2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
18 #include "socket-util.h"
19 #include <arpa/inet.h>
29 #include <sys/ioctl.h>
30 #include <sys/socket.h>
35 #include "dynamic-string.h"
36 #include "fatal-signal.h"
37 #include "ovs-thread.h"
39 #include "poll-loop.h"
43 #include <linux/if_packet.h>
46 #include "netlink-protocol.h"
47 #include "netlink-socket.h"
50 VLOG_DEFINE_THIS_MODULE(socket_util);
52 /* #ifdefs make it a pain to maintain code: you have to try to build both ways.
53 * Thus, this file compiles all of the code regardless of the target, by
54 * writing "if (LINUX)" instead of "#ifdef __linux__". */
65 /* Maximum length of the sun_path member in a struct sockaddr_un, excluding
66 * space for a null terminator. */
67 #define MAX_UN_LEN (sizeof(((struct sockaddr_un *) 0)->sun_path) - 1)
69 static int getsockopt_int(int fd, int level, int option, const char *optname,
72 /* Sets 'fd' to non-blocking mode. Returns 0 if successful, otherwise a
73 * positive errno value. */
75 set_nonblocking(int fd)
78 int flags = fcntl(fd, F_GETFL, 0);
80 if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) != -1) {
83 VLOG_ERR("fcntl(F_SETFL) failed: %s", ovs_strerror(errno));
87 VLOG_ERR("fcntl(F_GETFL) failed: %s", ovs_strerror(errno));
91 unsigned long arg = 1;
92 if (ioctlsocket(fd, FIONBIO, &arg)) {
93 int error = sock_errno();
94 VLOG_ERR("set_nonblocking failed: %s", sock_strerror(error));
102 xset_nonblocking(int fd)
104 if (set_nonblocking(fd)) {
110 set_dscp(int fd, uint8_t dscp)
119 /* Note: this function is used for both of IPv4 and IPv6 sockets */
122 if (setsockopt(fd, IPPROTO_IP, IP_TOS, &val, sizeof val)) {
124 if (sock_errno() != ENOPROTOOPT) {
126 if (sock_errno() != WSAENOPROTOOPT) {
133 if (setsockopt(fd, IPPROTO_IPV6, IPV6_TCLASS, &val, sizeof val)) {
135 if (sock_errno() != ENOPROTOOPT) {
137 if (sock_errno() != WSAENOPROTOOPT) {
151 /* Translates 'host_name', which must be a string representation of an IP
152 * address, into a numeric IP address in '*addr'. Returns 0 if successful,
153 * otherwise a positive errno value. */
155 lookup_ip(const char *host_name, struct in_addr *addr)
157 if (!inet_pton(AF_INET, host_name, addr)) {
158 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
159 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name);
165 /* Translates 'host_name', which must be a string representation of an IPv6
166 * address, into a numeric IPv6 address in '*addr'. Returns 0 if successful,
167 * otherwise a positive errno value. */
169 lookup_ipv6(const char *host_name, struct in6_addr *addr)
171 if (inet_pton(AF_INET6, host_name, addr) != 1) {
172 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
173 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IPv6 address", host_name);
179 /* Translates 'host_name', which must be a host name or a string representation
180 * of an IP address, into a numeric IP address in '*addr'. Returns 0 if
181 * successful, otherwise a positive errno value.
183 * Most Open vSwitch code should not use this because it causes deadlocks:
184 * getaddrinfo() sends out a DNS request but that starts a new flow for which
185 * OVS must set up a flow, but it can't because it's waiting for a DNS reply.
