]>
Commit | Line | Data |
---|---|---|
623a4c97 LP |
1 | /*** |
2 | This file is part of systemd. | |
3 | ||
4 | Copyright 2014 Lennart Poettering | |
5 | ||
6 | systemd is free software; you can redistribute it and/or modify it | |
7 | under the terms of the GNU Lesser General Public License as published by | |
8 | the Free Software Foundation; either version 2.1 of the License, or | |
9 | (at your option) any later version. | |
10 | ||
11 | systemd is distributed in the hope that it will be useful, but | |
12 | WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
14 | Lesser General Public License for more details. | |
15 | ||
16 | You should have received a copy of the GNU Lesser General Public License | |
17 | along with systemd; If not, see <http://www.gnu.org/licenses/>. | |
18 | ***/ | |
19 | ||
20 | #include <netinet/tcp.h> | |
21 | ||
b5efdb8a | 22 | #include "alloc-util.h" |
3ffd4af2 | 23 | #include "fd-util.h" |
afc5dbf3 | 24 | #include "io-util.h" |
623a4c97 LP |
25 | #include "missing.h" |
26 | #include "resolved-dns-stream.h" | |
27 | ||
28 | #define DNS_STREAM_TIMEOUT_USEC (10 * USEC_PER_SEC) | |
29 | #define DNS_STREAMS_MAX 128 | |
30 | ||
31 | static void dns_stream_stop(DnsStream *s) { | |
32 | assert(s); | |
33 | ||
34 | s->io_event_source = sd_event_source_unref(s->io_event_source); | |
35 | s->timeout_event_source = sd_event_source_unref(s->timeout_event_source); | |
36 | s->fd = safe_close(s->fd); | |
37 | } | |
38 | ||
39 | static int dns_stream_update_io(DnsStream *s) { | |
40 | int f = 0; | |
41 | ||
42 | assert(s); | |
43 | ||
44 | if (s->write_packet && s->n_written < sizeof(s->write_size) + s->write_packet->size) | |
45 | f |= EPOLLOUT; | |
46 | if (!s->read_packet || s->n_read < sizeof(s->read_size) + s->read_packet->size) | |
47 | f |= EPOLLIN; | |
48 | ||
49 | return sd_event_source_set_io_events(s->io_event_source, f); | |
50 | } | |
51 | ||
b914e211 | 52 | static int dns_stream_complete(DnsStream *s, int error) { |
623a4c97 LP |
53 | assert(s); |
54 | ||
55 | dns_stream_stop(s); | |
56 | ||
57 | if (s->complete) | |
58 | s->complete(s, error); | |
b30bf55d LP |
59 | else /* the default action if no completion function is set is to close the stream */ |
60 | dns_stream_unref(s); | |
623a4c97 LP |
61 | |
62 | return 0; | |
63 | } | |
64 | ||
b914e211 LP |
65 | static int dns_stream_identify(DnsStream *s) { |
66 | union { | |
67 | struct cmsghdr header; /* For alignment */ | |
40a1eebd | 68 | uint8_t buffer[CMSG_SPACE(MAXSIZE(struct in_pktinfo, struct in6_pktinfo)) |
b914e211 LP |
69 | + EXTRA_CMSG_SPACE /* kernel appears to require extra space */]; |
70 | } control; | |
71 | struct msghdr mh = {}; | |
72 | struct cmsghdr *cmsg; | |
73 | socklen_t sl; | |
74 | int r; | |
75 | ||
76 | assert(s); | |
77 | ||
78 | if (s->identified) | |
79 | return 0; | |
80 | ||
81 | /* Query the local side */ | |
82 | s->local_salen = sizeof(s->local); | |
83 | r = getsockname(s->fd, &s->local.sa, &s->local_salen); | |
