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"
42 #if AF_PACKET && LINUX_DATAPATH
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_DATAPATH)" instead of "#ifdef __linux__". */
55 #ifndef LINUX_DATAPATH
56 #define LINUX_DATAPATH 0
63 /* Maximum length of the sun_path member in a struct sockaddr_un, excluding
64 * space for a null terminator. */
65 #define MAX_UN_LEN (sizeof(((struct sockaddr_un *) 0)->sun_path) - 1)
67 static int getsockopt_int(int fd, int level, int option, const char *optname,
70 /* Sets 'fd' to non-blocking mode. Returns 0 if successful, otherwise a
71 * positive errno value. */
73 set_nonblocking(int fd)
76 int flags = fcntl(fd, F_GETFL, 0);
78 if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) != -1) {
81 VLOG_ERR("fcntl(F_SETFL) failed: %s", ovs_strerror(errno));
85 VLOG_ERR("fcntl(F_GETFL) failed: %s", ovs_strerror(errno));
89 unsigned long arg = 1;
90 if (ioctlsocket(fd, FIONBIO, &arg)) {
91 int error = sock_errno();
92 VLOG_ERR("set_nonblocking failed: %s", sock_strerror(error));
100 xset_nonblocking(int fd)
102 if (set_nonblocking(fd)) {
108 set_dscp(int fd, uint8_t dscp)
117 if (setsockopt(fd, IPPROTO_IP, IP_TOS, &val, sizeof val)) {
124 /* Translates 'host_name', which must be a string representation of an IP
125 * address, into a numeric IP address in '*addr'. Returns 0 if successful,
126 * otherwise a positive errno value. */
128 lookup_ip(const char *host_name, struct in_addr *addr)
130 if (!inet_pton(AF_INET, host_name, addr)) {
131 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
132 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name);
138 /* Translates 'host_name', which must be a string representation of an IPv6
139 * address, into a numeric IPv6 address in '*addr'. Returns 0 if successful,
140 * otherwise a positive errno value. */
142 lookup_ipv6(const char *host_name, struct in6_addr *addr)
144 if (inet_pton(AF_INET6, host_name, addr) != 1) {
145 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
146 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IPv6 address", host_name);
152 /* Translates 'host_name', which must be a host name or a string representation
153 * of an IP address, into a numeric IP address in '*addr'. Returns 0 if
154 * successful, otherwise a positive errno value.
156 * Most Open vSwitch code should not use this because it causes deadlocks:
157 * getaddrinfo() sends out a DNS request but that starts a new flow for which
158 * OVS must set up a flow, but it can't because it's waiting for a DNS reply.
159 * The synchronous lookup also delays other activity. (Of course we can solve
160 * this but it doesn't seem worthwhile quite yet.) */
162 lookup_hostname(const char *host_name, struct in_addr *addr)
164 struct addrinfo *result;
165 struct addrinfo hints;
167 if (inet_pton(AF_INET, host_name, addr)) {
171 memset(&hints, 0, sizeof hints);
172 hints.ai_family = AF_INET;
174 switch (getaddrinfo(host_name, NULL, &hints, &result)) {
176 *addr = ALIGNED_CAST(struct sockaddr_in *,
177 result->ai_addr)->sin_addr;
178 freeaddrinfo(result);
181 #ifdef EAI_ADDRFAMILY
202 #if defined (EAI_NODATA) && EAI_NODATA != EAI_NONAME
218 check_connection_completion(int fd)
220 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
225 pfd.events = POLLOUT;
229 retval = poll(&pfd, 1, 0);
230 } while (retval < 0 && errno == EINTR);
232 retval = WSAPoll(&pfd, 1, 0);
235 if (pfd.revents & POLLERR) {
236 ssize_t n = send(fd, "", 1, 0);
240 VLOG_ERR_RL(&rl, "poll return POLLERR but send succeeded");
245 } else if (retval < 0) {
246 VLOG_ERR_RL(&rl, "poll: %s", sock_strerror(sock_errno()));
253 /* Drain all the data currently in the receive queue of a datagram socket (and
254 * possibly additional data). There is no way to know how many packets are in
255 * the receive queue, but we do know that the total number of bytes queued does
256 * not exceed the receive buffer size, so we pull packets until none are left
257 * or we've read that many bytes. */
263 rcvbuf = get_socket_rcvbuf(fd);
269 /* In Linux, specifying MSG_TRUNC in the flags argument causes the
270 * datagram length to be returned, even if that is longer than the
271 * buffer provided. Thus, we can use a 1-byte buffer to discard the
272 * incoming datagram and still be able to account how many bytes were
273 * removed from the receive buffer.
