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>
28 #include <sys/resource.h>
29 #include <sys/socket.h>
33 #include "fatal-signal.h"
37 VLOG_DEFINE_THIS_MODULE(socket_util);
39 /* #ifdefs make it a pain to maintain code: you have to try to build both ways.
40 * Thus, this file compiles all of the code regardless of the target, by
41 * writing "if (LINUX)" instead of "#ifdef __linux__". */
52 /* Sets 'fd' to non-blocking mode. Returns 0 if successful, otherwise a
53 * positive errno value. */
55 set_nonblocking(int fd)
57 int flags = fcntl(fd, F_GETFL, 0);
59 if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) != -1) {
62 VLOG_ERR("fcntl(F_SETFL) failed: %s", strerror(errno));
66 VLOG_ERR("fcntl(F_GETFL) failed: %s", strerror(errno));
72 rlim_is_finite(rlim_t limit)
74 if (limit == RLIM_INFINITY) {
78 #ifdef RLIM_SAVED_CUR /* FreeBSD 8.0 lacks RLIM_SAVED_CUR. */
79 if (limit == RLIM_SAVED_CUR) {
84 #ifdef RLIM_SAVED_MAX /* FreeBSD 8.0 lacks RLIM_SAVED_MAX. */
85 if (limit == RLIM_SAVED_MAX) {
93 /* Returns the maximum valid FD value, plus 1. */
97 static int max_fds = -1;
100 if (!getrlimit(RLIMIT_NOFILE, &r) && rlim_is_finite(r.rlim_cur)) {
101 max_fds = r.rlim_cur;
103 VLOG_WARN("failed to obtain fd limit, defaulting to 1024");
110 /* Translates 'host_name', which must be a string representation of an IP
111 * address, into a numeric IP address in '*addr'. Returns 0 if successful,
112 * otherwise a positive errno value. */
114 lookup_ip(const char *host_name, struct in_addr *addr)
116 if (!inet_aton(host_name, addr)) {
117 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
118 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name);
124 /* Translates 'host_name', which must be a string representation of an IPv6
125 * address, into a numeric IPv6 address in '*addr'. Returns 0 if successful,
126 * otherwise a positive errno value. */
128 lookup_ipv6(const char *host_name, struct in6_addr *addr)
130 if (inet_pton(AF_INET6, host_name, addr) != 1) {
131 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
132 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IPv6 address", host_name);
138 /* Returns the error condition associated with socket 'fd' and resets the
139 * socket's error status. */
141 get_socket_error(int fd)
144 socklen_t len = sizeof(error);
145 if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &error, &len) < 0) {
146 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
148 VLOG_ERR_RL(&rl, "getsockopt(SO_ERROR): %s", strerror(error));
154 check_connection_completion(int fd)
160 pfd.events = POLLOUT;
162 retval = poll(&pfd, 1, 0);
163 } while (retval < 0 && errno == EINTR);
165 return get_socket_error(fd);
166 } else if (retval < 0) {
167 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
168 VLOG_ERR_RL(&rl, "poll: %s", strerror(errno));
175 /* Drain all the data currently in the receive queue of a datagram socket (and
176 * possibly additional data). There is no way to know how many packets are in
177 * the receive queue, but we do know that the total number of bytes queued does
178 * not exceed the receive buffer size, so we pull packets until none are left
179 * or we've read that many bytes. */
183 socklen_t rcvbuf_len;
186 rcvbuf_len = sizeof rcvbuf;
187 if (getsockopt(fd, SOL_SOCKET, SO_RCVBUF, &rcvbuf, &rcvbuf_len) < 0) {
188 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
189 VLOG_ERR_RL(&rl, "getsockopt(SO_RCVBUF) failed: %s", strerror(errno));
193 /* In Linux, specifying MSG_TRUNC in the flags argument causes the
194 * datagram length to be returned, even if that is longer than the
195 * buffer provided. Thus, we can use a 1-byte buffer to discard the
196 * incoming datagram and still be able to account how many bytes were
197 * removed from the receive buffer.
