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"
35 #if 0 /// UNNEEDED by elogind
37 # define IDN_FLAGS NI_IDN
42 static const char* const socket_address_type_table[] = {
43 [SOCK_STREAM] = "Stream",
44 [SOCK_DGRAM] = "Datagram",
46 [SOCK_RDM] = "ReliableDatagram",
47 [SOCK_SEQPACKET] = "SequentialPacket",
48 [SOCK_DCCP] = "DatagramCongestionControl",
51 DEFINE_STRING_TABLE_LOOKUP(socket_address_type, int);
53 int socket_address_parse(SocketAddress *a, const char *s) {
62 a->type = SOCK_STREAM;
67 /* IPv6 in [x:.....:z]:p notation */
73 n = strndupa(s+1, e-s-1);
76 if (inet_pton(AF_INET6, n, &a->sockaddr.in6.sin6_addr) <= 0)
77 return errno > 0 ? -errno : -EINVAL;
85 r = parse_ip_port(e, &port);
89 if (u <= 0 || u > 0xFFFF)
92 a->sockaddr.in6.sin6_family = AF_INET6;
93 a->sockaddr.in6.sin6_port = htobe16((uint16_t)u);
94 a->sockaddr.in6.sin6_port = htobe16(port);
95 a->size = sizeof(struct sockaddr_in6);
97 } else if (*s == '/') {
103 if (l >= sizeof(a->sockaddr.un.sun_path))
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;
110 } else if (*s == '@') {
111 /* Abstract AF_UNIX socket */
115 if (l >= sizeof(a->sockaddr.un.sun_path) - 1)
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;
122 } else if (startswith(s, "vsock:")) {
123 /* AF_VSOCK socket in vsock:cid:port notation */
124 const char *cid_start = s + STRLEN("vsock:");
127 e = strchr(cid_start, ':');
131 r = safe_atou(e+1, &u);
132 r = safe_atou(e+1, &port);
136 n = strndupa(cid_start, e - cid_start);
138 r = safe_atou(n, &a->sockaddr.vm.svm_cid);
142 a->sockaddr.vm.svm_cid = VMADDR_CID_ANY;
144 a->sockaddr.vm.svm_family = AF_VSOCK;
145 a->sockaddr.vm.svm_port = u;
146 a->sockaddr.vm.svm_port = port;
147 a->size = sizeof(struct sockaddr_vm);
154 r = safe_atou(e+1, &u);
155 r = parse_ip_port(e + 1, &port);
159 if (u <= 0 || u > 0xFFFF)
162 n = strndupa(s, e-s);
164 /* IPv4 in w.x.y.z:p notation? */
165 r = inet_pton(AF_INET, n, &a->sockaddr.in.sin_addr);
170 /* Gotcha, it's a traditional IPv4 address */
171 a->sockaddr.in.sin_family = AF_INET;
172 a->sockaddr.in.sin_port = htobe16((uint16_t)u);
173 a->sockaddr.in.sin_port = htobe16(port);
174 a->size = sizeof(struct sockaddr_in);
178 if (strlen(n) > IF_NAMESIZE-1)
181 /* Uh, our last resort, an interface name */
182 idx = if_nametoindex(n);
186 a->sockaddr.in6.sin6_family = AF_INET6;
187 a->sockaddr.in6.sin6_port = htobe16((uint16_t)u);
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 = safe_atou(s, &u);
197 r = parse_ip_port(s, &port);
201 if (u <= 0 || u > 0xFFFF)
204 if (socket_ipv6_is_supported()) {
205 a->sockaddr.in6.sin6_family = AF_INET6;
206 a->sockaddr.in6.sin6_port = htobe16((uint16_t)u);
207 a->sockaddr.in6.sin6_port = htobe16(port);
208 a->sockaddr.in6.sin6_addr = in6addr_any;
209 a->size = sizeof(struct sockaddr_in6);
211 a->sockaddr.in.sin_family = AF_INET;
212 a->sockaddr.in.sin_port = htobe16((uint16_t)u);
213 a->sockaddr.in.sin_port = htobe16(port);
214 a->sockaddr.in.sin_addr.s_addr = INADDR_ANY;
215 a->size = sizeof(struct sockaddr_in);
223 int socket_address_parse_and_warn(SocketAddress *a, const char *s) {
227 /* Similar to socket_address_parse() but warns for IPv6 sockets when we don't support them. */
229 r = socket_address_parse(&b, s);
