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 /* Binds Unix domain socket 'fd' to a file with permissions 0700. */
307 bind_unix_socket(int fd, struct sockaddr *sun, socklen_t sun_len)
309 /* According to _Unix Network Programming_, umask should affect bind(). */
310 mode_t old_umask = umask(0077);
311 int error = bind(fd, sun, sun_len) ? errno : 0;
316 /* Creates a Unix domain socket in the given 'style' (either SOCK_DGRAM or
317 * SOCK_STREAM) that is bound to '*bind_path' (if 'bind_path' is non-null) and
318 * connected to '*connect_path' (if 'connect_path' is non-null). If 'nonblock'
319 * is true, the socket is made non-blocking. If 'passcred' is true, the socket
320 * is configured to receive SCM_CREDENTIALS control messages.
322 * Returns the socket's fd if successful, otherwise a negative errno value. */
324 make_unix_socket(int style, bool nonblock, bool passcred OVS_UNUSED,
325 const char *bind_path, const char *connect_path)
330 fd = socket(PF_UNIX, style, 0);
335 /* Set nonblocking mode right away, if we want it. This prevents blocking
336 * in connect(), if connect_path != NULL. (In turn, that's a corner case:
337 * it will only happen if style is SOCK_STREAM or SOCK_SEQPACKET, and only
338 * if a backlog of un-accepted connections has built up in the kernel.) */
340 int flags = fcntl(fd, F_GETFL, 0);
345 if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) == -1) {
352 struct sockaddr_un un;
356 if (unlink(bind_path) && errno != ENOENT) {
357 VLOG_WARN("unlinking \"%s\": %s\n", bind_path, strerror(errno));
359 fatal_signal_add_file_to_unlink(bind_path);
361 error = make_sockaddr_un(bind_path, &un, &un_len, &dirfd);
363 error = bind_unix_socket(fd, (struct sockaddr *) &un, un_len);
374 struct sockaddr_un un;
378 error = make_sockaddr_un(connect_path, &un, &un_len, &dirfd);
380 && connect(fd, (struct sockaddr*) &un, un_len)
381 && errno != EINPROGRESS) {
392 #ifdef SCM_CREDENTIALS
395 if (setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &enable, sizeof(enable))) {
405 if (error == EAGAIN) {
409 fatal_signal_remove_file_to_unlink(bind_path);
416 get_unix_name_len(socklen_t sun_len)
418 return (sun_len >= offsetof(struct sockaddr_un, sun_path)
419 ? sun_len - offsetof(struct sockaddr_un, sun_path)
424 guess_netmask(ovs_be32 ip_)
426 uint32_t ip = ntohl(ip_);
427 return ((ip >> 31) == 0 ? htonl(0xff000000) /* Class A */
428 : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */
429 : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */
430 : htonl(0)); /* ??? */
433 /* Parses 'target', which should be a string in the format "<host>[:<port>]".
434 * <host> is required. If 'default_port' is nonzero then <port> is optional
435 * and defaults to 'default_port'.
437 * On success, returns true and stores the parsed remote address into '*sinp'.
438 * On failure, logs an error, stores zeros into '*sinp', and returns false. */
440 inet_parse_active(const char *target_, uint16_t default_port,
441 struct sockaddr_in *sinp)
443 char *target = xstrdup(target_);
444 char *save_ptr = NULL;
445 const char *host_name;
446 const char *port_string;
450 sinp->sin_family = AF_INET;
451 sinp->sin_port = htons(default_port);
454 host_name = strtok_r(target, ":", &save_ptr);
455 port_string = strtok_r(NULL, ":", &save_ptr);
457 VLOG_ERR("%s: bad peer name format", target_);
461 /* Look up IP, port. */
462 if (lookup_ip(host_name, &sinp->sin_addr)) {
465 if (port_string && atoi(port_string)) {
466 sinp->sin_port = htons(atoi(port_string));
467 } else if (!default_port) {
468 VLOG_ERR("%s: port number must be specified", target_);
476 memset(sinp, 0, sizeof *sinp);
482 /* Opens a non-blocking IPv4 socket of the specified 'style' and connects to
483 * 'target', which should be a string in the format "<host>[:<port>]". <host>
484 * is required. If 'default_port' is nonzero then <port> is optional and
485 * defaults to 'default_port'.
487 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
489 * On success, returns 0 (indicating connection complete) or EAGAIN (indicating
490 * connection in progress), in which case the new file descriptor is stored
491 * into '*fdp'. On failure, returns a positive errno value other than EAGAIN
492 * and stores -1 into '*fdp'.
494 * If 'sinp' is non-null, then on success the target address is stored into
497 inet_open_active(int style, const char *target, uint16_t default_port,
498 struct sockaddr_in *sinp, int *fdp)
500 struct sockaddr_in sin;
505 if (!inet_parse_active(target, default_port, &sin)) {
506 error = EAFNOSUPPORT;
510 /* Create non-blocking socket. */
511 fd = socket(AF_INET, style, 0);
513 VLOG_ERR("%s: socket: %s", target, strerror(errno));
517 error = set_nonblocking(fd);
523 error = connect(fd, (struct sockaddr *) &sin, sizeof sin) == 0 ? 0 : errno;
524 if (error == EINPROGRESS) {
526 } else if (error && error != EAGAIN) {
530 /* Success: error is 0 or EAGAIN. */
536 if (!error || error == EAGAIN) {
547 /* Opens a non-blocking IPv4 socket of the specified 'style', binds to
548 * 'target', and listens for incoming connections. 'target' should be a string
549 * in the format "[<port>][:<ip>]":
551 * - If 'default_port' is -1, then <port> is required. Otherwise, if
552 * <port> is omitted, then 'default_port' is used instead.
