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"
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)
75 int flags = fcntl(fd, F_GETFL, 0);
77 if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) != -1) {
80 VLOG_ERR("fcntl(F_SETFL) failed: %s", ovs_strerror(errno));
84 VLOG_ERR("fcntl(F_GETFL) failed: %s", ovs_strerror(errno));
90 xset_nonblocking(int fd)
92 if (set_nonblocking(fd)) {
98 set_dscp(int fd, uint8_t dscp)
107 if (setsockopt(fd, IPPROTO_IP, IP_TOS, &val, sizeof val)) {
115 rlim_is_finite(rlim_t limit)
117 if (limit == RLIM_INFINITY) {
121 #ifdef RLIM_SAVED_CUR /* FreeBSD 8.0 lacks RLIM_SAVED_CUR. */
122 if (limit == RLIM_SAVED_CUR) {
127 #ifdef RLIM_SAVED_MAX /* FreeBSD 8.0 lacks RLIM_SAVED_MAX. */
128 if (limit == RLIM_SAVED_MAX) {
136 /* Returns the maximum valid FD value, plus 1. */
140 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
143 if (ovsthread_once_start(&once)) {
145 if (!getrlimit(RLIMIT_NOFILE, &r) && rlim_is_finite(r.rlim_cur)) {
146 max_fds = r.rlim_cur;
148 VLOG_WARN("failed to obtain fd limit, defaulting to 1024");
151 ovsthread_once_done(&once);
157 /* Translates 'host_name', which must be a string representation of an IP
158 * address, into a numeric IP address in '*addr'. Returns 0 if successful,
159 * otherwise a positive errno value. */
161 lookup_ip(const char *host_name, struct in_addr *addr)
163 if (!inet_aton(host_name, addr)) {
164 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
165 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name);
171 /* Translates 'host_name', which must be a string representation of an IPv6
172 * address, into a numeric IPv6 address in '*addr'. Returns 0 if successful,
173 * otherwise a positive errno value. */
175 lookup_ipv6(const char *host_name, struct in6_addr *addr)
177 if (inet_pton(AF_INET6, host_name, addr) != 1) {
178 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
179 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IPv6 address", host_name);
185 /* Translates 'host_name', which must be a host name or a string representation
186 * of an IP address, into a numeric IP address in '*addr'. Returns 0 if
187 * successful, otherwise a positive errno value.
189 * Most Open vSwitch code should not use this because it causes deadlocks:
190 * getaddrinfo() sends out a DNS request but that starts a new flow for which
191 * OVS must set up a flow, but it can't because it's waiting for a DNS reply.
192 * The synchronous lookup also delays other activity. (Of course we can solve
193 * this but it doesn't seem worthwhile quite yet.) */
195 lookup_hostname(const char *host_name, struct in_addr *addr)
197 struct addrinfo *result;
198 struct addrinfo hints;
200 if (inet_aton(host_name, addr)) {
204 memset(&hints, 0, sizeof hints);
205 hints.ai_family = AF_INET;
207 switch (getaddrinfo(host_name, NULL, &hints, &result)) {
209 *addr = ALIGNED_CAST(struct sockaddr_in *,
210 result->ai_addr)->sin_addr;
211 freeaddrinfo(result);
214 #ifdef EAI_ADDRFAMILY
249 check_connection_completion(int fd)
251 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
256 pfd.events = POLLOUT;
258 retval = poll(&pfd, 1, 0);
259 } while (retval < 0 && errno == EINTR);
261 if (pfd.revents & POLLERR) {
262 ssize_t n = send(fd, "", 1, MSG_DONTWAIT);
266 VLOG_ERR_RL(&rl, "poll return POLLERR but send succeeded");
271 } else if (retval < 0) {
272 VLOG_ERR_RL(&rl, "poll: %s", ovs_strerror(errno));
279 /* Drain all the data currently in the receive queue of a datagram socket (and
280 * possibly additional data). There is no way to know how many packets are in
281 * the receive queue, but we do know that the total number of bytes queued does
282 * not exceed the receive buffer size, so we pull packets until none are left
283 * or we've read that many bytes. */
289 rcvbuf = get_socket_rcvbuf(fd);
295 /* In Linux, specifying MSG_TRUNC in the flags argument causes the
296 * datagram length to be returned, even if that is longer than the
297 * buffer provided. Thus, we can use a 1-byte buffer to discard the
298 * incoming datagram and still be able to account how many bytes were
299 * removed from the receive buffer.
