]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/resolve/resolved-dns-stream.c
Merge pull request #11105 from keszybz/path-parsing
[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 dns_stream_detach(s);
68
69 if (s->complete)
70 s->complete(s, error);
71 else /* the default action if no completion function is set is to close the stream */
72 dns_stream_unref(s);
73
74 return 0;
75 }
76
77 static int dns_stream_identify(DnsStream *s) {
78 union {
79 struct cmsghdr header; /* For alignment */
80 uint8_t buffer[CMSG_SPACE(MAXSIZE(struct in_pktinfo, struct in6_pktinfo))
81 + EXTRA_CMSG_SPACE /* kernel appears to require extra space */];
82 } control;
83 struct msghdr mh = {};
84 struct cmsghdr *cmsg;
85 socklen_t sl;
86 int r;
87
88 assert(s);
89
90 if (s->identified)
91 return 0;
92
93 /* Query the local side */
94 s->local_salen = sizeof(s->local);
95 r = getsockname(s->fd, &s->local.sa, &s->local_salen);
96 if (r < 0)
97 return -errno;
98 if (s->local.sa.sa_family == AF_INET6 && s->ifindex <= 0)
99 s->ifindex = s->local.in6.sin6_scope_id;
100
101 /* Query the remote side */
102 s->peer_salen = sizeof(s->peer);
103 r = getpeername(s->fd, &s->peer.sa, &s->peer_salen);
104 if (r < 0)
105 return -errno;
106 if (s->peer.sa.sa_family == AF_INET6 && s->ifindex <= 0)
107 s->ifindex = s->peer.in6.sin6_scope_id;
108
109 /* Check consistency */
110 assert(s->peer.sa.sa_family == s->local.sa.sa_family);
111 assert(IN_SET(s->peer.sa.sa_family, AF_INET, AF_INET6));
112
113 /* Query connection meta information */
114 sl = sizeof(control);
115 if (s->peer.sa.sa_family == AF_INET) {
116 r = getsockopt(s->fd, IPPROTO_IP, IP_PKTOPTIONS, &control, &sl);
117 if (r < 0)
118 return -errno;
119 } else if (s->peer.sa.sa_family == AF_INET6) {
120
121 r = getsockopt(s->fd, IPPROTO_IPV6, IPV6_2292PKTOPTIONS, &control, &sl);
122 if (r < 0)
123 return -errno;
124 } else
125 return -EAFNOSUPPORT;
126
127 mh.msg_control = &control;
128 mh.msg_controllen = sl;
129
130 CMSG_FOREACH(cmsg, &mh) {
131
132 if (cmsg->cmsg_level == IPPROTO_IPV6) {
133 assert(s->peer.sa.sa_family == AF_INET6);
134
135 switch (cmsg->cmsg_type) {
136
137 case IPV6_PKTINFO: {
138 struct in6_pktinfo *i = (struct in6_pktinfo*) CMSG_DATA(cmsg);
139
140 if (s->ifindex <= 0)
141 s->ifindex = i->ipi6_ifindex;
142 break;
143 }
144
145 case IPV6_HOPLIMIT:
146 s->ttl = *(int *) CMSG_DATA(cmsg);
147 break;
148 }
149
150 } else if (cmsg->cmsg_level == IPPROTO_IP) {
151 assert(s->peer.sa.sa_family == AF_INET);
152
153 switch (cmsg->cmsg_type) {
154
155 case IP_PKTINFO: {
156 struct in_pktinfo *i = (struct in_pktinfo*) CMSG_DATA(cmsg);
157
158 if (s->ifindex <= 0)
159 s->ifindex = i->ipi_ifindex;
160 break;
161 }
162
163 case IP_TTL:
164 s->ttl = *(int *) CMSG_DATA(cmsg);
165 break;
166 }
167 }
168 }
169
170 /* The Linux kernel sets the interface index to the loopback
171 * device if the connection came from the local host since it
172 * avoids the routing table in such a case. Let's unset the
173 * interface index in such a case. */
174 if (s->ifindex == LOOPBACK_IFINDEX)
175 s->ifindex = 0;
176
177 /* If we don't know the interface index still, we look for the
178 * first local interface with a matching address. Yuck! */
179 if (s->ifindex <= 0)
180 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);
181
