2 * Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks.
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/resource.h>
30 #include <sys/socket.h>
34 #include "dynamic-string.h"
35 #include "fatal-signal.h"
39 #if AF_PACKET && __linux__
40 #include <linux/if_packet.h>
43 #include "netlink-protocol.h"
44 #include "netlink-socket.h"
47 VLOG_DEFINE_THIS_MODULE(socket_util);
49 /* #ifdefs make it a pain to maintain code: you have to try to build both ways.
50 * Thus, this file compiles all of the code regardless of the target, by
51 * writing "if (LINUX)" instead of "#ifdef __linux__". */
62 /* Sets 'fd' to non-blocking mode. Returns 0 if successful, otherwise a
63 * positive errno value. */
65 set_nonblocking(int fd)
67 int flags = fcntl(fd, F_GETFL, 0);
69 if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) != -1) {
72 VLOG_ERR("fcntl(F_SETFL) failed: %s", strerror(errno));
76 VLOG_ERR("fcntl(F_GETFL) failed: %s", strerror(errno));
82 rlim_is_finite(rlim_t limit)
84 if (limit == RLIM_INFINITY) {
88 #ifdef RLIM_SAVED_CUR /* FreeBSD 8.0 lacks RLIM_SAVED_CUR. */
89 if (limit == RLIM_SAVED_CUR) {
94 #ifdef RLIM_SAVED_MAX /* FreeBSD 8.0 lacks RLIM_SAVED_MAX. */
95 if (limit == RLIM_SAVED_MAX) {
103 /* Returns the maximum valid FD value, plus 1. */
107 static int max_fds = -1;
110 if (!getrlimit(RLIMIT_NOFILE, &r) && rlim_is_finite(r.rlim_cur)) {
111 max_fds = r.rlim_cur;
113 VLOG_WARN("failed to obtain fd limit, defaulting to 1024");
120 /* Translates 'host_name', which must be a string representation of an IP
121 * address, into a numeric IP address in '*addr'. Returns 0 if successful,
122 * otherwise a positive errno value. */
124 lookup_ip(const char *host_name, struct in_addr *addr)
126 if (!inet_aton(host_name, addr)) {
127 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
128 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name);
134 /* Translates 'host_name', which must be a string representation of an IPv6
135 * address, into a numeric IPv6 address in '*addr'. Returns 0 if successful,
136 * otherwise a positive errno value. */
138 lookup_ipv6(const char *host_name, struct in6_addr *addr)
140 if (inet_pton(AF_INET6, host_name, addr) != 1) {
141 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
142 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IPv6 address", host_name);
148 /* Returns the error condition associated with socket 'fd' and resets the
149 * socket's error status. */
151 get_socket_error(int fd)
154 socklen_t len = sizeof(error);
155 if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &error, &len) < 0) {
156 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
158 VLOG_ERR_RL(&rl, "getsockopt(SO_ERROR): %s", strerror(error));
164 check_connection_completion(int fd)
170 pfd.events = POLLOUT;
172 retval = poll(&pfd, 1, 0);
173 } while (retval < 0 && errno == EINTR);
175 return get_socket_error(fd);
176 } else if (retval < 0) {
177 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
178 VLOG_ERR_RL(&rl, "poll: %s", strerror(errno));
185 /* Drain all the data currently in the receive queue of a datagram socket (and
186 * possibly additional data). There is no way to know how many packets are in
187 * the receive queue, but we do know that the total number of bytes queued does
188 * not exceed the receive buffer size, so we pull packets until none are left
189 * or we've read that many bytes. */
193 socklen_t rcvbuf_len;
196 rcvbuf_len = sizeof rcvbuf;
197 if (getsockopt(fd, SOL_SOCKET, SO_RCVBUF, &rcvbuf, &rcvbuf_len) < 0) {
198 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
199 VLOG_ERR_RL(&rl, "getsockopt(SO_RCVBUF) failed: %s", strerror(errno));
203 /* In Linux, specifying MSG_TRUNC in the flags argument causes the
204 * datagram length to be returned, even if that is longer than the
205 * buffer provided. Thus, we can use a 1-byte buffer to discard the
206 * incoming datagram and still be able to account how many bytes were
207 * removed from the receive buffer.
