]>
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 | ||
398ce0bc | 44 | static const char* const socket_address_type_table[] = { |
955bb7fa ZJS |
45 | [SOCK_STREAM] = "Stream", |
46 | [SOCK_DGRAM] = "Datagram", | |
47 | [SOCK_RAW] = "Raw", | |
48 | [SOCK_RDM] = "ReliableDatagram", | |
398ce0bc | 49 | [SOCK_SEQPACKET] = "SequentialPacket", |
955bb7fa | 50 | [SOCK_DCCP] = "DatagramCongestionControl", |
398ce0bc YW |
51 | }; |
52 | ||
53 | DEFINE_STRING_TABLE_LOOKUP(socket_address_type, int); | |
54 | ||
15dca371 | 55 | int socket_address_verify(const SocketAddress *a, bool strict) { |
42f4e3c4 LP |
56 | assert(a); |
57 | ||
15dca371 ZJS |
58 | /* With 'strict' we enforce additional sanity constraints which are not set by the standard, |
59 | * but should only apply to sockets we create ourselves. */ | |
60 | ||
542563ba | 61 | switch (socket_address_family(a)) { |
42f4e3c4 | 62 | |
7a22745a LP |
63 | case AF_INET: |
64 | if (a->size != sizeof(struct sockaddr_in)) | |
65 | return -EINVAL; | |
42f4e3c4 | 66 | |
4d49b48c | 67 | if (a->sockaddr.in.sin_port == 0) |
7a22745a | 68 | return -EINVAL; |
42f4e3c4 | 69 | |
44ab2347 | 70 | if (!IN_SET(a->type, 0, SOCK_STREAM, SOCK_DGRAM)) |
7a22745a | 71 | return -EINVAL; |
42f4e3c4 | 72 | |
7a22745a LP |
73 | return 0; |
74 | ||
75 | case AF_INET6: | |
76 | if (a->size != sizeof(struct sockaddr_in6)) | |
77 | return -EINVAL; | |
42f4e3c4 | 78 | |
7a22745a LP |
79 | if (a->sockaddr.in6.sin6_port == 0) |
80 | return -EINVAL; | |
42f4e3c4 | 81 | |
44ab2347 | 82 | if (!IN_SET(a->type, 0, SOCK_STREAM, SOCK_DGRAM)) |
7a22745a | 83 | return -EINVAL; |
42f4e3c4 | 84 | |
7a22745a | 85 | return 0; |
42f4e3c4 | 86 | |
7a22745a LP |
87 | case AF_UNIX: |
88 | if (a->size < offsetof(struct sockaddr_un, sun_path)) | |
89 | return -EINVAL; | |
15dca371 ZJS |
90 | if (a->size > sizeof(struct sockaddr_un) + !strict) |
91 | /* If !strict, allow one extra byte, since getsockname() on Linux will append | |
92 | * a NUL byte if we have path sockets that are above sun_path's full size. */ | |
8e8132c6 | 93 | return -EINVAL; |
42f4e3c4 | 94 | |
8e8132c6 | 95 | if (a->size > offsetof(struct sockaddr_un, sun_path) && |
15dca371 ZJS |
96 | a->sockaddr.un.sun_path[0] != 0 && |
97 | strict) { | |
98 | /* Only validate file system sockets here, and only in strict mode */ | |
8e8132c6 | 99 | const char *e; |
7a22745a | 100 | |
8e8132c6 LP |
101 | e = memchr(a->sockaddr.un.sun_path, 0, sizeof(a->sockaddr.un.sun_path)); |
102 | if (e) { | |
15dca371 | 103 | /* If there's an embedded NUL byte, make sure the size of the socket address matches it */ |
7a22745a LP |
104 | if (a->size != offsetof(struct sockaddr_un, sun_path) + (e - a->sockaddr.un.sun_path) + 1) |
105 | return -EINVAL; | |
8e8132c6 | 106 | } else { |
d7b34e38 | 107 | /* If there's no embedded NUL byte, then the size needs to match the whole |
8e8132c6 LP |
108 | * structure or the structure with one extra NUL byte suffixed. (Yeah, Linux is awful, |
109 | * and considers both equivalent: getsockname() even extends sockaddr_un beyond its | |
7802194a | 110 | * size if the path is non NUL terminated.) */ |
8e8132c6 LP |
111 | if (!IN_SET(a->size, sizeof(a->sockaddr.un.sun_path), sizeof(a->sockaddr.un.sun_path)+1)) |
112 | return -EINVAL; | |
42f4e3c4 | 113 | } |
7a22745a | 114 | } |
42f4e3c4 | 115 | |
44ab2347 | 116 | if (!IN_SET(a->type, 0, SOCK_STREAM, SOCK_DGRAM, SOCK_SEQPACKET)) |
7a22745a | 117 | return -EINVAL; |
42f4e3c4 | 118 | |
7a22745a LP |
119 | return 0; |
120 | ||
121 | case AF_NETLINK: | |
122 | ||
123 | if (a->size != sizeof(struct sockaddr_nl)) | |
124 | return -EINVAL; | |
125 | ||
44ab2347 | 126 | if (!IN_SET(a->type, 0, SOCK_RAW, SOCK_DGRAM)) |
7a22745a LP |
127 | return -EINVAL; |
128 | ||
129 | return 0; | |
130 | ||
0fc0f14b SH |
131 | case AF_VSOCK: |
132 | if (a->size != sizeof(struct sockaddr_vm)) | |
133 | return -EINVAL; | |
134 | ||
44ab2347 | 135 | if (!IN_SET(a->type, 0, SOCK_STREAM, SOCK_DGRAM)) |
0fc0f14b SH |
136 | return -EINVAL; |
137 | ||
138 | return 0; | |
139 | ||
7a22745a LP |
140 | default: |
141 | return -EAFNOSUPPORT; | |
42f4e3c4 LP |
142 | } |
143 | } | |
144 | ||
4d49b48c | 145 | int socket_address_print(const SocketAddress *a, char **ret) { |
42f4e3c4 | 146 | int r; |
4d49b48c | 147 | |
42f4e3c4 | 148 | assert(a); |
4d49b48c | 149 | assert(ret); |
42f4e3c4 | 150 | |
15dca371 ZJS |
151 | r = socket_address_verify(a, false); /* We do non-strict validation, because we want to be |
152 | * able to pretty-print any socket the kernel considers | |
153 | * valid. We still need to do validation to know if we | |
154 | * can meaningfully print the address. */ | |
4d49b48c | 155 | if (r < 0) |
42f4e3c4 LP |
156 | return r; |
157 | ||
4d49b48c | 158 | if (socket_address_family(a) == AF_NETLINK) { |
7fd1b19b | 159 | _cleanup_free_ char *sfamily = NULL; |
7a22745a | 160 | |
f8b69d1d | 161 | r = netlink_family_to_string_alloc(a->protocol, &sfamily); |
7a22745a | 162 | if (r < 0) |
f8b69d1d | 163 | return r; |
4d49b48c LP |
164 | |
165 | r = asprintf(ret, "%s %u", sfamily, a->sockaddr.nl.nl_groups); | |
8520cfa5 MS |
166 | if (r < 0) |
167 | return -ENOMEM; | |
7a22745a LP |
168 | |
169 | return 0; | |
170 | } | |
171 | ||
3b1c5241 | 172 | return sockaddr_pretty(&a->sockaddr.sa, a->size, false, true, ret); |
42f4e3c4 LP |
173 | } |
174 | ||
4f2d528d LP |
175 | bool socket_address_can_accept(const SocketAddress *a) { |
176 | assert(a); | |
177 | ||
178 | return | |
3742095b | 179 | IN_SET(a->type, SOCK_STREAM, SOCK_SEQPACKET); |
4f2d528d | 180 | } |
a16e1123 LP |
181 | |
182 | bool socket_address_equal(const SocketAddress *a, const SocketAddress *b) { | |
183 | assert(a); | |
184 | assert(b); | |
185 | ||
186 | /* Invalid addresses are unequal to all */ | |
15dca371 ZJS |
187 | if (socket_address_verify(a, false) < 0 || |
188 | socket_address_verify(b, false) < 0) | |
a16e1123 LP |
189 | return false; |
190 | ||
191 | if (a->type != b->type) | |
192 | return false; | |
193 | ||
a16e1123 LP |
194 | if (socket_address_family(a) != socket_address_family(b)) |
195 | return false; | |
196 | ||
197 | switch (socket_address_family(a)) { | |
198 | ||
