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 #define THIS_MODULE VLM_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));
58 /* Returns the maximum valid FD value, plus 1. */
62 static int max_fds = -1;
65 if (!getrlimit(RLIMIT_NOFILE, &r)
66 && r.rlim_cur != RLIM_INFINITY
67 && r.rlim_cur != RLIM_SAVED_MAX
68 && r.rlim_cur != RLIM_SAVED_CUR) {
71 VLOG_WARN("failed to obtain fd limit, defaulting to 1024");
78 /* Translates 'host_name', which must be a string representation of an IP
79 * address, into a numeric IP address in '*addr'. Returns 0 if successful,
80 * otherwise a positive errno value. */
82 lookup_ip(const char *host_name, struct in_addr *addr)
84 if (!inet_aton(host_name, addr)) {
85 struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
86 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name);
92 /* Returns the error condition associated with socket 'fd' and resets the
93 * socket's error status. */
95 get_socket_error(int fd)
98 socklen_t len = sizeof(error);
99 if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &error, &len) < 0) {
100 struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
102 VLOG_ERR_RL(&rl, "getsockopt(SO_ERROR): %s", strerror(error));
108 check_connection_completion(int fd)
114 pfd.events = POLLOUT;
116 retval = poll(&pfd, 1, 0);
117 } while (retval < 0 && errno == EINTR);
119 return get_socket_error(fd);
120 } else if (retval < 0) {
121 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
122 VLOG_ERR_RL(&rl, "poll: %s", strerror(errno));
129 /* Drain all the data currently in the receive queue of a datagram socket (and
130 * possibly additional data). There is no way to know how many packets are in
131 * the receive queue, but we do know that the total number of bytes queued does
132 * not exceed the receive buffer size, so we pull packets until none are left
133 * or we've read that many bytes. */
137 socklen_t rcvbuf_len;
140 rcvbuf_len = sizeof rcvbuf;
141 if (getsockopt(fd, SOL_SOCKET, SO_RCVBUF, &rcvbuf, &rcvbuf_len) < 0) {
142 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
143 VLOG_ERR_RL(&rl, "getsockopt(SO_RCVBUF) failed: %s", strerror(errno));
147 /* In Linux, specifying MSG_TRUNC in the flags argument causes the
148 * datagram length to be returned, even if that is longer than the
149 * buffer provided. Thus, we can use a 1-byte buffer to discard the
150 * incoming datagram and still be able to account how many bytes were
151 * removed from the receive buffer.
153 * On other Unix-like OSes, MSG_TRUNC has no effect in the flags
156 #define BUFFER_SIZE 1
158 #define BUFFER_SIZE 2048
160 char buffer[BUFFER_SIZE];
161 ssize_t n_bytes = recv(fd, buffer, sizeof buffer,
162 MSG_TRUNC | MSG_DONTWAIT);
163 if (n_bytes <= 0 || n_bytes >= rcvbuf) {
171 /* Reads and discards up to 'n' datagrams from 'fd', stopping as soon as no
172 * more data can be immediately read. ('fd' should therefore be in
173 * non-blocking mode.)*/
175 drain_fd(int fd, size_t n_packets)
177 for (; n_packets > 0; n_packets--) {
178 /* 'buffer' only needs to be 1 byte long in most circumstances. This
179 * size is defensive against the possibility that we someday want to
180 * use a Linux tap device without TUN_NO_PI, in which case a buffer
181 * smaller than sizeof(struct tun_pi) will give EINVAL on read. */
183 if (read(fd, buffer, sizeof buffer) <= 0) {
189 /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in
190 * '*un_len' the size of the sockaddr_un. */
192 make_sockaddr_un(const char *name, struct sockaddr_un* un, socklen_t *un_len)
194 un->sun_family = AF_UNIX;
195 strncpy(un->sun_path, name, sizeof un->sun_path);
196 un->sun_path[sizeof un->sun_path - 1] = '\0';
197 *un_len = (offsetof(struct sockaddr_un, sun_path)
198 + strlen (un->sun_path) + 1);
201 /* Creates a Unix domain socket in the given 'style' (either SOCK_DGRAM or
202 * SOCK_STREAM) that is bound to '*bind_path' (if 'bind_path' is non-null) and
203 * connected to '*connect_path' (if 'connect_path' is non-null). If 'nonblock'
204 * is true, the socket is made non-blocking. If 'passcred' is true, the socket
205 * is configured to receive SCM_CREDENTIALS control messages.
