]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/basic/socket-util.c
projects / thirdparty / systemd.git / blob
? search:


3d929f5418dcda72f1391f605473bb8756a9d942
[thirdparty/systemd.git] / src / basic / socket-util.c
1 /* SPDX-License-Identifier: LGPL-2.1+ */
2
3 #include <arpa/inet.h>
4 #include <errno.h>
5 #include <limits.h>
6 #include <net/if.h>
7 #include <netdb.h>
8 #include <netinet/ip.h>
9 #include <poll.h>
10 #include <stddef.h>
11 #include <stdint.h>
12 #include <stdio.h>
13 #include <stdlib.h>
14 #include <string.h>
15 #include <unistd.h>
16
17 #include "alloc-util.h"
18 #include "escape.h"
19 #include "fd-util.h"
20 #include "fileio.h"
21 #include "format-util.h"
22 #include "log.h"
23 #include "macro.h"
24 #include "memory-util.h"
25 #include "missing.h"
26 #include "parse-util.h"
27 #include "path-util.h"
28 #include "process-util.h"
29 #include "socket-util.h"
30 #include "string-table.h"
31 #include "string-util.h"
32 #include "strv.h"
33 #include "user-util.h"
34 #include "utf8.h"
35
36 #if ENABLE_IDN
37 # define IDN_FLAGS NI_IDN
38 #else
39 # define IDN_FLAGS 0
40 #endif
41
42 static const char* const socket_address_type_table[] = {
43 [SOCK_STREAM] = "Stream",
44 [SOCK_DGRAM] = "Datagram",
45 [SOCK_RAW] = "Raw",
46 [SOCK_RDM] = "ReliableDatagram",
47 [SOCK_SEQPACKET] = "SequentialPacket",
48 [SOCK_DCCP] = "DatagramCongestionControl",
49 };
50
51 DEFINE_STRING_TABLE_LOOKUP(socket_address_type, int);
52
53 int socket_address_parse(SocketAddress *a, const char *s) {
54 _cleanup_free_ char *n = NULL;
55 char *e;
56 int r;
57
58 assert(a);
59 assert(s);
60
61 *a = (SocketAddress) {
62 .type = SOCK_STREAM,
63 };
64
65 if (*s == '[') {
66 uint16_t port;
67
68 /* IPv6 in [x:.....:z]:p notation */
69
70 e = strchr(s+1, ']');
71 if (!e)
72 return -EINVAL;
73
74 n = strndup(s+1, e-s-1);
75 if (!n)
76 return -ENOMEM;
77
78 errno = 0;
79 if (inet_pton(AF_INET6, n, &a->sockaddr.in6.sin6_addr) <= 0)
80 return errno > 0 ? -errno : -EINVAL;
81
82 e++;
83 if (*e != ':')
84 return -EINVAL;
85
86 e++;
87 r = parse_ip_port(e, &port);
88 if (r < 0)
89 return r;
90
91 a->sockaddr.in6.sin6_family = AF_INET6;
92 a->sockaddr.in6.sin6_port = htobe16(port);
93 a->size = sizeof(struct sockaddr_in6);
94
95 } else if (*s == '/') {
96 /* AF_UNIX socket */
97
98 size_t l;
99
100 l = strlen(s);
101 if (l >= sizeof(a->sockaddr.un.sun_path)) /* Note that we refuse non-NUL-terminated sockets when
102 * parsing (the kernel itself is less strict here in what it
103 * accepts) */
104 return -EINVAL;
105
106 a->sockaddr.un.sun_family = AF_UNIX;
107 memcpy(a->sockaddr.un.sun_path, s, l);
108 a->size = offsetof(struct sockaddr_un, sun_path) + l + 1;
109
110 } else if (*s == '@') {
111 /* Abstract AF_UNIX socket */
112 size_t l;
113
114 l = strlen(s+1);
115 if (l >= sizeof(a->sockaddr.un.sun_path) - 1) /* Note that we refuse non-NUL-terminated sockets here
116 * when parsing, even though abstract namespace sockets
117 * explicitly allow embedded NUL bytes and don't consider
118 * them special. But it's simply annoying to debug such
119 * sockets. */
120 return -EINVAL;
121
122 a->sockaddr.un.sun_family = AF_UNIX;
123 memcpy(a->sockaddr.un.sun_path+1, s+1, l);
124 a->size = offsetof(struct sockaddr_un, sun_path) + 1 + l;
125
126 } else if (startswith(s, "vsock:")) {
127 /* AF_VSOCK socket in vsock:cid:port notation */
128 const char *cid_start = s + STRLEN("vsock:");
129 unsigned port;
130
131 e = strchr(cid_start, ':');
132 if (!e)
133 return -EINVAL;
134
135 r = safe_atou(e+1, &port);
136 if (r < 0)
137 return r;
138
139 n = strndup(cid_start, e - cid_start);
140 if (!n)
141 return -ENOMEM;
142
143 if (!isempty(n)) {
144 r = safe_atou(n, &a->sockaddr.vm.svm_cid);
145 if (r < 0)
146 return r;
147 } else
148 a->sockaddr.vm.svm_cid = VMADDR_CID_ANY;
149
150 a->sockaddr.vm.svm_family = AF_VSOCK;
151 a->sockaddr.vm.svm_port = port;
152 a->size = sizeof(struct sockaddr_vm);
153
154 } else {
155 uint16_t port;
156
157 e = strchr(s, ':');
158 if (e) {
159 r = parse_ip_port(e + 1, &port);
160 if (r < 0)
161 return r;
162
163 n = strndup(s, e-s);
164 if (!n)
165 return -ENOMEM;
166
167 /* IPv4 in w.x.y.z:p notation? */
168 r = inet_pton(AF_INET, n, &a->sockaddr.in.sin_addr);
169 if (r < 0)
170 return -errno;
171
172 if (r > 0) {
173 /* Gotcha, it's a traditional IPv4 address */
174 a->sockaddr.in.sin_family = AF_INET;