275 * On other Unix-like OSes, MSG_TRUNC has no effect in the flags
277 char buffer[LINUX_DATAPATH ? 1 : 2048];
278 ssize_t n_bytes = recv(fd, buffer, sizeof buffer, MSG_TRUNC);
279 if (n_bytes <= 0 || n_bytes >= rcvbuf) {
287 /* Returns the size of socket 'sock''s receive buffer (SO_RCVBUF), or a
288 * negative errno value if an error occurs. */
290 get_socket_rcvbuf(int sock)
295 error = getsockopt_int(sock, SOL_SOCKET, SO_RCVBUF, "SO_RCVBUF", &rcvbuf);
296 return error ? -error : rcvbuf;
299 /* Reads and discards up to 'n' datagrams from 'fd', stopping as soon as no
300 * more data can be immediately read. ('fd' should therefore be in
301 * non-blocking mode.)*/
303 drain_fd(int fd, size_t n_packets)
305 for (; n_packets > 0; n_packets--) {
306 /* 'buffer' only needs to be 1 byte long in most circumstances. This
307 * size is defensive against the possibility that we someday want to
308 * use a Linux tap device without TUN_NO_PI, in which case a buffer
309 * smaller than sizeof(struct tun_pi) will give EINVAL on read. */
311 if (read(fd, buffer, sizeof buffer) <= 0) {
318 /* Attempts to shorten 'name' by opening a file descriptor for the directory
319 * part of the name and indirecting through /proc/self/fd/<dirfd>/<basename>.
320 * On systems with Linux-like /proc, this works as long as <basename> isn't too
323 * On success, returns 0 and stores the short name in 'short_name' and a
324 * directory file descriptor to eventually be closed in '*dirfpd'. */
326 shorten_name_via_proc(const char *name, char short_name[MAX_UN_LEN + 1],
333 if (!LINUX_DATAPATH) {
337 dir = dir_name(name);
338 dirfd = open(dir, O_DIRECTORY | O_RDONLY);
340 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
343 VLOG_WARN_RL(&rl, "%s: open failed (%s)", dir, ovs_strerror(error));
350 base = base_name(name);
351 len = snprintf(short_name, MAX_UN_LEN + 1,
352 "/proc/self/fd/%d/%s", dirfd, base);
355 if (len >= 0 && len <= MAX_UN_LEN) {
364 /* Attempts to shorten 'name' by creating a symlink for the directory part of
365 * the name and indirecting through <symlink>/<basename>. This works on
366 * systems that support symlinks, as long as <basename> isn't too long.
368 * On success, returns 0 and stores the short name in 'short_name' and the
369 * symbolic link to eventually delete in 'linkname'. */
371 shorten_name_via_symlink(const char *name, char short_name[MAX_UN_LEN + 1],
372 char linkname[MAX_UN_LEN + 1])
374 char *abs, *dir, *base;
379 abs = abs_file_name(NULL, name);
381 base = base_name(abs);
384 tmpdir = getenv("TMPDIR");
385 if (tmpdir == NULL) {
389 for (i = 0; i < 1000; i++) {
392 len = snprintf(linkname, MAX_UN_LEN + 1,
393 "%s/ovs-un-c-%"PRIu32, tmpdir, random_uint32());
394 error = (len < 0 || len > MAX_UN_LEN ? ENAMETOOLONG
395 : symlink(dir, linkname) ? errno
397 if (error != EEXIST) {
405 fatal_signal_add_file_to_unlink(linkname);
407 len = snprintf(short_name, MAX_UN_LEN + 1, "%s/%s", linkname, base);
408 if (len < 0 || len > MAX_UN_LEN) {
409 fatal_signal_unlink_file_now(linkname);
410 error = ENAMETOOLONG;
423 /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in
424 * '*un_len' the size of the sockaddr_un.
