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>
33 #include "alloc-util.h"
36 #include "formats-util.h"
40 #include "parse-util.h"
41 #include "path-util.h"
42 #include "socket-util.h"
43 #include "string-table.h"
44 #include "string-util.h"
46 #include "user-util.h"
49 #if 0 /// UNNEEDED by elogind
50 int socket_address_parse(SocketAddress *a, const char *s) {
59 a->type = SOCK_STREAM;
62 /* IPv6 in [x:.....:z]:p notation */
68 n = strndupa(s+1, e-s-1);
71 if (inet_pton(AF_INET6, n, &a->sockaddr.in6.sin6_addr) <= 0)
72 return errno > 0 ? -errno : -EINVAL;
83 if (u <= 0 || u > 0xFFFF)
86 a->sockaddr.in6.sin6_family = AF_INET6;
87 a->sockaddr.in6.sin6_port = htobe16((uint16_t)u);
88 a->size = sizeof(struct sockaddr_in6);
90 } else if (*s == '/') {
96 if (l >= sizeof(a->sockaddr.un.sun_path))
99 a->sockaddr.un.sun_family = AF_UNIX;
100 memcpy(a->sockaddr.un.sun_path, s, l);
101 a->size = offsetof(struct sockaddr_un, sun_path) + l + 1;
103 } else if (*s == '@') {
104 /* Abstract AF_UNIX socket */
108 if (l >= sizeof(a->sockaddr.un.sun_path) - 1)
111 a->sockaddr.un.sun_family = AF_UNIX;
112 memcpy(a->sockaddr.un.sun_path+1, s+1, l);
113 a->size = offsetof(struct sockaddr_un, sun_path) + 1 + l;
118 r = safe_atou(e+1, &u);
122 if (u <= 0 || u > 0xFFFF)
125 n = strndupa(s, e-s);
127 /* IPv4 in w.x.y.z:p notation? */
128 r = inet_pton(AF_INET, n, &a->sockaddr.in.sin_addr);
133 /* Gotcha, it's a traditional IPv4 address */
134 a->sockaddr.in.sin_family = AF_INET;
135 a->sockaddr.in.sin_port = htobe16((uint16_t)u);
136 a->size = sizeof(struct sockaddr_in);
140 if (strlen(n) > IF_NAMESIZE-1)
143 /* Uh, our last resort, an interface name */
144 idx = if_nametoindex(n);
148 a->sockaddr.in6.sin6_family = AF_INET6;
149 a->sockaddr.in6.sin6_port = htobe16((uint16_t)u);
150 a->sockaddr.in6.sin6_scope_id = idx;
151 a->sockaddr.in6.sin6_addr = in6addr_any;
152 a->size = sizeof(struct sockaddr_in6);
157 r = safe_atou(s, &u);
161 if (u <= 0 || u > 0xFFFF)
164 if (socket_ipv6_is_supported()) {
165 a->sockaddr.in6.sin6_family = AF_INET6;
166 a->sockaddr.in6.sin6_port = htobe16((uint16_t)u);
167 a->sockaddr.in6.sin6_addr = in6addr_any;
168 a->size = sizeof(struct sockaddr_in6);
170 a->sockaddr.in.sin_family = AF_INET;
171 a->sockaddr.in.sin_port = htobe16((uint16_t)u);
172 a->sockaddr.in.sin_addr.s_addr = INADDR_ANY;
173 a->size = sizeof(struct sockaddr_in);
181 int socket_address_parse_and_warn(SocketAddress *a, const char *s) {
185 /* Similar to socket_address_parse() but warns for IPv6 sockets when we don't support them. */
187 r = socket_address_parse(&b, s);
191 if (!socket_ipv6_is_supported() && b.sockaddr.sa.sa_family == AF_INET6) {
192 log_warning("Binding to IPv6 address not available since kernel does not support IPv6.");
193 return -EAFNOSUPPORT;
200 int socket_address_parse_netlink(SocketAddress *a, const char *s) {
203 _cleanup_free_ char *sfamily = NULL;
211 if (sscanf(s, "%ms %u", &sfamily, &group) < 1)
212 return errno > 0 ? -errno : -EINVAL;
214 family = netlink_family_from_string(sfamily);
218 a->sockaddr.nl.nl_family = AF_NETLINK;
219 a->sockaddr.nl.nl_groups = group;
222 a->size = sizeof(struct sockaddr_nl);
223 a->protocol = family;
228 int socket_address_verify(const SocketAddress *a) {
231 switch (socket_address_family(a)) {
234 if (a->size != sizeof(struct sockaddr_in))
237 if (a->sockaddr.in.sin_port == 0)
240 if (a->type != SOCK_STREAM && a->type != SOCK_DGRAM)
