1 /* SPDX-License-Identifier: LGPL-2.1+ */
8 #include <netinet/ip.h>
17 #include "alloc-util.h"
20 #include "format-util.h"
24 #include "parse-util.h"
25 #include "path-util.h"
26 #include "process-util.h"
27 #include "socket-util.h"
28 #include "string-table.h"
29 #include "string-util.h"
31 #include "user-util.h"
36 # define IDN_FLAGS NI_IDN
41 static const char* const socket_address_type_table
[] = {
42 [SOCK_STREAM
] = "Stream",
43 [SOCK_DGRAM
] = "Datagram",
45 [SOCK_RDM
] = "ReliableDatagram",
46 [SOCK_SEQPACKET
] = "SequentialPacket",
47 [SOCK_DCCP
] = "DatagramCongestionControl",
50 DEFINE_STRING_TABLE_LOOKUP(socket_address_type
, int);
52 int socket_address_parse(SocketAddress
*a
, const char *s
) {
53 _cleanup_free_
char *n
= NULL
;
60 *a
= (SocketAddress
) {
67 /* IPv6 in [x:.....:z]:p notation */
73 n
= strndup(s
+1, e
-s
-1);
78 if (inet_pton(AF_INET6
, n
, &a
->sockaddr
.in6
.sin6_addr
) <= 0)
79 return errno
> 0 ? -errno
: -EINVAL
;
86 r
= parse_ip_port(e
, &port
);
90 a
->sockaddr
.in6
.sin6_family
= AF_INET6
;
91 a
->sockaddr
.in6
.sin6_port
= htobe16(port
);
92 a
->size
= sizeof(struct sockaddr_in6
);
94 } else if (*s
== '/') {
100 if (l
>= sizeof(a
->sockaddr
.un
.sun_path
)) /* Note that we refuse non-NUL-terminated sockets when
101 * parsing (the kernel itself is less strict here in what it
105 a
->sockaddr
.un
.sun_family
= AF_UNIX
;
106 memcpy(a
->sockaddr
.un
.sun_path
, s
, l
);
107 a
->size
= offsetof(struct sockaddr_un
, sun_path
) + l
+ 1;
109 } else if (*s
== '@') {
110 /* Abstract AF_UNIX socket */
114 if (l
>= sizeof(a
->sockaddr
.un
.sun_path
) - 1) /* Note that we refuse non-NUL-terminate sockets here
115 * when parsing, even though abstract namespace sockets
116 * explicitly allow embedded NUL bytes and don't consider
117 * them special. But it's simply annoying to debug such
121 a
->sockaddr
.un
.sun_family
= AF_UNIX
;
122 memcpy(a
->sockaddr
.un
.sun_path
+1, s
+1, l
);
123 a
->size
= offsetof(struct sockaddr_un
, sun_path
) + 1 + l
;
125 } else if (startswith(s
, "vsock:")) {
126 /* AF_VSOCK socket in vsock:cid:port notation */
127 const char *cid_start
= s
+ STRLEN("vsock:");
130 e
= strchr(cid_start
, ':');
134 r
= safe_atou(e
+1, &port
);
138 n
= strndup(cid_start
, e
- cid_start
);
143 r
= safe_atou(n
, &a
->sockaddr
.vm
.svm_cid
);
147 a
->sockaddr
.vm
.svm_cid
= VMADDR_CID_ANY
;
149 a
->sockaddr
.vm
.svm_family
= AF_VSOCK
;
150 a
->sockaddr
.vm
.svm_port
= port
;
151 a
->size
= sizeof(struct sockaddr_vm
);
158 r
= parse_ip_port(e
+ 1, &port
);
166 /* IPv4 in w.x.y.z:p notation? */
167 r
= inet_pton(AF_INET
, n
, &a
->sockaddr
.in
.sin_addr
);
172 /* Gotcha, it's a traditional IPv4 address */
173 a
->sockaddr
.in
.sin_family
= AF_INET
;
174 a
->sockaddr
.in
.sin_port
= htobe16(port
);
175 a
->size
= sizeof(struct sockaddr_in
);
179 if (strlen(n
) > IF_NAMESIZE
-1)
182 /* Uh, our last resort, an interface name */
183 idx
= if_nametoindex(n
);
187 a
->sockaddr
.in6
.sin6_family
= AF_INET6
;
188 a
->sockaddr
.in6
.sin6_port
= htobe16(port
);
189 a
->sockaddr
.in6
