]>
Commit | Line | Data |
---|---|---|
53e1b683 | 1 | /* SPDX-License-Identifier: LGPL-2.1+ */ |
a7334b09 | 2 | |
42f4e3c4 | 3 | #include <arpa/inet.h> |
07630cea | 4 | #include <errno.h> |
11c3a366 | 5 | #include <limits.h> |
542563ba | 6 | #include <net/if.h> |
b31f535c | 7 | #include <netdb.h> |
2583fbea | 8 | #include <netinet/ip.h> |
60d9771c | 9 | #include <poll.h> |
07630cea | 10 | #include <stddef.h> |
11c3a366 | 11 | #include <stdint.h> |
07630cea | 12 | #include <stdio.h> |
11c3a366 | 13 | #include <stdlib.h> |
f5947a5e | 14 | #include <sys/ioctl.h> |
07630cea | 15 | #include <unistd.h> |
4252696a | 16 | #include <linux/if.h> |
42f4e3c4 | 17 | |
b5efdb8a | 18 | #include "alloc-util.h" |
4ff9bc2e | 19 | #include "errno-util.h" |
15dca371 | 20 | #include "escape.h" |
2583fbea | 21 | #include "fd-util.h" |
07630cea | 22 | #include "fileio.h" |
f97b34a6 | 23 | #include "format-util.h" |
0f2d351f | 24 | #include "io-util.h" |
93cc7779 | 25 | #include "log.h" |
42f4e3c4 | 26 | #include "macro.h" |
0a970718 | 27 | #include "memory-util.h" |
f5947a5e | 28 | #include "missing_socket.h" |
5d0fe423 | 29 | #include "missing_network.h" |
6bedfcbb | 30 | #include "parse-util.h" |
9eb977db | 31 | #include "path-util.h" |
dccca82b | 32 | #include "process-util.h" |
2583fbea | 33 | #include "socket-util.h" |
8b43440b | 34 | #include "string-table.h" |
07630cea | 35 | #include "string-util.h" |
ef76dff2 | 36 | #include "strv.h" |
ee104e11 | 37 | #include "user-util.h" |
ef76dff2 | 38 | #include "utf8.h" |
42f4e3c4 | 39 | |
349cc4a5 | 40 | #if ENABLE_IDN |
cadc80b8 | 41 | # define IDN_FLAGS NI_IDN |
6326a143 WB |
42 | #else |
43 | # define IDN_FLAGS 0 | |
44 | #endif | |
45 | ||
398ce0bc | 46 | static const char* const socket_address_type_table[] = { |
955bb7fa ZJS |
47 | [SOCK_STREAM] = "Stream", |
48 | [SOCK_DGRAM] = "Datagram", | |
49 | [SOCK_RAW] = "Raw", | |
50 | [SOCK_RDM] = "ReliableDatagram", | |
398ce0bc | 51 | [SOCK_SEQPACKET] = "SequentialPacket", |
955bb7fa | 52 | [SOCK_DCCP] = "DatagramCongestionControl", |
398ce0bc YW |
53 | }; |
54 | ||
55 | DEFINE_STRING_TABLE_LOOKUP(socket_address_type, int); | |
56 | ||
15dca371 | 57 | int socket_address_verify(const SocketAddress *a, bool strict) { |
42f4e3c4 LP |
58 | assert(a); |
59 | ||
15dca371 ZJS |
60 | /* With 'strict' we enforce additional sanity constraints which are not set by the standard, |
61 | * but should only apply to sockets we create ourselves. */ | |
62 | ||
542563ba | 63 | switch (socket_address_family(a)) { |
42f4e3c4 | 64 | |
7a22745a LP |
65 | case AF_INET: |
66 | if (a->size != sizeof(struct sockaddr_in)) | |
67 | return -EINVAL; | |
42f4e3c4 | 68 | |
4d49b48c | 69 | if (a->sockaddr.in.sin_port == 0) |
7a22745a | 70 | return -EINVAL; |
42f4e3c4 | 71 | |
44ab2347 | 72 | if (!IN_SET(a->type, 0, SOCK_STREAM, SOCK_DGRAM)) |
7a22745a | 73 | return -EINVAL; |
42f4e3c4 | 74 | |
7a22745a LP |
75 | return 0; |
76 | ||
77 | case AF_INET6: | |
78 | if (a->size != sizeof(struct sockaddr_in6)) | |
79 | return -EINVAL; | |
42f4e3c4 | 80 | |
7a22745a LP |
81 | if (a->sockaddr.in6.sin6_port == 0) |
82 | return -EINVAL; | |
42f4e3c4 | 83 | |
44ab2347 | 84 | if (!IN_SET(a->type, 0, SOCK_STREAM, SOCK_DGRAM)) |
7a22745a | 85 | return -EINVAL; |
42f4e3c4 | 86 | |
7a22745a | 87 | return 0; |
42f4e3c4 | 88 | |
7a22745a LP |
89 | case AF_UNIX: |
90 | if (a->size < offsetof(struct sockaddr_un, sun_path)) | |
91 | return -EINVAL; | |
15dca371 ZJS |
92 | if (a->size > sizeof(struct sockaddr_un) + !strict) |
93 | /* If !strict, allow one extra byte, since getsockname() on Linux will append | |
94 | * a NUL byte if we have path sockets that are above sun_path's full size. */ | |
8e8132c6 | 95 | return -EINVAL; |
42f4e3c4 | 96 | |
8e8132c6 | 97 | if (a->size > offsetof(struct sockaddr_un, sun_path) && |
15dca371 ZJS |
98 | a->sockaddr.un.sun_path[0] != 0 && |
99 | strict) { | |
100 | /* Only validate file system sockets here, and only in strict mode */ | |
8e8132c6 | 101 | const char *e; |
7a22745a | 102 | |
8e8132c6 LP |
103 | e = memchr(a->sockaddr.un.sun_path, 0, sizeof(a->sockaddr.un.sun_path)); |
104 | if (e) { | |
15dca371 | 105 | /* If there's an embedded NUL byte, make sure the size of the socket address matches it */ |
7a22745a LP |
106 | if (a->size != offsetof(struct sockaddr_un, sun_path) + (e - a->sockaddr.un.sun_path) + 1) |
107 | return -EINVAL; | |
8e8132c6 | 108 | } else { |
d7b34e38 | 109 | /* If there's no embedded NUL byte, then the size needs to match the whole |
8e8132c6 LP |
110 | * structure or the structure with one extra NUL byte suffixed. (Yeah, Linux is awful, |
111 | * and considers both equivalent: getsockname() even extends sockaddr_un beyond its | |
112 | * size if the path is non NUL terminated.)*/ | |
113 | if (!IN_SET(a->size, sizeof(a->sockaddr.un.sun_path), sizeof(a->sockaddr.un.sun_path)+1)) | |
114 | return -EINVAL; | |
42f4e3c4 | 115 | } |
7a22745a | 116 | } |
42f4e3c4 | 117 | |
44ab2347 | 118 | if (!IN_SET(a->type, 0, SOCK_STREAM, SOCK_DGRAM, SOCK_SEQPACKET)) |
7a22745a | 119 | return -EINVAL; |
42f4e3c4 | 120 | |
7a22745a LP |
121 | return 0; |
122 | ||
123 | case AF_NETLINK: | |
124 | ||
125 | if (a->size != sizeof(struct sockaddr_nl)) | |
126 | return -EINVAL; | |
127 | ||
44ab2347 | 128 | if (!IN_SET(a->type, 0, SOCK_RAW, SOCK_DGRAM)) |
7a22745a LP |
129 | return -EINVAL; |
130 | ||
131 | return 0; | |
132 | ||
0fc0f14b SH |
133 | case AF_VSOCK: |
134 | if (a->size != sizeof(struct sockaddr_vm)) | |
135 | return -EINVAL; | |
136 | ||
44ab2347 | 137 | if (!IN_SET(a->type, 0, SOCK_STREAM, SOCK_DGRAM)) |
0fc0f14b SH |
138 | return -EINVAL; |
139 | ||
140 | return 0; | |
141 | ||
7a22745a LP |
142 | default: |
143 | return -EAFNOSUPPORT; | |
42f4e3c4 LP |
144 | } |
145 | } | |
146 | ||
4d49b48c | 147 | int socket_address_print(const SocketAddress *a, char **ret) { |
42f4e3c4 | 148 | int r; |
4d49b48c | 149 | |
42f4e3c4 | 150 | assert(a); |
4d49b48c | 151 | assert(ret); |
42f4e3c4 | 152 | |
15dca371 ZJS |
153 | r = socket_address_verify(a, false); /* We do non-strict validation, because we want to be |
154 | * able to pretty-print any socket the kernel considers | |
