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/resource.h>
31 #include <sys/socket.h>
36 #include "dynamic-string.h"
37 #include "fatal-signal.h"
38 #include "ovs-thread.h"
40 #include "poll-loop.h"
43 #if AF_PACKET && LINUX_DATAPATH
44 #include <linux/if_packet.h>
47 #include "netlink-protocol.h"
48 #include "netlink-socket.h"
51 VLOG_DEFINE_THIS_MODULE(socket_util);
53 /* #ifdefs make it a pain to maintain code: you have to try to build both ways.
54 * Thus, this file compiles all of the code regardless of the target, by
55 * writing "if (LINUX_DATAPATH)" instead of "#ifdef __linux__". */
56 #ifndef LINUX_DATAPATH
57 #define LINUX_DATAPATH 0
64 /* Maximum length of the sun_path member in a struct sockaddr_un, excluding
65 * space for a null terminator. */
66 #define MAX_UN_LEN (sizeof(((struct sockaddr_un *) 0)->sun_path) - 1)
69 /* Buffer used by sock_strerror(). */
70 DEFINE_STATIC_PER_THREAD_DATA(struct { char s[128]; },
75 static int getsockopt_int(int fd, int level, int option, const char *optname,
78 /* Sets 'fd' to non-blocking mode. Returns 0 if successful, otherwise a
79 * positive errno value. */
81 set_nonblocking(int fd)
83 int flags = fcntl(fd, F_GETFL, 0);
85 if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) != -1) {
88 VLOG_ERR("fcntl(F_SETFL) failed: %s", ovs_strerror(errno));
92 VLOG_ERR("fcntl(F_GETFL) failed: %s", ovs_strerror(errno));
98 xset_nonblocking(int fd)
100 if (set_nonblocking(fd)) {
106 set_dscp(int fd, uint8_t dscp)
115 if (setsockopt(fd, IPPROTO_IP, IP_TOS, &val, sizeof val)) {
123 rlim_is_finite(rlim_t limit)
125 if (limit == RLIM_INFINITY) {
129 #ifdef RLIM_SAVED_CUR /* FreeBSD 8.0 lacks RLIM_SAVED_CUR. */
130 if (limit == RLIM_SAVED_CUR) {
135 #ifdef RLIM_SAVED_MAX /* FreeBSD 8.0 lacks RLIM_SAVED_MAX. */
136 if (limit == RLIM_SAVED_MAX) {
144 /* Returns the maximum valid FD value, plus 1. */
148 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
151 if (ovsthread_once_start(&once)) {
153 if (!getrlimit(RLIMIT_NOFILE, &r) && rlim_is_finite(r.rlim_cur)) {
154 max_fds = r.rlim_cur;
156 VLOG_WARN("failed to obtain fd limit, defaulting to 1024");
159 ovsthread_once_done(&once);
165 /* Translates 'host_name', which must be a string representation of an IP
166 * address, into a numeric IP address in '*addr'. Returns 0 if successful,
167 * otherwise a positive errno value. */
169 lookup_ip(const char *host_name, struct in_addr *addr)
171 if (!inet_aton(host_name, addr)) {
172 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
173 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name);
179 /* Translates 'host_name', which must be a string representation of an IPv6
180 * address, into a numeric IPv6 address in '*addr'. Returns 0 if successful,
181 * otherwise a positive errno value. */
183 lookup_ipv6(const char *host_name, struct in6_addr *addr)
185 if (inet_pton(AF_INET6, host_name, addr) != 1) {
186 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
187 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IPv6 address", host_name);
193 /* Translates 'host_name', which must be a host name or a string representation
194 * of an IP address, into a numeric IP address in '*addr'. Returns 0 if
195 * successful, otherwise a positive errno value.
197 * Most Open vSwitch code should not use this because it causes deadlocks:
198 * getaddrinfo() sends out a DNS request but that starts a new flow for which
199 * OVS must set up a flow, but it can't because it's waiting for a DNS reply.
