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>
31 #include "fatal-signal.h"
35 #define THIS_MODULE VLM_socket_util
37 /* Sets 'fd' to non-blocking mode. Returns 0 if successful, otherwise a
38 * positive errno value. */
40 set_nonblocking(int fd)
42 int flags = fcntl(fd, F_GETFL, 0);
44 if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) != -1) {
47 VLOG_ERR("fcntl(F_SETFL) failed: %s", strerror(errno));
51 VLOG_ERR("fcntl(F_GETFL) failed: %s", strerror(errno));
56 /* Returns the maximum valid FD value, plus 1. */
60 static int max_fds = -1;
63 if (!getrlimit(RLIMIT_NOFILE, &r)
64 && r.rlim_cur != RLIM_INFINITY
65 && r.rlim_cur != RLIM_SAVED_MAX
66 && r.rlim_cur != RLIM_SAVED_CUR) {
69 VLOG_WARN("failed to obtain fd limit, defaulting to 1024");
76 /* Translates 'host_name', which must be a string representation of an IP
77 * address, into a numeric IP address in '*addr'. Returns 0 if successful,
78 * otherwise a positive errno value. */
80 lookup_ip(const char *host_name, struct in_addr *addr)
82 if (!inet_aton(host_name, addr)) {
83 struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
84 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name);
90 /* Returns the error condition associated with socket 'fd' and resets the
91 * socket's error status. */
93 get_socket_error(int fd)
96 socklen_t len = sizeof(error);
97 if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &error, &len) < 0) {
98 struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
100 VLOG_ERR_RL(&rl, "getsockopt(SO_ERROR): %s", strerror(error));
106 check_connection_completion(int fd)
112 pfd.events = POLLOUT;
114 retval = poll(&pfd, 1, 0);
115 } while (retval < 0 && errno == EINTR);
117 return get_socket_error(fd);
118 } else if (retval < 0) {
119 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
120 VLOG_ERR_RL(&rl, "poll: %s", strerror(errno));
127 /* Drain all the data currently in the receive queue of a datagram socket (and
128 * possibly additional data). There is no way to know how many packets are in
129 * the receive queue, but we do know that the total number of bytes queued does
130 * not exceed the receive buffer size, so we pull packets until none are left
131 * or we've read that many bytes. */
135 socklen_t rcvbuf_len;
138 rcvbuf_len = sizeof rcvbuf;
139 if (getsockopt(fd, SOL_SOCKET, SO_RCVBUF, &rcvbuf, &rcvbuf_len) < 0) {
140 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
141 VLOG_ERR_RL(&rl, "getsockopt(SO_RCVBUF) failed: %s", strerror(errno));
145 /* In Linux, specifying MSG_TRUNC in the flags argument causes the
146 * datagram length to be returned, even if that is longer than the
147 * buffer provided. Thus, we can use a 1-byte buffer to discard the
148 * incoming datagram and still be able to account how many bytes were
149 * removed from the receive buffer.
151 * On other Unix-like OSes, MSG_TRUNC has no effect in the flags
154 #define BUFFER_SIZE 1
156 #define BUFFER_SIZE 2048
158 char buffer[BUFFER_SIZE];
159 ssize_t n_bytes = recv(fd, buffer, sizeof buffer,
160 MSG_TRUNC | MSG_DONTWAIT);
161 if (n_bytes <= 0 || n_bytes >= rcvbuf) {
169 /* Reads and discards up to 'n' datagrams from 'fd', stopping as soon as no
170 * more data can be immediately read. ('fd' should therefore be in
171 * non-blocking mode.)*/
173 drain_fd(int fd, size_t n_packets)
175 for (; n_packets > 0; n_packets--) {
176 /* 'buffer' only needs to be 1 byte long in most circumstances. This
177 * size is defensive against the possibility that we someday want to
178 * use a Linux tap device without TUN_NO_PI, in which case a buffer
179 * smaller than sizeof(struct tun_pi) will give EINVAL on read. */
181 if (read(fd, buffer, sizeof buffer) <= 0) {
187 /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in
188 * '*un_len' the size of the sockaddr_un. */
190 make_sockaddr_un(const char *name, struct sockaddr_un* un, socklen_t *un_len)
192 un->sun_family = AF_UNIX;
193 strncpy(un->sun_path, name, sizeof un->sun_path);
194 un->sun_path[sizeof un->sun_path - 1] = '\0';
195 *un_len = (offsetof(struct sockaddr_un, sun_path)
196 + strlen (un->sun_path) + 1);
199 /* Creates a Unix domain socket in the given 'style' (either SOCK_DGRAM or
200 * SOCK_STREAM) that is bound to '*bind_path' (if 'bind_path' is non-null) and
201 * connected to '*connect_path' (if 'connect_path' is non-null). If 'nonblock'
202 * is true, the socket is made non-blocking. If 'passcred' is true, the socket
203 * is configured to receive SCM_CREDENTIALS control messages.