155 | * valid. We still need to do validation to know if we | |
156 | * can meaningfully print the address. */ | |
4d49b48c | 157 | if (r < 0) |
42f4e3c4 LP |
158 | return r; |
159 | ||
4d49b48c | 160 | if (socket_address_family(a) == AF_NETLINK) { |
7fd1b19b | 161 | _cleanup_free_ char *sfamily = NULL; |
7a22745a | 162 | |
f8b69d1d | 163 | r = netlink_family_to_string_alloc(a->protocol, &sfamily); |
7a22745a | 164 | if (r < 0) |
f8b69d1d | 165 | return r; |
4d49b48c LP |
166 | |
167 | r = asprintf(ret, "%s %u", sfamily, a->sockaddr.nl.nl_groups); | |
8520cfa5 MS |
168 | if (r < 0) |
169 | return -ENOMEM; | |
7a22745a LP |
170 | |
171 | return 0; | |
172 | } | |
173 | ||
3b1c5241 | 174 | return sockaddr_pretty(&a->sockaddr.sa, a->size, false, true, ret); |
42f4e3c4 LP |
175 | } |
176 | ||
4f2d528d LP |
177 | bool socket_address_can_accept(const SocketAddress *a) { |
178 | assert(a); | |
179 | ||
180 | return | |
3742095b | 181 | IN_SET(a->type, SOCK_STREAM, SOCK_SEQPACKET); |
4f2d528d | 182 | } |
a16e1123 LP |
183 | |
184 | bool socket_address_equal(const SocketAddress *a, const SocketAddress *b) { | |
185 | assert(a); | |
186 | assert(b); | |
187 | ||
188 | /* Invalid addresses are unequal to all */ | |
15dca371 ZJS |
189 | if (socket_address_verify(a, false) < 0 || |
190 | socket_address_verify(b, false) < 0) | |
a16e1123 LP |
191 | return false; |
192 | ||
193 | if (a->type != b->type) | |
194 | return false; | |
195 | ||
a16e1123 LP |
196 | if (socket_address_family(a) != socket_address_family(b)) |
197 | return false; | |
198 | ||
199 | switch (socket_address_family(a)) { | |
200 | ||
201 | case AF_INET: | |
4d49b48c | 202 | if (a->sockaddr.in.sin_addr.s_addr != b->sockaddr.in.sin_addr.s_addr) |
a16e1123 LP |
203 | return false; |
204 | ||
4d49b48c | 205 | if (a->sockaddr.in.sin_port != b->sockaddr.in.sin_port) |
a16e1123 LP |
206 | return false; |
207 | ||
208 | break; | |
209 | ||
210 | case AF_INET6: | |
211 | if (memcmp(&a->sockaddr.in6.sin6_addr, &b->sockaddr.in6.sin6_addr, sizeof(a->sockaddr.in6.sin6_addr)) != 0) | |
212 | return false; | |
213 | ||
214 | if (a->sockaddr.in6.sin6_port != b->sockaddr.in6.sin6_port) | |
215 | return false; | |
216 | ||
217 | break; | |
218 | ||
219 | case AF_UNIX: | |
710708a5 MS |
220 | if (a->size <= offsetof(struct sockaddr_un, sun_path) || |
221 | b->size <= offsetof(struct sockaddr_un, sun_path)) | |
222 | return false; | |
223 | ||
a16e1123 LP |
224 | if ((a->sockaddr.un.sun_path[0] == 0) != (b->sockaddr.un.sun_path[0] == 0)) |
225 | return false; | |
226 | ||
227 | if (a->sockaddr.un.sun_path[0]) { | |
e3f791a2 | 228 | if (!path_equal_or_files_same(a->sockaddr.un.sun_path, b->sockaddr.un.sun_path, 0)) |
a16e1123 LP |
229 | return false; |
230 | } else { | |
c78e47a6 MS |
231 | if (a->size != b->size) |
232 | return false; | |
233 | ||
b12c1e7c | 234 | if (memcmp(a->sockaddr.un.sun_path, b->sockaddr.un.sun_path, a->size) != 0) |
a16e1123 LP |
235 | return false; |
236 | } | |
237 | ||
238 | break; | |
239 | ||
7a22745a | 240 | case AF_NETLINK: |
7a22745a LP |
241 | if (a->protocol != b->protocol) |
242 | return false; | |
243 | ||
244 | if (a->sockaddr.nl.nl_groups != b->sockaddr.nl.nl_groups) | |
245 | return false; | |
246 | ||
247 | break; | |
248 | ||
0fc0f14b SH |
249 | case AF_VSOCK: |
250 | if (a->sockaddr.vm.svm_cid != b->sockaddr.vm.svm_cid) | |
251 | return false; | |
252 | ||
253 | if (a->sockaddr.vm.svm_port != b->sockaddr.vm.svm_port) | |
254 | return false; | |
255 | ||
256 | break; | |
257 | ||
a16e1123 LP |
258 | default: |
259 | /* Cannot compare, so we assume the addresses are different */ | |
260 | return false; | |
261 | } | |
262 | ||
263 | return true; | |
264 | } | |
265 | ||
a57f7e2c | 266 | const char* socket_address_get_path(const SocketAddress *a) { |
6e2ef85b LP |
267 | assert(a); |
268 | ||
269 | if (socket_address_family(a) != AF_UNIX) | |
a57f7e2c | 270 | return NULL; |
6e2ef85b LP |
271 | |
272 | if (a->sockaddr.un.sun_path[0] == 0) | |
a57f7e2c | 273 | return NULL; |
a16e1123 | 274 | |
48e6a2dc LP |
275 | /* Note that this is only safe because we know that there's an extra NUL byte after the sockaddr_un |
276 | * structure. On Linux AF_UNIX file system socket addresses don't have to be NUL terminated if they take up the | |
277 | * full sun_path space. */ | |
278 | assert_cc(sizeof(union sockaddr_union) >= sizeof(struct sockaddr_un)+1); | |
a57f7e2c | 279 | return a->sockaddr.un.sun_path; |
a16e1123 | 280 | } |
c0120d99 | 281 | |
5bfcc1c6 | 282 | bool socket_ipv6_is_supported(void) { |
629abfc2 | 283 | if (access("/proc/net/if_inet6", F_OK) != 0) |
90ab5042 | 284 | return false; |
f89f1e8f | 285 | |
7377964d | 286 | return true; |
5bfcc1c6 FF |
287 | } |
288 | ||
01e10de3 | 289 | bool socket_address_matches_fd(const SocketAddress *a, int fd) { |
dbafedac MS |
290 | SocketAddress b; |
291 | socklen_t solen; | |
01e10de3 LP |
292 | |
293 | assert(a); | |
294 | assert(fd >= 0); | |
295 | ||
dbafedac MS |
296 | b.size = sizeof(b.sockaddr); |
297 | if (getsockname(fd, &b.sockaddr.sa, &b.size) < 0) | |
01e10de3 LP |
298 | return false; |
299 | ||
dbafedac | 300 | if (b.sockaddr.sa.sa_family != a->sockaddr.sa.sa_family) |
01e10de3 LP |
301 | return false; |
302 | ||
dbafedac MS |
303 | solen = sizeof(b.type); |
304 | if (getsockopt(fd, SOL_SOCKET, SO_TYPE, &b.type, &solen) < 0) | |
01e10de3 LP |
305 | return false; |
306 | ||
dbafedac | 307 | if (b.type != a->type) |
01e10de3 LP |
308 | return false; |
309 | ||
310 | if (a->protocol != 0) { | |
dbafedac MS |
311 | solen = sizeof(b.protocol); |
312 | if (getsockopt(fd, SOL_SOCKET, SO_PROTOCOL, &b.protocol, &solen) < 0) | |
01e10de3 LP |
313 | return false; |
314 | ||
dbafedac | 315 | if (b.protocol != a->protocol) |
01e10de3 LP |
316 | return false; |
317 | } | |
318 | ||
02233928 | 319 | return socket_address_equal(a, &b); |
01e10de3 LP |
320 | } |
321 | ||
f6aac5bf | 322 | int sockaddr_port(const struct sockaddr *_sa, unsigned *ret_port) { |
3b1c5241 SL |
323 | union sockaddr_union *sa = (union sockaddr_union*) _sa; |
324 | ||
f6aac5bf LP |
325 | /* Note, this returns the port as 'unsigned' rather than 'uint16_t', as AF_VSOCK knows larger ports */ |