209 * On other Unix-like OSes, MSG_TRUNC has no effect in the flags
211 char buffer[LINUX ? 1 : 2048];
212 ssize_t n_bytes = recv(fd, buffer, sizeof buffer,
213 MSG_TRUNC | MSG_DONTWAIT);
214 if (n_bytes <= 0 || n_bytes >= rcvbuf) {
222 /* Reads and discards up to 'n' datagrams from 'fd', stopping as soon as no
223 * more data can be immediately read. ('fd' should therefore be in
224 * non-blocking mode.)*/
226 drain_fd(int fd, size_t n_packets)
228 for (; n_packets > 0; n_packets--) {
229 /* 'buffer' only needs to be 1 byte long in most circumstances. This
230 * size is defensive against the possibility that we someday want to
231 * use a Linux tap device without TUN_NO_PI, in which case a buffer
232 * smaller than sizeof(struct tun_pi) will give EINVAL on read. */
234 if (read(fd, buffer, sizeof buffer) <= 0) {
240 /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in
241 * '*un_len' the size of the sockaddr_un. */
243 make_sockaddr_un__(const char *name, struct sockaddr_un *un, socklen_t *un_len)
245 un->sun_family = AF_UNIX;
246 ovs_strzcpy(un->sun_path, name, sizeof un->sun_path);
247 *un_len = (offsetof(struct sockaddr_un, sun_path)
248 + strlen (un->sun_path) + 1);
251 /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in
252 * '*un_len' the size of the sockaddr_un.
254 * Returns 0 on success, otherwise a positive errno value. On success,
255 * '*dirfdp' is either -1 or a nonnegative file descriptor that the caller
256 * should close after using '*un' to bind or connect. On failure, '*dirfdp' is
259 make_sockaddr_un(const char *name, struct sockaddr_un *un, socklen_t *un_len,
262 enum { MAX_UN_LEN = sizeof un->sun_path - 1 };
265 if (strlen(name) > MAX_UN_LEN) {
266 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
269 /* 'name' is too long to fit in a sockaddr_un, but we have a
270 * workaround for that on Linux: shorten it by opening a file
271 * descriptor for the directory part of the name and indirecting
272 * through /proc/self/fd/<dirfd>/<basename>. */
277 dir = dir_name(name);
278 base = base_name(name);
280 dirfd = open(dir, O_DIRECTORY | O_RDONLY);
287 short_name = xasprintf("/proc/self/fd/%d/%s", dirfd, base);
291 if (strlen(short_name) <= MAX_UN_LEN) {
292 make_sockaddr_un__(short_name, un, un_len);
300 VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum "
301 "%d bytes (even shortened)", name, MAX_UN_LEN);
303 /* 'name' is too long and we have no workaround. */
304 VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum "
305 "%d bytes", name, MAX_UN_LEN);
310 make_sockaddr_un__(name, un, un_len);
315 /* Binds Unix domain socket 'fd' to a file with permissions 0700. */
317 bind_unix_socket(int fd, struct sockaddr *sun, socklen_t sun_len)
319 /* According to _Unix Network Programming_, umask should affect bind(). */
320 mode_t old_umask = umask(0077);
321 int error = bind(fd, sun, sun_len) ? errno : 0;
326 /* Creates a Unix domain socket in the given 'style' (either SOCK_DGRAM or
327 * SOCK_STREAM) that is bound to '*bind_path' (if 'bind_path' is non-null) and
328 * connected to '*connect_path' (if 'connect_path' is non-null). If 'nonblock'
329 * is true, the socket is made non-blocking. If 'passcred' is true, the socket
330 * is configured to receive SCM_CREDENTIALS control messages.
