2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014, 2015 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)) {
109 /* Sets the DSCP value of socket 'fd' to 'dscp', which must be 63 or less.
110 * 'family' must indicate the socket's address family (AF_INET or AF_INET6, to
111 * do anything useful). */
113 set_dscp(int fd, int family, uint8_t dscp)
125 retval = setsockopt(fd, IPPROTO_IP, IP_TOS, &val, sizeof val);
129 retval = setsockopt(fd, IPPROTO_IPV6, IPV6_TCLASS, &val, sizeof val);
136 return retval ? sock_errno() : 0;
139 /* Translates 'host_name', which must be a string representation of an IP
140 * address, into a numeric IP address in '*addr'. Returns 0 if successful,
141 * otherwise a positive errno value. */
143 lookup_ip(const char *host_name, struct in_addr *addr)
145 if (!inet_pton(AF_INET, host_name, addr)) {
146 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
147 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name);
153 /* Translates 'host_name', which must be a string representation of an IPv6
154 * address, into a numeric IPv6 address in '*addr'. Returns 0 if successful,
155 * otherwise a positive errno value. */
157 lookup_ipv6(const char *host_name, struct in6_addr *addr)
159 if (inet_pton(AF_INET6, host_name, addr) != 1) {
160 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
161 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IPv6 address", host_name);
167 /* Translates 'host_name', which must be a host name or a string representation
168 * of an IP address, into a numeric IP address in '*addr'. Returns 0 if
169 * successful, otherwise a positive errno value.
171 * Most Open vSwitch code should not use this because it causes deadlocks:
172 * getaddrinfo() sends out a DNS request but that starts a new flow for which
173 * OVS must set up a flow, but it can't because it's waiting for a DNS reply.
174 * The synchronous lookup also delays other activity. (Of course we can solve
175 * this but it doesn't seem worthwhile quite yet.) */
177 lookup_hostname(const char *host_name, struct in_addr *addr)
179 struct addrinfo *result;
180 struct addrinfo hints;
182 if (inet_pton(AF_INET, host_name, addr)) {
186 memset(&hints, 0, sizeof hints);
187 hints.ai_family = AF_INET;
189 switch (getaddrinfo(host_name, NULL, &hints, &result)) {
191 *addr = ALIGNED_CAST(struct sockaddr_in *,
192 result->ai_addr)->sin_addr;
193 freeaddrinfo(result);
196 #ifdef EAI_ADDRFAMILY
217 #if defined (EAI_NODATA) && EAI_NODATA != EAI_NONAME
233 check_connection_completion(int fd)
235 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
240 pfd.events = POLLOUT;
244 retval = poll(&pfd, 1, 0);
245 } while (retval < 0 && errno == EINTR);
247 retval = WSAPoll(&pfd, 1, 0);
250 if (pfd.revents & POLLERR) {
251 ssize_t n = send(fd, "", 1, 0);
255 VLOG_ERR_RL(&rl, "poll return POLLERR but send succeeded");
260 } else if (retval < 0) {
261 VLOG_ERR_RL(&rl, "poll: %s", sock_strerror(sock_errno()));
269 /* Drain all the data currently in the receive queue of a datagram socket (and
270 * possibly additional data). There is no way to know how many packets are in
271 * the receive queue, but we do know that the total number of bytes queued does
272 * not exceed the receive buffer size, so we pull packets until none are left
273 * or we've read that many bytes. */
279 rcvbuf = get_socket_rcvbuf(fd);
285 /* In Linux, specifying MSG_TRUNC in the flags argument causes the
286 * datagram length to be returned, even if that is longer than the
287 * buffer provided. Thus, we can use a 1-byte buffer to discard the
288 * incoming datagram and still be able to account how many bytes were
289 * removed from the receive buffer.
