]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/basic/socket-util.c
cgroup: rework which files we chown() on delegation
[thirdparty/systemd.git] / src / basic / socket-util.c
1 /***
2 This file is part of systemd.
3
4 Copyright 2010 Lennart Poettering
5
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
10
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
15
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
18 ***/
19
20 #include <arpa/inet.h>
21 #include <errno.h>
22 #include <limits.h>
23 #include <net/if.h>
24 #include <netdb.h>
25 #include <netinet/ip.h>
26 #include <poll.h>
27 #include <stddef.h>
28 #include <stdint.h>
29 #include <stdio.h>
30 #include <stdlib.h>
31 #include <string.h>
32 #include <unistd.h>
33
34 #include "alloc-util.h"
35 #include "fd-util.h"
36 #include "fileio.h"
37 #include "format-util.h"
38 #include "log.h"
39 #include "macro.h"
40 #include "missing.h"
41 #include "parse-util.h"
42 #include "path-util.h"
43 #include "socket-util.h"
44 #include "string-table.h"
45 #include "string-util.h"
46 #include "strv.h"
47 #include "user-util.h"
48 #include "utf8.h"
49 #include "util.h"
50
51 #ifdef ENABLE_IDN
52 # define IDN_FLAGS (NI_IDN|NI_IDN_USE_STD3_ASCII_RULES)
53 #else
54 # define IDN_FLAGS 0
55 #endif
56
57 int socket_address_parse(SocketAddress *a, const char *s) {
58 char *e, *n;
59 unsigned u;
60 int r;
61
62 assert(a);
63 assert(s);
64
65 zero(*a);
66 a->type = SOCK_STREAM;
67
68 if (*s == '[') {
69 /* IPv6 in [x:.....:z]:p notation */
70
71 e = strchr(s+1, ']');
72 if (!e)
73 return -EINVAL;
74
75 n = strndupa(s+1, e-s-1);
76
77 errno = 0;
78 if (inet_pton(AF_INET6, n, &a->sockaddr.in6.sin6_addr) <= 0)
79 return errno > 0 ? -errno : -EINVAL;
80
81 e++;
82 if (*e != ':')
83 return -EINVAL;
84
85 e++;
86 r = safe_atou(e, &u);
87 if (r < 0)
88 return r;
89
90 if (u <= 0 || u > 0xFFFF)
91 return -EINVAL;
92
93 a->sockaddr.in6.sin6_family = AF_INET6;
94 a->sockaddr.in6.sin6_port = htobe16((uint16_t)u);
95 a->size = sizeof(struct sockaddr_in6);
96
97 } else if (*s == '/') {
98 /* AF_UNIX socket */
99
100 size_t l;
101
102 l = strlen(s);
103 if (l >= sizeof(a->sockaddr.un.sun_path))
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)
116 return -EINVAL;
117
118 a->sockaddr.un.sun_family = AF_UNIX;
119 memcpy(a->sockaddr.un.sun_path+1, s+1, l);
120 a->size = offsetof(struct sockaddr_un, sun_path) + 1 + l;
121
122 } else if (startswith(s, "vsock:")) {
123 /* AF_VSOCK socket in vsock:cid:port notation */
124 const char *cid_start = s + strlen("vsock:");
125
126 e = strchr(cid_start, ':');
127 if (!e)
128 return -EINVAL;
129
130 r = safe_atou(e+1, &u);
131 if (r < 0)
132 return r;
133
134 n = strndupa(cid_start, e - cid_start);
135 if (!isempty(n)) {
136 r = safe_atou(n, &a->sockaddr.vm.svm_cid);
137 if (r < 0)
138 return r;
139 } else
140 a->sockaddr.vm.svm_cid = VMADDR_CID_ANY;
141
142 a->sockaddr.vm.svm_family = AF_VSOCK;
143 a->sockaddr.vm.svm_port = u;
144 a->size = sizeof(struct sockaddr_vm);
145
146 } else {
147 e = strchr(s, ':');
148 if (e) {
149 r = safe_atou(e+1, &u);
150 if (r < 0)
151 return r;
152
153 if (u <= 0 || u > 0xFFFF)
154 return -EINVAL;
155
156 n = strndupa(s, e-s);
157
158 /* IPv4 in w.x.y.z:p notation? */
159 r = inet_pton(AF_INET, n, &a->sockaddr.in.sin_addr);
160 if (r < 0)
161 return -errno;
162
163 if (r > 0) {
164 /* Gotcha, it's a traditional IPv4 address */
165 a->sockaddr.in.sin_family = AF_INET;
166 a->sockaddr.in.sin_port = htobe16((uint16_t)u);