332 * Returns the socket's fd if successful, otherwise a negative errno value. */
334 make_unix_socket(int style, bool nonblock, bool passcred OVS_UNUSED,
335 const char *bind_path, const char *connect_path)
340 fd = socket(PF_UNIX, style, 0);
345 /* Set nonblocking mode right away, if we want it. This prevents blocking
346 * in connect(), if connect_path != NULL. (In turn, that's a corner case:
347 * it will only happen if style is SOCK_STREAM or SOCK_SEQPACKET, and only
348 * if a backlog of un-accepted connections has built up in the kernel.) */
350 int flags = fcntl(fd, F_GETFL, 0);
355 if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) == -1) {
362 struct sockaddr_un un;
366 if (unlink(bind_path) && errno != ENOENT) {
367 VLOG_WARN("unlinking \"%s\": %s\n", bind_path, strerror(errno));
369 fatal_signal_add_file_to_unlink(bind_path);
371 error = make_sockaddr_un(bind_path, &un, &un_len, &dirfd);
373 error = bind_unix_socket(fd, (struct sockaddr *) &un, un_len);
384 struct sockaddr_un un;
388 error = make_sockaddr_un(connect_path, &un, &un_len, &dirfd);
390 && connect(fd, (struct sockaddr*) &un, un_len)
391 && errno != EINPROGRESS) {
402 #ifdef SCM_CREDENTIALS
405 if (setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &enable, sizeof(enable))) {
415 if (error == EAGAIN) {
419 fatal_signal_remove_file_to_unlink(bind_path);
426 get_unix_name_len(socklen_t sun_len)
428 return (sun_len >= offsetof(struct sockaddr_un, sun_path)
429 ? sun_len - offsetof(struct sockaddr_un, sun_path)
434 guess_netmask(uint32_t ip)
437 return ((ip >> 31) == 0 ? htonl(0xff000000) /* Class A */
438 : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */
439 : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */
440 : htonl(0)); /* ??? */
443 /* Parses 'target', which should be a string in the format "<host>[:<port>]".
444 * <host> is required. If 'default_port' is nonzero then <port> is optional
445 * and defaults to 'default_port'.
447 * On success, returns true and stores the parsed remote address into '*sinp'.
448 * On failure, logs an error, stores zeros into '*sinp', and returns false. */
450 inet_parse_active(const char *target_, uint16_t default_port,
451 struct sockaddr_in *sinp)
453 char *target = xstrdup(target_);
454 char *save_ptr = NULL;
455 const char *host_name;
456 const char *port_string;
460 sinp->sin_family = AF_INET;
461 sinp->sin_port = htons(default_port);
464 host_name = strtok_r(target, ":", &save_ptr);
465 port_string = strtok_r(NULL, ":", &save_ptr);
467 VLOG_ERR("%s: bad peer name format", target_);
471 /* Look up IP, port. */
472 if (lookup_ip(host_name, &sinp->sin_addr)) {
475 if (port_string && atoi(port_string)) {
476 sinp->sin_port = htons(atoi(port_string));
477 } else if (!default_port) {
478 VLOG_ERR("%s: port number must be specified", target_);
486 memset(sinp, 0, sizeof *sinp);
492 /* Opens a non-blocking IPv4 socket of the specified 'style' and connects to
493 * 'target', which should be a string in the format "<host>[:<port>]". <host>
494 * is required. If 'default_port' is nonzero then <port> is optional and
495 * defaults to 'default_port'.
497 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
499 * On success, returns 0 (indicating connection complete) or EAGAIN (indicating
500 * connection in progress), in which case the new file descriptor is stored
501 * into '*fdp'. On failure, returns a positive errno value other than EAGAIN
502 * and stores -1 into '*fdp'.
504 * If 'sinp' is non-null, then on success the target address is stored into
507 inet_open_active(int style, const char *target, uint16_t default_port,
508 struct sockaddr_in *sinp, int *fdp)
510 struct sockaddr_in sin;
515 if (!inet_parse_active(target, default_port, &sin)) {
516 error = EAFNOSUPPORT;
520 /* Create non-blocking socket. */
521 fd = socket(AF_INET, style, 0);
523 VLOG_ERR("%s: socket: %s", target, strerror(errno));
527 error = set_nonblocking(fd);
533 error = connect(fd, (struct sockaddr *) &sin, sizeof sin) == 0 ? 0 : errno;
534 if (error == EINPROGRESS) {
536 } else if (error && error != EAGAIN) {
540 /* Success: error is 0 or EAGAIN. */
546 if (!error || error == EAGAIN) {
557 /* Opens a non-blocking IPv4 socket of the specified 'style', binds to
558 * 'target', and listens for incoming connections. 'target' should be a string
559 * in the format "[<port>][:<ip>]":
561 * - If 'default_port' is -1, then <port> is required. Otherwise, if
562 * <port> is omitted, then 'default_port' is used instead.
564 * - If <port> (or 'default_port', if used) is 0, then no port is bound
565 * and the TCP/IP stack will select a port.
567 * - If <ip> is omitted then the IP address is wildcarded.
569 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
571 * For TCP, the socket will have SO_REUSEADDR turned on.