291 * On other Unix-like OSes, MSG_TRUNC has no effect in the flags
293 char buffer[LINUX ? 1 : 2048];
294 ssize_t n_bytes = recv(fd, buffer, sizeof buffer,
295 MSG_TRUNC | MSG_DONTWAIT);
296 if (n_bytes <= 0 || n_bytes >= rcvbuf) {
305 /* Returns the size of socket 'sock''s receive buffer (SO_RCVBUF), or a
306 * negative errno value if an error occurs. */
308 get_socket_rcvbuf(int sock)
313 error = getsockopt_int(sock, SOL_SOCKET, SO_RCVBUF, "SO_RCVBUF", &rcvbuf);
314 return error ? -error : rcvbuf;
317 /* Reads and discards up to 'n' datagrams from 'fd', stopping as soon as no
318 * more data can be immediately read. ('fd' should therefore be in
319 * non-blocking mode.)*/
321 drain_fd(int fd, size_t n_packets)
323 for (; n_packets > 0; n_packets--) {
324 /* 'buffer' only needs to be 1 byte long in most circumstances. This
325 * size is defensive against the possibility that we someday want to
326 * use a Linux tap device without TUN_NO_PI, in which case a buffer
327 * smaller than sizeof(struct tun_pi) will give EINVAL on read. */
329 if (read(fd, buffer, sizeof buffer) <= 0) {
336 /* Attempts to shorten 'name' by opening a file descriptor for the directory
337 * part of the name and indirecting through /proc/self/fd/<dirfd>/<basename>.
338 * On systems with Linux-like /proc, this works as long as <basename> isn't too
341 * On success, returns 0 and stores the short name in 'short_name' and a
342 * directory file descriptor to eventually be closed in '*dirfpd'. */
344 shorten_name_via_proc(const char *name, char short_name[MAX_UN_LEN + 1],
355 dir = dir_name(name);
356 dirfd = open(dir, O_DIRECTORY | O_RDONLY);
358 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
361 VLOG_WARN_RL(&rl, "%s: open failed (%s)", dir, ovs_strerror(error));
368 base = base_name(name);
369 len = snprintf(short_name, MAX_UN_LEN + 1,
370 "/proc/self/fd/%d/%s", dirfd, base);
373 if (len >= 0 && len <= MAX_UN_LEN) {
382 /* Attempts to shorten 'name' by creating a symlink for the directory part of
383 * the name and indirecting through <symlink>/<basename>. This works on
384 * systems that support symlinks, as long as <basename> isn't too long.
386 * On success, returns 0 and stores the short name in 'short_name' and the
387 * symbolic link to eventually delete in 'linkname'. */
389 shorten_name_via_symlink(const char *name, char short_name[MAX_UN_LEN + 1],
390 char linkname[MAX_UN_LEN + 1])
392 char *abs, *dir, *base;
397 abs = abs_file_name(NULL, name);
399 base = base_name(abs);
402 tmpdir = getenv("TMPDIR");
403 if (tmpdir == NULL) {
407 for (i = 0; i < 1000; i++) {
410 len = snprintf(linkname, MAX_UN_LEN + 1,
411 "%s/ovs-un-c-%"PRIu32, tmpdir, random_uint32());
412 error = (len < 0 || len > MAX_UN_LEN ? ENAMETOOLONG
413 : symlink(dir, linkname) ? errno
415 if (error != EEXIST) {
423 fatal_signal_add_file_to_unlink(linkname);
425 len = snprintf(short_name, MAX_UN_LEN + 1, "%s/%s", linkname, base);
426 if (len < 0 || len > MAX_UN_LEN) {
427 fatal_signal_unlink_file_now(linkname);
428 error = ENAMETOOLONG;
441 /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in
442 * '*un_len' the size of the sockaddr_un.
444 * Returns 0 on success, otherwise a positive errno value.
446 * Uses '*dirfdp' and 'linkname' to store references to data when the caller no
447 * longer needs to use 'un'. On success, freeing these references with
448 * free_sockaddr_un() is mandatory to avoid a leak; on failure, freeing them is
449 * unnecessary but harmless. */
451 make_sockaddr_un(const char *name, struct sockaddr_un *un, socklen_t *un_len,
452 int *dirfdp, char linkname[MAX_UN_LEN + 1])
454 char short_name[MAX_UN_LEN + 1];
458 if (strlen(name) > MAX_UN_LEN) {
459 /* 'name' is too long to fit in a sockaddr_un. Try a workaround. */
460 int error = shorten_name_via_proc(name, short_name, dirfdp);
461 if (error == ENAMETOOLONG) {
462 error = shorten_name_via_symlink(name, short_name, linkname);
465 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
467 VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum "
468 "%"PRIuSIZE" bytes", name, MAX_UN_LEN);
475 un->sun_family = AF_UNIX;
476 ovs_strzcpy(un->sun_path, name, sizeof un->sun_path);
477 *un_len = (offsetof(struct sockaddr_un, sun_path)
478 + strlen (un->sun_path) + 1);
482 /* Clean up after make_sockaddr_un(). */
484 free_sockaddr_un(int dirfd, const char *linkname)
490 fatal_signal_unlink_file_now(linkname);
494 /* Binds Unix domain socket 'fd' to a file with permissions 0700. */
496 bind_unix_socket(int fd, struct sockaddr *sun, socklen_t sun_len)
498 /* According to _Unix Network Programming_, umask should affect bind(). */
499 mode_t old_umask = umask(0077);
500 int error = bind(fd, sun, sun_len) ? errno : 0;
505 /* Creates a Unix domain socket in the given 'style' (either SOCK_DGRAM or
506 * SOCK_STREAM) that is bound to '*bind_path' (if 'bind_path' is non-null) and
507 * connected to '*connect_path' (if 'connect_path' is non-null). If 'nonblock'
508 * is true, the socket is made non-blocking.