233 if (!socket_ipv6_is_supported() && b.sockaddr.sa.sa_family == AF_INET6) {
234 log_warning("Binding to IPv6 address not available since kernel does not support IPv6.");
235 return -EAFNOSUPPORT;
242 int socket_address_parse_netlink(SocketAddress *a, const char *s) {
245 _cleanup_free_ char *sfamily = NULL;
253 if (sscanf(s, "%ms %u", &sfamily, &group) < 1)
254 return errno > 0 ? -errno : -EINVAL;
256 family = netlink_family_from_string(sfamily);
260 a->sockaddr.nl.nl_family = AF_NETLINK;
261 a->sockaddr.nl.nl_groups = group;
264 a->size = sizeof(struct sockaddr_nl);
265 a->protocol = family;
270 int socket_address_verify(const SocketAddress *a) {
273 switch (socket_address_family(a)) {
276 if (a->size != sizeof(struct sockaddr_in))
279 if (a->sockaddr.in.sin_port == 0)
282 if (!IN_SET(a->type, SOCK_STREAM, SOCK_DGRAM))
288 if (a->size != sizeof(struct sockaddr_in6))
291 if (a->sockaddr.in6.sin6_port == 0)
294 if (!IN_SET(a->type, SOCK_STREAM, SOCK_DGRAM))
300 if (a->size < offsetof(struct sockaddr_un, sun_path))
303 if (a->size > offsetof(struct sockaddr_un, sun_path)) {
305 if (a->sockaddr.un.sun_path[0] != 0) {
309 e = memchr(a->sockaddr.un.sun_path, 0, sizeof(a->sockaddr.un.sun_path));
313 if (a->size != offsetof(struct sockaddr_un, sun_path) + (e - a->sockaddr.un.sun_path) + 1)
318 if (!IN_SET(a->type, SOCK_STREAM, SOCK_DGRAM, SOCK_SEQPACKET))
325 if (a->size != sizeof(struct sockaddr_nl))
328 if (!IN_SET(a->type, SOCK_RAW, SOCK_DGRAM))
334 if (a->size != sizeof(struct sockaddr_vm))
337 if (!IN_SET(a->type, SOCK_STREAM, SOCK_DGRAM))
343 return -EAFNOSUPPORT;
347 int socket_address_print(const SocketAddress *a, char **ret) {
353 r = socket_address_verify(a);
357 if (socket_address_family(a) == AF_NETLINK) {
358 _cleanup_free_ char *sfamily = NULL;
360 r = netlink_family_to_string_alloc(a->protocol, &sfamily);
364 r = asprintf(ret, "%s %u", sfamily, a->sockaddr.nl.nl_groups);
371 return sockaddr_pretty(&a->sockaddr.sa, a->size, false, true, ret);
374 bool socket_address_can_accept(const SocketAddress *a) {
378 IN_SET(a->type, SOCK_STREAM, SOCK_SEQPACKET);
381 bool socket_address_equal(const SocketAddress *a, const SocketAddress *b) {
385 /* Invalid addresses are unequal to all */
386 if (socket_address_verify(a) < 0 ||
387 socket_address_verify(b) < 0)
390 if (a->type != b->type)
393 if (socket_address_family(a) != socket_address_family(b))
396 switch (socket_address_family(a)) {
399 if (a->sockaddr.in.sin_addr.s_addr != b->sockaddr.in.sin_addr.s_addr)
402 if (a->sockaddr.in.sin_port != b->sockaddr.in.sin_port)
408 if (memcmp(&a->sockaddr.in6.sin6_addr, &b->sockaddr.in6.sin6_addr, sizeof(a->sockaddr.in6.sin6_addr)) != 0)
411 if (a->sockaddr.in6.sin6_port != b->sockaddr.in6.sin6_port)
417 if (a->size <= offsetof(struct sockaddr_un, sun_path) ||
418 b->size <= offsetof(struct sockaddr_un, sun_path))
421 if ((a->sockaddr.un.sun_path[0] == 0) != (b->sockaddr.un.sun_path[0] == 0))
424 if (a->sockaddr.un.sun_path[0]) {
425 if (!path_equal_or_files_same(a->sockaddr.un.sun_path, b->sockaddr.un.sun_path, 0))
428 if (a->size != b->size)
431 if (memcmp(a->sockaddr.un.sun_path, b->sockaddr.un.sun_path, a->size) != 0)
438 if (a->protocol != b->protocol)
441 if (a->sockaddr.nl.nl_groups != b->sockaddr.nl.nl_groups)