.sin6_scope_id
= idx
;
190 a
->sockaddr
.in6
.sin6_addr
= in6addr_any
;
191 a
->size
= sizeof(struct sockaddr_in6
);
196 r
= parse_ip_port(s
, &port
);
200 if (socket_ipv6_is_supported()) {
201 a
->sockaddr
.in6
.sin6_family
= AF_INET6
;
202 a
->sockaddr
.in6
.sin6_port
= htobe16(port
);
203 a
->sockaddr
.in6
.sin6_addr
= in6addr_any
;
204 a
->size
= sizeof(struct sockaddr_in6
);
206 a
->sockaddr
.in
.sin_family
= AF_INET
;
207 a
->sockaddr
.in
.sin_port
= htobe16(port
);
208 a
->sockaddr
.in
.sin_addr
.s_addr
= INADDR_ANY
;
209 a
->size
= sizeof(struct sockaddr_in
);
217 int socket_address_parse_and_warn(SocketAddress
*a
, const char *s
) {
221 /* Similar to socket_address_parse() but warns for IPv6 sockets when we don't support them. */
223 r
= socket_address_parse(&b
, s
);
227 if (!socket_ipv6_is_supported() && b
.sockaddr
.sa
.sa_family
== AF_INET6
) {
228 log_warning("Binding to IPv6 address not available since kernel does not support IPv6.");
229 return -EAFNOSUPPORT
;
236 int socket_address_parse_netlink(SocketAddress
*a
, const char *s
) {
239 _cleanup_free_
char *sfamily
= NULL
;
247 if (sscanf(s
, "%ms %u", &sfamily
, &group
) < 1)
248 return errno
> 0 ? -errno
: -EINVAL
;
250 family
= netlink_family_from_string(sfamily
);
254 a
->sockaddr
.nl
.nl_family
= AF_NETLINK
;
255 a
->sockaddr
.nl
.nl_groups
= group
;
258 a
->size
= sizeof(struct sockaddr_nl
);
259 a
->protocol
= family
;
264 int socket_address_verify(const SocketAddress
*a
) {
267 switch (socket_address_family(a
)) {
270 if (a
->size
!= sizeof(struct sockaddr_in
))
273 if (a
->sockaddr
.in
.sin_port
== 0)
276 if (!IN_SET(a
->type
, SOCK_STREAM
, SOCK_DGRAM
))
282 if (a
->size
!= sizeof(struct sockaddr_in6
))
285 if (a
->sockaddr
.in6
.sin6_port
== 0)
288 if (!IN_SET(a
->type
, SOCK_STREAM
, SOCK_DGRAM
))
294 if (a
->size
< offsetof(struct sockaddr_un
, sun_path
))
296 if (a
->size
> sizeof(struct sockaddr_un
)+1) /* Allow one extra byte, since getsockname() on Linux will
297 * append a NUL byte if we have path sockets that are above
298 * sun_path' full size */
301 if (a
->size
> offsetof(struct sockaddr_un
, sun_path
) &&
302 a
->sockaddr
.un
.sun_path
[0] != 0) { /* Only validate file system sockets here */
306 e
= memchr(a
->sockaddr
.un
.sun_path
, 0, sizeof(a
->sockaddr
.un
.sun_path
));
308 /* If there's an embedded NUL byte, make sure the size of the socket addresses matches it */
309 if (a
->size
!= offsetof(struct sockaddr_un
, sun_path
) + (e
- a
->sockaddr
.un
.sun_path
) + 1)
312 /* If there's no embedded NUL byte, then then the size needs to match the whole
313 * structure or the structure with one extra NUL byte suffixed. (Yeah, Linux is awful,
314 * and considers both equivalent: getsockname() even extends sockaddr_un beyond its
315 * size if the path is non NUL terminated.)*/
316 if (!IN_SET(a
->size
, sizeof(a
->sockaddr
.un
.sun_path
), sizeof(a
->sockaddr
.un
.sun_path
)+1))
321 if (!IN_SET(a
->type
, SOCK_STREAM
, SOCK_DGRAM
, SOCK_SEQPACKET
))
328 if (a
->size
!= sizeof(struct sockaddr_nl
))
331 if (!IN_SET(a