84 | if (r < 0) | |
85 | return -errno; | |
86 | if (s->local.sa.sa_family == AF_INET6 && s->ifindex <= 0) | |
87 | s->ifindex = s->local.in6.sin6_scope_id; | |
88 | ||
89 | /* Query the remote side */ | |
90 | s->peer_salen = sizeof(s->peer); | |
91 | r = getpeername(s->fd, &s->peer.sa, &s->peer_salen); | |
92 | if (r < 0) | |
93 | return -errno; | |
94 | if (s->peer.sa.sa_family == AF_INET6 && s->ifindex <= 0) | |
95 | s->ifindex = s->peer.in6.sin6_scope_id; | |
96 | ||
97 | /* Check consistency */ | |
98 | assert(s->peer.sa.sa_family == s->local.sa.sa_family); | |
99 | assert(IN_SET(s->peer.sa.sa_family, AF_INET, AF_INET6)); | |
100 | ||
101 | /* Query connection meta information */ | |
102 | sl = sizeof(control); | |
103 | if (s->peer.sa.sa_family == AF_INET) { | |
104 | r = getsockopt(s->fd, IPPROTO_IP, IP_PKTOPTIONS, &control, &sl); | |
105 | if (r < 0) | |
106 | return -errno; | |
107 | } else if (s->peer.sa.sa_family == AF_INET6) { | |
108 | ||
109 | r = getsockopt(s->fd, IPPROTO_IPV6, IPV6_2292PKTOPTIONS, &control, &sl); | |
110 | if (r < 0) | |
111 | return -errno; | |
112 | } else | |
113 | return -EAFNOSUPPORT; | |
114 | ||
115 | mh.msg_control = &control; | |
116 | mh.msg_controllen = sl; | |
2a1288ff LP |
117 | |
118 | CMSG_FOREACH(cmsg, &mh) { | |
b914e211 LP |
119 | |
120 | if (cmsg->cmsg_level == IPPROTO_IPV6) { | |
121 | assert(s->peer.sa.sa_family == AF_INET6); | |
122 | ||
123 | switch (cmsg->cmsg_type) { | |
124 | ||
125 | case IPV6_PKTINFO: { | |
126 | struct in6_pktinfo *i = (struct in6_pktinfo*) CMSG_DATA(cmsg); | |
127 | ||
128 | if (s->ifindex <= 0) | |
129 | s->ifindex = i->ipi6_ifindex; | |
130 | break; | |
131 | } | |
132 | ||
133 | case IPV6_HOPLIMIT: | |
134 | s->ttl = *(int *) CMSG_DATA(cmsg); | |
135 | break; | |
136 | } | |
137 | ||
138 | } else if (cmsg->cmsg_level == IPPROTO_IP) { | |
139 | assert(s->peer.sa.sa_family == AF_INET); | |
140 | ||
141 | switch (cmsg->cmsg_type) { | |
142 | ||
143 | case IP_PKTINFO: { | |
144 | struct in_pktinfo *i = (struct in_pktinfo*) CMSG_DATA(cmsg); | |
145 | ||
146 | if (s->ifindex <= 0) | |
147 | s->ifindex = i->ipi_ifindex; | |
148 | break; | |
149 | } | |
150 | ||
151 | case IP_TTL: | |
152 | s->ttl = *(int *) CMSG_DATA(cmsg); | |
153 | break; | |
154 | } | |
155 | } | |
156 | } | |
157 | ||
158 | /* The Linux kernel sets the interface index to the loopback | |
159 | * device if the connection came from the local host since it | |
160 | * avoids the routing table in such a case. Let's unset the | |
161 | * interface index in such a case. */ | |
a5f03596 | 162 | if (s->ifindex == LOOPBACK_IFINDEX) |
b914e211 LP |
163 | s->ifindex = 0; |
164 | ||
165 | /* If we don't know the interface index still, we look for the | |
166 | * first local interface with a matching address. Yuck! */ | |
167 | if (s->ifindex <= 0) | |
168 | 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); | |
169 | ||