447 if (a->sockaddr.vm.svm_cid != b->sockaddr.vm.svm_cid)
450 if (a->sockaddr.vm.svm_port != b->sockaddr.vm.svm_port)
456 /* Cannot compare, so we assume the addresses are different */
463 bool socket_address_is(const SocketAddress *a, const char *s, int type) {
464 struct SocketAddress b;
469 if (socket_address_parse(&b, s) < 0)
474 return socket_address_equal(a, &b);
477 bool socket_address_is_netlink(const SocketAddress *a, const char *s) {
478 struct SocketAddress b;
483 if (socket_address_parse_netlink(&b, s) < 0)
486 return socket_address_equal(a, &b);
489 const char* socket_address_get_path(const SocketAddress *a) {
492 if (socket_address_family(a) != AF_UNIX)
495 if (a->sockaddr.un.sun_path[0] == 0)
498 return a->sockaddr.un.sun_path;
501 bool socket_ipv6_is_supported(void) {
502 if (access("/proc/net/if_inet6", F_OK) != 0)
508 bool socket_address_matches_fd(const SocketAddress *a, int fd) {
515 b.size = sizeof(b.sockaddr);
516 if (getsockname(fd, &b.sockaddr.sa, &b.size) < 0)
519 if (b.sockaddr.sa.sa_family != a->sockaddr.sa.sa_family)
522 solen = sizeof(b.type);
523 if (getsockopt(fd, SOL_SOCKET, SO_TYPE, &b.type, &solen) < 0)
526 if (b.type != a->type)
529 if (a->protocol != 0) {
530 solen = sizeof(b.protocol);
531 if (getsockopt(fd, SOL_SOCKET, SO_PROTOCOL, &b.protocol, &solen) < 0)
534 if (b.protocol != a->protocol)
538 return socket_address_equal(a, &b);
542 int sockaddr_port(const struct sockaddr *_sa, unsigned *ret_port) {
543 union sockaddr_union *sa = (union sockaddr_union*) _sa;
545 /* Note, this returns the port as 'unsigned' rather than 'uint16_t', as AF_VSOCK knows larger ports */
549 switch (sa->sa.sa_family) {
552 *ret_port = be16toh(sa->in.sin_port);
556 *ret_port = be16toh(sa->in6.sin6_port);
560 *ret_port = sa->vm.svm_port;
564 return -EAFNOSUPPORT;
568 #if 0 /// UNNEEDED by elogind
569 int sockaddr_pretty(const struct sockaddr *_sa, socklen_t salen, bool translate_ipv6, bool include_port, char **ret) {
570 union sockaddr_union *sa = (union sockaddr_union*) _sa;
575 assert(salen >= sizeof(sa->sa.sa_family));
577 switch (sa->sa.sa_family) {
582 a = be32toh(sa->in.sin_addr.s_addr);
587 a >> 24, (a >> 16) & 0xFF, (a >> 8) & 0xFF, a & 0xFF,
588 be16toh(sa->in.sin_port));
592 a >> 24, (a >> 16) & 0xFF, (a >> 8) & 0xFF, a & 0xFF);
599 static const unsigned char ipv4_prefix[] = {
600 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0xFF, 0xFF
603 if (translate_ipv6 &&
604 memcmp(&sa->in6.sin6_addr, ipv4_prefix, sizeof(ipv4_prefix)) == 0) {
605 const uint8_t *a = sa->in6.sin6_addr.s6_addr+12;
609 a[0], a[1], a[2], a[3],
610 be16toh(sa->in6.sin6_port));
614 a[0], a[1], a[2], a[3]);
618 char a[INET6_ADDRSTRLEN];
620 inet_ntop(AF_INET6, &sa->in6.sin6_addr, a, sizeof(a));
626 be16toh(sa->in6.sin6_port));
640 if (salen <= offsetof(struct sockaddr_un, sun_path)) {
641 p = strdup("<unnamed>");
645 } else if (sa->un.sun_path[0] == 0) {
648 /* FIXME: We assume we can print the
649 * socket path here and that it hasn't
650 * more than one NUL byte. That is
651 * actually an invalid assumption */
653 p = new(char, sizeof(sa->un.sun_path)+1);
658 memcpy(p+1, sa->un.sun_path+1, sizeof(sa->un.sun_path)-1);
659 p[sizeof(sa->un.sun_path)] = 0;