->type
, SOCK_RAW
, SOCK_DGRAM
))
337 if (a
->size
!= sizeof(struct sockaddr_vm
))
340 if (!IN_SET(a
->type
, SOCK_STREAM
, SOCK_DGRAM
))
346 return -EAFNOSUPPORT
;
350 int socket_address_print(const SocketAddress
*a
, char **ret
) {
356 r
= socket_address_verify(a
);
360 if (socket_address_family(a
) == AF_NETLINK
) {
361 _cleanup_free_
char *sfamily
= NULL
;
363 r
= netlink_family_to_string_alloc(a
->protocol
, &sfamily
);
367 r
= asprintf(ret
, "%s %u", sfamily
, a
->sockaddr
.nl
.nl_groups
);
374 return sockaddr_pretty(&a
->sockaddr
.sa
, a
->size
, false, true, ret
);
377 bool socket_address_can_accept(const SocketAddress
*a
) {
381 IN_SET(a
->type
, SOCK_STREAM
, SOCK_SEQPACKET
);
384 bool socket_address_equal(const SocketAddress
*a
, const SocketAddress
*b
) {
388 /* Invalid addresses are unequal to all */
389 if (socket_address_verify(a
) < 0 ||
390 socket_address_verify(b
) < 0)
393 if (a
->type
!= b
->type
)
396 if (socket_address_family(a
) != socket_address_family(b
))
399 switch (socket_address_family(a
)) {
402 if (a
->sockaddr
.in
.sin_addr
.s_addr
!= b
->sockaddr
.in
.sin_addr
.s_addr
)
405 if (a
->sockaddr
.in
.sin_port
!= b
->sockaddr
.in
.sin_port
)
411 if (memcmp(&a
->sockaddr
.in6
.sin6_addr
, &b
->sockaddr
.in6
.sin6_addr
, sizeof(a
->sockaddr
.in6
.sin6_addr
)) != 0)
414 if (a
->sockaddr
.in6
.sin6_port
!= b
->sockaddr
.in6
.sin6_port
)
420 if (a
->size
<= offsetof(struct sockaddr_un
, sun_path
) ||
421 b
->size
<= offsetof(struct sockaddr_un
, sun_path
))
424 if ((a
->sockaddr
.un
.sun_path
[0] == 0) != (b
->sockaddr
.un
.sun_path
[0] == 0))
427 if (a
->sockaddr
.un
.sun_path
[0]) {
428 if (!path_equal_or_files_same(a
->sockaddr
.un
.sun_path
, b
->sockaddr
.un
.sun_path
, 0))
431 if (a
->size
!= b
->size
)
434 if (memcmp(a
->sockaddr
.un
.sun_path
, b
->sockaddr
.un
.sun_path
, a
->size
) != 0)
441 if (a
->protocol
!= b
->protocol
)
444 if (a
->sockaddr
.nl
.nl_groups
!= b
->sockaddr
.nl
.nl_groups
)
450 if (a
->sockaddr
.vm
.svm_cid
!= b
->sockaddr
.vm
.svm_cid
)
453 if (a
->sockaddr
.vm
.svm_port
!= b
->sockaddr
.vm
.svm_port
)
459 /* Cannot compare, so we assume the addresses are different */
466 bool socket_address_is(const SocketAddress
*a
, const char *s
, int type
) {
467 struct SocketAddress b
;
472 if (socket_address_parse(&b
, s
) < 0)
477 return socket_address_equal(a
, &b
);
480 bool socket_address_is_netlink(const SocketAddress
*a
, const char *s
) {
481 struct SocketAddress b
;
486 if (socket_address_parse_netlink(&b
, s
) < 0)
489 return socket_address_equal(a
, &b
);
492 const char* socket_address_get_path(const SocketAddress
*a
) {
495 if (socket_address_family(a
) != AF_UNIX
)
498 if (a
->sockaddr
.un
.sun_path
[0] == 0)
501 /* Note that this is only safe because we know that there's an extra NUL byte after the sockaddr_un
502 * structure. On Linux AF_UNIX file system socket addresses don't have to be NUL terminated if they take up the
503 * full sun_path space. */
504 assert_cc(sizeof(union sockaddr_union
) >= sizeof(struct sockaddr_un
)+1);
505 return a