199 | case AF_INET: | |
4d49b48c | 200 | if (a->sockaddr.in.sin_addr.s_addr != b->sockaddr.in.sin_addr.s_addr) |
a16e1123 LP |
201 | return false; |
202 | ||
4d49b48c | 203 | if (a->sockaddr.in.sin_port != b->sockaddr.in.sin_port) |
a16e1123 LP |
204 | return false; |
205 | ||
206 | break; | |
207 | ||
208 | case AF_INET6: | |
209 | if (memcmp(&a->sockaddr.in6.sin6_addr, &b->sockaddr.in6.sin6_addr, sizeof(a->sockaddr.in6.sin6_addr)) != 0) | |
210 | return false; | |
211 | ||
212 | if (a->sockaddr.in6.sin6_port != b->sockaddr.in6.sin6_port) | |
213 | return false; | |
214 | ||
215 | break; | |
216 | ||
217 | case AF_UNIX: | |
710708a5 MS |
218 | if (a->size <= offsetof(struct sockaddr_un, sun_path) || |
219 | b->size <= offsetof(struct sockaddr_un, sun_path)) | |
220 | return false; | |
221 | ||
a16e1123 LP |
222 | if ((a->sockaddr.un.sun_path[0] == 0) != (b->sockaddr.un.sun_path[0] == 0)) |
223 | return false; | |
224 | ||
225 | if (a->sockaddr.un.sun_path[0]) { | |
e3f791a2 | 226 | if (!path_equal_or_files_same(a->sockaddr.un.sun_path, b->sockaddr.un.sun_path, 0)) |
a16e1123 LP |
227 | return false; |
228 | } else { | |
c78e47a6 MS |
229 | if (a->size != b->size) |
230 | return false; | |
231 | ||
b12c1e7c | 232 | if (memcmp(a->sockaddr.un.sun_path, b->sockaddr.un.sun_path, a->size) != 0) |
a16e1123 LP |
233 | return false; |
234 | } | |
235 | ||
236 | break; | |
237 | ||
7a22745a | 238 | case AF_NETLINK: |
7a22745a LP |
239 | if (a->protocol != b->protocol) |
240 | return false; | |
241 | ||
242 | if (a->sockaddr.nl.nl_groups != b->sockaddr.nl.nl_groups) | |
243 | return false; | |
244 | ||
245 | break; | |
246 | ||
0fc0f14b SH |
247 | case AF_VSOCK: |
248 | if (a->sockaddr.vm.svm_cid != b->sockaddr.vm.svm_cid) | |
249 | return false; | |
250 | ||
251 | if (a->sockaddr.vm.svm_port != b->sockaddr.vm.svm_port) | |
252 | return false; | |
253 | ||
254 | break; | |
255 | ||
a16e1123 LP |
256 | default: |
257 | /* Cannot compare, so we assume the addresses are different */ | |
258 | return false; | |
259 | } | |
260 | ||
261 | return true; | |
262 | } | |
263 | ||
a57f7e2c | 264 | const char* socket_address_get_path(const SocketAddress *a) { |
6e2ef85b LP |
265 | assert(a); |
266 | ||
267 | if (socket_address_family(a) != AF_UNIX) | |
a57f7e2c | 268 | return NULL; |
6e2ef85b LP |
269 | |
270 | if (a->sockaddr.un.sun_path[0] == 0) | |
a57f7e2c | 271 | return NULL; |
a16e1123 | 272 | |
48e6a2dc LP |
273 | /* Note that this is only safe because we know that there's an extra NUL byte after the sockaddr_un |
274 | * structure. On Linux AF_UNIX file system socket addresses don't have to be NUL terminated if they take up the | |
275 | * full sun_path space. */ | |
276 | assert_cc(sizeof(union sockaddr_union) >= sizeof(struct sockaddr_un)+1); | |
a57f7e2c | 277 | return a->sockaddr.un.sun_path; |
a16e1123 | 278 | } |
c0120d99 | 279 | |
5bfcc1c6 | 280 | bool socket_ipv6_is_supported(void) { |
571ec995 | 281 | static int cached = -1; |
f89f1e8f | 282 | |
571ec995 LP |
283 | if (cached < 0) { |
284 | ||
285 | if (access("/proc/net/if_inet6", F_OK) < 0) { | |
286 | ||
287 | if (errno != ENOENT) { | |
288 | log_debug_errno(errno, "Unexpected error when checking whether /proc/net/if_inet6 exists: %m"); | |
289 | return false; | |
290 | } | |
291 | ||
292 | cached = false; | |
293 | } else | |
294 | cached = true; | |
295 | } | |
296 | ||
297 | return cached; | |
5bfcc1c6 FF |
298 | } |
299 | ||
83e03c4f | 300 | bool socket_ipv6_is_enabled(void) { |
f96f5d54 | 301 | _cleanup_free_ char *v = NULL; |
83e03c4f LP |
302 | int r; |
303 | ||
304 | /* Much like socket_ipv6_is_supported(), but also checks that the sysctl that disables IPv6 on all | |
305 | * interfaces isn't turned on */ | |
306 | ||
307 | if (!socket_ipv6_is_supported()) | |
90ab5042 | 308 | return false; |
f89f1e8f | 309 | |
83e03c4f LP |
310 | r = sysctl_read_ip_property(AF_INET6, "all", "disable_ipv6", &v); |
311 | if (r < 0) { | |
312 | log_debug_errno(r, "Unexpected error reading 'net.ipv6.conf.all.disable_ipv6' sysctl: %m"); | |
313 | return true; | |
314 | } | |
315 | ||
316 | r = parse_boolean(v); | |
317 | if (r < 0) { | |
318 | log_debug_errno(r, "Failed to pare 'net.ipv6.conf.all.disable_ipv6' sysctl: %m"); | |
319 | return true; | |
320 | } | |
321 | ||
322 | return !r; | |
5bfcc1c6 FF |
323 | } |
324 | ||
01e10de3 | 325 | bool socket_address_matches_fd(const SocketAddress *a, int fd) { |
dbafedac MS |
326 | SocketAddress b; |
327 | socklen_t solen; | |
01e10de3 LP |
328 | |
329 | assert(a); | |
330 | assert(fd >= 0); | |
331 | ||
dbafedac MS |
332 | b.size = sizeof(b.sockaddr); |
333 | if (getsockname(fd, &b.sockaddr.sa, &b.size) < 0) | |
01e10de3 LP |
334 | return false; |
335 | ||
dbafedac | 336 | if (b.sockaddr.sa.sa_family != a->sockaddr.sa.sa_family) |
01e10de3 LP |
337 | return false; |
338 | ||
dbafedac MS |
339 | solen = sizeof(b.type); |
340 | if (getsockopt(fd, SOL_SOCKET, SO_TYPE, &b.type, &solen) < 0) | |
01e10de3 LP |
341 | return false; |
342 | ||
dbafedac | 343 | if (b.type != a->type) |
01e10de3 LP |
344 | return false; |
345 | ||
346 | if (a->protocol != 0) { | |
dbafedac MS |
347 | solen = sizeof(b.protocol); |
348 | if (getsockopt(fd, SOL_SOCKET, SO_PROTOCOL, &b.protocol, &solen) < 0) | |
01e10de3 LP |
349 | return false; |
350 | ||
dbafedac | 351 | if (b.protocol != a->protocol) |
01e10de3 LP |
352 | return false; |
353 | } | |
354 | ||
02233928 | 355 | return socket_address_equal(a, &b); |
01e10de3 LP |
356 | } |
357 | ||
f6aac5bf | 358 | int sockaddr_port(const struct sockaddr *_sa, unsigned *ret_port) { |
31325971 | 359 | const union sockaddr_union *sa = (const union sockaddr_union*) _sa; |
3b1c5241 | 360 | |
f6aac5bf LP |
361 | /* Note, this returns the port as 'unsigned' rather than 'uint16_t', as AF_VSOCK knows larger ports */ |
362 | ||
3b1c5241 SL |
363 | assert(sa); |
364 | ||
0fc0f14b | 365 | switch (sa->sa.sa_family) { |
f6aac5bf | 366 | |
0fc0f14b | 367 | case AF_INET: |
f6aac5bf | 368 | *ret_port = be16toh(sa->in.sin_port); |
0fc0f14b | 369 | return 0; |
3b1c5241 | 370 | |
0fc0f14b | 371 | case AF_INET6: |
f6aac5bf | 372 | *ret_port = be16toh(sa->in6.sin6_port); |