200 * The synchronous lookup also delays other activity. (Of course we can solve
201 * this but it doesn't seem worthwhile quite yet.) */
203 lookup_hostname(const char *host_name, struct in_addr *addr)
205 struct addrinfo *result;
206 struct addrinfo hints;
208 if (inet_aton(host_name, addr)) {
212 memset(&hints, 0, sizeof hints);
213 hints.ai_family = AF_INET;
215 switch (getaddrinfo(host_name, NULL, &hints, &result)) {
217 *addr = ALIGNED_CAST(struct sockaddr_in *,
218 result->ai_addr)->sin_addr;
219 freeaddrinfo(result);
222 #ifdef EAI_ADDRFAMILY
257 check_connection_completion(int fd)
259 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
264 pfd.events = POLLOUT;
266 retval = poll(&pfd, 1, 0);
267 } while (retval < 0 && errno == EINTR);
269 if (pfd.revents & POLLERR) {
270 ssize_t n = send(fd, "", 1, MSG_DONTWAIT);
274 VLOG_ERR_RL(&rl, "poll return POLLERR but send succeeded");
279 } else if (retval < 0) {
280 VLOG_ERR_RL(&rl, "poll: %s", ovs_strerror(errno));
287 /* Drain all the data currently in the receive queue of a datagram socket (and
288 * possibly additional data). There is no way to know how many packets are in
289 * the receive queue, but we do know that the total number of bytes queued does
290 * not exceed the receive buffer size, so we pull packets until none are left
291 * or we've read that many bytes. */
297 rcvbuf = get_socket_rcvbuf(fd);
303 /* In Linux, specifying MSG_TRUNC in the flags argument causes the
304 * datagram length to be returned, even if that is longer than the
305 * buffer provided. Thus, we can use a 1-byte buffer to discard the
306 * incoming datagram and still be able to account how many bytes were
307 * removed from the receive buffer.
309 * On other Unix-like OSes, MSG_TRUNC has no effect in the flags
311 char buffer[LINUX_DATAPATH ? 1 : 2048];
312 ssize_t n_bytes = recv(fd, buffer, sizeof buffer,
313 MSG_TRUNC | MSG_DONTWAIT);
314 if (n_bytes <= 0 || n_bytes >= rcvbuf) {
322 /* Returns the size of socket 'sock''s receive buffer (SO_RCVBUF), or a
323 * negative errno value if an error occurs. */
325 get_socket_rcvbuf(int sock)
330 error = getsockopt_int(sock, SOL_SOCKET, SO_RCVBUF, "SO_RCVBUF", &rcvbuf);
331 return error ? -error : rcvbuf;
334 /* Reads and discards up to 'n' datagrams from 'fd', stopping as soon as no
335 * more data can be immediately read. ('fd' should therefore be in
336 * non-blocking mode.)*/
338 drain_fd(int fd, size_t n_packets)
340 for (; n_packets > 0; n_packets--) {
341 /* 'buffer' only needs to be 1 byte long in most circumstances. This
342 * size is defensive against the possibility that we someday want to
343 * use a Linux tap device without TUN_NO_PI, in which case a buffer
344 * smaller than sizeof(struct tun_pi) will give EINVAL on read. */
346 if (read(fd, buffer, sizeof buffer) <= 0) {
352 /* Attempts to shorten 'name' by opening a file descriptor for the directory
353 * part of the name and indirecting through /proc/self/fd/<dirfd>/<basename>.
354 * On systems with Linux-like /proc, this works as long as <basename> isn't too
357 * On success, returns 0 and stores the short name in 'short_name' and a
358 * directory file descriptor to eventually be closed in '*dirfpd'. */
360 shorten_name_via_proc(const char *name, char short_name[MAX_UN_LEN + 1],
367 if (!LINUX_DATAPATH) {
371 dir = dir_name(name);
372 dirfd = open(dir, O_DIRECTORY | O_RDONLY);
374 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
377 VLOG_WARN_RL(&rl, "%s: open failed (%s)", dir, ovs_strerror(error));
384 base = base_name(name);
385 len = snprintf(short_name, MAX_UN_LEN + 1,
386 "/proc/self/fd/%d/%s", dirfd, base);
389 if (len >= 0 && len <= MAX_UN_LEN) {
398 /* Attempts to shorten 'name' by creating a symlink for the directory part of
399 * the name and indirecting through <symlink>/<basename>. This works on
400 * systems that support symlinks, as long as <basename> isn't too long.