510 * Returns the socket's fd if successful, otherwise a negative errno value. */
512 make_unix_socket(int style, bool nonblock,
513 const char *bind_path, const char *connect_path)
518 fd = socket(PF_UNIX, style, 0);
523 /* Set nonblocking mode right away, if we want it. This prevents blocking
524 * in connect(), if connect_path != NULL. (In turn, that's a corner case:
525 * it will only happen if style is SOCK_STREAM or SOCK_SEQPACKET, and only
526 * if a backlog of un-accepted connections has built up in the kernel.) */
528 error = set_nonblocking(fd);
535 char linkname[MAX_UN_LEN + 1];
536 struct sockaddr_un un;
540 if (unlink(bind_path) && errno != ENOENT) {
541 VLOG_WARN("unlinking \"%s\": %s\n",
542 bind_path, ovs_strerror(errno));
544 fatal_signal_add_file_to_unlink(bind_path);
546 error = make_sockaddr_un(bind_path, &un, &un_len, &dirfd, linkname);
548 error = bind_unix_socket(fd, (struct sockaddr *) &un, un_len);
550 free_sockaddr_un(dirfd, linkname);
558 char linkname[MAX_UN_LEN + 1];
559 struct sockaddr_un un;
563 error = make_sockaddr_un(connect_path, &un, &un_len, &dirfd, linkname);
565 && connect(fd, (struct sockaddr*) &un, un_len)
566 && errno != EINPROGRESS) {
569 free_sockaddr_un(dirfd, linkname);
579 if (error == EAGAIN) {
583 fatal_signal_unlink_file_now(bind_path);
590 get_unix_name_len(socklen_t sun_len)
592 return (sun_len >= offsetof(struct sockaddr_un, sun_path)
593 ? sun_len - offsetof(struct sockaddr_un, sun_path)
599 guess_netmask(ovs_be32 ip_)
601 uint32_t ip = ntohl(ip_);
602 return ((ip >> 31) == 0 ? htonl(0xff000000) /* Class A */
603 : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */
604 : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */
605 : htonl(0)); /* ??? */
608 /* This is like strsep() except:
610 * - The separator string is ":".
612 * - Square brackets [] quote ":" separators and are removed from the
615 parse_bracketed_token(char **pp)
621 } else if (*p == '\0') {
624 } else if (*p == '[') {
626 char *end = start + strcspn(start, "]");
627 *pp = (*end == '\0' ? NULL
628 : end[1] == ':' ? end + 2
634 char *end = start + strcspn(start, ":");
635 *pp = *end == '\0' ? NULL : end + 1;
642 parse_sockaddr_components(struct sockaddr_storage *ss,
644 const char *port_s, uint16_t default_port,
647 struct sockaddr_in *sin = ALIGNED_CAST(struct sockaddr_in *, ss);
650 if (port_s && port_s[0]) {
651 if (!str_to_int(port_s, 10, &port) || port < 0 || port > 65535) {
652 VLOG_ERR("%s: bad port number \"%s\"", s, port_s);
658 memset(ss, 0, sizeof *ss);
659 if (strchr(host_s, ':')) {
660 struct sockaddr_in6 *sin6
661 = ALIGNED_CAST(struct sockaddr_in6 *, ss);
663 sin6->sin6_family = AF_INET6;
664 sin6->sin6_port = htons(port);
665 if (!inet_pton(AF_INET6, host_s, sin6->sin6_addr.s6_addr)) {
666 VLOG_ERR("%s: bad IPv6 address \"%s\"", s, host_s);
670 sin->sin_family = AF_INET;
671 sin->sin_port = htons(port);
672 if (!inet_pton(AF_INET, host_s, &sin->sin_addr.s_addr)) {
673 VLOG_ERR("%s: bad IPv4 address \"%s\"", s, host_s);
681 memset(ss, 0, sizeof *ss);
685 /* Parses 'target', which should be a string in the format "<host>[:<port>]".