301 * On other Unix-like OSes, MSG_TRUNC has no effect in the flags
303 char buffer[LINUX_DATAPATH ? 1 : 2048];
304 ssize_t n_bytes = recv(fd, buffer, sizeof buffer,
305 MSG_TRUNC | MSG_DONTWAIT);
306 if (n_bytes <= 0 || n_bytes >= rcvbuf) {
314 /* Returns the size of socket 'sock''s receive buffer (SO_RCVBUF), or a
315 * negative errno value if an error occurs. */
317 get_socket_rcvbuf(int sock)
322 error = getsockopt_int(sock, SOL_SOCKET, SO_RCVBUF, "SO_RCVBUF", &rcvbuf);
323 return error ? -error : rcvbuf;
326 /* Reads and discards up to 'n' datagrams from 'fd', stopping as soon as no
327 * more data can be immediately read. ('fd' should therefore be in
328 * non-blocking mode.)*/
330 drain_fd(int fd, size_t n_packets)
332 for (; n_packets > 0; n_packets--) {
333 /* 'buffer' only needs to be 1 byte long in most circumstances. This
334 * size is defensive against the possibility that we someday want to
335 * use a Linux tap device without TUN_NO_PI, in which case a buffer
336 * smaller than sizeof(struct tun_pi) will give EINVAL on read. */
338 if (read(fd, buffer, sizeof buffer) <= 0) {
344 /* Attempts to shorten 'name' by opening a file descriptor for the directory
345 * part of the name and indirecting through /proc/self/fd/<dirfd>/<basename>.
346 * On systems with Linux-like /proc, this works as long as <basename> isn't too
349 * On success, returns 0 and stores the short name in 'short_name' and a
350 * directory file descriptor to eventually be closed in '*dirfpd'. */
352 shorten_name_via_proc(const char *name, char short_name[MAX_UN_LEN + 1],
359 if (!LINUX_DATAPATH) {
363 dir = dir_name(name);
364 dirfd = open(dir, O_DIRECTORY | O_RDONLY);
366 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
369 VLOG_WARN_RL(&rl, "%s: open failed (%s)", dir, ovs_strerror(error));
376 base = base_name(name);
377 len = snprintf(short_name, MAX_UN_LEN + 1,
378 "/proc/self/fd/%d/%s", dirfd, base);
381 if (len >= 0 && len <= MAX_UN_LEN) {
390 /* Attempts to shorten 'name' by creating a symlink for the directory part of
391 * the name and indirecting through <symlink>/<basename>. This works on
392 * systems that support symlinks, as long as <basename> isn't too long.
394 * On success, returns 0 and stores the short name in 'short_name' and the
395 * symbolic link to eventually delete in 'linkname'. */
397 shorten_name_via_symlink(const char *name, char short_name[MAX_UN_LEN + 1],
398 char linkname[MAX_UN_LEN + 1])
400 char *abs, *dir, *base;
405 abs = abs_file_name(NULL, name);
407 base = base_name(abs);
410 tmpdir = getenv("TMPDIR");
411 if (tmpdir == NULL) {
415 for (i = 0; i < 1000; i++) {
418 len = snprintf(linkname, MAX_UN_LEN + 1,
419 "%s/ovs-un-c-%"PRIu32, tmpdir, random_uint32());
420 error = (len < 0 || len > MAX_UN_LEN ? ENAMETOOLONG
421 : symlink(dir, linkname) ? errno
423 if (error != EEXIST) {
431 fatal_signal_add_file_to_unlink(linkname);
433 len = snprintf(short_name, MAX_UN_LEN + 1, "%s/%s", linkname, base);
434 if (len < 0 || len > MAX_UN_LEN) {
435 fatal_signal_unlink_file_now(linkname);
436 error = ENAMETOOLONG;
449 /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in
450 * '*un_len' the size of the sockaddr_un.