0fc0f14b SH |
373 | return 0; |
374 | ||
375 | case AF_VSOCK: | |
f6aac5bf | 376 | *ret_port = sa->vm.svm_port; |
0fc0f14b SH |
377 | return 0; |
378 | ||
379 | default: | |
380 | return -EAFNOSUPPORT; | |
381 | } | |
3b1c5241 SL |
382 | } |
383 | ||
31325971 LP |
384 | const union in_addr_union *sockaddr_in_addr(const struct sockaddr *_sa) { |
385 | const union sockaddr_union *sa = (const union sockaddr_union*) _sa; | |
386 | ||
387 | if (!sa) | |
388 | return NULL; | |
389 | ||
390 | switch (sa->sa.sa_family) { | |
391 | ||
392 | case AF_INET: | |
393 | return (const union in_addr_union*) &sa->in.sin_addr; | |
394 | ||
395 | case AF_INET6: | |
396 | return (const union in_addr_union*) &sa->in6.sin6_addr; | |
397 | ||
398 | default: | |
399 | return NULL; | |
400 | } | |
401 | } | |
402 | ||
836f9cfe LP |
403 | int sockaddr_pretty( |
404 | const struct sockaddr *_sa, | |
405 | socklen_t salen, | |
406 | bool translate_ipv6, | |
407 | bool include_port, | |
408 | char **ret) { | |
409 | ||
4d49b48c | 410 | union sockaddr_union *sa = (union sockaddr_union*) _sa; |
8569a776 | 411 | char *p; |
fc25ad25 | 412 | int r; |
8569a776 | 413 | |
4d49b48c LP |
414 | assert(sa); |
415 | assert(salen >= sizeof(sa->sa.sa_family)); | |
8569a776 | 416 | |
4d49b48c | 417 | switch (sa->sa.sa_family) { |
8569a776 LP |
418 | |
419 | case AF_INET: { | |
420 | uint32_t a; | |
421 | ||
8e38570e | 422 | a = be32toh(sa->in.sin_addr.s_addr); |
8569a776 | 423 | |
fc25ad25 ZJS |
424 | if (include_port) |
425 | r = asprintf(&p, | |
3b1c5241 SL |
426 | "%u.%u.%u.%u:%u", |
427 | a >> 24, (a >> 16) & 0xFF, (a >> 8) & 0xFF, a & 0xFF, | |
8e38570e | 428 | be16toh(sa->in.sin_port)); |
fc25ad25 ZJS |
429 | else |
430 | r = asprintf(&p, | |
3b1c5241 | 431 | "%u.%u.%u.%u", |
fc25ad25 ZJS |
432 | a >> 24, (a >> 16) & 0xFF, (a >> 8) & 0xFF, a & 0xFF); |
433 | if (r < 0) | |
434 | return -ENOMEM; | |
8569a776 LP |
435 | break; |
436 | } | |
437 | ||
438 | case AF_INET6: { | |
439 | static const unsigned char ipv4_prefix[] = { | |
440 | 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0xFF, 0xFF | |
441 | }; | |
442 | ||
fc25ad25 ZJS |
443 | if (translate_ipv6 && |
444 | memcmp(&sa->in6.sin6_addr, ipv4_prefix, sizeof(ipv4_prefix)) == 0) { | |
4d49b48c | 445 | const uint8_t *a = sa->in6.sin6_addr.s6_addr+12; |
fc25ad25 ZJS |
446 | if (include_port) |
447 | r = asprintf(&p, | |
3b1c5241 SL |
448 | "%u.%u.%u.%u:%u", |
449 | a[0], a[1], a[2], a[3], | |
8e38570e | 450 | be16toh(sa->in6.sin6_port)); |
fc25ad25 ZJS |
451 | else |
452 | r = asprintf(&p, | |
3b1c5241 | 453 | "%u.%u.%u.%u", |
fc25ad25 ZJS |
454 | a[0], a[1], a[2], a[3]); |
455 | if (r < 0) | |
456 | return -ENOMEM; | |
8569a776 | 457 | } else { |
b16d17a6 | 458 | char a[INET6_ADDRSTRLEN], ifname[IF_NAMESIZE + 1]; |
8569a776 | 459 | |
3b1c5241 | 460 | inet_ntop(AF_INET6, &sa->in6.sin6_addr, a, sizeof(a)); |
b16d17a6 ZJS |
461 | if (sa->in6.sin6_scope_id != 0) |
462 | format_ifname_full(sa->in6.sin6_scope_id, ifname, FORMAT_IFNAME_IFINDEX); | |
3b1c5241 SL |
463 | |
464 | if (include_port) { | |
fc25ad25 | 465 | r = asprintf(&p, |
b16d17a6 | 466 | "[%s]:%u%s%s", |
3b1c5241 | 467 | a, |
b16d17a6 ZJS |
468 | be16toh(sa->in6.sin6_port), |
469 | sa->in6.sin6_scope_id != 0 ? "%" : "", | |
470 | sa->in6.sin6_scope_id != 0 ? ifname : ""); | |
fc25ad25 | 471 | if (r < 0) |
3b1c5241 SL |
472 | return -ENOMEM; |
473 | } else { | |
b16d17a6 | 474 | p = sa->in6.sin6_scope_id != 0 ? strjoin(a, "%", ifname) : strdup(a); |
3b1c5241 SL |
475 | if (!p) |
476 | return -ENOMEM; | |
477 | } | |
8569a776 LP |
478 | } |
479 | ||
480 | break; | |
481 | } | |
482 | ||
4d49b48c | 483 | case AF_UNIX: |
15dca371 | 484 | if (salen <= offsetof(struct sockaddr_un, sun_path) || |
994b9d4e | 485 | (sa->un.sun_path[0] == 0 && salen == offsetof(struct sockaddr_un, sun_path) + 1)) |
15dca371 | 486 | /* The name must have at least one character (and the leading NUL does not count) */ |
4d49b48c | 487 | p = strdup("<unnamed>"); |
994b9d4e | 488 | else { |
085b39e9 LP |
489 | /* Note that we calculate the path pointer here through the .un_buffer[] field, in order to |
490 | * outtrick bounds checking tools such as ubsan, which are too smart for their own good: on | |
491 | * Linux the kernel may return sun_path[] data one byte longer than the declared size of the | |
492 | * field. */ | |
493 | char *path = (char*) sa->un_buffer + offsetof(struct sockaddr_un, sun_path); | |
15dca371 | 494 | size_t path_len = salen - offsetof(struct sockaddr_un, sun_path); |
8569a776 | 495 | |
085b39e9 | 496 | if (path[0] == 0) { |
15dca371 ZJS |
497 | /* Abstract socket. When parsing address information from, we |
498 | * explicitly reject overly long paths and paths with embedded NULs. | |
499 | * But we might get such a socket from the outside. Let's return | |
500 | * something meaningful and printable in this case. */ | |
4d49b48c | 501 | |
15dca371 | 502 | _cleanup_free_ char *e = NULL; |
4d49b48c | 503 | |
085b39e9 | 504 | e = cescape_length(path + 1, path_len - 1); |
15dca371 ZJS |
505 | if (!e) |
506 | return -ENOMEM; | |
4d49b48c | 507 | |
15dca371 ZJS |
508 | p = strjoin("@", e); |
509 | } else { | |
085b39e9 | 510 | if (path[path_len - 1] == '\0') |
15dca371 ZJS |
511 | /* We expect a terminating NUL and don't print it */ |
512 | path_len --; | |
513 | ||
085b39e9 | 514 | p = cescape_length(path, path_len); |
15dca371 | 515 | } |
4d49b48c | 516 | } |
994b9d4e LP |
517 | if (!p) |
518 | return -ENOMEM; | |
8569a776 LP |
519 | |
520 | break; | |
8569a776 | 521 | |
0fc0f14b | 522 | case AF_VSOCK: |
3a484991 ZJS |
523 | if (include_port) { |
524 | if (sa->vm.svm_cid == VMADDR_CID_ANY) | |
525 | r = asprintf(&p, "vsock::%u", sa->vm.svm_port); | |
526 | else | |
527 | r = asprintf(&p, "vsock:%u:%u", sa->vm.svm_cid, sa->vm.svm_port); | |
528 | } else | |
0fc0f14b SH |
529 | r = asprintf(&p, "vsock:%u", sa->vm.svm_cid); |
530 | if (r < 0) | |
531 | return -ENOMEM; | |
532 | break; | |
533 | ||
8569a776 | 534 | default: |
15411c0c | 535 | return -EOPNOTSUPP; |
8569a776 LP |
536 | } |
537 | ||
538 | *ret = p; | |
539 | return 0; | |
540 | } | |
541 | ||