686 * <host>, which is required, may be an IPv4 address or an IPv6 address
687 * enclosed in square brackets. If 'default_port' is nonzero then <port> is
688 * optional and defaults to 'default_port'.
690 * On success, returns true and stores the parsed remote address into '*ss'.
691 * On failure, logs an error, stores zeros into '*ss', and returns false. */
693 inet_parse_active(const char *target_, uint16_t default_port,
694 struct sockaddr_storage *ss)
696 char *target = xstrdup(target_);
703 host = parse_bracketed_token(&p);
704 port = parse_bracketed_token(&p);
706 VLOG_ERR("%s: host must be specified", target_);
708 } else if (!port && !default_port) {
709 VLOG_ERR("%s: port must be specified", target_);
712 ok = parse_sockaddr_components(ss, host, port, default_port, target_);
715 memset(ss, 0, sizeof *ss);
722 /* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style' and
723 * connects to 'target', which should be a string in the format
724 * "<host>[:<port>]". <host>, which is required, may be an IPv4 address or an
725 * IPv6 address enclosed in square brackets. If 'default_port' is nonzero then
726 * <port> is optional and defaults to 'default_port'.
728 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
730 * On success, returns 0 (indicating connection complete) or EAGAIN (indicating
731 * connection in progress), in which case the new file descriptor is stored
732 * into '*fdp'. On failure, returns a positive errno value other than EAGAIN
733 * and stores -1 into '*fdp'.
735 * If 'ss' is non-null, then on success stores the target address into '*ss'.
737 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
738 * should be in the range [0, 63] and will automatically be shifted to the
739 * appropriately place in the IP tos field. */
741 inet_open_active(int style, const char *target, uint16_t default_port,
742 struct sockaddr_storage *ssp, int *fdp, uint8_t dscp)
744 struct sockaddr_storage ss;
749 if (!inet_parse_active(target, default_port, &ss)) {
750 error = EAFNOSUPPORT;
754 /* Create non-blocking socket. */
755 fd = socket(ss.ss_family, style, 0);
757 error = sock_errno();
758 VLOG_ERR("%s: socket: %s", target, sock_strerror(error));
761 error = set_nonblocking(fd);
766 /* The dscp bits must be configured before connect() to ensure that the
767 * TOS field is set during the connection establishment. If set after
768 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
769 error = set_dscp(fd, ss.ss_family, dscp);
771 VLOG_ERR("%s: set_dscp: %s", target, sock_strerror(error));
776 error = connect(fd, (struct sockaddr *) &ss, ss_length(&ss)) == 0
779 if (error == EINPROGRESS
781 || error == WSAEALREADY || error == WSAEWOULDBLOCK
788 if (error && error != EAGAIN) {
790 memset(ssp, 0, sizeof *ssp);
805 /* Parses 'target', which should be a string in the format "[<port>][:<host>]":
807 * - If 'default_port' is -1, then <port> is required. Otherwise, if
808 * <port> is omitted, then 'default_port' is used instead.
810 * - If <port> (or 'default_port', if used) is 0, then no port is bound
811 * and the TCP/IP stack will select a port.
813 * - <host> is optional. If supplied, it may be an IPv4 address or an
814 * IPv6 address enclosed in square brackets. If omitted, the IP address
817 * If successful, stores the address into '*ss' and returns true; otherwise
818 * zeros '*ss' and returns false. */
820 inet_parse_passive(const char *target_, int default_port,
821 struct sockaddr_storage *ss)
823 char *target = xstrdup(target_);
830 port = parse_bracketed_token(&p);
831 host = parse_bracketed_token(&p);
832 if (!port && default_port < 0) {
833 VLOG_ERR("%s: port must be specified", target_);
836 ok = parse_sockaddr_components(ss, host ? host : "0.0.0.0",
837 port, default_port, target_);
840 memset(ss, 0, sizeof *ss);
847 /* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style', binds to
848 * 'target', and listens for incoming connections. Parses 'target' in the same
849 * way was inet_parse_passive().
