1 /* SPDX-License-Identifier: LGPL-2.1+ */
3 Copyright 2010 Lennart Poettering
11 #include <netinet/ip.h>
20 #include "alloc-util.h"
23 #include "format-util.h"
27 #include "parse-util.h"
28 #include "path-util.h"
29 #include "process-util.h"
30 #include "socket-util.h"
31 #include "string-table.h"
32 #include "string-util.h"
34 #include "user-util.h"
38 #if 0 /// UNNEEDED by elogind
40 # define IDN_FLAGS NI_IDN
45 static const char* const socket_address_type_table[] = {
46 [SOCK_STREAM] = "Stream",
47 [SOCK_DGRAM] = "Datagram",
49 [SOCK_RDM] = "ReliableDatagram",
50 [SOCK_SEQPACKET] = "SequentialPacket",
51 [SOCK_DCCP] = "DatagramCongestionControl",
54 DEFINE_STRING_TABLE_LOOKUP(socket_address_type, int);
56 int socket_address_parse(SocketAddress *a, const char *s) {
65 a->type = SOCK_STREAM;
70 /* IPv6 in [x:.....:z]:p notation */
76 n = strndupa(s+1, e-s-1);
79 if (inet_pton(AF_INET6, n, &a->sockaddr.in6.sin6_addr) <= 0)
80 return errno > 0 ? -errno : -EINVAL;
88 r = parse_ip_port(e, &port);
92 if (u <= 0 || u > 0xFFFF)
95 a->sockaddr.in6.sin6_family = AF_INET6;
96 a->sockaddr.in6.sin6_port = htobe16((uint16_t)u);
97 a->sockaddr.in6.sin6_port = htobe16(port);
98 a->size = sizeof(struct sockaddr_in6);
100 } else if (*s == '/') {
106 if (l >= sizeof(a->sockaddr.un.sun_path))
109 a->sockaddr.un.sun_family = AF_UNIX;
110 memcpy(a->sockaddr.un.sun_path, s, l);
111 a->size = offsetof(struct sockaddr_un, sun_path) + l + 1;
113 } else if (*s == '@') {
114 /* Abstract AF_UNIX socket */
118 if (l >= sizeof(a->sockaddr.un.sun_path) - 1)
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, &u);
135 r = safe_atou(e+1, &port);
139 n = strndupa(cid_start, e - cid_start);
141 r = safe_atou(n, &a->sockaddr.vm.svm_cid);
145 a->sockaddr.vm.svm_cid = VMADDR_CID_ANY;
147 a->sockaddr.vm.svm_family = AF_VSOCK;
148 a->sockaddr.vm.svm_port = u;
149 a->sockaddr.vm.svm_port = port;
150 a->size = sizeof(struct sockaddr_vm);
157 r = safe_atou(e+1, &u);
158 r = parse_ip_port(e + 1, &port);
162 if (u <= 0 || u > 0xFFFF)
165 n = strndupa(s, e-s);
167 /* IPv4 in w.x.y.z:p notation? */
168 r = inet_pton(AF_INET, n, &a->sockaddr.in.sin_addr);
173 /* Gotcha, it's a traditional IPv4 address */
174 a->sockaddr.in.sin_family = AF_INET;
175 a->sockaddr.in.sin_port = htobe16((uint16_t)u);
176 a->sockaddr.in.sin_port = htobe16(port);
177 a->size = sizeof(struct sockaddr_in);
181 if (strlen(n) > IF_NAMESIZE-1)
184 /* Uh, our last resort, an interface name */
185 idx = if_nametoindex(n);
189 a->sockaddr.in6.sin6_family = AF_INET6;
190 a->sockaddr.in6.sin6_port = htobe16((uint16_t)u);
191 a->sockaddr.in6.sin6_port = htobe16(port);
192 a->sockaddr.in6.sin6_scope_id = idx;
193 a->sockaddr.in6.sin6_addr = in6addr_any;
194 a->size = sizeof(struct sockaddr_in6);
199 r = safe_atou(s, &u);
200 r = parse_ip_port(s, &port);
204 if (u <= 0 || u > 0xFFFF)
207 if (socket_ipv6_is_supported()) {
208 a->sockaddr.in6.sin6_family = AF_INET6;
209 a->sockaddr.in6.sin6_port = htobe16((uint16_t)u);
210 a->sockaddr.in6.sin6_port = htobe16(port);
211 a->sockaddr.in6.sin6_addr = in6addr_any;
212 a->size = sizeof(struct sockaddr_in6);
214 a->sockaddr.in.sin_family = AF_INET;
215 a->sockaddr.in.sin_port = htobe16((uint16_t)u);
216 a->sockaddr.in.sin_port = htobe16(port);
217 a->sockaddr.in.sin_addr.s_addr = INADDR_ANY;
218 a->size = sizeof(struct sockaddr_in);
226 int socket_address_parse_and_warn(SocketAddress *a, const char *s) {