662 p = strndup(sa->un.sun_path, sizeof(sa->un.sun_path));
676 r = asprintf(&p, "vsock:%u", sa->vm.svm_cid);
689 int getpeername_pretty(int fd, bool include_port, char **ret) {
690 union sockaddr_union sa;
691 socklen_t salen = sizeof(sa);
697 if (getpeername(fd, &sa.sa, &salen) < 0)
700 if (sa.sa.sa_family == AF_UNIX) {
701 struct ucred ucred = {};
703 /* UNIX connection sockets are anonymous, so let's use
704 * PID/UID as pretty credentials instead */
706 r = getpeercred(fd, &ucred);
710 if (asprintf(ret, "PID "PID_FMT"/UID "UID_FMT, ucred.pid, ucred.uid) < 0)
716 /* For remote sockets we translate IPv6 addresses back to IPv4
717 * if applicable, since that's nicer. */
719 return sockaddr_pretty(&sa.sa, salen, true, include_port, ret);
722 int getsockname_pretty(int fd, char **ret) {
723 union sockaddr_union sa;
724 socklen_t salen = sizeof(sa);
729 if (getsockname(fd, &sa.sa, &salen) < 0)
732 /* For local sockets we do not translate IPv6 addresses back
733 * to IPv6 if applicable, since this is usually used for
734 * listening sockets where the difference between IPv4 and
737 return sockaddr_pretty(&sa.sa, salen, false, true, ret);
740 int socknameinfo_pretty(union sockaddr_union *sa, socklen_t salen, char **_ret) {
742 char host[NI_MAXHOST], *ret;
746 r = getnameinfo(&sa->sa, salen, host, sizeof(host), NULL, 0, IDN_FLAGS);
748 int saved_errno = errno;
750 r = sockaddr_pretty(&sa->sa, salen, true, true, &ret);
754 log_debug_errno(saved_errno, "getnameinfo(%s) failed: %m", ret);
765 int socket_address_unlink(SocketAddress *a) {
768 if (socket_address_family(a) != AF_UNIX)
771 if (a->sockaddr.un.sun_path[0] == 0)
774 if (unlink(a->sockaddr.un.sun_path) < 0)
780 static const char* const netlink_family_table[] = {
781 [NETLINK_ROUTE] = "route",
782 [NETLINK_FIREWALL] = "firewall",
783 [NETLINK_INET_DIAG] = "inet-diag",
784 [NETLINK_NFLOG] = "nflog",
785 [NETLINK_XFRM] = "xfrm",
786 [NETLINK_SELINUX] = "selinux",
787 [NETLINK_ISCSI] = "iscsi",
788 [NETLINK_AUDIT] = "audit",
789 [NETLINK_FIB_LOOKUP] = "fib-lookup",
790 [NETLINK_CONNECTOR] = "connector",
791 [NETLINK_NETFILTER] = "netfilter",
792 [NETLINK_IP6_FW] = "ip6-fw",
793 [NETLINK_DNRTMSG] = "dnrtmsg",
794 [NETLINK_KOBJECT_UEVENT] = "kobject-uevent",
795 [NETLINK_GENERIC] = "generic",
796 [NETLINK_SCSITRANSPORT] = "scsitransport",
797 [NETLINK_ECRYPTFS] = "ecryptfs",
798 [NETLINK_RDMA] = "rdma",
801 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(netlink_family, int, INT_MAX);
803 static const char* const socket_address_bind_ipv6_only_table[_SOCKET_ADDRESS_BIND_IPV6_ONLY_MAX] = {
804 [SOCKET_ADDRESS_DEFAULT] = "default",
805 [SOCKET_ADDRESS_BOTH] = "both",
806 [SOCKET_ADDRESS_IPV6_ONLY] = "ipv6-only"
809 DEFINE_STRING_TABLE_LOOKUP(socket_address_bind_ipv6_only, SocketAddressBindIPv6Only);
811 SocketAddressBindIPv6Only socket_address_bind_ipv6_only_or_bool_from_string(const char *n) {
814 r = parse_boolean(n);
816 return SOCKET_ADDRESS_IPV6_ONLY;
818 return SOCKET_ADDRESS_BOTH;
820 return socket_address_bind_ipv6_only_from_string(n);
823 bool sockaddr_equal(const union sockaddr_union *a, const union sockaddr_union *b) {
827 if (a->sa.sa_family != b->sa.sa_family)
830 if (a->sa.sa_family == AF_INET)
831 return a->in.sin_addr.s_addr == b->in.sin_addr.s_addr;