402 * On success, returns 0 and stores the short name in 'short_name' and the
403 * symbolic link to eventually delete in 'linkname'. */
405 shorten_name_via_symlink(const char *name, char short_name[MAX_UN_LEN + 1],
406 char linkname[MAX_UN_LEN + 1])
408 char *abs, *dir, *base;
413 abs = abs_file_name(NULL, name);
415 base = base_name(abs);
418 tmpdir = getenv("TMPDIR");
419 if (tmpdir == NULL) {
423 for (i = 0; i < 1000; i++) {
426 len = snprintf(linkname, MAX_UN_LEN + 1,
427 "%s/ovs-un-c-%"PRIu32, tmpdir, random_uint32());
428 error = (len < 0 || len > MAX_UN_LEN ? ENAMETOOLONG
429 : symlink(dir, linkname) ? errno
431 if (error != EEXIST) {
439 fatal_signal_add_file_to_unlink(linkname);
441 len = snprintf(short_name, MAX_UN_LEN + 1, "%s/%s", linkname, base);
442 if (len < 0 || len > MAX_UN_LEN) {
443 fatal_signal_unlink_file_now(linkname);
444 error = ENAMETOOLONG;
457 /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in
458 * '*un_len' the size of the sockaddr_un.
460 * Returns 0 on success, otherwise a positive errno value.
462 * Uses '*dirfdp' and 'linkname' to store references to data when the caller no
463 * longer needs to use 'un'. On success, freeing these references with
464 * free_sockaddr_un() is mandatory to avoid a leak; on failure, freeing them is
465 * unnecessary but harmless. */
467 make_sockaddr_un(const char *name, struct sockaddr_un *un, socklen_t *un_len,
468 int *dirfdp, char linkname[MAX_UN_LEN + 1])
470 char short_name[MAX_UN_LEN + 1];
474 if (strlen(name) > MAX_UN_LEN) {
475 /* 'name' is too long to fit in a sockaddr_un. Try a workaround. */
476 int error = shorten_name_via_proc(name, short_name, dirfdp);
477 if (error == ENAMETOOLONG) {
478 error = shorten_name_via_symlink(name, short_name, linkname);
481 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
483 VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum "
484 "%"PRIuSIZE" bytes", name, MAX_UN_LEN);
491 un->sun_family = AF_UNIX;
492 ovs_strzcpy(un->sun_path, name, sizeof un->sun_path);
493 *un_len = (offsetof(struct sockaddr_un, sun_path)
494 + strlen (un->sun_path) + 1);
498 /* Clean up after make_sockaddr_un(). */
500 free_sockaddr_un(int dirfd, const char *linkname)
506 fatal_signal_unlink_file_now(linkname);
510 /* Binds Unix domain socket 'fd' to a file with permissions 0700. */
512 bind_unix_socket(int fd, struct sockaddr *sun, socklen_t sun_len)
514 /* According to _Unix Network Programming_, umask should affect bind(). */
515 mode_t old_umask = umask(0077);
516 int error = bind(fd, sun, sun_len) ? errno : 0;
521 /* Creates a Unix domain socket in the given 'style' (either SOCK_DGRAM or
522 * SOCK_STREAM) that is bound to '*bind_path' (if 'bind_path' is non-null) and
523 * connected to '*connect_path' (if 'connect_path' is non-null). If 'nonblock'
524 * is true, the socket is made non-blocking.