199 * On other Unix-like OSes, MSG_TRUNC has no effect in the flags
201 char buffer[LINUX ? 1 : 2048];
202 ssize_t n_bytes = recv(fd, buffer, sizeof buffer,
203 MSG_TRUNC | MSG_DONTWAIT);
204 if (n_bytes <= 0 || n_bytes >= rcvbuf) {
212 /* Reads and discards up to 'n' datagrams from 'fd', stopping as soon as no
213 * more data can be immediately read. ('fd' should therefore be in
214 * non-blocking mode.)*/
216 drain_fd(int fd, size_t n_packets)
218 for (; n_packets > 0; n_packets--) {
219 /* 'buffer' only needs to be 1 byte long in most circumstances. This
220 * size is defensive against the possibility that we someday want to
221 * use a Linux tap device without TUN_NO_PI, in which case a buffer
222 * smaller than sizeof(struct tun_pi) will give EINVAL on read. */
224 if (read(fd, buffer, sizeof buffer) <= 0) {
230 /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in
231 * '*un_len' the size of the sockaddr_un. */
233 make_sockaddr_un__(const char *name, struct sockaddr_un *un, socklen_t *un_len)
235 un->sun_family = AF_UNIX;
236 ovs_strzcpy(un->sun_path, name, sizeof un->sun_path);
237 *un_len = (offsetof(struct sockaddr_un, sun_path)
238 + strlen (un->sun_path) + 1);
241 /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in
242 * '*un_len' the size of the sockaddr_un.
244 * Returns 0 on success, otherwise a positive errno value. On success,
245 * '*dirfdp' is either -1 or a nonnegative file descriptor that the caller
246 * should close after using '*un' to bind or connect. On failure, '*dirfdp' is
249 make_sockaddr_un(const char *name, struct sockaddr_un *un, socklen_t *un_len,
252 enum { MAX_UN_LEN = sizeof un->sun_path - 1 };
255 if (strlen(name) > MAX_UN_LEN) {
256 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
259 /* 'name' is too long to fit in a sockaddr_un, but we have a
260 * workaround for that on Linux: shorten it by opening a file
261 * descriptor for the directory part of the name and indirecting
262 * through /proc/self/fd/<dirfd>/<basename>. */
267 dir = dir_name(name);
268 base = base_name(name);
270 dirfd = open(dir, O_DIRECTORY | O_RDONLY);
277 short_name = xasprintf("/proc/self/fd/%d/%s", dirfd, base);
281 if (strlen(short_name) <= MAX_UN_LEN) {
282 make_sockaddr_un__(short_name, un, un_len);
290 VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum "
291 "%d bytes (even shortened)", name, MAX_UN_LEN);
293 /* 'name' is too long and we have no workaround. */
294 VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum "
295 "%d bytes", name, MAX_UN_LEN);
300 make_sockaddr_un__(name, un, un_len);
305 /* Creates a Unix domain socket in the given 'style' (either SOCK_DGRAM or
306 * SOCK_STREAM) that is bound to '*bind_path' (if 'bind_path' is non-null) and
307 * connected to '*connect_path' (if 'connect_path' is non-null). If 'nonblock'
308 * is true, the socket is made non-blocking. If 'passcred' is true, the socket
309 * is configured to receive SCM_CREDENTIALS control messages.
311 * Returns the socket's fd if successful, otherwise a negative errno value. */
313 make_unix_socket(int style, bool nonblock, bool passcred OVS_UNUSED,
314 const char *bind_path, const char *connect_path)
319 fd = socket(PF_UNIX, style, 0);
324 /* Set nonblocking mode right away, if we want it. This prevents blocking
325 * in connect(), if connect_path != NULL. (In turn, that's a corner case:
326 * it will only happen if style is SOCK_STREAM or SOCK_SEQPACKET, and only
327 * if a backlog of un-accepted connections has built up in the kernel.) */
329 int flags = fcntl(fd, F_GETFL, 0);
334 if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) == -1) {
341 struct sockaddr_un un;
345 if (unlink(bind_path) && errno != ENOENT) {
346 VLOG_WARN("unlinking \"%s\": %s\n", bind_path, strerror(errno));
348 fatal_signal_add_file_to_unlink(bind_path);
350 error = make_sockaddr_un(bind_path, &un, &un_len, &dirfd);
351 if (!error && (bind(fd, (struct sockaddr*) &un, un_len)
352 || fchmod(fd, S_IRWXU))) {
364 struct sockaddr_un un;
368 error = make_sockaddr_un(connect_path, &un, &un_len, &dirfd);
370 && connect(fd, (struct sockaddr*) &un, un_len)
371 && errno != EINPROGRESS) {
382 #ifdef SCM_CREDENTIALS
385 if (setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &enable, sizeof(enable))) {
395 if (error == EAGAIN) {
399 fatal_signal_remove_file_to_unlink(bind_path);
406 get_unix_name_len(socklen_t sun_len)
408 return (sun_len >= offsetof(struct sockaddr_un, sun_path)
409 ? sun_len - offsetof(struct sockaddr_un, sun_path)
414 guess_netmask(uint32_t ip)
417 return ((ip >> 31) == 0 ? htonl(0xff000000) /* Class A */
418 : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */
419 : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */
420 : htonl(0)); /* ??? */
423 /* Parses 'target', which should be a string in the format "<host>[:<port>]".