426 * Returns 0 on success, otherwise a positive errno value.
428 * Uses '*dirfdp' and 'linkname' to store references to data when the caller no
429 * longer needs to use 'un'. On success, freeing these references with
430 * free_sockaddr_un() is mandatory to avoid a leak; on failure, freeing them is
431 * unnecessary but harmless. */
433 make_sockaddr_un(const char *name, struct sockaddr_un *un, socklen_t *un_len,
434 int *dirfdp, char linkname[MAX_UN_LEN + 1])
436 char short_name[MAX_UN_LEN + 1];
440 if (strlen(name) > MAX_UN_LEN) {
441 /* 'name' is too long to fit in a sockaddr_un. Try a workaround. */
442 int error = shorten_name_via_proc(name, short_name, dirfdp);
443 if (error == ENAMETOOLONG) {
444 error = shorten_name_via_symlink(name, short_name, linkname);
447 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
449 VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum "
450 "%"PRIuSIZE" bytes", name, MAX_UN_LEN);
457 un->sun_family = AF_UNIX;
458 ovs_strzcpy(un->sun_path, name, sizeof un->sun_path);
459 *un_len = (offsetof(struct sockaddr_un, sun_path)
460 + strlen (un->sun_path) + 1);
464 /* Clean up after make_sockaddr_un(). */
466 free_sockaddr_un(int dirfd, const char *linkname)
472 fatal_signal_unlink_file_now(linkname);
476 /* Binds Unix domain socket 'fd' to a file with permissions 0700. */
478 bind_unix_socket(int fd, struct sockaddr *sun, socklen_t sun_len)
480 /* According to _Unix Network Programming_, umask should affect bind(). */
481 mode_t old_umask = umask(0077);
482 int error = bind(fd, sun, sun_len) ? errno : 0;
487 /* Creates a Unix domain socket in the given 'style' (either SOCK_DGRAM or
488 * SOCK_STREAM) that is bound to '*bind_path' (if 'bind_path' is non-null) and
489 * connected to '*connect_path' (if 'connect_path' is non-null). If 'nonblock'
490 * is true, the socket is made non-blocking.
492 * Returns the socket's fd if successful, otherwise a negative errno value. */
494 make_unix_socket(int style, bool nonblock,
495 const char *bind_path, const char *connect_path)
500 fd = socket(PF_UNIX, style, 0);
505 /* Set nonblocking mode right away, if we want it. This prevents blocking
506 * in connect(), if connect_path != NULL. (In turn, that's a corner case:
507 * it will only happen if style is SOCK_STREAM or SOCK_SEQPACKET, and only
508 * if a backlog of un-accepted connections has built up in the kernel.) */
510 error = set_nonblocking(fd);
517 char linkname[MAX_UN_LEN + 1];
518 struct sockaddr_un un;
522 if (unlink(bind_path) && errno != ENOENT) {
523 VLOG_WARN("unlinking \"%s\": %s\n",
524 bind_path, ovs_strerror(errno));
526 fatal_signal_add_file_to_unlink(bind_path);
528 error = make_sockaddr_un(bind_path, &un, &un_len, &dirfd, linkname);
530 error = bind_unix_socket(fd, (struct sockaddr *) &un, un_len);
532 free_sockaddr_un(dirfd, linkname);
540 char linkname[MAX_UN_LEN + 1];
541 struct sockaddr_un un;
545 error = make_sockaddr_un(connect_path, &un, &un_len, &dirfd, linkname);
547 && connect(fd, (struct sockaddr*) &un, un_len)
548 && errno != EINPROGRESS) {
551 free_sockaddr_un(dirfd, linkname);
561 if (error == EAGAIN) {
565 fatal_signal_unlink_file_now(bind_path);
572 get_unix_name_len(socklen_t sun_len)
574 return (sun_len >= offsetof(struct sockaddr_un, sun_path)
575 ? sun_len - offsetof(struct sockaddr_un, sun_path)
581 guess_netmask(ovs_be32 ip_)
583 uint32_t ip = ntohl(ip_);
584 return ((ip >> 31) == 0 ? htonl(0xff000000) /* Class A */
585 : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */
586 : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */
587 : htonl(0)); /* ??? */
590 /* This is like strsep() except:
592 * - The separator string is ":".