526 * Returns the socket's fd if successful, otherwise a negative errno value. */
528 make_unix_socket(int style, bool nonblock,
529 const char *bind_path, const char *connect_path)
534 fd = socket(PF_UNIX, style, 0);
539 /* Set nonblocking mode right away, if we want it. This prevents blocking
540 * in connect(), if connect_path != NULL. (In turn, that's a corner case:
541 * it will only happen if style is SOCK_STREAM or SOCK_SEQPACKET, and only
542 * if a backlog of un-accepted connections has built up in the kernel.) */
544 error = set_nonblocking(fd);
551 char linkname[MAX_UN_LEN + 1];
552 struct sockaddr_un un;
556 if (unlink(bind_path) && errno != ENOENT) {
557 VLOG_WARN("unlinking \"%s\": %s\n",
558 bind_path, ovs_strerror(errno));
560 fatal_signal_add_file_to_unlink(bind_path);
562 error = make_sockaddr_un(bind_path, &un, &un_len, &dirfd, linkname);
564 error = bind_unix_socket(fd, (struct sockaddr *) &un, un_len);
566 free_sockaddr_un(dirfd, linkname);
574 char linkname[MAX_UN_LEN + 1];
575 struct sockaddr_un un;
579 error = make_sockaddr_un(connect_path, &un, &un_len, &dirfd, linkname);
581 && connect(fd, (struct sockaddr*) &un, un_len)
582 && errno != EINPROGRESS) {
585 free_sockaddr_un(dirfd, linkname);
595 if (error == EAGAIN) {
599 fatal_signal_unlink_file_now(bind_path);
606 get_unix_name_len(socklen_t sun_len)
608 return (sun_len >= offsetof(struct sockaddr_un, sun_path)
609 ? sun_len - offsetof(struct sockaddr_un, sun_path)
614 guess_netmask(ovs_be32 ip_)
616 uint32_t ip = ntohl(ip_);
617 return ((ip >> 31) == 0 ? htonl(0xff000000) /* Class A */
618 : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */
619 : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */
620 : htonl(0)); /* ??? */
623 /* This is like strsep() except:
625 * - The separator string is ":".
627 * - Square brackets [] quote ":" separators and are removed from the
630 parse_bracketed_token(char **pp)
636 } else if (*p == '\0') {
639 } else if (*p == '[') {
641 char *end = start + strcspn(start, "]");
642 *pp = (*end == '\0' ? NULL
643 : end[1] == ':' ? end + 2
649 char *end = start + strcspn(start, ":");
650 *pp = *end == '\0' ? NULL : end + 1;
657 parse_sockaddr_components(struct sockaddr_storage *ss,
659 const char *port_s, uint16_t default_port,
662 struct sockaddr_in *sin = ALIGNED_CAST(struct sockaddr_in *, ss);
665 if (port_s && port_s[0]) {
666 if (!str_to_int(port_s, 10, &port) || port < 0 || port > 65535) {
667 VLOG_ERR("%s: bad port number \"%s\"", s, port_s);
673 memset(ss, 0, sizeof *ss);
674 if (strchr(host_s, ':')) {
675 struct sockaddr_in6 *sin6
676 = ALIGNED_CAST(struct sockaddr_in6 *, ss);
678 sin6->sin6_family = AF_INET6;
679 sin6->sin6_port = htons(port);
680 if (!inet_pton(AF_INET6, host_s, sin6->sin6_addr.s6_addr)) {
681 VLOG_ERR("%s: bad IPv6 address \"%s\"", s, host_s);
685 sin->sin_family = AF_INET;
686 sin->sin_port = htons(port);
687 if (!inet_pton(AF_INET, host_s, &sin->sin_addr.s_addr)) {
688 VLOG_ERR("%s: bad IPv4 address \"%s\"", s, host_s);
696 memset(ss, 0, sizeof *ss);
700 /* Parses 'target', which should be a string in the format "<host>[:<port>]".
701 * <host>, which is required, may be an IPv4 address or an IPv6 address
702 * enclosed in square brackets. If 'default_port' is nonzero then <port> is
703 * optional and defaults to 'default_port'.
705 * On success, returns true and stores the parsed remote address into '*ss'.
706 * On failure, logs an error, stores zeros into '*ss', and returns false. */
708 inet_parse_active(const char *target_, uint16_t default_port,
709 struct sockaddr_storage *ss)
711 char *target = xstrdup(target_);
718 host = parse_bracketed_token(&p);
719 port = parse_bracketed_token(&p);
721 VLOG_ERR("%s: host must be specified", target_);
723 } else if (!port && !default_port) {
724 VLOG_ERR("%s: port must be specified", target_);
727 ok = parse_sockaddr_components(ss, host, port, default_port, target_);
730 memset(ss, 0, sizeof *ss);
737 /* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style' and
738 * connects to 'target', which should be a string in the format
739 * "<host>[:<port>]". <host>, which is required, may be an IPv4 address or an
740 * IPv6 address enclosed in square brackets. If 'default_port' is nonzero then
741 * <port> is optional and defaults to 'default_port'.
