]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/resolve/resolved-dns-stream.c
resolved: add new helper for carefully detach a stream from any server
[thirdparty/systemd.git] / src / resolve / resolved-dns-stream.c
1 /* SPDX-License-Identifier: LGPL-2.1+ */
2
3 #include <netinet/tcp.h>
4
5 #include "alloc-util.h"
6 #include "fd-util.h"
7 #include "io-util.h"
8 #include "missing.h"
9 #include "resolved-dns-stream.h"
10
11 #define DNS_STREAM_TIMEOUT_USEC (10 * USEC_PER_SEC)
12 #define DNS_STREAMS_MAX 128
13
14 static void dns_stream_stop(DnsStream *s) {
15 assert(s);
16
17 s->io_event_source = sd_event_source_unref(s->io_event_source);
18 s->timeout_event_source = sd_event_source_unref(s->timeout_event_source);
19 s->fd = safe_close(s->fd);
20
21 /* Disconnect us from the server object if we are now not usable anymore */
22 dns_stream_detach(s);
23 }
24
25 static int dns_stream_update_io(DnsStream *s) {
26 int f = 0;
27
28 assert(s);
29
30 if (s->write_packet && s->n_written < sizeof(s->write_size) + s->write_packet->size)
31 f |= EPOLLOUT;
32 else if (!ordered_set_isempty(s->write_queue)) {
33 dns_packet_unref(s->write_packet);
34 s->write_packet = ordered_set_steal_first(s->write_queue);
35 s->write_size = htobe16(s->write_packet->size);
36 s->n_written = 0;
37 f |= EPOLLOUT;
38 }
39 if (!s->read_packet || s->n_read < sizeof(s->read_size) + s->read_packet->size)
40 f |= EPOLLIN;
41
42 #if ENABLE_DNS_OVER_TLS
43 /* For handshake and clean closing purposes, TLS can override requested events */
44 if (s->dnstls_events)
45 f = s->dnstls_events;
46 #endif
47
48 return sd_event_source_set_io_events(s->io_event_source, f);
49 }
50
51 static int dns_stream_complete(DnsStream *s, int error) {
52 _cleanup_(dns_stream_unrefp) _unused_ DnsStream *ref = dns_stream_ref(s); /* Protect stream while we process it */
53
54 assert(s);
55
56 #if ENABLE_DNS_OVER_TLS
57 if (s->encrypted) {
58 int r;
59
60 r = dnstls_stream_shutdown(s, error);
61 if (r != -EAGAIN)
62 dns_stream_stop(s);
63 } else
64 #endif
65 dns_stream_stop(s);
66
67 if (s->complete)
68 s->complete(s, error);
69 else /* the default action if no completion function is set is to close the stream */
70 dns_stream_unref(s);
71
72 return 0;
73 }
74
75 static int dns_stream_identify(DnsStream *s) {
76 union {
77 struct cmsghdr header; /* For alignment */
78 uint8_t buffer[CMSG_SPACE(MAXSIZE(struct in_pktinfo, struct in6_pktinfo))
79 + EXTRA_CMSG_SPACE /* kernel appears to require extra space */];
80 } control;
81 struct msghdr mh = {};
82 struct cmsghdr *cmsg;
83 socklen_t sl;
84 int r;
85
86 assert(s);
87
88 if (s->identified)
89 return 0;
90
91 /* Query the local side */
92 s->local_salen = sizeof(s->local);
93 r = getsockname(s->fd, &s->local.sa, &s->local_salen);
94 if (r < 0)
95 return -errno;
96 if (s->local.sa.sa_family == AF_INET6 && s->ifindex <= 0)
97 s->ifindex = s->local.in6.sin6_scope_id;
98
99 /* Query the remote side */
100 s->peer_salen = sizeof(s->peer);
101 r = getpeername(s->fd, &s->peer.sa, &s->peer_salen);
102 if (r < 0)