851 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
853 * For TCP, the socket will have SO_REUSEADDR turned on.
855 * On success, returns a non-negative file descriptor. On failure, returns a
856 * negative errno value.
858 * If 'ss' is non-null, then on success stores the bound address into '*ss'.
860 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
861 * should be in the range [0, 63] and will automatically be shifted to the
862 * appropriately place in the IP tos field. */
864 inet_open_passive(int style, const char *target, int default_port,
865 struct sockaddr_storage *ssp, uint8_t dscp)
867 bool kernel_chooses_port;
868 struct sockaddr_storage ss;
870 unsigned int yes = 1;
872 if (!inet_parse_passive(target, default_port, &ss)) {
873 return -EAFNOSUPPORT;
875 kernel_chooses_port = ss_get_port(&ss) == 0;
877 /* Create non-blocking socket, set SO_REUSEADDR. */
878 fd = socket(ss.ss_family, style, 0);
880 error = sock_errno();
881 VLOG_ERR("%s: socket: %s", target, sock_strerror(error));
884 error = set_nonblocking(fd);
888 if (style == SOCK_STREAM
889 && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) {
890 error = sock_errno();
891 VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s",
892 target, sock_strerror(error));
897 if (bind(fd, (struct sockaddr *) &ss, ss_length(&ss)) < 0) {
898 error = sock_errno();
899 VLOG_ERR("%s: bind: %s", target, sock_strerror(error));
903 /* The dscp bits must be configured before connect() to ensure that the TOS
904 * field is set during the connection establishment. If set after
905 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
906 error = set_dscp(fd, ss.ss_family, dscp);
908 VLOG_ERR("%s: set_dscp: %s", target, sock_strerror(error));
913 if (style == SOCK_STREAM && listen(fd, 10) < 0) {
914 error = sock_errno();
915 VLOG_ERR("%s: listen: %s", target, sock_strerror(error));
919 if (ssp || kernel_chooses_port) {
920 socklen_t ss_len = sizeof ss;
921 if (getsockname(fd, (struct sockaddr *) &ss, &ss_len) < 0) {
922 error = sock_errno();
923 VLOG_ERR("%s: getsockname: %s", target, sock_strerror(error));
926 if (kernel_chooses_port) {
927 VLOG_INFO("%s: listening on port %"PRIu16,
928 target, ss_get_port(&ss));
939 memset(ssp, 0, sizeof *ssp);
946 read_fully(int fd, void *p_, size_t size, size_t *bytes_read)
952 ssize_t retval = read(fd, p, size);
954 *bytes_read += retval;
957 } else if (retval == 0) {
959 } else if (errno != EINTR) {
967 write_fully(int fd, const void *p_, size_t size, size_t *bytes_written)
969 const uint8_t *p = p_;
973 ssize_t retval = write(fd, p, size);
975 *bytes_written += retval;
978 } else if (retval == 0) {
979 VLOG_WARN("write returned 0");
981 } else if (errno != EINTR) {
988 /* Given file name 'file_name', fsyncs the directory in which it is contained.