230 /* Similar to socket_address_parse() but warns for IPv6 sockets when we don't support them. */
232 r = socket_address_parse(&b, s);
236 if (!socket_ipv6_is_supported() && b.sockaddr.sa.sa_family == AF_INET6) {
237 log_warning("Binding to IPv6 address not available since kernel does not support IPv6.");
238 return -EAFNOSUPPORT;
245 int socket_address_parse_netlink(SocketAddress *a, const char *s) {
248 _cleanup_free_ char *sfamily = NULL;
256 if (sscanf(s, "%ms %u", &sfamily, &group) < 1)
257 return errno > 0 ? -errno : -EINVAL;
259 family = netlink_family_from_string(sfamily);
263 a->sockaddr.nl.nl_family = AF_NETLINK;
264 a->sockaddr.nl.nl_groups = group;
267 a->size = sizeof(struct sockaddr_nl);
268 a->protocol = family;
273 int socket_address_verify(const SocketAddress *a) {
276 switch (socket_address_family(a)) {
279 if (a->size != sizeof(struct sockaddr_in))
282 if (a->sockaddr.in.sin_port == 0)
285 if (!IN_SET(a->type, SOCK_STREAM, SOCK_DGRAM))
291 if (a->size != sizeof(struct sockaddr_in6))
294 if (a->sockaddr.in6.sin6_port == 0)
297 if (!IN_SET(a->type, SOCK_STREAM, SOCK_DGRAM))
303 if (a->size < offsetof(struct sockaddr_un, sun_path))
306 if (a->size > offsetof(struct sockaddr_un, sun_path)) {
308 if (a->sockaddr.un.sun_path[0] != 0) {
312 e = memchr(a->sockaddr.un.sun_path, 0, sizeof(a->sockaddr.un.sun_path));
316 if (a->size != offsetof(struct sockaddr_un, sun_path) + (e - 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 return a->sockaddr.un.sun_path;
504 bool socket_ipv6_is_supported(void) {
505 if (access("/proc/net/if_inet6", F_OK) != 0)
511 bool socket_address_matches_fd(const SocketAddress *a, int fd) {
518 b.size = sizeof(b.sockaddr);
519 if (getsockname(fd, &b.sockaddr.sa, &b.size) < 0)
522 if (b.sockaddr.sa.sa_family != a->sockaddr.sa.sa_family)
525 solen = sizeof(b.type);
526 if (getsockopt(fd, SOL_SOCKET, SO_TYPE, &b.type, &solen) < 0)
529 if (b.type != a->type)
532 if (a->protocol != 0) {
533 solen = sizeof(b.protocol);
534 if (getsockopt(fd, SOL_SOCKET, SO_PROTOCOL, &b.protocol, &solen) < 0)
537 if (b.protocol != a->protocol)
541 return socket_address_equal(a, &b);
545 int sockaddr_port(const struct sockaddr *_sa, unsigned *ret_port) {
546 union sockaddr_union *sa = (union sockaddr_union*) _sa;
548 /* Note, this returns the port as 'unsigned' rather than 'uint16_t', as AF_VSOCK knows larger ports */
552 switch (sa->sa.sa_family) {
555 *ret_port = be16toh(sa->in.sin_port);
559 *ret_port = be16toh(sa->in6.sin6_port);
563 *ret_port = sa->vm.svm_port;
567 return -EAFNOSUPPORT;
571 #if 0 /// UNNEEDED by elogind
572 int sockaddr_pretty(const struct sockaddr *_sa, socklen_t salen, bool translate_ipv6, bool include_port, char **ret) {
573 union sockaddr_union *sa = (union sockaddr_union*) _sa;
578 assert(salen >= sizeof(sa->sa.sa_family));
580 switch (sa->sa.sa_family) {
585 a = be32toh(sa->in.sin_addr.s_addr);
590 a >> 24, (a >> 16) & 0xFF, (a >> 8) & 0xFF, a & 0xFF,
591 be16toh(sa->in.sin_port));
595 a >> 24, (a >> 16) & 0xFF, (a >> 8) & 0xFF, a & 0xFF);
602 static const unsigned char ipv4_prefix[] = {
603 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0xFF, 0xFF
606 if (translate_ipv6 &&
607 memcmp(&sa->in6.sin6_addr, ipv4_prefix, sizeof(ipv4_prefix)) == 0) {
608 const uint8_t *a = sa->in6.sin6_addr.s6_addr+12;
612 a[0], a[1], a[2], a[3],
613 be16toh(sa->in6.sin6_port));