170 | if (s->protocol == DNS_PROTOCOL_LLMNR && s->ifindex > 0) { | |
171 | uint32_t ifindex = htobe32(s->ifindex); | |
172 | ||
173 | /* Make sure all packets for this connection are sent on the same interface */ | |
174 | if (s->local.sa.sa_family == AF_INET) { | |
175 | r = setsockopt(s->fd, IPPROTO_IP, IP_UNICAST_IF, &ifindex, sizeof(ifindex)); | |
176 | if (r < 0) | |
56f64d95 | 177 | log_debug_errno(errno, "Failed to invoke IP_UNICAST_IF: %m"); |
b914e211 LP |
178 | } else if (s->local.sa.sa_family == AF_INET6) { |
179 | r = setsockopt(s->fd, IPPROTO_IPV6, IPV6_UNICAST_IF, &ifindex, sizeof(ifindex)); | |
180 | if (r < 0) | |
56f64d95 | 181 | log_debug_errno(errno, "Failed to invoke IPV6_UNICAST_IF: %m"); |
b914e211 LP |
182 | } |
183 | } | |
184 | ||
185 | s->identified = true; | |
186 | ||
187 | return 0; | |
188 | } | |
189 | ||
623a4c97 LP |
190 | static int on_stream_timeout(sd_event_source *es, usec_t usec, void *userdata) { |
191 | DnsStream *s = userdata; | |
192 | ||
193 | assert(s); | |
194 | ||
b914e211 | 195 | return dns_stream_complete(s, ETIMEDOUT); |
623a4c97 LP |
196 | } |
197 | ||
198 | static int on_stream_io(sd_event_source *es, int fd, uint32_t revents, void *userdata) { | |
199 | DnsStream *s = userdata; | |
200 | int r; | |
201 | ||
202 | assert(s); | |
203 | ||
b914e211 LP |
204 | r = dns_stream_identify(s); |
205 | if (r < 0) | |
206 | return dns_stream_complete(s, -r); | |
207 | ||
623a4c97 LP |
208 | if ((revents & EPOLLOUT) && |
209 | s->write_packet && | |
210 | s->n_written < sizeof(s->write_size) + s->write_packet->size) { | |
211 | ||
212 | struct iovec iov[2]; | |
213 | ssize_t ss; | |
214 | ||
215 | iov[0].iov_base = &s->write_size; | |
216 | iov[0].iov_len = sizeof(s->write_size); | |
217 | iov[1].iov_base = DNS_PACKET_DATA(s->write_packet); | |
218 | iov[1].iov_len = s->write_packet->size; | |
219 | ||
220 | IOVEC_INCREMENT(iov, 2, s->n_written); | |
221 | ||
222 | ss = writev(fd, iov, 2); | |
223 | if (ss < 0) { | |
ec2ce0c5 | 224 | if (!IN_SET(errno, EINTR, EAGAIN)) |
b914e211 | 225 | return dns_stream_complete(s, errno); |
623a4c97 LP |
226 | } else |
227 | s->n_written += ss; | |
228 | ||
229 | /* Are we done? If so, disable the event source for EPOLLOUT */ | |
230 | if (s->n_written >= sizeof(s->write_size) + s->write_packet->size) { | |
231 | r = dns_stream_update_io(s); | |
232 | if (r < 0) | |
b914e211 | 233 | return dns_stream_complete(s, -r); |
623a4c97 LP |
234 | } |
235 | } | |
236 | ||
237 | if ((revents & (EPOLLIN|EPOLLHUP|EPOLLRDHUP)) && | |
238 | (!s->read_packet || | |
239 | s->n_read < sizeof(s->read_size) + s->read_packet->size)) { | |
240 | ||
241 | if (s->n_read < sizeof(s->read_size)) { | |
242 | ssize_t ss; | |
243 | ||
244 | ss = read(fd, (uint8_t*) &s->read_size + s->n_read, sizeof(s->read_size) - s->n_read); | |
245 | if (ss < 0) { | |
ec2ce0c5 | 246 | if (!IN_SET(errno, EINTR, EAGAIN)) |
b914e211 | 247 | return dns_stream_complete(s, errno); |
623a4c97 | 248 | } else if (ss == 0) |