175 a->sockaddr.in.sin_port = htobe16(port);
176 a->size = sizeof(struct sockaddr_in);
177 } else {
178 unsigned idx;
179
180 if (strlen(n) > IF_NAMESIZE-1)
181 return -EINVAL;
182
183 /* Uh, our last resort, an interface name */
184 idx = if_nametoindex(n);
185 if (idx == 0)
186 return -EINVAL;
187
188 a->sockaddr.in6.sin6_family = AF_INET6;
189 a->sockaddr.in6.sin6_port = htobe16(port);
190 a->sockaddr.in6.sin6_scope_id = idx;
191 a->sockaddr.in6.sin6_addr = in6addr_any;
192 a->size = sizeof(struct sockaddr_in6);
193 }
194 } else {
195
196 /* Just a port */
197 r = parse_ip_port(s, &port);
198 if (r < 0)
199 return r;
200
201 if (socket_ipv6_is_supported()) {
202 a->sockaddr.in6.sin6_family = AF_INET6;
203 a->sockaddr.in6.sin6_port = htobe16(port);
204 a->sockaddr.in6.sin6_addr = in6addr_any;
205 a->size = sizeof(struct sockaddr_in6);
206 } else {
207 a->sockaddr.in.sin_family = AF_INET;
208 a->sockaddr.in.sin_port = htobe16(port);
209 a->sockaddr.in.sin_addr.s_addr = INADDR_ANY;
210 a->size = sizeof(struct sockaddr_in);
211 }
212 }
213 }
214
215 return 0;
216 }
217
218 int socket_address_parse_and_warn(SocketAddress *a, const char *s) {
219 SocketAddress b;
220 int r;
221
222 /* Similar to socket_address_parse() but warns for IPv6 sockets when we don't support them. */
223
224 r = socket_address_parse(&b, s);
225 if (r < 0)
226 return r;
227
228 if (!socket_ipv6_is_supported() && b.sockaddr.sa.sa_family == AF_INET6) {
229 log_warning("Binding to IPv6 address not available since kernel does not support IPv6.");
230 return -EAFNOSUPPORT;
231 }
232
233 *a = b;
234 return 0;
235 }
236
237 int socket_address_parse_netlink(SocketAddress *a, const char *s) {
238 _cleanup_free_ char *word = NULL;
239 unsigned group = 0;
240 int family, r;
241
242 assert(a);
243 assert(s);
244
245 zero(*a);
246 a->type = SOCK_RAW;
247
248 r = extract_first_word(&s, &word, NULL, 0);
249 if (r < 0)
250 return r;
251 if (r == 0)
252 return -EINVAL;
253
254 family = netlink_family_from_string(word);
255 if (family < 0)
256 return -EINVAL;
257
258 if (!isempty(s)) {
259 r = safe_atou(s, &group);
260 if (r < 0)
261 return r;
262 }
263
264 a->sockaddr.nl.nl_family = AF_NETLINK;
265 a->sockaddr.nl.nl_groups = group;
266
267 a->type = SOCK_RAW;
268 a->size = sizeof(struct sockaddr_nl);
269 a->protocol = family;
270
271 return 0;
272 }
273
274 int socket_address_verify(const SocketAddress *a, bool strict) {
275 assert(a);
276
277 /* With 'strict' we enforce additional sanity constraints which are not set by the standard,
278 * but should only apply to sockets we create ourselves. */
279
280 switch (socket_address_family(a)) {
281
282 case AF_INET:
283 if (a->size != sizeof(struct sockaddr_in))
284 return -EINVAL;
285
286 if (a->sockaddr.in.sin_port == 0)
287 return -EINVAL;
288
289 if (!IN_SET(a->type, SOCK_STREAM, SOCK_DGRAM))
290 return -EINVAL;
291
292 return 0;
293
294 case AF_INET6:
295 if (a->size != sizeof(struct sockaddr_in6))
296 return -EINVAL;
297
298 if (a->sockaddr.in6.sin6_port == 0)
299 return -EINVAL;
300
301 if (!IN_SET(a->type, SOCK_STREAM, SOCK_DGRAM))
302 return -EINVAL;
303
304 return 0;
305
306 case AF_UNIX:
307 if (a->size < offsetof(struct sockaddr_un, sun_path))
308 return -EINVAL;
309 if (a->size > sizeof(struct sockaddr_un) + !strict)
310 /* If !strict, allow one extra byte, since getsockname() on Linux will append
311 * a NUL byte if we have path sockets that are above sun_path's full size. */
312 return -EINVAL;
313
314 if (a->size > offsetof(struct sockaddr_un, sun_path) &&
315 a->sockaddr.un.sun_path[0] != 0 &&
316 strict) {
317 /* Only validate file system sockets here, and only in strict mode */
318 const char *e;
319
320 e = memchr(a->sockaddr.un.sun_path, 0, sizeof(a->sockaddr.un.sun_path));
321 if (e) {
322 /* If there's an embedded NUL byte, make sure the size of the socket address matches it */
323 if (a->size != offsetof(struct sockaddr_un, sun_path) + (e - a->sockaddr.un.sun_path) + 1)
324 return -EINVAL;
325 } else {
326 /* If there's no embedded NUL byte, then then the size needs to match the whole
327 * structure or the structure with one extra NUL byte suffixed. (Yeah, Linux is awful,
328 * and considers both equivalent: getsockname() even extends sockaddr_un beyond its
329 * size if the path is non NUL terminated.)*/
330 if (!IN_SET(a->size, sizeof(a->sockaddr.un.sun_path), sizeof(a->sockaddr.un.sun_path)+1))