167 a->size = sizeof(struct sockaddr_in);
168 } else {
169 unsigned idx;
170
171 if (strlen(n) > IF_NAMESIZE-1)
172 return -EINVAL;
173
174 /* Uh, our last resort, an interface name */
175 idx = if_nametoindex(n);
176 if (idx == 0)
177 return -EINVAL;
178
179 a->sockaddr.in6.sin6_family = AF_INET6;
180 a->sockaddr.in6.sin6_port = htobe16((uint16_t)u);
181 a->sockaddr.in6.sin6_scope_id = idx;
182 a->sockaddr.in6.sin6_addr = in6addr_any;
183 a->size = sizeof(struct sockaddr_in6);
184 }
185 } else {
186
187 /* Just a port */
188 r = safe_atou(s, &u);
189 if (r < 0)
190 return r;
191
192 if (u <= 0 || u > 0xFFFF)
193 return -EINVAL;
194
195 if (socket_ipv6_is_supported()) {
196 a->sockaddr.in6.sin6_family = AF_INET6;
197 a->sockaddr.in6.sin6_port = htobe16((uint16_t)u);
198 a->sockaddr.in6.sin6_addr = in6addr_any;
199 a->size = sizeof(struct sockaddr_in6);
200 } else {
201 a->sockaddr.in.sin_family = AF_INET;
202 a->sockaddr.in.sin_port = htobe16((uint16_t)u);
203 a->sockaddr.in.sin_addr.s_addr = INADDR_ANY;
204 a->size = sizeof(struct sockaddr_in);
205 }
206 }
207 }
208
209 return 0;
210 }
211
212 int socket_address_parse_and_warn(SocketAddress *a, const char *s) {
213 SocketAddress b;
214 int r;
215
216 /* Similar to socket_address_parse() but warns for IPv6 sockets when we don't support them. */
217
218 r = socket_address_parse(&b, s);
219 if (r < 0)
220 return r;
221
222 if (!socket_ipv6_is_supported() && b.sockaddr.sa.sa_family == AF_INET6) {
223 log_warning("Binding to IPv6 address not available since kernel does not support IPv6.");
224 return -EAFNOSUPPORT;
225 }
226
227 *a = b;
228 return 0;
229 }
230
231 int socket_address_parse_netlink(SocketAddress *a, const char *s) {
232 int family;
233 unsigned group = 0;
234 _cleanup_free_ char *sfamily = NULL;
235 assert(a);
236 assert(s);
237
238 zero(*a);
239 a->type = SOCK_RAW;
240
241 errno = 0;
242 if (sscanf(s, "%ms %u", &sfamily, &group) < 1)
243 return errno > 0 ? -errno : -EINVAL;
244
245 family = netlink_family_from_string(sfamily);
246 if (family < 0)
247 return -EINVAL;
248
249 a->sockaddr.nl.nl_family = AF_NETLINK;
250 a->sockaddr.nl.nl_groups = group;
251
252 a->type = SOCK_RAW;
253 a->size = sizeof(struct sockaddr_nl);
254 a->protocol = family;
255
256 return 0;
257 }
258
259 int socket_address_verify(const SocketAddress *a) {
260 assert(a);
261
262 switch (socket_address_family(a)) {
263
264 case AF_INET:
265 if (a->size != sizeof(struct sockaddr_in))
266 return -EINVAL;
267
268 if (a->sockaddr.in.sin_port == 0)
269 return -EINVAL;
270
271 if (a->type != SOCK_STREAM && a->type != SOCK_DGRAM)
272 return -EINVAL;
273
274 return 0;
275
276 case AF_INET6:
277 if (a->size != sizeof(struct sockaddr_in6))
278 return -EINVAL;
279
280 if (a->sockaddr.in6.sin6_port == 0)
281 return -EINVAL;
282
283 if (a->type != SOCK_STREAM && a->type != SOCK_DGRAM)
284 return -EINVAL;
285
286 return 0;
287
288 case AF_UNIX:
289 if (a->size < offsetof(struct sockaddr_un, sun_path))
290 return -EINVAL;
291
292 if (a->size > offsetof(struct sockaddr_un, sun_path)) {
293
294 if (a->sockaddr.un.sun_path[0] != 0) {
295 char *e;
296
297 /* path */
298 e = memchr(a->sockaddr.un.sun_path, 0, sizeof(a->sockaddr.un.sun_path));
299 if (!e)
300 return -EINVAL;
301
302 if (a->size != offsetof(struct sockaddr_un, sun_path) + (e - a->sockaddr.un.sun_path) + 1)
303 return -EINVAL;
304 }
305 }
306
307 if (a->type != SOCK_STREAM && a->type != SOCK_DGRAM && a->type != SOCK_SEQPACKET)
308 return -EINVAL;