b914e211 | 249 | return dns_stream_complete(s, ECONNRESET); |
623a4c97 LP |
250 | else |
251 | s->n_read += ss; | |
252 | } | |
253 | ||
254 | if (s->n_read >= sizeof(s->read_size)) { | |
255 | ||
256 | if (be16toh(s->read_size) < DNS_PACKET_HEADER_SIZE) | |
b914e211 | 257 | return dns_stream_complete(s, EBADMSG); |
623a4c97 LP |
258 | |
259 | if (s->n_read < sizeof(s->read_size) + be16toh(s->read_size)) { | |
260 | ssize_t ss; | |
261 | ||
262 | if (!s->read_packet) { | |
263 | r = dns_packet_new(&s->read_packet, s->protocol, be16toh(s->read_size)); | |
264 | if (r < 0) | |
b914e211 | 265 | return dns_stream_complete(s, -r); |
623a4c97 LP |
266 | |
267 | s->read_packet->size = be16toh(s->read_size); | |
268 | s->read_packet->ipproto = IPPROTO_TCP; | |
269 | s->read_packet->family = s->peer.sa.sa_family; | |
270 | s->read_packet->ttl = s->ttl; | |
271 | s->read_packet->ifindex = s->ifindex; | |
272 | ||
273 | if (s->read_packet->family == AF_INET) { | |
274 | s->read_packet->sender.in = s->peer.in.sin_addr; | |
275 | s->read_packet->sender_port = be16toh(s->peer.in.sin_port); | |
276 | s->read_packet->destination.in = s->local.in.sin_addr; | |
277 | s->read_packet->destination_port = be16toh(s->local.in.sin_port); | |
278 | } else { | |
279 | assert(s->read_packet->family == AF_INET6); | |
280 | s->read_packet->sender.in6 = s->peer.in6.sin6_addr; | |
281 | s->read_packet->sender_port = be16toh(s->peer.in6.sin6_port); | |
282 | s->read_packet->destination.in6 = s->local.in6.sin6_addr; | |
283 | s->read_packet->destination_port = be16toh(s->local.in6.sin6_port); | |
284 | ||
285 | if (s->read_packet->ifindex == 0) | |
286 | s->read_packet->ifindex = s->peer.in6.sin6_scope_id; | |
287 | if (s->read_packet->ifindex == 0) | |
288 | s->read_packet->ifindex = s->local.in6.sin6_scope_id; | |
289 | } | |
290 | } | |
291 | ||
292 | ss = read(fd, | |
293 | (uint8_t*) DNS_PACKET_DATA(s->read_packet) + s->n_read - sizeof(s->read_size), | |
294 | sizeof(s->read_size) + be16toh(s->read_size) - s->n_read); | |
295 | if (ss < 0) { | |
ec2ce0c5 | 296 | if (!IN_SET(errno, EINTR, EAGAIN)) |
b914e211 | 297 | return dns_stream_complete(s, errno); |
623a4c97 | 298 | } else if (ss == 0) |
b914e211 | 299 | return dns_stream_complete(s, ECONNRESET); |
623a4c97 LP |
300 | else |
301 | s->n_read += ss; | |
302 | } | |
303 | ||
304 | /* Are we done? If so, disable the event source for EPOLLIN */ | |
305 | if (s->n_read >= sizeof(s->read_size) + be16toh(s->read_size)) { | |
306 | r = dns_stream_update_io(s); | |
307 | if (r < 0) | |
b914e211 | 308 | return dns_stream_complete(s, -r); |
623a4c97 LP |
309 | |
310 | /* If there's a packet handler | |
311 | * installed, call that. Note that | |
312 | * this is optional... */ | |
313 | if (s->on_packet) | |
314 | return s->on_packet(s); | |
315 | } | |
316 | } | |
317 | } | |
318 | ||
319 | if ((s->write_packet && s->n_written >= sizeof(s->write_size) + s->write_packet->size) && | |