833 if (a->sa.sa_family == AF_INET6)
834 return memcmp(&a->in6.sin6_addr, &b->in6.sin6_addr, sizeof(a->in6.sin6_addr)) == 0;
836 if (a->sa.sa_family == AF_VSOCK)
837 return a->vm.svm_cid == b->vm.svm_cid;
843 int fd_inc_sndbuf(int fd, size_t n) {
845 socklen_t l = sizeof(value);
847 r = getsockopt(fd, SOL_SOCKET, SO_SNDBUF, &value, &l);
848 if (r >= 0 && l == sizeof(value) && (size_t) value >= n*2)
851 /* If we have the privileges we will ignore the kernel limit. */
854 if (setsockopt(fd, SOL_SOCKET, SO_SNDBUFFORCE, &value, sizeof(value)) < 0)
855 if (setsockopt(fd, SOL_SOCKET, SO_SNDBUF, &value, sizeof(value)) < 0)
861 int fd_inc_rcvbuf(int fd, size_t n) {
863 socklen_t l = sizeof(value);
865 r = getsockopt(fd, SOL_SOCKET, SO_RCVBUF, &value, &l);
866 if (r >= 0 && l == sizeof(value) && (size_t) value >= n*2)
869 /* If we have the privileges we will ignore the kernel limit. */
872 if (setsockopt(fd, SOL_SOCKET, SO_RCVBUFFORCE, &value, sizeof(value)) < 0)
873 if (setsockopt(fd, SOL_SOCKET, SO_RCVBUF, &value, sizeof(value)) < 0)
878 #if 0 /// UNNEEDED by elogind
879 static const char* const ip_tos_table[] = {
880 [IPTOS_LOWDELAY] = "low-delay",
881 [IPTOS_THROUGHPUT] = "throughput",
882 [IPTOS_RELIABILITY] = "reliability",
883 [IPTOS_LOWCOST] = "low-cost",
886 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(ip_tos, int, 0xff);
888 bool ifname_valid(const char *p) {
891 /* Checks whether a network interface name is valid. This is inspired by dev_valid_name() in the kernel sources
892 * but slightly stricter, as we only allow non-control, non-space ASCII characters in the interface name. We
893 * also don't permit names that only container numbers, to avoid confusion with numeric interface indexes. */
898 if (strlen(p) >= IFNAMSIZ)
901 if (dot_or_dot_dot(p))
905 if ((unsigned char) *p >= 127U)
908 if ((unsigned char) *p <= 32U)
911 if (IN_SET(*p, ':', '/'))
914 numeric = numeric && (*p >= '0' && *p <= '9');
924 bool address_label_valid(const char *p) {
929 if (strlen(p) >= IFNAMSIZ)
933 if ((uint8_t) *p >= 127U)
936 if ((uint8_t) *p <= 31U)
945 int getpeercred(int fd, struct ucred *ucred) {
946 socklen_t n = sizeof(struct ucred);
953 r = getsockopt(fd, SOL_SOCKET, SO_PEERCRED, &u, &n);
957 if (n != sizeof(struct ucred))
960 /* Check if the data is actually useful and not suppressed due to namespacing issues */
961 if (!pid_is_valid(u.pid))
964 /* Note that we don't check UID/GID here, as namespace translation works differently there: instead of
965 * receiving in "invalid" user/group we get the overflow UID/GID. */
971 int getpeersec(int fd, char **ret) {
972 _cleanup_free_ char *s = NULL;
983 if (getsockopt(fd, SOL_SOCKET, SO_PEERSEC, s, &n) >= 0)
1000 int getpeergroups(int fd, gid_t **ret) {
1001 socklen_t n = sizeof(gid_t) * 64;
1002 _cleanup_free_ gid_t *d = NULL;
1012 if (getsockopt(fd, SOL_SOCKET, SO_PEERGROUPS, d, &n) >= 0)
1015 if (errno != ERANGE)
1021 assert_se(n % sizeof(gid_t) == 0);
1024 if ((socklen_t) (int) n != n)
1035 const struct sockaddr *sa, socklen_t len,
1039 struct cmsghdr cmsghdr;
1040 uint8_t buf[CMSG_SPACE(sizeof(int))];
1042 struct msghdr mh = {
1043 .msg_name = (struct sockaddr*) sa,
1045 .msg_control = &control,
1046 .msg_controllen = sizeof(control),