331 return -EINVAL;
332 }
333 }
334
335 if (!IN_SET(a->type, SOCK_STREAM, SOCK_DGRAM, SOCK_SEQPACKET))
336 return -EINVAL;
337
338 return 0;
339
340 case AF_NETLINK:
341
342 if (a->size != sizeof(struct sockaddr_nl))
343 return -EINVAL;
344
345 if (!IN_SET(a->type, SOCK_RAW, SOCK_DGRAM))
346 return -EINVAL;
347
348 return 0;
349
350 case AF_VSOCK:
351 if (a->size != sizeof(struct sockaddr_vm))
352 return -EINVAL;
353
354 if (!IN_SET(a->type, SOCK_STREAM, SOCK_DGRAM))
355 return -EINVAL;
356
357 return 0;
358
359 default:
360 return -EAFNOSUPPORT;
361 }
362 }
363
364 int socket_address_print(const SocketAddress *a, char **ret) {
365 int r;
366
367 assert(a);
368 assert(ret);
369
370 r = socket_address_verify(a, false); /* We do non-strict validation, because we want to be
371 * able to pretty-print any socket the kernel considers
372 * valid. We still need to do validation to know if we
373 * can meaningfully print the address. */
374 if (r < 0)
375 return r;
376
377 if (socket_address_family(a) == AF_NETLINK) {
378 _cleanup_free_ char *sfamily = NULL;
379
380 r = netlink_family_to_string_alloc(a->protocol, &sfamily);
381 if (r < 0)
382 return r;
383
384 r = asprintf(ret, "%s %u", sfamily, a->sockaddr.nl.nl_groups);
385 if (r < 0)
386 return -ENOMEM;
387
388 return 0;
389 }
390
391 return sockaddr_pretty(&a->sockaddr.sa, a->size, false, true, ret);
392 }
393
394 bool socket_address_can_accept(const SocketAddress *a) {
395 assert(a);
396
397 return
398 IN_SET(a->type, SOCK_STREAM, SOCK_SEQPACKET);
399 }
400
401 bool socket_address_equal(const SocketAddress *a, const SocketAddress *b) {
402 assert(a);
403 assert(b);
404
405 /* Invalid addresses are unequal to all */
406 if (socket_address_verify(a, false) < 0 ||
407 socket_address_verify(b, false) < 0)
408 return false;
409
410 if (a->type != b->type)
411 return false;
412
413 if (socket_address_family(a) != socket_address_family(b))
414 return false;
415
416 switch (socket_address_family(a)) {
417
418 case AF_INET:
419 if (a->sockaddr.in.sin_addr.s_addr != b->sockaddr.in.sin_addr.s_addr)
420 return false;
421
422 if (a->sockaddr.in.sin_port != b->sockaddr.in.sin_port)
423 return false;
424
425 break;
426
427 case AF_INET6:
428 if (memcmp(&a->sockaddr.in6.sin6_addr, &b->sockaddr.in6.sin6_addr, sizeof(a->sockaddr.in6.sin6_addr)) != 0)
429 return false;
430
431 if (a->sockaddr.in6.sin6_port != b->sockaddr.in6.sin6_port)
432 return false;
433
434 break;
435
436 case AF_UNIX:
437 if (a->size <= offsetof(struct sockaddr_un, sun_path) ||
438 b->size <= offsetof(struct sockaddr_un, sun_path))
439 return false;
440
441 if ((a->sockaddr.un.sun_path[0] == 0) != (b->sockaddr.un.sun_path[0] == 0))
442 return false;
443
444 if (a->sockaddr.un.sun_path[0]) {
445 if (!path_equal_or_files_same(a->sockaddr.un.sun_path, b->sockaddr.un.sun_path, 0))
446 return false;
447 } else {
448 if (a->size != b->size)
449 return false;
450
451 if (memcmp(a->sockaddr.un.sun_path, b->sockaddr.un.sun_path, a->size) != 0)
452 return false;
453 }
454
455 break;
456
457 case AF_NETLINK:
458 if (a->protocol != b->protocol)
459 return false;
460
461 if (a->sockaddr.nl.nl_groups != b->sockaddr.nl.nl_groups)
462 return false;
463
464 break;
465
466 case AF_VSOCK:
467 if (a->sockaddr.vm.svm_cid != b->sockaddr.vm.svm_cid)
468 return false;
469
470 if (a->sockaddr.vm.svm_port != b->sockaddr.vm.svm_port)
471 return false;
472
473 break;
474
475 default:
476 /* Cannot compare, so we assume the addresses are different */
477 return false;
478 }
479
480 return true;
481 }
482
483 bool socket_address_is(const SocketAddress *a, const char *s, int type) {
484 struct SocketAddress b;
485
486 assert(a);
487 assert(s);
488
489 if (socket_address_parse(&b, s) < 0)
490 return false;
491
492 b.type = type;
493
494 return socket_address_equal(a, &b);
495 }
496
497 bool socket_address_is_netlink(const SocketAddress *a, const char *s) {
498 struct SocketAddress b;
499
500 assert(a);
501 assert(s);
502
503 if (socket_address_parse_netlink(&b, s) < 0)
504 return false;
505
506 return socket_address_equal(a, &b);
507 }
508
509 const char* socket_address_get_path(const SocketAddress *a) {
510 assert(a);
511
512 if (socket_address_family(a) != AF_UNIX)
513 return NULL;
514
515 if (a->sockaddr.un.sun_path[0] == 0)
516 return NULL;
517