320 | (s->read_packet && s->n_read >= sizeof(s->read_size) + s->read_packet->size)) | |
b914e211 | 321 | return dns_stream_complete(s, 0); |
623a4c97 LP |
322 | |
323 | return 0; | |
324 | } | |
325 | ||
b30bf55d | 326 | DnsStream *dns_stream_unref(DnsStream *s) { |
623a4c97 LP |
327 | if (!s) |
328 | return NULL; | |
329 | ||
b30bf55d LP |
330 | assert(s->n_ref > 0); |
331 | s->n_ref--; | |
332 | ||
333 | if (s->n_ref > 0) | |
334 | return NULL; | |
335 | ||
623a4c97 LP |
336 | dns_stream_stop(s); |
337 | ||
338 | if (s->manager) { | |
339 | LIST_REMOVE(streams, s->manager->dns_streams, s); | |
340 | s->manager->n_dns_streams--; | |
341 | } | |
342 | ||
343 | dns_packet_unref(s->write_packet); | |
344 | dns_packet_unref(s->read_packet); | |
345 | ||
6b430fdb | 346 | return mfree(s); |
623a4c97 LP |
347 | } |
348 | ||
b30bf55d LP |
349 | DEFINE_TRIVIAL_CLEANUP_FUNC(DnsStream*, dns_stream_unref); |
350 | ||
351 | DnsStream *dns_stream_ref(DnsStream *s) { | |
352 | if (!s) | |
353 | return NULL; | |
354 | ||
355 | assert(s->n_ref > 0); | |
356 | s->n_ref++; | |
357 | ||
358 | return s; | |
359 | } | |
623a4c97 LP |
360 | |
361 | int dns_stream_new(Manager *m, DnsStream **ret, DnsProtocol protocol, int fd) { | |
b30bf55d | 362 | _cleanup_(dns_stream_unrefp) DnsStream *s = NULL; |
623a4c97 LP |
363 | int r; |
364 | ||
365 | assert(m); | |
366 | assert(fd >= 0); | |
367 | ||
368 | if (m->n_dns_streams > DNS_STREAMS_MAX) | |
369 | return -EBUSY; | |
370 | ||
371 | s = new0(DnsStream, 1); | |
372 | if (!s) | |
373 | return -ENOMEM; | |
374 | ||
b30bf55d | 375 | s->n_ref = 1; |
623a4c97 LP |
376 | s->fd = -1; |
377 | s->protocol = protocol; | |
378 | ||
623a4c97 LP |
379 | r = sd_event_add_io(m->event, &s->io_event_source, fd, EPOLLIN, on_stream_io, s); |
380 | if (r < 0) | |
381 | return r; | |
382 | ||
aa4a9deb LP |
383 | (void) sd_event_source_set_description(s->io_event_source, "dns-stream-io"); |
384 | ||
9a015429 LP |
385 | r = sd_event_add_time( |
386 | m->event, | |
387 | &s->timeout_event_source, | |
388 | clock_boottime_or_monotonic(), | |
389 | now(clock_boottime_or_monotonic()) + DNS_STREAM_TIMEOUT_USEC, 0, | |
390 | on_stream_timeout, s); | |
623a4c97 LP |
391 | if (r < 0) |
392 | return r; | |
393 | ||
aa4a9deb LP |
394 | (void) sd_event_source_set_description(s->timeout_event_source, "dns-stream-timeout"); |
395 | ||
623a4c97 LP |
396 | LIST_PREPEND(streams, m->dns_streams, s); |
397 | s->manager = m; | |
398 | s->fd = fd; | |
399 | m->n_dns_streams++; | |
400 | ||
401 | *ret = s; | |
402 | s = NULL; | |
403 | ||
404 | return 0; | |
405 | } | |
406 | ||
407 | int dns_stream_write_packet(DnsStream *s, DnsPacket *p) { | |
408 | assert(s); | |
409 | ||
410 | if (s->write_packet) | |
411 | return -EBUSY; | |
412 | ||
413 | s->write_packet = dns_packet_ref(p); | |
414 | s->write_size = htobe16(p->size); | |
415 | s->n_written = 0; | |
416 | ||
417 | return dns_stream_update_io(s); | |
418 | } |