1048 struct cmsghdr *cmsg;
1050 assert(transport_fd >= 0);
1053 cmsg = CMSG_FIRSTHDR(&mh);
1054 cmsg->cmsg_level = SOL_SOCKET;
1055 cmsg->cmsg_type = SCM_RIGHTS;
1056 cmsg->cmsg_len = CMSG_LEN(sizeof(int));
1057 memcpy(CMSG_DATA(cmsg), &fd, sizeof(int));
1059 mh.msg_controllen = CMSG_SPACE(sizeof(int));
1060 if (sendmsg(transport_fd, &mh, MSG_NOSIGNAL | flags) < 0)
1066 #if 0 /// UNNEEDED by elogind
1067 int receive_one_fd(int transport_fd, int flags) {
1069 struct cmsghdr cmsghdr;
1070 uint8_t buf[CMSG_SPACE(sizeof(int))];
1072 struct msghdr mh = {
1073 .msg_control = &control,
1074 .msg_controllen = sizeof(control),
1076 struct cmsghdr *cmsg, *found = NULL;
1078 assert(transport_fd >= 0);
1081 * Receive a single FD via @transport_fd. We don't care for
1082 * the transport-type. We retrieve a single FD at most, so for
1083 * packet-based transports, the caller must ensure to send
1084 * only a single FD per packet. This is best used in
1085 * combination with send_one_fd().
1088 if (recvmsg(transport_fd, &mh, MSG_CMSG_CLOEXEC | flags) < 0)
1091 CMSG_FOREACH(cmsg, &mh) {
1092 if (cmsg->cmsg_level == SOL_SOCKET &&
1093 cmsg->cmsg_type == SCM_RIGHTS &&
1094 cmsg->cmsg_len == CMSG_LEN(sizeof(int))) {
1102 cmsg_close_all(&mh);
1106 return *(int*) CMSG_DATA(found);
1109 ssize_t next_datagram_size_fd(int fd) {
1113 /* This is a bit like FIONREAD/SIOCINQ, however a bit more powerful. The difference being: recv(MSG_PEEK) will
1114 * actually cause the next datagram in the queue to be validated regarding checksums, which FIONREAD doesn't
1115 * do. This difference is actually of major importance as we need to be sure that the size returned here
1116 * actually matches what we will read with recvmsg() next, as otherwise we might end up allocating a buffer of
1117 * the wrong size. */
1119 l = recv(fd, NULL, 0, MSG_PEEK|MSG_TRUNC);
1121 if (IN_SET(errno, EOPNOTSUPP, EFAULT))
1134 /* Some sockets (AF_PACKET) do not support null-sized recv() with MSG_TRUNC set, let's fall back to FIONREAD
1135 * for them. Checksums don't matter for raw sockets anyway, hence this should be fine. */
1137 if (ioctl(fd, FIONREAD, &k) < 0)
1143 int flush_accept(int fd) {
1145 struct pollfd pollfd = {
1151 /* Similar to flush_fd() but flushes all incoming connection by accepting them and immediately closing them. */
1156 r = poll(&pollfd, 1, 0);
1166 cfd = accept4(fd, NULL, NULL, SOCK_NONBLOCK|SOCK_CLOEXEC);
1171 if (errno == EAGAIN)
1181 struct cmsghdr* cmsg_find(struct msghdr *mh, int level, int type, socklen_t length) {
1182 struct cmsghdr *cmsg;
1186 CMSG_FOREACH(cmsg, mh)
1187 if (cmsg->cmsg_level == level &&
1188 cmsg->cmsg_type == type &&
1189 (length == (socklen_t) -1 || length == cmsg->cmsg_len))
1195 int socket_ioctl_fd(void) {
1198 /* Create a socket to invoke the various network interface ioctl()s on. Traditionally only AF_INET was good for
1199 * that. Since kernel 4.6 AF_NETLINK works for this too. We first try to use AF_INET hence, but if that's not
1200 * available (for example, because it is made unavailable via SECCOMP or such), we'll fall back to the more
1201 * generic AF_NETLINK. */
1203 fd = socket(AF_INET, SOCK_DGRAM|SOCK_CLOEXEC, 0);
1205 fd = socket(AF_NETLINK, SOCK_RAW|SOCK_CLOEXEC, NETLINK_GENERIC);