326 | ||
3b1c5241 SL |
327 | assert(sa); |
328 | ||
0fc0f14b | 329 | switch (sa->sa.sa_family) { |
f6aac5bf | 330 | |
0fc0f14b | 331 | case AF_INET: |
f6aac5bf | 332 | *ret_port = be16toh(sa->in.sin_port); |
0fc0f14b | 333 | return 0; |
3b1c5241 | 334 | |
0fc0f14b | 335 | case AF_INET6: |
f6aac5bf | 336 | *ret_port = be16toh(sa->in6.sin6_port); |
0fc0f14b SH |
337 | return 0; |
338 | ||
339 | case AF_VSOCK: | |
f6aac5bf | 340 | *ret_port = sa->vm.svm_port; |
0fc0f14b SH |
341 | return 0; |
342 | ||
343 | default: | |
344 | return -EAFNOSUPPORT; | |
345 | } | |
3b1c5241 SL |
346 | } |
347 | ||
836f9cfe LP |
348 | int sockaddr_pretty( |
349 | const struct sockaddr *_sa, | |
350 | socklen_t salen, | |
351 | bool translate_ipv6, | |
352 | bool include_port, | |
353 | char **ret) { | |
354 | ||
4d49b48c | 355 | union sockaddr_union *sa = (union sockaddr_union*) _sa; |
8569a776 | 356 | char *p; |
fc25ad25 | 357 | int r; |
8569a776 | 358 | |
4d49b48c LP |
359 | assert(sa); |
360 | assert(salen >= sizeof(sa->sa.sa_family)); | |
8569a776 | 361 | |
4d49b48c | 362 | switch (sa->sa.sa_family) { |
8569a776 LP |
363 | |
364 | case AF_INET: { | |
365 | uint32_t a; | |
366 | ||
8e38570e | 367 | a = be32toh(sa->in.sin_addr.s_addr); |
8569a776 | 368 | |
fc25ad25 ZJS |
369 | if (include_port) |
370 | r = asprintf(&p, | |
3b1c5241 SL |
371 | "%u.%u.%u.%u:%u", |
372 | a >> 24, (a >> 16) & 0xFF, (a >> 8) & 0xFF, a & 0xFF, | |
8e38570e | 373 | be16toh(sa->in.sin_port)); |
fc25ad25 ZJS |
374 | else |
375 | r = asprintf(&p, | |
3b1c5241 | 376 | "%u.%u.%u.%u", |
fc25ad25 ZJS |
377 | a >> 24, (a >> 16) & 0xFF, (a >> 8) & 0xFF, a & 0xFF); |
378 | if (r < 0) | |
379 | return -ENOMEM; | |
8569a776 LP |
380 | break; |
381 | } | |
382 | ||
383 | case AF_INET6: { | |
384 | static const unsigned char ipv4_prefix[] = { | |
385 | 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0xFF, 0xFF | |
386 | }; | |
387 | ||
fc25ad25 ZJS |
388 | if (translate_ipv6 && |
389 | memcmp(&sa->in6.sin6_addr, ipv4_prefix, sizeof(ipv4_prefix)) == 0) { | |
4d49b48c | 390 | const uint8_t *a = sa->in6.sin6_addr.s6_addr+12; |
fc25ad25 ZJS |
391 | if (include_port) |
392 | r = asprintf(&p, | |
3b1c5241 SL |
393 | "%u.%u.%u.%u:%u", |
394 | a[0], a[1], a[2], a[3], | |
8e38570e | 395 | be16toh(sa->in6.sin6_port)); |
fc25ad25 ZJS |
396 | else |
397 | r = asprintf(&p, | |
3b1c5241 | 398 | "%u.%u.%u.%u", |
fc25ad25 ZJS |
399 | a[0], a[1], a[2], a[3]); |
400 | if (r < 0) | |
401 | return -ENOMEM; | |
8569a776 | 402 | } else { |
b16d17a6 | 403 | char a[INET6_ADDRSTRLEN], ifname[IF_NAMESIZE + 1]; |
8569a776 | 404 | |
3b1c5241 | 405 | inet_ntop(AF_INET6, &sa->in6.sin6_addr, a, sizeof(a)); |
b16d17a6 ZJS |
406 | if (sa->in6.sin6_scope_id != 0) |
407 | format_ifname_full(sa->in6.sin6_scope_id, ifname, FORMAT_IFNAME_IFINDEX); | |
3b1c5241 SL |
408 | |
409 | if (include_port) { | |
fc25ad25 | 410 | r = asprintf(&p, |
b16d17a6 | 411 | "[%s]:%u%s%s", |
3b1c5241 | 412 | a, |
b16d17a6 ZJS |
413 | be16toh(sa->in6.sin6_port), |
414 | sa->in6.sin6_scope_id != 0 ? "%" : "", | |
415 | sa->in6.sin6_scope_id != 0 ? ifname : ""); | |
fc25ad25 | 416 | if (r < 0) |
3b1c5241 SL |
417 | return -ENOMEM; |
418 | } else { | |
b16d17a6 | 419 | p = sa->in6.sin6_scope_id != 0 ? strjoin(a, "%", ifname) : strdup(a); |
3b1c5241 SL |
420 | if (!p) |
421 | return -ENOMEM; | |
422 | } | |
8569a776 LP |
423 | } |
424 | ||
425 | break; | |
426 | } | |
427 | ||
4d49b48c | 428 | case AF_UNIX: |
15dca371 | 429 | if (salen <= offsetof(struct sockaddr_un, sun_path) || |
994b9d4e | 430 | (sa->un.sun_path[0] == 0 && salen == offsetof(struct sockaddr_un, sun_path) + 1)) |
15dca371 | 431 | /* The name must have at least one character (and the leading NUL does not count) */ |
4d49b48c | 432 | p = strdup("<unnamed>"); |
994b9d4e | 433 | else { |
085b39e9 LP |
434 | /* Note that we calculate the path pointer here through the .un_buffer[] field, in order to |
435 | * outtrick bounds checking tools such as ubsan, which are too smart for their own good: on | |
436 | * Linux the kernel may return sun_path[] data one byte longer than the declared size of the | |
437 | * field. */ | |
438 | char *path = (char*) sa->un_buffer + offsetof(struct sockaddr_un, sun_path); | |
15dca371 | 439 | size_t path_len = salen - offsetof(struct sockaddr_un, sun_path); |
8569a776 | 440 | |
085b39e9 | 441 | if (path[0] == 0) { |
15dca371 ZJS |
442 | /* Abstract socket. When parsing address information from, we |
443 | * explicitly reject overly long paths and paths with embedded NULs. | |
444 | * But we might get such a socket from the outside. Let's return | |
445 | * something meaningful and printable in this case. */ | |
4d49b48c | 446 | |
15dca371 | 447 | _cleanup_free_ char *e = NULL; |
4d49b48c | 448 | |
085b39e9 | 449 | e = cescape_length(path + 1, path_len - 1); |
15dca371 ZJS |
450 | if (!e) |
451 | return -ENOMEM; | |
4d49b48c | 452 | |
15dca371 ZJS |
453 | p = strjoin("@", e); |
454 | } else { | |
085b39e9 | 455 | if (path[path_len - 1] == '\0') |
15dca371 ZJS |
456 | /* We expect a terminating NUL and don't print it */ |
457 | path_len --; | |
458 | ||
085b39e9 | 459 | p = cescape_length(path, path_len); |
15dca371 | 460 | } |
4d49b48c | 461 | } |
994b9d4e LP |
462 | if (!p) |
463 | return -ENOMEM; | |
8569a776 LP |
464 | |
465 | break; | |
8569a776 | 466 | |
0fc0f14b | 467 | case AF_VSOCK: |
3a484991 ZJS |
468 | if (include_port) { |
469 | if (sa->vm.svm_cid == VMADDR_CID_ANY) | |
470 | r = asprintf(&p, "vsock::%u", sa->vm.svm_port); | |
471 | else | |
472 | r = asprintf(&p, "vsock:%u:%u", sa->vm.svm_cid, sa->vm.svm_port); | |
473 | } else | |
0fc0f14b SH |
474 | r = asprintf(&p, "vsock:%u", sa->vm.svm_cid); |
475 | if (r < 0) | |
476 | return -ENOMEM; | |
477 | break; | |
478 | ||
8569a776 | 479 | default: |
15411c0c | 480 | return -EOPNOTSUPP; |
8569a776 LP |
481 | } |
482 | ||
483 | *ret = p; | |
484 | return 0; | |
485 | } | |
486 | ||
366b7db4 | 487 | int getpeername_pretty(int fd, bool include_port, char **ret) { |
4d49b48c | 488 | union sockaddr_union sa; |
b31f535c | 489 | socklen_t salen = sizeof(sa); |
eff05270 | 490 | int r; |
4d49b48c LP |
491 | |
492 | assert(fd >= 0); | |