309
310 return 0;
311
312 case AF_NETLINK:
313
314 if (a->size != sizeof(struct sockaddr_nl))
315 return -EINVAL;
316
317 if (a->type != SOCK_RAW && a->type != SOCK_DGRAM)
318 return -EINVAL;
319
320 return 0;
321
322 case AF_VSOCK:
323 if (a->size != sizeof(struct sockaddr_vm))
324 return -EINVAL;
325
326 if (a->type != SOCK_STREAM && a->type != SOCK_DGRAM)
327 return -EINVAL;
328
329 return 0;
330
331 default:
332 return -EAFNOSUPPORT;
333 }
334 }
335
336 int socket_address_print(const SocketAddress *a, char **ret) {
337 int r;
338
339 assert(a);
340 assert(ret);
341
342 r = socket_address_verify(a);
343 if (r < 0)
344 return r;
345
346 if (socket_address_family(a) == AF_NETLINK) {
347 _cleanup_free_ char *sfamily = NULL;
348
349 r = netlink_family_to_string_alloc(a->protocol, &sfamily);
350 if (r < 0)
351 return r;
352
353 r = asprintf(ret, "%s %u", sfamily, a->sockaddr.nl.nl_groups);
354 if (r < 0)
355 return -ENOMEM;
356
357 return 0;
358 }
359
360 return sockaddr_pretty(&a->sockaddr.sa, a->size, false, true, ret);
361 }
362
363 bool socket_address_can_accept(const SocketAddress *a) {
364 assert(a);
365
366 return
367 a->type == SOCK_STREAM ||
368 a->type == SOCK_SEQPACKET;
369 }
370
371 bool socket_address_equal(const SocketAddress *a, const SocketAddress *b) {
372 assert(a);
373 assert(b);
374
375 /* Invalid addresses are unequal to all */
376 if (socket_address_verify(a) < 0 ||
377 socket_address_verify(b) < 0)
378 return false;
379
380 if (a->type != b->type)
381 return false;
382
383 if (socket_address_family(a) != socket_address_family(b))
384 return false;
385
386 switch (socket_address_family(a)) {
387
388 case AF_INET:
389 if (a->sockaddr.in.sin_addr.s_addr != b->sockaddr.in.sin_addr.s_addr)
390 return false;
391
392 if (a->sockaddr.in.sin_port != b->sockaddr.in.sin_port)
393 return false;
394
395 break;
396
397 case AF_INET6:
398 if (memcmp(&a->sockaddr.in6.sin6_addr, &b->sockaddr.in6.sin6_addr, sizeof(a->sockaddr.in6.sin6_addr)) != 0)
399 return false;
400
401 if (a->sockaddr.in6.sin6_port != b->sockaddr.in6.sin6_port)
402 return false;
403
404 break;
405
406 case AF_UNIX:
407 if (a->size <= offsetof(struct sockaddr_un, sun_path) ||
408 b->size <= offsetof(struct sockaddr_un, sun_path))
409 return false;
410
411 if ((a->sockaddr.un.sun_path[0] == 0) != (b->sockaddr.un.sun_path[0] == 0))
412 return false;
413
414 if (a->sockaddr.un.sun_path[0]) {
415 if (!path_equal_or_files_same(a->sockaddr.un.sun_path, b->sockaddr.un.sun_path, 0))
416 return false;
417 } else {
418 if (a->size != b->size)
419 return false;
420
421 if (memcmp(a->sockaddr.un.sun_path, b->sockaddr.un.sun_path, a->size) != 0)
422 return false;
423 }
424
425 break;
426
427 case AF_NETLINK:
428 if (a->protocol != b->protocol)
429 return false;
430
431 if (a->sockaddr.nl.nl_groups != b->sockaddr.nl.nl_groups)
432 return false;
433
434 break;
435
436 case AF_VSOCK:
437 if (a->sockaddr.vm.svm_cid != b->sockaddr.vm.svm_cid)
438 return false;
439
440 if (a->sockaddr.vm.svm_port != b->sockaddr.vm.svm_port)
441 return false;
442
443 break;
444
445 default:
446 /* Cannot compare, so we assume the addresses are different */
447 return false;
448 }
449
450 return true;
451 }
452
453 bool socket_address_is(const SocketAddress *a, const char *s, int type) {
454 struct SocketAddress b;
455
456 assert(a);
457 assert(s);
458
459 if (socket_address_parse(&b, s) < 0)
460 return false;
461
462 b.type = type;
463
464 return socket_address_equal(a, &b);
465 }