186 * The synchronous lookup also delays other activity. (Of course we can solve
187 * this but it doesn't seem worthwhile quite yet.) */
189 lookup_hostname(const char *host_name, struct in_addr *addr)
191 struct addrinfo *result;
192 struct addrinfo hints;
194 if (inet_pton(AF_INET, host_name, addr)) {
198 memset(&hints, 0, sizeof hints);
199 hints.ai_family = AF_INET;
201 switch (getaddrinfo(host_name, NULL, &hints, &result)) {
203 *addr = ALIGNED_CAST(struct sockaddr_in *,
204 result->ai_addr)->sin_addr;
205 freeaddrinfo(result);
208 #ifdef EAI_ADDRFAMILY
229 #if defined (EAI_NODATA) && EAI_NODATA != EAI_NONAME
245 check_connection_completion(int fd)
247 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
252 pfd.events = POLLOUT;
256 retval = poll(&pfd, 1, 0);
257 } while (retval < 0 && errno == EINTR);
259 retval = WSAPoll(&pfd, 1, 0);
262 if (pfd.revents & POLLERR) {
263 ssize_t n = send(fd, "", 1, 0);
267 VLOG_ERR_RL(&rl, "poll return POLLERR but send succeeded");
272 } else if (retval < 0) {
273 VLOG_ERR_RL(&rl, "poll: %s", sock_strerror(sock_errno()));
281 /* Drain all the data currently in the receive queue of a datagram socket (and
282 * possibly additional data). There is no way to know how many packets are in
283 * the receive queue, but we do know that the total number of bytes queued does
284 * not exceed the receive buffer size, so we pull packets until none are left
285 * or we've read that many bytes. */
291 rcvbuf = get_socket_rcvbuf(fd);
297 /* In Linux, specifying MSG_TRUNC in the flags argument causes the
298 * datagram length to be returned, even if that is longer than the
299 * buffer provided. Thus, we can use a 1-byte buffer to discard the
300 * incoming datagram and still be able to account how many bytes were
301 * removed from the receive buffer.
303 * On other Unix-like OSes, MSG_TRUNC has no effect in the flags
305 char buffer[LINUX ? 1 : 2048];
306 ssize_t n_bytes = recv(fd, buffer, sizeof buffer,
307 MSG_TRUNC | MSG_DONTWAIT);
308 if (n_bytes <= 0 || n_bytes >= rcvbuf) {
317 /* Returns the size of socket 'sock''s receive buffer (SO_RCVBUF), or a
318 * negative errno value if an error occurs. */
320 get_socket_rcvbuf(int sock)
325 error = getsockopt_int(sock, SOL_SOCKET, SO_RCVBUF, "SO_RCVBUF", &rcvbuf);
326 return error ? -error : rcvbuf;
329 /* Reads and discards up to 'n' datagrams from 'fd', stopping as soon as no
330 * more data can be immediately read. ('fd' should therefore be in
331 * non-blocking mode.)*/
333 drain_fd(int fd, size_t n_packets)
335 for (; n_packets > 0; n_packets--) {
336 /* 'buffer' only needs to be 1 byte long in most circumstances. This
337 * size is defensive against the possibility that we someday want to
338 * use a Linux tap device without TUN_NO_PI, in which case a buffer
339 * smaller than sizeof(struct tun_pi) will give EINVAL on read. */
341 if (read(fd, buffer, sizeof buffer) <= 0) {
348 /* Attempts to shorten 'name' by opening a file descriptor for the directory
349 * part of the name and indirecting through /proc/self/fd/<dirfd>/<basename>.
350 * On systems with Linux-like /proc, this works as long as <basename> isn't too
353 * On success, returns 0 and stores the short name in 'short_name' and a
354 * directory file descriptor to eventually be closed in '*dirfpd'. */
356 shorten_name_via_proc(const char *name, char short_name[MAX_UN_LEN + 1],
367 dir = dir_name(name);
368 dirfd = open(dir, O_DIRECTORY | O_RDONLY);
370 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
373 VLOG_WARN_RL(&rl, "%s: open failed (%s)", dir, ovs_strerror(error));
380 base = base_name(name);
381 len = snprintf(short_name, MAX_UN_LEN + 1,
382 "/proc/self/fd/%d/%s", dirfd, base);
385 if (len >= 0 && len <= MAX_UN_LEN) {
394 /* Attempts to shorten 'name' by creating a symlink for the directory part of
395 * the name and indirecting through <symlink>/<basename>. This works on
396 * systems that support symlinks, as long as <basename> isn't too long.