594 * - Square brackets [] quote ":" separators and are removed from the
597 parse_bracketed_token(char **pp)
603 } else if (*p == '\0') {
606 } else if (*p == '[') {
608 char *end = start + strcspn(start, "]");
609 *pp = (*end == '\0' ? NULL
610 : end[1] == ':' ? end + 2
616 char *end = start + strcspn(start, ":");
617 *pp = *end == '\0' ? NULL : end + 1;
624 parse_sockaddr_components(struct sockaddr_storage *ss,
626 const char *port_s, uint16_t default_port,
629 struct sockaddr_in *sin = ALIGNED_CAST(struct sockaddr_in *, ss);
632 if (port_s && port_s[0]) {
633 if (!str_to_int(port_s, 10, &port) || port < 0 || port > 65535) {
634 VLOG_ERR("%s: bad port number \"%s\"", s, port_s);
640 memset(ss, 0, sizeof *ss);
641 if (strchr(host_s, ':')) {
642 struct sockaddr_in6 *sin6
643 = ALIGNED_CAST(struct sockaddr_in6 *, ss);
645 sin6->sin6_family = AF_INET6;
646 sin6->sin6_port = htons(port);
647 if (!inet_pton(AF_INET6, host_s, sin6->sin6_addr.s6_addr)) {
648 VLOG_ERR("%s: bad IPv6 address \"%s\"", s, host_s);
652 sin->sin_family = AF_INET;
653 sin->sin_port = htons(port);
654 if (!inet_pton(AF_INET, host_s, &sin->sin_addr.s_addr)) {
655 VLOG_ERR("%s: bad IPv4 address \"%s\"", s, host_s);
663 memset(ss, 0, sizeof *ss);
667 /* Parses 'target', which should be a string in the format "<host>[:<port>]".
668 * <host>, which is required, may be an IPv4 address or an IPv6 address
669 * enclosed in square brackets. If 'default_port' is nonzero then <port> is
670 * optional and defaults to 'default_port'.
672 * On success, returns true and stores the parsed remote address into '*ss'.
673 * On failure, logs an error, stores zeros into '*ss', and returns false. */
675 inet_parse_active(const char *target_, uint16_t default_port,
676 struct sockaddr_storage *ss)
678 char *target = xstrdup(target_);
685 host = parse_bracketed_token(&p);
686 port = parse_bracketed_token(&p);
688 VLOG_ERR("%s: host must be specified", target_);
690 } else if (!port && !default_port) {
691 VLOG_ERR("%s: port must be specified", target_);
694 ok = parse_sockaddr_components(ss, host, port, default_port, target_);
697 memset(ss, 0, sizeof *ss);
704 /* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style' and
705 * connects to 'target', which should be a string in the format
706 * "<host>[:<port>]". <host>, which is required, may be an IPv4 address or an
707 * IPv6 address enclosed in square brackets. If 'default_port' is nonzero then
708 * <port> is optional and defaults to 'default_port'.
