,
3120                     sq->addrlen, sq->zone, sq->zonelen, sq->qtype,
3121                     -1, sq->last_rtt, (time_t)now.tv_sec))
3122                     log_err("out of memory in TCP exponential backoff.");
3123             } else if(now.tv_sec > sq->last_sent_time.tv_sec ||
3124                 (now.tv_sec == sq->last_sent_time.tv_sec &&
3125                 now.tv_usec > sq->last_sent_time.tv_usec)) {
3126                 /* convert from microseconds to milliseconds */
3127                 int roundtime = ((int)(now.tv_sec - sq->last_sent_time.tv_sec))*1000
3128                   + ((int)now.tv_usec - (int)sq->last_sent_time.tv_usec)/1000;
3129                 verbose(VERB_ALGO, "measured TCP-time at %d msec", roundtime);
3130                 log_assert(roundtime >= 0);
3131                 /* only store if less then AUTH_TIMEOUT seconds, it could be
3132                  * huge due to system-hibernated and we woke up */
3133                 if(roundtime < 60000) {
3134                     if(!infra_rtt_update(sq->outnet->infra, &sq->addr,
3135                         sq->addrlen, sq->zone, sq->zonelen, sq->qtype,
3136                         roundtime, sq->last_rtt, (time_t)now.tv_sec))
3137                         log_err("out of memory noting rtt.");
3138                 }
3139             }
3140         }
3141         /* insert address into reply info */
3142         if(!rep) {
3143                 /* create one if there isn't (on errors) */
3144                 rep = &r2;
3145                 r2.c = c;
3146         }
3147         memcpy(&rep->remote_addr, &sq->addr, sq->addrlen);
3148         rep->remote_addrlen = sq->addrlen;
3149         serviced_callbacks(sq, error, c, rep);
3150         return 0;
3151 }
3152
3153 static void
3154 serviced_tcp_initiate(struct serviced_query* sq, sldns_buffer* buff)
3155 {
3156         verbose(VERB_ALGO, "initiate TCP query %s", 
3157                 sq->status==serviced_query_TCP_EDNS?"EDNS":"");
3158         serviced_encode(sq, buff, sq->status == serviced_query_TCP_EDNS);
3159         sq->last_sent_time = *sq->outnet->now_tv;
3160         log_assert(!sq->busy);
3161         sq->busy = 1;
3162         sq->pending = pending_tcp_query(sq, buff, sq->outnet->tcp_auth_query_timeout,
3163                 serviced_tcp_callback, sq);
3164         sq->busy = 0;
3165         if(!sq->pending) {
3166                 /* delete from tree so that a retry by above layer does not
3167                  * clash with this entry */
3168                 verbose(VERB_ALGO, "serviced_tcp_initiate: failed to send tcp query");
3169                 serviced_callbacks(sq, NETEVENT_CLOSED, NULL, NULL);
3170         }
3171 }
3172
3173 /** Send serviced query over TCP return false on initial failure */
3174 static int
3175 serviced_tcp_send(struct serviced_query* sq, sldns_buffer* buff)
3176 {
3177         int vs, rtt, timeout;
3178         uint8_t edns_lame_known;
3179         if(!infra_host(sq->outnet->infra, &sq->addr, sq->addrlen, sq->zone,
3180                 sq->zonelen, *sq->outnet->now_secs, &vs, &edns_lame_known,
3181                 &rtt))
3182                 return 0;
3183         sq->last_rtt = rtt;
3184         if(vs != -1)
3185                 sq->status = serviced_query_TCP_EDNS;
3186         else    sq->status = serviced_query_TCP;
3187         serviced_encode(sq, buff, sq->status == serviced_query_TCP_EDNS);
3188         sq->last_sent_time = *sq->outnet->now_tv;
3189         if(sq->tcp_upstream || sq->ssl_upstream) {
3190                 timeout = rtt;
3191                 if(rtt >= UNKNOWN_SERVER_NICENESS && rtt < sq->outnet->tcp_auth_query_timeout)
3192                         timeout = sq->outnet->tcp_auth_query_timeout;
3193         } else {
3194                 timeout = sq->outnet->tcp_auth_query_timeout;
3195         }
3196         log_assert(!sq->busy);
3197         sq->busy = 1;
3198         sq->pending = pending_tcp_query(sq, buff, timeout,
3199                 serviced_tcp_callback, sq);
3200         sq->busy = 0;
3201         return sq->pending != NULL;
3202 }
3203
3204 /* see if packet is edns malformed; got zeroes at start.
3205  * This is from servers that return malformed packets to EDNS0 queries,
3206  * but they return good packets for nonEDNS0 queries.