->sockaddr
.un
.sun_path
;
508 bool socket_ipv6_is_supported(void) {
509 if (access("/proc/net/if_inet6", F_OK
) != 0)
515 bool socket_address_matches_fd(const SocketAddress
*a
, int fd
) {
522 b
.size
= sizeof(b
.sockaddr
);
523 if (getsockname(fd
, &b
.sockaddr
.sa
, &b
.size
) < 0)
526 if (b
.sockaddr
.sa
.sa_family
!= a
->sockaddr
.sa
.sa_family
)
529 solen
= sizeof(b
.type
);
530 if (getsockopt(fd
, SOL_SOCKET
, SO_TYPE
, &b
.type
, &solen
) < 0)
533 if (b
.type
!= a
->type
)
536 if (a
->protocol
!= 0) {
537 solen
= sizeof(b
.protocol
);
538 if (getsockopt(fd
, SOL_SOCKET
, SO_PROTOCOL
, &b
.protocol
, &solen
) < 0)
541 if (b
.protocol
!= a
->protocol
)
545 return socket_address_equal(a
, &b
);
548 int sockaddr_port(const struct sockaddr
*_sa
, unsigned *ret_port
) {
549 union sockaddr_union
*sa
= (union sockaddr_union
*) _sa
;
551 /* Note, this returns the port as 'unsigned' rather than 'uint16_t', as AF_VSOCK knows larger ports */
555 switch (sa
->sa
.sa_family
) {
558 *ret_port
= be16toh(sa
->in
.sin_port
);
562 *ret_port
= be16toh(sa
->in6
.sin6_port
);
566 *ret_port
= sa
->vm
.svm_port
;
570 return -EAFNOSUPPORT
;
574 int sockaddr_pretty(const struct sockaddr
*_sa
, socklen_t salen
, bool translate_ipv6
, bool include_port
, char **ret
) {
575 union sockaddr_union
*sa
= (union sockaddr_union
*) _sa
;
580 assert(salen
>= sizeof(sa
->sa
.sa_family
));
582 switch (sa
->sa
.sa_family
) {
587 a
= be32toh(sa
->in
.sin_addr
.s_addr
);
592 a
>> 24, (a
>> 16) & 0xFF, (a
>> 8) & 0xFF, a
& 0xFF,
593 be16toh(sa
->in
.sin_port
));
597 a
>> 24, (a
>> 16) & 0xFF, (a
>> 8) & 0xFF, a
& 0xFF);
604 static const unsigned char ipv4_prefix
[] = {
605 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0xFF, 0xFF
608 if (translate_ipv6
&&
609 memcmp(&sa
->in6
.sin6_addr
, ipv4_prefix
, sizeof(ipv4_prefix
)) == 0) {
610 const uint8_t *a
= sa
->in6
.sin6_addr
.s6_addr
+12;
614 a
[0], a
[1], a
[2], a
[3],
615 be16toh(sa
->in6
.sin6_port
));
619 a
[0], a
[1], a
[2], a
[3]);
623 char a
[INET6_ADDRSTRLEN
];
625 inet_ntop(AF_INET6
, &sa
->in6
.sin6_addr
, a
, sizeof(a
));
631 be16toh(sa
->in6
.sin6_port
));
645 if (salen
<= offsetof(struct sockaddr_un
, sun_path
)) {
646 p
= strdup("<unnamed>");
650 } else if (sa
->un
.sun_path
[0] == 0) {
653 /* FIXME: We assume we can print the
654 * socket path here and that it hasn't
655 * more than one NUL byte. That is
656 * actually an invalid assumption */
658 p
= new(char, sizeof(sa
->un
.sun_path
)+1);
663 memcpy(p
+1, sa
->un
.sun_path
+1, sizeof(sa
->un
.sun_path
)-1);
664 p
[sizeof(sa
->un
.sun_path
)] = 0;
667 p
= strndup(sa
->un
.sun_path
, sizeof(sa
->un
.sun_path
));
681 r
= asprintf(&p
, "vsock:%u", sa
->vm
.svm_cid
);
694 int getpeername_pretty(int fd
, bool include_port
, char **ret
) {
695 union sockaddr_union sa
;
696 socklen_t salen
= sizeof(sa
);
702 if (getpeername(fd
, &sa
.sa
, &salen
) < 0)
705 if (sa
.sa
.sa_family
== AF_UNIX
) {
706 struct ucred ucred
= {};
708 /* UNIX connection sockets are anonymous, so let's use
709 * PID/UID as pretty credentials instead */