398 * On success, returns 0 and stores the short name in 'short_name' and the
399 * symbolic link to eventually delete in 'linkname'. */
401 shorten_name_via_symlink(const char *name, char short_name[MAX_UN_LEN + 1],
402 char linkname[MAX_UN_LEN + 1])
404 char *abs, *dir, *base;
409 abs = abs_file_name(NULL, name);
411 base = base_name(abs);
414 tmpdir = getenv("TMPDIR");
415 if (tmpdir == NULL) {
419 for (i = 0; i < 1000; i++) {
422 len = snprintf(linkname, MAX_UN_LEN + 1,
423 "%s/ovs-un-c-%"PRIu32, tmpdir, random_uint32());
424 error = (len < 0 || len > MAX_UN_LEN ? ENAMETOOLONG
425 : symlink(dir, linkname) ? errno
427 if (error != EEXIST) {
435 fatal_signal_add_file_to_unlink(linkname);
437 len = snprintf(short_name, MAX_UN_LEN + 1, "%s/%s", linkname, base);
438 if (len < 0 || len > MAX_UN_LEN) {
439 fatal_signal_unlink_file_now(linkname);
440 error = ENAMETOOLONG;
453 /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in
454 * '*un_len' the size of the sockaddr_un.
456 * Returns 0 on success, otherwise a positive errno value.
458 * Uses '*dirfdp' and 'linkname' to store references to data when the caller no
459 * longer needs to use 'un'. On success, freeing these references with
460 * free_sockaddr_un() is mandatory to avoid a leak; on failure, freeing them is
461 * unnecessary but harmless. */
463 make_sockaddr_un(const char *name, struct sockaddr_un *un, socklen_t *un_len,
464 int *dirfdp, char linkname[MAX_UN_LEN + 1])
466 char short_name[MAX_UN_LEN + 1];
470 if (strlen(name) > MAX_UN_LEN) {
471 /* 'name' is too long to fit in a sockaddr_un. Try a workaround. */
472 int error = shorten_name_via_proc(name, short_name, dirfdp);
473 if (error == ENAMETOOLONG) {
474 error = shorten_name_via_symlink(name, short_name, linkname);
477 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
479 VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum "
480 "%"PRIuSIZE" bytes", name, MAX_UN_LEN);
487 un->sun_family = AF_UNIX;
488 ovs_strzcpy(un->sun_path, name, sizeof un->sun_path);
489 *un_len = (offsetof(struct sockaddr_un, sun_path)
490 + strlen (un->sun_path) + 1);
494 /* Clean up after make_sockaddr_un(). */
496 free_sockaddr_un(int dirfd, const char *linkname)
502 fatal_signal_unlink_file_now(linkname);
506 /* Binds Unix domain socket 'fd' to a file with permissions 0700. */
508 bind_unix_socket(int fd, struct sockaddr *sun, socklen_t sun_len)
510 /* According to _Unix Network Programming_, umask should affect bind(). */
511 mode_t old_umask = umask(0077);
512 int error = bind(fd, sun, sun_len) ? errno : 0;
517 /* Creates a Unix domain socket in the given 'style' (either SOCK_DGRAM or
518 * SOCK_STREAM) that is bound to '*bind_path' (if 'bind_path' is non-null) and
519 * connected to '*connect_path' (if 'connect_path' is non-null). If 'nonblock'
520 * is true, the socket is made non-blocking.