452 * Returns 0 on success, otherwise a positive errno value.
454 * Uses '*dirfdp' and 'linkname' to store references to data when the caller no
455 * longer needs to use 'un'. On success, freeing these references with
456 * free_sockaddr_un() is mandatory to avoid a leak; on failure, freeing them is
457 * unnecessary but harmless. */
459 make_sockaddr_un(const char *name, struct sockaddr_un *un, socklen_t *un_len,
460 int *dirfdp, char linkname[MAX_UN_LEN + 1])
462 char short_name[MAX_UN_LEN + 1];
466 if (strlen(name) > MAX_UN_LEN) {
467 /* 'name' is too long to fit in a sockaddr_un. Try a workaround. */
468 int error = shorten_name_via_proc(name, short_name, dirfdp);
469 if (error == ENAMETOOLONG) {
470 error = shorten_name_via_symlink(name, short_name, linkname);
473 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
475 VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum "
476 "%"PRIuSIZE" bytes", name, MAX_UN_LEN);
483 un->sun_family = AF_UNIX;
484 ovs_strzcpy(un->sun_path, name, sizeof un->sun_path);
485 *un_len = (offsetof(struct sockaddr_un, sun_path)
486 + strlen (un->sun_path) + 1);
490 /* Clean up after make_sockaddr_un(). */
492 free_sockaddr_un(int dirfd, const char *linkname)
498 fatal_signal_unlink_file_now(linkname);
502 /* Binds Unix domain socket 'fd' to a file with permissions 0700. */
504 bind_unix_socket(int fd, struct sockaddr *sun, socklen_t sun_len)
506 /* According to _Unix Network Programming_, umask should affect bind(). */
507 mode_t old_umask = umask(0077);
508 int error = bind(fd, sun, sun_len) ? errno : 0;
513 /* Creates a Unix domain socket in the given 'style' (either SOCK_DGRAM or
514 * SOCK_STREAM) that is bound to '*bind_path' (if 'bind_path' is non-null) and
515 * connected to '*connect_path' (if 'connect_path' is non-null). If 'nonblock'
516 * is true, the socket is made non-blocking.
518 * Returns the socket's fd if successful, otherwise a negative errno value. */
520 make_unix_socket(int style, bool nonblock,
521 const char *bind_path, const char *connect_path)
526 fd = socket(PF_UNIX, style, 0);
531 /* Set nonblocking mode right away, if we want it. This prevents blocking
532 * in connect(), if connect_path != NULL. (In turn, that's a corner case:
533 * it will only happen if style is SOCK_STREAM or SOCK_SEQPACKET, and only
534 * if a backlog of un-accepted connections has built up in the kernel.) */
536 error = set_nonblocking(fd);
543 char linkname[MAX_UN_LEN + 1];
544 struct sockaddr_un un;
548 if (unlink(bind_path) && errno != ENOENT) {
549 VLOG_WARN("unlinking \"%s\": %s\n",
550 bind_path, ovs_strerror(errno));
552 fatal_signal_add_file_to_unlink(bind_path);
554 error = make_sockaddr_un(bind_path, &un, &un_len, &dirfd, linkname);
556 error = bind_unix_socket(fd, (struct sockaddr *) &un, un_len);
558 free_sockaddr_un(dirfd, linkname);
566 char linkname[MAX_UN_LEN + 1];
567 struct sockaddr_un un;
571 error = make_sockaddr_un(connect_path, &un, &un_len, &dirfd, linkname);
573 && connect(fd, (struct sockaddr*) &un, un_len)
574 && errno != EINPROGRESS) {
577 free_sockaddr_un(dirfd, linkname);
587 if (error == EAGAIN) {
591 fatal_signal_unlink_file_now(bind_path);
598 get_unix_name_len(socklen_t sun_len)
600 return (sun_len >= offsetof(struct sockaddr_un, sun_path)
601 ? sun_len - offsetof(struct sockaddr_un, sun_path)
606 guess_netmask(ovs_be32 ip_)
608 uint32_t ip = ntohl(ip_);
609 return ((ip >> 31) == 0 ? htonl(0xff000000) /* Class A */
610 : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */
611 : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */
612 : htonl(0)); /* ??? */
615 /* This is like strsep() except:
617 * - The separator string is ":".
