2 This file is part of systemd.
4 Copyright 2010 Lennart Poettering
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.
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.
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/>.
20 #include <arpa/inet.h>
25 #include <netinet/ip.h>
34 #include "alloc-util.h"
37 #include "format-util.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"
47 #include "user-util.h"
51 int socket_address_parse(SocketAddress
*a
, const char *s
) {
60 a
->type
= SOCK_STREAM
;
63 /* IPv6 in [x:.....:z]:p notation */
69 n
= strndupa(s
+1, e
-s
-1);
72 if (inet_pton(AF_INET6
, n
, &a
->sockaddr
.in6
.sin6_addr
) <= 0)
73 return errno
> 0 ? -errno
: -EINVAL
;
84 if (u
<= 0 || u
> 0xFFFF)
87 a
->sockaddr
.in6
.sin6_family
= AF_INET6
;
88 a
->sockaddr
.in6
.sin6_port
= htobe16((uint16_t)u
);
89 a
->size
= sizeof(struct sockaddr_in6
);
91 } else if (*s
== '/') {
97 if (l
>= sizeof(a
->sockaddr
.un
.sun_path
))
100 a
->sockaddr
.un
.sun_family
= AF_UNIX
;
101 memcpy(a
->sockaddr
.un
.sun_path
, s
, l
);
102 a
->size
= offsetof(struct sockaddr_un
, sun_path
) + l
+ 1;
104 } else if (*s
== '@') {
105 /* Abstract AF_UNIX socket */
109 if (l
>= sizeof(a
->sockaddr
.un
.sun_path
) - 1)
112 a
->sockaddr
.un
.sun_family
= AF_UNIX
;
113 memcpy(a
->sockaddr
.un
.sun_path
+1, s
+1, l
);
114 a
->size
= offsetof(struct sockaddr_un
, sun_path
) + 1 + l
;
116 } else if (startswith(s
, "vsock:")) {
117 /* AF_VSOCK socket in vsock:cid:port notation */
118 const char *cid_start
= s
+ strlen("vsock:");
120 e
= strchr(cid_start
, ':');
124 r
= safe_atou(e
+1, &u
);
128 n
= strndupa(cid_start
, e
- cid_start
);
130 r
= safe_atou(n
, &a
->sockaddr
.vm
.svm_cid
);
134 a
->sockaddr
.vm
.svm_cid
= VMADDR_CID_ANY
;
136 a
->sockaddr
.vm
.svm_family
= AF_VSOCK
;
137 a
->sockaddr
.vm
.svm_port
= u
;
138 a
->size
= sizeof(struct sockaddr_vm
);
143 r
= safe_atou(e
+1, &u
);
147 if (u
<= 0 || u
> 0xFFFF)
150 n
= strndupa(s
, e
-s
);
152 /* IPv4 in w.x.y.z:p notation? */
153 r
= inet_pton(AF_INET
, n
, &a
->sockaddr
.in
.sin_addr
);
158 /* Gotcha, it's a traditional IPv4 address */
159 a
->sockaddr
.in
.sin_family
= AF_INET
;
160 a
->sockaddr
.in
.sin_port
= htobe16((uint16_t)u
);
161 a
->size
= sizeof(struct sockaddr_in
);
165 if (strlen(n
) > IF_NAMESIZE
-1)
168 /* Uh, our last resort, an interface name */
169 idx
= if_nametoindex(n
);
173 a
->sockaddr
.in6
.sin6_family
= AF_INET6
;
174 a
->sockaddr
.in6
.sin6_port
= htobe16((uint16_t)u
);
175 a
->sockaddr
.in6
.sin6_scope_id
= idx
;
176 a
->sockaddr
.in6
.sin6_addr
= in6addr_any
;
177 a
->size
= sizeof(struct sockaddr_in6
);
182 r
= safe_atou(s
, &u
);
186 if (u
<= 0 || u
> 0xFFFF)
189 if (socket_ipv6_is_supported()) {
190 a
->sockaddr
.in6
.sin6_family
= AF_INET6
;
191 a
->sockaddr
.in6
.sin6_port
= htobe16((uint16_t)u
);
192 a
->sockaddr
.in6
.sin6_addr
