]>
Commit | Line | Data |
---|---|---|
db9ecf05 | 1 | /* SPDX-License-Identifier: LGPL-2.1-or-later */ |
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" |
0a970718 | 26 | #include "memory-util.h" |
6bedfcbb | 27 | #include "parse-util.h" |
9eb977db | 28 | #include "path-util.h" |
dccca82b | 29 | #include "process-util.h" |
2583fbea | 30 | #include "socket-util.h" |
8b43440b | 31 | #include "string-table.h" |
07630cea | 32 | #include "string-util.h" |
ef76dff2 | 33 | #include "strv.h" |
83e03c4f | 34 | #include "sysctl-util.h" |
ee104e11 | 35 | #include "user-util.h" |
ef76dff2 | 36 | #include "utf8.h" |
42f4e3c4 | 37 | |
349cc4a5 | 38 | #if ENABLE_IDN |
cadc80b8 | 39 | # define IDN_FLAGS NI_IDN |
6326a143 WB |
40 | #else |
41 | # define IDN_FLAGS 0 | |
42 | #endif | |
43 | ||
598d2428 LB |
44 | /* From the kernel's include/net/scm.h */ |
45 | #ifndef SCM_MAX_FD | |
46 | # define SCM_MAX_FD 253 | |
47 | #endif | |
48 | ||
398ce0bc | 49 | static const char* const socket_address_type_table[] = { |
955bb7fa ZJS |
50 | [SOCK_STREAM] = "Stream", |
51 | [SOCK_DGRAM] = "Datagram", | |
52 | [SOCK_RAW] = "Raw", | |
53 | [SOCK_RDM] = "ReliableDatagram", | |
398ce0bc | 54 | [SOCK_SEQPACKET] = "SequentialPacket", |
955bb7fa | 55 | [SOCK_DCCP] = "DatagramCongestionControl", |
398ce0bc YW |
56 | }; |
57 | ||
58 | DEFINE_STRING_TABLE_LOOKUP(socket_address_type, int); | |
59 | ||
15dca371 | 60 | int socket_address_verify(const SocketAddress *a, bool strict) { |
42f4e3c4 LP |
61 | assert(a); |
62 | ||
15dca371 ZJS |
63 | /* With 'strict' we enforce additional sanity constraints which are not set by the standard, |
64 | * but should only apply to sockets we create ourselves. */ | |
65 | ||
542563ba | 66 | switch (socket_address_family(a)) { |
42f4e3c4 | 67 | |
7a22745a LP |
68 | case AF_INET: |
69 | if (a->size != sizeof(struct sockaddr_in)) | |
70 | return -EINVAL; | |
42f4e3c4 | 71 | |
4d49b48c | 72 | if (a->sockaddr.in.sin_port == 0) |
7a22745a | 73 | return -EINVAL; |
42f4e3c4 | 74 | |
44ab2347 | 75 | if (!IN_SET(a->type, 0, SOCK_STREAM, SOCK_DGRAM)) |
7a22745a | 76 | return -EINVAL; |
42f4e3c4 | 77 | |
7a22745a LP |
78 | return 0; |
79 | ||
80 | case AF_INET6: | |
81 | if (a->size != sizeof(struct sockaddr_in6)) | |
82 | return -EINVAL; | |
42f4e3c4 | 83 | |
7a22745a LP |
84 | if (a->sockaddr.in6.sin6_port == 0) |
85 | return -EINVAL; | |
42f4e3c4 | 86 | |
44ab2347 | 87 | if (!IN_SET(a->type, 0, SOCK_STREAM, SOCK_DGRAM)) |
7a22745a | 88 | return -EINVAL; |
42f4e3c4 | 89 | |
7a22745a | 90 | return 0; |
42f4e3c4 | 91 | |
7a22745a LP |
92 | case AF_UNIX: |
93 | if (a->size < offsetof(struct sockaddr_un, sun_path)) | |
94 | return -EINVAL; | |
15dca371 ZJS |
95 | if (a->size > sizeof(struct sockaddr_un) + !strict) |
96 | /* If !strict, allow one extra byte, since getsockname() on Linux will append | |
97 | * a NUL byte if we have path sockets that are above sun_path's full size. */ | |
8e8132c6 | 98 | return -EINVAL; |
42f4e3c4 | 99 | |
8e8132c6 | 100 | if (a->size > offsetof(struct sockaddr_un, sun_path) && |
15dca371 ZJS |
101 | a->sockaddr.un.sun_path[0] != 0 && |
102 | strict) { | |
103 | /* Only validate file system sockets here, and only in strict mode */ | |
8e8132c6 | 104 | const char *e; |
7a22745a | 105 | |
8e8132c6 LP |
106 | e = memchr(a->sockaddr.un.sun_path, 0, sizeof(a->sockaddr.un.sun_path)); |
107 | if (e) { | |
15dca371 | 108 | /* If there's an embedded NUL byte, make sure the size of the socket address matches it */ |
7a22745a LP |
109 | if (a->size != offsetof(struct sockaddr_un, sun_path) + (e - a->sockaddr.un.sun_path) + 1) |
110 | return -EINVAL; | |
8e8132c6 | 111 | } else { |
d7b34e38 | 112 | /* If there's no embedded NUL byte, then the size needs to match the whole |
8e8132c6 LP |
113 | * structure or the structure with one extra NUL byte suffixed. (Yeah, Linux is awful, |
114 | * and considers both equivalent: getsockname() even extends sockaddr_un beyond its | |
7802194a | 115 | * size if the path is non NUL terminated.) */ |
8e8132c6 LP |
116 | if (!IN_SET(a->size, sizeof(a->sockaddr.un.sun_path), sizeof(a->sockaddr.un.sun_path)+1)) |
117 | return -EINVAL; | |
42f4e3c4 | 118 | } |
7a22745a | 119 | } |
42f4e3c4 | 120 | |
44ab2347 | 121 | if (!IN_SET(a->type, 0, SOCK_STREAM, SOCK_DGRAM, SOCK_SEQPACKET)) |
7a22745a | 122 | return -EINVAL; |
42f4e3c4 | 123 | |
7a22745a LP |
124 | return 0; |
125 | ||
126 | case AF_NETLINK: | |
127 | ||
128 | if (a->size != sizeof(struct sockaddr_nl)) | |
129 | return -EINVAL; | |
130 | ||
44ab2347 | 131 | if (!IN_SET(a->type, 0, SOCK_RAW, SOCK_DGRAM)) |
7a22745a LP |
132 | return -EINVAL; |
133 | ||
134 | return 0; | |
135 | ||
0fc0f14b SH |
136 | case AF_VSOCK: |
137 | if (a->size != sizeof(struct sockaddr_vm)) | |
138 | return -EINVAL; | |
139 | ||
44ab2347 | 140 | if (!IN_SET(a->type, 0, SOCK_STREAM, SOCK_DGRAM)) |
0fc0f14b SH |
141 | return -EINVAL; |
142 | ||
143 | return 0; | |
144 | ||
7a22745a LP |
145 | default: |
146 | return -EAFNOSUPPORT; | |
42f4e3c4 LP |
147 | } |
148 | } | |
149 | ||
4d49b48c | 150 | int socket_address_print(const SocketAddress *a, char **ret) { |
42f4e3c4 | 151 | int r; |
4d49b48c | 152 | |
42f4e3c4 | 153 | assert(a); |
4d49b48c | 154 | assert(ret); |
42f4e3c4 | 155 | |
15dca371 ZJS |
156 | r = socket_address_verify(a, false); /* We do non-strict validation, because we want to be |
157 | * able to pretty-print any socket the kernel considers | |
158 | * valid. We still need to do validation to know if we | |
159 | * can meaningfully print the address. */ | |
4d49b48c | 160 | if (r < 0) |
42f4e3c4 LP |
161 | return r; |
162 | ||
4d49b48c | 163 | if (socket_address_family(a) == AF_NETLINK) { |
7fd1b19b | 164 | _cleanup_free_ char *sfamily = NULL; |
7a22745a | 165 | |
f8b69d1d | 166 | r = netlink_family_to_string_alloc(a->protocol, &sfamily); |
7a22745a | 167 | if (r < 0) |
f8b69d1d | 168 | return r; |
4d49b48c LP |
169 | |
170 | r = asprintf(ret, "%s %u", sfamily, a->sockaddr.nl.nl_groups); | |
8520cfa5 MS |
171 | if (r < 0) |
172 | return -ENOMEM; | |
7a22745a LP |
173 | |
174 | return 0; | |
175 | } | |
176 | ||
3b1c5241 | 177 | return sockaddr_pretty(&a->sockaddr.sa, a->size, false, true, ret); |
42f4e3c4 LP |
178 | } |
179 | ||
4f2d528d LP |
180 | bool socket_address_can_accept(const SocketAddress *a) { |
181 | assert(a); | |
182 | ||
183 | return | |
3742095b | 184 | IN_SET(a->type, SOCK_STREAM, SOCK_SEQPACKET); |
4f2d528d | 185 | } |
a16e1123 LP |
186 | |
187 | bool socket_address_equal(const SocketAddress *a, const SocketAddress *b) { | |
188 | assert(a); | |
189 | assert(b); | |
190 | ||
191 | /* Invalid addresses are unequal to all */ | |
15dca371 ZJS |
192 | if (socket_address_verify(a, false) < 0 || |
193 | socket_address_verify(b, false) < 0) | |
a16e1123 LP |
194 | return false; |
195 | ||
196 | if (a->type != b->type) | |
197 | return false; | |
198 | ||
a16e1123 LP |
199 | if (socket_address_family(a) != socket_address_family(b)) |
200 | return false; | |
201 | ||
202 | switch (socket_address_family(a)) { | |
203 | ||
204 | case AF_INET: | |
4d49b48c | 205 | if (a->sockaddr.in.sin_addr.s_addr != b->sockaddr.in.sin_addr.s_addr) |
a16e1123 LP |
206 | return false; |
207 | ||
4d49b48c | 208 | if (a->sockaddr.in.sin_port != b->sockaddr.in.sin_port) |
a16e1123 LP |
209 | return false; |
210 | ||
211 | break; | |
212 | ||
213 | case AF_INET6: | |
214 | if (memcmp(&a->sockaddr.in6.sin6_addr, &b->sockaddr.in6.sin6_addr, sizeof(a->sockaddr.in6.sin6_addr)) != 0) | |
215 | return false; | |
216 | ||
217 | if (a->sockaddr.in6.sin6_port != b->sockaddr.in6.sin6_port) | |
218 | return false; | |
219 | ||
220 | break; | |
221 | ||
222 | case AF_UNIX: | |
710708a5 MS |
223 | if (a->size <= offsetof(struct sockaddr_un, sun_path) || |
224 | b->size <= offsetof(struct sockaddr_un, sun_path)) | |
225 | return false; | |
226 | ||
a16e1123 LP |
227 | if ((a->sockaddr.un.sun_path[0] == 0) != (b->sockaddr.un.sun_path[0] == 0)) |
228 | return false; | |