518 /* Note that this is only safe because we know that there's an extra NUL byte after the sockaddr_un
519 * structure. On Linux AF_UNIX file system socket addresses don't have to be NUL terminated if they take up the
520 * full sun_path space. */
521 assert_cc(sizeof(union sockaddr_union) >= sizeof(struct sockaddr_un)+1);
522 return a->sockaddr.un.sun_path;
523 }
524
525 bool socket_ipv6_is_supported(void) {
526 if (access("/proc/net/if_inet6", F_OK) != 0)
527 return false;
528
529 return true;
530 }
531
532 bool socket_address_matches_fd(const SocketAddress *a, int fd) {
533 SocketAddress b;
534 socklen_t solen;
535
536 assert(a);
537 assert(fd >= 0);
538
539 b.size = sizeof(b.sockaddr);
540 if (getsockname(fd, &b.sockaddr.sa, &b.size) < 0)
541 return false;
542
543 if (b.sockaddr.sa.sa_family != a->sockaddr.sa.sa_family)
544 return false;
545
546 solen = sizeof(b.type);
547 if (getsockopt(fd, SOL_SOCKET, SO_TYPE, &b.type, &solen) < 0)
548 return false;
549
550 if (b.type != a->type)
551 return false;
552
553 if (a->protocol != 0) {
554 solen = sizeof(b.protocol);
555 if (getsockopt(fd, SOL_SOCKET, SO_PROTOCOL, &b.protocol, &solen) < 0)
556 return false;
557
558 if (b.protocol != a->protocol)
559 return false;
560 }
561
562 return socket_address_equal(a, &b);
563 }
564
565 int sockaddr_port(const struct sockaddr *_sa, unsigned *ret_port) {
566 union sockaddr_union *sa = (union sockaddr_union*) _sa;
567
568 /* Note, this returns the port as 'unsigned' rather than 'uint16_t', as AF_VSOCK knows larger ports */
569
570 assert(sa);
571
572 switch (sa->sa.sa_family) {
573
574 case AF_INET:
575 *ret_port = be16toh(sa->in.sin_port);
576 return 0;
577
578 case AF_INET6:
579 *ret_port = be16toh(sa->in6.sin6_port);
580 return 0;
581
582 case AF_VSOCK:
583 *ret_port = sa->vm.svm_port;
584 return 0;
585
586 default:
587 return -EAFNOSUPPORT;
588 }
589 }
590
591 int sockaddr_pretty(
592 const struct sockaddr *_sa,
593 socklen_t salen,
594 bool translate_ipv6,
595 bool include_port,
596 char **ret) {
597
598 union sockaddr_union *sa = (union sockaddr_union*) _sa;
599 char *p;
600 int r;
601
602 assert(sa);
603 assert(salen >= sizeof(sa->sa.sa_family));
604
605 switch (sa->sa.sa_family) {
606
607 case AF_INET: {
608 uint32_t a;
609
610 a = be32toh(sa->in.sin_addr.s_addr);
611
612 if (include_port)
613 r = asprintf(&p,
614 "%u.%u.%u.%u:%u",
615 a >> 24, (a >> 16) & 0xFF, (a >> 8) & 0xFF, a & 0xFF,
616 be16toh(sa->in.sin_port));
617 else
618 r = asprintf(&p,
619 "%u.%u.%u.%u",
620 a >> 24, (a >> 16) & 0xFF, (a >> 8) & 0xFF, a & 0xFF);
621 if (r < 0)
622 return -ENOMEM;
623 break;
624 }
625
626 case AF_INET6: {
627 static const unsigned char ipv4_prefix[] = {
628 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0xFF, 0xFF
629 };
630
631 if (translate_ipv6 &&
632 memcmp(&sa->in6.sin6_addr, ipv4_prefix, sizeof(ipv4_prefix)) == 0) {
633 const uint8_t *a = sa->in6.sin6_addr.s6_addr+12;
634 if (include_port)
635 r = asprintf(&p,
636 "%u.%u.%u.%u:%u",
637 a[0], a[1], a[2], a[3],
638 be16toh(sa->in6.sin6_port));
639 else
640 r = asprintf(&p,
641 "%u.%u.%u.%u",
642 a[0], a[1], a[2], a[3]);
643 if (r < 0)
644 return -ENOMEM;
645 } else {
646 char a[INET6_ADDRSTRLEN];
647
648 inet_ntop(AF_INET6, &sa->in6.sin6_addr, a, sizeof(a));
649
650 if (include_port) {
651 r = asprintf(&p,
652 "[%s]:%u",
653 a,
654 be16toh(sa->in6.sin6_port));
655 if (r < 0)
656 return -ENOMEM;
657 } else {
658 p = strdup(a);
659 if (!p)
660 return -ENOMEM;
661 }
662 }
663
664 break;
665 }
666
667 case AF_UNIX:
668 if (salen <= offsetof(struct sockaddr_un, sun_path) ||
669 (sa->un.sun_path[0] == 0 && salen == offsetof(struct sockaddr_un, sun_path) + 1))
670 /* The name must have at least one character (and the leading NUL does not count) */
671 p = strdup("<unnamed>");
672 else {
673 /* Note that we calculate the path pointer here through the .un_buffer[] field, in order to
674 * outtrick bounds checking tools such as ubsan, which are too smart for their own good: on
675 * Linux the kernel may return sun_path[] data one byte longer than the declared size of the
676 * field. */
677 char *path = (char*) sa->un_buffer + offsetof(struct sockaddr_un, sun_path);
678 size_t path_len = salen - offsetof(struct sockaddr_un, sun_path);
679
680 if (path[0] == 0) {
681 /* Abstract socket. When parsing address information from, we
682 * explicitly reject overly long paths and paths with embedded NULs.