743 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
745 * On success, returns 0 (indicating connection complete) or EAGAIN (indicating
746 * connection in progress), in which case the new file descriptor is stored
747 * into '*fdp'. On failure, returns a positive errno value other than EAGAIN
748 * and stores -1 into '*fdp'.
750 * If 'ss' is non-null, then on success stores the target address into '*ss'.
752 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
753 * should be in the range [0, 63] and will automatically be shifted to the
754 * appropriately place in the IP tos field. */
756 inet_open_active(int style, const char *target, uint16_t default_port,
757 struct sockaddr_storage *ssp, int *fdp, uint8_t dscp)
759 struct sockaddr_storage ss;
764 if (!inet_parse_active(target, default_port, &ss)) {
765 error = EAFNOSUPPORT;
769 /* Create non-blocking socket. */
770 fd = socket(ss.ss_family, style, 0);
772 VLOG_ERR("%s: socket: %s", target, ovs_strerror(errno));
776 error = set_nonblocking(fd);
781 /* The dscp bits must be configured before connect() to ensure that the
782 * TOS field is set during the connection establishment. If set after
783 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
784 error = set_dscp(fd, dscp);
786 VLOG_ERR("%s: socket: %s", target, ovs_strerror(error));
791 error = connect(fd, (struct sockaddr *) &ss, ss_length(&ss)) == 0
794 if (error == EINPROGRESS) {
799 if (error && error != EAGAIN) {
801 memset(ssp, 0, sizeof *ssp);
816 /* Parses 'target', which should be a string in the format "[<port>][:<host>]":
818 * - If 'default_port' is -1, then <port> is required. Otherwise, if
819 * <port> is omitted, then 'default_port' is used instead.
821 * - If <port> (or 'default_port', if used) is 0, then no port is bound
822 * and the TCP/IP stack will select a port.
824 * - <host> is optional. If supplied, it may be an IPv4 address or an
825 * IPv6 address enclosed in square brackets. If omitted, the IP address
828 * If successful, stores the address into '*ss' and returns true; otherwise
829 * zeros '*ss' and returns false. */
831 inet_parse_passive(const char *target_, int default_port,
832 struct sockaddr_storage *ss)
834 char *target = xstrdup(target_);
841 port = parse_bracketed_token(&p);
842 host = parse_bracketed_token(&p);
843 if (!port && default_port < 0) {
844 VLOG_ERR("%s: port must be specified", target_);
847 ok = parse_sockaddr_components(ss, host ? host : "0.0.0.0",
848 port, default_port, target_);
851 memset(ss, 0, sizeof *ss);
858 /* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style', binds to
859 * 'target', and listens for incoming connections. Parses 'target' in the same
860 * way was inet_parse_passive().
862 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
864 * For TCP, the socket will have SO_REUSEADDR turned on.
866 * On success, returns a non-negative file descriptor. On failure, returns a
867 * negative errno value.
869 * If 'ss' is non-null, then on success stores the bound address into '*ss'.