229 | ||
230 | if (a->sockaddr.un.sun_path[0]) { | |
563e6846 | 231 | if (!path_equal_or_inode_same(a->sockaddr.un.sun_path, b->sockaddr.un.sun_path, 0)) |
a16e1123 LP |
232 | return false; |
233 | } else { | |
c78e47a6 MS |
234 | if (a->size != b->size) |
235 | return false; | |
236 | ||
b12c1e7c | 237 | if (memcmp(a->sockaddr.un.sun_path, b->sockaddr.un.sun_path, a->size) != 0) |
a16e1123 LP |
238 | return false; |
239 | } | |
240 | ||
241 | break; | |
242 | ||
7a22745a | 243 | case AF_NETLINK: |
7a22745a LP |
244 | if (a->protocol != b->protocol) |
245 | return false; | |
246 | ||
247 | if (a->sockaddr.nl.nl_groups != b->sockaddr.nl.nl_groups) | |
248 | return false; | |
249 | ||
250 | break; | |
251 | ||
0fc0f14b SH |
252 | case AF_VSOCK: |
253 | if (a->sockaddr.vm.svm_cid != b->sockaddr.vm.svm_cid) | |
254 | return false; | |
255 | ||
256 | if (a->sockaddr.vm.svm_port != b->sockaddr.vm.svm_port) | |
257 | return false; | |
258 | ||
259 | break; | |
260 | ||
a16e1123 LP |
261 | default: |
262 | /* Cannot compare, so we assume the addresses are different */ | |
263 | return false; | |
264 | } | |
265 | ||
266 | return true; | |
267 | } | |
268 | ||
a57f7e2c | 269 | const char* socket_address_get_path(const SocketAddress *a) { |
6e2ef85b LP |
270 | assert(a); |
271 | ||
272 | if (socket_address_family(a) != AF_UNIX) | |
a57f7e2c | 273 | return NULL; |
6e2ef85b LP |
274 | |
275 | if (a->sockaddr.un.sun_path[0] == 0) | |
a57f7e2c | 276 | return NULL; |
a16e1123 | 277 | |
48e6a2dc LP |
278 | /* Note that this is only safe because we know that there's an extra NUL byte after the sockaddr_un |
279 | * structure. On Linux AF_UNIX file system socket addresses don't have to be NUL terminated if they take up the | |
280 | * full sun_path space. */ | |
281 | assert_cc(sizeof(union sockaddr_union) >= sizeof(struct sockaddr_un)+1); | |
a57f7e2c | 282 | return a->sockaddr.un.sun_path; |
a16e1123 | 283 | } |
c0120d99 | 284 | |
5bfcc1c6 | 285 | bool socket_ipv6_is_supported(void) { |
571ec995 | 286 | static int cached = -1; |
f89f1e8f | 287 | |
571ec995 LP |
288 | if (cached < 0) { |
289 | ||
290 | if (access("/proc/net/if_inet6", F_OK) < 0) { | |
291 | ||
292 | if (errno != ENOENT) { | |
293 | log_debug_errno(errno, "Unexpected error when checking whether /proc/net/if_inet6 exists: %m"); | |
294 | return false; | |
295 | } | |
296 | ||
297 | cached = false; | |
298 | } else | |
299 | cached = true; | |
300 | } | |
301 | ||
302 | return cached; | |
5bfcc1c6 FF |
303 | } |
304 | ||
83e03c4f | 305 | bool socket_ipv6_is_enabled(void) { |
f96f5d54 | 306 | _cleanup_free_ char *v = NULL; |
83e03c4f LP |
307 | int r; |
308 | ||
309 | /* Much like socket_ipv6_is_supported(), but also checks that the sysctl that disables IPv6 on all | |
310 | * interfaces isn't turned on */ | |
311 | ||
312 | if (!socket_ipv6_is_supported()) | |
90ab5042 | 313 | return false; |
f89f1e8f | 314 | |
83e03c4f LP |
315 | r = sysctl_read_ip_property(AF_INET6, "all", "disable_ipv6", &v); |
316 | if (r < 0) { | |
317 | log_debug_errno(r, "Unexpected error reading 'net.ipv6.conf.all.disable_ipv6' sysctl: %m"); | |
318 | return true; | |
319 | } | |
320 | ||
321 | r = parse_boolean(v); | |
322 | if (r < 0) { | |
323 | log_debug_errno(r, "Failed to pare 'net.ipv6.conf.all.disable_ipv6' sysctl: %m"); | |
324 | return true; | |
325 | } | |
326 | ||
327 | return !r; | |
5bfcc1c6 FF |
328 | } |
329 | ||
01e10de3 | 330 | bool socket_address_matches_fd(const SocketAddress *a, int fd) { |
dbafedac MS |
331 | SocketAddress b; |
332 | socklen_t solen; | |
01e10de3 LP |
333 | |
334 | assert(a); | |
335 | assert(fd >= 0); | |
336 | ||
dbafedac MS |
337 | b.size = sizeof(b.sockaddr); |
338 | if (getsockname(fd, &b.sockaddr.sa, &b.size) < 0) | |
01e10de3 LP |
339 | return false; |
340 | ||
dbafedac | 341 | if (b.sockaddr.sa.sa_family != a->sockaddr.sa.sa_family) |
01e10de3 LP |
342 | return false; |
343 | ||
dbafedac MS |
344 | solen = sizeof(b.type); |
345 | if (getsockopt(fd, SOL_SOCKET, SO_TYPE, &b.type, &solen) < 0) | |
01e10de3 LP |
346 | return false; |
347 | ||
dbafedac | 348 | if (b.type != a->type) |
01e10de3 LP |
349 | return false; |
350 | ||
351 | if (a->protocol != 0) { | |
dbafedac MS |
352 | solen = sizeof(b.protocol); |
353 | if (getsockopt(fd, SOL_SOCKET, SO_PROTOCOL, &b.protocol, &solen) < 0) | |
01e10de3 LP |
354 | return false; |
355 | ||
dbafedac | 356 | if (b.protocol != a->protocol) |
01e10de3 LP |
357 | return false; |
358 | } | |
359 | ||
02233928 | 360 | return socket_address_equal(a, &b); |
01e10de3 LP |
361 | } |
362 | ||
f6aac5bf | 363 | int sockaddr_port(const struct sockaddr *_sa, unsigned *ret_port) { |
31325971 | 364 | const union sockaddr_union *sa = (const union sockaddr_union*) _sa; |
3b1c5241 | 365 | |
f6aac5bf LP |
366 | /* Note, this returns the port as 'unsigned' rather than 'uint16_t', as AF_VSOCK knows larger ports */ |
367 | ||
3b1c5241 SL |
368 | assert(sa); |
369 | ||
0fc0f14b | 370 | switch (sa->sa.sa_family) { |
f6aac5bf | 371 | |
0fc0f14b | 372 | case AF_INET: |
f6aac5bf | 373 | *ret_port = be16toh(sa->in.sin_port); |
0fc0f14b | 374 | return 0; |
3b1c5241 | 375 | |
0fc0f14b | 376 | case AF_INET6: |
f6aac5bf | 377 | *ret_port = be16toh(sa->in6.sin6_port); |
0fc0f14b SH |
378 | return 0; |
379 | ||
380 | case AF_VSOCK: | |
f6aac5bf | 381 | *ret_port = sa->vm.svm_port; |
0fc0f14b SH |
382 | return 0; |
383 | ||
384 | default: | |
385 | return -EAFNOSUPPORT; | |
386 | } | |
3b1c5241 SL |
387 | } |
388 | ||
31325971 LP |
389 | const union in_addr_union *sockaddr_in_addr(const struct sockaddr *_sa) { |
390 | const union sockaddr_union *sa = (const union sockaddr_union*) _sa; | |
391 | ||
392 | if (!sa) | |
393 | return NULL; | |
394 | ||
395 | switch (sa->sa.sa_family) { | |
396 | ||
397 | case AF_INET: | |
398 | return (const union in_addr_union*) &sa->in.sin_addr; | |
399 | ||
400 | case AF_INET6: | |
401 | return (const union in_addr_union*) &sa->in6.sin6_addr; | |
402 | ||
403 | default: | |
404 | return NULL; | |
405 | } | |
406 | } | |
407 | ||
c1b91f06 LP |
408 | int sockaddr_set_in_addr( |
409 | union sockaddr_union *u, | |
410 | int family, | |
411 | const union in_addr_union *a, | |
412 | uint16_t port) { | |
413 | ||
414 | assert(u); | |
415 | assert(a); | |
416 | ||
417 | switch (family) { | |
418 | ||
419 | case AF_INET: | |
420 | u->in = (struct sockaddr_in) { | |
421 | .sin_family = AF_INET, | |
422 | .sin_addr = a->in, | |
423 | .sin_port = htobe16(port), | |
424 | }; | |
425 | ||
426 | return 0; | |
427 | ||
428 | case AF_INET6: | |
429 | u->in6 = (struct sockaddr_in6) { | |
430 | .sin6_family = AF_INET6, | |
431 | .sin6_addr = a->in6, | |
432 | .sin6_port = htobe16(port), | |
433 | }; | |
434 | ||
435 | return 0; | |
436 | ||
437 | default: | |
438 | return -EAFNOSUPPORT; | |
439 | ||
440 | } | |
441 | } | |
442 | ||
836f9cfe LP |
443 | int sockaddr_pretty( |
444 | const struct sockaddr *_sa, | |
445 | socklen_t salen, | |
446 | bool translate_ipv6, | |
447 | bool include_port, | |
448 | char **ret) { | |
449 | ||
4d49b48c | 450 | union sockaddr_union *sa = (union sockaddr_union*) _sa; |
8569a776 | 451 | char *p; |
fc25ad25 | 452 | int r; |
8569a776 | 453 | |
4d49b48c LP |
454 | assert(sa); |
455 | assert(salen >= sizeof(sa->sa.sa_family)); | |
4eb3ec63 | 456 | assert(ret); |
8569a776 | 457 | |
4d49b48c | 458 | switch (sa->sa.sa_family) { |
8569a776 LP |
459 | |
460 | case AF_INET: { | |
461 | uint32_t a; | |
462 | ||
8e38570e | 463 | a = be32toh(sa->in.sin_addr.s_addr); |
8569a776 | 464 | |
fc25ad25 ZJS |
465 | if (include_port) |
466 | r = asprintf(&p, | |
3b1c5241 SL |
467 | "%u.%u.%u.%u:%u", |
468 | a >> 24, (a >> 16) & 0xFF, (a >> 8) & 0xFF, a & 0xFF, | |
8e38570e | 469 | be16toh(sa->in.sin_port)); |
fc25ad25 ZJS |
470 | else |
471 | r = asprintf(&p, | |
3b1c5241 | 472 | "%u.%u.%u.%u", |
fc25ad25 ZJS |
473 | a >> 24, (a >> 16) & 0xFF, (a >> 8) & 0xFF, a & 0xFF); |
474 | if (r < 0) | |
475 | return -ENOMEM; | |
8569a776 LP |