= in6addr_any
;
193 a
->size
= sizeof(struct sockaddr_in6
);
195 a
->sockaddr
.in
.sin_family
= AF_INET
;
196 a
->sockaddr
.in
.sin_port
= htobe16((uint16_t)u
);
197 a
->sockaddr
.in
.sin_addr
.s_addr
= INADDR_ANY
;
198 a
->size
= sizeof(struct sockaddr_in
);
206 int socket_address_parse_and_warn(SocketAddress
*a
, const char *s
) {
210 /* Similar to socket_address_parse() but warns for IPv6 sockets when we don't support them. */
212 r
= socket_address_parse(&b
, s
);
216 if (!socket_ipv6_is_supported() && b
.sockaddr
.sa
.sa_family
== AF_INET6
) {
217 log_warning("Binding to IPv6 address not available since kernel does not support IPv6.");
218 return -EAFNOSUPPORT
;
225 int socket_address_parse_netlink(SocketAddress
*a
, const char *s
) {
228 _cleanup_free_
char *sfamily
= NULL
;
236 if (sscanf(s
, "%ms %u", &sfamily
, &group
) < 1)
237 return errno
> 0 ? -errno
: -EINVAL
;
239 family
= netlink_family_from_string(sfamily
);
243 a
->sockaddr
.nl
.nl_family
= AF_NETLINK
;
244 a
->sockaddr
.nl
.nl_groups
= group
;
247 a
->size
= sizeof(struct sockaddr_nl
);
248 a
->protocol
= family
;
253 int socket_address_verify(const SocketAddress
*a
) {
256 switch (socket_address_family(a
)) {
259 if (a
->size
!= sizeof(struct sockaddr_in
))
262 if (a
->sockaddr
.in
.sin_port
== 0)
265 if (a
->type
!= SOCK_STREAM
&& a
->type
!= SOCK_DGRAM
)
271 if (a
->size
!= sizeof(struct sockaddr_in6
))
274 if (a
->sockaddr
.in6
.sin6_port
== 0)
277 if (a
->type
!= SOCK_STREAM
&& a
->type
!= SOCK_DGRAM
)
283 if (a
->size
< offsetof(struct sockaddr_un
, sun_path
))
286 if (a
->size
> offsetof(struct sockaddr_un
, sun_path
)) {
288 if (a
->sockaddr
.un
.sun_path
[0] != 0) {
292 e
= memchr(a
->sockaddr
.un
.sun_path
, 0, sizeof(a
->sockaddr
.un
.sun_path
));
296 if (a
->size
!= offsetof(struct sockaddr_un
, sun_path
) + (e
- a
->sockaddr
.un
.sun_path
) + 1)
301 if (a
->type
!= SOCK_STREAM
&& a
->type
!= SOCK_DGRAM
&& a
->type
!= SOCK_SEQPACKET
)
308 if (a
->size
!= sizeof(struct sockaddr_nl
))
311 if (a
->type
!= SOCK_RAW
&& a
->type
!= SOCK_DGRAM
)
317 if (a
->size
!= sizeof(struct sockaddr_vm
))
320 if (a
->type
!= SOCK_STREAM
&& a
->type
!= SOCK_DGRAM
)
326 return -EAFNOSUPPORT
;
330 int socket_address_print(const SocketAddress
*a
, char **ret
) {
336 r
= socket_address_verify(a
);
340 if (socket_address_family(a
) == AF_NETLINK
) {
341 _cleanup_free_
char *sfamily
= NULL
;
343 r
= netlink_family_to_string_alloc(a
->protocol
, &sfamily
);
347 r
= asprintf(ret
, "%s %u", sfamily
, a
->sockaddr
.nl
.nl_groups
);
354 return sockaddr_pretty(&a
->sockaddr
.sa
, a
->size
, false, true, ret
);
357 bool socket_address_can_accept(const SocketAddress
*a
) {
361 a
->type
== SOCK_STREAM
||
362 a
->type
== SOCK_SEQPACKET
;
365 bool socket_address_equal(const SocketAddress
*a
, const SocketAddress
*b
) {
369 /* Invalid addresses are unequal to all */