103 return -errno;
104 if (s->peer.sa.sa_family == AF_INET6 && s->ifindex <= 0)
105 s->ifindex = s->peer.in6.sin6_scope_id;
106
107 /* Check consistency */
108 assert(s->peer.sa.sa_family == s->local.sa.sa_family);
109 assert(IN_SET(s->peer.sa.sa_family, AF_INET, AF_INET6));
110
111 /* Query connection meta information */
112 sl = sizeof(control);
113 if (s->peer.sa.sa_family == AF_INET) {
114 r = getsockopt(s->fd, IPPROTO_IP, IP_PKTOPTIONS, &control, &sl);
115 if (r < 0)
116 return -errno;
117 } else if (s->peer.sa.sa_family == AF_INET6) {
118
119 r = getsockopt(s->fd, IPPROTO_IPV6, IPV6_2292PKTOPTIONS, &control, &sl);
120 if (r < 0)
121 return -errno;
122 } else
123 return -EAFNOSUPPORT;
124
125 mh.msg_control = &control;
126 mh.msg_controllen = sl;
127
128 CMSG_FOREACH(cmsg, &mh) {
129
130 if (cmsg->cmsg_level == IPPROTO_IPV6) {
131 assert(s->peer.sa.sa_family == AF_INET6);
132
133 switch (cmsg->cmsg_type) {
134
135 case IPV6_PKTINFO: {
136 struct in6_pktinfo *i = (struct in6_pktinfo*) CMSG_DATA(cmsg);
137
138 if (s->ifindex <= 0)
139 s->ifindex = i->ipi6_ifindex;
140 break;
141 }
142
143 case IPV6_HOPLIMIT:
144 s->ttl = *(int *) CMSG_DATA(cmsg);
145 break;
146 }
147
148 } else if (cmsg->cmsg_level == IPPROTO_IP) {
149 assert(s->peer.sa.sa_family == AF_INET);
150
151 switch (cmsg->cmsg_type) {
152
153 case IP_PKTINFO: {
154 struct in_pktinfo *i = (struct in_pktinfo*) CMSG_DATA(cmsg);
155
156 if (s->ifindex <= 0)
157 s->ifindex = i->ipi_ifindex;
158 break;
159 }
160
161 case IP_TTL:
162 s->ttl = *(int *) CMSG_DATA(cmsg);
163 break;
164 }
165 }
166 }
167
168 /* The Linux kernel sets the interface index to the loopback
169 * device if the connection came from the local host since it
170 * avoids the routing table in such a case. Let's unset the
171 * interface index in such a case. */
172 if (s->ifindex == LOOPBACK_IFINDEX)
173 s->ifindex = 0;
174
175 /* If we don't know the interface index still, we look for the
176 * first local interface with a matching address. Yuck! */
177 if (s->ifindex <= 0)
178 s->ifindex = manager_find_ifindex(s->manager, s->local.sa.sa_family, s->local.sa.sa_family == AF_INET ? (union in_addr_union*) &s->local.in.sin_addr : (union in_addr_union*) &s->local.in6.sin6_addr);
179
180 if (s->protocol == DNS_PROTOCOL_LLMNR && s->ifindex > 0) {
181 uint32_t ifindex = htobe32(s->ifindex);
182
183 /* Make sure all packets for this connection are sent on the same interface */
184 if (s->local.sa.sa_family == AF_INET) {
185 r = setsockopt(s->fd, IPPROTO_IP, IP_UNICAST_IF, &ifindex, sizeof(ifindex));
186 if (r < 0)
187 log_debug_errno(errno, "Failed to invoke IP_UNICAST_IF: %m");
188 } else if (s->local.sa.sa_family == AF_INET6) {
189 r = setsockopt(s->fd, IPPROTO_IPV6, IPV6_UNICAST_IF, &ifindex, sizeof(ifindex));
190 if (r < 0)
191 log_debug_errno(errno, "Failed to invoke IPV6_UNICAST_IF: %m");
192 }
193 }
194
195 s->identified = true;
196
197 return 0;
198 }
199
200 ssize_t dns_stream_writev(DnsStream *s, const struct iovec *iov, size_t iovcnt, int flags) {
201 ssize_t m;
202
203 assert(s);
