2 * Copyright (c) 2008, 2009 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 may be a DNS name or an IP address, into a
77 * numeric IP address in '*addr'. Returns 0 if successful, otherwise a
78 * positive errno value. */
80 lookup_ip(const char *host_name, struct in_addr *addr)
82 if (!inet_aton(host_name, addr)) {
83 struct hostent *he = gethostbyname(host_name);
85 struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
86 VLOG_ERR_RL(&rl, "gethostbyname(%s): %s", host_name,
87 (h_errno == HOST_NOT_FOUND ? "host not found"
88 : h_errno == TRY_AGAIN ? "try again"
89 : h_errno == NO_RECOVERY ? "non-recoverable error"
90 : h_errno == NO_ADDRESS ? "no address"
94 addr->s_addr = *(uint32_t *) he->h_addr;
99 /* Returns the error condition associated with socket 'fd' and resets the
100 * socket's error status. */
102 get_socket_error(int fd)
105 socklen_t len = sizeof(error);
106 if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &error, &len) < 0) {
107 struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
109 VLOG_ERR_RL(&rl, "getsockopt(SO_ERROR): %s", strerror(error));
115 check_connection_completion(int fd)
121 pfd.events = POLLOUT;
123 retval = poll(&pfd, 1, 0);
124 } while (retval < 0 && errno == EINTR);
126 return get_socket_error(fd);
127 } else if (retval < 0) {
128 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
129 VLOG_ERR_RL(&rl, "poll: %s", strerror(errno));
136 /* Drain all the data currently in the receive queue of a datagram socket (and
137 * possibly additional data). There is no way to know how many packets are in
138 * the receive queue, but we do know that the total number of bytes queued does
139 * not exceed the receive buffer size, so we pull packets until none are left
140 * or we've read that many bytes. */
144 socklen_t rcvbuf_len;
147 rcvbuf_len = sizeof rcvbuf;
148 if (getsockopt(fd, SOL_SOCKET, SO_RCVBUF, &rcvbuf, &rcvbuf_len) < 0) {
149 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
150 VLOG_ERR_RL(&rl, "getsockopt(SO_RCVBUF) failed: %s", strerror(errno));
154 /* In Linux, specifying MSG_TRUNC in the flags argument causes the
155 * datagram length to be returned, even if that is longer than the
156 * buffer provided. Thus, we can use a 1-byte buffer to discard the
157 * incoming datagram and still be able to account how many bytes were
158 * removed from the receive buffer.
160 * On other Unix-like OSes, MSG_TRUNC has no effect in the flags
163 #define BUFFER_SIZE 1
165 #define BUFFER_SIZE 2048
167 char buffer[BUFFER_SIZE];
168 ssize_t n_bytes = recv(fd, buffer, sizeof buffer,
169 MSG_TRUNC | MSG_DONTWAIT);
170 if (n_bytes <= 0 || n_bytes >= rcvbuf) {
178 /* Reads and discards up to 'n' datagrams from 'fd', stopping as soon as no
179 * more data can be immediately read. ('fd' should therefore be in
180 * non-blocking mode.)*/
182 drain_fd(int fd, size_t n_packets)
184 for (; n_packets > 0; n_packets--) {
185 /* 'buffer' only needs to be 1 byte long in most circumstances. This
186 * size is defensive against the possibility that we someday want to
187 * use a Linux tap device without TUN_NO_PI, in which case a buffer
188 * smaller than sizeof(struct tun_pi) will give EINVAL on read. */
190 if (read(fd, buffer, sizeof buffer) <= 0) {
196 /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in
197 * '*un_len' the size of the sockaddr_un. */
199 make_sockaddr_un(const char *name, struct sockaddr_un* un, socklen_t *un_len)
201 un->sun_family = AF_UNIX;
202 strncpy(un->sun_path, name, sizeof un->sun_path);
203 un->sun_path[sizeof un->sun_path - 1] = '\0';
204 *un_len = (offsetof(struct sockaddr_un, sun_path)
205 + strlen (un->sun_path) + 1);
208 /* Creates a Unix domain socket in the given 'style' (either SOCK_DGRAM or
209 * SOCK_STREAM) that is bound to '*bind_path' (if 'bind_path' is non-null) and
210 * connected to '*connect_path' (if 'connect_path' is non-null). If 'nonblock'
211 * is true, the socket is made non-blocking. If 'passcred' is true, the socket
212 * is configured to receive SCM_CREDENTIALS control messages.