522 * Returns the socket's fd if successful, otherwise a negative errno value. */
524 make_unix_socket(int style, bool nonblock,
525 const char *bind_path, const char *connect_path)
530 fd = socket(PF_UNIX, style, 0);
535 /* Set nonblocking mode right away, if we want it. This prevents blocking
536 * in connect(), if connect_path != NULL. (In turn, that's a corner case:
537 * it will only happen if style is SOCK_STREAM or SOCK_SEQPACKET, and only
538 * if a backlog of un-accepted connections has built up in the kernel.) */
540 error = set_nonblocking(fd);
547 char linkname[MAX_UN_LEN + 1];
548 struct sockaddr_un un;
552 if (unlink(bind_path) && errno != ENOENT) {
553 VLOG_WARN("unlinking \"%s\": %s\n",
554 bind_path, ovs_strerror(errno));
556 fatal_signal_add_file_to_unlink(bind_path);
558 error = make_sockaddr_un(bind_path, &un, &un_len, &dirfd, linkname);
560 error = bind_unix_socket(fd, (struct sockaddr *) &un, un_len);
562 free_sockaddr_un(dirfd, linkname);
570 char linkname[MAX_UN_LEN + 1];
571 struct sockaddr_un un;
575 error = make_sockaddr_un(connect_path, &un, &un_len, &dirfd, linkname);
577 && connect(fd, (struct sockaddr*) &un, un_len)
578 && errno != EINPROGRESS) {
581 free_sockaddr_un(dirfd, linkname);
591 if (error == EAGAIN) {
595 fatal_signal_unlink_file_now(bind_path);
602 get_unix_name_len(socklen_t sun_len)
604 return (sun_len >= offsetof(struct sockaddr_un, sun_path)
605 ? sun_len - offsetof(struct sockaddr_un, sun_path)
611 guess_netmask(ovs_be32 ip_)
613 uint32_t ip = ntohl(ip_);
614 return ((ip >> 31) == 0 ? htonl(0xff000000) /* Class A */
615 : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */
616 : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */
617 : htonl(0)); /* ??? */
620 /* This is like strsep() except:
622 * - The separator string is ":".
624 * - Square brackets [] quote ":" separators and are removed from the
627 parse_bracketed_token(char **pp)
633 } else if (*p == '\0') {
636 } else if (*p == '[') {
638 char *end = start + strcspn(start, "]");
639 *pp = (*end == '\0' ? NULL
640 : end[1] == ':' ? end + 2
646 char *end = start + strcspn(start, ":");
647 *pp = *end == '\0' ? NULL : end + 1;
654 parse_sockaddr_components(struct sockaddr_storage *ss,
656 const char *port_s, uint16_t default_port,
659 struct sockaddr_in *sin = ALIGNED_CAST(struct sockaddr_in *, ss);
662 if (port_s && port_s[0]) {
663 if (!str_to_int(port_s, 10, &port) || port < 0 || port > 65535) {
664 VLOG_ERR("%s: bad port number \"%s\"", s, port_s);
670 memset(ss, 0, sizeof *ss);
671 if (strchr(host_s, ':')) {
672 struct sockaddr_in6 *sin6
673 = ALIGNED_CAST(struct sockaddr_in6 *, ss);
675 sin6->sin6_family = AF_INET6;
676 sin6->sin6_port = htons(port);
677 if (!inet_pton(AF_INET6, host_s, sin6->sin6_addr.s6_addr)) {
678 VLOG_ERR("%s: bad IPv6 address \"%s\"", s, host_s);
682 sin->sin_family = AF_INET;
683 sin->sin_port = htons(port);
684 if (!inet_pton(AF_INET, host_s, &sin->sin_addr.s_addr)) {
685 VLOG_ERR("%s: bad IPv4 address \"%s\"", s, host_s);
693 memset(ss, 0, sizeof *ss);
697 /* Parses 'target', which should be a string in the format "<host>[:<port>]".
698 * <host>, which is required, may be an IPv4 address or an IPv6 address
699 * enclosed in square brackets. If 'default_port' is nonzero then <port> is
700 * optional and defaults to 'default_port'.
702 * On success, returns true and stores the parsed remote address into '*ss'.