366b7db4 | 542 | int getpeername_pretty(int fd, bool include_port, char **ret) { |
4d49b48c | 543 | union sockaddr_union sa; |
b31f535c | 544 | socklen_t salen = sizeof(sa); |
eff05270 | 545 | int r; |
4d49b48c LP |
546 | |
547 | assert(fd >= 0); | |
548 | assert(ret); | |
549 | ||
4d49b48c LP |
550 | if (getpeername(fd, &sa.sa, &salen) < 0) |
551 | return -errno; | |
552 | ||
553 | if (sa.sa.sa_family == AF_UNIX) { | |
39883f62 | 554 | struct ucred ucred = {}; |
4d49b48c LP |
555 | |
556 | /* UNIX connection sockets are anonymous, so let's use | |
557 | * PID/UID as pretty credentials instead */ | |
558 | ||
eff05270 LP |
559 | r = getpeercred(fd, &ucred); |
560 | if (r < 0) | |
561 | return r; | |
4d49b48c | 562 | |
de0671ee | 563 | if (asprintf(ret, "PID "PID_FMT"/UID "UID_FMT, ucred.pid, ucred.uid) < 0) |
4d49b48c LP |
564 | return -ENOMEM; |
565 | ||
566 | return 0; | |
567 | } | |
568 | ||
569 | /* For remote sockets we translate IPv6 addresses back to IPv4 | |
570 | * if applicable, since that's nicer. */ | |
571 | ||
366b7db4 | 572 | return sockaddr_pretty(&sa.sa, salen, true, include_port, ret); |
4d49b48c LP |
573 | } |
574 | ||
575 | int getsockname_pretty(int fd, char **ret) { | |
576 | union sockaddr_union sa; | |
b31f535c | 577 | socklen_t salen = sizeof(sa); |
4d49b48c LP |
578 | |
579 | assert(fd >= 0); | |
580 | assert(ret); | |
581 | ||
4d49b48c LP |
582 | if (getsockname(fd, &sa.sa, &salen) < 0) |
583 | return -errno; | |
584 | ||
585 | /* For local sockets we do not translate IPv6 addresses back | |
586 | * to IPv6 if applicable, since this is usually used for | |
587 | * listening sockets where the difference between IPv4 and | |
588 | * IPv6 matters. */ | |
589 | ||
3b1c5241 | 590 | return sockaddr_pretty(&sa.sa, salen, false, true, ret); |
4d49b48c LP |
591 | } |
592 | ||
b31f535c ZJS |
593 | int socknameinfo_pretty(union sockaddr_union *sa, socklen_t salen, char **_ret) { |
594 | int r; | |
595 | char host[NI_MAXHOST], *ret; | |
596 | ||
597 | assert(_ret); | |
598 | ||
6326a143 | 599 | r = getnameinfo(&sa->sa, salen, host, sizeof(host), NULL, 0, IDN_FLAGS); |
b31f535c | 600 | if (r != 0) { |
b31f535c ZJS |
601 | int saved_errno = errno; |
602 | ||
3b1c5241 | 603 | r = sockaddr_pretty(&sa->sa, salen, true, true, &ret); |
f647962d | 604 | if (r < 0) |
1938ac51 | 605 | return r; |
b31f535c | 606 | |
279d3c9c | 607 | log_debug_errno(saved_errno, "getnameinfo(%s) failed: %m", ret); |
cb651834 ZJS |
608 | } else { |
609 | ret = strdup(host); | |
610 | if (!ret) | |
1938ac51 | 611 | return -ENOMEM; |
cb651834 | 612 | } |
b31f535c ZJS |
613 | |
614 | *_ret = ret; | |
615 | return 0; | |
616 | } | |
617 | ||
7a22745a LP |
618 | static const char* const netlink_family_table[] = { |
619 | [NETLINK_ROUTE] = "route", | |
620 | [NETLINK_FIREWALL] = "firewall", | |
621 | [NETLINK_INET_DIAG] = "inet-diag", | |
622 | [NETLINK_NFLOG] = "nflog", | |
623 | [NETLINK_XFRM] = "xfrm", | |
624 | [NETLINK_SELINUX] = "selinux", | |
625 | [NETLINK_ISCSI] = "iscsi", | |
626 | [NETLINK_AUDIT] = "audit", | |
627 | [NETLINK_FIB_LOOKUP] = "fib-lookup", | |
628 | [NETLINK_CONNECTOR] = "connector", | |
629 | [NETLINK_NETFILTER] = "netfilter", | |
630 | [NETLINK_IP6_FW] = "ip6-fw", | |
631 | [NETLINK_DNRTMSG] = "dnrtmsg", | |
632 | [NETLINK_KOBJECT_UEVENT] = "kobject-uevent", | |
633 | [NETLINK_GENERIC] = "generic", | |
634 | [NETLINK_SCSITRANSPORT] = "scsitransport", | |
5570d7f9 JG |
635 | [NETLINK_ECRYPTFS] = "ecryptfs", |
636 | [NETLINK_RDMA] = "rdma", | |
7a22745a LP |
637 | }; |
638 | ||
f8b69d1d | 639 | DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(netlink_family, int, INT_MAX); |
7a22745a | 640 | |
c0120d99 LP |
641 | static const char* const socket_address_bind_ipv6_only_table[_SOCKET_ADDRESS_BIND_IPV6_ONLY_MAX] = { |
642 | [SOCKET_ADDRESS_DEFAULT] = "default", | |
643 | [SOCKET_ADDRESS_BOTH] = "both", | |
644 | [SOCKET_ADDRESS_IPV6_ONLY] = "ipv6-only" | |
645 | }; | |
646 | ||
647 | DEFINE_STRING_TABLE_LOOKUP(socket_address_bind_ipv6_only, SocketAddressBindIPv6Only); | |
f01e5736 | 648 | |
b54e98ef | 649 | SocketAddressBindIPv6Only socket_address_bind_ipv6_only_or_bool_from_string(const char *n) { |
6f90844f YW |
650 | int r; |
651 | ||
652 | r = parse_boolean(n); | |
653 | if (r > 0) | |
654 | return SOCKET_ADDRESS_IPV6_ONLY; | |
655 | if (r == 0) | |
656 | return SOCKET_ADDRESS_BOTH; | |
657 | ||
658 | return socket_address_bind_ipv6_only_from_string(n); | |
659 | } | |
660 | ||
f01e5736 LP |
661 | bool sockaddr_equal(const union sockaddr_union *a, const union sockaddr_union *b) { |
662 | assert(a); | |
663 | assert(b); | |
664 | ||
665 | if (a->sa.sa_family != b->sa.sa_family) | |
666 | return false; | |
667 | ||
668 | if (a->sa.sa_family == AF_INET) | |
669 | return a->in.sin_addr.s_addr == b->in.sin_addr.s_addr; | |
670 | ||
671 | if (a->sa.sa_family == AF_INET6) | |
672 | return memcmp(&a->in6.sin6_addr, &b->in6.sin6_addr, sizeof(a->in6.sin6_addr)) == 0; | |
673 | ||
0fc0f14b SH |
674 | if (a->sa.sa_family == AF_VSOCK) |
675 | return a->vm.svm_cid == b->vm.svm_cid; | |
676 | ||
f01e5736 LP |
677 | return false; |
678 | } | |
2583fbea | 679 | |
d9d9b2a0 | 680 | int fd_set_sndbuf(int fd, size_t n, bool increase) { |
2583fbea LP |
681 | int r, value; |
682 | socklen_t l = sizeof(value); | |
683 | ||
1263c85e YW |
684 | if (n > INT_MAX) |
685 | return -ERANGE; | |
686 | ||
2583fbea | 687 | r = getsockopt(fd, SOL_SOCKET, SO_SNDBUF, &value, &l); |
d9d9b2a0 | 688 | if (r >= 0 && l == sizeof(value) && increase ? (size_t) value >= n*2 : (size_t) value == n*2) |
2583fbea LP |
689 | return 0; |
690 | ||
b92f3507 YW |
691 | /* First, try to set the buffer size with SO_SNDBUF. */ |
692 | r = setsockopt_int(fd, SOL_SOCKET, SO_SNDBUF, n); | |
693 | if (r < 0) | |
694 | return r; | |
2583fbea | 695 | |
b92f3507 YW |
696 | /* SO_SNDBUF above may set to the kernel limit, instead of the requested size. |
697 | * So, we need to check the actual buffer size here. */ | |
67f5ae2d | 698 | l = sizeof(value); |
b92f3507 | 699 | r = getsockopt(fd, SOL_SOCKET, SO_SNDBUF, &value, &l); |
d9d9b2a0 | 700 | if (r >= 0 && l == sizeof(value) && increase ? (size_t) value >= n*2 : (size_t) value == n*2) |