466
467 bool socket_address_is_netlink(const SocketAddress *a, const char *s) {
468 struct SocketAddress b;
469
470 assert(a);
471 assert(s);
472
473 if (socket_address_parse_netlink(&b, s) < 0)
474 return false;
475
476 return socket_address_equal(a, &b);
477 }
478
479 const char* socket_address_get_path(const SocketAddress *a) {
480 assert(a);
481
482 if (socket_address_family(a) != AF_UNIX)
483 return NULL;
484
485 if (a->sockaddr.un.sun_path[0] == 0)
486 return NULL;
487
488 return a->sockaddr.un.sun_path;
489 }
490
491 bool socket_ipv6_is_supported(void) {
492 if (access("/proc/net/if_inet6", F_OK) != 0)
493 return false;
494
495 return true;
496 }
497
498 bool socket_address_matches_fd(const SocketAddress *a, int fd) {
499 SocketAddress b;
500 socklen_t solen;
501
502 assert(a);
503 assert(fd >= 0);
504
505 b.size = sizeof(b.sockaddr);
506 if (getsockname(fd, &b.sockaddr.sa, &b.size) < 0)
507 return false;
508
509 if (b.sockaddr.sa.sa_family != a->sockaddr.sa.sa_family)
510 return false;
511
512 solen = sizeof(b.type);
513 if (getsockopt(fd, SOL_SOCKET, SO_TYPE, &b.type, &solen) < 0)
514 return false;
515
516 if (b.type != a->type)
517 return false;
518
519 if (a->protocol != 0) {
520 solen = sizeof(b.protocol);
521 if (getsockopt(fd, SOL_SOCKET, SO_PROTOCOL, &b.protocol, &solen) < 0)
522 return false;
523
524 if (b.protocol != a->protocol)
525 return false;
526 }
527
528 return socket_address_equal(a, &b);
529 }
530
531 int sockaddr_port(const struct sockaddr *_sa, unsigned *port) {
532 union sockaddr_union *sa = (union sockaddr_union*) _sa;
533
534 assert(sa);
535
536 switch (sa->sa.sa_family) {
537 case AF_INET:
538 *port = be16toh(sa->in.sin_port);
539 return 0;
540
541 case AF_INET6:
542 *port = be16toh(sa->in6.sin6_port);
543 return 0;
544
545 case AF_VSOCK:
546 *port = sa->vm.svm_port;
547 return 0;
548
549 default:
550 return -EAFNOSUPPORT;
551 }
552 }
553
554 int sockaddr_pretty(const struct sockaddr *_sa, socklen_t salen, bool translate_ipv6, bool include_port, char **ret) {
555 union sockaddr_union *sa = (union sockaddr_union*) _sa;
556 char *p;
557 int r;
558
559 assert(sa);
560 assert(salen >= sizeof(sa->sa.sa_family));
561
562 switch (sa->sa.sa_family) {
563
564 case AF_INET: {
565 uint32_t a;
566
567 a = be32toh(sa->in.sin_addr.s_addr);
568
569 if (include_port)
570 r = asprintf(&p,
571 "%u.%u.%u.%u:%u",
572 a >> 24, (a >> 16) & 0xFF, (a >> 8) & 0xFF, a & 0xFF,
573 be16toh(sa->in.sin_port));
574 else
575 r = asprintf(&p,
576 "%u.%u.%u.%u",
577 a >> 24, (a >> 16) & 0xFF, (a >> 8) & 0xFF, a & 0xFF);
578 if (r < 0)
579 return -ENOMEM;
580 break;
581 }
582
583 case AF_INET6: {
584 static const unsigned char ipv4_prefix[] = {
585 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0xFF, 0xFF
586 };
587
588 if (translate_ipv6 &&
589 memcmp(&sa->in6.sin6_addr, ipv4_prefix, sizeof(ipv4_prefix)) == 0) {
590 const uint8_t *a = sa->in6.sin6_addr.s6_addr+12;
591 if (include_port)
592 r = asprintf(&p,
593 "%u.%u.%u.%u:%u",
594 a[0], a[1], a[2], a[3],
595 be16toh(sa->in6.sin6_port));
596 else
597 r = asprintf(&p,
598 "%u.%u.%u.%u",
599 a[0], a[1], a[2], a[3]);
600 if (r < 0)
601 return -ENOMEM;
602 } else {
603 char a[INET6_ADDRSTRLEN];
604
605 inet_ntop(AF_INET6, &sa->in6.sin6_addr, a, sizeof(a));
606
607 if (include_port) {
608 r = asprintf(&p,
609 "[%s]:%u",
610 a,
611 be16toh(sa->in6.sin6_port));
612 if (r < 0)
613 return -ENOMEM;
614 } else {
615 p = strdup(a);
616 if (!p)
617 return -ENOMEM;
618 }
619 }