703 * On failure, logs an error, stores zeros into '*ss', and returns false. */
705 inet_parse_active(const char *target_, uint16_t default_port,
706 struct sockaddr_storage *ss)
708 char *target = xstrdup(target_);
715 host = parse_bracketed_token(&p);
716 port = parse_bracketed_token(&p);
718 VLOG_ERR("%s: host must be specified", target_);
720 } else if (!port && !default_port) {
721 VLOG_ERR("%s: port must be specified", target_);
724 ok = parse_sockaddr_components(ss, host, port, default_port, target_);
727 memset(ss, 0, sizeof *ss);
734 /* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style' and
735 * connects to 'target', which should be a string in the format
736 * "<host>[:<port>]". <host>, which is required, may be an IPv4 address or an
737 * IPv6 address enclosed in square brackets. If 'default_port' is nonzero then
738 * <port> is optional and defaults to 'default_port'.
740 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
742 * On success, returns 0 (indicating connection complete) or EAGAIN (indicating
743 * connection in progress), in which case the new file descriptor is stored
744 * into '*fdp'. On failure, returns a positive errno value other than EAGAIN
745 * and stores -1 into '*fdp'.
747 * If 'ss' is non-null, then on success stores the target address into '*ss'.
749 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
750 * should be in the range [0, 63] and will automatically be shifted to the
751 * appropriately place in the IP tos field. */
753 inet_open_active(int style, const char *target, uint16_t default_port,
754 struct sockaddr_storage *ssp, int *fdp, uint8_t dscp)
756 struct sockaddr_storage ss;
761 if (!inet_parse_active(target, default_port, &ss)) {
762 error = EAFNOSUPPORT;
766 /* Create non-blocking socket. */
767 fd = socket(ss.ss_family, style, 0);
769 error = sock_errno();
770 VLOG_ERR("%s: socket: %s", target, sock_strerror(error));
773 error = set_nonblocking(fd);
778 /* The dscp bits must be configured before connect() to ensure that the
779 * TOS field is set during the connection establishment. If set after
780 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
781 error = set_dscp(fd, dscp);
783 VLOG_ERR("%s: set_dscp: %s", target, sock_strerror(error));
788 error = connect(fd, (struct sockaddr *) &ss, ss_length(&ss)) == 0
791 if (error == EINPROGRESS
793 || error == WSAEALREADY || error == WSAEWOULDBLOCK
800 if (error && error != EAGAIN) {
802 memset(ssp, 0, sizeof *ssp);
817 /* Parses 'target', which should be a string in the format "[<port>][:<host>]":
819 * - If 'default_port' is -1, then <port> is required. Otherwise, if
820 * <port> is omitted, then 'default_port' is used instead.
822 * - If <port> (or 'default_port', if used) is 0, then no port is bound
823 * and the TCP/IP stack will select a port.
825 * - <host> is optional. If supplied, it may be an IPv4 address or an
826 * IPv6 address enclosed in square brackets. If omitted, the IP address
829 * If successful, stores the address into '*ss' and returns true; otherwise
830 * zeros '*ss' and returns false. */
832 inet_parse_passive(const char *target_, int default_port,
833 struct sockaddr_storage *ss)
835 char *target = xstrdup(target_);
842 port = parse_bracketed_token(&p);
843 host = parse_bracketed_token(&p);
844 if (!port && default_port < 0) {
845 VLOG_ERR("%s: port must be specified", target_);
848 ok = parse_sockaddr_components(ss, host ? host : "0.0.0.0",
849 port, default_port, target_);
852 memset(ss, 0, sizeof *ss);
859 /* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style', binds to
860 * 'target', and listens for incoming connections. Parses 'target' in the same
861 * way was inet_parse_passive().
863 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
865 * For TCP, the socket will have SO_REUSEADDR turned on.
867 * On success, returns a non-negative file descriptor. On failure, returns a
868 * negative errno value.
870 * If 'ss' is non-null, then on success stores the bound address into '*ss'.