619 * - Square brackets [] quote ":" separators and are removed from the
622 parse_bracketed_token(char **pp)
628 } else if (*p == '\0') {
631 } else if (*p == '[') {
633 char *end = start + strcspn(start, "]");
634 *pp = (*end == '\0' ? NULL
635 : end[1] == ':' ? end + 2
641 char *end = start + strcspn(start, ":");
642 *pp = *end == '\0' ? NULL : end + 1;
649 parse_sockaddr_components(struct sockaddr_storage *ss,
651 const char *port_s, uint16_t default_port,
654 struct sockaddr_in *sin = ALIGNED_CAST(struct sockaddr_in *, ss);
657 if (port_s && port_s[0]) {
658 if (!str_to_int(port_s, 10, &port) || port < 0 || port > 65535) {
659 VLOG_ERR("%s: bad port number \"%s\"", s, port_s);
665 memset(ss, 0, sizeof *ss);
666 if (strchr(host_s, ':')) {
667 struct sockaddr_in6 *sin6
668 = ALIGNED_CAST(struct sockaddr_in6 *, ss);
670 sin6->sin6_family = AF_INET6;
671 sin6->sin6_port = htons(port);
672 if (!inet_pton(AF_INET6, host_s, sin6->sin6_addr.s6_addr)) {
673 VLOG_ERR("%s: bad IPv6 address \"%s\"", s, host_s);
677 sin->sin_family = AF_INET;
678 sin->sin_port = htons(port);
679 if (!inet_pton(AF_INET, host_s, &sin->sin_addr.s_addr)) {
680 VLOG_ERR("%s: bad IPv4 address \"%s\"", s, host_s);
688 memset(ss, 0, sizeof *ss);
692 /* Parses 'target', which should be a string in the format "<host>[:<port>]".
693 * <host>, which is required, may be an IPv4 address or an IPv6 address
694 * enclosed in square brackets. If 'default_port' is nonzero then <port> is
695 * optional and defaults to 'default_port'.
697 * On success, returns true and stores the parsed remote address into '*ss'.
698 * On failure, logs an error, stores zeros into '*ss', and returns false. */
700 inet_parse_active(const char *target_, uint16_t default_port,
701 struct sockaddr_storage *ss)
703 char *target = xstrdup(target_);
710 host = parse_bracketed_token(&p);
711 port = parse_bracketed_token(&p);
713 VLOG_ERR("%s: host must be specified", target_);
715 } else if (!port && !default_port) {
716 VLOG_ERR("%s: port must be specified", target_);
719 ok = parse_sockaddr_components(ss, host, port, default_port, target_);
722 memset(ss, 0, sizeof *ss);
729 /* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style' and
730 * connects to 'target', which should be a string in the format
731 * "<host>[:<port>]". <host>, which is required, may be an IPv4 address or an
732 * IPv6 address enclosed in square brackets. If 'default_port' is nonzero then
733 * <port> is optional and defaults to 'default_port'.
735 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
737 * On success, returns 0 (indicating connection complete) or EAGAIN (indicating
738 * connection in progress), in which case the new file descriptor is stored
739 * into '*fdp'. On failure, returns a positive errno value other than EAGAIN
740 * and stores -1 into '*fdp'.
742 * If 'ss' is non-null, then on success stores the target address into '*ss'.