424 * <host> is required. If 'default_port' is nonzero then <port> is optional
425 * and defaults to 'default_port'.
427 * On success, returns true and stores the parsed remote address into '*sinp'.
428 * On failure, logs an error, stores zeros into '*sinp', and returns false. */
430 inet_parse_active(const char *target_, uint16_t default_port,
431 struct sockaddr_in *sinp)
433 char *target = xstrdup(target_);
434 char *save_ptr = NULL;
435 const char *host_name;
436 const char *port_string;
440 sinp->sin_family = AF_INET;
441 sinp->sin_port = htons(default_port);
444 host_name = strtok_r(target, ":", &save_ptr);
445 port_string = strtok_r(NULL, ":", &save_ptr);
447 VLOG_ERR("%s: bad peer name format", target_);
451 /* Look up IP, port. */
452 if (lookup_ip(host_name, &sinp->sin_addr)) {
455 if (port_string && atoi(port_string)) {
456 sinp->sin_port = htons(atoi(port_string));
457 } else if (!default_port) {
458 VLOG_ERR("%s: port number must be specified", target_);
466 memset(sinp, 0, sizeof *sinp);
472 /* Opens a non-blocking IPv4 socket of the specified 'style' and connects to
473 * 'target', which should be a string in the format "<host>[:<port>]". <host>
474 * is required. If 'default_port' is nonzero then <port> is optional and
475 * defaults to 'default_port'.
477 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
479 * On success, returns 0 (indicating connection complete) or EAGAIN (indicating
480 * connection in progress), in which case the new file descriptor is stored
481 * into '*fdp'. On failure, returns a positive errno value other than EAGAIN
482 * and stores -1 into '*fdp'.
484 * If 'sinp' is non-null, then on success the target address is stored into
487 inet_open_active(int style, const char *target, uint16_t default_port,
488 struct sockaddr_in *sinp, int *fdp)
490 struct sockaddr_in sin;
495 if (!inet_parse_active(target, default_port, &sin)) {
496 error = EAFNOSUPPORT;
500 /* Create non-blocking socket. */
501 fd = socket(AF_INET, style, 0);
503 VLOG_ERR("%s: socket: %s", target, strerror(errno));
507 error = set_nonblocking(fd);
513 error = connect(fd, (struct sockaddr *) &sin, sizeof sin) == 0 ? 0 : errno;
514 if (error == EINPROGRESS) {
516 } else if (error && error != EAGAIN) {
520 /* Success: error is 0 or EAGAIN. */
526 if (!error || error == EAGAIN) {
537 /* Opens a non-blocking IPv4 socket of the specified 'style', binds to
538 * 'target', and listens for incoming connections. 'target' should be a string
539 * in the format "[<port>][:<ip>]":
541 * - If 'default_port' is -1, then <port> is required. Otherwise, if
542 * <port> is omitted, then 'default_port' is used instead.
544 * - If <port> (or 'default_port', if used) is 0, then no port is bound
545 * and the TCP/IP stack will select a port.
547 * - If <ip> is omitted then the IP address is wildcarded.
549 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
551 * For TCP, the socket will have SO_REUSEADDR turned on.
553 * On success, returns a non-negative file descriptor. On failure, returns a
554 * negative errno value.
