]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/resolve/resolved-dns-stream.c
Merge pull request #8676 from keszybz/drop-license-boilerplate
[thirdparty/systemd.git] / src / resolve / resolved-dns-stream.c
1 /* SPDX-License-Identifier: LGPL-2.1+ */
2 /***
3 This file is part of systemd.
4
5 Copyright 2014 Lennart Poettering
6 ***/
7
8 #include <netinet/tcp.h>
9
10 #include "alloc-util.h"
11 #include "fd-util.h"
12 #include "io-util.h"
13 #include "missing.h"
14 #include "resolved-dns-stream.h"
15
16 #define DNS_STREAM_TIMEOUT_USEC (10 * USEC_PER_SEC)
17 #define DNS_STREAMS_MAX 128
18
19 static void dns_stream_stop(DnsStream *s) {
20 assert(s);
21
22 s->io_event_source = sd_event_source_unref(s->io_event_source);
23 s->timeout_event_source = sd_event_source_unref(s->timeout_event_source);
24 s->fd = safe_close(s->fd);
25 }
26
27 static int dns_stream_update_io(DnsStream *s) {
28 int f = 0;
29
30 assert(s);
31
32 if (s->write_packet && s->n_written < sizeof(s->write_size) + s->write_packet->size)
33 f |= EPOLLOUT;
34 if (!s->read_packet || s->n_read < sizeof(s->read_size) + s->read_packet->size)
35 f |= EPOLLIN;
36
37 return sd_event_source_set_io_events(s->io_event_source, f);
38 }
39
40 static int dns_stream_complete(DnsStream *s, int error) {
41 assert(s);
42
43 dns_stream_stop(s);
44
45 if (s->complete)
46 s->complete(s, error);
47 else /* the default action if no completion function is set is to close the stream */
48 dns_stream_unref(s);
49
50 return 0;
51 }
52
53 static int dns_stream_identify(DnsStream *s) {
54 union {
55 struct cmsghdr header; /* For alignment */
56 uint8_t buffer[CMSG_SPACE(MAXSIZE(struct in_pktinfo, struct in6_pktinfo))
57 + EXTRA_CMSG_SPACE /* kernel appears to require extra space */];
58 } control;
59 struct msghdr mh = {};
60 struct cmsghdr *cmsg;
61 socklen_t sl;
62 int r;
63
64 assert(s);
65
66 if (s->identified)
67 return 0;
68
69 /* Query the local side */
70 s->local_salen = sizeof(s->local);
71 r = getsockname(s->fd, &s->local.sa, &s->local_salen);
72 if (r < 0)
73 return -errno;
74 if (s->local.sa.sa_family == AF_INET6 && s->ifindex <= 0)
75 s->ifindex = s->local.in6.sin6_scope_id;
76
77 /* Query the remote side */
78 s->peer_salen = sizeof(s->peer);
79 r = getpeername(s->fd, &s->peer.sa, &s->peer_salen);
80 if (r < 0)
81 return -errno;
82 if (s->peer.sa.sa_family == AF_INET6 && s->ifindex <= 0)
83 s->ifindex = s->peer.in6.sin6_scope_id;
84
85 /* Check consistency */
86 assert(s->peer.sa.sa_family == s->local.sa.sa_family);
87 assert(IN_SET(s->peer.sa.sa_family, AF_INET, AF_INET6));
88
89 /* Query connection meta information */
90 sl = sizeof(control);
91 if (s->peer.sa.sa_family == AF_INET) {
92 r = getsockopt(s->fd, IPPROTO_IP, IP_PKTOPTIONS, &control, &sl);
93 if (r < 0)
94 return -errno;
95 } else if (s->peer.sa.sa_family == AF_INET6) {
96
97 r = getsockopt(s->fd, IPPROTO_IPV6, IPV6_2292PKTOPTIONS, &control, &sl);
98 if (r < 0)
99 return -errno;
100 } else
101 return -EAFNOSUPPORT;
102
103 mh.msg_control = &control;
104 mh.msg_controllen = sl;
105
106 CMSG_FOREACH(cmsg, &mh) {
107
108 if (cmsg->cmsg_level == IPPROTO_IPV6) {
109 assert(s->peer.sa.sa_family == AF_INET6);
110
111 switch (cmsg->cmsg_type) {
112
113 case IPV6_PKTINFO: {
114 struct in6_pktinfo *i = (struct in6_pktinfo*) CMSG_DATA(cmsg);
115
116 if (s->ifindex <= 0)
117 s->ifindex = i->ipi6_ifindex;
118 break;
119 }
120
121 case IPV6_HOPLIMIT:
122 s->ttl = *(int *) CMSG_DATA(cmsg);
123 break;
124 }
125
126 } else if (cmsg->cmsg_level == IPPROTO_IP) {
127 assert(s->peer.sa.sa_family == AF_INET);
128
129 switch (cmsg->cmsg_type) {
130
131 case IP_PKTINFO: {
132 struct in_pktinfo *i = (struct in_pktinfo*) CMSG_DATA(cmsg);
133
134 if (s->ifindex <= 0)
135 s->ifindex = i->ipi_ifindex;
136 break;
137 }
138