207 * Returns the socket's fd if successful, otherwise a negative errno value. */
209 make_unix_socket(int style, bool nonblock, bool passcred OVS_UNUSED,
210 const char *bind_path, const char *connect_path)
215 fd = socket(PF_UNIX, style, 0);
220 /* Set nonblocking mode right away, if we want it. This prevents blocking
221 * in connect(), if connect_path != NULL. (In turn, that's a corner case:
222 * it will only happen if style is SOCK_STREAM or SOCK_SEQPACKET, and only
223 * if a backlog of un-accepted connections has built up in the kernel.) */
225 int flags = fcntl(fd, F_GETFL, 0);
229 if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) == -1) {
235 struct sockaddr_un un;
237 make_sockaddr_un(bind_path, &un, &un_len);
238 if (unlink(un.sun_path) && errno != ENOENT) {
239 VLOG_WARN("unlinking \"%s\": %s\n", un.sun_path, strerror(errno));
241 fatal_signal_add_file_to_unlink(bind_path);
242 if (bind(fd, (struct sockaddr*) &un, un_len)
243 || fchmod(fd, S_IRWXU)) {
249 struct sockaddr_un un;
251 make_sockaddr_un(connect_path, &un, &un_len);
252 if (connect(fd, (struct sockaddr*) &un, un_len)
253 && errno != EINPROGRESS) {
254 printf("connect failed with %s\n", strerror(errno));
259 #ifdef SCM_CREDENTIALS
262 if (setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &enable, sizeof(enable))) {
271 error = errno == EAGAIN ? EPROTO : errno;
273 fatal_signal_remove_file_to_unlink(bind_path);
280 get_unix_name_len(socklen_t sun_len)
282 return (sun_len >= offsetof(struct sockaddr_un, sun_path)
283 ? sun_len - offsetof(struct sockaddr_un, sun_path)
288 guess_netmask(uint32_t ip)
291 return ((ip >> 31) == 0 ? htonl(0xff000000) /* Class A */
292 : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */
293 : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */
294 : htonl(0)); /* ??? */
297 /* Parses 'target', which should be a string in the format "<host>[:<port>]".
298 * <host> is required. If 'default_port' is nonzero then <port> is optional
299 * and defaults to 'default_port'.
301 * On success, returns true and stores the parsed remote address into '*sinp'.
302 * On failure, logs an error, stores zeros into '*sinp', and returns false. */
304 inet_parse_active(const char *target_, uint16_t default_port,
305 struct sockaddr_in *sinp)
307 char *target = xstrdup(target_);
308 char *save_ptr = NULL;
309 const char *host_name;
310 const char *port_string;
314 sinp->sin_family = AF_INET;
315 sinp->sin_port = htons(default_port);
318 host_name = strtok_r(target, ":", &save_ptr);
319 port_string = strtok_r(NULL, ":", &save_ptr);
321 VLOG_ERR("%s: bad peer name format", target_);
325 /* Look up IP, port. */
326 if (lookup_ip(host_name, &sinp->sin_addr)) {
329 if (port_string && atoi(port_string)) {
330 sinp->sin_port = htons(atoi(port_string));
331 } else if (!default_port) {
332 VLOG_ERR("%s: port number must be specified", target_);
340 memset(sinp, 0, sizeof *sinp);
346 /* Opens a non-blocking IPv4 socket of the specified 'style' and connects to
347 * 'target', which should be a string in the format "<host>[:<port>]". <host>
348 * is required. If 'default_port' is nonzero then <port> is optional and
349 * defaults to 'default_port'.
351 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
353 * On success, returns 0 (indicating connection complete) or EAGAIN (indicating
354 * connection in progress), in which case the new file descriptor is stored
355 * into '*fdp'. On failure, returns a positive errno value other than EAGAIN
356 * and stores -1 into '*fdp'.
358 * If 'sinp' is non-null, then on success the target address is stored into
361 inet_open_active(int style, const char *target, uint16_t default_port,
362 struct sockaddr_in *sinp, int *fdp)
364 struct sockaddr_in sin;
369 if (!inet_parse_active(target, default_port, &sin)) {
370 error = EAFNOSUPPORT;
374 /* Create non-blocking socket. */
375 fd = socket(AF_INET, style, 0);
377 VLOG_ERR("%s: socket: %s", target, strerror(errno));
381 error = set_nonblocking(fd);
387 error = connect(fd, (struct sockaddr *) &sin, sizeof sin) == 0 ? 0 : errno;
388 if (error == EINPROGRESS) {
390 } else if (error && error != EAGAIN) {
394 /* Success: error is 0 or EAGAIN. */
400 if (!error || error == EAGAIN) {
411 /* Opens a non-blocking IPv4 socket of the specified 'style', binds to
412 * 'target', and listens for incoming connections. 'target' should be a string
413 * in the format "[<port>][:<ip>]":
415 * - If 'default_port' is -1, then <port> is required. Otherwise, if
416 * <port> is omitted, then 'default_port' is used instead.