3207  * We try to detect their output; without resorting to a full parse or
3208  * check for too many bytes after the end of the packet. */
3209 static int
3210 packet_edns_malformed(struct sldns_buffer* buf, int qtype)
3211 {
3212         size_t len;
3213         if(sldns_buffer_limit(buf) < LDNS_HEADER_SIZE)
3214                 return 1; /* malformed */
3215         /* they have NOERROR rcode, 1 answer. */
3216         if(LDNS_RCODE_WIRE(sldns_buffer_begin(buf)) != LDNS_RCODE_NOERROR)
3217                 return 0;
3218         /* one query (to skip) and answer records */
3219         if(LDNS_QDCOUNT(sldns_buffer_begin(buf)) != 1 ||
3220                 LDNS_ANCOUNT(sldns_buffer_begin(buf)) == 0)
3221                 return 0;
3222         /* skip qname */
3223         len = dname_valid(sldns_buffer_at(buf, LDNS_HEADER_SIZE),
3224                 sldns_buffer_limit(buf)-LDNS_HEADER_SIZE);
3225         if(len == 0)
3226                 return 0;
3227         if(len == 1 && qtype == 0)
3228                 return 0; /* we asked for '.' and type 0 */
3229         /* and then 4 bytes (type and class of query) */
3230         if(sldns_buffer_limit(buf) < LDNS_HEADER_SIZE + len + 4 + 3)
3231                 return 0;
3232
3233         /* and start with 11 zeroes as the answer RR */
3234         /* so check the qtype of the answer record, qname=0, type=0 */
3235         if(sldns_buffer_at(buf, LDNS_HEADER_SIZE+len+4)[0] == 0 &&
3236            sldns_buffer_at(buf, LDNS_HEADER_SIZE+len+4)[1] == 0 &&
3237            sldns_buffer_at(buf, LDNS_HEADER_SIZE+len+4)[2] == 0)
3238                 return 1;
3239         return 0;
3240 }
3241
3242 int 
3243 serviced_udp_callback(struct comm_point* c, void* arg, int error,
3244         struct comm_reply* rep)
3245 {
3246         struct serviced_query* sq = (struct serviced_query*)arg;
3247         struct outside_network* outnet = sq->outnet;
3248         struct timeval now = *sq->outnet->now_tv;
3249 #ifdef USE_DNSTAP
3250         struct pending* p = (struct pending*)sq->pending;
3251 #endif
3252
3253         sq->pending = NULL; /* removed after callback */
3254         if(error == NETEVENT_TIMEOUT) {
3255                 if(sq->status == serviced_query_UDP_EDNS && sq->last_rtt < 5000) {
3256                         /* fallback to 1480/1280 */
3257                         sq->status = serviced_query_UDP_EDNS_FRAG;
3258                         log_name_addr(VERB_ALGO, "try edns1xx0", sq->qbuf+10,
3259                                 &sq->addr, sq->addrlen);
3260                         if(!serviced_udp_send(sq, c->buffer)) {
3261                                 serviced_callbacks(sq, NETEVENT_CLOSED, c, rep);
3262                         }
3263                         return 0;
3264                 }
3265                 if(sq->status == serviced_query_UDP_EDNS_FRAG) {
3266                         /* fragmentation size did not fix it */
3267                         sq->status = serviced_query_UDP_EDNS;
3268                 }
3269                 sq->retry++;
3270                 if(!infra_rtt_update(outnet->infra, &sq->addr, sq->addrlen,
3271                         sq->zone, sq->zonelen, sq->qtype, -1, sq->last_rtt,
3272                         (time_t)now.tv_sec))
3273                         log_err("out of memory in UDP exponential backoff");
3274                 if(sq->retry < OUTBOUND_UDP_RETRY) {
3275                         log_name_addr(VERB_ALGO, "retry query", sq->qbuf+10,
3276                                 &sq->addr, sq->addrlen);
3277                         if(!serviced_udp_send(sq, c->buffer)) {
3278                                 serviced_callbacks(sq, NETEVENT_CLOSED, c, rep);
3279                         }
3280                         return 0;
3281                 }
3282         }
3283         if(error != NETEVENT_NOERROR) {
3284                 /* udp returns error (due to no ID or interface available) */
3285                 serviced_callbacks(sq, error, c, rep);
3286                 return 0;
3287         }
3288 #ifdef USE_DNSTAP
3289         /*
3290          * sending src (local service)/dst (upstream) addresses over DNSTAP
3291          */
3292         if(error == NETEVENT_NOERROR && outnet->dtenv && p->pc &&
3293                 (outnet->dtenv->log_resolver_response_messages ||
3294                 outnet->dtenv->log_forwarder_response_messages)) {
3295                 log_addr(VERB_ALGO, "response from upstream", &sq->addr, sq->addrlen);
3296                 log_addr(VERB_ALGO, "to local addr", &p->pc->pif->addr,
3297                         p->pc->pif->addrlen);
3298                 dt_msg_send_outside_response(outnet->dtenv, &sq->addr,
3299                         &p->pc->pif->addr, c->type, sq->zone, sq->zonelen,
3300                         sq->qbuf, sq->qbuflen, &sq->last_sent_time,