871 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
872 * should be in the range [0, 63] and will automatically be shifted to the
873 * appropriately place in the IP tos field. */
875 inet_open_passive(int style, const char *target, int default_port,
876 struct sockaddr_storage *ssp, uint8_t dscp)
878 bool kernel_chooses_port;
879 struct sockaddr_storage ss;
881 unsigned int yes = 1;
883 if (!inet_parse_passive(target, default_port, &ss)) {
884 return -EAFNOSUPPORT;
886 kernel_chooses_port = ss_get_port(&ss) == 0;
888 /* Create non-blocking socket, set SO_REUSEADDR. */
889 fd = socket(ss.ss_family, style, 0);
892 VLOG_ERR("%s: socket: %s", target, ovs_strerror(error));
895 error = set_nonblocking(fd);
899 if (style == SOCK_STREAM
900 && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) {
902 VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s",
903 target, ovs_strerror(error));
908 if (bind(fd, (struct sockaddr *) &ss, ss_length(&ss)) < 0) {
910 VLOG_ERR("%s: bind: %s", target, ovs_strerror(error));
914 /* The dscp bits must be configured before connect() to ensure that the TOS
915 * field is set during the connection establishment. If set after
916 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
917 error = set_dscp(fd, dscp);
919 VLOG_ERR("%s: socket: %s", target, ovs_strerror(error));
924 if (style == SOCK_STREAM && listen(fd, 10) < 0) {
926 VLOG_ERR("%s: listen: %s", target, ovs_strerror(error));
930 if (ssp || kernel_chooses_port) {
931 socklen_t ss_len = sizeof ss;
932 if (getsockname(fd, (struct sockaddr *) &ss, &ss_len) < 0) {
934 VLOG_ERR("%s: getsockname: %s", target, ovs_strerror(error));
937 if (kernel_chooses_port) {
938 VLOG_INFO("%s: listening on port %"PRIu16,
939 target, ss_get_port(&ss));
950 memset(ssp, 0, sizeof *ssp);
956 /* Returns a readable and writable fd for /dev/null, if successful, otherwise
957 * a negative errno value. The caller must not close the returned fd (because
958 * the same fd will be handed out to subsequent callers). */
962 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
965 if (ovsthread_once_start(&once)) {
966 null_fd = open("/dev/null", O_RDWR);
969 VLOG_ERR("could not open /dev/null: %s", ovs_strerror(error));
972 ovsthread_once_done(&once);
979 read_fully(int fd, void *p_, size_t size, size_t *bytes_read)
985 ssize_t retval = read(fd, p, size);
987 *bytes_read += retval;
990 } else if (retval == 0) {
992 } else if (errno != EINTR) {
1000 write_fully(int fd, const void *p_, size_t size, size_t *bytes_written)
1002 const uint8_t *p = p_;
1006 ssize_t retval = write(fd, p, size);
1008 *bytes_written += retval;
1011 } else if (retval == 0) {
1012 VLOG_WARN("write returned 0");
1014 } else if (errno != EINTR) {
1021 /* Given file name 'file_name', fsyncs the directory in which it is contained.
1022 * Returns 0 if successful, otherwise a positive errno value. */
1024 fsync_parent_dir(const char *file_name)
1030 dir = dir_name(file_name);
1031 fd = open(dir, O_RDONLY);
1034 if (errno == EINVAL || errno == EROFS) {
1035 /* This directory does not support synchronization. Not
1036 * really an error. */
1039 VLOG_ERR("%s: fsync failed (%s)", dir, ovs_strerror(error));
1045 VLOG_ERR("%s: open failed (%s)", dir, ovs_strerror(error));
1052 /* Obtains the modification time of the file named 'file_name' to the greatest
1053 * supported precision. If successful, stores the mtime in '*mtime' and