204 assert(iov);
205
206 #if ENABLE_DNS_OVER_TLS
207 if (s->encrypted && !(flags & DNS_STREAM_WRITE_TLS_DATA)) {
208 ssize_t ss;
209 size_t i;
210
211 m = 0;
212 for (i = 0; i < iovcnt; i++) {
213 ss = dnstls_stream_write(s, iov[i].iov_base, iov[i].iov_len);
214 if (ss < 0)
215 return ss;
216
217 m += ss;
218 if (ss != (ssize_t) iov[i].iov_len)
219 continue;
220 }
221 } else
222 #endif
223 if (s->tfo_salen > 0) {
224 struct msghdr hdr = {
225 .msg_iov = (struct iovec*) iov,
226 .msg_iovlen = iovcnt,
227 .msg_name = &s->tfo_address.sa,
228 .msg_namelen = s->tfo_salen
229 };
230
231 m = sendmsg(s->fd, &hdr, MSG_FASTOPEN);
232 if (m < 0) {
233 if (errno == EOPNOTSUPP) {
234 s->tfo_salen = 0;
235 if (connect(s->fd, &s->tfo_address.sa, s->tfo_salen) < 0)
236 return -errno;
237
238 return -EAGAIN;
239 }
240 if (errno == EINPROGRESS)
241 return -EAGAIN;
242
243 return -errno;
244 } else
245 s->tfo_salen = 0; /* connection is made */
246 } else {
247 m = writev(s->fd, iov, iovcnt);
248 if (m < 0)
249 return -errno;
250 }
251
252 return m;
253 }
254
255 static ssize_t dns_stream_read(DnsStream *s, void *buf, size_t count) {
256 ssize_t ss;
257
258 #if ENABLE_DNS_OVER_TLS
259 if (s->encrypted)
260 ss = dnstls_stream_read(s, buf, count);
261 else
262 #endif
263 {
264 ss = read(s->fd, buf, count);
265 if (ss < 0)
266 return -errno;
267 }
268
269 return ss;
270 }
271
272 static int on_stream_timeout(sd_event_source *es, usec_t usec, void *userdata) {
273 DnsStream *s = userdata;
274
275 assert(s);
276
277 return dns_stream_complete(s, ETIMEDOUT);
278 }
279
280 static int on_stream_io(sd_event_source *es, int fd, uint32_t revents, void *userdata) {
281 _cleanup_(dns_stream_unrefp) DnsStream *s = dns_stream_ref(userdata); /* Protect stream while we process it */
282 int r;
283
284 assert(s);
285
286 #if ENABLE_DNS_OVER_TLS
287 if (s->encrypted) {
288 r = dnstls_stream_on_io(s, revents);
289 if (r == DNSTLS_STREAM_CLOSED)
290 return 0;
291 if (r == -EAGAIN)
292 return dns_stream_update_io(s);
293 if (r < 0)
294 return dns_stream_complete(s, -r);
295
296 r = dns_stream_update_io(s);
297 if (r < 0)
298 return r;
299 }
300 #endif
301
302 /* only identify after connecting */
303 if (s->tfo_salen == 0) {
304 r = dns_stream_identify(s);
305 if (r < 0)
306 return dns_stream_complete(s, -r);
307 }
308
309 if ((revents & EPOLLOUT) &&
310 s->write_packet &&
311 s->n_written < sizeof(s->write_size) + s->write_packet->size) {
312
313 struct iovec iov[2];
314 ssize_t ss;
315
316 iov[0] = IOVEC_MAKE(&s->write_size, sizeof(s->write_size));
317 iov[1] = IOVEC_MAKE(DNS_PACKET_DATA(s->write_packet), s->write_packet->size);
318
319 IOVEC_INCREMENT(iov, 2, s->n_written);
320
321 ss = dns_stream_writev(s, iov, 2, 0);
322 if (ss < 0) {
323 if (!IN_SET(-ss, EINTR, EAGAIN))
324 return dns_stream_complete(s, -ss);
325 } else
326 s->n_written += ss;
327
328 /* Are we done? If so, disable the event source for EPOLLOUT */
329 if (s->n_written >= sizeof(s->write_size) + s->write_packet->size) {
330 r = dns_stream_update_io(s);
331 if (r < 0)