3301                         sq->outnet->now_tv, c->buffer);
3302         }
3303 #endif
3304         if( (sq->status == serviced_query_UDP_EDNS 
3305                 ||sq->status == serviced_query_UDP_EDNS_FRAG)
3306                 && (LDNS_RCODE_WIRE(sldns_buffer_begin(c->buffer)) 
3307                         == LDNS_RCODE_FORMERR || LDNS_RCODE_WIRE(
3308                         sldns_buffer_begin(c->buffer)) == LDNS_RCODE_NOTIMPL
3309                     || packet_edns_malformed(c->buffer, sq->qtype)
3310                         )) {
3311                 /* try to get an answer by falling back without EDNS */
3312                 verbose(VERB_ALGO, "serviced query: attempt without EDNS");
3313                 sq->status = serviced_query_UDP_EDNS_fallback;
3314                 sq->retry = 0;
3315                 if(!serviced_udp_send(sq, c->buffer)) {
3316                         serviced_callbacks(sq, NETEVENT_CLOSED, c, rep);
3317                 }
3318                 return 0;
3319         } else if(sq->status == serviced_query_UDP_EDNS && 
3320                 !sq->edns_lame_known) {
3321                 /* now we know that edns queries received answers store that */
3322                 log_addr(VERB_ALGO, "serviced query: EDNS works for",
3323                         &sq->addr, sq->addrlen);
3324                 if(!infra_edns_update(outnet->infra, &sq->addr, sq->addrlen, 
3325                         sq->zone, sq->zonelen, 0, (time_t)now.tv_sec)) {
3326                         log_err("Out of memory caching edns works");
3327                 }
3328                 sq->edns_lame_known = 1;
3329         } else if(sq->status == serviced_query_UDP_EDNS_fallback &&
3330                 !sq->edns_lame_known && (LDNS_RCODE_WIRE(
3331                 sldns_buffer_begin(c->buffer)) == LDNS_RCODE_NOERROR || 
3332                 LDNS_RCODE_WIRE(sldns_buffer_begin(c->buffer)) == 
3333                 LDNS_RCODE_NXDOMAIN || LDNS_RCODE_WIRE(sldns_buffer_begin(
3334                 c->buffer)) == LDNS_RCODE_YXDOMAIN)) {
3335                 /* the fallback produced a result that looks promising, note
3336                  * that this server should be approached without EDNS */
3337                 /* only store noEDNS in cache if domain is noDNSSEC */
3338                 if(!sq->want_dnssec) {
3339                   log_addr(VERB_ALGO, "serviced query: EDNS fails for",
3340                         &sq->addr, sq->addrlen);
3341                   if(!infra_edns_update(outnet->infra, &sq->addr, sq->addrlen,
3342                         sq->zone, sq->zonelen, -1, (time_t)now.tv_sec)) {
3343                         log_err("Out of memory caching no edns for host");
3344                   }
3345                 } else {
3346                   log_addr(VERB_ALGO, "serviced query: EDNS fails, but "
3347                         "not stored because need DNSSEC for", &sq->addr,
3348                         sq->addrlen);
3349                 }
3350                 sq->status = serviced_query_UDP;
3351         }
3352         if(now.tv_sec > sq->last_sent_time.tv_sec ||
3353                 (now.tv_sec == sq->last_sent_time.tv_sec &&
3354                 now.tv_usec > sq->last_sent_time.tv_usec)) {
3355                 /* convert from microseconds to milliseconds */
3356                 int roundtime = ((int)(now.tv_sec - sq->last_sent_time.tv_sec))*1000
3357                   + ((int)now.tv_usec - (int)sq->last_sent_time.tv_usec)/1000;
3358                 verbose(VERB_ALGO, "measured roundtrip at %d msec", roundtime);
3359                 log_assert(roundtime >= 0);
3360                 /* in case the system hibernated, do not enter a huge value,
3361                  * above this value gives trouble with server selection */
3362                 if(roundtime < 60000) {
3363                     if(!infra_rtt_update(outnet->infra, &sq->addr, sq->addrlen, 
3364                         sq->zone, sq->zonelen, sq->qtype, roundtime,
3365                         sq->last_rtt, (time_t)now.tv_sec))
3366                         log_err("out of memory noting rtt.");
3367                 }
3368         }
3369         /* perform TC flag check and TCP fallback after updating our
3370          * cache entries for EDNS status and RTT times */
3371         if(LDNS_TC_WIRE(sldns_buffer_begin(c->buffer))) {
3372                 /* fallback to TCP */
3373                 /* this discards partial UDP contents */
3374                 if(sq->status == serviced_query_UDP_EDNS ||
3375                         sq->status == serviced_query_UDP_EDNS_FRAG ||
3376                         sq->status == serviced_query_UDP_EDNS_fallback)
3377                         /* if we have unfinished EDNS_fallback, start again */
3378                         sq->status = serviced_query_TCP_EDNS;