476 | break; |
477 | } | |
478 | ||
479 | case AF_INET6: { | |
480 | static const unsigned char ipv4_prefix[] = { | |
481 | 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0xFF, 0xFF | |
482 | }; | |
483 | ||
fc25ad25 ZJS |
484 | if (translate_ipv6 && |
485 | memcmp(&sa->in6.sin6_addr, ipv4_prefix, sizeof(ipv4_prefix)) == 0) { | |
4d49b48c | 486 | const uint8_t *a = sa->in6.sin6_addr.s6_addr+12; |
fc25ad25 ZJS |
487 | if (include_port) |
488 | r = asprintf(&p, | |
3b1c5241 SL |
489 | "%u.%u.%u.%u:%u", |
490 | a[0], a[1], a[2], a[3], | |
8e38570e | 491 | be16toh(sa->in6.sin6_port)); |
fc25ad25 ZJS |
492 | else |
493 | r = asprintf(&p, | |
3b1c5241 | 494 | "%u.%u.%u.%u", |
fc25ad25 ZJS |
495 | a[0], a[1], a[2], a[3]); |
496 | if (r < 0) | |
497 | return -ENOMEM; | |
8569a776 | 498 | } else { |
071e522e | 499 | const char *a = IN6_ADDR_TO_STRING(&sa->in6.sin6_addr); |
3b1c5241 SL |
500 | |
501 | if (include_port) { | |
01afd0f7 | 502 | if (asprintf(&p, |
b16d17a6 | 503 | "[%s]:%u%s%s", |
3b1c5241 | 504 | a, |
b16d17a6 ZJS |
505 | be16toh(sa->in6.sin6_port), |
506 | sa->in6.sin6_scope_id != 0 ? "%" : "", | |
01afd0f7 | 507 | FORMAT_IFNAME_FULL(sa->in6.sin6_scope_id, FORMAT_IFNAME_IFINDEX)) < 0) |
3b1c5241 SL |
508 | return -ENOMEM; |
509 | } else { | |
01afd0f7 YW |
510 | if (sa->in6.sin6_scope_id != 0) |
511 | p = strjoin(a, "%", FORMAT_IFNAME_FULL(sa->in6.sin6_scope_id, FORMAT_IFNAME_IFINDEX)); | |
512 | else | |
513 | p = strdup(a); | |
3b1c5241 SL |
514 | if (!p) |
515 | return -ENOMEM; | |
516 | } | |
8569a776 LP |
517 | } |
518 | ||
519 | break; | |
520 | } | |
521 | ||
4d49b48c | 522 | case AF_UNIX: |
15dca371 | 523 | if (salen <= offsetof(struct sockaddr_un, sun_path) || |
994b9d4e | 524 | (sa->un.sun_path[0] == 0 && salen == offsetof(struct sockaddr_un, sun_path) + 1)) |
15dca371 | 525 | /* The name must have at least one character (and the leading NUL does not count) */ |
4d49b48c | 526 | p = strdup("<unnamed>"); |
994b9d4e | 527 | else { |
085b39e9 LP |
528 | /* Note that we calculate the path pointer here through the .un_buffer[] field, in order to |
529 | * outtrick bounds checking tools such as ubsan, which are too smart for their own good: on | |
530 | * Linux the kernel may return sun_path[] data one byte longer than the declared size of the | |
531 | * field. */ | |
532 | char *path = (char*) sa->un_buffer + offsetof(struct sockaddr_un, sun_path); | |
15dca371 | 533 | size_t path_len = salen - offsetof(struct sockaddr_un, sun_path); |
8569a776 | 534 | |
085b39e9 | 535 | if (path[0] == 0) { |
15dca371 ZJS |
536 | /* Abstract socket. When parsing address information from, we |
537 | * explicitly reject overly long paths and paths with embedded NULs. | |
538 | * But we might get such a socket from the outside. Let's return | |
539 | * something meaningful and printable in this case. */ | |
4d49b48c | 540 | |
15dca371 | 541 | _cleanup_free_ char *e = NULL; |
4d49b48c | 542 | |
085b39e9 | 543 | e = cescape_length(path + 1, path_len - 1); |
15dca371 ZJS |
544 | if (!e) |
545 | return -ENOMEM; | |
4d49b48c | 546 | |
15dca371 ZJS |
547 | p = strjoin("@", e); |
548 | } else { | |
085b39e9 | 549 | if (path[path_len - 1] == '\0') |
15dca371 | 550 | /* We expect a terminating NUL and don't print it */ |
b3a9d980 | 551 | path_len--; |
15dca371 | 552 | |
085b39e9 | 553 | p = cescape_length(path, path_len); |
15dca371 | 554 | } |
4d49b48c | 555 | } |
994b9d4e LP |
556 | if (!p) |
557 | return -ENOMEM; | |
8569a776 LP |
558 | |
559 | break; | |
8569a776 | 560 | |
0fc0f14b | 561 | case AF_VSOCK: |
3a484991 ZJS |
562 | if (include_port) { |
563 | if (sa->vm.svm_cid == VMADDR_CID_ANY) | |
564 | r = asprintf(&p, "vsock::%u", sa->vm.svm_port); | |
565 | else | |
566 | r = asprintf(&p, "vsock:%u:%u", sa->vm.svm_cid, sa->vm.svm_port); | |
567 | } else | |
0fc0f14b SH |
568 | r = asprintf(&p, "vsock:%u", sa->vm.svm_cid); |
569 | if (r < 0) | |
570 | return -ENOMEM; | |
571 | break; | |
572 | ||
8569a776 | 573 | default: |
15411c0c | 574 | return -EOPNOTSUPP; |
8569a776 LP |
575 | } |
576 | ||
577 | *ret = p; | |
578 | return 0; | |
579 | } | |
580 | ||
366b7db4 | 581 | int getpeername_pretty(int fd, bool include_port, char **ret) { |
4d49b48c | 582 | union sockaddr_union sa; |
b31f535c | 583 | socklen_t salen = sizeof(sa); |
eff05270 | 584 | int r; |
4d49b48c LP |
585 | |
586 | assert(fd >= 0); | |
587 | assert(ret); | |
588 | ||
4d49b48c LP |
589 | if (getpeername(fd, &sa.sa, &salen) < 0) |
590 | return -errno; | |
591 | ||
592 | if (sa.sa.sa_family == AF_UNIX) { | |
a995ce47 | 593 | struct ucred ucred = UCRED_INVALID; |
4d49b48c LP |
594 | |
595 | /* UNIX connection sockets are anonymous, so let's use | |
596 | * PID/UID as pretty credentials instead */ | |
597 | ||
eff05270 LP |
598 | r = getpeercred(fd, &ucred); |
599 | if (r < 0) | |
600 | return r; | |
4d49b48c | 601 | |
de0671ee | 602 | if (asprintf(ret, "PID "PID_FMT"/UID "UID_FMT, ucred.pid, ucred.uid) < 0) |
4d49b48c LP |
603 | return -ENOMEM; |
604 | ||
605 | return 0; | |
606 | } | |
607 | ||
608 | /* For remote sockets we translate IPv6 addresses back to IPv4 | |
609 | * if applicable, since that's nicer. */ | |
610 | ||
366b7db4 | 611 | return sockaddr_pretty(&sa.sa, salen, true, include_port, ret); |
4d49b48c LP |
612 | } |
613 | ||
614 | int getsockname_pretty(int fd, char **ret) { | |
615 | union sockaddr_union sa; | |
b31f535c | 616 | socklen_t salen = sizeof(sa); |
4d49b48c LP |
617 | |
618 | assert(fd >= 0); | |
619 | assert(ret); | |
620 | ||
4d49b48c LP |
621 | if (getsockname(fd, &sa.sa, &salen) < 0) |
622 | return -errno; | |
623 | ||
624 | /* For local sockets we do not translate IPv6 addresses back | |
625 | * to IPv6 if applicable, since this is usually used for | |
626 | * listening sockets where the difference between IPv4 and | |
627 | * IPv6 matters. */ | |
628 | ||
3b1c5241 | 629 | return sockaddr_pretty(&sa.sa, salen, false, true, ret); |
4d49b48c LP |
630 | } |
631 | ||
fc1f05eb LP |
632 | int socknameinfo_pretty(const struct sockaddr *sa, socklen_t salen, char **ret) { |
633 | char host[NI_MAXHOST]; | |
b31f535c | 634 | int r; |
b31f535c | 635 | |
fc1f05eb | 636 | assert(sa); |
cde08933 | 637 | assert(salen >= sizeof(sa_family_t)); |
4eb3ec63 | 638 | assert(ret); |
b31f535c | 639 | |
fc1f05eb | 640 | r = getnameinfo(sa, salen, host, sizeof(host), /* service= */ NULL, /* service_len= */ 0, IDN_FLAGS); |
b31f535c | 641 | if (r != 0) { |
fc1f05eb LP |
642 | if (r == EAI_MEMORY) |
643 | return log_oom_debug(); | |
644 | if (r == EAI_SYSTEM) | |
645 | log_debug_errno(errno, "getnameinfo() failed, ignoring: %m"); | |
646 | else | |
647 | log_debug("getnameinfo() failed, ignoring: %s", gai_strerror(r)); | |
b31f535c | 648 | |
fc1f05eb LP |
649 | return sockaddr_pretty(sa, salen, /* translate_ipv6= */ true, /* include_port= */ true, ret); |
650 | } | |
b31f535c | 651 | |
4eb3ec63 | 652 | return strdup_to(ret, host); |
b31f535c ZJS |
653 | } |
654 | ||
7a22745a | 655 | static const char* const netlink_family_table[] = { |
0d7e34e3 ZJS |
656 | [NETLINK_ROUTE] = "route", |
657 | [NETLINK_FIREWALL] = "firewall", | |
658 | [NETLINK_INET_DIAG] = "inet-diag", | |
659 | [NETLINK_NFLOG] = "nflog", | |
660 | [NETLINK_XFRM] = "xfrm", | |
661 | [NETLINK_SELINUX] = "selinux", | |
662 | [NETLINK_ISCSI] = "iscsi", | |
663 | [NETLINK_AUDIT] = "audit", | |
664 | [NETLINK_FIB_LOOKUP] = "fib-lookup", | |
665 | [NETLINK_CONNECTOR] = "connector", | |
666 | [NETLINK_NETFILTER] = "netfilter", | |
667 | [NETLINK_IP6_FW] = "ip6-fw", | |
668 | [NETLINK_DNRTMSG] = "dnrtmsg", | |
7a22745a | 669 | [NETLINK_KOBJECT_UEVENT] = "kobject-uevent", |
0d7e34e3 ZJS |
670 | [NETLINK_GENERIC] = "generic", |
671 | [NETLINK_SCSITRANSPORT] = "scsitransport", | |
672 | [NETLINK_ECRYPTFS] = "ecryptfs", | |
673 | [NETLINK_RDMA] = "rdma", | |
7a22745a LP |
674 | }; |
675 | ||
f8b69d1d | 676 | DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(netlink_family, int, INT_MAX); |
7a22745a | 677 | |