872 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
873 * should be in the range [0, 63] and will automatically be shifted to the
874 * appropriately place in the IP tos field. */
876 inet_open_passive(int style, const char *target, int default_port,
877 struct sockaddr_storage *ssp, uint8_t dscp)
879 bool kernel_chooses_port;
880 struct sockaddr_storage ss;
882 unsigned int yes = 1;
884 if (!inet_parse_passive(target, default_port, &ss)) {
885 return -EAFNOSUPPORT;
887 kernel_chooses_port = ss_get_port(&ss) == 0;
889 /* Create non-blocking socket, set SO_REUSEADDR. */
890 fd = socket(ss.ss_family, style, 0);
892 error = sock_errno();
893 VLOG_ERR("%s: socket: %s", target, sock_strerror(error));
896 error = set_nonblocking(fd);
900 if (style == SOCK_STREAM
901 && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) {
902 error = sock_errno();
903 VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s",
904 target, sock_strerror(error));
909 if (bind(fd, (struct sockaddr *) &ss, ss_length(&ss)) < 0) {
910 error = sock_errno();
911 VLOG_ERR("%s: bind: %s", target, sock_strerror(error));
915 /* The dscp bits must be configured before connect() to ensure that the TOS
916 * field is set during the connection establishment. If set after
917 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
918 error = set_dscp(fd, dscp);
920 VLOG_ERR("%s: set_dscp: %s", target, sock_strerror(error));
925 if (style == SOCK_STREAM && listen(fd, 10) < 0) {
926 error = sock_errno();
927 VLOG_ERR("%s: listen: %s", target, sock_strerror(error));
931 if (ssp || kernel_chooses_port) {
932 socklen_t ss_len = sizeof ss;
933 if (getsockname(fd, (struct sockaddr *) &ss, &ss_len) < 0) {
934 error = sock_errno();
935 VLOG_ERR("%s: getsockname: %s", target, sock_strerror(error));
938 if (kernel_chooses_port) {
939 VLOG_INFO("%s: listening on port %"PRIu16,
940 target, ss_get_port(&ss));
951 memset(ssp, 0, sizeof *ssp);
958 read_fully(int fd, void *p_, size_t size, size_t *bytes_read)
964 ssize_t retval = read(fd, p, size);
966 *bytes_read += retval;
969 } else if (retval == 0) {
971 } else if (errno != EINTR) {
979 write_fully(int fd, const void *p_, size_t size, size_t *bytes_written)
981 const uint8_t *p = p_;
985 ssize_t retval = write(fd, p, size);
987 *bytes_written += retval;
990 } else if (retval == 0) {
991 VLOG_WARN("write returned 0");
993 } else if (errno != EINTR) {
1000 /* Given file name 'file_name', fsyncs the directory in which it is contained.
1001 * Returns 0 if successful, otherwise a positive errno value. */
1003 fsync_parent_dir(const char *file_name)
1010 dir = dir_name(file_name);
1011 fd = open(dir, O_RDONLY);
1014 if (errno == EINVAL || errno == EROFS) {
1015 /* This directory does not support synchronization. Not
1016 * really an error. */
1019 VLOG_ERR("%s: fsync failed (%s)", dir, ovs_strerror(error));
1025 VLOG_ERR("%s: open failed (%s)", dir, ovs_strerror(error));
1033 /* Obtains the modification time of the file named 'file_name' to the greatest
1034 * supported precision. If successful, stores the mtime in '*mtime' and
1035 * returns 0. On error, returns a positive errno value and stores zeros in
1038 get_mtime(const char *file_name, struct timespec *mtime)
1042 if (!stat(file_name, &s)) {