370 if (socket_address_verify(a
) < 0 ||
371 socket_address_verify(b
) < 0)
374 if (a
->type
!= b
->type
)
377 if (socket_address_family(a
) != socket_address_family(b
))
380 switch (socket_address_family(a
)) {
383 if (a
->sockaddr
.in
.sin_addr
.s_addr
!= b
->sockaddr
.in
.sin_addr
.s_addr
)
386 if (a
->sockaddr
.in
.sin_port
!= b
->sockaddr
.in
.sin_port
)
392 if (memcmp(&a
->sockaddr
.in6
.sin6_addr
, &b
->sockaddr
.in6
.sin6_addr
, sizeof(a
->sockaddr
.in6
.sin6_addr
)) != 0)
395 if (a
->sockaddr
.in6
.sin6_port
!= b
->sockaddr
.in6
.sin6_port
)
401 if (a
->size
<= offsetof(struct sockaddr_un
, sun_path
) ||
402 b
->size
<= offsetof(struct sockaddr_un
, sun_path
))
405 if ((a
->sockaddr
.un
.sun_path
[0] == 0) != (b
->sockaddr
.un
.sun_path
[0] == 0))
408 if (a
->sockaddr
.un
.sun_path
[0]) {
409 if (!path_equal_or_files_same(a
->sockaddr
.un
.sun_path
, b
->sockaddr
.un
.sun_path
))
412 if (a
->size
!= b
->size
)
415 if (memcmp(a
->sockaddr
.un
.sun_path
, b
->sockaddr
.un
.sun_path
, a
->size
) != 0)
422 if (a
->protocol
!= b
->protocol
)
425 if (a
->sockaddr
.nl
.nl_groups
!= b
->sockaddr
.nl
.nl_groups
)
431 if (a
->sockaddr
.vm
.svm_cid
!= b
->sockaddr
.vm
.svm_cid
)
434 if (a
->sockaddr
.vm
.svm_port
!= b
->sockaddr
.vm
.svm_port
)
440 /* Cannot compare, so we assume the addresses are different */
447 bool socket_address_is(const SocketAddress
*a
, const char *s
, int type
) {
448 struct SocketAddress b
;
453 if (socket_address_parse(&b
, s
) < 0)
458 return socket_address_equal(a
, &b
);
461 bool socket_address_is_netlink(const SocketAddress
*a
, const char *s
) {
462 struct SocketAddress b
;
467 if (socket_address_parse_netlink(&b
, s
) < 0)
470 return socket_address_equal(a
, &b
);
473 const char* socket_address_get_path(const SocketAddress
*a
) {
476 if (socket_address_family(a
) != AF_UNIX
)
479 if (a
->sockaddr
.un
.sun_path
[0] == 0)
482 return a
->sockaddr
.un
.sun_path
;
485 bool socket_ipv6_is_supported(void) {
486 if (access("/proc/net/if_inet6", F_OK
) != 0)
492 bool socket_address_matches_fd(const SocketAddress
*a
, int fd
) {
499 b
.size
= sizeof(b
.sockaddr
);
500 if (getsockname(fd
, &b
.sockaddr
.sa
, &b
.size
) < 0)
503 if (b
.sockaddr
.sa
.sa_family
!= a
->sockaddr
.sa
.sa_family
)
506 solen
= sizeof(b
.type
);
507 if (getsockopt(fd
, SOL_SOCKET
, SO_TYPE
, &b
.type
, &solen
) < 0)
510 if (b
.type
!= a
->type
)
513 if (a
->protocol
!= 0) {
514 solen
= sizeof(b
.protocol
);
515 if (getsockopt(fd
, SOL_SOCKET
, SO_PROTOCOL
, &b
.protocol
, &solen
) < 0)
518 if (b
.protocol
!= a
->protocol
)
522 return socket_address_equal(a
, &b
);
525 int sockaddr_port(const struct sockaddr
*_sa
, unsigned *port
) {
526 union sockaddr_union
*sa
= (union sockaddr_union
*) _sa
;
530 switch (sa
->sa
.sa_family
) {
532 *port
= be16toh(sa
->in
.sin_port
);
536 *port
= be16toh(sa
->in6
.sin6_port