617 a[0], a[1], a[2], a[3]);
621 char a[INET6_ADDRSTRLEN];
623 inet_ntop(AF_INET6, &sa->in6.sin6_addr, a, sizeof(a));
629 be16toh(sa->in6.sin6_port));
643 if (salen <= offsetof(struct sockaddr_un, sun_path)) {
644 p = strdup("<unnamed>");
648 } else if (sa->un.sun_path[0] == 0) {
651 /* FIXME: We assume we can print the
652 * socket path here and that it hasn't
653 * more than one NUL byte. That is
654 * actually an invalid assumption */
656 p = new(char, sizeof(sa->un.sun_path)+1);
661 memcpy(p+1, sa->un.sun_path+1, sizeof(sa->un.sun_path)-1);
662 p[sizeof(sa->un.sun_path)] = 0;
665 p = strndup(sa->un.sun_path, sizeof(sa->un.sun_path));
679 r = asprintf(&p, "vsock:%u", sa->vm.svm_cid);
692 int getpeername_pretty(int fd, bool include_port, char **ret) {
693 union sockaddr_union sa;
694 socklen_t salen = sizeof(sa);
700 if (getpeername(fd, &sa.sa, &salen) < 0)
703 if (sa.sa.sa_family == AF_UNIX) {
704 struct ucred ucred = {};
706 /* UNIX connection sockets are anonymous, so let's use
707 * PID/UID as pretty credentials instead */
709 r = getpeercred(fd, &ucred);
713 if (asprintf(ret, "PID "PID_FMT"/UID "UID_FMT, ucred.pid, ucred.uid) < 0)
719 /* For remote sockets we translate IPv6 addresses back to IPv4
720 * if applicable, since that's nicer. */
722 return sockaddr_pretty(&sa.sa, salen, true, include_port, ret);
725 int getsockname_pretty(int fd, char **ret) {
726 union sockaddr_union sa;
727 socklen_t salen = sizeof(sa);
732 if (getsockname(fd, &sa.sa, &salen) < 0)
735 /* For local sockets we do not translate IPv6 addresses back
736 * to IPv6 if applicable, since this is usually used for
737 * listening sockets where the difference between IPv4 and
740 return sockaddr_pretty(&sa.sa, salen, false, true, ret);
743 int socknameinfo_pretty(union sockaddr_union *sa, socklen_t salen, char **_ret) {
745 char host[NI_MAXHOST], *ret;
749 r = getnameinfo(&sa->sa, salen, host, sizeof(host), NULL, 0, IDN_FLAGS);
751 int saved_errno = errno;
753 r = sockaddr_pretty(&sa->sa, salen, true, true, &ret);
757 log_debug_errno(saved_errno, "getnameinfo(%s) failed: %m", ret);
768 int socket_address_unlink(SocketAddress *a) {
771 if (socket_address_family(a) != AF_UNIX)
774 if (a->sockaddr.un.sun_path[0] == 0)
777 if (unlink(a->sockaddr.un.sun_path) < 0)
783 static const char* const netlink_family_table[] = {
784 [NETLINK_ROUTE] = "route",
785 [NETLINK_FIREWALL] = "firewall",
786 [NETLINK_INET_DIAG] = "inet-diag",
787 [NETLINK_NFLOG] = "nflog",
788 [NETLINK_XFRM] = "xfrm",
789 [NETLINK_SELINUX] = "selinux",
790 [NETLINK_ISCSI] = "iscsi",
791 [NETLINK_AUDIT] = "audit",
792 [NETLINK_FIB_LOOKUP] = "fib-lookup",
793 [NETLINK_CONNECTOR] = "connector",
794 [NETLINK_NETFILTER] = "netfilter",
795 [NETLINK_IP6_FW] = "ip6-fw",
796 [NETLINK_DNRTMSG] = "dnrtmsg",
797 [NETLINK_KOBJECT_UEVENT] = "kobject-uevent",
798 [NETLINK_GENERIC] = "generic",
799 [NETLINK_SCSITRANSPORT] = "scsitransport",
800 [NETLINK_ECRYPTFS] = "ecryptfs",
801 [NETLINK_RDMA] = "rdma",
804 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(netlink_family, int, INT_MAX);
806 static const char* const socket_address_bind_ipv6_only_table[_SOCKET_ADDRESS_BIND_IPV6_ONLY_MAX] = {
807 [SOCKET_ADDRESS_DEFAULT] = "default",
808 [SOCKET_ADDRESS_BOTH] = "both",
809 [SOCKET_ADDRESS_IPV6_ONLY] = "ipv6-only"
812 DEFINE_STRING_TABLE_LOOKUP(socket_address_bind_ipv6_only, SocketAddressBindIPv6Only);