c0120d99 LP |
678 | static const char* const socket_address_bind_ipv6_only_table[_SOCKET_ADDRESS_BIND_IPV6_ONLY_MAX] = { |
679 | [SOCKET_ADDRESS_DEFAULT] = "default", | |
680 | [SOCKET_ADDRESS_BOTH] = "both", | |
681 | [SOCKET_ADDRESS_IPV6_ONLY] = "ipv6-only" | |
682 | }; | |
683 | ||
684 | DEFINE_STRING_TABLE_LOOKUP(socket_address_bind_ipv6_only, SocketAddressBindIPv6Only); | |
f01e5736 | 685 | |
b54e98ef | 686 | SocketAddressBindIPv6Only socket_address_bind_ipv6_only_or_bool_from_string(const char *n) { |
6f90844f YW |
687 | int r; |
688 | ||
689 | r = parse_boolean(n); | |
690 | if (r > 0) | |
691 | return SOCKET_ADDRESS_IPV6_ONLY; | |
692 | if (r == 0) | |
693 | return SOCKET_ADDRESS_BOTH; | |
694 | ||
695 | return socket_address_bind_ipv6_only_from_string(n); | |
696 | } | |
697 | ||
f01e5736 LP |
698 | bool sockaddr_equal(const union sockaddr_union *a, const union sockaddr_union *b) { |
699 | assert(a); | |
700 | assert(b); | |
701 | ||
702 | if (a->sa.sa_family != b->sa.sa_family) | |
703 | return false; | |
704 | ||
705 | if (a->sa.sa_family == AF_INET) | |
706 | return a->in.sin_addr.s_addr == b->in.sin_addr.s_addr; | |
707 | ||
708 | if (a->sa.sa_family == AF_INET6) | |
709 | return memcmp(&a->in6.sin6_addr, &b->in6.sin6_addr, sizeof(a->in6.sin6_addr)) == 0; | |
710 | ||
0fc0f14b SH |
711 | if (a->sa.sa_family == AF_VSOCK) |
712 | return a->vm.svm_cid == b->vm.svm_cid; | |
713 | ||
f01e5736 LP |
714 | return false; |
715 | } | |
2583fbea | 716 | |
d9d9b2a0 | 717 | int fd_set_sndbuf(int fd, size_t n, bool increase) { |
2583fbea LP |
718 | int r, value; |
719 | socklen_t l = sizeof(value); | |
720 | ||
1263c85e YW |
721 | if (n > INT_MAX) |
722 | return -ERANGE; | |
723 | ||
2583fbea | 724 | r = getsockopt(fd, SOL_SOCKET, SO_SNDBUF, &value, &l); |
d9d9b2a0 | 725 | if (r >= 0 && l == sizeof(value) && increase ? (size_t) value >= n*2 : (size_t) value == n*2) |
2583fbea LP |
726 | return 0; |
727 | ||
b92f3507 YW |
728 | /* First, try to set the buffer size with SO_SNDBUF. */ |
729 | r = setsockopt_int(fd, SOL_SOCKET, SO_SNDBUF, n); | |
730 | if (r < 0) | |
731 | return r; | |
2583fbea | 732 | |
b92f3507 YW |
733 | /* SO_SNDBUF above may set to the kernel limit, instead of the requested size. |
734 | * So, we need to check the actual buffer size here. */ | |
67f5ae2d | 735 | l = sizeof(value); |
b92f3507 | 736 | r = getsockopt(fd, SOL_SOCKET, SO_SNDBUF, &value, &l); |
d9d9b2a0 | 737 | if (r >= 0 && l == sizeof(value) && increase ? (size_t) value >= n*2 : (size_t) value == n*2) |
b92f3507 YW |
738 | return 1; |
739 | ||
740 | /* If we have the privileges we will ignore the kernel limit. */ | |
741 | r = setsockopt_int(fd, SOL_SOCKET, SO_SNDBUFFORCE, n); | |
742 | if (r < 0) | |
743 | return r; | |
2583fbea LP |
744 | |
745 | return 1; | |
746 | } | |
747 | ||
d9d9b2a0 | 748 | int fd_set_rcvbuf(int fd, size_t n, bool increase) { |
2583fbea LP |
749 | int r, value; |
750 | socklen_t l = sizeof(value); | |
751 | ||
1263c85e YW |
752 | if (n > INT_MAX) |
753 | return -ERANGE; | |
754 | ||
2583fbea | 755 | r = getsockopt(fd, SOL_SOCKET, SO_RCVBUF, &value, &l); |
d9d9b2a0 | 756 | if (r >= 0 && l == sizeof(value) && increase ? (size_t) value >= n*2 : (size_t) value == n*2) |
2583fbea LP |
757 | return 0; |
758 | ||
b92f3507 YW |
759 | /* First, try to set the buffer size with SO_RCVBUF. */ |
760 | r = setsockopt_int(fd, SOL_SOCKET, SO_RCVBUF, n); | |
761 | if (r < 0) | |
762 | return r; | |
2583fbea | 763 | |
b92f3507 YW |
764 | /* SO_RCVBUF above may set to the kernel limit, instead of the requested size. |
765 | * So, we need to check the actual buffer size here. */ | |
67f5ae2d | 766 | l = sizeof(value); |
b92f3507 | 767 | r = getsockopt(fd, SOL_SOCKET, SO_RCVBUF, &value, &l); |
d9d9b2a0 | 768 | if (r >= 0 && l == sizeof(value) && increase ? (size_t) value >= n*2 : (size_t) value == n*2) |
b92f3507 YW |
769 | return 1; |
770 | ||
771 | /* If we have the privileges we will ignore the kernel limit. */ | |
772 | r = setsockopt_int(fd, SOL_SOCKET, SO_RCVBUFFORCE, n); | |
773 | if (r < 0) | |
774 | return r; | |
9e5b6496 | 775 | |
2583fbea LP |
776 | return 1; |
777 | } | |
778 | ||
779 | static const char* const ip_tos_table[] = { | |
0d7e34e3 ZJS |
780 | [IPTOS_LOWDELAY] = "low-delay", |
781 | [IPTOS_THROUGHPUT] = "throughput", | |
2583fbea | 782 | [IPTOS_RELIABILITY] = "reliability", |
0d7e34e3 | 783 | [IPTOS_LOWCOST] = "low-cost", |
2583fbea LP |
784 | }; |
785 | ||
786 | DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(ip_tos, int, 0xff); | |
787 | ||
5a3586db YW |
788 | bool ifname_valid_char(char a) { |
789 | if ((unsigned char) a >= 127U) | |
790 | return false; | |
791 | ||
792 | if ((unsigned char) a <= 32U) | |
793 | return false; | |
794 | ||
795 | if (IN_SET(a, | |
796 | ':', /* colons are used by the legacy "alias" interface logic */ | |
797 | '/', /* slashes cannot work, since we need to use network interfaces in sysfs paths, and in paths slashes are separators */ | |
798 | '%')) /* %d is used in the kernel's weird foo%d format string naming feature which we really really don't want to ever run into by accident */ | |
799 | return false; | |
800 | ||
801 | return true; | |
802 | } | |
803 | ||
2313524a | 804 | bool ifname_valid_full(const char *p, IfnameValidFlags flags) { |
ef76dff2 LP |
805 | bool numeric = true; |
806 | ||
807 | /* Checks whether a network interface name is valid. This is inspired by dev_valid_name() in the kernel sources | |
808 | * but slightly stricter, as we only allow non-control, non-space ASCII characters in the interface name. We | |
809 | * also don't permit names that only container numbers, to avoid confusion with numeric interface indexes. */ | |
810 | ||
2313524a ZJS |
811 | assert(!(flags & ~_IFNAME_VALID_ALL)); |
812 | ||
ef76dff2 LP |
813 | if (isempty(p)) |
814 | return false; | |
815 | ||
ee29bd5d LP |
816 | /* A valid ifindex? If so, it's valid iff IFNAME_VALID_NUMERIC is set */ |
817 | if (parse_ifindex(p) >= 0) | |
818 | return flags & IFNAME_VALID_NUMERIC; | |
819 | ||
2313524a | 820 | if (flags & IFNAME_VALID_ALTERNATIVE) { |
4252696a YW |
821 | if (strlen(p) >= ALTIFNAMSIZ) |
822 | return false; | |
823 | } else { | |
824 | if (strlen(p) >= IFNAMSIZ) | |
825 | return false; | |
826 | } | |
ef76dff2 | 827 | |
49bfc877 | 828 | if (dot_or_dot_dot(p)) |
ef76dff2 LP |
829 | return false; |
830 | ||
07a7441a LP |
831 | /* Let's refuse "all" and "default" as interface name, to avoid collisions with the special sysctl |
832 | * directories /proc/sys/net/{ipv4,ipv6}/conf/{all,default} */ | |
6aebfec3 | 833 | if (!FLAGS_SET(flags, IFNAME_VALID_SPECIAL) && STR_IN_SET(p, "all", "default")) |
07a7441a LP |
834 | return false; |
835 | ||
2313524a | 836 | for (const char *t = p; *t; t++) { |
5a3586db | 837 | if (!ifname_valid_char(*t)) |
ef76dff2 LP |
838 | return false; |
839 | ||
ff25d338 | 840 | numeric = numeric && ascii_isdigit(*t); |
ef76dff2 LP |
841 | } |
842 | ||
ee29bd5d LP |
843 | /* It's fully numeric but didn't parse as valid ifindex above? if so, it must be too large or zero or |
844 | * so, let's refuse that. */ | |
845 | if (numeric) | |
846 | return false; | |
ef76dff2 LP |
847 | |
848 | return true; | |
849 | } | |
850 | ||
26808948 SS |
851 | bool address_label_valid(const char *p) { |
852 | ||
853 | if (isempty(p)) | |
854 | return false; | |
855 | ||
856 | if (strlen(p) >= IFNAMSIZ) | |
857 | return false; | |
858 | ||
859 | while (*p) { | |
860 | if ((uint8_t) *p >= 127U) | |
861 | return false; | |
862 | ||
863 | if ((uint8_t) *p <= 31U) | |
864 | return false; | |
865 | p++; | |
866 | } | |
867 | ||
868 | return true; | |
869 | } | |
870 | ||
2583fbea LP |
871 | int getpeercred(int fd, struct ucred *ucred) { |
872 | socklen_t n = sizeof(struct ucred); | |
873 | struct ucred u; | |
2583fbea LP |
874 | |
875 | assert(fd >= 0); | |
876 | assert(ucred); | |
877 | ||
fccad706 | 878 | if (getsockopt(fd, SOL_SOCKET, SO_PEERCRED, &u, &n) < 0) |
2583fbea LP |