246 if (a->size != sizeof(struct sockaddr_in6))
249 if (a->sockaddr.in6.sin6_port == 0)
252 if (a->type != SOCK_STREAM && a->type != SOCK_DGRAM)
258 if (a->size < offsetof(struct sockaddr_un, sun_path))
261 if (a->size > offsetof(struct sockaddr_un, sun_path)) {
263 if (a->sockaddr.un.sun_path[0] != 0) {
267 e = memchr(a->sockaddr.un.sun_path, 0, sizeof(a->sockaddr.un.sun_path));
271 if (a->size != offsetof(struct sockaddr_un, sun_path) + (e - a->sockaddr.un.sun_path) + 1)
276 if (a->type != SOCK_STREAM && a->type != SOCK_DGRAM && a->type != SOCK_SEQPACKET)
283 if (a->size != sizeof(struct sockaddr_nl))
286 if (a->type != SOCK_RAW && a->type != SOCK_DGRAM)
292 return -EAFNOSUPPORT;
296 int socket_address_print(const SocketAddress *a, char **ret) {
302 r = socket_address_verify(a);
306 if (socket_address_family(a) == AF_NETLINK) {
307 _cleanup_free_ char *sfamily = NULL;
309 r = netlink_family_to_string_alloc(a->protocol, &sfamily);
313 r = asprintf(ret, "%s %u", sfamily, a->sockaddr.nl.nl_groups);
320 return sockaddr_pretty(&a->sockaddr.sa, a->size, false, true, ret);
323 bool socket_address_can_accept(const SocketAddress *a) {
327 a->type == SOCK_STREAM ||
328 a->type == SOCK_SEQPACKET;
331 bool socket_address_equal(const SocketAddress *a, const SocketAddress *b) {
335 /* Invalid addresses are unequal to all */
336 if (socket_address_verify(a) < 0 ||
337 socket_address_verify(b) < 0)
340 if (a->type != b->type)
343 if (socket_address_family(a) != socket_address_family(b))
346 switch (socket_address_family(a)) {
349 if (a->sockaddr.in.sin_addr.s_addr != b->sockaddr.in.sin_addr.s_addr)
352 if (a->sockaddr.in.sin_port != b->sockaddr.in.sin_port)
358 if (memcmp(&a->sockaddr.in6.sin6_addr, &b->sockaddr.in6.sin6_addr, sizeof(a->sockaddr.in6.sin6_addr)) != 0)
361 if (a->sockaddr.in6.sin6_port != b->sockaddr.in6.sin6_port)
367 if (a->size <= offsetof(struct sockaddr_un, sun_path) ||
368 b->size <= offsetof(struct sockaddr_un, sun_path))
371 if ((a->sockaddr.un.sun_path[0] == 0) != (b->sockaddr.un.sun_path[0] == 0))
374 if (a->sockaddr.un.sun_path[0]) {
375 if (!path_equal_or_files_same(a->sockaddr.un.sun_path, b->sockaddr.un.sun_path))
378 if (a->size != b->size)
381 if (memcmp(a->sockaddr.un.sun_path, b->sockaddr.un.sun_path, a->size) != 0)
388 if (a->protocol != b->protocol)
391 if (a->sockaddr.nl.nl_groups != b->sockaddr.nl.nl_groups)
397 /* Cannot compare, so we assume the addresses are different */
404 bool socket_address_is(const SocketAddress *a, const char *s, int type) {
405 struct SocketAddress b;
410 if (socket_address_parse(&b, s) < 0)
415 return socket_address_equal(a, &b);
418 bool socket_address_is_netlink(const SocketAddress *a, const char *s) {
419 struct SocketAddress b;
424 if (socket_address_parse_netlink(&b, s) < 0)
427 return socket_address_equal(a, &b);
430 const char* socket_address_get_path(const SocketAddress *a) {
433 if (socket_address_family(a) != AF_UNIX)
436 if (a->sockaddr.un.sun_path[0] == 0)
439 return a->sockaddr.un.sun_path;
443 bool socket_ipv6_is_supported(void) {
444 if (access("/proc/net/sockstat6", F_OK) != 0)
450 #if 0 /// UNNEEDED by elogind
451 bool socket_address_matches_fd(const SocketAddress *a, int fd) {
458 b.size = sizeof(b.sockaddr);
459 if (getsockname(fd, &b.sockaddr.sa, &b.size) < 0)
462 if (b.sockaddr.sa.sa_family != a->sockaddr.sa.sa_family)
465 solen = sizeof(b.type);