814 SocketAddressBindIPv6Only socket_address_bind_ipv6_only_or_bool_from_string(const char *n) {
817 r = parse_boolean(n);
819 return SOCKET_ADDRESS_IPV6_ONLY;
821 return SOCKET_ADDRESS_BOTH;
823 return socket_address_bind_ipv6_only_from_string(n);
826 bool sockaddr_equal(const union sockaddr_union *a, const union sockaddr_union *b) {
830 if (a->sa.sa_family != b->sa.sa_family)
833 if (a->sa.sa_family == AF_INET)
834 return a->in.sin_addr.s_addr == b->in.sin_addr.s_addr;
836 if (a->sa.sa_family == AF_INET6)
837 return memcmp(&a->in6.sin6_addr, &b->in6.sin6_addr, sizeof(a->in6.sin6_addr)) == 0;
839 if (a->sa.sa_family == AF_VSOCK)
840 return a->vm.svm_cid == b->vm.svm_cid;
846 int fd_inc_sndbuf(int fd, size_t n) {
848 socklen_t l = sizeof(value);
850 r = getsockopt(fd, SOL_SOCKET, SO_SNDBUF, &value, &l);
851 if (r >= 0 && l == sizeof(value) && (size_t) value >= n*2)
854 /* If we have the privileges we will ignore the kernel limit. */
857 if (setsockopt(fd, SOL_SOCKET, SO_SNDBUFFORCE, &value, sizeof(value)) < 0)
858 if (setsockopt(fd, SOL_SOCKET, SO_SNDBUF, &value, sizeof(value)) < 0)
864 int fd_inc_rcvbuf(int fd, size_t n) {
866 socklen_t l = sizeof(value);
868 r = getsockopt(fd, SOL_SOCKET, SO_RCVBUF, &value, &l);
869 if (r >= 0 && l == sizeof(value) && (size_t) value >= n*2)
872 /* If we have the privileges we will ignore the kernel limit. */
875 if (setsockopt(fd, SOL_SOCKET, SO_RCVBUFFORCE, &value, sizeof(value)) < 0)
876 if (setsockopt(fd, SOL_SOCKET, SO_RCVBUF, &value, sizeof(value)) < 0)
881 #if 0 /// UNNEEDED by elogind
882 static const char* const ip_tos_table[] = {
883 [IPTOS_LOWDELAY] = "low-delay",
884 [IPTOS_THROUGHPUT] = "throughput",
885 [IPTOS_RELIABILITY] = "reliability",
886 [IPTOS_LOWCOST] = "low-cost",
889 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(ip_tos, int, 0xff);
891 bool ifname_valid(const char *p) {
894 /* Checks whether a network interface name is valid. This is inspired by dev_valid_name() in the kernel sources
895 * but slightly stricter, as we only allow non-control, non-space ASCII characters in the interface name. We
896 * also don't permit names that only container numbers, to avoid confusion with numeric interface indexes. */
901 if (strlen(p) >= IFNAMSIZ)
904 if (dot_or_dot_dot(p))
908 if ((unsigned char) *p >= 127U)
911 if ((unsigned char) *p <= 32U)
914 if (IN_SET(*p, ':', '/'))
917 numeric = numeric && (*p >= '0' && *p <= '9');
927 bool address_label_valid(const char *p) {
932 if (strlen(p) >= IFNAMSIZ)
936 if ((uint8_t) *p >= 127U)
939 if ((uint8_t) *p <= 31U)
948 int getpeercred(int fd, struct ucred *ucred) {
949 socklen_t n = sizeof(struct ucred);
956 r = getsockopt(fd, SOL_SOCKET, SO_PEERCRED, &u, &n);
960 if (n != sizeof(struct ucred))
963 /* Check if the data is actually useful and not suppressed due to namespacing issues */
964 if (!pid_is_valid(u.pid))
967 /* Note that we don't check UID/GID here, as namespace translation works differently there: instead of
968 * receiving in "invalid" user/group we get the overflow UID/GID. */
974 int getpeersec(int fd, char **ret) {
975 _cleanup_free_ char *s = NULL;
986 if (getsockopt(fd, SOL_SOCKET, SO_PEERSEC, s, &n) >= 0)
1003 int getpeergroups(int fd, gid_t **ret) {
1004 socklen_t n = sizeof(gid_t) * 64;
1005 _cleanup_free_ gid_t *d = NULL;
1015 if (getsockopt(fd, SOL_SOCKET, SO_PEERGROUPS, d, &n) >= 0)
1018 if (errno != ERANGE)