711 r
= getpeercred(fd
, &ucred
);
715 if (asprintf(ret
, "PID "PID_FMT
"/UID "UID_FMT
, ucred
.pid
, ucred
.uid
) < 0)
721 /* For remote sockets we translate IPv6 addresses back to IPv4
722 * if applicable, since that's nicer. */
724 return sockaddr_pretty(&sa
.sa
, salen
, true, include_port
, ret
);
727 int getsockname_pretty(int fd
, char **ret
) {
728 union sockaddr_union sa
;
729 socklen_t salen
= sizeof(sa
);
734 if (getsockname(fd
, &sa
.sa
, &salen
) < 0)
737 /* For local sockets we do not translate IPv6 addresses back
738 * to IPv6 if applicable, since this is usually used for
739 * listening sockets where the difference between IPv4 and
742 return sockaddr_pretty(&sa
.sa
, salen
, false, true, ret
);
745 int socknameinfo_pretty(union sockaddr_union
*sa
, socklen_t salen
, char **_ret
) {
747 char host
[NI_MAXHOST
], *ret
;
751 r
= getnameinfo(&sa
->sa
, salen
, host
, sizeof(host
), NULL
, 0, IDN_FLAGS
);
753 int saved_errno
= errno
;
755 r
= sockaddr_pretty(&sa
->sa
, salen
, true, true, &ret
);
759 log_debug_errno(saved_errno
, "getnameinfo(%s) failed: %m", ret
);
770 static const char* const netlink_family_table
[] = {
771 [NETLINK_ROUTE
] = "route",
772 [NETLINK_FIREWALL
] = "firewall",
773 [NETLINK_INET_DIAG
] = "inet-diag",
774 [NETLINK_NFLOG
] = "nflog",
775 [NETLINK_XFRM
] = "xfrm",
776 [NETLINK_SELINUX
] = "selinux",
777 [NETLINK_ISCSI
] = "iscsi",
778 [NETLINK_AUDIT
] = "audit",
779 [NETLINK_FIB_LOOKUP
] = "fib-lookup",
780 [NETLINK_CONNECTOR
] = "connector",
781 [NETLINK_NETFILTER
] = "netfilter",
782 [NETLINK_IP6_FW
] = "ip6-fw",
783 [NETLINK_DNRTMSG
] = "dnrtmsg",
784 [NETLINK_KOBJECT_UEVENT
] = "kobject-uevent",
785 [NETLINK_GENERIC
] = "generic",
786 [NETLINK_SCSITRANSPORT
] = "scsitransport",
787 [NETLINK_ECRYPTFS
] = "ecryptfs",
788 [NETLINK_RDMA
] = "rdma",
791 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(netlink_family
, int, INT_MAX
);
793 static const char* const socket_address_bind_ipv6_only_table
[_SOCKET_ADDRESS_BIND_IPV6_ONLY_MAX
] = {
794 [SOCKET_ADDRESS_DEFAULT
] = "default",
795 [SOCKET_ADDRESS_BOTH
] = "both",
796 [SOCKET_ADDRESS_IPV6_ONLY
] = "ipv6-only"
799 DEFINE_STRING_TABLE_LOOKUP(socket_address_bind_ipv6_only
, SocketAddressBindIPv6Only
);
801 SocketAddressBindIPv6Only
socket_address_bind_ipv6_only_or_bool_from_string(const char *n
) {
804 r
= parse_boolean(n
);
806 return SOCKET_ADDRESS_IPV6_ONLY
;
808 return SOCKET_ADDRESS_BOTH
;
810 return socket_address_bind_ipv6_only_from_string(n
);
813 bool sockaddr_equal(const union sockaddr_union
*a
, const union sockaddr_union
*b
) {
817 if (a
->sa
.sa_family
!= b
->sa
.sa_family
)
820 if (a
->sa
.sa_family
== AF_INET
)
821 return a
->in
.sin_addr
.s_addr
== b
->in
.sin_addr
.s_addr
;
823 if (a
->sa
.sa_family
== AF_INET6
)
824 return memcmp(&a
->in6
.sin6_addr
, &b
->in6
.sin6_addr
, sizeof(a
->in6
.sin6_addr
)) == 0;
826 if (a
->sa
.sa_family
== AF_VSOCK
)
827 return a
->vm
.svm_cid
== b
->vm
.svm_cid
;
832 int fd_inc_sndbuf(int fd
, size_t n
) {
834 socklen_t l
= sizeof(value
);
836 r
= getsockopt(fd