b92f3507 YW |
701 | return 1; |
702 | ||
703 | /* If we have the privileges we will ignore the kernel limit. */ | |
704 | r = setsockopt_int(fd, SOL_SOCKET, SO_SNDBUFFORCE, n); | |
705 | if (r < 0) | |
706 | return r; | |
2583fbea LP |
707 | |
708 | return 1; | |
709 | } | |
710 | ||
d9d9b2a0 | 711 | int fd_set_rcvbuf(int fd, size_t n, bool increase) { |
2583fbea LP |
712 | int r, value; |
713 | socklen_t l = sizeof(value); | |
714 | ||
1263c85e YW |
715 | if (n > INT_MAX) |
716 | return -ERANGE; | |
717 | ||
2583fbea | 718 | r = getsockopt(fd, SOL_SOCKET, SO_RCVBUF, &value, &l); |
d9d9b2a0 | 719 | if (r >= 0 && l == sizeof(value) && increase ? (size_t) value >= n*2 : (size_t) value == n*2) |
2583fbea LP |
720 | return 0; |
721 | ||
b92f3507 YW |
722 | /* First, try to set the buffer size with SO_RCVBUF. */ |
723 | r = setsockopt_int(fd, SOL_SOCKET, SO_RCVBUF, n); | |
724 | if (r < 0) | |
725 | return r; | |
2583fbea | 726 | |
b92f3507 YW |
727 | /* SO_RCVBUF above may set to the kernel limit, instead of the requested size. |
728 | * So, we need to check the actual buffer size here. */ | |
67f5ae2d | 729 | l = sizeof(value); |
b92f3507 | 730 | r = getsockopt(fd, SOL_SOCKET, SO_RCVBUF, &value, &l); |
d9d9b2a0 | 731 | if (r >= 0 && l == sizeof(value) && increase ? (size_t) value >= n*2 : (size_t) value == n*2) |
b92f3507 YW |
732 | return 1; |
733 | ||
734 | /* If we have the privileges we will ignore the kernel limit. */ | |
735 | r = setsockopt_int(fd, SOL_SOCKET, SO_RCVBUFFORCE, n); | |
736 | if (r < 0) | |
737 | return r; | |
9e5b6496 | 738 | |
2583fbea LP |
739 | return 1; |
740 | } | |
741 | ||
742 | static const char* const ip_tos_table[] = { | |
743 | [IPTOS_LOWDELAY] = "low-delay", | |
744 | [IPTOS_THROUGHPUT] = "throughput", | |
745 | [IPTOS_RELIABILITY] = "reliability", | |
746 | [IPTOS_LOWCOST] = "low-cost", | |
747 | }; | |
748 | ||
749 | DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(ip_tos, int, 0xff); | |
750 | ||
5a3586db YW |
751 | bool ifname_valid_char(char a) { |
752 | if ((unsigned char) a >= 127U) | |
753 | return false; | |
754 | ||
755 | if ((unsigned char) a <= 32U) | |
756 | return false; | |
757 | ||
758 | if (IN_SET(a, | |
759 | ':', /* colons are used by the legacy "alias" interface logic */ | |
760 | '/', /* slashes cannot work, since we need to use network interfaces in sysfs paths, and in paths slashes are separators */ | |
761 | '%')) /* %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 */ | |
762 | return false; | |
763 | ||
764 | return true; | |
765 | } | |
766 | ||
2313524a | 767 | bool ifname_valid_full(const char *p, IfnameValidFlags flags) { |
ef76dff2 LP |
768 | bool numeric = true; |
769 | ||
770 | /* Checks whether a network interface name is valid. This is inspired by dev_valid_name() in the kernel sources | |
771 | * but slightly stricter, as we only allow non-control, non-space ASCII characters in the interface name. We | |
772 | * also don't permit names that only container numbers, to avoid confusion with numeric interface indexes. */ | |
773 | ||
2313524a ZJS |
774 | assert(!(flags & ~_IFNAME_VALID_ALL)); |
775 | ||
ef76dff2 LP |
776 | if (isempty(p)) |
777 | return false; | |
778 | ||
ee29bd5d LP |
779 | /* A valid ifindex? If so, it's valid iff IFNAME_VALID_NUMERIC is set */ |
780 | if (parse_ifindex(p) >= 0) | |
781 | return flags & IFNAME_VALID_NUMERIC; | |
782 | ||
2313524a | 783 | if (flags & IFNAME_VALID_ALTERNATIVE) { |
4252696a YW |
784 | if (strlen(p) >= ALTIFNAMSIZ) |
785 | return false; | |
786 | } else { | |
787 | if (strlen(p) >= IFNAMSIZ) | |
788 | return false; | |
789 | } | |
ef76dff2 | 790 | |
49bfc877 | 791 | if (dot_or_dot_dot(p)) |
ef76dff2 LP |
792 | return false; |
793 | ||
07a7441a LP |
794 | /* Let's refuse "all" and "default" as interface name, to avoid collisions with the special sysctl |
795 | * directories /proc/sys/net/{ipv4,ipv6}/conf/{all,default} */ | |
6aebfec3 | 796 | if (!FLAGS_SET(flags, IFNAME_VALID_SPECIAL) && STR_IN_SET(p, "all", "default")) |
07a7441a LP |
797 | return false; |
798 | ||
2313524a | 799 | for (const char *t = p; *t; t++) { |
5a3586db | 800 | if (!ifname_valid_char(*t)) |
ef76dff2 LP |
801 | return false; |
802 | ||
2313524a | 803 | numeric = numeric && (*t >= '0' && *t <= '9'); |
ef76dff2 LP |
804 | } |
805 | ||
ee29bd5d LP |
806 | /* It's fully numeric but didn't parse as valid ifindex above? if so, it must be too large or zero or |
807 | * so, let's refuse that. */ | |
808 | if (numeric) | |
809 | return false; | |
ef76dff2 LP |
810 | |
811 | return true; | |
812 | } | |
813 | ||
26808948 SS |
814 | bool address_label_valid(const char *p) { |
815 | ||
816 | if (isempty(p)) | |
817 | return false; | |
818 | ||
819 | if (strlen(p) >= IFNAMSIZ) | |
820 | return false; | |
821 | ||
822 | while (*p) { | |
823 | if ((uint8_t) *p >= 127U) | |
824 | return false; | |
825 | ||
826 | if ((uint8_t) *p <= 31U) | |
827 | return false; | |
828 | p++; | |
829 | } | |
830 | ||
831 | return true; | |
832 | } | |
833 | ||
2583fbea LP |
834 | int getpeercred(int fd, struct ucred *ucred) { |
835 | socklen_t n = sizeof(struct ucred); | |
836 | struct ucred u; | |
837 | int r; | |
838 | ||
839 | assert(fd >= 0); | |
840 | assert(ucred); | |
841 | ||
842 | r = getsockopt(fd, SOL_SOCKET, SO_PEERCRED, &u, &n); | |
843 | if (r < 0) | |
844 | return -errno; | |
845 | ||
846 | if (n != sizeof(struct ucred)) | |
847 | return -EIO; | |
848 | ||
bbcc701e LP |
849 | /* Check if the data is actually useful and not suppressed due to namespacing issues */ |
850 | if (!pid_is_valid(u.pid)) | |
2583fbea LP |
851 | return -ENODATA; |
852 | ||
bbcc701e LP |
853 | /* Note that we don't check UID/GID here, as namespace translation works differently there: instead of |
854 | * receiving in "invalid" user/group we get the overflow UID/GID. */ | |
855 | ||
2583fbea LP |
856 | *ucred = u; |
857 | return 0; | |
858 | } | |
859 | ||
860 | int getpeersec(int fd, char **ret) { | |
217d8967 | 861 | _cleanup_free_ char *s = NULL; |
2583fbea | 862 | socklen_t n = 64; |
2583fbea LP |
863 | |
864 | assert(fd >= 0); | |
865 | assert(ret); | |
866 | ||
217d8967 LP |
867 | for (;;) { |