710 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
712 * On success, returns 0 (indicating connection complete) or EAGAIN (indicating
713 * connection in progress), in which case the new file descriptor is stored
714 * into '*fdp'. On failure, returns a positive errno value other than EAGAIN
715 * and stores -1 into '*fdp'.
717 * If 'ss' is non-null, then on success stores the target address into '*ss'.
719 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
720 * should be in the range [0, 63] and will automatically be shifted to the
721 * appropriately place in the IP tos field. */
723 inet_open_active(int style, const char *target, uint16_t default_port,
724 struct sockaddr_storage *ssp, int *fdp, uint8_t dscp)
726 struct sockaddr_storage ss;
731 if (!inet_parse_active(target, default_port, &ss)) {
732 error = EAFNOSUPPORT;
736 /* Create non-blocking socket. */
737 fd = socket(ss.ss_family, style, 0);
739 error = sock_errno();
740 VLOG_ERR("%s: socket: %s", target, sock_strerror(error));
743 error = set_nonblocking(fd);
748 /* The dscp bits must be configured before connect() to ensure that the
749 * TOS field is set during the connection establishment. If set after
750 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
751 error = set_dscp(fd, dscp);
753 VLOG_ERR("%s: socket: %s", target, sock_strerror(error));
758 error = connect(fd, (struct sockaddr *) &ss, ss_length(&ss)) == 0
761 if (error == EINPROGRESS
763 || error == WSAEALREADY || error == WSAEWOULDBLOCK
770 if (error && error != EAGAIN) {
772 memset(ssp, 0, sizeof *ssp);
787 /* Parses 'target', which should be a string in the format "[<port>][:<host>]":
789 * - If 'default_port' is -1, then <port> is required. Otherwise, if
790 * <port> is omitted, then 'default_port' is used instead.
792 * - If <port> (or 'default_port', if used) is 0, then no port is bound
793 * and the TCP/IP stack will select a port.
795 * - <host> is optional. If supplied, it may be an IPv4 address or an
796 * IPv6 address enclosed in square brackets. If omitted, the IP address
799 * If successful, stores the address into '*ss' and returns true; otherwise
800 * zeros '*ss' and returns false. */
802 inet_parse_passive(const char *target_, int default_port,
803 struct sockaddr_storage *ss)
805 char *target = xstrdup(target_);
812 port = parse_bracketed_token(&p);
813 host = parse_bracketed_token(&p);
814 if (!port && default_port < 0) {
815 VLOG_ERR("%s: port must be specified", target_);
818 ok = parse_sockaddr_components(ss, host ? host : "0.0.0.0",
819 port, default_port, target_);
822 memset(ss, 0, sizeof *ss);
829 /* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style', binds to
830 * 'target', and listens for incoming connections. Parses 'target' in the same
831 * way was inet_parse_passive().
833 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
835 * For TCP, the socket will have SO_REUSEADDR turned on.
837 * On success, returns a non-negative file descriptor. On failure, returns a
838 * negative errno value.
840 * If 'ss' is non-null, then on success stores the bound address into '*ss'.
