2 * Copyright (c) 2008, 2009, 2010 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 /* Sets 'fd' to non-blocking mode. Returns 0 if successful, otherwise a
40 * positive errno value. */
42 set_nonblocking(int fd)
44 int flags = fcntl(fd, F_GETFL, 0);
46 if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) != -1) {
49 VLOG_ERR("fcntl(F_SETFL) failed: %s", strerror(errno));
53 VLOG_ERR("fcntl(F_GETFL) failed: %s", strerror(errno));
59 rlim_is_finite(rlim_t limit)
61 if (limit == RLIM_INFINITY) {
65 #ifdef RLIM_SAVED_CUR /* FreeBSD 8.0 lacks RLIM_SAVED_CUR. */
66 if (limit == RLIM_SAVED_CUR) {
71 #ifdef RLIM_SAVED_MAX /* FreeBSD 8.0 lacks RLIM_SAVED_MAX. */
72 if (limit == RLIM_SAVED_MAX) {
80 /* Returns the maximum valid FD value, plus 1. */
84 static int max_fds = -1;
87 if (!getrlimit(RLIMIT_NOFILE, &r) && rlim_is_finite(r.rlim_cur)) {
90 VLOG_WARN("failed to obtain fd limit, defaulting to 1024");
97 /* Translates 'host_name', which must be a string representation of an IP
98 * address, into a numeric IP address in '*addr'. Returns 0 if successful,
99 * otherwise a positive errno value. */
101 lookup_ip(const char *host_name, struct in_addr *addr)
103 if (!inet_aton(host_name, addr)) {
104 struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
105 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name);
111 /* Returns the error condition associated with socket 'fd' and resets the
112 * socket's error status. */
114 get_socket_error(int fd)
117 socklen_t len = sizeof(error);
118 if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &error, &len) < 0) {
119 struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
121 VLOG_ERR_RL(&rl, "getsockopt(SO_ERROR): %s", strerror(error));
127 check_connection_completion(int fd)
133 pfd.events = POLLOUT;
135 retval = poll(&pfd, 1, 0);
136 } while (retval < 0 && errno == EINTR);
138 return get_socket_error(fd);
139 } else if (retval < 0) {
140 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
141 VLOG_ERR_RL(&rl, "poll: %s", strerror(errno));
148 /* Drain all the data currently in the receive queue of a datagram socket (and
149 * possibly additional data). There is no way to know how many packets are in
150 * the receive queue, but we do know that the total number of bytes queued does
151 * not exceed the receive buffer size, so we pull packets until none are left
152 * or we've read that many bytes. */
156 socklen_t rcvbuf_len;
159 rcvbuf_len = sizeof rcvbuf;
160 if (getsockopt(fd, SOL_SOCKET, SO_RCVBUF, &rcvbuf, &rcvbuf_len) < 0) {
161 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
162 VLOG_ERR_RL(&rl, "getsockopt(SO_RCVBUF) failed: %s", strerror(errno));
166 /* In Linux, specifying MSG_TRUNC in the flags argument causes the
167 * datagram length to be returned, even if that is longer than the
168 * buffer provided. Thus, we can use a 1-byte buffer to discard the
169 * incoming datagram and still be able to account how many bytes were
170 * removed from the receive buffer.
172 * On other Unix-like OSes, MSG_TRUNC has no effect in the flags
175 #define BUFFER_SIZE 1
177 #define BUFFER_SIZE 2048
179 char buffer[BUFFER_SIZE];
180 ssize_t n_bytes = recv(fd, buffer, sizeof buffer,
181 MSG_TRUNC | MSG_DONTWAIT);
182 if (n_bytes <= 0 || n_bytes >= rcvbuf) {
190 /* Reads and discards up to 'n' datagrams from 'fd', stopping as soon as no
191 * more data can be immediately read. ('fd' should therefore be in
192 * non-blocking mode.)*/
194 drain_fd(int fd, size_t n_packets)
196 for (; n_packets > 0; n_packets--) {
197 /* 'buffer' only needs to be 1 byte long in most circumstances. This
198 * size is defensive against the possibility that we someday want to
199 * use a Linux tap device without TUN_NO_PI, in which case a buffer
200 * smaller than sizeof(struct tun_pi) will give EINVAL on read. */
202 if (read(fd, buffer, sizeof buffer) <= 0) {
208 /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in
209 * '*un_len' the size of the sockaddr_un. */
211 make_sockaddr_un(const char *name, struct sockaddr_un* un, socklen_t *un_len)
213 un->sun_family = AF_UNIX;
214 strncpy(un->sun_path, name, sizeof un->sun_path);
215 un->sun_path[sizeof un->sun_path - 1] = '\0';
216 *un_len = (offsetof(struct sockaddr_un, sun_path)
217 + strlen (un->sun_path) + 1);
220 /* Creates a Unix domain socket in the given 'style' (either SOCK_DGRAM or
221 * SOCK_STREAM) that is bound to '*bind_path' (if 'bind_path' is non-null) and
222 * connected to '*connect_path' (if 'connect_path' is non-null). If 'nonblock'
223 * is true, the socket is made non-blocking. If 'passcred' is true, the socket
224 * is configured to receive SCM_CREDENTIALS control messages.