, SOL_SOCKET
, SO_SNDBUF
, &value
, &l
);
837 if (r
>= 0 && l
== sizeof(value
) && (size_t) value
>= n
*2)
840 /* If we have the privileges we will ignore the kernel limit. */
842 if (setsockopt_int(fd
, SOL_SOCKET
, SO_SNDBUF
, n
) < 0) {
843 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_SNDBUFFORCE
, n
);
851 int fd_inc_rcvbuf(int fd
, size_t n
) {
853 socklen_t l
= sizeof(value
);
855 r
= getsockopt(fd
, SOL_SOCKET
, SO_RCVBUF
, &value
, &l
);
856 if (r
>= 0 && l
== sizeof(value
) && (size_t) value
>= n
*2)
859 /* If we have the privileges we will ignore the kernel limit. */
861 if (setsockopt_int(fd
, SOL_SOCKET
, SO_RCVBUF
, n
) < 0) {
862 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_RCVBUFFORCE
, n
);
870 static const char* const ip_tos_table
[] = {
871 [IPTOS_LOWDELAY
] = "low-delay",
872 [IPTOS_THROUGHPUT
] = "throughput",
873 [IPTOS_RELIABILITY
] = "reliability",
874 [IPTOS_LOWCOST
] = "low-cost",
877 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(ip_tos
, int, 0xff);
879 bool ifname_valid(const char *p
) {
882 /* Checks whether a network interface name is valid. This is inspired by dev_valid_name() in the kernel sources
883 * but slightly stricter, as we only allow non-control, non-space ASCII characters in the interface name. We
884 * also don't permit names that only container numbers, to avoid confusion with numeric interface indexes. */
889 if (strlen(p
) >= IFNAMSIZ
)
892 if (dot_or_dot_dot(p
))
896 if ((unsigned char) *p
>= 127U)
899 if ((unsigned char) *p
<= 32U)
902 if (IN_SET(*p
, ':', '/'))
905 numeric
= numeric
&& (*p
>= '0' && *p
<= '9');
915 bool address_label_valid(const char *p
) {
920 if (strlen(p
) >= IFNAMSIZ
)
924 if ((uint8_t) *p
>= 127U)
927 if ((uint8_t) *p
<= 31U)
935 int getpeercred(int fd
, struct ucred
*ucred
) {
936 socklen_t n
= sizeof(struct ucred
);
943 r
= getsockopt(fd
, SOL_SOCKET
, SO_PEERCRED
, &u
, &n
);
947 if (n
!= sizeof(struct ucred
))
950 /* Check if the data is actually useful and not suppressed due to namespacing issues */
951 if (!pid_is_valid(u
.pid
))
954 /* Note that we don't check UID/GID here, as namespace translation works differently there: instead of
955 * receiving in "invalid" user/group we get the overflow UID/GID. */
961 int getpeersec(int fd
, char **ret
) {
962 _cleanup_free_
char *s
= NULL
;
973 if (getsockopt(fd
, SOL_SOCKET
, SO_PEERSEC
, s
, &n
) >= 0)
990 int getpeergroups(int fd
, gid_t
**ret
) {
991 socklen_t n
= sizeof(gid_t
) * 64;
992 _cleanup_free_ gid_t
*d
= NULL
;
1002 if (getsockopt(fd
, SOL_SOCKET
, SO_PEERGROUPS
, d
, &n
) >= 0)
1005 if (errno
!= ERANGE
)
1011 assert_se(n
% sizeof(gid_t
) == 0);
1014 if ((socklen_t
) (int) n
!= n
)
1022 ssize_t
send_one_fd_iov_sa(
1025 struct iovec
*iov
, size_t iovlen
,
1026 const struct sockaddr
*sa
, socklen_t len
,
1030 struct cmsghdr cmsghdr
;
1031 uint8_t buf
[CMSG_SPACE(sizeof(int))];
1033 struct msghdr mh
= {
1034 .msg_name
= (struct sockaddr
*) sa
,
1037 .msg_iovlen
= iovlen
,
1041 assert(transport_fd
>= 0);
1044 * We need either an FD or data to send.