842 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
843 * should be in the range [0, 63] and will automatically be shifted to the
844 * appropriately place in the IP tos field. */
846 inet_open_passive(int style, const char *target, int default_port,
847 struct sockaddr_storage *ssp, uint8_t dscp)
849 bool kernel_chooses_port;
850 struct sockaddr_storage ss;
852 unsigned int yes = 1;
854 if (!inet_parse_passive(target, default_port, &ss)) {
855 return -EAFNOSUPPORT;
857 kernel_chooses_port = ss_get_port(&ss) == 0;
859 /* Create non-blocking socket, set SO_REUSEADDR. */
860 fd = socket(ss.ss_family, style, 0);
862 error = sock_errno();
863 VLOG_ERR("%s: socket: %s", target, sock_strerror(error));
866 error = set_nonblocking(fd);
870 if (style == SOCK_STREAM
871 && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) {
872 error = sock_errno();
873 VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s",
874 target, sock_strerror(error));
879 if (bind(fd, (struct sockaddr *) &ss, ss_length(&ss)) < 0) {
880 error = sock_errno();
881 VLOG_ERR("%s: bind: %s", target, sock_strerror(error));
885 /* The dscp bits must be configured before connect() to ensure that the TOS
886 * field is set during the connection establishment. If set after
887 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
888 error = set_dscp(fd, dscp);
890 VLOG_ERR("%s: socket: %s", target, sock_strerror(error));
895 if (style == SOCK_STREAM && listen(fd, 10) < 0) {
896 error = sock_errno();
897 VLOG_ERR("%s: listen: %s", target, sock_strerror(error));
901 if (ssp || kernel_chooses_port) {
902 socklen_t ss_len = sizeof ss;
903 if (getsockname(fd, (struct sockaddr *) &ss, &ss_len) < 0) {
904 error = sock_errno();
905 VLOG_ERR("%s: getsockname: %s", target, sock_strerror(error));
908 if (kernel_chooses_port) {
909 VLOG_INFO("%s: listening on port %"PRIu16,
910 target, ss_get_port(&ss));
921 memset(ssp, 0, sizeof *ssp);
928 read_fully(int fd, void *p_, size_t size, size_t *bytes_read)
934 ssize_t retval = read(fd, p, size);
936 *bytes_read += retval;
939 } else if (retval == 0) {
941 } else if (errno != EINTR) {
949 write_fully(int fd, const void *p_, size_t size, size_t *bytes_written)
951 const uint8_t *p = p_;
955 ssize_t retval = write(fd, p, size);
957 *bytes_written += retval;
960 } else if (retval == 0) {
961 VLOG_WARN("write returned 0");
963 } else if (errno != EINTR) {
970 /* Given file name 'file_name', fsyncs the directory in which it is contained.
971 * Returns 0 if successful, otherwise a positive errno value. */
973 fsync_parent_dir(const char *file_name)
980 dir = dir_name(file_name);
981 fd = open(dir, O_RDONLY);
984 if (errno == EINVAL || errno == EROFS) {
985 /* This directory does not support synchronization. Not
986 * really an error. */
989 VLOG_ERR("%s: fsync failed (%s)", dir, ovs_strerror(error));
995 VLOG_ERR("%s: open failed (%s)", dir, ovs_strerror(error));
1003 /* Obtains the modification time of the file named 'file_name' to the greatest
1004 * supported precision. If successful, stores the mtime in '*mtime' and
1005 * returns 0. On error, returns a positive errno value and stores zeros in
1008 get_mtime(const char *file_name, struct timespec *mtime)
1012 if (!stat(file_name, &s)) {
1013 mtime->tv_sec = s.st_mtime;