620
621 break;
622 }
623
624 case AF_UNIX:
625 if (salen <= offsetof(struct sockaddr_un, sun_path)) {
626 p = strdup("<unnamed>");
627 if (!p)
628 return -ENOMEM;
629
630 } else if (sa->un.sun_path[0] == 0) {
631 /* abstract */
632
633 /* FIXME: We assume we can print the
634 * socket path here and that it hasn't
635 * more than one NUL byte. That is
636 * actually an invalid assumption */
637
638 p = new(char, sizeof(sa->un.sun_path)+1);
639 if (!p)
640 return -ENOMEM;
641
642 p[0] = '@';
643 memcpy(p+1, sa->un.sun_path+1, sizeof(sa->un.sun_path)-1);
644 p[sizeof(sa->un.sun_path)] = 0;
645
646 } else {
647 p = strndup(sa->un.sun_path, sizeof(sa->un.sun_path));
648 if (!p)
649 return -ENOMEM;
650 }
651
652 break;
653
654 case AF_VSOCK:
655 if (include_port)
656 r = asprintf(&p,
657 "vsock:%u:%u",
658 sa->vm.svm_cid,
659 sa->vm.svm_port);
660 else
661 r = asprintf(&p, "vsock:%u", sa->vm.svm_cid);
662 if (r < 0)
663 return -ENOMEM;
664 break;
665
666 default:
667 return -EOPNOTSUPP;
668 }
669
670
671 *ret = p;
672 return 0;
673 }
674
675 int getpeername_pretty(int fd, bool include_port, char **ret) {
676 union sockaddr_union sa;
677 socklen_t salen = sizeof(sa);
678 int r;
679
680 assert(fd >= 0);
681 assert(ret);
682
683 if (getpeername(fd, &sa.sa, &salen) < 0)
684 return -errno;
685
686 if (sa.sa.sa_family == AF_UNIX) {
687 struct ucred ucred = {};
688
689 /* UNIX connection sockets are anonymous, so let's use
690 * PID/UID as pretty credentials instead */
691
692 r = getpeercred(fd, &ucred);
693 if (r < 0)
694 return r;
695
696 if (asprintf(ret, "PID "PID_FMT"/UID "UID_FMT, ucred.pid, ucred.uid) < 0)
697 return -ENOMEM;
698
699 return 0;
700 }
701
702 /* For remote sockets we translate IPv6 addresses back to IPv4
703 * if applicable, since that's nicer. */
704
705 return sockaddr_pretty(&sa.sa, salen, true, include_port, ret);
706 }
707
708 int getsockname_pretty(int fd, char **ret) {
709 union sockaddr_union sa;
710 socklen_t salen = sizeof(sa);
711
712 assert(fd >= 0);
713 assert(ret);
714
715 if (getsockname(fd, &sa.sa, &salen) < 0)
716 return -errno;
717
718 /* For local sockets we do not translate IPv6 addresses back
719 * to IPv6 if applicable, since this is usually used for
720 * listening sockets where the difference between IPv4 and
721 * IPv6 matters. */
722
723 return sockaddr_pretty(&sa.sa, salen, false, true, ret);
724 }
725
726 int socknameinfo_pretty(union sockaddr_union *sa, socklen_t salen, char **_ret) {
727 int r;
728 char host[NI_MAXHOST], *ret;
729
730 assert(_ret);
731
732 r = getnameinfo(&sa->sa, salen, host, sizeof(host), NULL, 0, IDN_FLAGS);
733 if (r != 0) {
734 int saved_errno = errno;
735
736 r = sockaddr_pretty(&sa->sa, salen, true, true, &ret);
737 if (r < 0)
738 return r;
739
740 log_debug_errno(saved_errno, "getnameinfo(%s) failed: %m", ret);
741 } else {
742 ret = strdup(host);
743 if (!ret)
744 return -ENOMEM;
745 }
746
747 *_ret = ret;
748 return 0;
749 }
750
751 int getnameinfo_pretty(int fd, char **ret) {
752 union sockaddr_union sa;
753 socklen_t salen = sizeof(sa);
754
755 assert(fd >= 0);
756 assert(ret);
757
758 if (getsockname(fd, &sa.sa, &salen) < 0)
759 return -errno;
760
761 return socknameinfo_pretty(&sa, salen, ret);
762 }
763
764 int socket_address_unlink(SocketAddress *a) {
765 assert(a);
766
767 if (socket_address_family(a) != AF_UNIX)
768 return 0;
769
770 if (a->sockaddr.un.sun_path[0] == 0)
771 return 0;
772
773 if (unlink(a->sockaddr.un.sun_path) < 0)
774 return -errno;