573 * On success, returns a non-negative file descriptor. On failure, returns a
574 * negative errno value.
576 * If 'sinp' is non-null, then on success the bound address is stored into
579 inet_open_passive(int style, const char *target_, int default_port,
580 struct sockaddr_in *sinp)
582 char *target = xstrdup(target_);
583 char *string_ptr = target;
584 struct sockaddr_in sin;
585 const char *host_name;
586 const char *port_string;
587 int fd = 0, error, port;
588 unsigned int yes = 1;
590 /* Address defaults. */
591 memset(&sin, 0, sizeof sin);
592 sin.sin_family = AF_INET;
593 sin.sin_addr.s_addr = htonl(INADDR_ANY);
594 sin.sin_port = htons(default_port);
596 /* Parse optional port number. */
597 port_string = strsep(&string_ptr, ":");
598 if (port_string && str_to_int(port_string, 10, &port)) {
599 sin.sin_port = htons(port);
600 } else if (default_port < 0) {
601 VLOG_ERR("%s: port number must be specified", target_);
602 error = EAFNOSUPPORT;
606 /* Parse optional bind IP. */
607 host_name = strsep(&string_ptr, ":");
608 if (host_name && host_name[0]) {
609 error = lookup_ip(host_name, &sin.sin_addr);
615 /* Create non-blocking socket, set SO_REUSEADDR. */
616 fd = socket(AF_INET, style, 0);
619 VLOG_ERR("%s: socket: %s", target_, strerror(error));
622 error = set_nonblocking(fd);
626 if (style == SOCK_STREAM
627 && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) {
629 VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s", target_, strerror(error));
634 if (bind(fd, (struct sockaddr *) &sin, sizeof sin) < 0) {
636 VLOG_ERR("%s: bind: %s", target_, strerror(error));
641 if (listen(fd, 10) < 0) {
643 VLOG_ERR("%s: listen: %s", target_, strerror(error));
648 socklen_t sin_len = sizeof sin;
649 if (getsockname(fd, (struct sockaddr *) &sin, &sin_len) < 0){
651 VLOG_ERR("%s: getsockname: %s", target_, strerror(error));
654 if (sin.sin_family != AF_INET || sin_len != sizeof sin) {
655 VLOG_ERR("%s: getsockname: invalid socket name", target_);
668 return error ? -error : fd;
671 /* Returns a readable and writable fd for /dev/null, if successful, otherwise
672 * a negative errno value. The caller must not close the returned fd (because
673 * the same fd will be handed out to subsequent callers). */
677 static int null_fd = -1;
679 null_fd = open("/dev/null", O_RDWR);
682 VLOG_ERR("could not open /dev/null: %s", strerror(error));
690 read_fully(int fd, void *p_, size_t size, size_t *bytes_read)
696 ssize_t retval = read(fd, p, size);
698 *bytes_read += retval;
701 } else if (retval == 0) {
703 } else if (errno != EINTR) {
711 write_fully(int fd, const void *p_, size_t size, size_t *bytes_written)
713 const uint8_t *p = p_;
717 ssize_t retval = write(fd, p, size);
719 *bytes_written += retval;
722 } else if (retval == 0) {
723 VLOG_WARN("write returned 0");
725 } else if (errno != EINTR) {
732 /* Given file name 'file_name', fsyncs the directory in which it is contained.