493 | assert(ret); | |
494 | ||
4d49b48c LP |
495 | if (getpeername(fd, &sa.sa, &salen) < 0) |
496 | return -errno; | |
497 | ||
498 | if (sa.sa.sa_family == AF_UNIX) { | |
39883f62 | 499 | struct ucred ucred = {}; |
4d49b48c LP |
500 | |
501 | /* UNIX connection sockets are anonymous, so let's use | |
502 | * PID/UID as pretty credentials instead */ | |
503 | ||
eff05270 LP |
504 | r = getpeercred(fd, &ucred); |
505 | if (r < 0) | |
506 | return r; | |
4d49b48c | 507 | |
de0671ee | 508 | if (asprintf(ret, "PID "PID_FMT"/UID "UID_FMT, ucred.pid, ucred.uid) < 0) |
4d49b48c LP |
509 | return -ENOMEM; |
510 | ||
511 | return 0; | |
512 | } | |
513 | ||
514 | /* For remote sockets we translate IPv6 addresses back to IPv4 | |
515 | * if applicable, since that's nicer. */ | |
516 | ||
366b7db4 | 517 | return sockaddr_pretty(&sa.sa, salen, true, include_port, ret); |
4d49b48c LP |
518 | } |
519 | ||
520 | int getsockname_pretty(int fd, char **ret) { | |
521 | union sockaddr_union sa; | |
b31f535c | 522 | socklen_t salen = sizeof(sa); |
4d49b48c LP |
523 | |
524 | assert(fd >= 0); | |
525 | assert(ret); | |
526 | ||
4d49b48c LP |
527 | if (getsockname(fd, &sa.sa, &salen) < 0) |
528 | return -errno; | |
529 | ||
530 | /* For local sockets we do not translate IPv6 addresses back | |
531 | * to IPv6 if applicable, since this is usually used for | |
532 | * listening sockets where the difference between IPv4 and | |
533 | * IPv6 matters. */ | |
534 | ||
3b1c5241 | 535 | return sockaddr_pretty(&sa.sa, salen, false, true, ret); |
4d49b48c LP |
536 | } |
537 | ||
b31f535c ZJS |
538 | int socknameinfo_pretty(union sockaddr_union *sa, socklen_t salen, char **_ret) { |
539 | int r; | |
540 | char host[NI_MAXHOST], *ret; | |
541 | ||
542 | assert(_ret); | |
543 | ||
6326a143 | 544 | r = getnameinfo(&sa->sa, salen, host, sizeof(host), NULL, 0, IDN_FLAGS); |
b31f535c | 545 | if (r != 0) { |
b31f535c ZJS |
546 | int saved_errno = errno; |
547 | ||
3b1c5241 | 548 | r = sockaddr_pretty(&sa->sa, salen, true, true, &ret); |
f647962d | 549 | if (r < 0) |
1938ac51 | 550 | return r; |
b31f535c | 551 | |
279d3c9c | 552 | log_debug_errno(saved_errno, "getnameinfo(%s) failed: %m", ret); |
cb651834 ZJS |
553 | } else { |
554 | ret = strdup(host); | |
555 | if (!ret) | |
1938ac51 | 556 | return -ENOMEM; |
cb651834 | 557 | } |
b31f535c ZJS |
558 | |
559 | *_ret = ret; | |
560 | return 0; | |
561 | } | |
562 | ||
7a22745a LP |
563 | static const char* const netlink_family_table[] = { |
564 | [NETLINK_ROUTE] = "route", | |
565 | [NETLINK_FIREWALL] = "firewall", | |
566 | [NETLINK_INET_DIAG] = "inet-diag", | |
567 | [NETLINK_NFLOG] = "nflog", | |
568 | [NETLINK_XFRM] = "xfrm", | |
569 | [NETLINK_SELINUX] = "selinux", | |
570 | [NETLINK_ISCSI] = "iscsi", | |
571 | [NETLINK_AUDIT] = "audit", | |
572 | [NETLINK_FIB_LOOKUP] = "fib-lookup", | |
573 | [NETLINK_CONNECTOR] = "connector", | |
574 | [NETLINK_NETFILTER] = "netfilter", | |
575 | [NETLINK_IP6_FW] = "ip6-fw", | |
576 | [NETLINK_DNRTMSG] = "dnrtmsg", | |
577 | [NETLINK_KOBJECT_UEVENT] = "kobject-uevent", | |
578 | [NETLINK_GENERIC] = "generic", | |
579 | [NETLINK_SCSITRANSPORT] = "scsitransport", | |
5570d7f9 JG |
580 | [NETLINK_ECRYPTFS] = "ecryptfs", |
581 | [NETLINK_RDMA] = "rdma", | |
7a22745a LP |
582 | }; |
583 | ||
f8b69d1d | 584 | DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(netlink_family, int, INT_MAX); |
7a22745a | 585 | |
c0120d99 LP |
586 | static const char* const socket_address_bind_ipv6_only_table[_SOCKET_ADDRESS_BIND_IPV6_ONLY_MAX] = { |
587 | [SOCKET_ADDRESS_DEFAULT] = "default", | |
588 | [SOCKET_ADDRESS_BOTH] = "both", | |
589 | [SOCKET_ADDRESS_IPV6_ONLY] = "ipv6-only" | |
590 | }; | |
591 | ||
592 | DEFINE_STRING_TABLE_LOOKUP(socket_address_bind_ipv6_only, SocketAddressBindIPv6Only); | |
f01e5736 | 593 | |
b54e98ef | 594 | SocketAddressBindIPv6Only socket_address_bind_ipv6_only_or_bool_from_string(const char *n) { |
6f90844f YW |
595 | int r; |
596 | ||
597 | r = parse_boolean(n); | |
598 | if (r > 0) | |
599 | return SOCKET_ADDRESS_IPV6_ONLY; | |
600 | if (r == 0) | |
601 | return SOCKET_ADDRESS_BOTH; | |
602 | ||
603 | return socket_address_bind_ipv6_only_from_string(n); | |
604 | } | |
605 | ||
f01e5736 LP |
606 | bool sockaddr_equal(const union sockaddr_union *a, const union sockaddr_union *b) { |
607 | assert(a); | |
608 | assert(b); | |
609 | ||
610 | if (a->sa.sa_family != b->sa.sa_family) | |
611 | return false; | |
612 | ||
613 | if (a->sa.sa_family == AF_INET) | |
614 | return a->in.sin_addr.s_addr == b->in.sin_addr.s_addr; | |
615 | ||
616 | if (a->sa.sa_family == AF_INET6) | |
617 | return memcmp(&a->in6.sin6_addr, &b->in6.sin6_addr, sizeof(a->in6.sin6_addr)) == 0; | |
618 | ||
0fc0f14b SH |
619 | if (a->sa.sa_family == AF_VSOCK) |
620 | return a->vm.svm_cid == b->vm.svm_cid; | |
621 | ||
f01e5736 LP |
622 | return false; |
623 | } | |
2583fbea | 624 | |
d9d9b2a0 | 625 | int fd_set_sndbuf(int fd, size_t n, bool increase) { |
2583fbea LP |
626 | int r, value; |
627 | socklen_t l = sizeof(value); | |
628 | ||
1263c85e YW |
629 | if (n > INT_MAX) |
630 | return -ERANGE; | |
631 | ||
2583fbea | 632 | r = getsockopt(fd, SOL_SOCKET, SO_SNDBUF, &value, &l); |
d9d9b2a0 | 633 | if (r >= 0 && l == sizeof(value) && increase ? (size_t) value >= n*2 : (size_t) value == n*2) |
2583fbea LP |
634 | return 0; |
635 | ||
b92f3507 YW |
636 | /* First, try to set the buffer size with SO_SNDBUF. */ |
637 | r = setsockopt_int(fd, SOL_SOCKET, SO_SNDBUF, n); | |
638 | if (r < 0) | |
639 | return r; | |
2583fbea | 640 | |
b92f3507 YW |
641 | /* SO_SNDBUF above may set to the kernel limit, instead of the requested size. |
642 | * So, we need to check the actual buffer size here. */ | |
67f5ae2d | 643 | l = sizeof(value); |
b92f3507 | 644 | r = getsockopt(fd, SOL_SOCKET, SO_SNDBUF, &value, &l); |
d9d9b2a0 | 645 | if (r >= 0 && l == sizeof(value) && increase ? (size_t) value >= n*2 : (size_t) value == n*2) |
b92f3507 YW |
646 | return 1; |
647 | ||
648 | /* If we have the privileges we will ignore the kernel limit. */ | |
649 | r = setsockopt_int(fd, SOL_SOCKET, SO_SNDBUFFORCE, n); | |
650 | if (r < 0) | |