775
776 return 1;
777 }
778
779 static const char* const netlink_family_table[] = {
780 [NETLINK_ROUTE] = "route",
781 [NETLINK_FIREWALL] = "firewall",
782 [NETLINK_INET_DIAG] = "inet-diag",
783 [NETLINK_NFLOG] = "nflog",
784 [NETLINK_XFRM] = "xfrm",
785 [NETLINK_SELINUX] = "selinux",
786 [NETLINK_ISCSI] = "iscsi",
787 [NETLINK_AUDIT] = "audit",
788 [NETLINK_FIB_LOOKUP] = "fib-lookup",
789 [NETLINK_CONNECTOR] = "connector",
790 [NETLINK_NETFILTER] = "netfilter",
791 [NETLINK_IP6_FW] = "ip6-fw",
792 [NETLINK_DNRTMSG] = "dnrtmsg",
793 [NETLINK_KOBJECT_UEVENT] = "kobject-uevent",
794 [NETLINK_GENERIC] = "generic",
795 [NETLINK_SCSITRANSPORT] = "scsitransport",
796 [NETLINK_ECRYPTFS] = "ecryptfs",
797 [NETLINK_RDMA] = "rdma",
798 };
799
800 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(netlink_family, int, INT_MAX);
801
802 static const char* const socket_address_bind_ipv6_only_table[_SOCKET_ADDRESS_BIND_IPV6_ONLY_MAX] = {
803 [SOCKET_ADDRESS_DEFAULT] = "default",
804 [SOCKET_ADDRESS_BOTH] = "both",
805 [SOCKET_ADDRESS_IPV6_ONLY] = "ipv6-only"
806 };
807
808 DEFINE_STRING_TABLE_LOOKUP(socket_address_bind_ipv6_only, SocketAddressBindIPv6Only);
809
810 bool sockaddr_equal(const union sockaddr_union *a, const union sockaddr_union *b) {
811 assert(a);
812 assert(b);
813
814 if (a->sa.sa_family != b->sa.sa_family)
815 return false;
816
817 if (a->sa.sa_family == AF_INET)
818 return a->in.sin_addr.s_addr == b->in.sin_addr.s_addr;
819
820 if (a->sa.sa_family == AF_INET6)
821 return memcmp(&a->in6.sin6_addr, &b->in6.sin6_addr, sizeof(a->in6.sin6_addr)) == 0;
822
823 if (a->sa.sa_family == AF_VSOCK)
824 return a->vm.svm_cid == b->vm.svm_cid;
825
826 return false;
827 }
828
829 int fd_inc_sndbuf(int fd, size_t n) {
830 int r, value;
831 socklen_t l = sizeof(value);
832
833 r = getsockopt(fd, SOL_SOCKET, SO_SNDBUF, &value, &l);
834 if (r >= 0 && l == sizeof(value) && (size_t) value >= n*2)
835 return 0;
836
837 /* If we have the privileges we will ignore the kernel limit. */
838
839 value = (int) n;
840 if (setsockopt(fd, SOL_SOCKET, SO_SNDBUFFORCE, &value, sizeof(value)) < 0)
841 if (setsockopt(fd, SOL_SOCKET, SO_SNDBUF, &value, sizeof(value)) < 0)
842 return -errno;
843
844 return 1;
845 }
846
847 int fd_inc_rcvbuf(int fd, size_t n) {
848 int r, value;
849 socklen_t l = sizeof(value);
850
851 r = getsockopt(fd, SOL_SOCKET, SO_RCVBUF, &value, &l);
852 if (r >= 0 && l == sizeof(value) && (size_t) value >= n*2)
853 return 0;
854
855 /* If we have the privileges we will ignore the kernel limit. */
856
857 value = (int) n;
858 if (setsockopt(fd, SOL_SOCKET, SO_RCVBUFFORCE, &value, sizeof(value)) < 0)
859 if (setsockopt(fd, SOL_SOCKET, SO_RCVBUF, &value, sizeof(value)) < 0)
860 return -errno;
861 return 1;
862 }
863
864 static const char* const ip_tos_table[] = {
865 [IPTOS_LOWDELAY] = "low-delay",
866 [IPTOS_THROUGHPUT] = "throughput",
867 [IPTOS_RELIABILITY] = "reliability",
868 [IPTOS_LOWCOST] = "low-cost",
869 };
870
871 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(ip_tos, int, 0xff);
872
873 bool ifname_valid(const char *p) {
874 bool numeric = true;
875
876 /* Checks whether a network interface name is valid. This is inspired by dev_valid_name() in the kernel sources
877 * but slightly stricter, as we only allow non-control, non-space ASCII characters in the interface name. We
878 * also don't permit names that only container numbers, to avoid confusion with numeric interface indexes. */
879
880 if (isempty(p))