1043 mtime->tv_sec = s.st_mtime;
1045 #if HAVE_STRUCT_STAT_ST_MTIM_TV_NSEC
1046 mtime->tv_nsec = s.st_mtim.tv_nsec;
1047 #elif HAVE_STRUCT_STAT_ST_MTIMENSEC
1048 mtime->tv_nsec = s.st_mtimensec;
1055 mtime->tv_sec = mtime->tv_nsec = 0;
1065 VLOG_FATAL("failed to create pipe (%s)", ovs_strerror(errno));
1070 xpipe_nonblocking(int fds[2])
1073 xset_nonblocking(fds[0]);
1074 xset_nonblocking(fds[1]);
1079 getsockopt_int(int fd, int level, int option, const char *optname, int *valuep)
1081 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
1087 if (getsockopt(fd, level, option, &value, &len)) {
1088 error = sock_errno();
1089 VLOG_ERR_RL(&rl, "getsockopt(%s): %s", optname, sock_strerror(error));
1090 } else if (len != sizeof value) {
1092 VLOG_ERR_RL(&rl, "getsockopt(%s): value is %u bytes (expected %"PRIuSIZE")",
1093 optname, (unsigned int) len, sizeof value);
1098 *valuep = error ? 0 : value;
1103 describe_sockaddr(struct ds *string, int fd,
1104 int (*getaddr)(int, struct sockaddr *, socklen_t *))
1106 struct sockaddr_storage ss;
1107 socklen_t len = sizeof ss;
1109 if (!getaddr(fd, (struct sockaddr *) &ss, &len)) {
1110 if (ss.ss_family == AF_INET || ss.ss_family == AF_INET6) {
1111 char addrbuf[SS_NTOP_BUFSIZE];
1113 ds_put_format(string, "%s:%"PRIu16,
1114 ss_format_address(&ss, addrbuf, sizeof addrbuf),
1117 } else if (ss.ss_family == AF_UNIX) {
1118 struct sockaddr_un sun;
1122 memcpy(&sun, &ss, sizeof sun);
1123 maxlen = len - offsetof(struct sockaddr_un, sun_path);
1124 null = memchr(sun.sun_path, '\0', maxlen);
1125 ds_put_buffer(string, sun.sun_path,
1126 null ? null - sun.sun_path : maxlen);
1130 else if (ss.ss_family == AF_NETLINK) {
1133 /* SO_PROTOCOL was introduced in 2.6.32. Support it regardless of the version
1134 * of the Linux kernel headers in use at build time. */
1136 #define SO_PROTOCOL 38
1139 if (!getsockopt_int(fd, SOL_SOCKET, SO_PROTOCOL, "SO_PROTOCOL",
1143 ds_put_cstr(string, "NETLINK_ROUTE");
1146 case NETLINK_GENERIC:
1147 ds_put_cstr(string, "NETLINK_GENERIC");
1151 ds_put_format(string, "AF_NETLINK family %d", protocol);
1155 ds_put_cstr(string, "AF_NETLINK");
1160 else if (ss.ss_family == AF_PACKET) {
1161 struct sockaddr_ll sll;
1163 memcpy(&sll, &ss, sizeof sll);
1164 ds_put_cstr(string, "AF_PACKET");
1165 if (sll.sll_ifindex) {
1166 char name[IFNAMSIZ];
1168 if (if_indextoname(sll.sll_ifindex, name)) {
1169 ds_put_format(string, "(%s)", name);
1171 ds_put_format(string, "(ifindex=%d)", sll.sll_ifindex);
1174 if (sll.sll_protocol) {
1175 ds_put_format(string, "(protocol=0x%"PRIu16")",
1176 ntohs(sll.sll_protocol));
1180 else if (ss.ss_family == AF_UNSPEC) {
1181 ds_put_cstr(string, "AF_UNSPEC");
1183 ds_put_format(string, "AF_%d", (int) ss.ss_family);
1191 put_fd_filename(struct ds *string, int fd)
1197 linkname = xasprintf("/proc/self/fd/%d", fd);
1198 n = readlink(linkname, buf, sizeof buf);
1200 ds_put_char(string, ' ');
1201 ds_put_buffer(string, buf, n);
1202 if (n > sizeof buf) {
1203 ds_put_cstr(string, "...");
1210 /* Returns a malloc()'d string describing 'fd', for use in logging. */
1219 if (fstat(fd, &s)) {
1220 ds_put_format(&string, "fstat failed (%s)", ovs_strerror(errno));
1221 } else if (S_ISSOCK(s.st_mode)) {
1222 describe_sockaddr(&string, fd, getsockname);