651 | return r; | |
2583fbea LP |
652 | |
653 | return 1; | |
654 | } | |
655 | ||
d9d9b2a0 | 656 | int fd_set_rcvbuf(int fd, size_t n, bool increase) { |
2583fbea LP |
657 | int r, value; |
658 | socklen_t l = sizeof(value); | |
659 | ||
1263c85e YW |
660 | if (n > INT_MAX) |
661 | return -ERANGE; | |
662 | ||
2583fbea | 663 | r = getsockopt(fd, SOL_SOCKET, SO_RCVBUF, &value, &l); |
d9d9b2a0 | 664 | if (r >= 0 && l == sizeof(value) && increase ? (size_t) value >= n*2 : (size_t) value == n*2) |
2583fbea LP |
665 | return 0; |
666 | ||
b92f3507 YW |
667 | /* First, try to set the buffer size with SO_RCVBUF. */ |
668 | r = setsockopt_int(fd, SOL_SOCKET, SO_RCVBUF, n); | |
669 | if (r < 0) | |
670 | return r; | |
2583fbea | 671 | |
b92f3507 YW |
672 | /* SO_RCVBUF above may set to the kernel limit, instead of the requested size. |
673 | * So, we need to check the actual buffer size here. */ | |
67f5ae2d | 674 | l = sizeof(value); |
b92f3507 | 675 | r = getsockopt(fd, SOL_SOCKET, SO_RCVBUF, &value, &l); |
d9d9b2a0 | 676 | if (r >= 0 && l == sizeof(value) && increase ? (size_t) value >= n*2 : (size_t) value == n*2) |
b92f3507 YW |
677 | return 1; |
678 | ||
679 | /* If we have the privileges we will ignore the kernel limit. */ | |
680 | r = setsockopt_int(fd, SOL_SOCKET, SO_RCVBUFFORCE, n); | |
681 | if (r < 0) | |
682 | return r; | |
9e5b6496 | 683 | |
2583fbea LP |
684 | return 1; |
685 | } | |
686 | ||
687 | static const char* const ip_tos_table[] = { | |
688 | [IPTOS_LOWDELAY] = "low-delay", | |
689 | [IPTOS_THROUGHPUT] = "throughput", | |
690 | [IPTOS_RELIABILITY] = "reliability", | |
691 | [IPTOS_LOWCOST] = "low-cost", | |
692 | }; | |
693 | ||
694 | DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(ip_tos, int, 0xff); | |
695 | ||
2313524a | 696 | bool ifname_valid_full(const char *p, IfnameValidFlags flags) { |
ef76dff2 LP |
697 | bool numeric = true; |
698 | ||
699 | /* Checks whether a network interface name is valid. This is inspired by dev_valid_name() in the kernel sources | |
700 | * but slightly stricter, as we only allow non-control, non-space ASCII characters in the interface name. We | |
701 | * also don't permit names that only container numbers, to avoid confusion with numeric interface indexes. */ | |
702 | ||
2313524a ZJS |
703 | assert(!(flags & ~_IFNAME_VALID_ALL)); |
704 | ||
ef76dff2 LP |
705 | if (isempty(p)) |
706 | return false; | |
707 | ||
2313524a | 708 | if (flags & IFNAME_VALID_ALTERNATIVE) { |
4252696a YW |
709 | if (strlen(p) >= ALTIFNAMSIZ) |
710 | return false; | |
711 | } else { | |
712 | if (strlen(p) >= IFNAMSIZ) | |
713 | return false; | |
714 | } | |
ef76dff2 | 715 | |
49bfc877 | 716 | if (dot_or_dot_dot(p)) |
ef76dff2 LP |
717 | return false; |
718 | ||
2313524a ZJS |
719 | for (const char *t = p; *t; t++) { |
720 | if ((unsigned char) *t >= 127U) | |
ef76dff2 LP |
721 | return false; |
722 | ||
2313524a | 723 | if ((unsigned char) *t <= 32U) |
ef76dff2 LP |
724 | return false; |
725 | ||
2313524a | 726 | if (IN_SET(*t, ':', '/')) |
ef76dff2 LP |
727 | return false; |
728 | ||
2313524a | 729 | numeric = numeric && (*t >= '0' && *t <= '9'); |
ef76dff2 LP |
730 | } |
731 | ||
2313524a ZJS |
732 | if (numeric) { |
733 | if (!(flags & IFNAME_VALID_NUMERIC)) | |
734 | return false; | |
735 | ||
736 | /* Verify that the number is well-formatted and in range. */ | |
737 | if (parse_ifindex(p) < 0) | |
738 | return false; | |
739 | } | |
ef76dff2 LP |
740 | |
741 | return true; | |
742 | } | |
743 | ||
26808948 SS |
744 | bool address_label_valid(const char *p) { |
745 | ||
746 | if (isempty(p)) | |
747 | return false; | |
748 | ||
749 | if (strlen(p) >= IFNAMSIZ) | |
750 | return false; | |
751 | ||
752 | while (*p) { | |
753 | if ((uint8_t) *p >= 127U) | |
754 | return false; | |
755 | ||
756 | if ((uint8_t) *p <= 31U) | |
757 | return false; | |
758 | p++; | |
759 | } | |
760 | ||
761 | return true; | |
762 | } | |
763 | ||
2583fbea LP |
764 | int getpeercred(int fd, struct ucred *ucred) { |
765 | socklen_t n = sizeof(struct ucred); | |
766 | struct ucred u; | |
767 | int r; | |
768 | ||
769 | assert(fd >= 0); | |
770 | assert(ucred); | |
771 | ||
772 | r = getsockopt(fd, SOL_SOCKET, SO_PEERCRED, &u, &n); | |
773 | if (r < 0) | |
774 | return -errno; | |
775 | ||
776 | if (n != sizeof(struct ucred)) | |
777 | return -EIO; | |
778 | ||
bbcc701e LP |
779 | /* Check if the data is actually useful and not suppressed due to namespacing issues */ |
780 | if (!pid_is_valid(u.pid)) | |
2583fbea LP |
781 | return -ENODATA; |
782 | ||
bbcc701e LP |
783 | /* Note that we don't check UID/GID here, as namespace translation works differently there: instead of |
784 | * receiving in "invalid" user/group we get the overflow UID/GID. */ | |
785 | ||
2583fbea LP |
786 | *ucred = u; |
787 | return 0; | |
788 | } | |
789 | ||
790 | int getpeersec(int fd, char **ret) { | |
217d8967 | 791 | _cleanup_free_ char *s = NULL; |
2583fbea | 792 | socklen_t n = 64; |
2583fbea LP |
793 | |
794 | assert(fd >= 0); | |
795 | assert(ret); | |
796 | ||
217d8967 LP |
797 | for (;;) { |
798 | s = new0(char, n+1); | |
799 | if (!s) | |
800 | return -ENOMEM; | |
2583fbea | 801 | |
217d8967 LP |
802 | if (getsockopt(fd, SOL_SOCKET, SO_PEERSEC, s, &n) >= 0) |
803 | break; | |
2583fbea LP |
804 | |
805 | if (errno != ERANGE) | |
806 | return -errno; | |
807 | ||
217d8967 | 808 | s = mfree(s); |
2583fbea LP |
809 | } |
810 | ||
217d8967 | 811 | if (isempty(s)) |
2583fbea | 812 | return -EOPNOTSUPP; |
2583fbea | 813 | |
ae2a15bc | 814 | *ret = TAKE_PTR(s); |
217d8967 | 815 | |
2583fbea LP |
816 | return 0; |
817 | } | |
818 | ||
43f2c88d LP |
819 | int getpeergroups(int fd, gid_t **ret) { |
820 | socklen_t n = sizeof(gid_t) * 64; | |
821 | _cleanup_free_ gid_t *d = NULL; | |
822 | ||
75f40779 | 823 | assert(fd >= 0); |
43f2c88d LP |
824 | assert(ret); |
825 | ||
826 | for (;;) { | |
827 | d = malloc(n); | |
828 | if (!d) | |
829 | return -ENOMEM; | |
830 | ||
831 | if (getsockopt(fd, SOL_SOCKET, SO_PEERGROUPS, d, &n) >= 0) | |
832 | break; | |
833 | ||
834 | if (errno != ERANGE) | |
835 | return -errno; | |
836 | ||
837 | d = mfree(d); | |
838 | } | |
839 | ||