466 if (getsockopt(fd, SOL_SOCKET, SO_TYPE, &b.type, &solen) < 0)
469 if (b.type != a->type)
472 if (a->protocol != 0) {
473 solen = sizeof(b.protocol);
474 if (getsockopt(fd, SOL_SOCKET, SO_PROTOCOL, &b.protocol, &solen) < 0)
477 if (b.protocol != a->protocol)
481 return socket_address_equal(a, &b);
484 int sockaddr_port(const struct sockaddr *_sa) {
485 union sockaddr_union *sa = (union sockaddr_union*) _sa;
489 if (!IN_SET(sa->sa.sa_family, AF_INET, AF_INET6))
490 return -EAFNOSUPPORT;
492 return be16toh(sa->sa.sa_family == AF_INET6 ? sa->in6.sin6_port : sa->in.sin_port);
495 int sockaddr_pretty(const struct sockaddr *_sa, socklen_t salen, bool translate_ipv6, bool include_port, char **ret) {
496 union sockaddr_union *sa = (union sockaddr_union*) _sa;
501 assert(salen >= sizeof(sa->sa.sa_family));
503 switch (sa->sa.sa_family) {
508 a = be32toh(sa->in.sin_addr.s_addr);
513 a >> 24, (a >> 16) & 0xFF, (a >> 8) & 0xFF, a & 0xFF,
514 be16toh(sa->in.sin_port));
518 a >> 24, (a >> 16) & 0xFF, (a >> 8) & 0xFF, a & 0xFF);
525 static const unsigned char ipv4_prefix[] = {
526 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0xFF, 0xFF
529 if (translate_ipv6 &&
530 memcmp(&sa->in6.sin6_addr, ipv4_prefix, sizeof(ipv4_prefix)) == 0) {
531 const uint8_t *a = sa->in6.sin6_addr.s6_addr+12;
535 a[0], a[1], a[2], a[3],
536 be16toh(sa->in6.sin6_port));
540 a[0], a[1], a[2], a[3]);
544 char a[INET6_ADDRSTRLEN];
546 inet_ntop(AF_INET6, &sa->in6.sin6_addr, a, sizeof(a));
552 be16toh(sa->in6.sin6_port));
566 if (salen <= offsetof(struct sockaddr_un, sun_path)) {
567 p = strdup("<unnamed>");
571 } else if (sa->un.sun_path[0] == 0) {
574 /* FIXME: We assume we can print the
575 * socket path here and that it hasn't
576 * more than one NUL byte. That is
577 * actually an invalid assumption */
579 p = new(char, sizeof(sa->un.sun_path)+1);
584 memcpy(p+1, sa->un.sun_path+1, sizeof(sa->un.sun_path)-1);
585 p[sizeof(sa->un.sun_path)] = 0;
588 p = strndup(sa->un.sun_path, sizeof(sa->un.sun_path));
604 int getpeername_pretty(int fd, bool include_port, char **ret) {
605 union sockaddr_union sa;
606 socklen_t salen = sizeof(sa);
612 if (getpeername(fd, &sa.sa, &salen) < 0)
615 if (sa.sa.sa_family == AF_UNIX) {
616 struct ucred ucred = {};
618 /* UNIX connection sockets are anonymous, so let's use
619 * PID/UID as pretty credentials instead */
621 r = getpeercred(fd, &ucred);
625 if (asprintf(ret, "PID "PID_FMT"/UID "UID_FMT, ucred.pid, ucred.uid) < 0)
631 /* For remote sockets we translate IPv6 addresses back to IPv4
632 * if applicable, since that's nicer. */
634 return sockaddr_pretty(&sa.sa, salen, true, include_port, ret);
637 int getsockname_pretty(int fd, char **ret) {
638 union sockaddr_union sa;
639 socklen_t salen = sizeof(sa);
644 if (getsockname(fd, &sa.sa, &salen) < 0)
647 /* For local sockets we do not translate IPv6 addresses back
648 * to IPv6 if applicable, since this is usually used for
649 * listening sockets where the difference between IPv4 and
652 return sockaddr_pretty(&sa.sa, salen, false, true, ret);
655 int socknameinfo_pretty(union sockaddr_union *sa, socklen_t salen, char **_ret) {
657 char host[NI_MAXHOST], *ret;
661 r = getnameinfo(&sa->sa, salen, host, sizeof(host), NULL, 0,
662 NI_IDN|NI_IDN_USE_STD3_ASCII_RULES);
664 int saved_errno = errno;
666 r = sockaddr_pretty(&sa->sa, salen, true, true, &ret);
670 log_debug_errno(saved_errno, "getnameinfo(%s) failed: %m", ret);