868 | s = new0(char, n+1); | |
869 | if (!s) | |
870 | return -ENOMEM; | |
2583fbea | 871 | |
217d8967 LP |
872 | if (getsockopt(fd, SOL_SOCKET, SO_PEERSEC, s, &n) >= 0) |
873 | break; | |
2583fbea LP |
874 | |
875 | if (errno != ERANGE) | |
876 | return -errno; | |
877 | ||
217d8967 | 878 | s = mfree(s); |
2583fbea LP |
879 | } |
880 | ||
217d8967 | 881 | if (isempty(s)) |
2583fbea | 882 | return -EOPNOTSUPP; |
2583fbea | 883 | |
ae2a15bc | 884 | *ret = TAKE_PTR(s); |
217d8967 | 885 | |
2583fbea LP |
886 | return 0; |
887 | } | |
888 | ||
43f2c88d LP |
889 | int getpeergroups(int fd, gid_t **ret) { |
890 | socklen_t n = sizeof(gid_t) * 64; | |
891 | _cleanup_free_ gid_t *d = NULL; | |
892 | ||
75f40779 | 893 | assert(fd >= 0); |
43f2c88d LP |
894 | assert(ret); |
895 | ||
896 | for (;;) { | |
897 | d = malloc(n); | |
898 | if (!d) | |
899 | return -ENOMEM; | |
900 | ||
901 | if (getsockopt(fd, SOL_SOCKET, SO_PEERGROUPS, d, &n) >= 0) | |
902 | break; | |
903 | ||
904 | if (errno != ERANGE) | |
905 | return -errno; | |
906 | ||
907 | d = mfree(d); | |
908 | } | |
909 | ||
910 | assert_se(n % sizeof(gid_t) == 0); | |
911 | n /= sizeof(gid_t); | |
912 | ||
913 | if ((socklen_t) (int) n != n) | |
914 | return -E2BIG; | |
915 | ||
1cc6c93a | 916 | *ret = TAKE_PTR(d); |
43f2c88d LP |
917 | |
918 | return (int) n; | |
919 | } | |
920 | ||
d34673ec | 921 | ssize_t send_one_fd_iov_sa( |
726f4c47 ZJS |
922 | int transport_fd, |
923 | int fd, | |
f621b8d7 | 924 | const struct iovec *iov, size_t iovlen, |
726f4c47 ZJS |
925 | const struct sockaddr *sa, socklen_t len, |
926 | int flags) { | |
927 | ||
fb29cdbe | 928 | CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(int))) control = {}; |
2583fbea | 929 | struct msghdr mh = { |
726f4c47 ZJS |
930 | .msg_name = (struct sockaddr*) sa, |
931 | .msg_namelen = len, | |
f621b8d7 | 932 | .msg_iov = (struct iovec *)iov, |
d34673ec | 933 | .msg_iovlen = iovlen, |
2583fbea | 934 | }; |
d34673ec | 935 | ssize_t k; |
2583fbea LP |
936 | |
937 | assert(transport_fd >= 0); | |
2583fbea | 938 | |
d34673ec FB |
939 | /* |
940 | * We need either an FD or data to send. | |
941 | * If there's nothing, return an error. | |
942 | */ | |
943 | if (fd < 0 && !iov) | |
944 | return -EINVAL; | |
2583fbea | 945 | |
d34673ec FB |
946 | if (fd >= 0) { |
947 | struct cmsghdr *cmsg; | |
2583fbea | 948 | |
d34673ec FB |
949 | mh.msg_control = &control; |
950 | mh.msg_controllen = sizeof(control); | |
951 | ||
952 | cmsg = CMSG_FIRSTHDR(&mh); | |
953 | cmsg->cmsg_level = SOL_SOCKET; | |
954 | cmsg->cmsg_type = SCM_RIGHTS; | |
955 | cmsg->cmsg_len = CMSG_LEN(sizeof(int)); | |
956 | memcpy(CMSG_DATA(cmsg), &fd, sizeof(int)); | |
d34673ec FB |
957 | } |
958 | k = sendmsg(transport_fd, &mh, MSG_NOSIGNAL | flags); | |
959 | if (k < 0) | |
960 | return (ssize_t) -errno; | |
961 | ||
962 | return k; | |
2583fbea LP |
963 | } |
964 | ||
d34673ec FB |
965 | int send_one_fd_sa( |
966 | int transport_fd, | |
967 | int fd, | |
968 | const struct sockaddr *sa, socklen_t len, | |
969 | int flags) { | |
970 | ||
971 | assert(fd >= 0); | |
972 | ||
973 | return (int) send_one_fd_iov_sa(transport_fd, fd, NULL, 0, sa, len, flags); | |
974 | } | |
975 | ||
976 | ssize_t receive_one_fd_iov( | |
977 | int transport_fd, | |
978 | struct iovec *iov, size_t iovlen, | |
979 | int flags, | |
980 | int *ret_fd) { | |
981 | ||
fb29cdbe | 982 | CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(int))) control; |
2583fbea LP |
983 | struct msghdr mh = { |
984 | .msg_control = &control, | |
985 | .msg_controllen = sizeof(control), | |
d34673ec FB |
986 | .msg_iov = iov, |
987 | .msg_iovlen = iovlen, | |
2583fbea | 988 | }; |
dac556fa | 989 | struct cmsghdr *found; |
d34673ec | 990 | ssize_t k; |
2583fbea LP |
991 | |
992 | assert(transport_fd >= 0); | |
d34673ec | 993 | assert(ret_fd); |
2583fbea LP |
994 | |
995 | /* | |
996 | * Receive a single FD via @transport_fd. We don't care for | |
997 | * the transport-type. We retrieve a single FD at most, so for | |
998 | * packet-based transports, the caller must ensure to send | |
999 | * only a single FD per packet. This is best used in | |
1000 | * combination with send_one_fd(). | |
1001 | */ | |
1002 | ||
3691bcf3 | 1003 | k = recvmsg_safe(transport_fd, &mh, MSG_CMSG_CLOEXEC | flags); |
d34673ec | 1004 | if (k < 0) |
3691bcf3 | 1005 | return k; |
2583fbea | 1006 | |
dac556fa | 1007 | found = cmsg_find(&mh, SOL_SOCKET, SCM_RIGHTS, CMSG_LEN(sizeof(int))); |
3691bcf3 | 1008 | if (!found) { |
2583fbea | 1009 | cmsg_close_all(&mh); |
d34673ec | 1010 | |
3691bcf3 LP |
1011 | /* If didn't receive an FD or any data, return an error. */ |
1012 | if (k == 0) | |
1013 | return -EIO; | |
1014 | } | |
2583fbea | 1015 | |
d34673ec FB |
1016 | if (found) |
1017 | *ret_fd = *(int*) CMSG_DATA(found); | |
1018 | else | |
1019 | *ret_fd = -1; | |
1020 | ||
1021 | return k; | |
1022 | } | |
1023 | ||
1024 | int receive_one_fd(int transport_fd, int flags) { | |
1025 | int fd; | |
1026 | ssize_t k; | |
1027 | ||
1028 | k = receive_one_fd_iov(transport_fd, NULL, 0, flags, &fd); | |
1029 | if (k == 0) | |
1030 | return fd; | |
1031 | ||
1032 | /* k must be negative, since receive_one_fd_iov() only returns | |
1033 | * a positive value if data was received through the iov. */ | |
1034 | assert(k < 0); | |
1035 | return (int) k; | |
2583fbea | 1036 | } |
4edc2c9b LP |
1037 | |
1038 | ssize_t next_datagram_size_fd(int fd) { | |
1039 | ssize_t l; | |
1040 | int k; | |
1041 | ||
1042 | /* This is a bit like FIONREAD/SIOCINQ, however a bit more powerful. The difference being: recv(MSG_PEEK) will | |
96d49011 | 1043 | * actually cause the next datagram in the queue to be validated regarding checksums, which FIONREAD doesn't |
4edc2c9b LP |
1044 | * do. This difference is actually of major importance as we need to be sure that the size returned here |
1045 | * actually matches what we will read with recvmsg() next, as otherwise we might end up allocating a buffer of | |
1046 | * the wrong size. */ | |
1047 | ||
1048 | l = recv(fd, NULL, 0, MSG_PEEK|MSG_TRUNC); | |
1049 | if (l < 0) { | |
3742095b | 1050 | if (IN_SET(errno, EOPNOTSUPP, EFAULT)) |