683 * But we might get such a socket from the outside. Let's return
684 * something meaningful and printable in this case. */
685
686 _cleanup_free_ char *e = NULL;
687
688 e = cescape_length(path + 1, path_len - 1);
689 if (!e)
690 return -ENOMEM;
691
692 p = strjoin("@", e);
693 } else {
694 if (path[path_len - 1] == '\0')
695 /* We expect a terminating NUL and don't print it */
696 path_len --;
697
698 p = cescape_length(path, path_len);
699 }
700 }
701 if (!p)
702 return -ENOMEM;
703
704 break;
705
706 case AF_VSOCK:
707 if (include_port) {
708 if (sa->vm.svm_cid == VMADDR_CID_ANY)
709 r = asprintf(&p, "vsock::%u", sa->vm.svm_port);
710 else
711 r = asprintf(&p, "vsock:%u:%u", sa->vm.svm_cid, sa->vm.svm_port);
712 } else
713 r = asprintf(&p, "vsock:%u", sa->vm.svm_cid);
714 if (r < 0)
715 return -ENOMEM;
716 break;
717
718 default:
719 return -EOPNOTSUPP;
720 }
721
722 *ret = p;
723 return 0;
724 }
725
726 int getpeername_pretty(int fd, bool include_port, char **ret) {
727 union sockaddr_union sa;
728 socklen_t salen = sizeof(sa);
729 int r;
730
731 assert(fd >= 0);
732 assert(ret);
733
734 if (getpeername(fd, &sa.sa, &salen) < 0)
735 return -errno;
736
737 if (sa.sa.sa_family == AF_UNIX) {
738 struct ucred ucred = {};
739
740 /* UNIX connection sockets are anonymous, so let's use
741 * PID/UID as pretty credentials instead */
742
743 r = getpeercred(fd, &ucred);
744 if (r < 0)
745 return r;
746
747 if (asprintf(ret, "PID "PID_FMT"/UID "UID_FMT, ucred.pid, ucred.uid) < 0)
748 return -ENOMEM;
749
750 return 0;
751 }
752
753 /* For remote sockets we translate IPv6 addresses back to IPv4
754 * if applicable, since that's nicer. */
755
756 return sockaddr_pretty(&sa.sa, salen, true, include_port, ret);
757 }
758
759 int getsockname_pretty(int fd, char **ret) {
760 union sockaddr_union sa;
761 socklen_t salen = sizeof(sa);
762
763 assert(fd >= 0);
764 assert(ret);
765
766 if (getsockname(fd, &sa.sa, &salen) < 0)
767 return -errno;
768
769 /* For local sockets we do not translate IPv6 addresses back
770 * to IPv6 if applicable, since this is usually used for
771 * listening sockets where the difference between IPv4 and
772 * IPv6 matters. */
773
774 return sockaddr_pretty(&sa.sa, salen, false, true, ret);
775 }
776
777 int socknameinfo_pretty(union sockaddr_union *sa, socklen_t salen, char **_ret) {
778 int r;
779 char host[NI_MAXHOST], *ret;
780
781 assert(_ret);
782
783 r = getnameinfo(&sa->sa, salen, host, sizeof(host), NULL, 0, IDN_FLAGS);
784 if (r != 0) {
785 int saved_errno = errno;
786
787 r = sockaddr_pretty(&sa->sa, salen, true, true, &ret);
788 if (r < 0)
789 return r;
790
791 log_debug_errno(saved_errno, "getnameinfo(%s) failed: %m", ret);
792 } else {
793 ret = strdup(host);
794 if (!ret)
795 return -ENOMEM;
796 }
797
798 *_ret = ret;
799 return 0;
800 }
801
802 static const char* const netlink_family_table[] = {
803 [NETLINK_ROUTE] = "route",
804 [NETLINK_FIREWALL] = "firewall",
805 [NETLINK_INET_DIAG] = "inet-diag",
806 [NETLINK_NFLOG] = "nflog",
807 [NETLINK_XFRM] = "xfrm",
808 [NETLINK_SELINUX] = "selinux",
809 [NETLINK_ISCSI] = "iscsi",
810 [NETLINK_AUDIT] = "audit",
811 [NETLINK_FIB_LOOKUP] = "fib-lookup",
812 [NETLINK_CONNECTOR] = "connector",
813 [NETLINK_NETFILTER] = "netfilter",
814 [NETLINK_IP6_FW] = "ip6-fw",
815 [NETLINK_DNRTMSG] = "dnrtmsg",
816 [NETLINK_KOBJECT_UEVENT] = "kobject-uevent",
817 [NETLINK_GENERIC] = "generic",
818 [NETLINK_SCSITRANSPORT] = "scsitransport",
819 [NETLINK_ECRYPTFS] = "ecryptfs",
820 [NETLINK_RDMA] = "rdma",
821 };
822
823 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(netlink_family, int, INT_MAX);
824
825 static const char* const socket_address_bind_ipv6_only_table[_SOCKET_ADDRESS_BIND_IPV6_ONLY_MAX] = {
826 [SOCKET_ADDRESS_DEFAULT] = "default",
827 [SOCKET_ADDRESS_BOTH] = "both",
828 [SOCKET_ADDRESS_IPV6_ONLY] = "ipv6-only"
829 };
830
831 DEFINE_STRING_TABLE_LOOKUP(socket_address_bind_ipv6_only, SocketAddressBindIPv6Only);
832
833 SocketAddressBindIPv6Only socket_address_bind_ipv6_only_or_bool_from_string(const char *n) {
834 int r;
835
836 r = parse_boolean(n);
837 if (r > 0)
838 return SOCKET_ADDRESS_IPV6_ONLY;
839 if (r == 0)
840 return SOCKET_ADDRESS_BOTH;
841
842 return socket_address_bind_ipv6_only_from_string(n);
843 }
844
845 bool sockaddr_equal(const union sockaddr_union *a, const union sockaddr_union *b) {
846 assert(a);
847 assert(b);
848
849 if (a->sa.sa_family != b->sa.sa_family)
850 return false;
851
852 if (a->sa.sa_family == AF_INET)
853 return a->in.sin_addr.s_addr == b->in.sin_addr.s_addr;
854
855 if (a->sa.sa_family == AF_INET6)
856 return memcmp(&a->in6.sin6_addr, &b->in6.sin6_addr, sizeof(a->in6.sin6_addr)) == 0;
857
858 if (a->sa.sa_family == AF_VSOCK)
859 return a->vm.svm_cid == b->vm.svm_cid;
860
861 return false;
862 }