);
540 *port
= sa
->vm
.svm_port
;
544 return -EAFNOSUPPORT
;
548 int sockaddr_pretty(const struct sockaddr
*_sa
, socklen_t salen
, bool translate_ipv6
, bool include_port
, char **ret
) {
549 union sockaddr_union
*sa
= (union sockaddr_union
*) _sa
;
554 assert(salen
>= sizeof(sa
->sa
.sa_family
));
556 switch (sa
->sa
.sa_family
) {
561 a
= be32toh(sa
->in
.sin_addr
.s_addr
);
566 a
>> 24, (a
>> 16) & 0xFF, (a
>> 8) & 0xFF, a
& 0xFF,
567 be16toh(sa
->in
.sin_port
));
571 a
>> 24, (a
>> 16) & 0xFF, (a
>> 8) & 0xFF, a
& 0xFF);
578 static const unsigned char ipv4_prefix
[] = {
579 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0xFF, 0xFF
582 if (translate_ipv6
&&
583 memcmp(&sa
->in6
.sin6_addr
, ipv4_prefix
, sizeof(ipv4_prefix
)) == 0) {
584 const uint8_t *a
= sa
->in6
.sin6_addr
.s6_addr
+12;
588 a
[0], a
[1], a
[2], a
[3],
589 be16toh(sa
->in6
.sin6_port
));
593 a
[0], a
[1], a
[2], a
[3]);
597 char a
[INET6_ADDRSTRLEN
];
599 inet_ntop(AF_INET6
, &sa
->in6
.sin6_addr
, a
, sizeof(a
));
605 be16toh(sa
->in6
.sin6_port
));
619 if (salen
<= offsetof(struct sockaddr_un
, sun_path
)) {
620 p
= strdup("<unnamed>");
624 } else if (sa
->un
.sun_path
[0] == 0) {
627 /* FIXME: We assume we can print the
628 * socket path here and that it hasn't
629 * more than one NUL byte. That is
630 * actually an invalid assumption */
632 p
= new(char, sizeof(sa
->un
.sun_path
)+1);
637 memcpy(p
+1, sa
->un
.sun_path
+1, sizeof(sa
->un
.sun_path
)-1);
638 p
[sizeof(sa
->un
.sun_path
)] = 0;
641 p
= strndup(sa
->un
.sun_path
, sizeof(sa
->un
.sun_path
));
655 r
= asprintf(&p
, "vsock:%u", sa
->vm
.svm_cid
);
669 int getpeername_pretty(int fd
, bool include_port
, char **ret
) {
670 union sockaddr_union sa
;
671 socklen_t salen
= sizeof(sa
);
677 if (getpeername(fd
, &sa
.sa
, &salen
) < 0)
680 if (sa
.sa
.sa_family
== AF_UNIX
) {
681 struct ucred ucred
= {};
683 /* UNIX connection sockets are anonymous, so let's use
684 * PID/UID as pretty credentials instead */
686 r
= getpeercred(fd
, &ucred
);
690 if (asprintf(ret
, "PID "PID_FMT
"/UID "UID_FMT
, ucred
.pid
, ucred
.uid
) < 0)
696 /* For remote sockets we translate IPv6 addresses back to IPv4
697 * if applicable, since that's nicer. */
699 return sockaddr_pretty(&sa
.sa
, salen
, true, include_port
, ret
);
702 int getsockname_pretty(int fd
, char **ret
) {
703 union sockaddr_union sa
;
704 socklen_t salen
= sizeof(sa
);
709 if (getsockname(fd
, &sa
.sa
, &salen
) < 0)
712 /* For local sockets we do not translate IPv6 addresses back
713 * to IPv6 if applicable, since this is usually used for
714 * listening sockets where the difference between IPv4 and
717 return sockaddr_pretty(&sa
.sa
, salen
, false, true, ret
);
720 int socknameinfo_pretty(union sockaddr_union
*sa
, socklen_t salen
, char **_ret
) {
722 char host
[NI_MAXHOST
], *ret
;
726 r
= getnameinfo(&sa
->sa