4edc2c9b LP |
1051 | goto fallback; |
1052 | ||
1053 | return -errno; | |
1054 | } | |
1055 | if (l == 0) | |
1056 | goto fallback; | |
1057 | ||
1058 | return l; | |
1059 | ||
1060 | fallback: | |
1061 | k = 0; | |
1062 | ||
1063 | /* Some sockets (AF_PACKET) do not support null-sized recv() with MSG_TRUNC set, let's fall back to FIONREAD | |
1064 | * for them. Checksums don't matter for raw sockets anyway, hence this should be fine. */ | |
1065 | ||
1066 | if (ioctl(fd, FIONREAD, &k) < 0) | |
1067 | return -errno; | |
1068 | ||
1069 | return (ssize_t) k; | |
1070 | } | |
60d9771c | 1071 | |
67962036 ZJS |
1072 | /* Put a limit on how many times will attempt to call accept4(). We loop |
1073 | * only on "transient" errors, but let's make sure we don't loop forever. */ | |
1074 | #define MAX_FLUSH_ITERATIONS 1024 | |
1075 | ||
60d9771c LP |
1076 | int flush_accept(int fd) { |
1077 | ||
f3d75364 LP |
1078 | int r, b; |
1079 | socklen_t l = sizeof(b); | |
1080 | ||
644cb4cc ZJS |
1081 | /* Similar to flush_fd() but flushes all incoming connections by accepting and immediately closing |
1082 | * them. */ | |
f3d75364 LP |
1083 | |
1084 | if (getsockopt(fd, SOL_SOCKET, SO_ACCEPTCONN, &b, &l) < 0) | |
1085 | return -errno; | |
60d9771c | 1086 | |
f3d75364 | 1087 | assert(l == sizeof(b)); |
644cb4cc ZJS |
1088 | if (!b) /* Let's check if this socket accepts connections before calling accept(). accept4() can |
1089 | * return EOPNOTSUPP if the fd is not a listening socket, which we should treat as a fatal | |
1090 | * error, or in case the incoming TCP connection triggered a network issue, which we want to | |
1091 | * treat as a transient error. Thus, let's rule out the first reason for EOPNOTSUPP early, so | |
1092 | * we can loop safely on transient errors below. */ | |
f3d75364 | 1093 | return -ENOTTY; |
60d9771c | 1094 | |
67962036 | 1095 | for (unsigned iteration = 0;; iteration++) { |
60d9771c LP |
1096 | int cfd; |
1097 | ||
0f2d351f | 1098 | r = fd_wait_for_event(fd, POLLIN, 0); |
60d9771c | 1099 | if (r < 0) { |
0f2d351f | 1100 | if (r == -EINTR) |
60d9771c LP |
1101 | continue; |
1102 | ||
0f2d351f | 1103 | return r; |
4ff9bc2e LP |
1104 | } |
1105 | if (r == 0) | |
60d9771c LP |
1106 | return 0; |
1107 | ||
67962036 ZJS |
1108 | if (iteration >= MAX_FLUSH_ITERATIONS) |
1109 | return log_debug_errno(SYNTHETIC_ERRNO(EBUSY), | |
1110 | "Failed to flush connections within " STRINGIFY(MAX_FLUSH_ITERATIONS) " iterations."); | |
1111 | ||
60d9771c LP |
1112 | cfd = accept4(fd, NULL, NULL, SOCK_NONBLOCK|SOCK_CLOEXEC); |
1113 | if (cfd < 0) { | |
60d9771c LP |
1114 | if (errno == EAGAIN) |
1115 | return 0; | |
1116 | ||
4ff9bc2e LP |
1117 | if (ERRNO_IS_ACCEPT_AGAIN(errno)) |
1118 | continue; | |
1119 | ||
60d9771c LP |
1120 | return -errno; |
1121 | } | |
1122 | ||
4ff9bc2e | 1123 | safe_close(cfd); |
60d9771c LP |
1124 | } |
1125 | } | |
29206d46 LP |
1126 | |
1127 | struct cmsghdr* cmsg_find(struct msghdr *mh, int level, int type, socklen_t length) { | |
1128 | struct cmsghdr *cmsg; | |
1129 | ||
1130 | assert(mh); | |
1131 | ||
1132 | CMSG_FOREACH(cmsg, mh) | |
1133 | if (cmsg->cmsg_level == level && | |
1134 | cmsg->cmsg_type == type && | |
1135 | (length == (socklen_t) -1 || length == cmsg->cmsg_len)) | |
1136 | return cmsg; | |
1137 | ||
1138 | return NULL; | |
1139 | } | |
429b4350 LP |
1140 | |
1141 | int socket_ioctl_fd(void) { | |
1142 | int fd; | |
1143 | ||
1144 | /* Create a socket to invoke the various network interface ioctl()s on. Traditionally only AF_INET was good for | |
1145 | * that. Since kernel 4.6 AF_NETLINK works for this too. We first try to use AF_INET hence, but if that's not | |
1146 | * available (for example, because it is made unavailable via SECCOMP or such), we'll fall back to the more | |
1147 | * generic AF_NETLINK. */ | |
1148 | ||
1149 | fd = socket(AF_INET, SOCK_DGRAM|SOCK_CLOEXEC, 0); | |
1150 | if (fd < 0) | |
1151 | fd = socket(AF_NETLINK, SOCK_RAW|SOCK_CLOEXEC, NETLINK_GENERIC); | |
1152 | if (fd < 0) | |
1153 | return -errno; | |
1154 | ||
1155 | return fd; | |
1156 | } | |
9f20fc28 LP |
1157 | |
1158 | int sockaddr_un_unlink(const struct sockaddr_un *sa) { | |
1159 | const char *p, * nul; | |
1160 | ||
1161 | assert(sa); | |
1162 | ||
1163 | if (sa->sun_family != AF_UNIX) | |
1164 | return -EPROTOTYPE; | |
1165 | ||
1166 | if (sa->sun_path[0] == 0) /* Nothing to do for abstract sockets */ | |
1167 | return 0; | |
1168 | ||
1169 | /* The path in .sun_path is not necessarily NUL terminated. Let's fix that. */ | |
1170 | nul = memchr(sa->sun_path, 0, sizeof(sa->sun_path)); | |
1171 | if (nul) | |
1172 | p = sa->sun_path; | |
1173 | else | |
1174 | p = memdupa_suffix0(sa->sun_path, sizeof(sa->sun_path)); | |
1175 | ||
1176 | if (unlink(p) < 0) | |
1177 | return -errno; | |
1178 | ||
1179 | return 1; | |
1180 | } | |
5cf91ea9 LP |
1181 | |
1182 | int sockaddr_un_set_path(struct sockaddr_un *ret, const char *path) { | |
1183 | size_t l; | |
1184 | ||
1185 | assert(ret); | |
1186 | assert(path); | |
1187 | ||
1188 | /* Initialize ret->sun_path from the specified argument. This will interpret paths starting with '@' as | |
1189 | * abstract namespace sockets, and those starting with '/' as regular filesystem sockets. It won't accept | |
1190 | * anything else (i.e. no relative paths), to avoid ambiguities. Note that this function cannot be used to | |
1191 | * reference paths in the abstract namespace that include NUL bytes in the name. */ | |
1192 | ||
1193 | l = strlen(path); | |
c097bf1f | 1194 | if (l < 2) |
5cf91ea9 LP |
1195 | return -EINVAL; |
1196 | if (!IN_SET(path[0], '/', '@')) | |
1197 | return -EINVAL; | |
5cf91ea9 LP |
1198 | |
1199 | /* Don't allow paths larger than the space in sockaddr_un. Note that we are a tiny bit more restrictive than | |
1200 | * the kernel is: we insist on NUL termination (both for abstract namespace and regular file system socket | |
1201 | * addresses!), which the kernel doesn't. We do this to reduce chance of incompatibility with other apps that | |
7802194a | 1202 | * do not expect non-NUL terminated file system path. */ |
5cf91ea9 LP |
1203 | if (l+1 > sizeof(ret->sun_path)) |