139 case IP_TTL:
140 s->ttl = *(int *) CMSG_DATA(cmsg);
141 break;
142 }
143 }
144 }
145
146 /* The Linux kernel sets the interface index to the loopback
147 * device if the connection came from the local host since it
148 * avoids the routing table in such a case. Let's unset the
149 * interface index in such a case. */
150 if (s->ifindex == LOOPBACK_IFINDEX)
151 s->ifindex = 0;
152
153 /* If we don't know the interface index still, we look for the
154 * first local interface with a matching address. Yuck! */
155 if (s->ifindex <= 0)
156 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);
157
158 if (s->protocol == DNS_PROTOCOL_LLMNR && s->ifindex > 0) {
159 uint32_t ifindex = htobe32(s->ifindex);
160
161 /* Make sure all packets for this connection are sent on the same interface */
162 if (s->local.sa.sa_family == AF_INET) {
163 r = setsockopt(s->fd, IPPROTO_IP, IP_UNICAST_IF, &ifindex, sizeof(ifindex));
164 if (r < 0)
165 log_debug_errno(errno, "Failed to invoke IP_UNICAST_IF: %m");
166 } else if (s->local.sa.sa_family == AF_INET6) {
167 r = setsockopt(s->fd, IPPROTO_IPV6, IPV6_UNICAST_IF, &ifindex, sizeof(ifindex));
168 if (r < 0)
169 log_debug_errno(errno, "Failed to invoke IPV6_UNICAST_IF: %m");
170 }
171 }
172
173 s->identified = true;
174
175 return 0;
176 }
177
178 static int on_stream_timeout(sd_event_source *es, usec_t usec, void *userdata) {
179 DnsStream *s = userdata;
180
181 assert(s);
182
183 return dns_stream_complete(s, ETIMEDOUT);
184 }
185
186 static int on_stream_io(sd_event_source *es, int fd, uint32_t revents, void *userdata) {
187 DnsStream *s = userdata;
188 int r;
189
190 assert(s);
191
192 r = dns_stream_identify(s);
193 if (r < 0)
194 return dns_stream_complete(s, -r);
195
196 if ((revents & EPOLLOUT) &&
197 s->write_packet &&
198 s->n_written < sizeof(s->write_size) + s->write_packet->size) {
199
200 struct iovec iov[2];
201 ssize_t ss;
202
203 iov[0].iov_base = &s->write_size;
204 iov[0].iov_len = sizeof(s->write_size);
205 iov[1].iov_base = DNS_PACKET_DATA(s->write_packet);
206 iov[1].iov_len = s->write_packet->size;
207
208 IOVEC_INCREMENT(iov, 2, s->n_written);
209
210 ss = writev(fd, iov, 2);
211 if (ss < 0) {
212 if (!IN_SET(errno, EINTR, EAGAIN))
213 return dns_stream_complete(s, errno);
214 } else
215 s->n_written += ss;
216
217 /* Are we done? If so, disable the event source for EPOLLOUT */
218 if (s->n_written >= sizeof(s->write_size) + s->write_packet->size) {
219 r = dns_stream_update_io(s);
220 if (r < 0)
221 return dns_stream_complete(s, -r);
222 }
223 }
224
225 if ((revents & (EPOLLIN|EPOLLHUP|EPOLLRDHUP)) &&
226 (!s->read_packet ||
227 s->n_read < sizeof(s->read_size) + s->read_packet->size)) {
228
229 if (s->n_read < sizeof(s->read_size)) {
230 ssize_t ss;
231
232 ss = read(fd, (uint8_t*) &s->read_size + s->n_read, sizeof(s->read_size) - s->n_read);
233 if (ss < 0) {
234 if (!IN_SET(errno, EINTR, EAGAIN))
235 return dns_stream_complete(s, errno);
236 } else if (ss == 0)
237 return dns_stream_complete(s, ECONNRESET);
238 else
239 s->n_read += ss;
240 }
241
242 if (s->n_read >= sizeof(s->read_size)) {
243
244 if (be16toh(s->read_size) < DNS_PACKET_HEADER_SIZE)
245 return dns_stream_complete(s, EBADMSG);
246
247 if (s->n_read < sizeof(s->read_size) + be16toh(s->read_size)) {
248 ssize_t ss;
249
250 if (!s->read_packet) {
251 r = dns_packet_new(&s->read_packet, s->protocol, be16toh(s->read_size), DNS_PACKET_SIZE_MAX);
252 if (r < 0)
253 return dns_stream_complete(s, -r);
254
255 s->read_packet->size = be16toh(s->read_size);
256 s->read_packet->ipproto = IPPROTO_TCP;
257 s->read_packet->family = s->peer.sa.sa_family;
258 s->read_packet->ttl = s->ttl;
259 s->read_packet->ifindex = s->ifindex;
260
261 if (s->read_packet->family == AF_INET) {