881 return false;
882
883 if (strlen(p) >= IFNAMSIZ)
884 return false;
885
886 if (dot_or_dot_dot(p))
887 return false;
888
889 while (*p) {
890 if ((unsigned char) *p >= 127U)
891 return false;
892
893 if ((unsigned char) *p <= 32U)
894 return false;
895
896 if (*p == ':' || *p == '/')
897 return false;
898
899 numeric = numeric && (*p >= '0' && *p <= '9');
900 p++;
901 }
902
903 if (numeric)
904 return false;
905
906 return true;
907 }
908
909 bool address_label_valid(const char *p) {
910
911 if (isempty(p))
912 return false;
913
914 if (strlen(p) >= IFNAMSIZ)
915 return false;
916
917 while (*p) {
918 if ((uint8_t) *p >= 127U)
919 return false;
920
921 if ((uint8_t) *p <= 31U)
922 return false;
923 p++;
924 }
925
926 return true;
927 }
928
929 int getpeercred(int fd, struct ucred *ucred) {
930 socklen_t n = sizeof(struct ucred);
931 struct ucred u;
932 int r;
933
934 assert(fd >= 0);
935 assert(ucred);
936
937 r = getsockopt(fd, SOL_SOCKET, SO_PEERCRED, &u, &n);
938 if (r < 0)
939 return -errno;
940
941 if (n != sizeof(struct ucred))
942 return -EIO;
943
944 /* Check if the data is actually useful and not suppressed due
945 * to namespacing issues */
946 if (u.pid <= 0)
947 return -ENODATA;
948 if (u.uid == UID_INVALID)
949 return -ENODATA;
950 if (u.gid == GID_INVALID)
951 return -ENODATA;
952
953 *ucred = u;
954 return 0;
955 }
956
957 int getpeersec(int fd, char **ret) {
958 socklen_t n = 64;
959 char *s;
960 int r;
961
962 assert(fd >= 0);
963 assert(ret);
964
965 s = new0(char, n);
966 if (!s)
967 return -ENOMEM;
968
969 r = getsockopt(fd, SOL_SOCKET, SO_PEERSEC, s, &n);
970 if (r < 0) {
971 free(s);
972
973 if (errno != ERANGE)
974 return -errno;
975
976 s = new0(char, n);
977 if (!s)
978 return -ENOMEM;
979
980 r = getsockopt(fd, SOL_SOCKET, SO_PEERSEC, s, &n);
981 if (r < 0) {
982 free(s);
983 return -errno;
984 }
985 }
986
987 if (isempty(s)) {
988 free(s);
989 return -EOPNOTSUPP;
990 }
991
992 *ret = s;
993 return 0;
994 }
995
996 int send_one_fd_sa(
997 int transport_fd,
998 int fd,
999 const struct sockaddr *sa, socklen_t len,
1000 int flags) {
1001
1002 union {
1003 struct cmsghdr cmsghdr;
1004 uint8_t buf[CMSG_SPACE(sizeof(int))];
1005 } control = {};
1006 struct msghdr mh = {
1007 .msg_name = (struct sockaddr*) sa,
1008 .msg_namelen = len,
1009 .msg_control = &control,
1010 .msg_controllen = sizeof(control),
1011 };
1012 struct cmsghdr *cmsg;
1013
1014 assert(transport_fd >= 0);
1015 assert(fd >= 0);
1016
1017 cmsg = CMSG_FIRSTHDR(&mh);
1018 cmsg->cmsg_level = SOL_SOCKET;
1019 cmsg->cmsg_type = SCM_RIGHTS;
1020 cmsg->cmsg_len = CMSG_LEN(sizeof(int));
1021 memcpy(CMSG_DATA(cmsg), &fd, sizeof(int));
1022
1023 mh.msg_controllen = CMSG_SPACE(sizeof(int));
1024 if (sendmsg(transport_fd, &mh, MSG_NOSIGNAL | flags) < 0)
1025 return -errno;
1026
1027 return 0;
1028 }
1029
1030 int receive_one_fd(int transport_fd, int flags) {
1031 union {
1032 struct cmsghdr cmsghdr;
1033 uint8_t buf[CMSG_SPACE(sizeof(int))];
1034 } control = {};
1035 struct msghdr mh = {
1036 .msg_control = &control,
1037 .msg_controllen = sizeof(control),
1038 };
1039 struct cmsghdr *cmsg, *found = NULL;
1040
1041 assert(transport_fd >= 0);
1042
1043 /*
1044 * Receive a single FD via @transport_fd. We don't care for
1045 * the transport-type. We retrieve a single FD at most, so for
1046 * packet-based transports, the caller must ensure to send
1047 * only a single FD per packet. This is best used in
1048 * combination with send_one_fd().