418 * - If <port> (or 'default_port', if used) is 0, then no port is bound
419 * and the TCP/IP stack will select a port.
421 * - If <ip> is omitted then the IP address is wildcarded.
423 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
425 * For TCP, the socket will have SO_REUSEADDR turned on.
427 * On success, returns a non-negative file descriptor. On failure, returns a
428 * negative errno value.
430 * If 'sinp' is non-null, then on success the bound address is stored into
433 inet_open_passive(int style, const char *target_, int default_port,
434 struct sockaddr_in *sinp)
436 char *target = xstrdup(target_);
437 char *string_ptr = target;
438 struct sockaddr_in sin;
439 const char *host_name;
440 const char *port_string;
442 unsigned int yes = 1;
444 /* Address defaults. */
445 memset(&sin, 0, sizeof sin);
446 sin.sin_family = AF_INET;
447 sin.sin_addr.s_addr = htonl(INADDR_ANY);
448 sin.sin_port = htons(default_port);
450 /* Parse optional port number. */
451 port_string = strsep(&string_ptr, ":");
452 if (port_string && str_to_int(port_string, 10, &port)) {
453 sin.sin_port = htons(port);
454 } else if (default_port < 0) {
455 VLOG_ERR("%s: port number must be specified", target_);
456 error = EAFNOSUPPORT;
460 /* Parse optional bind IP. */
461 host_name = strsep(&string_ptr, ":");
462 if (host_name && host_name[0]) {
463 error = lookup_ip(host_name, &sin.sin_addr);
469 /* Create non-blocking socket, set SO_REUSEADDR. */
470 fd = socket(AF_INET, style, 0);
473 VLOG_ERR("%s: socket: %s", target_, strerror(error));
476 error = set_nonblocking(fd);
480 if (style == SOCK_STREAM
481 && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) {
483 VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s", target_, strerror(error));
488 if (bind(fd, (struct sockaddr *) &sin, sizeof sin) < 0) {
490 VLOG_ERR("%s: bind: %s", target_, strerror(error));
495 if (listen(fd, 10) < 0) {
497 VLOG_ERR("%s: listen: %s", target_, strerror(error));
502 socklen_t sin_len = sizeof sin;
503 if (getsockname(fd, (struct sockaddr *) &sin, &sin_len) < 0){
505 VLOG_ERR("%s: getsockname: %s", target_, strerror(error));
508 if (sin.sin_family != AF_INET || sin_len != sizeof sin) {
509 VLOG_ERR("%s: getsockname: invalid socket name", target_);
522 return error ? -error : fd;
525 /* Returns a readable and writable fd for /dev/null, if successful, otherwise
526 * a negative errno value. The caller must not close the returned fd (because
527 * the same fd will be handed out to subsequent callers). */
531 static int null_fd = -1;
533 null_fd = open("/dev/null", O_RDWR);
536 VLOG_ERR("could not open /dev/null: %s", strerror(error));
544 read_fully(int fd, void *p_, size_t size, size_t *bytes_read)
550 ssize_t retval = read(fd, p, size);
552 *bytes_read += retval;
555 } else if (retval == 0) {
557 } else if (errno != EINTR) {
565 write_fully(int fd, const void *p_, size_t size, size_t *bytes_written)
567 const uint8_t *p = p_;
571 ssize_t retval = write(fd, p, size);
573 *bytes_written += retval;
576 } else if (retval == 0) {
577 VLOG_WARN("write returned 0");
579 } else if (errno != EINTR) {
586 /* Given file name 'file_name', fsyncs the directory in which it is contained.
587 * Returns 0 if successful, otherwise a positive errno value. */
589 fsync_parent_dir(const char *file_name)
595 dir = dir_name(file_name);
596 fd = open(dir, O_RDONLY);
599 if (errno == EINVAL || errno == EROFS) {
600 /* This directory does not support synchronization. Not
601 * really an error. */
604 VLOG_ERR("%s: fsync failed (%s)", dir, strerror(error));
610 VLOG_ERR("%s: open failed (%s)", dir, strerror(error));
617 /* Obtains the modification time of the file named 'file_name' to the greatest
618 * supported precision. If successful, stores the mtime in '*mtime' and
619 * returns 0. On error, returns a positive errno value and stores zeros in
622 get_mtime(const char *file_name, struct timespec *mtime)
626 if (!stat(file_name, &s)) {
627 mtime->tv_sec = s.st_mtime;
629 #if HAVE_STRUCT_STAT_ST_MTIM_TV_NSEC
630 mtime->tv_nsec = s.st_mtim.tv_nsec;
631 #elif HAVE_STRUCT_STAT_ST_MTIMENSEC
632 mtime->tv_nsec = s.st_mtimensec;
639 mtime->tv_sec = mtime->tv_nsec = 0;