3379                 else    sq->status = serviced_query_TCP;
3380                 serviced_tcp_initiate(sq, c->buffer);
3381                 return 0;
3382         }
3383         /* yay! an answer */
3384         serviced_callbacks(sq, error, c, rep);
3385         return 0;
3386 }
3387
3388 struct serviced_query* 
3389 outnet_serviced_query(struct outside_network* outnet,
3390         struct query_info* qinfo, uint16_t flags, int dnssec, int want_dnssec,
3391         int nocaps, int check_ratelimit, int tcp_upstream, int ssl_upstream,
3392         char* tls_auth_name, struct sockaddr_storage* addr, socklen_t addrlen,
3393         uint8_t* zone, size_t zonelen, struct module_qstate* qstate,
3394         comm_point_callback_type* callback, void* callback_arg,
3395         sldns_buffer* buff, struct module_env* env, int* was_ratelimited)
3396 {
3397         struct serviced_query* sq;
3398         struct service_callback* cb;
3399         struct edns_string_addr* client_string_addr;
3400         struct regional* region;
3401         struct edns_option* backed_up_opt_list = qstate->edns_opts_back_out;
3402         struct edns_option* per_upstream_opt_list = NULL;
3403         time_t timenow = 0;
3404
3405         /* If we have an already populated EDNS option list make a copy since
3406          * we may now add upstream specific EDNS options. */
3407         /* Use a region that could be attached to a serviced_query, if it needs
3408          * to be created. If an existing one is found then this region will be
3409          * destroyed here. */
3410         region = alloc_reg_obtain(env->alloc);
3411         if(!region) return NULL;
3412         if(qstate->edns_opts_back_out) {
3413                 per_upstream_opt_list = edns_opt_copy_region(
3414                         qstate->edns_opts_back_out, region);
3415                 if(!per_upstream_opt_list) {
3416                         alloc_reg_release(env->alloc, region);
3417                         return NULL;
3418                 }
3419                 qstate->edns_opts_back_out = per_upstream_opt_list;
3420         }
3421
3422         if(!inplace_cb_query_call(env, qinfo, flags, addr, addrlen, zone,
3423                 zonelen, qstate, region)) {
3424                 alloc_reg_release(env->alloc, region);
3425                 return NULL;
3426         }
3427         /* Restore the option list; we can explicitly use the copied one from
3428          * now on. */
3429         per_upstream_opt_list = qstate->edns_opts_back_out;
3430         qstate->edns_opts_back_out = backed_up_opt_list;
3431
3432         if((client_string_addr = edns_string_addr_lookup(
3433                 &env->edns_strings->client_strings, addr, addrlen))) {
3434                 edns_opt_list_append(&per_upstream_opt_list,
3435                         env->edns_strings->client_string_opcode,
3436                         client_string_addr->string_len,
3437                         client_string_addr->string, region);
3438         }
3439
3440         serviced_gen_query(buff, qinfo->qname, qinfo->qname_len, qinfo->qtype,
3441                 qinfo->qclass, flags);
3442         sq = lookup_serviced(outnet, buff, dnssec, addr, addrlen,
3443                 per_upstream_opt_list);
3444         if(!sq) {
3445                 /* Check ratelimit only for new serviced_query */
3446                 if(check_ratelimit) {
3447                         timenow = *env->now;
3448                         if(!infra_ratelimit_inc(env->infra_cache, zone,
3449                                 zonelen, timenow, env->cfg->ratelimit_backoff,
3450                                 &qstate->qinfo, qstate->reply)) {
3451                                 /* Can we pass through with slip factor? */
3452                                 if(env->cfg->ratelimit_factor == 0 ||
3453                                         ub_random_max(env->rnd,
3454                                         env->cfg->ratelimit_factor) != 1) {
3455                                         *was_ratelimited = 1;
3456                                         alloc_reg_release(env->alloc, region);
3457                                         return NULL;
3458                                 }
3459                                 log_nametypeclass(VERB_ALGO,
3460                                         "ratelimit allowed through for "
3461                                         "delegation point", zone,
3462                                         LDNS_RR_TYPE_NS, LDNS_RR_CLASS_IN);
3463                         }
3464                 }
3465                 /* make new serviced query entry */
3466                 sq = serviced_create(outnet, buff, dnssec, want_dnssec, nocaps,
3467                         tcp_upstream, ssl_upstream, tls_auth_name, addr,
3468                         addrlen, zone, zonelen, (int)qinfo->qtype,