733 * Returns 0 if successful, otherwise a positive errno value. */
735 fsync_parent_dir(const char *file_name)
741 dir = dir_name(file_name);
742 fd = open(dir, O_RDONLY);
745 if (errno == EINVAL || errno == EROFS) {
746 /* This directory does not support synchronization. Not
747 * really an error. */
750 VLOG_ERR("%s: fsync failed (%s)", dir, strerror(error));
756 VLOG_ERR("%s: open failed (%s)", dir, strerror(error));
763 /* Obtains the modification time of the file named 'file_name' to the greatest
764 * supported precision. If successful, stores the mtime in '*mtime' and
765 * returns 0. On error, returns a positive errno value and stores zeros in
768 get_mtime(const char *file_name, struct timespec *mtime)
772 if (!stat(file_name, &s)) {
773 mtime->tv_sec = s.st_mtime;
775 #if HAVE_STRUCT_STAT_ST_MTIM_TV_NSEC
776 mtime->tv_nsec = s.st_mtim.tv_nsec;
777 #elif HAVE_STRUCT_STAT_ST_MTIMENSEC
778 mtime->tv_nsec = s.st_mtimensec;
785 mtime->tv_sec = mtime->tv_nsec = 0;
794 VLOG_FATAL("failed to create pipe (%s)", strerror(errno));
799 getsockopt_int(int fd, int level, int optname, int *valuep)
801 socklen_t len = sizeof *valuep;
803 return (getsockopt(fd, level, optname, valuep, &len) ? errno
804 : len == sizeof *valuep ? 0
809 describe_sockaddr(struct ds *string, int fd,
810 int (*getaddr)(int, struct sockaddr *, socklen_t *))
812 struct sockaddr_storage ss;
813 socklen_t len = sizeof ss;
815 if (!getaddr(fd, (struct sockaddr *) &ss, &len)) {
816 if (ss.ss_family == AF_INET) {
817 struct sockaddr_in sin;
819 memcpy(&sin, &ss, sizeof sin);
820 ds_put_format(string, IP_FMT":%"PRIu16,
821 IP_ARGS(&sin.sin_addr.s_addr), ntohs(sin.sin_port));
822 } else if (ss.ss_family == AF_UNIX) {
823 struct sockaddr_un sun;
827 memcpy(&sun, &ss, sizeof sun);
828 maxlen = len - offsetof(struct sockaddr_un, sun_path);
829 null = memchr(sun.sun_path, '\0', maxlen);
830 ds_put_buffer(string, sun.sun_path,
831 null ? null - sun.sun_path : maxlen);
834 else if (ss.ss_family == AF_NETLINK) {
837 /* SO_PROTOCOL was introduced in 2.6.32. Support it regardless of the version
838 * of the Linux kernel headers in use at build time. */
840 #define SO_PROTOCOL 38
843 if (!getsockopt_int(fd, SOL_SOCKET, SO_PROTOCOL, &protocol)) {
846 ds_put_cstr(string, "NETLINK_ROUTE");
849 case NETLINK_GENERIC:
850 ds_put_cstr(string, "NETLINK_GENERIC");
854 ds_put_format(string, "AF_NETLINK family %d", protocol);
858 ds_put_cstr(string, "AF_NETLINK");
862 #if AF_PACKET && __linux__
863 else if (ss.ss_family == AF_PACKET) {
864 struct sockaddr_ll sll;
866 memcpy(&sll, &ss, sizeof sll);
867 ds_put_cstr(string, "AF_PACKET");
868 if (sll.sll_ifindex) {
871 if (if_indextoname(sll.sll_ifindex, name)) {
872 ds_put_format(string, "(%s)", name);
874 ds_put_format(string, "(ifindex=%d)", sll.sll_ifindex);
877 if (sll.sll_protocol) {
878 ds_put_format(string, "(protocol=0x%"PRIu16")",
879 ntohs(sll.sll_protocol));
883 else if (ss.ss_family == AF_UNSPEC) {
884 ds_put_cstr(string, "AF_UNSPEC");
886 ds_put_format(string, "AF_%d", (int) ss.ss_family);
894 put_fd_filename(struct ds *string, int fd)
900 linkname = xasprintf("/proc/self/fd/%d", fd);
901 n = readlink(linkname, buf, sizeof buf);
903 ds_put_char(string, ' ');
904 ds_put_buffer(string, buf, n);
905 if (n > sizeof buf) {
906 ds_put_cstr(string, "...");
913 /* Returns a malloc()'d string describing 'fd', for use in logging. */
922 ds_put_format(&string, "fstat failed (%s)", strerror(errno));
923 } else if (S_ISSOCK(s.st_mode)) {
924 describe_sockaddr(&string, fd, getsockname);
925 ds_put_cstr(&string, "<->");
926 describe_sockaddr(&string, fd, getpeername);
928 ds_put_cstr(&string, (isatty(fd) ? "tty"
929 : S_ISDIR(s.st_mode) ? "directory"
930 : S_ISCHR(s.st_mode) ? "character device"
931 : S_ISBLK(s.st_mode) ? "block device"
932 : S_ISREG(s.st_mode) ? "file"
933 : S_ISFIFO(s.st_mode) ? "FIFO"
934 : S_ISLNK(s.st_mode) ? "symbolic link"
937 put_fd_filename(&string, fd);
940 return ds_steal_cstr(&string);