1054 * returns 0. On error, returns a positive errno value and stores zeros in
1057 get_mtime(const char *file_name, struct timespec *mtime)
1061 if (!stat(file_name, &s)) {
1062 mtime->tv_sec = s.st_mtime;
1064 #if HAVE_STRUCT_STAT_ST_MTIM_TV_NSEC
1065 mtime->tv_nsec = s.st_mtim.tv_nsec;
1066 #elif HAVE_STRUCT_STAT_ST_MTIMENSEC
1067 mtime->tv_nsec = s.st_mtimensec;
1074 mtime->tv_sec = mtime->tv_nsec = 0;
1083 VLOG_FATAL("failed to create pipe (%s)", ovs_strerror(errno));
1088 xpipe_nonblocking(int fds[2])
1091 xset_nonblocking(fds[0]);
1092 xset_nonblocking(fds[1]);
1096 getsockopt_int(int fd, int level, int option, const char *optname, int *valuep)
1098 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
1104 if (getsockopt(fd, level, option, &value, &len)) {
1106 VLOG_ERR_RL(&rl, "getsockopt(%s): %s", optname, ovs_strerror(error));
1107 } else if (len != sizeof value) {
1109 VLOG_ERR_RL(&rl, "getsockopt(%s): value is %u bytes (expected %"PRIuSIZE")",
1110 optname, (unsigned int) len, sizeof value);
1115 *valuep = error ? 0 : value;
1120 describe_sockaddr(struct ds *string, int fd,
1121 int (*getaddr)(int, struct sockaddr *, socklen_t *))
1123 struct sockaddr_storage ss;
1124 socklen_t len = sizeof ss;
1126 if (!getaddr(fd, (struct sockaddr *) &ss, &len)) {
1127 if (ss.ss_family == AF_INET || ss.ss_family == AF_INET6) {
1128 char addrbuf[SS_NTOP_BUFSIZE];
1130 ds_put_format(string, "%s:%"PRIu16,
1131 ss_format_address(&ss, addrbuf, sizeof addrbuf),
1133 } else if (ss.ss_family == AF_UNIX) {
1134 struct sockaddr_un sun;
1138 memcpy(&sun, &ss, sizeof sun);
1139 maxlen = len - offsetof(struct sockaddr_un, sun_path);
1140 null = memchr(sun.sun_path, '\0', maxlen);
1141 ds_put_buffer(string, sun.sun_path,
1142 null ? null - sun.sun_path : maxlen);
1145 else if (ss.ss_family == AF_NETLINK) {
1148 /* SO_PROTOCOL was introduced in 2.6.32. Support it regardless of the version
1149 * of the Linux kernel headers in use at build time. */
1151 #define SO_PROTOCOL 38
1154 if (!getsockopt_int(fd, SOL_SOCKET, SO_PROTOCOL, "SO_PROTOCOL",
1158 ds_put_cstr(string, "NETLINK_ROUTE");
1161 case NETLINK_GENERIC:
1162 ds_put_cstr(string, "NETLINK_GENERIC");
1166 ds_put_format(string, "AF_NETLINK family %d", protocol);
1170 ds_put_cstr(string, "AF_NETLINK");
1174 #if AF_PACKET && LINUX_DATAPATH
1175 else if (ss.ss_family == AF_PACKET) {
1176 struct sockaddr_ll sll;
1178 memcpy(&sll, &ss, sizeof sll);
1179 ds_put_cstr(string, "AF_PACKET");
1180 if (sll.sll_ifindex) {
1181 char name[IFNAMSIZ];
1183 if (if_indextoname(sll.sll_ifindex, name)) {
1184 ds_put_format(string, "(%s)", name);
1186 ds_put_format(string, "(ifindex=%d)", sll.sll_ifindex);
1189 if (sll.sll_protocol) {
1190 ds_put_format(string, "(protocol=0x%"PRIu16")",
1191 ntohs(sll.sll_protocol));
1195 else if (ss.ss_family == AF_UNSPEC) {
1196 ds_put_cstr(string, "AF_UNSPEC");
1198 ds_put_format(string, "AF_%d", (int) ss.ss_family);
1204 #ifdef LINUX_DATAPATH
1206 put_fd_filename(struct ds *string, int fd)
1212 linkname = xasprintf("/proc/self/fd/%d", fd);
1213 n = readlink(linkname, buf, sizeof buf);
1215 ds_put_char(string, ' ');
1216 ds_put_buffer(string, buf, n);
1217 if (n > sizeof buf) {
1218 ds_put_cstr(string, "...");
1225 /* Returns a malloc()'d string describing 'fd', for use in logging. */
1233 if (fstat(fd, &s)) {
1234 ds_put_format(&string, "fstat failed (%s)", ovs_strerror(errno));
1235 } else if (S_ISSOCK(s.st_mode)) {
1236 describe_sockaddr(&string, fd, getsockname);