840 | assert_se(n % sizeof(gid_t) == 0); | |
841 | n /= sizeof(gid_t); | |
842 | ||
843 | if ((socklen_t) (int) n != n) | |
844 | return -E2BIG; | |
845 | ||
1cc6c93a | 846 | *ret = TAKE_PTR(d); |
43f2c88d LP |
847 | |
848 | return (int) n; | |
849 | } | |
850 | ||
d34673ec | 851 | ssize_t send_one_fd_iov_sa( |
726f4c47 ZJS |
852 | int transport_fd, |
853 | int fd, | |
d34673ec | 854 | struct iovec *iov, size_t iovlen, |
726f4c47 ZJS |
855 | const struct sockaddr *sa, socklen_t len, |
856 | int flags) { | |
857 | ||
fb29cdbe | 858 | CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(int))) control = {}; |
2583fbea | 859 | struct msghdr mh = { |
726f4c47 ZJS |
860 | .msg_name = (struct sockaddr*) sa, |
861 | .msg_namelen = len, | |
d34673ec FB |
862 | .msg_iov = iov, |
863 | .msg_iovlen = iovlen, | |
2583fbea | 864 | }; |
d34673ec | 865 | ssize_t k; |
2583fbea LP |
866 | |
867 | assert(transport_fd >= 0); | |
2583fbea | 868 | |
d34673ec FB |
869 | /* |
870 | * We need either an FD or data to send. | |
871 | * If there's nothing, return an error. | |
872 | */ | |
873 | if (fd < 0 && !iov) | |
874 | return -EINVAL; | |
2583fbea | 875 | |
d34673ec FB |
876 | if (fd >= 0) { |
877 | struct cmsghdr *cmsg; | |
2583fbea | 878 | |
d34673ec FB |
879 | mh.msg_control = &control; |
880 | mh.msg_controllen = sizeof(control); | |
881 | ||
882 | cmsg = CMSG_FIRSTHDR(&mh); | |
883 | cmsg->cmsg_level = SOL_SOCKET; | |
884 | cmsg->cmsg_type = SCM_RIGHTS; | |
885 | cmsg->cmsg_len = CMSG_LEN(sizeof(int)); | |
886 | memcpy(CMSG_DATA(cmsg), &fd, sizeof(int)); | |
d34673ec FB |
887 | } |
888 | k = sendmsg(transport_fd, &mh, MSG_NOSIGNAL | flags); | |
889 | if (k < 0) | |
890 | return (ssize_t) -errno; | |
891 | ||
892 | return k; | |
2583fbea LP |
893 | } |
894 | ||
d34673ec FB |
895 | int send_one_fd_sa( |
896 | int transport_fd, | |
897 | int fd, | |
898 | const struct sockaddr *sa, socklen_t len, | |
899 | int flags) { | |
900 | ||
901 | assert(fd >= 0); | |
902 | ||
903 | return (int) send_one_fd_iov_sa(transport_fd, fd, NULL, 0, sa, len, flags); | |
904 | } | |
905 | ||
906 | ssize_t receive_one_fd_iov( | |
907 | int transport_fd, | |
908 | struct iovec *iov, size_t iovlen, | |
909 | int flags, | |
910 | int *ret_fd) { | |
911 | ||
fb29cdbe | 912 | CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(int))) control; |
2583fbea LP |
913 | struct msghdr mh = { |
914 | .msg_control = &control, | |
915 | .msg_controllen = sizeof(control), | |
d34673ec FB |
916 | .msg_iov = iov, |
917 | .msg_iovlen = iovlen, | |
2583fbea | 918 | }; |
dac556fa | 919 | struct cmsghdr *found; |
d34673ec | 920 | ssize_t k; |
2583fbea LP |
921 | |
922 | assert(transport_fd >= 0); | |
d34673ec | 923 | assert(ret_fd); |
2583fbea LP |
924 | |
925 | /* | |
926 | * Receive a single FD via @transport_fd. We don't care for | |
927 | * the transport-type. We retrieve a single FD at most, so for | |
928 | * packet-based transports, the caller must ensure to send | |
929 | * only a single FD per packet. This is best used in | |
930 | * combination with send_one_fd(). | |
931 | */ | |
932 | ||
3691bcf3 | 933 | k = recvmsg_safe(transport_fd, &mh, MSG_CMSG_CLOEXEC | flags); |
d34673ec | 934 | if (k < 0) |
3691bcf3 | 935 | return k; |
2583fbea | 936 | |
dac556fa | 937 | found = cmsg_find(&mh, SOL_SOCKET, SCM_RIGHTS, CMSG_LEN(sizeof(int))); |
3691bcf3 | 938 | if (!found) { |
2583fbea | 939 | cmsg_close_all(&mh); |
d34673ec | 940 | |
3691bcf3 LP |
941 | /* If didn't receive an FD or any data, return an error. */ |
942 | if (k == 0) | |
943 | return -EIO; | |
944 | } | |
2583fbea | 945 | |
d34673ec FB |
946 | if (found) |
947 | *ret_fd = *(int*) CMSG_DATA(found); | |
948 | else | |
949 | *ret_fd = -1; | |
950 | ||
951 | return k; | |
952 | } | |
953 | ||
954 | int receive_one_fd(int transport_fd, int flags) { | |
955 | int fd; | |
956 | ssize_t k; | |
957 | ||
958 | k = receive_one_fd_iov(transport_fd, NULL, 0, flags, &fd); | |
959 | if (k == 0) | |
960 | return fd; | |
961 | ||
962 | /* k must be negative, since receive_one_fd_iov() only returns | |
963 | * a positive value if data was received through the iov. */ | |
964 | assert(k < 0); | |
965 | return (int) k; | |
2583fbea | 966 | } |
4edc2c9b LP |
967 | |
968 | ssize_t next_datagram_size_fd(int fd) { | |
969 | ssize_t l; | |
970 | int k; | |
971 | ||
972 | /* This is a bit like FIONREAD/SIOCINQ, however a bit more powerful. The difference being: recv(MSG_PEEK) will | |
96d49011 | 973 | * actually cause the next datagram in the queue to be validated regarding checksums, which FIONREAD doesn't |
4edc2c9b LP |
974 | * do. This difference is actually of major importance as we need to be sure that the size returned here |
975 | * actually matches what we will read with recvmsg() next, as otherwise we might end up allocating a buffer of | |
976 | * the wrong size. */ | |
977 | ||
978 | l = recv(fd, NULL, 0, MSG_PEEK|MSG_TRUNC); | |
979 | if (l < 0) { | |
3742095b | 980 | if (IN_SET(errno, EOPNOTSUPP, EFAULT)) |
4edc2c9b LP |
981 | goto fallback; |
982 | ||
983 | return -errno; | |
984 | } | |
985 | if (l == 0) | |
986 | goto fallback; | |
987 | ||
988 | return l; | |
989 | ||
990 | fallback: | |
991 | k = 0; | |
992 | ||
993 | /* Some sockets (AF_PACKET) do not support null-sized recv() with MSG_TRUNC set, let's fall back to FIONREAD | |
994 | * for them. Checksums don't matter for raw sockets anyway, hence this should be fine. */ | |
995 | ||
996 | if (ioctl(fd, FIONREAD, &k) < 0) | |
997 | return -errno; | |
998 | ||
999 | return (ssize_t) k; | |
1000 | } | |
60d9771c | 1001 | |
67962036 ZJS |
1002 | /* Put a limit on how many times will attempt to call accept4(). We loop |
1003 | * only on "transient" errors, but let's make sure we don't loop forever. */ | |
1004 | #define MAX_FLUSH_ITERATIONS 1024 | |
1005 | ||
60d9771c LP |
1006 | int flush_accept(int fd) { |
1007 | ||
f3d75364 LP |
1008 | int r, b; |
1009 | socklen_t l = sizeof(b); | |
1010 | ||
644cb4cc ZJS |
1011 | /* Similar to flush_fd() but flushes all incoming connections by accepting and immediately closing |
1012 | * them. */ | |
f3d75364 LP |
1013 | |
1014 | if (getsockopt(fd, SOL_SOCKET, SO_ACCEPTCONN, &b, &l) < 0) | |