863
864 int fd_inc_sndbuf(int fd, size_t n) {
865 int r, value;
866 socklen_t l = sizeof(value);
867
868 r = getsockopt(fd, SOL_SOCKET, SO_SNDBUF, &value, &l);
869 if (r >= 0 && l == sizeof(value) && (size_t) value >= n*2)
870 return 0;
871
872 /* If we have the privileges we will ignore the kernel limit. */
873
874 if (setsockopt_int(fd, SOL_SOCKET, SO_SNDBUF, n) < 0) {
875 r = setsockopt_int(fd, SOL_SOCKET, SO_SNDBUFFORCE, n);
876 if (r < 0)
877 return r;
878 }
879
880 return 1;
881 }
882
883 int fd_inc_rcvbuf(int fd, size_t n) {
884 int r, value;
885 socklen_t l = sizeof(value);
886
887 r = getsockopt(fd, SOL_SOCKET, SO_RCVBUF, &value, &l);
888 if (r >= 0 && l == sizeof(value) && (size_t) value >= n*2)
889 return 0;
890
891 /* If we have the privileges we will ignore the kernel limit. */
892
893 if (setsockopt_int(fd, SOL_SOCKET, SO_RCVBUF, n) < 0) {
894 r = setsockopt_int(fd, SOL_SOCKET, SO_RCVBUFFORCE, n);
895 if (r < 0)
896 return r;
897 }
898
899 return 1;
900 }
901
902 static const char* const ip_tos_table[] = {
903 [IPTOS_LOWDELAY] = "low-delay",
904 [IPTOS_THROUGHPUT] = "throughput",
905 [IPTOS_RELIABILITY] = "reliability",
906 [IPTOS_LOWCOST] = "low-cost",
907 };
908
909 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(ip_tos, int, 0xff);
910
911 bool ifname_valid(const char *p) {
912 bool numeric = true;
913
914 /* Checks whether a network interface name is valid. This is inspired by dev_valid_name() in the kernel sources
915 * but slightly stricter, as we only allow non-control, non-space ASCII characters in the interface name. We
916 * also don't permit names that only container numbers, to avoid confusion with numeric interface indexes. */
917
918 if (isempty(p))
919 return false;
920
921 if (strlen(p) >= IFNAMSIZ)
922 return false;
923
924 if (dot_or_dot_dot(p))
925 return false;
926
927 while (*p) {
928 if ((unsigned char) *p >= 127U)
929 return false;
930
931 if ((unsigned char) *p <= 32U)
932 return false;
933
934 if (IN_SET(*p, ':', '/'))
935 return false;
936
937 numeric = numeric && (*p >= '0' && *p <= '9');
938 p++;
939 }
940
941 if (numeric)
942 return false;
943
944 return true;
945 }
946
947 bool address_label_valid(const char *p) {
948
949 if (isempty(p))
950 return false;
951
952 if (strlen(p) >= IFNAMSIZ)
953 return false;
954
955 while (*p) {
956 if ((uint8_t) *p >= 127U)
957 return false;
958
959 if ((uint8_t) *p <= 31U)
960 return false;
961 p++;
962 }
963
964 return true;
965 }
966
967 int getpeercred(int fd, struct ucred *ucred) {
968 socklen_t n = sizeof(struct ucred);
969 struct ucred u;
970 int r;
971
972 assert(fd >= 0);
973 assert(ucred);
974
975 r = getsockopt(fd, SOL_SOCKET, SO_PEERCRED, &u, &n);
976 if (r < 0)
977 return -errno;
978
979 if (n != sizeof(struct ucred))
980 return -EIO;
981
982 /* Check if the data is actually useful and not suppressed due to namespacing issues */
983 if (!pid_is_valid(u.pid))
984 return -ENODATA;
985
986 /* Note that we don't check UID/GID here, as namespace translation works differently there: instead of
987 * receiving in "invalid" user/group we get the overflow UID/GID. */
988
989 *ucred = u;
990 return 0;
991 }
992
993 int getpeersec(int fd, char **ret) {
994 _cleanup_free_ char *s = NULL;
995 socklen_t n = 64;
996
997 assert(fd >= 0);
998 assert(ret);
999
1000 for (;;) {
1001 s = new0(char, n+1);
1002 if (!s)
1003 return -ENOMEM;
1004
1005 if (getsockopt(fd, SOL_SOCKET, SO_PEERSEC, s, &n) >= 0)
1006 break;
1007
1008 if (errno != ERANGE)
1009 return -errno;
1010
1011 s = mfree(s);
1012 }
1013
1014 if (isempty(s))
1015 return -EOPNOTSUPP;
1016
1017 *ret = TAKE_PTR(s);
1018
1019 return 0;
1020 }
1021
1022 int getpeergroups(int fd, gid_t **ret) {
1023 socklen_t n = sizeof(gid_t) * 64;
1024 _cleanup_free_ gid_t *d = NULL;
1025
1026 assert(fd >= 0);
1027 assert(ret);
1028
1029 for (;;) {
1030 d = malloc(n);
1031 if (!d)
1032 return -ENOMEM;
1033
1034 if (getsockopt(fd, SOL_SOCKET, SO_PEERGROUPS, d, &n) >= 0)
1035 break;
1036
1037 if (errno != ERANGE)
1038 return -errno;
1039
1040 d = mfree(d);
1041 }
1042
1043 assert_se(n % sizeof(gid_t) == 0);
1044 n /= sizeof(gid_t);
1045
1046 if ((socklen_t) (int) n != n)
1047 return -E2BIG;
1048
1049 *ret = TAKE_PTR(d);
1050
1051 return (int) n;
1052 }
1053
1054 ssize_t send_one_fd_iov_sa(
1055 int transport_fd,
1056 int fd,
1057 struct iovec *iov, size_t iovlen,
1058 const struct sockaddr *sa, socklen_t len,
1059 int flags) {
1060
1061 union {
1062 struct cmsghdr cmsghdr;
1063 uint8_t buf[CMSG_SPACE(sizeof(int))];
1064 } control = {};
1065 struct msghdr mh = {
1066 .msg_name = (struct sockaddr*) sa,
1067 .msg_namelen = len,
1068 .msg_iov = iov,
1069 .msg_iovlen = iovlen,
1070 };
1071 ssize_t k;
1072
1073 assert(transport_fd >= 0);
1074
1075 /*
1076 * We need either an FD or data to send.