989 * Returns 0 if successful, otherwise a positive errno value. */
991 fsync_parent_dir(const char *file_name)
998 dir = dir_name(file_name);
999 fd = open(dir, O_RDONLY);
1002 if (errno == EINVAL || errno == EROFS) {
1003 /* This directory does not support synchronization. Not
1004 * really an error. */
1007 VLOG_ERR("%s: fsync failed (%s)", dir, ovs_strerror(error));
1013 VLOG_ERR("%s: open failed (%s)", dir, ovs_strerror(error));
1021 /* Obtains the modification time of the file named 'file_name' to the greatest
1022 * supported precision. If successful, stores the mtime in '*mtime' and
1023 * returns 0. On error, returns a positive errno value and stores zeros in
1026 get_mtime(const char *file_name, struct timespec *mtime)
1030 if (!stat(file_name, &s)) {
1031 mtime->tv_sec = s.st_mtime;
1033 #if HAVE_STRUCT_STAT_ST_MTIM_TV_NSEC
1034 mtime->tv_nsec = s.st_mtim.tv_nsec;
1035 #elif HAVE_STRUCT_STAT_ST_MTIMENSEC
1036 mtime->tv_nsec = s.st_mtimensec;
1043 mtime->tv_sec = mtime->tv_nsec = 0;
1053 VLOG_FATAL("failed to create pipe (%s)", ovs_strerror(errno));
1058 xpipe_nonblocking(int fds[2])
1061 xset_nonblocking(fds[0]);
1062 xset_nonblocking(fds[1]);
1067 getsockopt_int(int fd, int level, int option, const char *optname, int *valuep)
1069 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
1075 if (getsockopt(fd, level, option, &value, &len)) {
1076 error = sock_errno();
1077 VLOG_ERR_RL(&rl, "getsockopt(%s): %s", optname, sock_strerror(error));
1078 } else if (len != sizeof value) {
1080 VLOG_ERR_RL(&rl, "getsockopt(%s): value is %u bytes (expected %"PRIuSIZE")",
1081 optname, (unsigned int) len, sizeof value);
1086 *valuep = error ? 0 : value;
1091 describe_sockaddr(struct ds *string, int fd,
1092 int (*getaddr)(int, struct sockaddr *, socklen_t *))
1094 struct sockaddr_storage ss;
1095 socklen_t len = sizeof ss;
1097 if (!getaddr(fd, (struct sockaddr *) &ss, &len)) {
1098 if (ss.ss_family == AF_INET || ss.ss_family == AF_INET6) {
1099 char addrbuf[SS_NTOP_BUFSIZE];
1101 ds_put_format(string, "%s:%"PRIu16,
1102 ss_format_address(&ss, addrbuf, sizeof addrbuf),
1105 } else if (ss.ss_family == AF_UNIX) {
1106 struct sockaddr_un sun;
1110 memcpy(&sun, &ss, sizeof sun);
1111 maxlen = len - offsetof(struct sockaddr_un, sun_path);
1112 null = memchr(sun.sun_path, '\0', maxlen);
1113 ds_put_buffer(string, sun.sun_path,
1114 null ? null - sun.sun_path : maxlen);
1118 else if (ss.ss_family == AF_NETLINK) {
1121 /* SO_PROTOCOL was introduced in 2.6.32. Support it regardless of the version
1122 * of the Linux kernel headers in use at build time. */
1124 #define SO_PROTOCOL 38
1127 if (!getsockopt_int(fd, SOL_SOCKET, SO_PROTOCOL, "SO_PROTOCOL",
1131 ds_put_cstr(string, "NETLINK_ROUTE");
1134 case NETLINK_GENERIC:
1135 ds_put_cstr(string, "NETLINK_GENERIC");
1139 ds_put_format(string, "AF_NETLINK family %d", protocol);
1143 ds_put_cstr(string, "AF_NETLINK");
1148 else if (ss.ss_family == AF_PACKET) {
1149 struct sockaddr_ll sll;
1151 memcpy(&sll, &ss, sizeof sll);
1152 ds_put_cstr(string, "AF_PACKET");
1153 if (sll.sll_ifindex) {
1154 char name[IFNAMSIZ];
1156 if (if_indextoname(sll.sll_ifindex, name)) {
1157 ds_put_format(string, "(%s)", name);
1159 ds_put_format(string, "(ifindex=%d)", sll.sll_ifindex);
1162 if (sll.sll_protocol) {
1163 ds_put_format(string, "(protocol=0x%"PRIu16")",
1164 ntohs(sll.sll_protocol));
1168 else if (ss.ss_family == AF_UNSPEC) {
1169 ds_put_cstr(string, "AF_UNSPEC");
1171 ds_put_format(string, "AF_%d", (int) ss.ss_family);
1179 put_fd_filename(struct ds *string, int fd)
1185 linkname = xasprintf("/proc/self/fd/%d", fd);
1186 n = readlink(linkname, buf, sizeof buf);