1049 */
1050
1051 if (recvmsg(transport_fd, &mh, MSG_NOSIGNAL | MSG_CMSG_CLOEXEC | flags) < 0)
1052 return -errno;
1053
1054 CMSG_FOREACH(cmsg, &mh) {
1055 if (cmsg->cmsg_level == SOL_SOCKET &&
1056 cmsg->cmsg_type == SCM_RIGHTS &&
1057 cmsg->cmsg_len == CMSG_LEN(sizeof(int))) {
1058 assert(!found);
1059 found = cmsg;
1060 break;
1061 }
1062 }
1063
1064 if (!found) {
1065 cmsg_close_all(&mh);
1066 return -EIO;
1067 }
1068
1069 return *(int*) CMSG_DATA(found);
1070 }
1071
1072 ssize_t next_datagram_size_fd(int fd) {
1073 ssize_t l;
1074 int k;
1075
1076 /* This is a bit like FIONREAD/SIOCINQ, however a bit more powerful. The difference being: recv(MSG_PEEK) will
1077 * actually cause the next datagram in the queue to be validated regarding checksums, which FIONREAD doesn't
1078 * do. This difference is actually of major importance as we need to be sure that the size returned here
1079 * actually matches what we will read with recvmsg() next, as otherwise we might end up allocating a buffer of
1080 * the wrong size. */
1081
1082 l = recv(fd, NULL, 0, MSG_PEEK|MSG_TRUNC);
1083 if (l < 0) {
1084 if (errno == EOPNOTSUPP || errno == EFAULT)
1085 goto fallback;
1086
1087 return -errno;
1088 }
1089 if (l == 0)
1090 goto fallback;
1091
1092 return l;
1093
1094 fallback:
1095 k = 0;
1096
1097 /* Some sockets (AF_PACKET) do not support null-sized recv() with MSG_TRUNC set, let's fall back to FIONREAD
1098 * for them. Checksums don't matter for raw sockets anyway, hence this should be fine. */
1099
1100 if (ioctl(fd, FIONREAD, &k) < 0)
1101 return -errno;
1102
1103 return (ssize_t) k;
1104 }
1105
1106 int flush_accept(int fd) {
1107
1108 struct pollfd pollfd = {
1109 .fd = fd,
1110 .events = POLLIN,
1111 };
1112 int r;
1113
1114
1115 /* Similar to flush_fd() but flushes all incoming connection by accepting them and immediately closing them. */
1116
1117 for (;;) {
1118 int cfd;
1119
1120 r = poll(&pollfd, 1, 0);
1121 if (r < 0) {
1122 if (errno == EINTR)
1123 continue;
1124
1125 return -errno;
1126
1127 } else if (r == 0)
1128 return 0;
1129
1130 cfd = accept4(fd, NULL, NULL, SOCK_NONBLOCK|SOCK_CLOEXEC);
1131 if (cfd < 0) {
1132 if (errno == EINTR)
1133 continue;
1134
1135 if (errno == EAGAIN)
1136 return 0;
1137
1138 return -errno;
1139 }
1140
1141 close(cfd);
1142 }
1143 }
1144
1145 struct cmsghdr* cmsg_find(struct msghdr *mh, int level, int type, socklen_t length) {
1146 struct cmsghdr *cmsg;
1147
1148 assert(mh);
1149
1150 CMSG_FOREACH(cmsg, mh)
1151 if (cmsg->cmsg_level == level &&
1152 cmsg->cmsg_type == type &&
1153 (length == (socklen_t) -1 || length == cmsg->cmsg_len))
1154 return cmsg;
1155
1156 return NULL;
1157 }
1158
1159 int socket_ioctl_fd(void) {
1160 int fd;
1161
1162 /* Create a socket to invoke the various network interface ioctl()s on. Traditionally only AF_INET was good for
1163 * that. Since kernel 4.6 AF_NETLINK works for this too. We first try to use AF_INET hence, but if that's not
1164 * available (for example, because it is made unavailable via SECCOMP or such), we'll fall back to the more
1165 * generic AF_NETLINK. */
1166
1167 fd = socket(AF_INET, SOCK_DGRAM|SOCK_CLOEXEC, 0);
1168 if (fd < 0)
1169 fd = socket(AF_NETLINK, SOCK_RAW|SOCK_CLOEXEC, NETLINK_GENERIC);
1170 if (fd < 0)
1171 return -errno;
1172
1173 return fd;
1174 }