1077 * If there's nothing, return an error.
1078 */
1079 if (fd < 0 && !iov)
1080 return -EINVAL;
1081
1082 if (fd >= 0) {
1083 struct cmsghdr *cmsg;
1084
1085 mh.msg_control = &control;
1086 mh.msg_controllen = sizeof(control);
1087
1088 cmsg = CMSG_FIRSTHDR(&mh);
1089 cmsg->cmsg_level = SOL_SOCKET;
1090 cmsg->cmsg_type = SCM_RIGHTS;
1091 cmsg->cmsg_len = CMSG_LEN(sizeof(int));
1092 memcpy(CMSG_DATA(cmsg), &fd, sizeof(int));
1093
1094 mh.msg_controllen = CMSG_SPACE(sizeof(int));
1095 }
1096 k = sendmsg(transport_fd, &mh, MSG_NOSIGNAL | flags);
1097 if (k < 0)
1098 return (ssize_t) -errno;
1099
1100 return k;
1101 }
1102
1103 int send_one_fd_sa(
1104 int transport_fd,
1105 int fd,
1106 const struct sockaddr *sa, socklen_t len,
1107 int flags) {
1108
1109 assert(fd >= 0);
1110
1111 return (int) send_one_fd_iov_sa(transport_fd, fd, NULL, 0, sa, len, flags);
1112 }
1113
1114 ssize_t receive_one_fd_iov(
1115 int transport_fd,
1116 struct iovec *iov, size_t iovlen,
1117 int flags,
1118 int *ret_fd) {
1119
1120 union {
1121 struct cmsghdr cmsghdr;
1122 uint8_t buf[CMSG_SPACE(sizeof(int))];
1123 } control = {};
1124 struct msghdr mh = {
1125 .msg_control = &control,
1126 .msg_controllen = sizeof(control),
1127 .msg_iov = iov,
1128 .msg_iovlen = iovlen,
1129 };
1130 struct cmsghdr *cmsg, *found = NULL;
1131 ssize_t k;
1132
1133 assert(transport_fd >= 0);
1134 assert(ret_fd);
1135
1136 /*
1137 * Receive a single FD via @transport_fd. We don't care for
1138 * the transport-type. We retrieve a single FD at most, so for
1139 * packet-based transports, the caller must ensure to send
1140 * only a single FD per packet. This is best used in
1141 * combination with send_one_fd().
1142 */
1143
1144 k = recvmsg(transport_fd, &mh, MSG_CMSG_CLOEXEC | flags);
1145 if (k < 0)
1146 return (ssize_t) -errno;
1147
1148 CMSG_FOREACH(cmsg, &mh) {
1149 if (cmsg->cmsg_level == SOL_SOCKET &&
1150 cmsg->cmsg_type == SCM_RIGHTS &&
1151 cmsg->cmsg_len == CMSG_LEN(sizeof(int))) {
1152 assert(!found);
1153 found = cmsg;
1154 break;
1155 }
1156 }
1157
1158 if (!found)
1159 cmsg_close_all(&mh);
1160
1161 /* If didn't receive an FD or any data, return an error. */
1162 if (k == 0 && !found)
1163 return -EIO;
1164
1165 if (found)
1166 *ret_fd = *(int*) CMSG_DATA(found);
1167 else
1168 *ret_fd = -1;
1169
1170 return k;
1171 }
1172
1173 int receive_one_fd(int transport_fd, int flags) {
1174 int fd;
1175 ssize_t k;
1176
1177 k = receive_one_fd_iov(transport_fd, NULL, 0, flags, &fd);
1178 if (k == 0)
1179 return fd;
1180
1181 /* k must be negative, since receive_one_fd_iov() only returns
1182 * a positive value if data was received through the iov. */
1183 assert(k < 0);
1184 return (int) k;
1185 }
1186
1187 ssize_t next_datagram_size_fd(int fd) {
1188 ssize_t l;
1189 int k;
1190
1191 /* This is a bit like FIONREAD/SIOCINQ, however a bit more powerful. The difference being: recv(MSG_PEEK) will
1192 * actually cause the next datagram in the queue to be validated regarding checksums, which FIONREAD doesn't
1193 * do. This difference is actually of major importance as we need to be sure that the size returned here
1194 * actually matches what we will read with recvmsg() next, as otherwise we might end up allocating a buffer of
1195 * the wrong size. */
1196
1197 l = recv(fd, NULL, 0, MSG_PEEK|MSG_TRUNC);
1198 if (l < 0) {
1199 if (IN_SET(errno, EOPNOTSUPP, EFAULT))
1200 goto fallback;
1201
1202 return -errno;
1203 }
1204 if (l == 0)
1205 goto fallback;
1206
1207 return l;
1208
1209 fallback:
1210 k = 0;
1211
1212 /* Some sockets (AF_PACKET) do not support null-sized recv() with MSG_TRUNC set, let's fall back to FIONREAD
1213 * for them. Checksums don't matter for raw sockets anyway, hence this should be fine. */
1214
1215 if (ioctl(fd, FIONREAD, &k) < 0)
1216 return -errno;
1217
1218 return (ssize_t) k;
1219 }
1220
1221 int flush_accept(int fd) {
1222
1223 struct pollfd pollfd = {
1224 .fd = fd,
1225 .events = POLLIN,
1226 };
1227 int r;
1228
1229 /* Similar to flush_fd() but flushes all incoming connection by accepting them and immediately closing them. */
1230
1231 for (;;) {
1232 int cfd;
1233
1234 r = poll(&pollfd, 1, 0);
1235 if (r < 0) {
1236 if (errno == EINTR)
1237 continue;
1238
1239 return -errno;
1240
1241 } else if (r == 0)
1242 return 0;
1243
1244 cfd = accept4(fd, NULL, NULL, SOCK_NONBLOCK|SOCK_CLOEXEC);
1245 if (cfd < 0) {
1246 if (errno == EINTR)
1247 continue;
1248
1249 if (errno == EAGAIN)
1250 return 0;
1251