262 s->read_packet->sender.in = s->peer.in.sin_addr;
263 s->read_packet->sender_port = be16toh(s->peer.in.sin_port);
264 s->read_packet->destination.in = s->local.in.sin_addr;
265 s->read_packet->destination_port = be16toh(s->local.in.sin_port);
266 } else {
267 assert(s->read_packet->family == AF_INET6);
268 s->read_packet->sender.in6 = s->peer.in6.sin6_addr;
269 s->read_packet->sender_port = be16toh(s->peer.in6.sin6_port);
270 s->read_packet->destination.in6 = s->local.in6.sin6_addr;
271 s->read_packet->destination_port = be16toh(s->local.in6.sin6_port);
272
273 if (s->read_packet->ifindex == 0)
274 s->read_packet->ifindex = s->peer.in6.sin6_scope_id;
275 if (s->read_packet->ifindex == 0)
276 s->read_packet->ifindex = s->local.in6.sin6_scope_id;
277 }
278 }
279
280 ss = read(fd,
281 (uint8_t*) DNS_PACKET_DATA(s->read_packet) + s->n_read - sizeof(s->read_size),
282 sizeof(s->read_size) + be16toh(s->read_size) - s->n_read);
283 if (ss < 0) {
284 if (!IN_SET(errno, EINTR, EAGAIN))
285 return dns_stream_complete(s, errno);
286 } else if (ss == 0)
287 return dns_stream_complete(s, ECONNRESET);
288 else
289 s->n_read += ss;
290 }
291
292 /* Are we done? If so, disable the event source for EPOLLIN */
293 if (s->n_read >= sizeof(s->read_size) + be16toh(s->read_size)) {
294 r = dns_stream_update_io(s);
295 if (r < 0)
296 return dns_stream_complete(s, -r);
297
298 /* If there's a packet handler
299 * installed, call that. Note that
300 * this is optional... */
301 if (s->on_packet)
302 return s->on_packet(s);
303 }
304 }
305 }
306
307 if ((s->write_packet && s->n_written >= sizeof(s->write_size) + s->write_packet->size) &&
308 (s->read_packet && s->n_read >= sizeof(s->read_size) + s->read_packet->size))
309 return dns_stream_complete(s, 0);
310
311 return 0;
312 }
313
314 DnsStream *dns_stream_unref(DnsStream *s) {
315 if (!s)
316 return NULL;
317
318 assert(s->n_ref > 0);
319 s->n_ref--;
320
321 if (s->n_ref > 0)
322 return NULL;
323
324 dns_stream_stop(s);
325
326 if (s->manager) {
327 LIST_REMOVE(streams, s->manager->dns_streams, s);
328 s->manager->n_dns_streams--;
329 }
330
331 dns_packet_unref(s->write_packet);
332 dns_packet_unref(s->read_packet);
333
334 return mfree(s);
335 }
336
337 DEFINE_TRIVIAL_CLEANUP_FUNC(DnsStream*, dns_stream_unref);
338
339 DnsStream *dns_stream_ref(DnsStream *s) {
340 if (!s)
341 return NULL;
342
343 assert(s->n_ref > 0);
344 s->n_ref++;
345
346 return s;
347 }
348
349 int dns_stream_new(Manager *m, DnsStream **ret, DnsProtocol protocol, int fd) {
350 _cleanup_(dns_stream_unrefp) DnsStream *s = NULL;
351 int r;
352
353 assert(m);
354 assert(fd >= 0);
355
356 if (m->n_dns_streams > DNS_STREAMS_MAX)
357 return -EBUSY;
358
359 s = new0(DnsStream, 1);
360 if (!s)
361 return -ENOMEM;
362
363 s->n_ref = 1;
364 s->fd = -1;
365 s->protocol = protocol;
366
367 r = sd_event_add_io(m->event, &s->io_event_source, fd, EPOLLIN, on_stream_io, s);
368 if (r < 0)
369 return r;
370
371 (void) sd_event_source_set_description(s->io_event_source, "dns-stream-io");
372
373 r = sd_event_add_time(
374 m->event,
375 &s->timeout_event_source,
376 clock_boottime_or_monotonic(),
377 now(clock_boottime_or_monotonic()) + DNS_STREAM_TIMEOUT_USEC, 0,
378 on_stream_timeout, s);
379 if (r < 0)
380 return r;
381
382 (void) sd_event_source_set_description(s->timeout_event_source, "dns-stream-timeout");
383
384 LIST_PREPEND(streams, m->dns_streams, s);
385 s->manager = m;
386 s->fd = fd;
387 m->n_dns_streams++;
388
389 *ret = TAKE_PTR(s);
390
391 return 0;
392 }
393
394 int dns_stream_write_packet(DnsStream *s, DnsPacket *p) {
395 assert(s);
396
397 if (s->write_packet)
398 return -EBUSY;
399
400 s->write_packet = dns_packet_ref(p);
401 s->write_size = htobe16(p->size);
402 s->n_written = 0;
403
404 return dns_stream_update_io(s);
405 }