1237 ds_put_cstr(&string, "<->");
1238 describe_sockaddr(&string, fd, getpeername);
1240 ds_put_cstr(&string, (isatty(fd) ? "tty"
1241 : S_ISDIR(s.st_mode) ? "directory"
1242 : S_ISCHR(s.st_mode) ? "character device"
1243 : S_ISBLK(s.st_mode) ? "block device"
1244 : S_ISREG(s.st_mode) ? "file"
1245 : S_ISFIFO(s.st_mode) ? "FIFO"
1246 : S_ISLNK(s.st_mode) ? "symbolic link"
1248 #ifdef LINUX_DATAPATH
1249 put_fd_filename(&string, fd);
1252 return ds_steal_cstr(&string);
1255 /* Calls ioctl() on an AF_INET sock, passing the specified 'command' and
1256 * 'arg'. Returns 0 if successful, otherwise a positive errno value. */
1258 af_inet_ioctl(unsigned long int command, const void *arg)
1260 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
1263 if (ovsthread_once_start(&once)) {
1264 sock = socket(AF_INET, SOCK_DGRAM, 0);
1267 VLOG_ERR("failed to create inet socket: %s", ovs_strerror(errno));
1269 ovsthread_once_done(&once);
1272 return (sock < 0 ? -sock
1273 : ioctl(sock, command, arg) == -1 ? errno
1278 af_inet_ifreq_ioctl(const char *name, struct ifreq *ifr, unsigned long int cmd,
1279 const char *cmd_name)
1283 ovs_strzcpy(ifr->ifr_name, name, sizeof ifr->ifr_name);
1284 error = af_inet_ioctl(cmd, ifr);
1286 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1287 VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name,
1288 ovs_strerror(error));
1293 /* sockaddr_storage helpers. */
1295 /* Returns the IPv4 or IPv6 port in 'ss'. */
1297 ss_get_port(const struct sockaddr_storage *ss)
1299 if (ss->ss_family == AF_INET) {
1300 const struct sockaddr_in *sin
1301 = ALIGNED_CAST(const struct sockaddr_in *, ss);
1302 return ntohs(sin->sin_port);
1303 } else if (ss->ss_family == AF_INET6) {
1304 const struct sockaddr_in6 *sin6
1305 = ALIGNED_CAST(const struct sockaddr_in6 *, ss);
1306 return ntohs(sin6->sin6_port);
1312 /* Formats the IPv4 or IPv6 address in 'ss' into the 'bufsize' bytes in 'buf'.
1313 * If 'ss' is an IPv6 address, puts square brackets around the address.
1314 * 'bufsize' should be at least SS_NTOP_BUFSIZE.
1318 ss_format_address(const struct sockaddr_storage *ss,
1319 char *buf, size_t bufsize)
1321 ovs_assert(bufsize >= SS_NTOP_BUFSIZE);
1322 if (ss->ss_family == AF_INET) {
1323 const struct sockaddr_in *sin
1324 = ALIGNED_CAST(const struct sockaddr_in *, ss);
1326 snprintf(buf, bufsize, IP_FMT, IP_ARGS(sin->sin_addr.s_addr));
1327 } else if (ss->ss_family == AF_INET6) {
1328 const struct sockaddr_in6 *sin6
1329 = ALIGNED_CAST(const struct sockaddr_in6 *, ss);
1332 inet_ntop(AF_INET6, sin6->sin6_addr.s6_addr, buf + 1, bufsize - 1);
1333 strcpy(strchr(buf, '\0'), "]");
1342 ss_length(const struct sockaddr_storage *ss)
1344 switch (ss->ss_family) {
1346 return sizeof(struct sockaddr_in);
1349 return sizeof(struct sockaddr_in6);
1356 /* For Windows socket calls, 'errno' is not set. One has to call
1357 * WSAGetLastError() to get the error number and then pass it to
1358 * FormatMessage() (through this function) to get the correct error string.
1360 * ovs_strerror() calls strerror_r() and would not get the correct error
1361 * string for Windows sockets, but is good for POSIX. */
1363 sock_strerror(int error)
1366 enum { BUFSIZE = sizeof sockerror_buffer_get()->s };
1367 char *buffer = sockerror_buffer_get()->s;
1369 FormatMessage(FORMAT_MESSAGE_FROM_SYSTEM
1370 | FORMAT_MESSAGE_IGNORE_INSERTS, NULL, error, 0,
1371 buffer, BUFSIZE, NULL);
1374 return ovs_strerror(error);