1252 return -errno;
1253 }
1254
1255 close(cfd);
1256 }
1257 }
1258
1259 struct cmsghdr* cmsg_find(struct msghdr *mh, int level, int type, socklen_t length) {
1260 struct cmsghdr *cmsg;
1261
1262 assert(mh);
1263
1264 CMSG_FOREACH(cmsg, mh)
1265 if (cmsg->cmsg_level == level &&
1266 cmsg->cmsg_type == type &&
1267 (length == (socklen_t) -1 || length == cmsg->cmsg_len))
1268 return cmsg;
1269
1270 return NULL;
1271 }
1272
1273 int socket_ioctl_fd(void) {
1274 int fd;
1275
1276 /* Create a socket to invoke the various network interface ioctl()s on. Traditionally only AF_INET was good for
1277 * that. Since kernel 4.6 AF_NETLINK works for this too. We first try to use AF_INET hence, but if that's not
1278 * available (for example, because it is made unavailable via SECCOMP or such), we'll fall back to the more
1279 * generic AF_NETLINK. */
1280
1281 fd = socket(AF_INET, SOCK_DGRAM|SOCK_CLOEXEC, 0);
1282 if (fd < 0)
1283 fd = socket(AF_NETLINK, SOCK_RAW|SOCK_CLOEXEC, NETLINK_GENERIC);
1284 if (fd < 0)
1285 return -errno;
1286
1287 return fd;
1288 }
1289
1290 int sockaddr_un_unlink(const struct sockaddr_un *sa) {
1291 const char *p, * nul;
1292
1293 assert(sa);
1294
1295 if (sa->sun_family != AF_UNIX)
1296 return -EPROTOTYPE;
1297
1298 if (sa->sun_path[0] == 0) /* Nothing to do for abstract sockets */
1299 return 0;
1300
1301 /* The path in .sun_path is not necessarily NUL terminated. Let's fix that. */
1302 nul = memchr(sa->sun_path, 0, sizeof(sa->sun_path));
1303 if (nul)
1304 p = sa->sun_path;
1305 else
1306 p = memdupa_suffix0(sa->sun_path, sizeof(sa->sun_path));
1307
1308 if (unlink(p) < 0)
1309 return -errno;
1310
1311 return 1;
1312 }
1313
1314 int sockaddr_un_set_path(struct sockaddr_un *ret, const char *path) {
1315 size_t l;
1316
1317 assert(ret);
1318 assert(path);
1319
1320 /* Initialize ret->sun_path from the specified argument. This will interpret paths starting with '@' as
1321 * abstract namespace sockets, and those starting with '/' as regular filesystem sockets. It won't accept
1322 * anything else (i.e. no relative paths), to avoid ambiguities. Note that this function cannot be used to
1323 * reference paths in the abstract namespace that include NUL bytes in the name. */
1324
1325 l = strlen(path);
1326 if (l == 0)
1327 return -EINVAL;
1328 if (!IN_SET(path[0], '/', '@'))
1329 return -EINVAL;
1330 if (path[1] == 0)
1331 return -EINVAL;
1332
1333 /* Don't allow paths larger than the space in sockaddr_un. Note that we are a tiny bit more restrictive than
1334 * the kernel is: we insist on NUL termination (both for abstract namespace and regular file system socket
1335 * addresses!), which the kernel doesn't. We do this to reduce chance of incompatibility with other apps that
1336 * do not expect non-NUL terminated file system path*/
1337 if (l+1 > sizeof(ret->sun_path))
1338 return -EINVAL;
1339
1340 *ret = (struct sockaddr_un) {
1341 .sun_family = AF_UNIX,
1342 };
1343
1344 if (path[0] == '@') {
1345 /* Abstract namespace socket */
1346 memcpy(ret->sun_path + 1, path + 1, l); /* copy *with* trailing NUL byte */
1347 return (int) (offsetof(struct sockaddr_un, sun_path) + l); /* 🔥 *don't* 🔥 include trailing NUL in size */
1348
1349 } else {
1350 assert(path[0] == '/');
1351
1352 /* File system socket */
1353 memcpy(ret->sun_path, path, l + 1); /* copy *with* trailing NUL byte */
1354 return (int) (offsetof(struct sockaddr_un, sun_path) + l + 1); /* include trailing NUL in size */
1355 }
1356 }
1357
1358 int socket_bind_to_ifname(int fd, const char *ifname) {
1359 assert(fd >= 0);
1360
1361 /* Call with NULL to drop binding */
1362
1363 if (setsockopt(fd, SOL_SOCKET, SO_BINDTODEVICE, ifname, strlen_ptr(ifname)) < 0)
1364 return -errno;
1365
1366 return 0;
1367 }
1368
1369 int socket_bind_to_ifindex(int fd, int ifindex) {
1370 char ifname[IFNAMSIZ] = "";
1371
1372 assert(fd >= 0);
1373
1374 if (ifindex <= 0) {
1375 /* Drop binding */
1376 if (setsockopt(fd, SOL_SOCKET, SO_BINDTODEVICE, NULL, 0) < 0)
1377 return -errno;
1378
1379 return 0;
1380 }
1381
1382 if (setsockopt(fd, SOL_SOCKET, SO_BINDTOIFINDEX, &ifindex, sizeof(ifindex)) >= 0)
1383 return 0;
1384 if (errno != ENOPROTOOPT)
1385 return -errno;
1386
1387 /* Fall back to SO_BINDTODEVICE on kernels < 5.0 which didn't have SO_BINDTOIFINDEX */
1388 if (!if_indextoname(ifindex, ifname))
1389 return -errno;
1390
1391 return socket_bind_to_ifname(fd, ifname);
1392 }
Unnamed repository; edit this file 'description' to name the repository.