182 if (s->protocol == DNS_PROTOCOL_LLMNR && s->ifindex > 0) {
183 uint32_t ifindex = htobe32(s->ifindex);
184
185 /* Make sure all packets for this connection are sent on the same interface */
186 if (s->local.sa.sa_family == AF_INET) {
187 r = setsockopt(s->fd, IPPROTO_IP, IP_UNICAST_IF, &ifindex, sizeof(ifindex));
188 if (r < 0)
189 log_debug_errno(errno, "Failed to invoke IP_UNICAST_IF: %m");
190 } else if (s->local.sa.sa_family == AF_INET6) {
191 r = setsockopt(s->fd, IPPROTO_IPV6, IPV6_UNICAST_IF, &ifindex, sizeof(ifindex));
192 if (r < 0)
193 log_debug_errno(errno, "Failed to invoke IPV6_UNICAST_IF: %m");
194 }
195 }
196
197 s->identified = true;
198
199 return 0;
200 }
201
202 ssize_t dns_stream_writev(DnsStream *s, const struct iovec *iov, size_t iovcnt, int flags) {
203 ssize_t m;
204
205 assert(s);
206 assert(iov);
207
208 #if ENABLE_DNS_OVER_TLS
209 if (s->encrypted && !(flags & DNS_STREAM_WRITE_TLS_DATA)) {
210 ssize_t ss;
211 size_t i;
212
213 m = 0;
214 for (i = 0; i < iovcnt; i++) {
215 ss = dnstls_stream_write(s, iov[i].iov_base, iov[i].iov_len);
216 if (ss < 0)
217 return ss;
218
219 m += ss;
220 if (ss != (ssize_t) iov[i].iov_len)
221 continue;
222 }
223 } else
224 #endif
225 if (s->tfo_salen > 0) {
226 struct msghdr hdr = {
227 .msg_iov = (struct iovec*) iov,
228 .msg_iovlen = iovcnt,
229 .msg_name = &s->tfo_address.sa,
230 .msg_namelen = s->tfo_salen
231 };
232
233 m = sendmsg(s->fd, &hdr, MSG_FASTOPEN);
234 if (m < 0) {
235 if (errno == EOPNOTSUPP) {
236 s->tfo_salen = 0;
237 if (connect(s->fd, &s->tfo_address.sa, s->tfo_salen) < 0)
238 return -errno;
239
240 return -EAGAIN;
241 }
242 if (errno == EINPROGRESS)
243 return -EAGAIN;
244
245 return -errno;
246 } else
247 s->tfo_salen = 0; /* connection is made */
248 } else {
249 m = writev(s->fd, iov, iovcnt);
250 if (m < 0)
251 return -errno;
252 }
253
254 return m;
255 }
256
257 static ssize_t dns_stream_read(DnsStream *s, void *buf, size_t count) {
258 ssize_t ss;
259
260 #if ENABLE_DNS_OVER_TLS
261 if (s->encrypted)
262 ss = dnstls_stream_read(s, buf, count);
263 else
264 #endif
265 {
266 ss = read(s->fd, buf, count);
267 if (ss < 0)
268 return -errno;
269 }
270
271 return ss;
272 }
273
274 static int on_stream_timeout(sd_event_source *es, usec_t usec, void *userdata) {
275 DnsStream *s = userdata;
276
277 assert(s);
278
279 return dns_stream_complete(s, ETIMEDOUT);
280 }
281
282 static int on_stream_io(sd_event_source *es, int fd, uint32_t revents, void *userdata) {
283 _cleanup_(dns_stream_unrefp) DnsStream *s = dns_stream_ref(userdata); /* Protect stream while we process it */
284 int r;
285
286 assert(s);
287
288 #if ENABLE_DNS_OVER_TLS
289 if (s->encrypted) {
290 r = dnstls_stream_on_io(s, revents);
291 if (r == DNSTLS_STREAM_CLOSED)
292 return 0;
293 if (r == -EAGAIN)
294 return dns_stream_update_io(s);
295 if (r < 0)
296 return dns_stream_complete(s, -r);
297
298 r = dns_stream_update_io(s);
299 if (r < 0)
300 return r;
301 }
302 #endif
303
304 /* only identify after connecting */
305 if (s->tfo_salen == 0) {
306 r = dns_stream_identify(s);
307 if (r < 0)
308 return dns_stream_complete(s, -r);
309 }
310
311 if ((revents & EPOLLOUT) &&
312 s->write_packet &&
313 s->n_written < sizeof(s->write_size) + s->write_packet->size) {
314
315 struct iovec iov[2];
316 ssize_t ss;
317