, salen
, host
, sizeof(host
), NULL
, 0,
727 NI_IDN
|NI_IDN_USE_STD3_ASCII_RULES
);
729 int saved_errno
= errno
;
731 r
= sockaddr_pretty(&sa
->sa
, salen
, true, true, &ret
);
735 log_debug_errno(saved_errno
, "getnameinfo(%s) failed: %m", ret
);
746 int getnameinfo_pretty(int fd
, char **ret
) {
747 union sockaddr_union sa
;
748 socklen_t salen
= sizeof(sa
);
753 if (getsockname(fd
, &sa
.sa
, &salen
) < 0)
756 return socknameinfo_pretty(&sa
, salen
, ret
);
759 int socket_address_unlink(SocketAddress
*a
) {
762 if (socket_address_family(a
) != AF_UNIX
)
765 if (a
->sockaddr
.un
.sun_path
[0] == 0)
768 if (unlink(a
->sockaddr
.un
.sun_path
) < 0)
774 static const char* const netlink_family_table
[] = {
775 [NETLINK_ROUTE
] = "route",
776 [NETLINK_FIREWALL
] = "firewall",
777 [NETLINK_INET_DIAG
] = "inet-diag",
778 [NETLINK_NFLOG
] = "nflog",
779 [NETLINK_XFRM
] = "xfrm",
780 [NETLINK_SELINUX
] = "selinux",
781 [NETLINK_ISCSI
] = "iscsi",
782 [NETLINK_AUDIT
] = "audit",
783 [NETLINK_FIB_LOOKUP
] = "fib-lookup",
784 [NETLINK_CONNECTOR
] = "connector",
785 [NETLINK_NETFILTER
] = "netfilter",
786 [NETLINK_IP6_FW
] = "ip6-fw",
787 [NETLINK_DNRTMSG
] = "dnrtmsg",
788 [NETLINK_KOBJECT_UEVENT
] = "kobject-uevent",
789 [NETLINK_GENERIC
] = "generic",
790 [NETLINK_SCSITRANSPORT
] = "scsitransport",
791 [NETLINK_ECRYPTFS
] = "ecryptfs"
794 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(netlink_family
, int, INT_MAX
);
796 static const char* const socket_address_bind_ipv6_only_table
[_SOCKET_ADDRESS_BIND_IPV6_ONLY_MAX
] = {
797 [SOCKET_ADDRESS_DEFAULT
] = "default",
798 [SOCKET_ADDRESS_BOTH
] = "both",
799 [SOCKET_ADDRESS_IPV6_ONLY
] = "ipv6-only"
802 DEFINE_STRING_TABLE_LOOKUP(socket_address_bind_ipv6_only
, SocketAddressBindIPv6Only
);
804 bool sockaddr_equal(const union sockaddr_union
*a
, const union sockaddr_union
*b
) {
808 if (a
->sa
.sa_family
!= b
->sa
.sa_family
)
811 if (a
->sa
.sa_family
== AF_INET
)
812 return a
->in
.sin_addr
.s_addr
== b
->in
.sin_addr
.s_addr
;
814 if (a
->sa
.sa_family
== AF_INET6
)
815 return memcmp(&a
->in6
.sin6_addr
, &b
->in6
.sin6_addr
, sizeof(a
->in6
.sin6_addr
)) == 0;
817 if (a
->sa
.sa_family
== AF_VSOCK
)
818 return a
->vm
.svm_cid
== b
->vm
.svm_cid
;
823 int fd_inc_sndbuf(int fd
, size_t n
) {
825 socklen_t l
= sizeof(value
);
827 r
= getsockopt(fd
, SOL_SOCKET
, SO_SNDBUF
, &value
, &l
);
828 if (r
>= 0 && l
== sizeof(value
) && (size_t) value
>= n
*2)
831 /* If we have the privileges we will ignore the kernel limit. */
834 if (setsockopt(fd
, SOL_SOCKET
, SO_SNDBUFFORCE
, &value
, sizeof(value
)) < 0)
835 if (setsockopt(fd
, SOL_SOCKET
, SO_SNDBUF
, &value
, sizeof(value
)) < 0)
841 int fd_inc_rcvbuf(int fd
, size_t n
) {
843 socklen_t l
= sizeof(value
);
845 r
= getsockopt(fd
, SOL_SOCKET
, SO_RCVBUF
, &value
, &l
);
846 if (r
>= 0 && l
== sizeof(value