1188 ds_put_char(string, ' ');
1189 ds_put_buffer(string, buf, n);
1190 if (n > sizeof buf) {
1191 ds_put_cstr(string, "...");
1198 /* Returns a malloc()'d string describing 'fd', for use in logging. */
1207 if (fstat(fd, &s)) {
1208 ds_put_format(&string, "fstat failed (%s)", ovs_strerror(errno));
1209 } else if (S_ISSOCK(s.st_mode)) {
1210 describe_sockaddr(&string, fd, getsockname);
1211 ds_put_cstr(&string, "<->");
1212 describe_sockaddr(&string, fd, getpeername);
1214 ds_put_cstr(&string, (isatty(fd) ? "tty"
1215 : S_ISDIR(s.st_mode) ? "directory"
1216 : S_ISCHR(s.st_mode) ? "character device"
1217 : S_ISBLK(s.st_mode) ? "block device"
1218 : S_ISREG(s.st_mode) ? "file"
1219 : S_ISFIFO(s.st_mode) ? "FIFO"
1220 : S_ISLNK(s.st_mode) ? "symbolic link"
1223 put_fd_filename(&string, fd);
1227 ds_put_format(&string,"file descriptor");
1229 return ds_steal_cstr(&string);
1233 /* Calls ioctl() on an AF_INET sock, passing the specified 'command' and
1234 * 'arg'. Returns 0 if successful, otherwise a positive errno value. */
1236 af_inet_ioctl(unsigned long int command, const void *arg)
1238 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
1241 if (ovsthread_once_start(&once)) {
1242 sock = socket(AF_INET, SOCK_DGRAM, 0);
1244 int error = sock_errno();
1245 VLOG_ERR("failed to create inet socket: %s", sock_strerror(error));
1248 ovsthread_once_done(&once);
1251 return (sock < 0 ? -sock
1252 : ioctl(sock, command, arg) == -1 ? errno
1257 af_inet_ifreq_ioctl(const char *name, struct ifreq *ifr, unsigned long int cmd,
1258 const char *cmd_name)
1262 ovs_strzcpy(ifr->ifr_name, name, sizeof ifr->ifr_name);
1263 error = af_inet_ioctl(cmd, ifr);
1265 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1266 VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name,
1267 ovs_strerror(error));
1273 /* sockaddr_storage helpers. */
1275 /* Returns the IPv4 or IPv6 port in 'ss'. */
1277 ss_get_port(const struct sockaddr_storage *ss)
1279 if (ss->ss_family == AF_INET) {
1280 const struct sockaddr_in *sin
1281 = ALIGNED_CAST(const struct sockaddr_in *, ss);
1282 return ntohs(sin->sin_port);
1283 } else if (ss->ss_family == AF_INET6) {
1284 const struct sockaddr_in6 *sin6
1285 = ALIGNED_CAST(const struct sockaddr_in6 *, ss);
1286 return ntohs(sin6->sin6_port);
1292 /* Formats the IPv4 or IPv6 address in 'ss' into the 'bufsize' bytes in 'buf'.
1293 * If 'ss' is an IPv6 address, puts square brackets around the address.
1294 * 'bufsize' should be at least SS_NTOP_BUFSIZE.
1298 ss_format_address(const struct sockaddr_storage *ss,
1299 char *buf, size_t bufsize)
1301 ovs_assert(bufsize >= SS_NTOP_BUFSIZE);
1302 if (ss->ss_family == AF_INET) {
1303 const struct sockaddr_in *sin
1304 = ALIGNED_CAST(const struct sockaddr_in *, ss);
1306 snprintf(buf, bufsize, IP_FMT, IP_ARGS(sin->sin_addr.s_addr));
1307 } else if (ss->ss_family == AF_INET6) {
1308 const struct sockaddr_in6 *sin6
1309 = ALIGNED_CAST(const struct sockaddr_in6 *, ss);
1312 inet_ntop(AF_INET6, sin6->sin6_addr.s6_addr, buf + 1, bufsize - 1);
1313 strcpy(strchr(buf, '\0'), "]");
1322 ss_length(const struct sockaddr_storage *ss)
1324 switch (ss->ss_family) {
1326 return sizeof(struct sockaddr_in);
1329 return sizeof(struct sockaddr_in6);
1336 /* For Windows socket calls, 'errno' is not set. One has to call
1337 * WSAGetLastError() to get the error number and then pass it to
1338 * this function to get the correct error string.
1340 * ovs_strerror() calls strerror_r() and would not get the correct error
1341 * string for Windows sockets, but is good for POSIX. */
1343 sock_strerror(int error)
1346 return ovs_format_message(error);
1348 return ovs_strerror(error);