318 iov[0] = IOVEC_MAKE(&s->write_size, sizeof(s->write_size));
319 iov[1] = IOVEC_MAKE(DNS_PACKET_DATA(s->write_packet), s->write_packet->size);
320
321 IOVEC_INCREMENT(iov, 2, s->n_written);
322
323 ss = dns_stream_writev(s, iov, 2, 0);
324 if (ss < 0) {
325 if (!IN_SET(-ss, EINTR, EAGAIN))
326 return dns_stream_complete(s, -ss);
327 } else
328 s->n_written += ss;
329
330 /* Are we done? If so, disable the event source for EPOLLOUT */
331 if (s->n_written >= sizeof(s->write_size) + s->write_packet->size) {
332 r = dns_stream_update_io(s);
333 if (r < 0)
334 return dns_stream_complete(s, -r);
335 }
336 }
337
338 if ((revents & (EPOLLIN|EPOLLHUP|EPOLLRDHUP)) &&
339 (!s->read_packet ||
340 s->n_read < sizeof(s->read_size) + s->read_packet->size)) {
341
342 if (s->n_read < sizeof(s->read_size)) {
343 ssize_t ss;
344
345 ss = dns_stream_read(s, (uint8_t*) &s->read_size + s->n_read, sizeof(s->read_size) - s->n_read);
346 if (ss < 0) {
347 if (!IN_SET(-ss, EINTR, EAGAIN))
348 return dns_stream_complete(s, -ss);
349 } else if (ss == 0)
350 return dns_stream_complete(s, ECONNRESET);
351 else
352 s->n_read += ss;
353 }
354
355 if (s->n_read >= sizeof(s->read_size)) {
356
357 if (be16toh(s->read_size) < DNS_PACKET_HEADER_SIZE)
358 return dns_stream_complete(s, EBADMSG);
359
360 if (s->n_read < sizeof(s->read_size) + be16toh(s->read_size)) {
361 ssize_t ss;
362
363 if (!s->read_packet) {
364 r = dns_packet_new(&s->read_packet, s->protocol, be16toh(s->read_size), DNS_PACKET_SIZE_MAX);
365 if (r < 0)
366 return dns_stream_complete(s, -r);
367
368 s->read_packet->size = be16toh(s->read_size);
369 s->read_packet->ipproto = IPPROTO_TCP;
370 s->read_packet->family = s->peer.sa.sa_family;
371 s->read_packet->ttl = s->ttl;
372 s->read_packet->ifindex = s->ifindex;
373
374 if (s->read_packet->family == AF_INET) {
375 s->read_packet->sender.in = s->peer.in.sin_addr;
376 s->read_packet->sender_port = be16toh(s->peer.in.sin_port);
377 s->read_packet->destination.in = s->local.in.sin_addr;
378 s->read_packet->destination_port = be16toh(s->local.in.sin_port);
379 } else {
380 assert(s->read_packet->family == AF_INET6);
381 s->read_packet->sender.in6 = s->peer.in6.sin6_addr;
382 s->read_packet->sender_port = be16toh(s->peer.in6.sin6_port);
383 s->read_packet->destination.in6 = s->local.in6.sin6_addr;
384 s->read_packet->destination_port = be16toh(s->local.in6.sin6_port);
385
386 if (s->read_packet->ifindex == 0)
387 s->read_packet->ifindex = s->peer.in6.sin6_scope_id;
388 if (s->read_packet->ifindex == 0)
389 s->read_packet->ifindex = s->local.in6.sin6_scope_id;
390 }
391 }
392
393 ss = dns_stream_read(s,
394 (uint8_t*) DNS_PACKET_DATA(s->read_packet) + s->n_read - sizeof(s->read_size),
395 sizeof(s->read_size) + be16toh(s->read_size) - s->n_read);
396 if (ss < 0) {
397 if (!IN_SET(-ss, EINTR, EAGAIN))
398 return dns_stream_complete(s, -ss);
399 } else if (ss == 0)
400 return dns_stream_complete(s, ECONNRESET);
401 else
402 s->n_read += ss;
403 }
404
405 /* Are we done? If so, disable the event source for EPOLLIN */
406 if (s->n_read >= sizeof(s->read_size) + be16toh(s->read_size)) {
407 /* If there's a packet handler
408 * installed, call that. Note that
409 * this is optional... */
410 if (s->on_packet) {
411 r = s->on_packet(s);
412 if (r < 0)
413 return r;
414 }
415
416 r = dns_stream_update_io(s);