879 | return -errno; |
880 | ||
881 | if (n != sizeof(struct ucred)) | |
882 | return -EIO; | |
883 | ||
bbcc701e LP |
884 | /* Check if the data is actually useful and not suppressed due to namespacing issues */ |
885 | if (!pid_is_valid(u.pid)) | |
2583fbea LP |
886 | return -ENODATA; |
887 | ||
bbcc701e LP |
888 | /* Note that we don't check UID/GID here, as namespace translation works differently there: instead of |
889 | * receiving in "invalid" user/group we get the overflow UID/GID. */ | |
890 | ||
2583fbea LP |
891 | *ucred = u; |
892 | return 0; | |
893 | } | |
894 | ||
895 | int getpeersec(int fd, char **ret) { | |
217d8967 | 896 | _cleanup_free_ char *s = NULL; |
2583fbea | 897 | socklen_t n = 64; |
2583fbea LP |
898 | |
899 | assert(fd >= 0); | |
900 | assert(ret); | |
901 | ||
217d8967 LP |
902 | for (;;) { |
903 | s = new0(char, n+1); | |
904 | if (!s) | |
905 | return -ENOMEM; | |
2583fbea | 906 | |
989740eb LP |
907 | if (getsockopt(fd, SOL_SOCKET, SO_PEERSEC, s, &n) >= 0) { |
908 | s[n] = 0; | |
217d8967 | 909 | break; |
989740eb | 910 | } |
2583fbea LP |
911 | |
912 | if (errno != ERANGE) | |
913 | return -errno; | |
914 | ||
217d8967 | 915 | s = mfree(s); |
2583fbea LP |
916 | } |
917 | ||
217d8967 | 918 | if (isempty(s)) |
2583fbea | 919 | return -EOPNOTSUPP; |
2583fbea | 920 | |
ae2a15bc | 921 | *ret = TAKE_PTR(s); |
217d8967 | 922 | |
2583fbea LP |
923 | return 0; |
924 | } | |
925 | ||
43f2c88d | 926 | int getpeergroups(int fd, gid_t **ret) { |
9c41e4eb | 927 | socklen_t n = sizeof(gid_t) * 64U; |
43f2c88d LP |
928 | _cleanup_free_ gid_t *d = NULL; |
929 | ||
75f40779 | 930 | assert(fd >= 0); |
43f2c88d LP |
931 | assert(ret); |
932 | ||
7e8aa5c2 | 933 | long ngroups_max = sysconf(_SC_NGROUPS_MAX); |
9c41e4eb LB |
934 | if (ngroups_max > 0) |
935 | n = MAX(n, sizeof(gid_t) * (socklen_t) ngroups_max); | |
7e8aa5c2 | 936 | |
43f2c88d LP |
937 | for (;;) { |
938 | d = malloc(n); | |
939 | if (!d) | |
940 | return -ENOMEM; | |
941 | ||
942 | if (getsockopt(fd, SOL_SOCKET, SO_PEERGROUPS, d, &n) >= 0) | |
943 | break; | |
944 | ||
945 | if (errno != ERANGE) | |
946 | return -errno; | |
947 | ||
948 | d = mfree(d); | |
949 | } | |
950 | ||
951 | assert_se(n % sizeof(gid_t) == 0); | |
952 | n /= sizeof(gid_t); | |
953 | ||
7e8aa5c2 | 954 | if (n > INT_MAX) |
43f2c88d LP |
955 | return -E2BIG; |
956 | ||
1cc6c93a | 957 | *ret = TAKE_PTR(d); |
43f2c88d LP |
958 | |
959 | return (int) n; | |
960 | } | |
961 | ||
da5e0c44 LP |
962 | int getpeerpidfd(int fd) { |
963 | socklen_t n = sizeof(int); | |
964 | int pidfd = -EBADF; | |
965 | ||
966 | assert(fd >= 0); | |
967 | ||
968 | if (getsockopt(fd, SOL_SOCKET, SO_PEERPIDFD, &pidfd, &n) < 0) | |
969 | return -errno; | |
970 | ||
971 | if (n != sizeof(int)) | |
972 | return -EIO; | |
973 | ||
974 | return pidfd; | |
975 | } | |
976 | ||
598d2428 LB |
977 | ssize_t send_many_fds_iov_sa( |
978 | int transport_fd, | |
979 | int *fds_array, size_t n_fds_array, | |
980 | const struct iovec *iov, size_t iovlen, | |
981 | const struct sockaddr *sa, socklen_t len, | |
982 | int flags) { | |
983 | ||
984 | _cleanup_free_ struct cmsghdr *cmsg = NULL; | |
985 | struct msghdr mh = { | |
986 | .msg_name = (struct sockaddr*) sa, | |
987 | .msg_namelen = len, | |
988 | .msg_iov = (struct iovec *)iov, | |
989 | .msg_iovlen = iovlen, | |
990 | }; | |
991 | ssize_t k; | |
992 | ||
993 | assert(transport_fd >= 0); | |
994 | assert(fds_array || n_fds_array == 0); | |
995 | ||
996 | /* The kernel will reject sending more than SCM_MAX_FD FDs at once */ | |
997 | if (n_fds_array > SCM_MAX_FD) | |
998 | return -E2BIG; | |
999 | ||
1000 | /* We need either an FD array or data to send. If there's nothing, return an error. */ | |
1001 | if (n_fds_array == 0 && !iov) | |
1002 | return -EINVAL; | |
1003 | ||
1004 | if (n_fds_array > 0) { | |
1005 | mh.msg_controllen = CMSG_SPACE(sizeof(int) * n_fds_array); | |
1006 | mh.msg_control = cmsg = malloc(mh.msg_controllen); | |
1007 | if (!cmsg) | |
1008 | return -ENOMEM; | |
1009 | ||
1010 | *cmsg = (struct cmsghdr) { | |
1011 | .cmsg_len = CMSG_LEN(sizeof(int) * n_fds_array), | |
1012 | .cmsg_level = SOL_SOCKET, | |
1013 | .cmsg_type = SCM_RIGHTS, | |
1014 | }; | |
1015 | memcpy(CMSG_DATA(cmsg), fds_array, sizeof(int) * n_fds_array); | |
1016 | } | |
1017 | k = sendmsg(transport_fd, &mh, MSG_NOSIGNAL | flags); | |
1018 | if (k < 0) | |
1019 | return (ssize_t) -errno; | |
1020 | ||
1021 | return k; | |
1022 | } | |
1023 | ||
d34673ec | 1024 | ssize_t send_one_fd_iov_sa( |
726f4c47 ZJS |
1025 | int transport_fd, |
1026 | int fd, | |
f621b8d7 | 1027 | const struct iovec *iov, size_t iovlen, |
726f4c47 ZJS |
1028 | const struct sockaddr *sa, socklen_t len, |
1029 | int flags) { | |
1030 | ||
fb29cdbe | 1031 | CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(int))) control = {}; |
2583fbea | 1032 | struct msghdr mh = { |
726f4c47 ZJS |
1033 | .msg_name = (struct sockaddr*) sa, |
1034 | .msg_namelen = len, | |
f621b8d7 | 1035 | .msg_iov = (struct iovec *)iov, |
d34673ec | 1036 | .msg_iovlen = iovlen, |
2583fbea | 1037 | }; |
d34673ec | 1038 | ssize_t k; |
2583fbea LP |
1039 | |
1040 | assert(transport_fd >= 0); | |
2583fbea | 1041 | |
d34673ec FB |
1042 | /* |
1043 | * We need either an FD or data to send. | |
1044 | * If there's nothing, return an error. | |
1045 | */ | |
1046 | if (fd < 0 && !iov) | |
1047 | return -EINVAL; | |
2583fbea | 1048 | |
d34673ec FB |
1049 | if (fd >= 0) { |
1050 | struct cmsghdr *cmsg; | |
2583fbea | 1051 | |
d34673ec FB |
1052 | mh.msg_control = &control; |
1053 | mh.msg_controllen = sizeof(control); | |
1054 | ||
1055 | cmsg = CMSG_FIRSTHDR(&mh); | |
1056 | cmsg->cmsg_level = SOL_SOCKET; | |
1057 | cmsg->cmsg_type = SCM_RIGHTS; | |
1058 | cmsg->cmsg_len = CMSG_LEN(sizeof(int)); | |
1059 | memcpy(CMSG_DATA(cmsg), &fd, sizeof(int)); | |
d34673ec FB |
1060 | } |
1061 | k = sendmsg(transport_fd, &mh, MSG_NOSIGNAL | flags); | |
1062 | if (k < 0) | |
1063 | return (ssize_t) -errno; | |
1064 | ||
1065 | return k; | |
2583fbea LP |
1066 | } |
1067 | ||
d34673ec FB |
1068 | int send_one_fd_sa( |
1069 | int transport_fd, | |
1070 | int fd, | |
1071 | const struct sockaddr *sa, socklen_t len, | |
1072 | int flags) { | |
1073 | ||
1074 | assert(fd >= 0); | |
1075 | ||
1076 | return (int) send_one_fd_iov_sa(transport_fd, fd, NULL, 0, sa, len, flags); | |
1077 | } | |
1078 | ||
598d2428 LB |
1079 | ssize_t receive_many_fds_iov( |
1080 | int transport_fd, | |
1081 | struct iovec *iov, size_t iovlen, | |
1082 | int **ret_fds_array, size_t *ret_n_fds_array, | |
1083 | int flags) { | |
1084 | ||
1085 | CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(int) * SCM_MAX_FD)) control; | |
1086 | struct msghdr mh = { | |
1087 | .msg_control = &control, | |
1088 | .msg_controllen = sizeof(control), | |
1089 | .msg_iov = iov, | |
1090 | .msg_iovlen = iovlen, | |
1091 | }; | |
1092 | _cleanup_free_ int *fds_array = NULL; | |
1093 | size_t n_fds_array = 0; | |
1094 | struct cmsghdr *cmsg; | |
1095 | ssize_t k; | |
1096 | ||
1097 | assert(transport_fd >= 0); | |
1098 | assert(ret_fds_array); | |
1099 | assert(ret_n_fds_array); | |
1100 | ||
1101 | /* | |
1102 | * Receive many FDs via @transport_fd. We don't care for the transport-type. We retrieve all the FDs | |
1103 | * at once. This is best used in combination with send_many_fds(). | |
1104 | */ | |
1105 | ||
1106 | k = recvmsg_safe(transport_fd, &mh, MSG_CMSG_CLOEXEC | flags); | |
1107 | if (k < 0) | |
1108 | return k; | |
1109 | ||
1110 | CMSG_FOREACH(cmsg, &mh) | |
1111 | if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) { | |
1112 | size_t n = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int); | |
1113 | ||
1114 | fds_array = GREEDY_REALLOC(fds_array, n_fds_array + n); | |
1115 | if (!fds_array) { | |
1116 | cmsg_close_all(&mh); | |
1117 | return -ENOMEM; | |
1118 | } | |
1119 | ||
1120 | memcpy(fds_array + n_fds_array, CMSG_TYPED_DATA(cmsg, int), sizeof(int) * n); | |