1024 assert_se(n % sizeof(gid_t) == 0);
1027 if ((socklen_t) (int) n != n)
1038 const struct sockaddr *sa, socklen_t len,
1042 struct cmsghdr cmsghdr;
1043 uint8_t buf[CMSG_SPACE(sizeof(int))];
1045 struct msghdr mh = {
1046 .msg_name = (struct sockaddr*) sa,
1048 .msg_control = &control,
1049 .msg_controllen = sizeof(control),
1051 struct cmsghdr *cmsg;
1053 assert(transport_fd >= 0);
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));
1063 if (sendmsg(transport_fd, &mh, MSG_NOSIGNAL | flags) < 0)
1069 #if 0 /// UNNEEDED by elogind
1070 int receive_one_fd(int transport_fd, int flags) {
1072 struct cmsghdr cmsghdr;
1073 uint8_t buf[CMSG_SPACE(sizeof(int))];
1075 struct msghdr mh = {
1076 .msg_control = &control,
1077 .msg_controllen = sizeof(control),
1079 struct cmsghdr *cmsg, *found = NULL;
1081 assert(transport_fd >= 0);
1084 * Receive a single FD via @transport_fd. We don't care for
1085 * the transport-type. We retrieve a single FD at most, so for
1086 * packet-based transports, the caller must ensure to send
1087 * only a single FD per packet. This is best used in
1088 * combination with send_one_fd().
1091 if (recvmsg(transport_fd, &mh, MSG_NOSIGNAL | MSG_CMSG_CLOEXEC | flags) < 0)
1094 CMSG_FOREACH(cmsg, &mh) {
1095 if (cmsg->cmsg_level == SOL_SOCKET &&
1096 cmsg->cmsg_type == SCM_RIGHTS &&
1097 cmsg->cmsg_len == CMSG_LEN(sizeof(int))) {
1105 cmsg_close_all(&mh);
1109 return *(int*) CMSG_DATA(found);
1112 ssize_t next_datagram_size_fd(int fd) {
1116 /* This is a bit like FIONREAD/SIOCINQ, however a bit more powerful. The difference being: recv(MSG_PEEK) will
1117 * actually cause the next datagram in the queue to be validated regarding checksums, which FIONREAD doesn't
1118 * do. This difference is actually of major importance as we need to be sure that the size returned here
1119 * actually matches what we will read with recvmsg() next, as otherwise we might end up allocating a buffer of
1120 * the wrong size. */
1122 l = recv(fd, NULL, 0, MSG_PEEK|MSG_TRUNC);
1124 if (IN_SET(errno, EOPNOTSUPP, EFAULT))
1137 /* Some sockets (AF_PACKET) do not support null-sized recv() with MSG_TRUNC set, let's fall back to FIONREAD
1138 * for them. Checksums don't matter for raw sockets anyway, hence this should be fine. */
1140 if (ioctl(fd, FIONREAD, &k) < 0)
1146 int flush_accept(int fd) {
1148 struct pollfd pollfd = {
1154 /* Similar to flush_fd() but flushes all incoming connection by accepting them and immediately closing them. */
1159 r = poll(&pollfd, 1, 0);
1169 cfd = accept4(fd, NULL, NULL, SOCK_NONBLOCK|SOCK_CLOEXEC);
1174 if (errno == EAGAIN)
1184 struct cmsghdr* cmsg_find(struct msghdr *mh, int level, int type, socklen_t length) {
1185 struct cmsghdr *cmsg;
1189 CMSG_FOREACH(cmsg, mh)
1190 if (cmsg->cmsg_level == level &&
1191 cmsg->cmsg_type == type &&
1192 (length == (socklen_t) -1 || length == cmsg->cmsg_len))
1198 int socket_ioctl_fd(void) {
1201 /* Create a socket to invoke the various network interface ioctl()s on. Traditionally only AF_INET was good for
1202 * that. Since kernel 4.6 AF_NETLINK works for this too. We first try to use AF_INET hence, but if that's not
1203 * available (for example, because it is made unavailable via SECCOMP or such), we'll fall back to the more
1204 * generic AF_NETLINK. */
1206 fd = socket(AF_INET, SOCK_DGRAM|SOCK_CLOEXEC, 0);
1208 fd = socket(AF_NETLINK, SOCK_RAW|SOCK_CLOEXEC, NETLINK_GENERIC);