226 * Returns the socket's fd if successful, otherwise a negative errno value. */
228 make_unix_socket(int style, bool nonblock, bool passcred OVS_UNUSED,
229 const char *bind_path, const char *connect_path)
234 fd = socket(PF_UNIX, style, 0);
239 /* Set nonblocking mode right away, if we want it. This prevents blocking
240 * in connect(), if connect_path != NULL. (In turn, that's a corner case:
241 * it will only happen if style is SOCK_STREAM or SOCK_SEQPACKET, and only
242 * if a backlog of un-accepted connections has built up in the kernel.) */
244 int flags = fcntl(fd, F_GETFL, 0);
248 if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) == -1) {
254 struct sockaddr_un un;
256 make_sockaddr_un(bind_path, &un, &un_len);
257 if (unlink(un.sun_path) && errno != ENOENT) {
258 VLOG_WARN("unlinking \"%s\": %s\n", un.sun_path, strerror(errno));
260 fatal_signal_add_file_to_unlink(bind_path);
261 if (bind(fd, (struct sockaddr*) &un, un_len)
262 || fchmod(fd, S_IRWXU)) {
268 struct sockaddr_un un;
270 make_sockaddr_un(connect_path, &un, &un_len);
271 if (connect(fd, (struct sockaddr*) &un, un_len)
272 && errno != EINPROGRESS) {
277 #ifdef SCM_CREDENTIALS
280 if (setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &enable, sizeof(enable))) {
289 error = errno == EAGAIN ? EPROTO : errno;
291 fatal_signal_remove_file_to_unlink(bind_path);
298 get_unix_name_len(socklen_t sun_len)
300 return (sun_len >= offsetof(struct sockaddr_un, sun_path)
301 ? sun_len - offsetof(struct sockaddr_un, sun_path)
306 guess_netmask(uint32_t ip)
309 return ((ip >> 31) == 0 ? htonl(0xff000000) /* Class A */
310 : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */
311 : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */
312 : htonl(0)); /* ??? */
315 /* Parses 'target', which should be a string in the format "<host>[:<port>]".
316 * <host> is required. If 'default_port' is nonzero then <port> is optional
317 * and defaults to 'default_port'.
319 * On success, returns true and stores the parsed remote address into '*sinp'.
320 * On failure, logs an error, stores zeros into '*sinp', and returns false. */
322 inet_parse_active(const char *target_, uint16_t default_port,
323 struct sockaddr_in *sinp)
325 char *target = xstrdup(target_);
326 char *save_ptr = NULL;
327 const char *host_name;
328 const char *port_string;
332 sinp->sin_family = AF_INET;
333 sinp->sin_port = htons(default_port);
336 host_name = strtok_r(target, ":", &save_ptr);
337 port_string = strtok_r(NULL, ":", &save_ptr);
339 VLOG_ERR("%s: bad peer name format", target_);
343 /* Look up IP, port. */
344 if (lookup_ip(host_name, &sinp->sin_addr)) {
347 if (port_string && atoi(port_string)) {
348 sinp->sin_port = htons(atoi(port_string));
349 } else if (!default_port) {
350 VLOG_ERR("%s: port number must be specified", target_);
358 memset(sinp, 0, sizeof *sinp);
364 /* Opens a non-blocking IPv4 socket of the specified 'style' and connects to
365 * 'target', which should be a string in the format "<host>[:<port>]". <host>
366 * is required. If 'default_port' is nonzero then <port> is optional and
367 * defaults to 'default_port'.
369 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
371 * On success, returns 0 (indicating connection complete) or EAGAIN (indicating
372 * connection in progress), in which case the new file descriptor is stored
373 * into '*fdp'. On failure, returns a positive errno value other than EAGAIN
374 * and stores -1 into '*fdp'.
376 * If 'sinp' is non-null, then on success the target address is stored into
379 inet_open_active(int style, const char *target, uint16_t default_port,
380 struct sockaddr_in *sinp, int *fdp)
382 struct sockaddr_in sin;
387 if (!inet_parse_active(target, default_port, &sin)) {
388 error = EAFNOSUPPORT;
392 /* Create non-blocking socket. */
393 fd = socket(AF_INET, style, 0);
395 VLOG_ERR("%s: socket: %s", target, strerror(errno));
399 error = set_nonblocking(fd);
405 error = connect(fd, (struct sockaddr *) &sin, sizeof sin) == 0 ? 0 : errno;
406 if (error == EINPROGRESS) {
408 } else if (error && error != EAGAIN) {
412 /* Success: error is 0 or EAGAIN. */
418 if (!error || error == EAGAIN) {
429 /* Opens a non-blocking IPv4 socket of the specified 'style', binds to
430 * 'target', and listens for incoming connections. 'target' should be a string
431 * in the format "[<port>][:<ip>]":
433 * - If 'default_port' is -1, then <port> is required. Otherwise, if
434 * <port> is omitted, then 'default_port' is used instead.