1121 | n_fds_array += n; | |
1122 | } | |
1123 | ||
1124 | if (n_fds_array == 0) { | |
1125 | cmsg_close_all(&mh); | |
1126 | ||
1127 | /* If didn't receive an FD or any data, return an error. */ | |
1128 | if (k == 0) | |
1129 | return -EIO; | |
1130 | } | |
1131 | ||
1132 | *ret_fds_array = TAKE_PTR(fds_array); | |
1133 | *ret_n_fds_array = n_fds_array; | |
1134 | ||
1135 | return k; | |
1136 | } | |
1137 | ||
1138 | int receive_many_fds(int transport_fd, int **ret_fds_array, size_t *ret_n_fds_array, int flags) { | |
1139 | ssize_t k; | |
1140 | ||
1141 | k = receive_many_fds_iov(transport_fd, NULL, 0, ret_fds_array, ret_n_fds_array, flags); | |
1142 | if (k == 0) | |
1143 | return 0; | |
1144 | ||
1145 | /* k must be negative, since receive_many_fds_iov() only returns a positive value if data was received | |
1146 | * through the iov. */ | |
1147 | assert(k < 0); | |
1148 | return (int) k; | |
1149 | } | |
1150 | ||
d34673ec FB |
1151 | ssize_t receive_one_fd_iov( |
1152 | int transport_fd, | |
1153 | struct iovec *iov, size_t iovlen, | |
1154 | int flags, | |
1155 | int *ret_fd) { | |
1156 | ||
fb29cdbe | 1157 | CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(int))) control; |
2583fbea LP |
1158 | struct msghdr mh = { |
1159 | .msg_control = &control, | |
1160 | .msg_controllen = sizeof(control), | |
d34673ec FB |
1161 | .msg_iov = iov, |
1162 | .msg_iovlen = iovlen, | |
2583fbea | 1163 | }; |
dac556fa | 1164 | struct cmsghdr *found; |
d34673ec | 1165 | ssize_t k; |
2583fbea LP |
1166 | |
1167 | assert(transport_fd >= 0); | |
d34673ec | 1168 | assert(ret_fd); |
2583fbea LP |
1169 | |
1170 | /* | |
1171 | * Receive a single FD via @transport_fd. We don't care for | |
1172 | * the transport-type. We retrieve a single FD at most, so for | |
1173 | * packet-based transports, the caller must ensure to send | |
1174 | * only a single FD per packet. This is best used in | |
1175 | * combination with send_one_fd(). | |
1176 | */ | |
1177 | ||
3691bcf3 | 1178 | k = recvmsg_safe(transport_fd, &mh, MSG_CMSG_CLOEXEC | flags); |
d34673ec | 1179 | if (k < 0) |
3691bcf3 | 1180 | return k; |
2583fbea | 1181 | |
dac556fa | 1182 | found = cmsg_find(&mh, SOL_SOCKET, SCM_RIGHTS, CMSG_LEN(sizeof(int))); |
3691bcf3 | 1183 | if (!found) { |
2583fbea | 1184 | cmsg_close_all(&mh); |
d34673ec | 1185 | |
3691bcf3 LP |
1186 | /* If didn't receive an FD or any data, return an error. */ |
1187 | if (k == 0) | |
1188 | return -EIO; | |
1189 | } | |
2583fbea | 1190 | |
d34673ec | 1191 | if (found) |
b1d02191 | 1192 | *ret_fd = *CMSG_TYPED_DATA(found, int); |
d34673ec | 1193 | else |
254d1313 | 1194 | *ret_fd = -EBADF; |
d34673ec FB |
1195 | |
1196 | return k; | |
1197 | } | |
1198 | ||
1199 | int receive_one_fd(int transport_fd, int flags) { | |
1200 | int fd; | |
1201 | ssize_t k; | |
1202 | ||
1203 | k = receive_one_fd_iov(transport_fd, NULL, 0, flags, &fd); | |
1204 | if (k == 0) | |
1205 | return fd; | |
1206 | ||
1207 | /* k must be negative, since receive_one_fd_iov() only returns | |
1208 | * a positive value if data was received through the iov. */ | |
1209 | assert(k < 0); | |
1210 | return (int) k; | |
2583fbea | 1211 | } |
4edc2c9b LP |
1212 | |
1213 | ssize_t next_datagram_size_fd(int fd) { | |
1214 | ssize_t l; | |
1215 | int k; | |
1216 | ||
1217 | /* This is a bit like FIONREAD/SIOCINQ, however a bit more powerful. The difference being: recv(MSG_PEEK) will | |
96d49011 | 1218 | * actually cause the next datagram in the queue to be validated regarding checksums, which FIONREAD doesn't |
4edc2c9b LP |
1219 | * do. This difference is actually of major importance as we need to be sure that the size returned here |
1220 | * actually matches what we will read with recvmsg() next, as otherwise we might end up allocating a buffer of | |
1221 | * the wrong size. */ | |
1222 | ||
1223 | l = recv(fd, NULL, 0, MSG_PEEK|MSG_TRUNC); | |
1224 | if (l < 0) { | |
3742095b | 1225 | if (IN_SET(errno, EOPNOTSUPP, EFAULT)) |
4edc2c9b LP |
1226 | goto fallback; |
1227 | ||
1228 | return -errno; | |
1229 | } | |
1230 | if (l == 0) | |
1231 | goto fallback; | |
1232 | ||
1233 | return l; | |
1234 | ||
1235 | fallback: | |
1236 | k = 0; | |
1237 | ||
1238 | /* Some sockets (AF_PACKET) do not support null-sized recv() with MSG_TRUNC set, let's fall back to FIONREAD | |
1239 | * for them. Checksums don't matter for raw sockets anyway, hence this should be fine. */ | |
1240 | ||
1241 | if (ioctl(fd, FIONREAD, &k) < 0) | |
1242 | return -errno; | |
1243 | ||
1244 | return (ssize_t) k; | |
1245 | } | |
60d9771c | 1246 | |
67962036 ZJS |
1247 | /* Put a limit on how many times will attempt to call accept4(). We loop |
1248 | * only on "transient" errors, but let's make sure we don't loop forever. */ | |
1249 | #define MAX_FLUSH_ITERATIONS 1024 | |
1250 | ||
60d9771c LP |
1251 | int flush_accept(int fd) { |
1252 | ||
f3d75364 LP |
1253 | int r, b; |
1254 | socklen_t l = sizeof(b); | |
1255 | ||
644cb4cc ZJS |
1256 | /* Similar to flush_fd() but flushes all incoming connections by accepting and immediately closing |
1257 | * them. */ | |
f3d75364 LP |
1258 | |
1259 | if (getsockopt(fd, SOL_SOCKET, SO_ACCEPTCONN, &b, &l) < 0) | |
1260 | return -errno; | |
60d9771c | 1261 | |
f3d75364 | 1262 | assert(l == sizeof(b)); |
644cb4cc ZJS |
1263 | if (!b) /* Let's check if this socket accepts connections before calling accept(). accept4() can |
1264 | * return EOPNOTSUPP if the fd is not a listening socket, which we should treat as a fatal | |
1265 | * error, or in case the incoming TCP connection triggered a network issue, which we want to | |
1266 | * treat as a transient error. Thus, let's rule out the first reason for EOPNOTSUPP early, so | |
1267 | * we can loop safely on transient errors below. */ | |
f3d75364 | 1268 | return -ENOTTY; |
60d9771c | 1269 | |
67962036 | 1270 | for (unsigned iteration = 0;; iteration++) { |
60d9771c LP |
1271 | int cfd; |
1272 | ||
0f2d351f | 1273 | r = fd_wait_for_event(fd, POLLIN, 0); |
60d9771c | 1274 | if (r < 0) { |
0f2d351f | 1275 | if (r == -EINTR) |
60d9771c LP |
1276 | continue; |
1277 | ||
0f2d351f | 1278 | return r; |
4ff9bc2e LP |
1279 | } |
1280 | if (r == 0) | |
60d9771c LP |
1281 | return 0; |
1282 | ||
67962036 ZJS |
1283 | if (iteration >= MAX_FLUSH_ITERATIONS) |
1284 | return log_debug_errno(SYNTHETIC_ERRNO(EBUSY), | |
1285 | "Failed to flush connections within " STRINGIFY(MAX_FLUSH_ITERATIONS) " iterations."); | |
1286 | ||
60d9771c LP |
1287 | cfd = accept4(fd, NULL, NULL, SOCK_NONBLOCK|SOCK_CLOEXEC); |
1288 | if (cfd < 0) { | |
60d9771c LP |
1289 | if (errno == EAGAIN) |
1290 | return 0; | |
1291 | ||
4ff9bc2e LP |
1292 | if (ERRNO_IS_ACCEPT_AGAIN(errno)) |
1293 | continue; | |
1294 | ||
60d9771c LP |
1295 | return -errno; |
1296 | } | |
1297 | ||
4ff9bc2e | 1298 | safe_close(cfd); |
60d9771c LP |
1299 | } |
1300 | } | |
29206d46 LP |
1301 | |
1302 | struct cmsghdr* cmsg_find(struct msghdr *mh, int level, int type, socklen_t length) { | |
1303 | struct cmsghdr *cmsg; | |
1304 | ||
1305 | assert(mh); | |
1306 | ||
1307 | CMSG_FOREACH(cmsg, mh) | |
1308 | if (cmsg->cmsg_level == level && | |
1309 | cmsg->cmsg_type == type && | |
1310 | (length == (socklen_t) -1 || length == cmsg->cmsg_len)) | |
1311 | return cmsg; | |
1312 | ||
1313 | return NULL; | |
1314 | } | |
429b4350 | 1315 | |
4836f4c6 YW |
1316 | void* cmsg_find_and_copy_data(struct msghdr *mh, int level, int type, void *buf, size_t buf_len) { |
1317 | struct cmsghdr *cmsg; | |
1318 | ||
1319 | assert(mh); | |
1320 | assert(buf); | |
1321 | assert(buf_len > 0); | |
1322 | ||
1323 | /* This is similar to cmsg_find_data(), but copy the found data to buf. This should be typically used | |
da890466 | 1324 | * when reading possibly unaligned data such as timestamp, as time_t is 64-bit and size_t is 32-bit on |
4836f4c6 YW |
1325 | * RISCV32. See issue #27241. */ |
1326 | ||
1327 | cmsg = cmsg_find(mh, level, type, CMSG_LEN(buf_len)); | |
1328 | if (!cmsg) | |
1329 | return NULL; | |