1045 * If there's nothing, return an error.
1051 struct cmsghdr
*cmsg
;
1053 mh
.msg_control
= &control
;
1054 mh
.msg_controllen
= sizeof(control
);
1056 cmsg
= CMSG_FIRSTHDR(&mh
);
1057 cmsg
->cmsg_level
= SOL_SOCKET
;
1058 cmsg
->cmsg_type
= SCM_RIGHTS
;
1059 cmsg
->cmsg_len
= CMSG_LEN(sizeof(int));
1060 memcpy(CMSG_DATA(cmsg
), &fd
, sizeof(int));
1062 mh
.msg_controllen
= CMSG_SPACE(sizeof(int));
1064 k
= sendmsg(transport_fd
, &mh
, MSG_NOSIGNAL
| flags
);
1066 return (ssize_t
) -errno
;
1074 const struct sockaddr
*sa
, socklen_t len
,
1079 return (int) send_one_fd_iov_sa(transport_fd
, fd
, NULL
, 0, sa
, len
, flags
);
1082 ssize_t
receive_one_fd_iov(
1084 struct iovec
*iov
, size_t iovlen
,
1089 struct cmsghdr cmsghdr
;
1090 uint8_t buf
[CMSG_SPACE(sizeof(int))];
1092 struct msghdr mh
= {
1093 .msg_control
= &control
,
1094 .msg_controllen
= sizeof(control
),
1096 .msg_iovlen
= iovlen
,
1098 struct cmsghdr
*cmsg
, *found
= NULL
;
1101 assert(transport_fd
>= 0);
1105 * Receive a single FD via @transport_fd. We don't care for
1106 * the transport-type. We retrieve a single FD at most, so for
1107 * packet-based transports, the caller must ensure to send
1108 * only a single FD per packet. This is best used in
1109 * combination with send_one_fd().
1112 k
= recvmsg(transport_fd
, &mh
, MSG_CMSG_CLOEXEC
| flags
);
1114 return (ssize_t
) -errno
;
1116 CMSG_FOREACH(cmsg
, &mh
) {
1117 if (cmsg
->cmsg_level
== SOL_SOCKET
&&
1118 cmsg
->cmsg_type
== SCM_RIGHTS
&&
1119 cmsg
->cmsg_len
== CMSG_LEN(sizeof(int))) {
1127 cmsg_close_all(&mh
);
1129 /* If didn't receive an FD or any data, return an error. */
1130 if (k
== 0 && !found
)
1134 *ret_fd
= *(int*) CMSG_DATA(found
);
1141 int receive_one_fd(int transport_fd
, int flags
) {
1145 k
= receive_one_fd_iov(transport_fd
, NULL
, 0, flags
, &fd
);
1149 /* k must be negative, since receive_one_fd_iov() only returns
1150 * a positive value if data was received through the iov. */
1155 ssize_t
next_datagram_size_fd(int fd
) {
1159 /* This is a bit like FIONREAD/SIOCINQ, however a bit more powerful. The difference being: recv(MSG_PEEK) will
1160 * actually cause the next datagram in the queue to be validated regarding checksums, which FIONREAD doesn't
1161 * do. This difference is actually of major importance as we need to be sure that the size returned here
1162 * actually matches what we will read with recvmsg() next, as otherwise we might end up allocating a buffer of
1163 * the wrong size. */
1165 l
= recv(fd
, NULL
, 0, MSG_PEEK
|MSG_TRUNC
);
1167 if (IN_SET(errno
, EOPNOTSUPP
, EFAULT
))
1180 /* Some sockets (AF_PACKET) do not support null-sized recv() with MSG_TRUNC set, let's fall back to FIONREAD
1181 * for them. Checksums don't matter for raw sockets anyway, hence this should be fine. */
1183 if (ioctl(fd
, FIONREAD
, &k
) < 0)
1189 int flush_accept(int fd
) {
1191 struct pollfd pollfd
= {
1197 /* Similar to flush_fd() but flushes all incoming connection by accepting them and immediately closing them. */