1015 | return -errno; | |
60d9771c | 1016 | |
f3d75364 | 1017 | assert(l == sizeof(b)); |
644cb4cc ZJS |
1018 | if (!b) /* Let's check if this socket accepts connections before calling accept(). accept4() can |
1019 | * return EOPNOTSUPP if the fd is not a listening socket, which we should treat as a fatal | |
1020 | * error, or in case the incoming TCP connection triggered a network issue, which we want to | |
1021 | * treat as a transient error. Thus, let's rule out the first reason for EOPNOTSUPP early, so | |
1022 | * we can loop safely on transient errors below. */ | |
f3d75364 | 1023 | return -ENOTTY; |
60d9771c | 1024 | |
67962036 | 1025 | for (unsigned iteration = 0;; iteration++) { |
60d9771c LP |
1026 | int cfd; |
1027 | ||
0f2d351f | 1028 | r = fd_wait_for_event(fd, POLLIN, 0); |
60d9771c | 1029 | if (r < 0) { |
0f2d351f | 1030 | if (r == -EINTR) |
60d9771c LP |
1031 | continue; |
1032 | ||
0f2d351f | 1033 | return r; |
4ff9bc2e LP |
1034 | } |
1035 | if (r == 0) | |
60d9771c LP |
1036 | return 0; |
1037 | ||
67962036 ZJS |
1038 | if (iteration >= MAX_FLUSH_ITERATIONS) |
1039 | return log_debug_errno(SYNTHETIC_ERRNO(EBUSY), | |
1040 | "Failed to flush connections within " STRINGIFY(MAX_FLUSH_ITERATIONS) " iterations."); | |
1041 | ||
60d9771c LP |
1042 | cfd = accept4(fd, NULL, NULL, SOCK_NONBLOCK|SOCK_CLOEXEC); |
1043 | if (cfd < 0) { | |
60d9771c LP |
1044 | if (errno == EAGAIN) |
1045 | return 0; | |
1046 | ||
4ff9bc2e LP |
1047 | if (ERRNO_IS_ACCEPT_AGAIN(errno)) |
1048 | continue; | |
1049 | ||
60d9771c LP |
1050 | return -errno; |
1051 | } | |
1052 | ||
4ff9bc2e | 1053 | safe_close(cfd); |
60d9771c LP |
1054 | } |
1055 | } | |
29206d46 LP |
1056 | |
1057 | struct cmsghdr* cmsg_find(struct msghdr *mh, int level, int type, socklen_t length) { | |
1058 | struct cmsghdr *cmsg; | |
1059 | ||
1060 | assert(mh); | |
1061 | ||
1062 | CMSG_FOREACH(cmsg, mh) | |
1063 | if (cmsg->cmsg_level == level && | |
1064 | cmsg->cmsg_type == type && | |
1065 | (length == (socklen_t) -1 || length == cmsg->cmsg_len)) | |
1066 | return cmsg; | |
1067 | ||
1068 | return NULL; | |
1069 | } | |
429b4350 LP |
1070 | |
1071 | int socket_ioctl_fd(void) { | |
1072 | int fd; | |
1073 | ||
1074 | /* Create a socket to invoke the various network interface ioctl()s on. Traditionally only AF_INET was good for | |
1075 | * that. Since kernel 4.6 AF_NETLINK works for this too. We first try to use AF_INET hence, but if that's not | |
1076 | * available (for example, because it is made unavailable via SECCOMP or such), we'll fall back to the more | |
1077 | * generic AF_NETLINK. */ | |
1078 | ||
1079 | fd = socket(AF_INET, SOCK_DGRAM|SOCK_CLOEXEC, 0); | |
1080 | if (fd < 0) | |
1081 | fd = socket(AF_NETLINK, SOCK_RAW|SOCK_CLOEXEC, NETLINK_GENERIC); | |
1082 | if (fd < 0) | |
1083 | return -errno; | |
1084 | ||
1085 | return fd; | |
1086 | } | |
9f20fc28 LP |
1087 | |
1088 | int sockaddr_un_unlink(const struct sockaddr_un *sa) { | |
1089 | const char *p, * nul; | |
1090 | ||
1091 | assert(sa); | |
1092 | ||
1093 | if (sa->sun_family != AF_UNIX) | |
1094 | return -EPROTOTYPE; | |
1095 | ||
1096 | if (sa->sun_path[0] == 0) /* Nothing to do for abstract sockets */ | |
1097 | return 0; | |
1098 | ||
1099 | /* The path in .sun_path is not necessarily NUL terminated. Let's fix that. */ | |
1100 | nul = memchr(sa->sun_path, 0, sizeof(sa->sun_path)); | |
1101 | if (nul) | |
1102 | p = sa->sun_path; | |
1103 | else | |
1104 | p = memdupa_suffix0(sa->sun_path, sizeof(sa->sun_path)); | |
1105 | ||
1106 | if (unlink(p) < 0) | |
1107 | return -errno; | |
1108 | ||
1109 | return 1; | |
1110 | } | |
5cf91ea9 LP |
1111 | |
1112 | int sockaddr_un_set_path(struct sockaddr_un *ret, const char *path) { | |
1113 | size_t l; | |
1114 | ||
1115 | assert(ret); | |
1116 | assert(path); | |
1117 | ||
1118 | /* Initialize ret->sun_path from the specified argument. This will interpret paths starting with '@' as | |
1119 | * abstract namespace sockets, and those starting with '/' as regular filesystem sockets. It won't accept | |
1120 | * anything else (i.e. no relative paths), to avoid ambiguities. Note that this function cannot be used to | |
1121 | * reference paths in the abstract namespace that include NUL bytes in the name. */ | |
1122 | ||
1123 | l = strlen(path); | |
c097bf1f | 1124 | if (l < 2) |
5cf91ea9 LP |
1125 | return -EINVAL; |
1126 | if (!IN_SET(path[0], '/', '@')) | |
1127 | return -EINVAL; | |
5cf91ea9 LP |
1128 | |
1129 | /* Don't allow paths larger than the space in sockaddr_un. Note that we are a tiny bit more restrictive than | |
1130 | * the kernel is: we insist on NUL termination (both for abstract namespace and regular file system socket | |
1131 | * addresses!), which the kernel doesn't. We do this to reduce chance of incompatibility with other apps that | |
1132 | * do not expect non-NUL terminated file system path*/ | |
1133 | if (l+1 > sizeof(ret->sun_path)) | |
1134 | return -EINVAL; | |
1135 | ||
1136 | *ret = (struct sockaddr_un) { | |
1137 | .sun_family = AF_UNIX, | |
1138 | }; | |
1139 | ||
1140 | if (path[0] == '@') { | |
1141 | /* Abstract namespace socket */ | |
1142 | memcpy(ret->sun_path + 1, path + 1, l); /* copy *with* trailing NUL byte */ | |
1143 | return (int) (offsetof(struct sockaddr_un, sun_path) + l); /* 🔥 *don't* 🔥 include trailing NUL in size */ | |
1144 | ||
1145 | } else { | |
1146 | assert(path[0] == '/'); | |
1147 | ||
1148 | /* File system socket */ | |
1149 | memcpy(ret->sun_path, path, l + 1); /* copy *with* trailing NUL byte */ | |
1150 | return (int) (offsetof(struct sockaddr_un, sun_path) + l + 1); /* include trailing NUL in size */ | |
1151 | } | |
1152 | } | |
5d594d01 LP |
1153 | |
1154 | int socket_bind_to_ifname(int fd, const char *ifname) { | |
1155 | assert(fd >= 0); | |
1156 | ||
1157 | /* Call with NULL to drop binding */ | |
1158 | ||
1159 | if (setsockopt(fd, SOL_SOCKET, SO_BINDTODEVICE, ifname, strlen_ptr(ifname)) < 0) | |
1160 | return -errno; | |
1161 | ||
1162 | return 0; | |
1163 | } | |
1164 | ||
1165 | int socket_bind_to_ifindex(int fd, int ifindex) { | |
518a66ec | 1166 | char ifname[IF_NAMESIZE + 1]; |
5e958e1d | 1167 | int r; |