1015 #if HAVE_STRUCT_STAT_ST_MTIM_TV_NSEC
1016 mtime->tv_nsec = s.st_mtim.tv_nsec;
1017 #elif HAVE_STRUCT_STAT_ST_MTIMENSEC
1018 mtime->tv_nsec = s.st_mtimensec;
1025 mtime->tv_sec = mtime->tv_nsec = 0;
1034 VLOG_FATAL("failed to create pipe (%s)", ovs_strerror(errno));
1039 xpipe_nonblocking(int fds[2])
1042 xset_nonblocking(fds[0]);
1043 xset_nonblocking(fds[1]);
1047 getsockopt_int(int fd, int level, int option, const char *optname, int *valuep)
1049 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
1055 if (getsockopt(fd, level, option, &value, &len)) {
1056 error = sock_errno();
1057 VLOG_ERR_RL(&rl, "getsockopt(%s): %s", optname, sock_strerror(error));
1058 } else if (len != sizeof value) {
1060 VLOG_ERR_RL(&rl, "getsockopt(%s): value is %u bytes (expected %"PRIuSIZE")",
1061 optname, (unsigned int) len, sizeof value);
1066 *valuep = error ? 0 : value;
1071 describe_sockaddr(struct ds *string, int fd,
1072 int (*getaddr)(int, struct sockaddr *, socklen_t *))
1074 struct sockaddr_storage ss;
1075 socklen_t len = sizeof ss;
1077 if (!getaddr(fd, (struct sockaddr *) &ss, &len)) {
1078 if (ss.ss_family == AF_INET || ss.ss_family == AF_INET6) {
1079 char addrbuf[SS_NTOP_BUFSIZE];
1081 ds_put_format(string, "%s:%"PRIu16,
1082 ss_format_address(&ss, addrbuf, sizeof addrbuf),
1085 } else if (ss.ss_family == AF_UNIX) {
1086 struct sockaddr_un sun;
1090 memcpy(&sun, &ss, sizeof sun);
1091 maxlen = len - offsetof(struct sockaddr_un, sun_path);
1092 null = memchr(sun.sun_path, '\0', maxlen);
1093 ds_put_buffer(string, sun.sun_path,
1094 null ? null - sun.sun_path : maxlen);
1098 else if (ss.ss_family == AF_NETLINK) {
1101 /* SO_PROTOCOL was introduced in 2.6.32. Support it regardless of the version
1102 * of the Linux kernel headers in use at build time. */
1104 #define SO_PROTOCOL 38
1107 if (!getsockopt_int(fd, SOL_SOCKET, SO_PROTOCOL, "SO_PROTOCOL",
1111 ds_put_cstr(string, "NETLINK_ROUTE");
1114 case NETLINK_GENERIC:
1115 ds_put_cstr(string, "NETLINK_GENERIC");
1119 ds_put_format(string, "AF_NETLINK family %d", protocol);
1123 ds_put_cstr(string, "AF_NETLINK");
1127 #if AF_PACKET && LINUX_DATAPATH
1128 else if (ss.ss_family == AF_PACKET) {
1129 struct sockaddr_ll sll;
1131 memcpy(&sll, &ss, sizeof sll);
1132 ds_put_cstr(string, "AF_PACKET");
1133 if (sll.sll_ifindex) {
1134 char name[IFNAMSIZ];
1136 if (if_indextoname(sll.sll_ifindex, name)) {
1137 ds_put_format(string, "(%s)", name);
1139 ds_put_format(string, "(ifindex=%d)", sll.sll_ifindex);
1142 if (sll.sll_protocol) {
1143 ds_put_format(string, "(protocol=0x%"PRIu16")",
1144 ntohs(sll.sll_protocol));
1148 else if (ss.ss_family == AF_UNSPEC) {
1149 ds_put_cstr(string, "AF_UNSPEC");
1151 ds_put_format(string, "AF_%d", (int) ss.ss_family);
1157 #ifdef LINUX_DATAPATH
1159 put_fd_filename(struct ds *string, int fd)
1165 linkname = xasprintf("/proc/self/fd/%d", fd);
1166 n = readlink(linkname, buf, sizeof buf);
1168 ds_put_char(string, ' ');
1169 ds_put_buffer(string, buf, n);
1170 if (n > sizeof buf) {
1171 ds_put_cstr(string, "...");
1178 /* Returns a malloc()'d string describing 'fd', for use in logging. */
1187 if (fstat(fd, &s)) {
1188 ds_put_format(&string, "fstat failed (%s)", ovs_strerror(errno));
1189 } else if (S_ISSOCK(s.st_mode)) {
1190 describe_sockaddr(&string, fd, getsockname);