332 return dns_stream_complete(s, -r);
333 }
334 }
335
336 if ((revents & (EPOLLIN|EPOLLHUP|EPOLLRDHUP)) &&
337 (!s->read_packet ||
338 s->n_read < sizeof(s->read_size) + s->read_packet->size)) {
339
340 if (s->n_read < sizeof(s->read_size)) {
341 ssize_t ss;
342
343 ss = dns_stream_read(s, (uint8_t*) &s->read_size + s->n_read, sizeof(s->read_size) - s->n_read);
344 if (ss < 0) {
345 if (!IN_SET(-ss, EINTR, EAGAIN))
346 return dns_stream_complete(s, -ss);
347 } else if (ss == 0)
348 return dns_stream_complete(s, ECONNRESET);
349 else
350 s->n_read += ss;
351 }
352
353 if (s->n_read >= sizeof(s->read_size)) {
354
355 if (be16toh(s->read_size) < DNS_PACKET_HEADER_SIZE)
356 return dns_stream_complete(s, EBADMSG);
357
358 if (s->n_read < sizeof(s->read_size) + be16toh(s->read_size)) {
359 ssize_t ss;
360
361 if (!s->read_packet) {
362 r = dns_packet_new(&s->read_packet, s->protocol, be16toh(s->read_size), DNS_PACKET_SIZE_MAX);
363 if (r < 0)
364 return dns_stream_complete(s, -r);
365
366 s->read_packet->size = be16toh(s->read_size);
367 s->read_packet->ipproto = IPPROTO_TCP;
368 s->read_packet->family = s->peer.sa.sa_family;
369 s->read_packet->ttl = s->ttl;
370 s->read_packet->ifindex = s->ifindex;
371
372 if (s->read_packet->family == AF_INET) {
373 s->read_packet->sender.in = s->peer.in.sin_addr;
374 s->read_packet->sender_port = be16toh(s->peer.in.sin_port);
375 s->read_packet->destination.in = s->local.in.sin_addr;
376 s->read_packet->destination_port = be16toh(s->local.in.sin_port);
377 } else {
378 assert(s->read_packet->family == AF_INET6);
379 s->read_packet->sender.in6 = s->peer.in6.sin6_addr;
380 s->read_packet->sender_port = be16toh(s->peer.in6.sin6_port);
381 s->read_packet->destination.in6 = s->local.in6.sin6_addr;
382 s->read_packet->destination_port = be16toh(s->local.in6.sin6_port);
383
384 if (s->read_packet->ifindex == 0)
385 s->read_packet->ifindex = s->peer.in6.sin6_scope_id;
386 if (s->read_packet->ifindex == 0)
387 s->read_packet->ifindex = s->local.in6.sin6_scope_id;
388 }
389 }
390
391 ss = dns_stream_read(s,
392 (uint8_t*) DNS_PACKET_DATA(s->read_packet) + s->n_read - sizeof(s->read_size),
393 sizeof(s->read_size) + be16toh(s->read_size) - s->n_read);
394 if (ss < 0) {
395 if (!IN_SET(-ss, EINTR, EAGAIN))
396 return dns_stream_complete(s, -ss);
397 } else if (ss == 0)
398 return dns_stream_complete(s, ECONNRESET);
399 else
400 s->n_read += ss;
401 }
402
403 /* Are we done? If so, disable the event source for EPOLLIN */
404 if (s->n_read >= sizeof(s->read_size) + be16toh(s->read_size)) {
405 /* If there's a packet handler
406 * installed, call that. Note that
407 * this is optional... */
408 if (s->on_packet) {
409 r = s->on_packet(s);
410 if (r < 0)
411 return r;
412 }
413
414 r = dns_stream_update_io(s);
415 if (r < 0)
416 return dns_stream_complete(s, -r);
417 }
418 }
419 }
420
421 if ((s->write_packet && s->n_written >= sizeof(s->write_size) + s->write_packet->size) &&
422 (s->read_packet && s->n_read >= sizeof(s->read_size) + s->read_packet->size))