744 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
745 * should be in the range [0, 63] and will automatically be shifted to the
746 * appropriately place in the IP tos field. */
748 inet_open_active(int style, const char *target, uint16_t default_port,
749 struct sockaddr_storage *ssp, int *fdp, uint8_t dscp)
751 struct sockaddr_storage ss;
756 if (!inet_parse_active(target, default_port, &ss)) {
757 error = EAFNOSUPPORT;
761 /* Create non-blocking socket. */
762 fd = socket(ss.ss_family, style, 0);
764 VLOG_ERR("%s: socket: %s", target, ovs_strerror(errno));
768 error = set_nonblocking(fd);
773 /* The dscp bits must be configured before connect() to ensure that the
774 * TOS field is set during the connection establishment. If set after
775 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
776 error = set_dscp(fd, dscp);
778 VLOG_ERR("%s: socket: %s", target, ovs_strerror(error));
783 error = connect(fd, (struct sockaddr *) &ss, ss_length(&ss)) == 0
786 if (error == EINPROGRESS) {
791 if (error && error != EAGAIN) {
793 memset(ssp, 0, sizeof *ssp);
808 /* Parses 'target', which should be a string in the format "[<port>][:<host>]":
810 * - If 'default_port' is -1, then <port> is required. Otherwise, if
811 * <port> is omitted, then 'default_port' is used instead.
813 * - If <port> (or 'default_port', if used) is 0, then no port is bound
814 * and the TCP/IP stack will select a port.
816 * - <host> is optional. If supplied, it may be an IPv4 address or an
817 * IPv6 address enclosed in square brackets. If omitted, the IP address
820 * If successful, stores the address into '*ss' and returns true; otherwise
821 * zeros '*ss' and returns false. */
823 inet_parse_passive(const char *target_, int default_port,
824 struct sockaddr_storage *ss)
826 char *target = xstrdup(target_);
833 port = parse_bracketed_token(&p);
834 host = parse_bracketed_token(&p);
835 if (!port && default_port < 0) {
836 VLOG_ERR("%s: port must be specified", target_);
839 ok = parse_sockaddr_components(ss, host ? host : "0.0.0.0",
840 port, default_port, target_);
843 memset(ss, 0, sizeof *ss);
850 /* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style', binds to
851 * 'target', and listens for incoming connections. Parses 'target' in the same
852 * way was inet_parse_passive().
854 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
856 * For TCP, the socket will have SO_REUSEADDR turned on.
858 * On success, returns a non-negative file descriptor. On failure, returns a
859 * negative errno value.
861 * If 'ss' is non-null, then on success stores the bound address into '*ss'.
863 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
864 * should be in the range [0, 63] and will automatically be shifted to the
865 * appropriately place in the IP tos field. */
867 inet_open_passive(int style, const char *target, int default_port,
868 struct sockaddr_storage *ssp, uint8_t dscp)
870 bool kernel_chooses_port;
871 struct sockaddr_storage ss;
873 unsigned int yes = 1;
875 if (!inet_parse_passive(target, default_port, &ss)) {
876 return -EAFNOSUPPORT;
878 kernel_chooses_port = ss_get_port(&ss) == 0;
880 /* Create non-blocking socket, set SO_REUSEADDR. */
881 fd = socket(ss.ss_family, style, 0);
884 VLOG_ERR("%s: socket: %s", target, ovs_strerror(error));
887 error = set_nonblocking(fd);
891 if (style == SOCK_STREAM
892 && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) {
894 VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s",
895 target, ovs_strerror(error));
900 if (bind(fd, (struct sockaddr *) &ss, ss_length(&ss)) < 0) {
902 VLOG_ERR("%s: bind: %s", target, ovs_strerror(error));
906 /* The dscp bits must be configured before connect() to ensure that the TOS
907 * field is set during the connection establishment. If set after
908 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
909 error = set_dscp(fd, dscp);
911 VLOG_ERR("%s: socket: %s", target, ovs_strerror(error));
916 if (style == SOCK_STREAM && listen(fd, 10) < 0) {
918 VLOG_ERR("%s: listen: %s", target, ovs_strerror(error));
922 if (ssp || kernel_chooses_port) {
923 socklen_t ss_len = sizeof ss;
924 if (getsockname(fd, (struct sockaddr *) &ss, &ss_len) < 0) {
926 VLOG_ERR("%s: getsockname: %s", target, ovs_strerror(error));
929 if (kernel_chooses_port) {
930 VLOG_INFO("%s: listening on port %"PRIu16,
931 target, ss_get_port(&ss));
942 memset(ssp, 0, sizeof *ssp);
948 /* Returns a readable and writable fd for /dev/null, if successful, otherwise
949 * a negative errno value. The caller must not close the returned fd (because
950 * the same fd will be handed out to subsequent callers). */
954 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
957 if (ovsthread_once_start(&once)) {
958 null_fd = open("/dev/null", O_RDWR);
961 VLOG_ERR("could not open /dev/null: %s", ovs_strerror(error));
964 ovsthread_once_done(&once);
971 read_fully(int fd, void *p_, size_t size, size_t *bytes_read)
977 ssize_t retval = read(fd, p, size);
979 *bytes_read += retval;
982 } else if (retval == 0) {
984 } else if (errno != EINTR) {
992 write_fully(int fd, const void *p_, size_t size, size_t *bytes_written)
994 const uint8_t *p = p_;
998 ssize_t retval = write(fd, p, size);
1000 *bytes_written += retval;
1003 } else if (retval == 0) {
1004 VLOG_WARN("write returned 0");
1006 } else if (errno != EINTR) {
1013 /* Given file name 'file_name', fsyncs the directory in which it is contained.