5d594d01 LP |
1168 | |
1169 | assert(fd >= 0); | |
1170 | ||
1171 | if (ifindex <= 0) { | |
1172 | /* Drop binding */ | |
1173 | if (setsockopt(fd, SOL_SOCKET, SO_BINDTODEVICE, NULL, 0) < 0) | |
1174 | return -errno; | |
1175 | ||
1176 | return 0; | |
1177 | } | |
1178 | ||
5e958e1d YW |
1179 | r = setsockopt_int(fd, SOL_SOCKET, SO_BINDTOIFINDEX, ifindex); |
1180 | if (r != -ENOPROTOOPT) | |
1181 | return r; | |
5d594d01 LP |
1182 | |
1183 | /* Fall back to SO_BINDTODEVICE on kernels < 5.0 which didn't have SO_BINDTOIFINDEX */ | |
518a66ec | 1184 | if (!format_ifname(ifindex, ifname)) |
5d594d01 LP |
1185 | return -errno; |
1186 | ||
1187 | return socket_bind_to_ifname(fd, ifname); | |
1188 | } | |
47eae6ce LP |
1189 | |
1190 | ssize_t recvmsg_safe(int sockfd, struct msghdr *msg, int flags) { | |
1191 | ssize_t n; | |
1192 | ||
1193 | /* A wrapper around recvmsg() that checks for MSG_CTRUNC, and turns it into an error, in a reasonably | |
1194 | * safe way, closing any SCM_RIGHTS fds in the error path. | |
1195 | * | |
1196 | * Note that unlike our usual coding style this might modify *msg on failure. */ | |
1197 | ||
1198 | n = recvmsg(sockfd, msg, flags); | |
1199 | if (n < 0) | |
1200 | return -errno; | |
1201 | ||
1202 | if (FLAGS_SET(msg->msg_flags, MSG_CTRUNC)) { | |
1203 | cmsg_close_all(msg); | |
1204 | return -EXFULL; /* a recognizable error code */ | |
1205 | } | |
1206 | ||
1207 | return n; | |
35a3eb9b LP |
1208 | } |
1209 | ||
5d0fe423 | 1210 | int socket_get_family(int fd, int *ret) { |
35a3eb9b LP |
1211 | int af; |
1212 | socklen_t sl = sizeof(af); | |
1213 | ||
1214 | if (getsockopt(fd, SOL_SOCKET, SO_DOMAIN, &af, &sl) < 0) | |
1215 | return -errno; | |
1216 | ||
5d0fe423 LP |
1217 | if (sl != sizeof(af)) |
1218 | return -EINVAL; | |
1219 | ||
1220 | return af; | |
1221 | } | |
1222 | ||
1223 | int socket_set_recvpktinfo(int fd, int af, bool b) { | |
1224 | int r; | |
1225 | ||
1226 | if (af == AF_UNSPEC) { | |
1227 | r = socket_get_family(fd, &af); | |
1228 | if (r < 0) | |
1229 | return r; | |
1230 | } | |
1231 | ||
35a3eb9b LP |
1232 | switch (af) { |
1233 | ||
1234 | case AF_INET: | |
1235 | return setsockopt_int(fd, IPPROTO_IP, IP_PKTINFO, b); | |
47eae6ce | 1236 | |
35a3eb9b LP |
1237 | case AF_INET6: |
1238 | return setsockopt_int(fd, IPPROTO_IPV6, IPV6_RECVPKTINFO, b); | |
1239 | ||
1240 | case AF_NETLINK: | |
1241 | return setsockopt_int(fd, SOL_NETLINK, NETLINK_PKTINFO, b); | |
1242 | ||
1243 | default: | |
1244 | return -EAFNOSUPPORT; | |
1245 | } | |
47eae6ce | 1246 | } |
5d0fe423 LP |
1247 | |
1248 | int socket_set_recverr(int fd, int af, bool b) { | |
1249 | int r; | |
1250 | ||
1251 | if (af == AF_UNSPEC) { | |
1252 | r = socket_get_family(fd, &af); | |
1253 | if (r < 0) | |
1254 | return r; | |
1255 | } | |
1256 | ||
1257 | switch (af) { | |
1258 | ||
1259 | case AF_INET: | |
1260 | return setsockopt_int(fd, IPPROTO_IP, IP_RECVERR, b); | |
1261 | ||
1262 | case AF_INET6: | |
1263 | return setsockopt_int(fd, IPPROTO_IPV6, IPV6_RECVERR, b); | |
1264 | ||
1265 | default: | |
1266 | return -EAFNOSUPPORT; | |
1267 | } | |
1268 | } | |
1269 | ||
1270 | int socket_set_recvttl(int fd, int af, bool b) { | |
1271 | int r; | |
1272 | ||
1273 | if (af == AF_UNSPEC) { | |
1274 | r = socket_get_family(fd, &af); | |
1275 | if (r < 0) | |
1276 | return r; | |
1277 | } | |
1278 | ||
1279 | switch (af) { | |
1280 | ||
1281 | case AF_INET: | |
1282 | return setsockopt_int(fd, IPPROTO_IP, IP_RECVTTL, b); | |
1283 | ||
1284 | case AF_INET6: | |
1285 | return setsockopt_int(fd, IPPROTO_IPV6, IPV6_RECVHOPLIMIT, b); | |
1286 | ||
1287 | default: | |
1288 | return -EAFNOSUPPORT; | |
1289 | } | |
1290 | } | |
1291 | ||
1292 | int socket_set_ttl(int fd, int af, int ttl) { | |
1293 | int r; | |
1294 | ||
1295 | if (af == AF_UNSPEC) { | |
1296 | r = socket_get_family(fd, &af); | |
1297 | if (r < 0) | |
1298 | return r; | |
1299 | } | |
1300 | ||
1301 | switch (af) { | |
1302 | ||
1303 | case AF_INET: | |
1304 | return setsockopt_int(fd, IPPROTO_IP, IP_TTL, ttl); | |
1305 | ||
1306 | case AF_INET6: | |
1307 | return setsockopt_int(fd, IPPROTO_IPV6, IPV6_UNICAST_HOPS, ttl); | |
1308 | ||
1309 | default: | |
1310 | return -EAFNOSUPPORT; | |
1311 | } | |
1312 | } | |
1313 | ||
1314 | int socket_set_unicast_if(int fd, int af, int ifi) { | |
1315 | be32_t ifindex_be = htobe32(ifi); | |
1316 | int r; | |
1317 | ||
1318 | if (af == AF_UNSPEC) { | |
1319 | r = socket_get_family(fd, &af); | |
1320 | if (r < 0) | |
1321 | return r; | |
1322 | } | |
1323 | ||
1324 | switch (af) { | |
1325 | ||
1326 | case AF_INET: | |
1327 | if (setsockopt(fd, IPPROTO_IP, IP_UNICAST_IF, &ifindex_be, sizeof(ifindex_be)) < 0) | |
1328 | return -errno; | |
1329 | ||
1330 | return 0; | |
1331 | ||
1332 | case AF_INET6: | |
1333 | if (setsockopt(fd, IPPROTO_IPV6, IPV6_UNICAST_IF, &ifindex_be, sizeof(ifindex_be)) < 0) | |
1334 | return -errno; | |
1335 | ||
1336 | return 0; | |
1337 | ||
1338 | default: | |
1339 | return -EAFNOSUPPORT; | |
1340 | } | |
1341 | } | |
1342 | ||
1343 | int socket_set_freebind(int fd, int af, bool b) { | |
1344 | int r; | |
1345 | ||
1346 | if (af == AF_UNSPEC) { | |
1347 | r = socket_get_family(fd, &af); | |
1348 | if (r < 0) | |
1349 | return r; | |
1350 | } | |
1351 | ||
1352 | switch (af) { | |
1353 | ||
1354 | case AF_INET: | |
1355 | return setsockopt_int(fd, IPPROTO_IP, IP_FREEBIND, b); | |
1356 | ||
1357 | case AF_INET6: | |
1358 | return setsockopt_int(fd, IPPROTO_IPV6, IPV6_FREEBIND, b); | |
1359 | ||
1360 | default: | |
1361 | return -EAFNOSUPPORT; | |
1362 | } | |
1363 | } | |
1364 | ||
1365 | int socket_set_transparent(int fd, int af, bool b) { | |
1366 | int r; | |
1367 | ||
1368 | if (af == AF_UNSPEC) { | |
1369 | r = socket_get_family(fd, &af); | |
1370 | if (r < 0) | |
1371 | return r; | |
1372 | } | |
1373 | ||
1374 | switch (af) { | |
1375 | ||
1376 | case AF_INET: | |
1377 | return setsockopt_int(fd, IPPROTO_IP, IP_TRANSPARENT, b); | |
1378 | ||
1379 | case AF_INET6: | |
1380 | return setsockopt_int(fd, IPPROTO_IPV6, IPV6_TRANSPARENT, b); | |
1381 | ||
1382 | default: | |
1383 | return -EAFNOSUPPORT; | |
1384 | } | |
1385 | } |