681 int getnameinfo_pretty(int fd, char **ret) {
682 union sockaddr_union sa;
683 socklen_t salen = sizeof(sa);
688 if (getsockname(fd, &sa.sa, &salen) < 0)
691 return socknameinfo_pretty(&sa, salen, ret);
694 int socket_address_unlink(SocketAddress *a) {
697 if (socket_address_family(a) != AF_UNIX)
700 if (a->sockaddr.un.sun_path[0] == 0)
703 if (unlink(a->sockaddr.un.sun_path) < 0)
709 static const char* const netlink_family_table[] = {
710 [NETLINK_ROUTE] = "route",
711 [NETLINK_FIREWALL] = "firewall",
712 [NETLINK_INET_DIAG] = "inet-diag",
713 [NETLINK_NFLOG] = "nflog",
714 [NETLINK_XFRM] = "xfrm",
715 [NETLINK_SELINUX] = "selinux",
716 [NETLINK_ISCSI] = "iscsi",
717 [NETLINK_AUDIT] = "audit",
718 [NETLINK_FIB_LOOKUP] = "fib-lookup",
719 [NETLINK_CONNECTOR] = "connector",
720 [NETLINK_NETFILTER] = "netfilter",
721 [NETLINK_IP6_FW] = "ip6-fw",
722 [NETLINK_DNRTMSG] = "dnrtmsg",
723 [NETLINK_KOBJECT_UEVENT] = "kobject-uevent",
724 [NETLINK_GENERIC] = "generic",
725 [NETLINK_SCSITRANSPORT] = "scsitransport",
726 [NETLINK_ECRYPTFS] = "ecryptfs"
729 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(netlink_family, int, INT_MAX);
731 static const char* const socket_address_bind_ipv6_only_table[_SOCKET_ADDRESS_BIND_IPV6_ONLY_MAX] = {
732 [SOCKET_ADDRESS_DEFAULT] = "default",
733 [SOCKET_ADDRESS_BOTH] = "both",
734 [SOCKET_ADDRESS_IPV6_ONLY] = "ipv6-only"
737 DEFINE_STRING_TABLE_LOOKUP(socket_address_bind_ipv6_only, SocketAddressBindIPv6Only);
739 bool sockaddr_equal(const union sockaddr_union *a, const union sockaddr_union *b) {
743 if (a->sa.sa_family != b->sa.sa_family)
746 if (a->sa.sa_family == AF_INET)
747 return a->in.sin_addr.s_addr == b->in.sin_addr.s_addr;
749 if (a->sa.sa_family == AF_INET6)
750 return memcmp(&a->in6.sin6_addr, &b->in6.sin6_addr, sizeof(a->in6.sin6_addr)) == 0;
756 int fd_inc_sndbuf(int fd, size_t n) {
758 socklen_t l = sizeof(value);
760 r = getsockopt(fd, SOL_SOCKET, SO_SNDBUF, &value, &l);
761 if (r >= 0 && l == sizeof(value) && (size_t) value >= n*2)
764 /* If we have the privileges we will ignore the kernel limit. */
767 if (setsockopt(fd, SOL_SOCKET, SO_SNDBUFFORCE, &value, sizeof(value)) < 0)
768 if (setsockopt(fd, SOL_SOCKET, SO_SNDBUF, &value, sizeof(value)) < 0)
774 int fd_inc_rcvbuf(int fd, size_t n) {
776 socklen_t l = sizeof(value);
778 r = getsockopt(fd, SOL_SOCKET, SO_RCVBUF, &value, &l);
779 if (r >= 0 && l == sizeof(value) && (size_t) value >= n*2)
782 /* If we have the privileges we will ignore the kernel limit. */
785 if (setsockopt(fd, SOL_SOCKET, SO_RCVBUFFORCE, &value, sizeof(value)) < 0)
786 if (setsockopt(fd, SOL_SOCKET, SO_RCVBUF, &value, sizeof(value)) < 0)
791 #if 0 /// UNNEEDED by elogind
792 static const char* const ip_tos_table[] = {
793 [IPTOS_LOWDELAY] = "low-delay",
794 [IPTOS_THROUGHPUT] = "throughput",
795 [IPTOS_RELIABILITY] = "reliability",
796 [IPTOS_LOWCOST] = "low-cost",
799 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(ip_tos, int, 0xff);
802 bool ifname_valid(const char *p) {
805 /* Checks whether a network interface name is valid. This is inspired by dev_valid_name() in the kernel sources
806 * but slightly stricter, as we only allow non-control, non-space ASCII characters in the interface name. We
807 * also don't permit names that only container numbers, to avoid confusion with numeric interface indexes. */
812 if (strlen(p) >= IFNAMSIZ)
815 if (STR_IN_SET(p, ".", ".."))