205 * Returns the socket's fd if successful, otherwise a negative errno value. */
207 make_unix_socket(int style, bool nonblock, bool passcred OVS_UNUSED,
208 const char *bind_path, const char *connect_path)
213 fd = socket(PF_UNIX, style, 0);
218 /* Set nonblocking mode right away, if we want it. This prevents blocking
219 * in connect(), if connect_path != NULL. (In turn, that's a corner case:
220 * it will only happen if style is SOCK_STREAM or SOCK_SEQPACKET, and only
221 * if a backlog of un-accepted connections has built up in the kernel.) */
223 int flags = fcntl(fd, F_GETFL, 0);
227 if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) == -1) {
233 struct sockaddr_un un;
235 make_sockaddr_un(bind_path, &un, &un_len);
236 if (unlink(un.sun_path) && errno != ENOENT) {
237 VLOG_WARN("unlinking \"%s\": %s\n", un.sun_path, strerror(errno));
239 fatal_signal_add_file_to_unlink(bind_path);
240 if (bind(fd, (struct sockaddr*) &un, un_len)
241 || fchmod(fd, S_IRWXU)) {
247 struct sockaddr_un un;
249 make_sockaddr_un(connect_path, &un, &un_len);
250 if (connect(fd, (struct sockaddr*) &un, un_len)
251 && errno != EINPROGRESS) {
256 #ifdef SCM_CREDENTIALS
259 if (setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &enable, sizeof(enable))) {
269 fatal_signal_remove_file_to_unlink(bind_path);
277 get_unix_name_len(socklen_t sun_len)
279 return (sun_len >= offsetof(struct sockaddr_un, sun_path)
280 ? sun_len - offsetof(struct sockaddr_un, sun_path)
285 guess_netmask(uint32_t ip)
288 return ((ip >> 31) == 0 ? htonl(0xff000000) /* Class A */
289 : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */
290 : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */
291 : htonl(0)); /* ??? */
294 /* Opens a non-blocking IPv4 socket of the specified 'style' and connects to
295 * 'target', which should be a string in the format "<host>[:<port>]". <host>
296 * is required. If 'default_port' is nonzero then <port> is optional and
297 * defaults to 'default_port'.
299 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
301 * On success, returns 0 (indicating connection complete) or EAGAIN (indicating
302 * connection in progress), in which case the new file descriptor is stored
303 * into '*fdp'. On failure, returns a positive errno value other than EAGAIN
304 * and stores -1 into '*fdp'.