1191 ds_put_cstr(&string, "<->");
1192 describe_sockaddr(&string, fd, getpeername);
1194 ds_put_cstr(&string, (isatty(fd) ? "tty"
1195 : S_ISDIR(s.st_mode) ? "directory"
1196 : S_ISCHR(s.st_mode) ? "character device"
1197 : S_ISBLK(s.st_mode) ? "block device"
1198 : S_ISREG(s.st_mode) ? "file"
1199 : S_ISFIFO(s.st_mode) ? "FIFO"
1200 : S_ISLNK(s.st_mode) ? "symbolic link"
1202 #ifdef LINUX_DATAPATH
1203 put_fd_filename(&string, fd);
1207 ds_put_format(&string,"file descriptor");
1209 return ds_steal_cstr(&string);
1213 /* Calls ioctl() on an AF_INET sock, passing the specified 'command' and
1214 * 'arg'. Returns 0 if successful, otherwise a positive errno value. */
1216 af_inet_ioctl(unsigned long int command, const void *arg)
1218 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
1221 if (ovsthread_once_start(&once)) {
1222 sock = socket(AF_INET, SOCK_DGRAM, 0);
1224 int error = sock_errno();
1225 VLOG_ERR("failed to create inet socket: %s", sock_strerror(error));
1228 ovsthread_once_done(&once);
1231 return (sock < 0 ? -sock
1232 : ioctl(sock, command, arg) == -1 ? errno
1237 af_inet_ifreq_ioctl(const char *name, struct ifreq *ifr, unsigned long int cmd,
1238 const char *cmd_name)
1242 ovs_strzcpy(ifr->ifr_name, name, sizeof ifr->ifr_name);
1243 error = af_inet_ioctl(cmd, ifr);
1245 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1246 VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name,
1247 ovs_strerror(error));
1253 /* sockaddr_storage helpers. */
1255 /* Returns the IPv4 or IPv6 port in 'ss'. */
1257 ss_get_port(const struct sockaddr_storage *ss)
1259 if (ss->ss_family == AF_INET) {
1260 const struct sockaddr_in *sin
1261 = ALIGNED_CAST(const struct sockaddr_in *, ss);
1262 return ntohs(sin->sin_port);
1263 } else if (ss->ss_family == AF_INET6) {
1264 const struct sockaddr_in6 *sin6
1265 = ALIGNED_CAST(const struct sockaddr_in6 *, ss);
1266 return ntohs(sin6->sin6_port);
1272 /* Formats the IPv4 or IPv6 address in 'ss' into the 'bufsize' bytes in 'buf'.
1273 * If 'ss' is an IPv6 address, puts square brackets around the address.
1274 * 'bufsize' should be at least SS_NTOP_BUFSIZE.
1278 ss_format_address(const struct sockaddr_storage *ss,
1279 char *buf, size_t bufsize)
1281 ovs_assert(bufsize >= SS_NTOP_BUFSIZE);
1282 if (ss->ss_family == AF_INET) {
1283 const struct sockaddr_in *sin
1284 = ALIGNED_CAST(const struct sockaddr_in *, ss);
1286 snprintf(buf, bufsize, IP_FMT, IP_ARGS(sin->sin_addr.s_addr));
1287 } else if (ss->ss_family == AF_INET6) {
1288 const struct sockaddr_in6 *sin6
1289 = ALIGNED_CAST(const struct sockaddr_in6 *, ss);
1292 inet_ntop(AF_INET6, sin6->sin6_addr.s6_addr, buf + 1, bufsize - 1);
1293 strcpy(strchr(buf, '\0'), "]");
1302 ss_length(const struct sockaddr_storage *ss)
1304 switch (ss->ss_family) {
1306 return sizeof(struct sockaddr_in);
1309 return sizeof(struct sockaddr_in6);
1316 /* For Windows socket calls, 'errno' is not set. One has to call
1317 * WSAGetLastError() to get the error number and then pass it to
1318 * this function to get the correct error string.
1320 * ovs_strerror() calls strerror_r() and would not get the correct error
1321 * string for Windows sockets, but is good for POSIX. */
1323 sock_strerror(int error)
1326 return ovs_format_message(error);
1328 return ovs_strerror(error);