1330 | ||
1331 | return memcpy_safe(buf, CMSG_DATA(cmsg), buf_len); | |
1332 | } | |
1333 | ||
429b4350 LP |
1334 | int socket_ioctl_fd(void) { |
1335 | int fd; | |
1336 | ||
1337 | /* Create a socket to invoke the various network interface ioctl()s on. Traditionally only AF_INET was good for | |
1338 | * that. Since kernel 4.6 AF_NETLINK works for this too. We first try to use AF_INET hence, but if that's not | |
1339 | * available (for example, because it is made unavailable via SECCOMP or such), we'll fall back to the more | |
1340 | * generic AF_NETLINK. */ | |
1341 | ||
1342 | fd = socket(AF_INET, SOCK_DGRAM|SOCK_CLOEXEC, 0); | |
1343 | if (fd < 0) | |
1344 | fd = socket(AF_NETLINK, SOCK_RAW|SOCK_CLOEXEC, NETLINK_GENERIC); | |
1345 | if (fd < 0) | |
1346 | return -errno; | |
1347 | ||
1348 | return fd; | |
1349 | } | |
9f20fc28 LP |
1350 | |
1351 | int sockaddr_un_unlink(const struct sockaddr_un *sa) { | |
1352 | const char *p, * nul; | |
1353 | ||
1354 | assert(sa); | |
1355 | ||
1356 | if (sa->sun_family != AF_UNIX) | |
1357 | return -EPROTOTYPE; | |
1358 | ||
1359 | if (sa->sun_path[0] == 0) /* Nothing to do for abstract sockets */ | |
1360 | return 0; | |
1361 | ||
1362 | /* The path in .sun_path is not necessarily NUL terminated. Let's fix that. */ | |
1363 | nul = memchr(sa->sun_path, 0, sizeof(sa->sun_path)); | |
1364 | if (nul) | |
1365 | p = sa->sun_path; | |
1366 | else | |
1367 | p = memdupa_suffix0(sa->sun_path, sizeof(sa->sun_path)); | |
1368 | ||
1369 | if (unlink(p) < 0) | |
1370 | return -errno; | |
1371 | ||
1372 | return 1; | |
1373 | } | |
5cf91ea9 LP |
1374 | |
1375 | int sockaddr_un_set_path(struct sockaddr_un *ret, const char *path) { | |
1376 | size_t l; | |
1377 | ||
1378 | assert(ret); | |
1379 | assert(path); | |
1380 | ||
1381 | /* Initialize ret->sun_path from the specified argument. This will interpret paths starting with '@' as | |
1382 | * abstract namespace sockets, and those starting with '/' as regular filesystem sockets. It won't accept | |
1383 | * anything else (i.e. no relative paths), to avoid ambiguities. Note that this function cannot be used to | |
1384 | * reference paths in the abstract namespace that include NUL bytes in the name. */ | |
1385 | ||
1386 | l = strlen(path); | |
c097bf1f | 1387 | if (l < 2) |
5cf91ea9 LP |
1388 | return -EINVAL; |
1389 | if (!IN_SET(path[0], '/', '@')) | |
1390 | return -EINVAL; | |
5cf91ea9 LP |
1391 | |
1392 | /* Don't allow paths larger than the space in sockaddr_un. Note that we are a tiny bit more restrictive than | |
1393 | * the kernel is: we insist on NUL termination (both for abstract namespace and regular file system socket | |
1394 | * addresses!), which the kernel doesn't. We do this to reduce chance of incompatibility with other apps that | |
7802194a | 1395 | * do not expect non-NUL terminated file system path. */ |
5cf91ea9 | 1396 | if (l+1 > sizeof(ret->sun_path)) |
dfa2b389 LP |
1397 | return path[0] == '@' ? -EINVAL : -ENAMETOOLONG; /* return a recognizable error if this is |
1398 | * too long to fit into a sockaddr_un, but | |
1399 | * is a file system path, and thus might be | |
1400 | * connectible via O_PATH indirection. */ | |
5cf91ea9 LP |
1401 | |
1402 | *ret = (struct sockaddr_un) { | |
1403 | .sun_family = AF_UNIX, | |
1404 | }; | |
1405 | ||
1406 | if (path[0] == '@') { | |
1407 | /* Abstract namespace socket */ | |
1408 | memcpy(ret->sun_path + 1, path + 1, l); /* copy *with* trailing NUL byte */ | |
1409 | return (int) (offsetof(struct sockaddr_un, sun_path) + l); /* 🔥 *don't* 🔥 include trailing NUL in size */ | |
1410 | ||
1411 | } else { | |
1412 | assert(path[0] == '/'); | |
1413 | ||
1414 | /* File system socket */ | |
1415 | memcpy(ret->sun_path, path, l + 1); /* copy *with* trailing NUL byte */ | |
1416 | return (int) (offsetof(struct sockaddr_un, sun_path) + l + 1); /* include trailing NUL in size */ | |
1417 | } | |
1418 | } | |
5d594d01 LP |
1419 | |
1420 | int socket_bind_to_ifname(int fd, const char *ifname) { | |
1421 | assert(fd >= 0); | |
1422 | ||
1423 | /* Call with NULL to drop binding */ | |
1424 | ||
7c248223 | 1425 | return RET_NERRNO(setsockopt(fd, SOL_SOCKET, SO_BINDTODEVICE, ifname, strlen_ptr(ifname))); |
5d594d01 LP |
1426 | } |
1427 | ||
1428 | int socket_bind_to_ifindex(int fd, int ifindex) { | |
01afd0f7 | 1429 | char ifname[IF_NAMESIZE]; |
5e958e1d | 1430 | int r; |
5d594d01 LP |
1431 | |
1432 | assert(fd >= 0); | |
1433 | ||
7c248223 | 1434 | if (ifindex <= 0) |
5d594d01 | 1435 | /* Drop binding */ |
7c248223 | 1436 | return RET_NERRNO(setsockopt(fd, SOL_SOCKET, SO_BINDTODEVICE, NULL, 0)); |
5d594d01 | 1437 | |
5e958e1d YW |
1438 | r = setsockopt_int(fd, SOL_SOCKET, SO_BINDTOIFINDEX, ifindex); |
1439 | if (r != -ENOPROTOOPT) | |
1440 | return r; | |
5d594d01 LP |
1441 | |
1442 | /* Fall back to SO_BINDTODEVICE on kernels < 5.0 which didn't have SO_BINDTOIFINDEX */ | |
01afd0f7 YW |
1443 | r = format_ifname(ifindex, ifname); |
1444 | if (r < 0) | |
1445 | return r; | |
5d594d01 LP |
1446 | |
1447 | return socket_bind_to_ifname(fd, ifname); | |
1448 | } | |
47eae6ce LP |
1449 | |
1450 | ssize_t recvmsg_safe(int sockfd, struct msghdr *msg, int flags) { | |
1451 | ssize_t n; | |
1452 | ||
1453 | /* A wrapper around recvmsg() that checks for MSG_CTRUNC, and turns it into an error, in a reasonably | |
1454 | * safe way, closing any SCM_RIGHTS fds in the error path. | |
1455 | * | |
1456 | * Note that unlike our usual coding style this might modify *msg on failure. */ | |
1457 | ||
1458 | n = recvmsg(sockfd, msg, flags); | |
1459 | if (n < 0) | |
1460 | return -errno; | |
1461 | ||
1462 | if (FLAGS_SET(msg->msg_flags, MSG_CTRUNC)) { | |
1463 | cmsg_close_all(msg); | |
1464 | return -EXFULL; /* a recognizable error code */ | |
1465 | } | |
1466 | ||
1467 | return n; | |
35a3eb9b LP |
1468 | } |
1469 | ||
5f64d2bf | 1470 | int socket_get_family(int fd) { |
35a3eb9b LP |
1471 | int af; |
1472 | socklen_t sl = sizeof(af); | |
1473 | ||
1474 | if (getsockopt(fd, SOL_SOCKET, SO_DOMAIN, &af, &sl) < 0) | |
1475 | return -errno; | |
1476 | ||
5d0fe423 LP |
1477 | if (sl != sizeof(af)) |
1478 | return -EINVAL; | |
1479 | ||
1480 | return af; | |
1481 | } | |
1482 | ||
1483 | int socket_set_recvpktinfo(int fd, int af, bool b) { | |
5d0fe423 LP |
1484 | |
1485 | if (af == AF_UNSPEC) { | |
5f64d2bf LP |
1486 | af = socket_get_family(fd); |
1487 | if (af < 0) | |
1488 | return af; | |
5d0fe423 LP |
1489 | } |
1490 | ||
35a3eb9b LP |
1491 | switch (af) { |
1492 | ||
1493 | case AF_INET: | |
1494 | return setsockopt_int(fd, IPPROTO_IP, IP_PKTINFO, b); | |
47eae6ce | 1495 | |
35a3eb9b LP |
1496 | case AF_INET6: |
1497 | return setsockopt_int(fd, IPPROTO_IPV6, IPV6_RECVPKTINFO, b); | |
1498 | ||
1499 | case AF_NETLINK: | |
1500 | return setsockopt_int(fd, SOL_NETLINK, NETLINK_PKTINFO, b); | |
1501 | ||
2d6d4136 LP |
1502 | case AF_PACKET: |
1503 | return setsockopt_int(fd, SOL_PACKET, PACKET_AUXDATA, b); | |
1504 | ||
35a3eb9b LP |
1505 | default: |
1506 | return -EAFNOSUPPORT; | |
1507 | } | |
47eae6ce | 1508 | } |
5d0fe423 | 1509 | |
5d0fe423 LP |
1510 | int socket_set_unicast_if(int fd, int af, int ifi) { |
1511 | be32_t ifindex_be = htobe32(ifi); | |
5d0fe423 LP |
1512 | |
1513 | if (af == AF_UNSPEC) { | |
5f64d2bf LP |
1514 | af = socket_get_family(fd); |
1515 | if (af < 0) | |
1516 | return af; | |
5d0fe423 LP |
1517 | } |
1518 | ||
1519 | switch (af) { | |
1520 | ||
1521 | case AF_INET: | |
7c248223 | 1522 | return RET_NERRNO(setsockopt(fd, IPPROTO_IP, IP_UNICAST_IF, &ifindex_be, sizeof(ifindex_be))); |
5d0fe423 LP |
1523 | |
1524 | case AF_INET6: | |
7c248223 | 1525 | return RET_NERRNO(setsockopt(fd, IPPROTO_IPV6, IPV6_UNICAST_IF, &ifindex_be, sizeof(ifindex_be))); |
5d0fe423 LP |
1526 | |
1527 | default: | |
1528 | return -EAFNOSUPPORT; | |
1529 | } | |
1530 | } | |
1531 | ||
402506ce | 1532 | int socket_set_option(int fd, int af, int opt_ipv4, int opt_ipv6, int val) { |
5d0fe423 | 1533 | if (af == AF_UNSPEC) { |