1014 * Returns 0 if successful, otherwise a positive errno value. */
1016 fsync_parent_dir(const char *file_name)
1022 dir = dir_name(file_name);
1023 fd = open(dir, O_RDONLY);
1026 if (errno == EINVAL || errno == EROFS) {
1027 /* This directory does not support synchronization. Not
1028 * really an error. */
1031 VLOG_ERR("%s: fsync failed (%s)", dir, ovs_strerror(error));
1037 VLOG_ERR("%s: open failed (%s)", dir, ovs_strerror(error));
1044 /* Obtains the modification time of the file named 'file_name' to the greatest
1045 * supported precision. If successful, stores the mtime in '*mtime' and
1046 * returns 0. On error, returns a positive errno value and stores zeros in
1049 get_mtime(const char *file_name, struct timespec *mtime)
1053 if (!stat(file_name, &s)) {
1054 mtime->tv_sec = s.st_mtime;
1056 #if HAVE_STRUCT_STAT_ST_MTIM_TV_NSEC
1057 mtime->tv_nsec = s.st_mtim.tv_nsec;
1058 #elif HAVE_STRUCT_STAT_ST_MTIMENSEC
1059 mtime->tv_nsec = s.st_mtimensec;
1066 mtime->tv_sec = mtime->tv_nsec = 0;
1075 VLOG_FATAL("failed to create pipe (%s)", ovs_strerror(errno));
1080 xpipe_nonblocking(int fds[2])
1083 xset_nonblocking(fds[0]);
1084 xset_nonblocking(fds[1]);
1088 getsockopt_int(int fd, int level, int option, const char *optname, int *valuep)
1090 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
1096 if (getsockopt(fd, level, option, &value, &len)) {
1098 VLOG_ERR_RL(&rl, "getsockopt(%s): %s", optname, ovs_strerror(error));
1099 } else if (len != sizeof value) {
1101 VLOG_ERR_RL(&rl, "getsockopt(%s): value is %u bytes (expected %"PRIuSIZE")",
1102 optname, (unsigned int) len, sizeof value);
1107 *valuep = error ? 0 : value;
1112 describe_sockaddr(struct ds *string, int fd,
1113 int (*getaddr)(int, struct sockaddr *, socklen_t *))
1115 struct sockaddr_storage ss;
1116 socklen_t len = sizeof ss;
1118 if (!getaddr(fd, (struct sockaddr *) &ss, &len)) {
1119 if (ss.ss_family == AF_INET || ss.ss_family == AF_INET6) {
1120 char addrbuf[SS_NTOP_BUFSIZE];
1122 ds_put_format(string, "%s:%"PRIu16,
1123 ss_format_address(&ss, addrbuf, sizeof addrbuf),
1125 } else if (ss.ss_family == AF_UNIX) {
1126 struct sockaddr_un sun;
1130 memcpy(&sun, &ss, sizeof sun);
1131 maxlen = len - offsetof(struct sockaddr_un, sun_path);
1132 null = memchr(sun.sun_path, '\0', maxlen);
1133 ds_put_buffer(string, sun.sun_path,
1134 null ? null - sun.sun_path : maxlen);
1137 else if (ss.ss_family == AF_NETLINK) {
1140 /* SO_PROTOCOL was introduced in 2.6.32. Support it regardless of the version
1141 * of the Linux kernel headers in use at build time. */
1143 #define SO_PROTOCOL 38
1146 if (!getsockopt_int(fd, SOL_SOCKET, SO_PROTOCOL, "SO_PROTOCOL",
1150 ds_put_cstr(string, "NETLINK_ROUTE");
1153 case NETLINK_GENERIC:
1154 ds_put_cstr(string, "NETLINK_GENERIC");
1158 ds_put_format(string, "AF_NETLINK family %d", protocol);
1162 ds_put_cstr(string, "AF_NETLINK");
1166 #if AF_PACKET && LINUX_DATAPATH
1167 else if (ss.ss_family == AF_PACKET) {
1168 struct sockaddr_ll sll;
1170 memcpy(&sll, &ss, sizeof sll);
1171 ds_put_cstr(string, "AF_PACKET");
1172 if (sll.sll_ifindex) {
1173 char name[IFNAMSIZ];
1175 if (if_indextoname(sll.sll_ifindex, name)) {
1176 ds_put_format(string, "(%s)", name);
1178 ds_put_format(string, "(ifindex=%d)", sll.sll_ifindex);
1181 if (sll.sll_protocol) {
1182 ds_put_format(string, "(protocol=0x%"PRIu16")",
1183 ntohs(sll.sll_protocol));
1187 else if (ss.ss_family == AF_UNSPEC) {
1188 ds_put_cstr(string, "AF_UNSPEC");
1190 ds_put_format(string, "AF_%d", (int) ss.ss_family);
1196 #ifdef LINUX_DATAPATH
1198 put_fd_filename(struct ds *string, int fd)
1204 linkname = xasprintf("/proc/self/fd/%d", fd);
1205 n = readlink(linkname, buf, sizeof buf);
1207 ds_put_char(string, ' ');
1208 ds_put_buffer(string, buf, n);
1209 if (n > sizeof buf) {
1210 ds_put_cstr(string, "...");
1217 /* Returns a malloc()'d string describing 'fd', for use in logging. */
1225 if (fstat(fd, &s)) {
1226 ds_put_format(&string, "fstat failed (%s)", ovs_strerror(errno));
1227 } else if (S_ISSOCK(s.st_mode)) {
1228 describe_sockaddr(&string, fd, getsockname);
1229 ds_put_cstr(&string, "<->");
1230 describe_sockaddr(&string, fd, getpeername);
1232 ds_put_cstr(&string, (isatty(fd) ? "tty"
1233 : S_ISDIR(s.st_mode) ? "directory"
1234 : S_ISCHR(s.st_mode) ? "character device"
1235 : S_ISBLK(s.st_mode) ? "block device"
1236 : S_ISREG(s.st_mode) ? "file"
1237 : S_ISFIFO(s.st_mode) ? "FIFO"
1238 : S_ISLNK(s.st_mode) ? "symbolic link"
1240 #ifdef LINUX_DATAPATH
1241 put_fd_filename(&string, fd);
1244 return ds_steal_cstr(&string);
1247 /* Calls ioctl() on an AF_INET sock, passing the specified 'command' and
1248 * 'arg'. Returns 0 if successful, otherwise a positive errno value. */
1250 af_inet_ioctl(unsigned long int command, const void *arg)
1252 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
1255 if (ovsthread_once_start(&once)) {
1256 sock = socket(AF_INET, SOCK_DGRAM, 0);
1259 VLOG_ERR("failed to create inet socket: %s", ovs_strerror(errno));
1261 ovsthread_once_done(&once);
1264 return (sock < 0 ? -sock
1265 : ioctl(sock, command, arg) == -1 ? errno
1270 af_inet_ifreq_ioctl(const char *name, struct ifreq *ifr, unsigned long int cmd,
1271 const char *cmd_name)
1275 ovs_strzcpy(ifr->ifr_name, name, sizeof ifr->ifr_name);
1276 error = af_inet_ioctl(cmd, ifr);
1278 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1279 VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name,
1280 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);