306 * If 'sinp' is non-null, then on success the target address is stored into
309 inet_open_active(int style, const char *target_, uint16_t default_port,
310 struct sockaddr_in *sinp, int *fdp)
312 char *target = xstrdup(target_);
313 char *save_ptr = NULL;
314 const char *host_name;
315 const char *port_string;
316 struct sockaddr_in sin;
321 memset(&sin, 0, sizeof sin);
322 sin.sin_family = AF_INET;
323 sin.sin_port = htons(default_port);
326 host_name = strtok_r(target, ":", &save_ptr);
327 port_string = strtok_r(NULL, ":", &save_ptr);
329 ovs_error(0, "%s: bad peer name format", target_);
330 error = EAFNOSUPPORT;
334 /* Look up IP, port. */
335 error = lookup_ip(host_name, &sin.sin_addr);
339 if (port_string && atoi(port_string)) {
340 sin.sin_port = htons(atoi(port_string));
341 } else if (!default_port) {
342 VLOG_ERR("%s: port number must be specified", target_);
343 error = EAFNOSUPPORT;
347 /* Create non-blocking socket. */
348 fd = socket(AF_INET, style, 0);
350 VLOG_ERR("%s: socket: %s", target_, strerror(errno));
354 error = set_nonblocking(fd);
360 error = connect(fd, (struct sockaddr *) &sin, sizeof sin) == 0 ? 0 : errno;
361 if (error == EINPROGRESS) {
363 } else if (error && error != EAGAIN) {
367 /* Success: error is 0 or EAGAIN. */
373 if (!error || error == EAGAIN) {
385 /* Opens a non-blocking IPv4 socket of the specified 'style', binds to
386 * 'target', and listens for incoming connections. 'target' should be a string
387 * in the format "[<port>][:<ip>]". <port> may be omitted if 'default_port' is
388 * nonzero, in which case it defaults to 'default_port'. If <ip> is omitted it
389 * defaults to the wildcard IP address.
391 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
393 * For TCP, the socket will have SO_REUSEADDR turned on.
395 * On success, returns a non-negative file descriptor. On failure, returns a
396 * negative errno value. */
398 inet_open_passive(int style, const char *target_, uint16_t default_port)
400 char *target = xstrdup(target_);
401 char *string_ptr = target;
402 struct sockaddr_in sin;
403 const char *host_name;
404 const char *port_string;
406 unsigned int yes = 1;
408 /* Address defaults. */
409 memset(&sin, 0, sizeof sin);
410 sin.sin_family = AF_INET;
411 sin.sin_addr.s_addr = htonl(INADDR_ANY);
412 sin.sin_port = htons(default_port);
414 /* Parse optional port number. */
415 port_string = strsep(&string_ptr, ":");
416 if (port_string && atoi(port_string)) {
417 sin.sin_port = htons(atoi(port_string));
418 } else if (!default_port) {
419 VLOG_ERR("%s: port number must be specified", target_);
420 error = EAFNOSUPPORT;
424 /* Parse optional bind IP. */
425 host_name = strsep(&string_ptr, ":");
426 if (host_name && host_name[0]) {
427 error = lookup_ip(host_name, &sin.sin_addr);
433 /* Create non-blocking socket, set SO_REUSEADDR. */
434 fd = socket(AF_INET, style, 0);
437 VLOG_ERR("%s: socket: %s", target_, strerror(error));
440 error = set_nonblocking(fd);
444 if (style == SOCK_STREAM
445 && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) {
447 VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s", target_, strerror(error));
452 if (bind(fd, (struct sockaddr *) &sin, sizeof sin) < 0) {
454 VLOG_ERR("%s: bind: %s", target_, strerror(error));
459 if (listen(fd, 10) < 0) {
461 VLOG_ERR("%s: listen: %s", target_, strerror(error));
471 return error ? -error : fd;
474 /* Returns a readable and writable fd for /dev/null, if successful, otherwise
475 * a negative errno value. The caller must not close the returned fd (because
476 * the same fd will be handed out to subsequent callers). */
480 static int null_fd = -1;
482 null_fd = open("/dev/null", O_RDWR);
485 VLOG_ERR("could not open /dev/null: %s", strerror(error));
493 read_fully(int fd, void *p_, size_t size, size_t *bytes_read)
499 ssize_t retval = read(fd, p, size);
501 *bytes_read += retval;
504 } else if (retval == 0) {
506 } else if (errno != EINTR) {
514 write_fully(int fd, const void *p_, size_t size, size_t *bytes_written)
516 const uint8_t *p = p_;
520 ssize_t retval = write(fd, p, size);
522 *bytes_written += retval;
525 } else if (retval == 0) {
526 VLOG_WARN("write returned 0");
528 } else if (errno != EINTR) {