1223 ds_put_cstr(&string, "<->");
1224 describe_sockaddr(&string, fd, getpeername);
1226 ds_put_cstr(&string, (isatty(fd) ? "tty"
1227 : S_ISDIR(s.st_mode) ? "directory"
1228 : S_ISCHR(s.st_mode) ? "character device"
1229 : S_ISBLK(s.st_mode) ? "block device"
1230 : S_ISREG(s.st_mode) ? "file"
1231 : S_ISFIFO(s.st_mode) ? "FIFO"
1232 : S_ISLNK(s.st_mode) ? "symbolic link"
1235 put_fd_filename(&string, fd);
1239 ds_put_format(&string,"file descriptor");
1241 return ds_steal_cstr(&string);
1245 /* Calls ioctl() on an AF_INET sock, passing the specified 'command' and
1246 * 'arg'. Returns 0 if successful, otherwise a positive errno value. */
1248 af_inet_ioctl(unsigned long int command, const void *arg)
1250 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
1253 if (ovsthread_once_start(&once)) {
1254 sock = socket(AF_INET, SOCK_DGRAM, 0);
1256 int error = sock_errno();
1257 VLOG_ERR("failed to create inet socket: %s", sock_strerror(error));
1260 ovsthread_once_done(&once);
1263 return (sock < 0 ? -sock
1264 : ioctl(sock, command, arg) == -1 ? errno
1269 af_inet_ifreq_ioctl(const char *name, struct ifreq *ifr, unsigned long int cmd,
1270 const char *cmd_name)
1274 ovs_strzcpy(ifr->ifr_name, name, sizeof ifr->ifr_name);
1275 error = af_inet_ioctl(cmd, ifr);
1277 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1278 VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name,
1279 ovs_strerror(error));
1285 /* sockaddr_storage helpers. */
1287 /* Returns the IPv4 or IPv6 port in 'ss'. */
1289 ss_get_port(const struct sockaddr_storage *ss)
1291 if (ss->ss_family == AF_INET) {
1292 const struct sockaddr_in *sin
1293 = ALIGNED_CAST(const struct sockaddr_in *, ss);
1294 return ntohs(sin->sin_port);
1295 } else if (ss->ss_family == AF_INET6) {
1296 const struct sockaddr_in6 *sin6
1297 = ALIGNED_CAST(const struct sockaddr_in6 *, ss);
1298 return ntohs(sin6->sin6_port);
1304 /* Formats the IPv4 or IPv6 address in 'ss' into the 'bufsize' bytes in 'buf'.
1305 * If 'ss' is an IPv6 address, puts square brackets around the address.
1306 * 'bufsize' should be at least SS_NTOP_BUFSIZE.
1310 ss_format_address(const struct sockaddr_storage *ss,
1311 char *buf, size_t bufsize)
1313 ovs_assert(bufsize >= SS_NTOP_BUFSIZE);
1314 if (ss->ss_family == AF_INET) {
1315 const struct sockaddr_in *sin
1316 = ALIGNED_CAST(const struct sockaddr_in *, ss);
1318 snprintf(buf, bufsize, IP_FMT, IP_ARGS(sin->sin_addr.s_addr));
1319 } else if (ss->ss_family == AF_INET6) {
1320 const struct sockaddr_in6 *sin6
1321 = ALIGNED_CAST(const struct sockaddr_in6 *, ss);
1324 inet_ntop(AF_INET6, sin6->sin6_addr.s6_addr, buf + 1, bufsize - 1);
1325 strcpy(strchr(buf, '\0'), "]");
1334 ss_length(const struct sockaddr_storage *ss)
1336 switch (ss->ss_family) {
1338 return sizeof(struct sockaddr_in);
1341 return sizeof(struct sockaddr_in6);
1348 /* For Windows socket calls, 'errno' is not set. One has to call
1349 * WSAGetLastError() to get the error number and then pass it to
1350 * this function to get the correct error string.
1352 * ovs_strerror() calls strerror_r() and would not get the correct error
1353 * string for Windows sockets, but is good for POSIX. */
1355 sock_strerror(int error)
1358 return ovs_format_message(error);
1360 return ovs_strerror(error);