3469                         per_upstream_opt_list,
3470                         ( ssl_upstream && env->cfg->pad_queries
3471                         ? env->cfg->pad_queries_block_size : 0 ),
3472                         env->alloc, region);
3473                 if(!sq) {
3474                         if(check_ratelimit) {
3475                                 infra_ratelimit_dec(env->infra_cache,
3476                                         zone, zonelen, timenow);
3477                         }
3478                         return NULL;
3479                 }
3480                 if(!(cb = (struct service_callback*)regional_alloc(
3481                         sq->region, sizeof(*cb)))) {
3482                         if(check_ratelimit) {
3483                                 infra_ratelimit_dec(env->infra_cache,
3484                                         zone, zonelen, timenow);
3485                         }
3486                         (void)rbtree_delete(outnet->serviced, sq);
3487                         serviced_node_del(&sq->node, NULL);
3488                         return NULL;
3489                 }
3490                 /* No network action at this point; it will be invoked with the
3491                  * serviced_query timer instead to run outside of the mesh. */
3492         } else {
3493                 /* We don't need this region anymore. */
3494                 alloc_reg_release(env->alloc, region);
3495                 /* duplicate entries are included in the callback list, because
3496                  * there is a counterpart registration by our caller that needs
3497                  * to be doubly-removed (with callbacks perhaps). */
3498                 if(!(cb = (struct service_callback*)regional_alloc(
3499                         sq->region, sizeof(*cb)))) {
3500                         return NULL;
3501                 }
3502         }
3503         /* add callback to list of callbacks */
3504         cb->cb = callback;
3505         cb->cb_arg = callback_arg;
3506         cb->next = sq->cblist;
3507         sq->cblist = cb;
3508         return sq;
3509 }
3510
3511 /** remove callback from list */
3512 static void
3513 callback_list_remove(struct serviced_query* sq, void* cb_arg)
3514 {
3515         struct service_callback** pp = &sq->cblist;
3516         while(*pp) {
3517                 if((*pp)->cb_arg == cb_arg) {
3518                         struct service_callback* del = *pp;
3519                         *pp = del->next;
3520                         return;
3521                 }
3522                 pp = &(*pp)->next;
3523         }
3524 }
3525
3526 void outnet_serviced_query_stop(struct serviced_query* sq, void* cb_arg)
3527 {
3528         if(!sq)
3529                 return;
3530         callback_list_remove(sq, cb_arg);
3531         /* if callbacks() routine scheduled deletion, let it do that */
3532         if(!sq->cblist && !sq->busy && !sq->to_be_deleted) {
3533                 (void)rbtree_delete(sq->outnet->serviced, sq);
3534                 serviced_delete(sq);
3535         }
3536 }
3537
3538 /** create fd to send to this destination */
3539 static int
3540 fd_for_dest(struct outside_network* outnet, struct sockaddr_storage* to_addr,
3541         socklen_t to_addrlen)
3542 {
3543         struct sockaddr_storage* addr;
3544         socklen_t addrlen;
3545         int i, try, pnum, dscp;
3546         struct port_if* pif;
3547
3548         /* create fd */
3549         dscp = outnet->ip_dscp;
3550         for(try = 0; try<1000; try++) {
3551                 int port = 0;
3552                 int freebind = 0;
3553                 int noproto = 0;
3554                 int inuse = 0;
3555                 int fd = -1;
3556
3557                 /* select interface */
3558                 if(addr_is_ip6(to_addr, to_addrlen)) {
3559                         if(outnet->num_ip6 == 0) {
3560                                 char to[64];
3561                                 addr_to_str(to_addr, to_addrlen, to, sizeof(to));
3562                                 verbose(VERB_QUERY, "need ipv6 to send, but no ipv6 outgoing interfaces, for %s", to);
3563                                 return -1;
3564                         }
3565                         i = ub_random_max(outnet->rnd, outnet->num_ip6);
3566                         pif = &outnet->ip6_ifs[i];
3567                 } else {
3568                         if(outnet->num_ip4 == 0) {
3569                                 char to[64];
3570                                 addr_to_str(to_addr, to_addrlen, to, sizeof(to));
3571                                 verbose(VERB_QUERY, "need ipv4 to send, but no ipv4 outgoing interfaces, for %s", to);
3572                                 return -1;
3573                         }
3574                         i = ub_random_max(outnet->rnd, outnet->num_ip4);