436 * - If <port> (or 'default_port', if used) is 0, then no port is bound
437 * and the TCP/IP stack will select a port.
439 * - If <ip> is omitted then the IP address is wildcarded.
441 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
443 * For TCP, the socket will have SO_REUSEADDR turned on.
445 * On success, returns a non-negative file descriptor. On failure, returns a
446 * negative errno value.
448 * If 'sinp' is non-null, then on success the bound address is stored into
451 inet_open_passive(int style, const char *target_, int default_port,
452 struct sockaddr_in *sinp)
454 char *target = xstrdup(target_);
455 char *string_ptr = target;
456 struct sockaddr_in sin;
457 const char *host_name;
458 const char *port_string;
459 int fd = 0, error, port;
460 unsigned int yes = 1;
462 /* Address defaults. */
463 memset(&sin, 0, sizeof sin);
464 sin.sin_family = AF_INET;
465 sin.sin_addr.s_addr = htonl(INADDR_ANY);
466 sin.sin_port = htons(default_port);
468 /* Parse optional port number. */
469 port_string = strsep(&string_ptr, ":");
470 if (port_string && str_to_int(port_string, 10, &port)) {
471 sin.sin_port = htons(port);
472 } else if (default_port < 0) {
473 VLOG_ERR("%s: port number must be specified", target_);
474 error = EAFNOSUPPORT;
478 /* Parse optional bind IP. */
479 host_name = strsep(&string_ptr, ":");
480 if (host_name && host_name[0]) {
481 error = lookup_ip(host_name, &sin.sin_addr);
487 /* Create non-blocking socket, set SO_REUSEADDR. */
488 fd = socket(AF_INET, style, 0);
491 VLOG_ERR("%s: socket: %s", target_, strerror(error));
494 error = set_nonblocking(fd);
498 if (style == SOCK_STREAM
499 && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) {
501 VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s", target_, strerror(error));
506 if (bind(fd, (struct sockaddr *) &sin, sizeof sin) < 0) {
508 VLOG_ERR("%s: bind: %s", target_, strerror(error));
513 if (listen(fd, 10) < 0) {
515 VLOG_ERR("%s: listen: %s", target_, strerror(error));
520 socklen_t sin_len = sizeof sin;
521 if (getsockname(fd, (struct sockaddr *) &sin, &sin_len) < 0){
523 VLOG_ERR("%s: getsockname: %s", target_, strerror(error));
526 if (sin.sin_family != AF_INET || sin_len != sizeof sin) {
527 VLOG_ERR("%s: getsockname: invalid socket name", target_);
540 return error ? -error : fd;
543 /* Returns a readable and writable fd for /dev/null, if successful, otherwise
544 * a negative errno value. The caller must not close the returned fd (because
545 * the same fd will be handed out to subsequent callers). */
549 static int null_fd = -1;
551 null_fd = open("/dev/null", O_RDWR);
554 VLOG_ERR("could not open /dev/null: %s", strerror(error));
562 read_fully(int fd, void *p_, size_t size, size_t *bytes_read)
568 ssize_t retval = read(fd, p, size);
570 *bytes_read += retval;
573 } else if (retval == 0) {
575 } else if (errno != EINTR) {
583 write_fully(int fd, const void *p_, size_t size, size_t *bytes_written)
585 const uint8_t *p = p_;
589 ssize_t retval = write(fd, p, size);
591 *bytes_written += retval;
594 } else if (retval == 0) {
595 VLOG_WARN("write returned 0");
597 } else if (errno != EINTR) {
604 /* Given file name 'file_name', fsyncs the directory in which it is contained.
605 * Returns 0 if successful, otherwise a positive errno value. */
607 fsync_parent_dir(const char *file_name)
613 dir = dir_name(file_name);
614 fd = open(dir, O_RDONLY);
617 if (errno == EINVAL || errno == EROFS) {
618 /* This directory does not support synchronization. Not
619 * really an error. */
622 VLOG_ERR("%s: fsync failed (%s)", dir, strerror(error));
628 VLOG_ERR("%s: open failed (%s)", dir, strerror(error));
635 /* Obtains the modification time of the file named 'file_name' to the greatest
636 * supported precision. If successful, stores the mtime in '*mtime' and
637 * returns 0. On error, returns a positive errno value and stores zeros in
640 get_mtime(const char *file_name, struct timespec *mtime)
644 if (!stat(file_name, &s)) {
645 mtime->tv_sec = s.st_mtime;
647 #if HAVE_STRUCT_STAT_ST_MTIM_TV_NSEC
648 mtime->tv_nsec = s.st_mtim.tv_nsec;
649 #elif HAVE_STRUCT_STAT_ST_MTIMENSEC
650 mtime->tv_nsec = s.st_mtimensec;
657 mtime->tv_sec = mtime->tv_nsec = 0;