554 * - If <port> (or 'default_port', if used) is 0, then no port is bound
555 * and the TCP/IP stack will select a port.
557 * - If <ip> is omitted then the IP address is wildcarded.
559 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
561 * For TCP, the socket will have SO_REUSEADDR turned on.
563 * On success, returns a non-negative file descriptor. On failure, returns a
564 * negative errno value.
566 * If 'sinp' is non-null, then on success the bound address is stored into
569 inet_open_passive(int style, const char *target_, int default_port,
570 struct sockaddr_in *sinp)
572 char *target = xstrdup(target_);
573 char *string_ptr = target;
574 struct sockaddr_in sin;
575 const char *host_name;
576 const char *port_string;
577 int fd = 0, error, port;
578 unsigned int yes = 1;
580 /* Address defaults. */
581 memset(&sin, 0, sizeof sin);
582 sin.sin_family = AF_INET;
583 sin.sin_addr.s_addr = htonl(INADDR_ANY);
584 sin.sin_port = htons(default_port);
586 /* Parse optional port number. */
587 port_string = strsep(&string_ptr, ":");
588 if (port_string && str_to_int(port_string, 10, &port)) {
589 sin.sin_port = htons(port);
590 } else if (default_port < 0) {
591 VLOG_ERR("%s: port number must be specified", target_);
592 error = EAFNOSUPPORT;
596 /* Parse optional bind IP. */
597 host_name = strsep(&string_ptr, ":");
598 if (host_name && host_name[0]) {
599 error = lookup_ip(host_name, &sin.sin_addr);
605 /* Create non-blocking socket, set SO_REUSEADDR. */
606 fd = socket(AF_INET, style, 0);
609 VLOG_ERR("%s: socket: %s", target_, strerror(error));
612 error = set_nonblocking(fd);
616 if (style == SOCK_STREAM
617 && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) {
619 VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s", target_, strerror(error));
624 if (bind(fd, (struct sockaddr *) &sin, sizeof sin) < 0) {
626 VLOG_ERR("%s: bind: %s", target_, strerror(error));
631 if (listen(fd, 10) < 0) {
633 VLOG_ERR("%s: listen: %s", target_, strerror(error));
638 socklen_t sin_len = sizeof sin;
639 if (getsockname(fd, (struct sockaddr *) &sin, &sin_len) < 0){
641 VLOG_ERR("%s: getsockname: %s", target_, strerror(error));
644 if (sin.sin_family != AF_INET || sin_len != sizeof sin) {
645 VLOG_ERR("%s: getsockname: invalid socket name", target_);
658 return error ? -error : fd;
661 /* Returns a readable and writable fd for /dev/null, if successful, otherwise
662 * a negative errno value. The caller must not close the returned fd (because
663 * the same fd will be handed out to subsequent callers). */
667 static int null_fd = -1;
669 null_fd = open("/dev/null", O_RDWR);
672 VLOG_ERR("could not open /dev/null: %s", strerror(error));
680 read_fully(int fd, void *p_, size_t size, size_t *bytes_read)
686 ssize_t retval = read(fd, p, size);
688 *bytes_read += retval;
691 } else if (retval == 0) {
693 } else if (errno != EINTR) {
701 write_fully(int fd, const void *p_, size_t size, size_t *bytes_written)
703 const uint8_t *p = p_;
707 ssize_t retval = write(fd, p, size);
709 *bytes_written += retval;
712 } else if (retval == 0) {
713 VLOG_WARN("write returned 0");
715 } else if (errno != EINTR) {
722 /* Given file name 'file_name', fsyncs the directory in which it is contained.
723 * Returns 0 if successful, otherwise a positive errno value. */
725 fsync_parent_dir(const char *file_name)
731 dir = dir_name(file_name);
732 fd = open(dir, O_RDONLY);
735 if (errno == EINVAL || errno == EROFS) {
736 /* This directory does not support synchronization. Not
737 * really an error. */
740 VLOG_ERR("%s: fsync failed (%s)", dir, strerror(error));
746 VLOG_ERR("%s: open failed (%s)", dir, strerror(error));
753 /* Obtains the modification time of the file named 'file_name' to the greatest
754 * supported precision. If successful, stores the mtime in '*mtime' and
755 * returns 0. On error, returns a positive errno value and stores zeros in
758 get_mtime(const char *file_name, struct timespec *mtime)
762 if (!stat(file_name, &s)) {
763 mtime->tv_sec = s.st_mtime;
765 #if HAVE_STRUCT_STAT_ST_MTIM_TV_NSEC
766 mtime->tv_nsec = s.st_mtim.tv_nsec;
767 #elif HAVE_STRUCT_STAT_ST_MTIMENSEC
768 mtime->tv_nsec = s.st_mtimensec;
775 mtime->tv_sec = mtime->tv_nsec = 0;
784 VLOG_FATAL("failed to create pipe (%s)", strerror(errno));