423 return dns_stream_complete(s, 0);
424
425 return 0;
426 }
427
428 static DnsStream *dns_stream_free(DnsStream *s) {
429 DnsPacket *p;
430 Iterator i;
431
432 assert(s);
433
434 dns_stream_stop(s);
435
436 if (s->manager) {
437 LIST_REMOVE(streams, s->manager->dns_streams, s);
438 s->manager->n_dns_streams--;
439 }
440
441 #if ENABLE_DNS_OVER_TLS
442 if (s->encrypted)
443 dnstls_stream_free(s);
444 #endif
445
446 ORDERED_SET_FOREACH(p, s->write_queue, i)
447 dns_packet_unref(ordered_set_remove(s->write_queue, p));
448
449 dns_packet_unref(s->write_packet);
450 dns_packet_unref(s->read_packet);
451 dns_server_unref(s->server);
452
453 ordered_set_free(s->write_queue);
454
455 return mfree(s);
456 }
457
458 DEFINE_TRIVIAL_REF_UNREF_FUNC(DnsStream, dns_stream, dns_stream_free);
459
460 int dns_stream_new(
461 Manager *m,
462 DnsStream **ret,
463 DnsProtocol protocol,
464 int fd,
465 const union sockaddr_union *tfo_address) {
466
467 _cleanup_(dns_stream_unrefp) DnsStream *s = NULL;
468 int r;
469
470 assert(m);
471 assert(ret);
472 assert(fd >= 0);
473
474 if (m->n_dns_streams > DNS_STREAMS_MAX)
475 return -EBUSY;
476
477 s = new(DnsStream, 1);
478 if (!s)
479 return -ENOMEM;
480
481 *s = (DnsStream) {
482 .n_ref = 1,
483 .fd = -1,
484 .protocol = protocol,
485 };
486
487 r = ordered_set_ensure_allocated(&s->write_queue, &dns_packet_hash_ops);
488 if (r < 0)
489 return r;
490
491 r = sd_event_add_io(m->event, &s->io_event_source, fd, EPOLLIN, on_stream_io, s);
492 if (r < 0)
493 return r;
494
495 (void) sd_event_source_set_description(s->io_event_source, "dns-stream-io");
496
497 r = sd_event_add_time(
498 m->event,
499 &s->timeout_event_source,
500 clock_boottime_or_monotonic(),
501 now(clock_boottime_or_monotonic()) + DNS_STREAM_TIMEOUT_USEC, 0,
502 on_stream_timeout, s);
503 if (r < 0)
504 return r;
505
506 (void) sd_event_source_set_description(s->timeout_event_source, "dns-stream-timeout");
507
508 LIST_PREPEND(streams, m->dns_streams, s);
509 m->n_dns_streams++;
510 s->manager = m;
511
512 s->fd = fd;
513
514 if (tfo_address) {
515 s->tfo_address = *tfo_address;
516 s->tfo_salen = tfo_address->sa.sa_family == AF_INET6 ? sizeof(tfo_address->in6) : sizeof(tfo_address->in);
517 }
518
519 *ret = TAKE_PTR(s);
520
521 return 0;
522 }
523
524 int dns_stream_write_packet(DnsStream *s, DnsPacket *p) {
525 int r;
526
527 assert(s);
528 assert(p);
529
530 r = ordered_set_put(s->write_queue, p);
531 if (r < 0)
532 return r;
533
534 dns_packet_ref(p);
535
536 return dns_stream_update_io(s);
537 }
538
539 DnsPacket *dns_stream_take_read_packet(DnsStream *s) {
540 assert(s);
541
542 if (!s->read_packet)
543 return NULL;
544
545 if (s->n_read < sizeof(s->read_size))
546 return NULL;
547
548 if (s->n_read < sizeof(s->read_size) + be16toh(s->read_size))
549 return NULL;
550
551 s->n_read = 0;
552 return TAKE_PTR(s->read_packet);
553 }
554
555 void dns_stream_detach(DnsStream *s) {
556 assert(s);
557
558 if (!s->server)
559 return;
560
561 if (s->server->stream != s)
562 return;
563
564 dns_server_unref_stream(s->server);
565 }