) && (size_t) value
>= n
*2)
849 /* If we have the privileges we will ignore the kernel limit. */
852 if (setsockopt(fd
, SOL_SOCKET
, SO_RCVBUFFORCE
, &value
, sizeof(value
)) < 0)
853 if (setsockopt(fd
, SOL_SOCKET
, SO_RCVBUF
, &value
, sizeof(value
)) < 0)
858 static const char* const ip_tos_table
[] = {
859 [IPTOS_LOWDELAY
] = "low-delay",
860 [IPTOS_THROUGHPUT
] = "throughput",
861 [IPTOS_RELIABILITY
] = "reliability",
862 [IPTOS_LOWCOST
] = "low-cost",
865 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(ip_tos
, int, 0xff);
867 bool ifname_valid(const char *p
) {
870 /* Checks whether a network interface name is valid. This is inspired by dev_valid_name() in the kernel sources
871 * but slightly stricter, as we only allow non-control, non-space ASCII characters in the interface name. We
872 * also don't permit names that only container numbers, to avoid confusion with numeric interface indexes. */
877 if (strlen(p
) >= IFNAMSIZ
)
880 if (STR_IN_SET(p
, ".", ".."))
884 if ((unsigned char) *p
>= 127U)
887 if ((unsigned char) *p
<= 32U)
890 if (*p
== ':' || *p
== '/')
893 numeric
= numeric
&& (*p
>= '0' && *p
<= '9');
903 int getpeercred(int fd
, struct ucred
*ucred
) {
904 socklen_t n
= sizeof(struct ucred
);
911 r
= getsockopt(fd
, SOL_SOCKET
, SO_PEERCRED
, &u
, &n
);
915 if (n
!= sizeof(struct ucred
))
918 /* Check if the data is actually useful and not suppressed due
919 * to namespacing issues */
922 if (u
.uid
== UID_INVALID
)
924 if (u
.gid
== GID_INVALID
)
931 int getpeersec(int fd
, char **ret
) {
943 r
= getsockopt(fd
, SOL_SOCKET
, SO_PEERSEC
, s
, &n
);
954 r
= getsockopt(fd
, SOL_SOCKET
, SO_PEERSEC
, s
, &n
);
973 const struct sockaddr
*sa
, socklen_t len
,
977 struct cmsghdr cmsghdr
;
978 uint8_t buf
[CMSG_SPACE(sizeof(int))];
981 .msg_name
= (struct sockaddr
*) sa
,
983 .msg_control
= &control
,
984 .msg_controllen
= sizeof(control
),
986 struct cmsghdr
*cmsg
;
988 assert(transport_fd
>= 0);
991 cmsg
= CMSG_FIRSTHDR(&mh
);
992 cmsg
->cmsg_level
= SOL_SOCKET
;
993 cmsg
->cmsg_type
= SCM_RIGHTS
;
994 cmsg
->cmsg_len
= CMSG_LEN(sizeof(int));
995 memcpy(CMSG_DATA(cmsg
), &fd
, sizeof(int));
997 mh
.msg_controllen
= CMSG_SPACE(sizeof(int));
998 if (sendmsg(transport_fd
, &mh
, MSG_NOSIGNAL
| flags
) < 0)
1004 int receive_one_fd(int transport_fd
, int flags
) {
1006 struct cmsghdr cmsghdr
;
1007 uint8_t buf
[CMSG_SPACE(sizeof(int))];
1009 struct msghdr mh
= {
1010 .msg_control
= &control
,
1011 .msg_controllen
= sizeof(control
),
1013 struct cmsghdr
*cmsg
, *found
= NULL
;
1015 assert(transport_fd
>= 0);
1018 * Receive a single FD via @transport_fd. We don't care for
1019 * the transport-type. We retrieve a single FD at most, so for
1020 * packet-based transports, the caller must ensure to send
1021 * only a single FD per packet. This is best used in
1022 * combination with send_one_fd().