417 if (r < 0)
418 return dns_stream_complete(s, -r);
419 }
420 }
421 }
422
423 if ((s->write_packet && s->n_written >= sizeof(s->write_size) + s->write_packet->size) &&
424 (s->read_packet && s->n_read >= sizeof(s->read_size) + s->read_packet->size))
425 return dns_stream_complete(s, 0);
426
427 return 0;
428 }
429
430 static DnsStream *dns_stream_free(DnsStream *s) {
431 DnsPacket *p;
432 Iterator i;
433
434 assert(s);
435
436 dns_stream_stop(s);
437
438 if (s->manager) {
439 LIST_REMOVE(streams, s->manager->dns_streams, s);
440 s->manager->n_dns_streams--;
441 }
442
443 #if ENABLE_DNS_OVER_TLS
444 if (s->encrypted)
445 dnstls_stream_free(s);
446 #endif
447
448 ORDERED_SET_FOREACH(p, s->write_queue, i)
449 dns_packet_unref(ordered_set_remove(s->write_queue, p));
450
451 dns_packet_unref(s->write_packet);
452 dns_packet_unref(s->read_packet);
453 dns_server_unref(s->server);
454
455 ordered_set_free(s->write_queue);
456
457 return mfree(s);
458 }
459
460 DEFINE_TRIVIAL_REF_UNREF_FUNC(DnsStream, dns_stream, dns_stream_free);
461
462 int dns_stream_new(
463 Manager *m,
464 DnsStream **ret,
465 DnsProtocol protocol,
466 int fd,
467 const union sockaddr_union *tfo_address) {
468
469 _cleanup_(dns_stream_unrefp) DnsStream *s = NULL;
470 int r;
471
472 assert(m);
473 assert(ret);
474 assert(fd >= 0);
475
476 if (m->n_dns_streams > DNS_STREAMS_MAX)
477 return -EBUSY;
478
479 s = new(DnsStream, 1);
480 if (!s)
481 return -ENOMEM;
482
483 *s = (DnsStream) {
484 .n_ref = 1,
485 .fd = -1,
486 .protocol = protocol,
487 };
488
489 r = ordered_set_ensure_allocated(&s->write_queue, &dns_packet_hash_ops);
490 if (r < 0)
491 return r;
492
493 r = sd_event_add_io(m->event, &s->io_event_source, fd, EPOLLIN, on_stream_io, s);
494 if (r < 0)
495 return r;
496
497 (void) sd_event_source_set_description(s->io_event_source, "dns-stream-io");
498
499 r = sd_event_add_time(
500 m->event,
501 &s->timeout_event_source,
502 clock_boottime_or_monotonic(),
503 now(clock_boottime_or_monotonic()) + DNS_STREAM_TIMEOUT_USEC, 0,
504 on_stream_timeout, s);
505 if (r < 0)
506 return r;
507
508 (void) sd_event_source_set_description(s->timeout_event_source, "dns-stream-timeout");
509
510 LIST_PREPEND(streams, m->dns_streams, s);
511 m->n_dns_streams++;
512 s->manager = m;
513
514 s->fd = fd;
515
516 if (tfo_address) {
517 s->tfo_address = *tfo_address;
518 s->tfo_salen = tfo_address->sa.sa_family == AF_INET6 ? sizeof(tfo_address->in6) : sizeof(tfo_address->in);
519 }
520
521 *ret = TAKE_PTR(s);
522
523 return 0;
524 }
525
526 int dns_stream_write_packet(DnsStream *s, DnsPacket *p) {
527 int r;
528
529 assert(s);
530 assert(p);
531
532 r = ordered_set_put(s->write_queue, p);
533 if (r < 0)
534 return r;
535
536 dns_packet_ref(p);
537
538 return dns_stream_update_io(s);
539 }
540
541 DnsPacket *dns_stream_take_read_packet(DnsStream *s) {
542 assert(s);
543
544 if (!s->read_packet)
545 return NULL;
546
547 if (s->n_read < sizeof(s->read_size))
548 return NULL;
549
550 if (s->n_read < sizeof(s->read_size) + be16toh(s->read_size))
551 return NULL;
552
553 s->n_read = 0;
554 return TAKE_PTR(s->read_packet);
555 }
556
557 void dns_stream_detach(DnsStream *s) {
558 assert(s);
559
560 if (!s->server)
561 return;
562
563 if (s->server->stream != s)
564 return;
565
566 dns_server_unref_stream(s->server);
567 }