819 if ((unsigned char) *p >= 127U)
822 if ((unsigned char) *p <= 32U)
825 if (*p == ':' || *p == '/')
828 numeric = numeric && (*p >= '0' && *p <= '9');
838 int getpeercred(int fd, struct ucred *ucred) {
839 socklen_t n = sizeof(struct ucred);
846 r = getsockopt(fd, SOL_SOCKET, SO_PEERCRED, &u, &n);
850 if (n != sizeof(struct ucred))
853 /* Check if the data is actually useful and not suppressed due
854 * to namespacing issues */
857 if (u.uid == UID_INVALID)
859 if (u.gid == GID_INVALID)
866 int getpeersec(int fd, char **ret) {
878 r = getsockopt(fd, SOL_SOCKET, SO_PEERSEC, s, &n);
889 r = getsockopt(fd, SOL_SOCKET, SO_PEERSEC, s, &n);
908 const struct sockaddr *sa, socklen_t len,
912 struct cmsghdr cmsghdr;
913 uint8_t buf[CMSG_SPACE(sizeof(int))];
916 .msg_name = (struct sockaddr*) sa,
918 .msg_control = &control,
919 .msg_controllen = sizeof(control),
921 struct cmsghdr *cmsg;
923 assert(transport_fd >= 0);
926 cmsg = CMSG_FIRSTHDR(&mh);
927 cmsg->cmsg_level = SOL_SOCKET;
928 cmsg->cmsg_type = SCM_RIGHTS;
929 cmsg->cmsg_len = CMSG_LEN(sizeof(int));
930 memcpy(CMSG_DATA(cmsg), &fd, sizeof(int));
932 mh.msg_controllen = CMSG_SPACE(sizeof(int));
933 if (sendmsg(transport_fd, &mh, MSG_NOSIGNAL | flags) < 0)
939 #if 0 /// UNNEEDED by elogind
940 int receive_one_fd(int transport_fd, int flags) {
942 struct cmsghdr cmsghdr;
943 uint8_t buf[CMSG_SPACE(sizeof(int))];
946 .msg_control = &control,
947 .msg_controllen = sizeof(control),
949 struct cmsghdr *cmsg, *found = NULL;
951 assert(transport_fd >= 0);
954 * Receive a single FD via @transport_fd. We don't care for
955 * the transport-type. We retrieve a single FD at most, so for
956 * packet-based transports, the caller must ensure to send
957 * only a single FD per packet. This is best used in
958 * combination with send_one_fd().
961 if (recvmsg(transport_fd, &mh, MSG_NOSIGNAL | MSG_CMSG_CLOEXEC | flags) < 0)
964 CMSG_FOREACH(cmsg, &mh) {
965 if (cmsg->cmsg_level == SOL_SOCKET &&
966 cmsg->cmsg_type == SCM_RIGHTS &&
967 cmsg->cmsg_len == CMSG_LEN(sizeof(int))) {
979 return *(int*) CMSG_DATA(found);
982 ssize_t next_datagram_size_fd(int fd) {
986 /* This is a bit like FIONREAD/SIOCINQ, however a bit more powerful. The difference being: recv(MSG_PEEK) will
987 * actually cause the next datagram in the queue to be validated regarding checksums, which FIONREAD doesn't
988 * do. This difference is actually of major importance as we need to be sure that the size returned here
989 * actually matches what we will read with recvmsg() next, as otherwise we might end up allocating a buffer of
992 l = recv(fd, NULL, 0, MSG_PEEK|MSG_TRUNC);
994 if (errno == EOPNOTSUPP || errno == EFAULT)
1007 /* Some sockets (AF_PACKET) do not support null-sized recv() with MSG_TRUNC set, let's fall back to FIONREAD
1008 * for them. Checksums don't matter for raw sockets anyway, hence this should be fine. */
1010 if (ioctl(fd, FIONREAD, &k) < 0)
1016 int flush_accept(int fd) {
1018 struct pollfd pollfd = {
1025 /* Similar to flush_fd() but flushes all incoming connection by accepting them and immediately closing them. */
1030 r = poll(&pollfd, 1, 0);
1040 cfd = accept4(fd, NULL, NULL, SOCK_NONBLOCK|SOCK_CLOEXEC);
1045 if (errno == EAGAIN)