3575                         pif = &outnet->ip4_ifs[i];
3576                 }
3577                 addr = &pif->addr;
3578                 addrlen = pif->addrlen;
3579 #ifndef DISABLE_EXPLICIT_PORT_RANDOMISATION
3580                 pnum = ub_random_max(outnet->rnd, pif->avail_total);
3581                 if(pnum < pif->inuse) {
3582                         /* port already open */
3583                         port = pif->out[pnum]->number;
3584                 } else {
3585                         /* unused ports in start part of array */
3586                         port = pif->avail_ports[pnum - pif->inuse];
3587                 }
3588 #else
3589                 pnum = port = 0;
3590 #endif
3591                 if(addr_is_ip6(to_addr, to_addrlen)) {
3592                         struct sockaddr_in6 sa = *(struct sockaddr_in6*)addr;
3593                         sa.sin6_port = (in_port_t)htons((uint16_t)port);
3594                         fd = create_udp_sock(AF_INET6, SOCK_DGRAM,
3595                                 (struct sockaddr*)&sa, addrlen, 1, &inuse, &noproto,
3596                                 0, 0, 0, NULL, 0, freebind, 0, dscp);
3597                 } else {
3598                         struct sockaddr_in* sa = (struct sockaddr_in*)addr;
3599                         sa->sin_port = (in_port_t)htons((uint16_t)port);
3600                         fd = create_udp_sock(AF_INET, SOCK_DGRAM, 
3601                                 (struct sockaddr*)addr, addrlen, 1, &inuse, &noproto,
3602                                 0, 0, 0, NULL, 0, freebind, 0, dscp);
3603                 }
3604                 if(fd != -1) {
3605                         return fd;
3606                 }
3607                 if(!inuse) {
3608                         return -1;
3609                 }
3610         }
3611         /* too many tries */
3612         log_err("cannot send probe, ports are in use");
3613         return -1;
3614 }
3615
3616 struct comm_point*
3617 outnet_comm_point_for_udp(struct outside_network* outnet,
3618         comm_point_callback_type* cb, void* cb_arg,
3619         struct sockaddr_storage* to_addr, socklen_t to_addrlen)
3620 {
3621         struct comm_point* cp;
3622         int fd = fd_for_dest(outnet, to_addr, to_addrlen);
3623         if(fd == -1) {
3624                 return NULL;
3625         }
3626         cp = comm_point_create_udp(outnet->base, fd, outnet->udp_buff, 0,
3627                 cb, cb_arg, NULL);
3628         if(!cp) {
3629                 log_err("malloc failure");
3630                 close(fd);
3631                 return NULL;
3632         }
3633         return cp;
3634 }
3635
3636 /** setup SSL for comm point */
3637 static int
3638 setup_comm_ssl(struct comm_point* cp, struct outside_network* outnet,
3639         int fd, char* host)
3640 {
3641         cp->ssl = outgoing_ssl_fd(outnet->sslctx, fd);
3642         if(!cp->ssl) {
3643                 log_err("cannot create SSL object");
3644                 return 0;
3645         }
3646 #ifdef USE_WINSOCK
3647         comm_point_tcp_win_bio_cb(cp, cp->ssl);
3648 #endif
3649         cp->ssl_shake_state = comm_ssl_shake_write;
3650         /* https verification */
3651 #ifdef HAVE_SSL
3652         if(outnet->tls_use_sni) {
3653                 (void)SSL_set_tlsext_host_name(cp->ssl, host);
3654         }
3655 #endif
3656 #ifdef HAVE_SSL_SET1_HOST
3657         if((SSL_CTX_get_verify_mode(outnet->sslctx)&SSL_VERIFY_PEER)) {
3658                 /* because we set SSL_VERIFY_PEER, in netevent in
3659                  * ssl_handshake, it'll check if the certificate
3660                  * verification has succeeded */
3661                 /* SSL_VERIFY_PEER is set on the sslctx */
3662                 /* and the certificates to verify with are loaded into
3663                  * it with SSL_load_verify_locations or
3664                  * SSL_CTX_set_default_verify_paths */
3665                 /* setting the hostname makes openssl verify the
3666                  * host name in the x509 certificate in the
3667                  * SSL connection*/
3668                 if(!SSL_set1_host(cp->ssl, host)) {
3669                         log_err("SSL_set1_host failed");
3670                         return 0;
3671                 }
3672         }
3673 #elif defined(HAVE_X509_VERIFY_PARAM_SET1_HOST)
3674         /* openssl 1.0.2 has this function that can be used for
3675          * set1_host like verification */
3676         if((SSL_CTX_get_verify_mode(outnet->sslctx)&SSL_VERIFY_PEER)) {
3677                 X509_VERIFY_PARAM* param = SSL_get0_param(cp->ssl);
3678 #  ifdef X509_CHECK_FLAG_NO_PARTIAL_WILDCARDS
3679                 X509_VERIFY_PARAM_set_hostflags(param, X509_CHECK_FLAG_NO_PARTIAL_WILDCARDS);
3680 #  endif
3681                 if(!X509_VERIFY_PARAM_set1_host(param, host, strlen(host))) {