1204 | return -EINVAL; | |
1205 | ||
1206 | *ret = (struct sockaddr_un) { | |
1207 | .sun_family = AF_UNIX, | |
1208 | }; | |
1209 | ||
1210 | if (path[0] == '@') { | |
1211 | /* Abstract namespace socket */ | |
1212 | memcpy(ret->sun_path + 1, path + 1, l); /* copy *with* trailing NUL byte */ | |
1213 | return (int) (offsetof(struct sockaddr_un, sun_path) + l); /* 🔥 *don't* 🔥 include trailing NUL in size */ | |
1214 | ||
1215 | } else { | |
1216 | assert(path[0] == '/'); | |
1217 | ||
1218 | /* File system socket */ | |
1219 | memcpy(ret->sun_path, path, l + 1); /* copy *with* trailing NUL byte */ | |
1220 | return (int) (offsetof(struct sockaddr_un, sun_path) + l + 1); /* include trailing NUL in size */ | |
1221 | } | |
1222 | } | |
5d594d01 LP |
1223 | |
1224 | int socket_bind_to_ifname(int fd, const char *ifname) { | |
1225 | assert(fd >= 0); | |
1226 | ||
1227 | /* Call with NULL to drop binding */ | |
1228 | ||
1229 | if (setsockopt(fd, SOL_SOCKET, SO_BINDTODEVICE, ifname, strlen_ptr(ifname)) < 0) | |
1230 | return -errno; | |
1231 | ||
1232 | return 0; | |
1233 | } | |
1234 | ||
1235 | int socket_bind_to_ifindex(int fd, int ifindex) { | |
518a66ec | 1236 | char ifname[IF_NAMESIZE + 1]; |
5e958e1d | 1237 | int r; |
5d594d01 LP |
1238 | |
1239 | assert(fd >= 0); | |
1240 | ||
1241 | if (ifindex <= 0) { | |
1242 | /* Drop binding */ | |
1243 | if (setsockopt(fd, SOL_SOCKET, SO_BINDTODEVICE, NULL, 0) < 0) | |
1244 | return -errno; | |
1245 | ||
1246 | return 0; | |
1247 | } | |
1248 | ||
5e958e1d YW |
1249 | r = setsockopt_int(fd, SOL_SOCKET, SO_BINDTOIFINDEX, ifindex); |
1250 | if (r != -ENOPROTOOPT) | |
1251 | return r; | |
5d594d01 LP |
1252 | |
1253 | /* Fall back to SO_BINDTODEVICE on kernels < 5.0 which didn't have SO_BINDTOIFINDEX */ | |
518a66ec | 1254 | if (!format_ifname(ifindex, ifname)) |
5d594d01 LP |
1255 | return -errno; |
1256 | ||
1257 | return socket_bind_to_ifname(fd, ifname); | |
1258 | } | |
47eae6ce LP |
1259 | |
1260 | ssize_t recvmsg_safe(int sockfd, struct msghdr *msg, int flags) { | |
1261 | ssize_t n; | |
1262 | ||
1263 | /* A wrapper around recvmsg() that checks for MSG_CTRUNC, and turns it into an error, in a reasonably | |
1264 | * safe way, closing any SCM_RIGHTS fds in the error path. | |
1265 | * | |
1266 | * Note that unlike our usual coding style this might modify *msg on failure. */ | |
1267 | ||
1268 | n = recvmsg(sockfd, msg, flags); | |
1269 | if (n < 0) | |
1270 | return -errno; | |
1271 | ||
1272 | if (FLAGS_SET(msg->msg_flags, MSG_CTRUNC)) { | |
1273 | cmsg_close_all(msg); | |
1274 | return -EXFULL; /* a recognizable error code */ | |
1275 | } | |
1276 | ||
1277 | return n; | |
35a3eb9b LP |
1278 | } |
1279 | ||
5d0fe423 | 1280 | int socket_get_family(int fd, int *ret) { |
35a3eb9b LP |
1281 | int af; |
1282 | socklen_t sl = sizeof(af); | |
1283 | ||
1284 | if (getsockopt(fd, SOL_SOCKET, SO_DOMAIN, &af, &sl) < 0) | |
1285 | return -errno; | |
1286 | ||
5d0fe423 LP |
1287 | if (sl != sizeof(af)) |
1288 | return -EINVAL; | |
1289 | ||
1290 | return af; | |
1291 | } | |
1292 | ||
1293 | int socket_set_recvpktinfo(int fd, int af, bool b) { | |
1294 | int r; | |
1295 | ||
1296 | if (af == AF_UNSPEC) { | |
1297 | r = socket_get_family(fd, &af); | |
1298 | if (r < 0) | |
1299 | return r; | |
1300 | } | |
1301 | ||
35a3eb9b LP |
1302 | switch (af) { |
1303 | ||
1304 | case AF_INET: | |
1305 | return setsockopt_int(fd, IPPROTO_IP, IP_PKTINFO, b); | |
47eae6ce | 1306 | |
35a3eb9b LP |
1307 | case AF_INET6: |
1308 | return setsockopt_int(fd, IPPROTO_IPV6, IPV6_RECVPKTINFO, b); | |
1309 | ||
1310 | case AF_NETLINK: | |
1311 | return setsockopt_int(fd, SOL_NETLINK, NETLINK_PKTINFO, b); | |
1312 | ||
2d6d4136 LP |
1313 | case AF_PACKET: |
1314 | return setsockopt_int(fd, SOL_PACKET, PACKET_AUXDATA, b); | |
1315 | ||
35a3eb9b LP |
1316 | default: |
1317 | return -EAFNOSUPPORT; | |
1318 | } | |
47eae6ce | 1319 | } |
5d0fe423 | 1320 | |
5d0fe423 LP |
1321 | int socket_set_unicast_if(int fd, int af, int ifi) { |
1322 | be32_t ifindex_be = htobe32(ifi); | |
1323 | int r; | |
1324 | ||
1325 | if (af == AF_UNSPEC) { | |
1326 | r = socket_get_family(fd, &af); | |
1327 | if (r < 0) | |
1328 | return r; | |
1329 | } | |
1330 | ||
1331 | switch (af) { | |
1332 | ||
1333 | case AF_INET: | |
1334 | if (setsockopt(fd, IPPROTO_IP, IP_UNICAST_IF, &ifindex_be, sizeof(ifindex_be)) < 0) | |
1335 | return -errno; | |
1336 | ||
1337 | return 0; | |
1338 | ||
1339 | case AF_INET6: | |
1340 | if (setsockopt(fd, IPPROTO_IPV6, IPV6_UNICAST_IF, &ifindex_be, sizeof(ifindex_be)) < 0) | |
1341 | return -errno; | |
1342 | ||
1343 | return 0; | |
1344 | ||
1345 | default: | |
1346 | return -EAFNOSUPPORT; | |
1347 | } | |
1348 | } | |
1349 | ||
402506ce | 1350 | int socket_set_option(int fd, int af, int opt_ipv4, int opt_ipv6, int val) { |
5d0fe423 LP |
1351 | int r; |
1352 | ||
1353 | if (af == AF_UNSPEC) { | |
1354 | r = socket_get_family(fd, &af); | |
1355 | if (r < 0) | |
1356 | return r; | |
1357 | } | |
1358 | ||
1359 | switch (af) { | |
1360 | ||
1361 | case AF_INET: | |
402506ce | 1362 | return setsockopt_int(fd, IPPROTO_IP, opt_ipv4, val); |
5d0fe423 LP |
1363 | |
1364 | case AF_INET6: | |
402506ce | 1365 | return setsockopt_int(fd, IPPROTO_IPV6, opt_ipv6, val); |
5d0fe423 LP |
1366 | |
1367 | default: | |
1368 | return -EAFNOSUPPORT; | |
1369 | } | |
1370 | } | |
52975f86 LP |
1371 | |
1372 | int socket_get_mtu(int fd, int af, size_t *ret) { | |
1373 | int mtu, r; | |
1374 | ||
1375 | if (af == AF_UNSPEC) { | |
1376 | r = socket_get_family(fd, &af); | |
1377 | if (r < 0) | |
1378 | return r; | |
1379 | } | |
1380 | ||
1381 | switch (af) { | |
1382 | ||
1383 | case AF_INET: | |
1384 | r = getsockopt_int(fd, IPPROTO_IP, IP_MTU, &mtu); | |
1385 | break; | |
1386 | ||
1387 | case AF_INET6: | |
1388 | r = getsockopt_int(fd, IPPROTO_IPV6, IPV6_MTU, &mtu); | |
1389 | break; | |
1390 | ||
1391 | default: | |
1392 | return -EAFNOSUPPORT; | |
1393 | } | |
1394 | ||
1395 | if (r < 0) | |
1396 | return r; | |
1397 | if (mtu <= 0) | |
1398 | return -EINVAL; | |
1399 | ||
1400 | *ret = (size_t) mtu; | |
1401 | return 0; | |
1402 | } |