1202 r
= poll(&pollfd
, 1, 0);
1212 cfd
= accept4(fd
, NULL
, NULL
, SOCK_NONBLOCK
|SOCK_CLOEXEC
);
1217 if (errno
== EAGAIN
)
1227 struct cmsghdr
* cmsg_find(struct msghdr
*mh
, int level
, int type
, socklen_t length
) {
1228 struct cmsghdr
*cmsg
;
1232 CMSG_FOREACH(cmsg
, mh
)
1233 if (cmsg
->cmsg_level
== level
&&
1234 cmsg
->cmsg_type
== type
&&
1235 (length
== (socklen_t
) -1 || length
== cmsg
->cmsg_len
))
1241 int socket_ioctl_fd(void) {
1244 /* Create a socket to invoke the various network interface ioctl()s on. Traditionally only AF_INET was good for
1245 * that. Since kernel 4.6 AF_NETLINK works for this too. We first try to use AF_INET hence, but if that's not
1246 * available (for example, because it is made unavailable via SECCOMP or such), we'll fall back to the more
1247 * generic AF_NETLINK. */
1249 fd
= socket(AF_INET
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0);
1251 fd
= socket(AF_NETLINK
, SOCK_RAW
|SOCK_CLOEXEC
, NETLINK_GENERIC
);
1258 int sockaddr_un_unlink(const struct sockaddr_un
*sa
) {
1259 const char *p
, * nul
;
1263 if (sa
->sun_family
!= AF_UNIX
)
1266 if (sa
->sun_path
[0] == 0) /* Nothing to do for abstract sockets */
1269 /* The path in .sun_path is not necessarily NUL terminated. Let's fix that. */
1270 nul
= memchr(sa
->sun_path
, 0, sizeof(sa
->sun_path
));
1274 p
= memdupa_suffix0(sa
->sun_path
, sizeof(sa
->sun_path
));
1282 int sockaddr_un_set_path(struct sockaddr_un
*ret
, const char *path
) {
1288 /* Initialize ret->sun_path from the specified argument. This will interpret paths starting with '@' as
1289 * abstract namespace sockets, and those starting with '/' as regular filesystem sockets. It won't accept
1290 * anything else (i.e. no relative paths), to avoid ambiguities. Note that this function cannot be used to
1291 * reference paths in the abstract namespace that include NUL bytes in the name. */
1296 if (!IN_SET(path
[0], '/', '@'))
1301 /* Don't allow paths larger than the space in sockaddr_un. Note that we are a tiny bit more restrictive than
1302 * the kernel is: we insist on NUL termination (both for abstract namespace and regular file system socket
1303 * addresses!), which the kernel doesn't. We do this to reduce chance of incompatibility with other apps that
1304 * do not expect non-NUL terminated file system path*/
1305 if (l
+1 > sizeof(ret
->sun_path
))
1308 *ret
= (struct sockaddr_un
) {
1309 .sun_family
= AF_UNIX
,
1312 if (path
[0] == '@') {
1313 /* Abstract namespace socket */
1314 memcpy(ret
->sun_path
+ 1, path
+ 1, l
); /* copy *with* trailing NUL byte */
1315 return (int) (offsetof(struct sockaddr_un
, sun_path
) + l
); /* 🔥 *don't* 🔥 include trailing NUL in size */
1318 assert(path
[0] == '/');
1320 /* File system socket */
1321 memcpy(ret
->sun_path
, path
, l
+ 1); /* copy *with* trailing NUL byte */
1322 return (int) (offsetof(struct sockaddr_un
, sun_path
) + l
+ 1); /* include trailing NUL in size */