1025 if (recvmsg(transport_fd
, &mh
, MSG_NOSIGNAL
| MSG_CMSG_CLOEXEC
| flags
) < 0)
1028 CMSG_FOREACH(cmsg
, &mh
) {
1029 if (cmsg
->cmsg_level
== SOL_SOCKET
&&
1030 cmsg
->cmsg_type
== SCM_RIGHTS
&&
1031 cmsg
->cmsg_len
== CMSG_LEN(sizeof(int))) {
1039 cmsg_close_all(&mh
);
1043 return *(int*) CMSG_DATA(found
);
1046 ssize_t
next_datagram_size_fd(int fd
) {
1050 /* This is a bit like FIONREAD/SIOCINQ, however a bit more powerful. The difference being: recv(MSG_PEEK) will
1051 * actually cause the next datagram in the queue to be validated regarding checksums, which FIONREAD doesn't
1052 * do. This difference is actually of major importance as we need to be sure that the size returned here
1053 * actually matches what we will read with recvmsg() next, as otherwise we might end up allocating a buffer of
1054 * the wrong size. */
1056 l
= recv(fd
, NULL
, 0, MSG_PEEK
|MSG_TRUNC
);
1058 if (errno
== EOPNOTSUPP
|| errno
== EFAULT
)
1071 /* Some sockets (AF_PACKET) do not support null-sized recv() with MSG_TRUNC set, let's fall back to FIONREAD
1072 * for them. Checksums don't matter for raw sockets anyway, hence this should be fine. */
1074 if (ioctl(fd
, FIONREAD
, &k
) < 0)
1080 int flush_accept(int fd
) {
1082 struct pollfd pollfd
= {
1089 /* Similar to flush_fd() but flushes all incoming connection by accepting them and immediately closing them. */
1094 r
= poll(&pollfd
, 1, 0);
1104 cfd
= accept4(fd
, NULL
, NULL
, SOCK_NONBLOCK
|SOCK_CLOEXEC
);
1109 if (errno
== EAGAIN
)
1119 struct cmsghdr
* cmsg_find(struct msghdr
*mh
, int level
, int type
, socklen_t length
) {
1120 struct cmsghdr
*cmsg
;
1124 CMSG_FOREACH(cmsg
, mh
)
1125 if (cmsg
->cmsg_level
== level
&&
1126 cmsg
->cmsg_type
== type
&&
1127 (length
== (socklen_t
) -1 || length
== cmsg
->cmsg_len
))
1133 int socket_ioctl_fd(void) {
1136 /* Create a socket to invoke the various network interface ioctl()s on. Traditionally only AF_INET was good for
1137 * that. Since kernel 4.6 AF_NETLINK works for this too. We first try to use AF_INET hence, but if that's not
1138 * available (for example, because it is made unavailable via SECCOMP or such), we'll fall back to the more
1139 * generic AF_NETLINK. */
1141 fd
= socket(AF_INET
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0);
1143 fd
= socket(AF_NETLINK
, SOCK_RAW
|SOCK_CLOEXEC
, NETLINK_GENERIC
);