3682                         log_err("X509_VERIFY_PARAM_set1_host failed");
3683                         return 0;
3684                 }
3685         }
3686 #else
3687         (void)host;
3688 #endif /* HAVE_SSL_SET1_HOST */
3689         return 1;
3690 }
3691
3692 struct comm_point*
3693 outnet_comm_point_for_tcp(struct outside_network* outnet,
3694         comm_point_callback_type* cb, void* cb_arg,
3695         struct sockaddr_storage* to_addr, socklen_t to_addrlen,
3696         sldns_buffer* query, int timeout, int ssl, char* host)
3697 {
3698         struct comm_point* cp;
3699         int fd = outnet_get_tcp_fd(to_addr, to_addrlen, outnet->tcp_mss, outnet->ip_dscp);
3700         if(fd == -1) {
3701                 return 0;
3702         }
3703         fd_set_nonblock(fd);
3704         if(!outnet_tcp_connect(fd, to_addr, to_addrlen)) {
3705                 /* outnet_tcp_connect has closed fd on error for us */
3706                 return 0;
3707         }
3708         cp = comm_point_create_tcp_out(outnet->base, 65552, cb, cb_arg);
3709         if(!cp) {
3710                 log_err("malloc failure");
3711                 close(fd);
3712                 return 0;
3713         }
3714         cp->repinfo.remote_addrlen = to_addrlen;
3715         memcpy(&cp->repinfo.remote_addr, to_addr, to_addrlen);
3716
3717         /* setup for SSL (if needed) */
3718         if(ssl) {
3719                 if(!setup_comm_ssl(cp, outnet, fd, host)) {
3720                         log_err("cannot setup XoT");
3721                         comm_point_delete(cp);
3722                         return NULL;
3723                 }
3724         }
3725
3726         /* set timeout on TCP connection */
3727         comm_point_start_listening(cp, fd, timeout);
3728         /* copy scratch buffer to cp->buffer */
3729         sldns_buffer_copy(cp->buffer, query);
3730         return cp;
3731 }
3732
3733 /** setup the User-Agent HTTP header based on http-user-agent configuration */
3734 static void
3735 setup_http_user_agent(sldns_buffer* buf, struct config_file* cfg)
3736 {
3737         if(cfg->hide_http_user_agent) return;
3738         if(cfg->http_user_agent==NULL || cfg->http_user_agent[0] == 0) {
3739                 sldns_buffer_printf(buf, "User-Agent: %s/%s\r\n", PACKAGE_NAME,
3740                         PACKAGE_VERSION);
3741         } else {
3742                 sldns_buffer_printf(buf, "User-Agent: %s\r\n", cfg->http_user_agent);
3743         }
3744 }
3745
3746 /** setup http request headers in buffer for sending query to destination */
3747 static int
3748 setup_http_request(sldns_buffer* buf, char* host, char* path,
3749         struct config_file* cfg)
3750 {
3751         sldns_buffer_clear(buf);
3752         sldns_buffer_printf(buf, "GET /%s HTTP/1.1\r\n", path);
3753         sldns_buffer_printf(buf, "Host: %s\r\n", host);
3754         setup_http_user_agent(buf, cfg);
3755         /* We do not really do multiple queries per connection,
3756          * but this header setting is also not needed.
3757          * sldns_buffer_printf(buf, "Connection: close\r\n") */
3758         sldns_buffer_printf(buf, "\r\n");
3759         if(sldns_buffer_position(buf)+10 > sldns_buffer_capacity(buf))
3760                 return 0; /* somehow buffer too short, but it is about 60K
3761                 and the request is only a couple bytes long. */
3762         sldns_buffer_flip(buf);
3763         return 1;
3764 }
3765
3766 struct comm_point*
3767 outnet_comm_point_for_http(struct outside_network* outnet,
3768         comm_point_callback_type* cb, void* cb_arg,
3769         struct sockaddr_storage* to_addr, socklen_t to_addrlen, int timeout,
3770         int ssl, char* host, char* path, struct config_file* cfg)
3771 {
3772         /* cp calls cb with err=NETEVENT_DONE when transfer is done */
3773         struct comm_point* cp;
3774         int fd = outnet_get_tcp_fd(to_addr, to_addrlen, outnet->tcp_mss, outnet->ip_dscp);
3775         if(fd == -1) {
3776                 return 0;
3777         }
3778         fd_set_nonblock(fd);
3779         if(!outnet_tcp_connect(fd, to_addr, to_addrlen)) {
3780                 /* outnet_tcp_connect has closed fd on error for us */
3781                 return 0;
3782         }
3783         cp = comm_point_create_http_out(outnet->base, 65552, cb, cb_arg,
3784                 outnet->udp_buff);
3785         if(!cp) {
3786                 log_err("malloc failure");
3787                 close(fd);
3788                 return 0;
3789         }
3790         cp->repinfo.remote_addrlen = to_addrlen;
3791         memcpy(&cp->repinfo.remote_addr, to_addr, to_addrlen);
3792
3793         /* setup for SSL (if needed) */
3794         if(ssl) {
3795                 if(!setup_comm_ssl(cp, outnet, fd, host)) {