556 * If 'sinp' is non-null, then on success the bound address is stored into
559 inet_open_passive(int style, const char *target_, int default_port,
560 struct sockaddr_in *sinp)
562 char *target = xstrdup(target_);
563 char *string_ptr = target;
564 struct sockaddr_in sin;
565 const char *host_name;
566 const char *port_string;
567 int fd = 0, error, port;
568 unsigned int yes = 1;
570 /* Address defaults. */
571 memset(&sin, 0, sizeof sin);
572 sin.sin_family = AF_INET;
573 sin.sin_addr.s_addr = htonl(INADDR_ANY);
574 sin.sin_port = htons(default_port);
576 /* Parse optional port number. */
577 port_string = strsep(&string_ptr, ":");
578 if (port_string && str_to_int(port_string, 10, &port)) {
579 sin.sin_port = htons(port);
580 } else if (default_port < 0) {
581 VLOG_ERR("%s: port number must be specified", target_);
582 error = EAFNOSUPPORT;
586 /* Parse optional bind IP. */
587 host_name = strsep(&string_ptr, ":");
588 if (host_name && host_name[0]) {
589 error = lookup_ip(host_name, &sin.sin_addr);
595 /* Create non-blocking socket, set SO_REUSEADDR. */
596 fd = socket(AF_INET, style, 0);
599 VLOG_ERR("%s: socket: %s", target_, strerror(error));
602 error = set_nonblocking(fd);
606 if (style == SOCK_STREAM
607 && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) {
609 VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s", target_, strerror(error));
614 if (bind(fd, (struct sockaddr *) &sin, sizeof sin) < 0) {
616 VLOG_ERR("%s: bind: %s", target_, strerror(error));
621 if (listen(fd, 10) < 0) {
623 VLOG_ERR("%s: listen: %s", target_, strerror(error));
628 socklen_t sin_len = sizeof sin;
629 if (getsockname(fd, (struct sockaddr *) &sin, &sin_len) < 0){
631 VLOG_ERR("%s: getsockname: %s", target_, strerror(error));
634 if (sin.sin_family != AF_INET || sin_len != sizeof sin) {
635 VLOG_ERR("%s: getsockname: invalid socket name", target_);
648 return error ? -error : fd;
651 /* Returns a readable and writable fd for /dev/null, if successful, otherwise
652 * a negative errno value. The caller must not close the returned fd (because
653 * the same fd will be handed out to subsequent callers). */
657 static int null_fd = -1;
659 null_fd = open("/dev/null", O_RDWR);
662 VLOG_ERR("could not open /dev/null: %s", strerror(error));
670 read_fully(int fd, void *p_, size_t size, size_t *bytes_read)
676 ssize_t retval = read(fd, p, size);
678 *bytes_read += retval;
681 } else if (retval == 0) {
683 } else if (errno != EINTR) {
691 write_fully(int fd, const void *p_, size_t size, size_t *bytes_written)
693 const uint8_t *p = p_;
697 ssize_t retval = write(fd, p, size);
699 *bytes_written += retval;
702 } else if (retval == 0) {
703 VLOG_WARN("write returned 0");
705 } else if (errno != EINTR) {
712 /* Given file name 'file_name', fsyncs the directory in which it is contained.
713 * Returns 0 if successful, otherwise a positive errno value. */
715 fsync_parent_dir(const char *file_name)
721 dir = dir_name(file_name);
722 fd = open(dir, O_RDONLY);
725 if (errno == EINVAL || errno == EROFS) {
726 /* This directory does not support synchronization. Not
727 * really an error. */
730 VLOG_ERR("%s: fsync failed (%s)", dir, strerror(error));
736 VLOG_ERR("%s: open failed (%s)", dir, strerror(error));
743 /* Obtains the modification time of the file named 'file_name' to the greatest
744 * supported precision. If successful, stores the mtime in '*mtime' and
745 * returns 0. On error, returns a positive errno value and stores zeros in
748 get_mtime(const char *file_name, struct timespec *mtime)
752 if (!stat(file_name, &s)) {
753 mtime->tv_sec = s.st_mtime;
755 #if HAVE_STRUCT_STAT_ST_MTIM_TV_NSEC
756 mtime->tv_nsec = s.st_mtim.tv_nsec;
757 #elif HAVE_STRUCT_STAT_ST_MTIMENSEC
758 mtime->tv_nsec = s.st_mtimensec;
765 mtime->tv_sec = mtime->tv_nsec = 0;