5f64d2bf LP |
1534 | af = socket_get_family(fd); |
1535 | if (af < 0) | |
1536 | return af; | |
5d0fe423 LP |
1537 | } |
1538 | ||
1539 | switch (af) { | |
1540 | ||
1541 | case AF_INET: | |
402506ce | 1542 | return setsockopt_int(fd, IPPROTO_IP, opt_ipv4, val); |
5d0fe423 LP |
1543 | |
1544 | case AF_INET6: | |
402506ce | 1545 | return setsockopt_int(fd, IPPROTO_IPV6, opt_ipv6, val); |
5d0fe423 LP |
1546 | |
1547 | default: | |
1548 | return -EAFNOSUPPORT; | |
1549 | } | |
1550 | } | |
52975f86 LP |
1551 | |
1552 | int socket_get_mtu(int fd, int af, size_t *ret) { | |
1553 | int mtu, r; | |
1554 | ||
1555 | if (af == AF_UNSPEC) { | |
5f64d2bf LP |
1556 | af = socket_get_family(fd); |
1557 | if (af < 0) | |
1558 | return af; | |
52975f86 LP |
1559 | } |
1560 | ||
1561 | switch (af) { | |
1562 | ||
1563 | case AF_INET: | |
1564 | r = getsockopt_int(fd, IPPROTO_IP, IP_MTU, &mtu); | |
1565 | break; | |
1566 | ||
1567 | case AF_INET6: | |
1568 | r = getsockopt_int(fd, IPPROTO_IPV6, IPV6_MTU, &mtu); | |
1569 | break; | |
1570 | ||
1571 | default: | |
1572 | return -EAFNOSUPPORT; | |
1573 | } | |
1574 | ||
1575 | if (r < 0) | |
1576 | return r; | |
1577 | if (mtu <= 0) | |
1578 | return -EINVAL; | |
1579 | ||
1580 | *ret = (size_t) mtu; | |
1581 | return 0; | |
1582 | } | |
2679aee4 | 1583 | |
9a603dc2 | 1584 | static int connect_unix_path_simple(int fd, const char *path) { |
2679aee4 LP |
1585 | union sockaddr_union sa = { |
1586 | .un.sun_family = AF_UNIX, | |
1587 | }; | |
9a603dc2 | 1588 | size_t l; |
2679aee4 LP |
1589 | |
1590 | assert(fd >= 0); | |
2679aee4 LP |
1591 | assert(path); |
1592 | ||
9a603dc2 LP |
1593 | l = strlen(path); |
1594 | assert(l > 0); | |
1595 | assert(l < sizeof(sa.un.sun_path)); | |
1596 | ||
1597 | memcpy(sa.un.sun_path, path, l + 1); | |
1598 | return RET_NERRNO(connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + l + 1)); | |
1599 | } | |
1600 | ||
1601 | static int connect_unix_inode(int fd, int inode_fd) { | |
1602 | assert(fd >= 0); | |
1603 | assert(inode_fd >= 0); | |
1604 | ||
1605 | return connect_unix_path_simple(fd, FORMAT_PROC_FD_PATH(inode_fd)); | |
1606 | } | |
1607 | ||
1608 | int connect_unix_path(int fd, int dir_fd, const char *path) { | |
1609 | _cleanup_close_ int inode_fd = -EBADF; | |
1610 | ||
1611 | assert(fd >= 0); | |
1612 | assert(dir_fd == AT_FDCWD || dir_fd >= 0); | |
1613 | ||
2679aee4 LP |
1614 | /* Connects to the specified AF_UNIX socket in the file system. Works around the 108 byte size limit |
1615 | * in sockaddr_un, by going via O_PATH if needed. This hence works for any kind of path. */ | |
1616 | ||
9a603dc2 LP |
1617 | if (!path) |
1618 | return connect_unix_inode(fd, dir_fd); /* If no path is specified, then dir_fd refers to the socket inode to connect to. */ | |
2679aee4 LP |
1619 | |
1620 | /* Refuse zero length path early, to make sure AF_UNIX stack won't mistake this for an abstract | |
1621 | * namespace path, since first char is NUL */ | |
9a603dc2 | 1622 | if (isempty(path)) |
2679aee4 LP |
1623 | return -EINVAL; |
1624 | ||
9a603dc2 LP |
1625 | /* Shortcut for the simple case */ |
1626 | if (dir_fd == AT_FDCWD && strlen(path) < sizeof_field(struct sockaddr_un, sun_path)) | |
1627 | return connect_unix_path_simple(fd, path); | |
2679aee4 | 1628 | |
9a603dc2 LP |
1629 | /* If dir_fd is specified, then we need to go the indirect O_PATH route, because connectat() does not |
1630 | * exist. If the path is too long, we also need to take the indirect route, since we can't fit this | |
1631 | * into a sockaddr_un directly. */ | |
2679aee4 | 1632 | |
9a603dc2 LP |
1633 | inode_fd = openat(dir_fd, path, O_PATH|O_CLOEXEC); |
1634 | if (inode_fd < 0) | |
1635 | return -errno; | |
2679aee4 | 1636 | |
9a603dc2 | 1637 | return connect_unix_inode(fd, inode_fd); |
2679aee4 | 1638 | } |
747b5d96 LB |
1639 | |
1640 | int socket_address_parse_unix(SocketAddress *ret_address, const char *s) { | |
1641 | struct sockaddr_un un; | |
1642 | int r; | |
1643 | ||
1644 | assert(ret_address); | |
1645 | assert(s); | |
1646 | ||
1647 | if (!IN_SET(*s, '/', '@')) | |
1648 | return -EPROTO; | |
1649 | ||
1650 | r = sockaddr_un_set_path(&un, s); | |
1651 | if (r < 0) | |
1652 | return r; | |
1653 | ||
1654 | *ret_address = (SocketAddress) { | |
1655 | .sockaddr.un = un, | |
1656 | .size = r, | |
1657 | }; | |
1658 | ||
1659 | return 0; | |
1660 | } | |
1661 | ||
8e471c6a LP |
1662 | int vsock_parse_port(const char *s, unsigned *ret) { |
1663 | int r; | |
1664 | ||
1665 | assert(ret); | |
1666 | ||
1667 | if (!s) | |
1668 | return -EINVAL; | |
1669 | ||
1670 | unsigned u; | |
1671 | r = safe_atou(s, &u); | |
1672 | if (r < 0) | |
1673 | return r; | |
1674 | ||
1675 | /* Port 0 is apparently valid and not special in AF_VSOCK (unlike on IP). But VMADDR_PORT_ANY | |
1676 | * (UINT32_MAX) is. Hence refuse that. */ | |
1677 | ||
1678 | if (u == VMADDR_PORT_ANY) | |
1679 | return -EINVAL; | |
1680 | ||
1681 | *ret = u; | |
1682 | return 0; | |
1683 | } | |
1684 | ||
1685 | int vsock_parse_cid(const char *s, unsigned *ret) { | |
1686 | assert(ret); | |
1687 | ||
1688 | if (!s) | |
1689 | return -EINVAL; | |
1690 | ||
1691 | /* Parsed an AF_VSOCK "CID". This is a 32bit entity, and the usual type is "unsigned". We recognize | |
1692 | * the three special CIDs as strings, and otherwise parse the numeric CIDs. */ | |
1693 | ||
1694 | if (streq(s, "hypervisor")) | |
1695 | *ret = VMADDR_CID_HYPERVISOR; | |
1696 | else if (streq(s, "local")) | |
1697 | *ret = VMADDR_CID_LOCAL; | |
1698 | else if (streq(s, "host")) | |
1699 | *ret = VMADDR_CID_HOST; | |
1700 | else | |
1701 | return safe_atou(s, ret); | |
1702 | ||
1703 | return 0; | |
1704 | } | |
1705 | ||
747b5d96 LB |
1706 | int socket_address_parse_vsock(SocketAddress *ret_address, const char *s) { |
1707 | /* AF_VSOCK socket in vsock:cid:port notation */ | |
1708 | _cleanup_free_ char *n = NULL; | |
1709 | char *e, *cid_start; | |
1710 | unsigned port, cid; | |
c31984e3 | 1711 | int type, r; |
747b5d96 LB |
1712 | |
1713 | assert(ret_address); | |
1714 | assert(s); | |
1715 | ||
c31984e3 DDM |
1716 | if ((cid_start = startswith(s, "vsock:"))) |
1717 | type = 0; | |
1718 | else if ((cid_start = startswith(s, "vsock-dgram:"))) | |
1719 | type = SOCK_DGRAM; | |
1720 | else if ((cid_start = startswith(s, "vsock-seqpacket:"))) | |
1721 | type = SOCK_SEQPACKET; | |
1722 | else if ((cid_start = startswith(s, "vsock-stream:"))) | |
1723 | type = SOCK_STREAM; | |
1724 | else | |
747b5d96 LB |
1725 | return -EPROTO; |
1726 | ||
1727 | e = strchr(cid_start, ':'); | |
1728 | if (!e) | |
1729 | return -EINVAL; | |
1730 | ||
8e471c6a | 1731 | r = vsock_parse_port(e+1, &port); |
747b5d96 LB |
1732 | if (r < 0) |
1733 | return r; | |
1734 | ||
1735 | n = strndup(cid_start, e - cid_start); | |
1736 | if (!n) | |
1737 | return -ENOMEM; | |
1738 | ||
1739 | if (isempty(n)) | |
1740 | cid = VMADDR_CID_ANY; | |
1741 | else { | |
8e471c6a | 1742 | r = vsock_parse_cid(n, &cid); |
747b5d96 LB |
1743 | if (r < 0) |
1744 | return r; | |
1745 | } | |
1746 | ||
1747 | *ret_address = (SocketAddress) { | |
1748 | .sockaddr.vm = { | |
747b5d96 | 1749 | .svm_family = AF_VSOCK, |
8e471c6a | 1750 | .svm_cid = cid, |
747b5d96 LB |
1751 | .svm_port = port, |
1752 | }, | |
c31984e3 | 1753 | .type = type, |
747b5d96 LB |
1754 | .size = sizeof(struct sockaddr_vm), |
1755 | }; | |
1756 | ||
1757 | return 0; | |
1758 | } | |
d3109d8d LP |
1759 | |
1760 | int vsock_get_local_cid(unsigned *ret) { | |
1761 | _cleanup_close_ int vsock_fd = -EBADF; | |
1762 | ||
1763 | assert(ret); | |
1764 | ||
1765 | vsock_fd = open("/dev/vsock", O_RDONLY|O_CLOEXEC); | |
1766 | if (vsock_fd < 0) | |
1767 | return log_debug_errno(errno, "Failed to open /dev/vsock: %m"); | |
1768 | ||
1769 | if (ioctl(vsock_fd, IOCTL_VM_SOCKETS_GET_LOCAL_CID, ret) < 0) | |
1770 | return log_debug_errno(errno, "Failed to query local AF_VSOCK CID: %m"); | |
1771 | ||
1772 | return 0; | |
1773 | } |