3796                         log_err("cannot setup https");
3797                         comm_point_delete(cp);
3798                         return NULL;
3799                 }
3800         }
3801
3802         /* set timeout on TCP connection */
3803         comm_point_start_listening(cp, fd, timeout);
3804
3805         /* setup http request in cp->buffer */
3806         if(!setup_http_request(cp->buffer, host, path, cfg)) {
3807                 log_err("error setting up http request");
3808                 comm_point_delete(cp);
3809                 return NULL;
3810         }
3811         return cp;
3812 }
3813
3814 /** get memory used by waiting tcp entry (in use or not) */
3815 static size_t
3816 waiting_tcp_get_mem(struct waiting_tcp* w)
3817 {
3818         size_t s;
3819         if(!w) return 0;
3820         s = sizeof(*w) + w->pkt_len;
3821         if(w->timer)
3822                 s += comm_timer_get_mem(w->timer);
3823         return s;
3824 }
3825
3826 /** get memory used by port if */
3827 static size_t
3828 if_get_mem(struct port_if* pif)
3829 {
3830         size_t s;
3831         int i;
3832         s = sizeof(*pif) +
3833 #ifndef DISABLE_EXPLICIT_PORT_RANDOMISATION
3834             sizeof(int)*pif->avail_total +
3835 #endif
3836                 sizeof(struct port_comm*)*pif->maxout;
3837         for(i=0; i<pif->inuse; i++)
3838                 s += sizeof(*pif->out[i]) + 
3839                         comm_point_get_mem(pif->out[i]->cp);
3840         return s;
3841 }
3842
3843 /** get memory used by waiting udp */
3844 static size_t
3845 waiting_udp_get_mem(struct pending* w)
3846 {
3847         size_t s;
3848         s = sizeof(*w) + comm_timer_get_mem(w->timer) + w->pkt_len;
3849         return s;
3850 }
3851
3852 size_t outnet_get_mem(struct outside_network* outnet)
3853 {
3854         size_t i;
3855         int k;
3856         struct waiting_tcp* w;
3857         struct pending* u;
3858         struct serviced_query* sq;
3859         struct service_callback* sb;
3860         struct port_comm* pc;
3861         size_t s = sizeof(*outnet) + sizeof(*outnet->base) + 
3862                 sizeof(*outnet->udp_buff) + 
3863                 sldns_buffer_capacity(outnet->udp_buff);
3864         /* second buffer is not ours */
3865         for(pc = outnet->unused_fds; pc; pc = pc->next) {
3866                 s += sizeof(*pc) + comm_point_get_mem(pc->cp);
3867         }
3868         for(k=0; k<outnet->num_ip4; k++)
3869                 s += if_get_mem(&outnet->ip4_ifs[k]);
3870         for(k=0; k<outnet->num_ip6; k++)
3871                 s += if_get_mem(&outnet->ip6_ifs[k]);
3872         for(u=outnet->udp_wait_first; u; u=u->next_waiting)
3873                 s += waiting_udp_get_mem(u);
3874         
3875         s += sizeof(struct pending_tcp*)*outnet->num_tcp;
3876         for(i=0; i<outnet->num_tcp; i++) {
3877                 s += sizeof(struct pending_tcp);
3878                 s += comm_point_get_mem(outnet->tcp_conns[i]->c);
3879                 if(outnet->tcp_conns[i]->query)
3880                         s += waiting_tcp_get_mem(outnet->tcp_conns[i]->query);
3881         }
3882         for(w=outnet->tcp_wait_first; w; w = w->next_waiting)
3883                 s += waiting_tcp_get_mem(w);
3884         s += sizeof(*outnet->pending);
3885         s += (sizeof(struct pending) + comm_timer_get_mem(NULL)) * 
3886                 outnet->pending->count;
3887         s += sizeof(*outnet->serviced);
3888         s += outnet->svcd_overhead;
3889         RBTREE_FOR(sq, struct serviced_query*, outnet->serviced) {
3890                 s += sizeof(*sq) + sq->qbuflen;
3891                 for(sb = sq->cblist; sb; sb = sb->next)
3892                         s += sizeof(*sb);
3893         }
3894         return s;
3895 }
3896
3897 size_t 
3898 serviced_get_mem(struct serviced_query* sq)
3899 {
3900         struct service_callback* sb;
3901         size_t s;
3902         s = sizeof(*sq) + sq->qbuflen;
3903         for(sb = sq->cblist; sb; sb = sb->next)
3904                 s += sizeof(*sb);
3905         if(sq->status == serviced_query_UDP_EDNS ||
3906                 sq->status == serviced_query_UDP ||
3907                 sq->status == serviced_query_UDP_EDNS_FRAG ||
3908                 sq->status == serviced_query_UDP_EDNS_fallback) {
3909                 s += sizeof(struct pending);
3910                 s += comm_timer_get_mem(NULL);
3911         } else {
3912                 /* does not have size of the pkt pointer */
3913                 /* always has a timer except on malloc failures */
3914
3915                 /* these sizes are part of the main outside network mem */
3916                 /*
3917                 s += sizeof(struct waiting_tcp);
3918                 s += comm_timer_get_mem(NULL);
3919                 */
3920         }
3921         return s;
3922 }
3923