214 * Returns the socket's fd if successful, otherwise a negative errno value. */
216 make_unix_socket(int style, bool nonblock, bool passcred UNUSED,
217 const char *bind_path, const char *connect_path)
222 fd = socket(PF_UNIX, style, 0);
227 /* Set nonblocking mode right away, if we want it. This prevents blocking
228 * in connect(), if connect_path != NULL. (In turn, that's a corner case:
229 * it will only happen if style is SOCK_STREAM or SOCK_SEQPACKET, and only
230 * if a backlog of un-accepted connections has built up in the kernel.) */
232 int flags = fcntl(fd, F_GETFL, 0);
236 if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) == -1) {
242 struct sockaddr_un un;
244 make_sockaddr_un(bind_path, &un, &un_len);
245 if (unlink(un.sun_path) && errno != ENOENT) {
246 VLOG_WARN("unlinking \"%s\": %s\n", un.sun_path, strerror(errno));
248 fatal_signal_add_file_to_unlink(bind_path);
249 if (bind(fd, (struct sockaddr*) &un, un_len)
250 || fchmod(fd, S_IRWXU)) {
256 struct sockaddr_un un;
258 make_sockaddr_un(connect_path, &un, &un_len);
259 if (connect(fd, (struct sockaddr*) &un, un_len)
260 && errno != EINPROGRESS) {
265 #ifdef SCM_CREDENTIALS
268 if (setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &enable, sizeof(enable))) {
278 fatal_signal_remove_file_to_unlink(bind_path);
286 get_unix_name_len(socklen_t sun_len)
288 return (sun_len >= offsetof(struct sockaddr_un, sun_path)
289 ? sun_len - offsetof(struct sockaddr_un, sun_path)
294 guess_netmask(uint32_t ip)
297 return ((ip >> 31) == 0 ? htonl(0xff000000) /* Class A */
298 : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */
299 : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */
300 : htonl(0)); /* ??? */
303 /* Opens a non-blocking TCP socket and connects to 'target', which should be a
304 * string in the format "<host>[:<port>]", where <host> is required and <port>
305 * is optional, with 'default_port' assumed if <port> is omitted.
307 * On success, returns 0 (indicating connection complete) or EAGAIN (indicating
308 * connection in progress), in which case the new file descriptor is stored
309 * into '*fdp'. On failure, returns a positive errno value other than EAGAIN
310 * and stores -1 into '*fdp'.
312 * If 'sinp' is non-null, then on success the target address is stored into
315 tcp_open_active(const char *target_, uint16_t default_port,
316 struct sockaddr_in *sinp, int *fdp)
318 char *target = xstrdup(target_);
319 char *save_ptr = NULL;
320 const char *host_name;
321 const char *port_string;
322 struct sockaddr_in sin;
327 memset(&sin, 0, sizeof sin);
328 sin.sin_family = AF_INET;
329 sin.sin_port = htons(default_port);
332 host_name = strtok_r(target, ":", &save_ptr);
333 port_string = strtok_r(NULL, ":", &save_ptr);
335 ovs_error(0, "%s: bad peer name format", target_);
336 error = EAFNOSUPPORT;
340 /* Look up IP, port. */
341 error = lookup_ip(host_name, &sin.sin_addr);
345 if (port_string && atoi(port_string)) {
346 sin.sin_port = htons(atoi(port_string));
349 /* Create non-blocking socket. */
350 fd = socket(AF_INET, SOCK_STREAM, 0);
352 VLOG_ERR("%s: socket: %s", target_, strerror(errno));
356 error = set_nonblocking(fd);
362 error = connect(fd, (struct sockaddr *) &sin, sizeof sin) == 0 ? 0 : errno;
363 if (error == EINPROGRESS) {
365 } else if (error && error != EAGAIN) {
369 /* Success: error is 0 or EAGAIN. */
375 if (!error || error == EAGAIN) {
387 /* Opens a non-blocking TCP socket, binds to 'target', and listens for incoming
388 * connections. 'target' should be a string in the format "[<port>][:<ip>]",
389 * where both <port> and <ip> are optional. If <port> is omitted, it defaults
390 * to 'default_port'; if <ip> is omitted it defaults to the wildcard IP
393 * 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 tcp_open_passive(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));
420 /* Parse optional bind IP. */
421 host_name = strsep(&string_ptr, ":");
422 if (host_name && host_name[0]) {
423 error = lookup_ip(host_name, &sin.sin_addr);
429 /* Create non-blocking socket, set SO_REUSEADDR. */
430 fd = socket(AF_INET, SOCK_STREAM, 0);
433 VLOG_ERR("%s: socket: %s", target_, strerror(error));
436 error = set_nonblocking(fd);
440 if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) {
442 VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s", target_, strerror(error));
447 if (bind(fd, (struct sockaddr *) &sin, sizeof sin) < 0) {
449 VLOG_ERR("%s: bind: %s", target_, strerror(error));
454 if (listen(fd, 10) < 0) {
456 VLOG_ERR("%s: listen: %s", target_, strerror(error));
466 return error ? -error : fd;
469 /* Returns a readable and writable fd for /dev/null, if successful, otherwise
470 * a negative errno value. The caller must not close the returned fd (because
471 * the same fd will be handed out to subsequent callers). */
475 static int null_fd = -1;
477 null_fd = open("/dev/null", O_RDWR);
480 VLOG_ERR("could not open /dev/null: %s", strerror(error));
488 read_fully(int fd, void *p_, size_t size, size_t *bytes_read)
494 ssize_t retval = read(fd, p, size);
496 *bytes_read += retval;
499 } else if (retval == 0) {
501 } else if (errno != EINTR) {
509 write_fully(int fd, const void *p_, size_t size, size_t *bytes_written)
511 const uint8_t *p = p_;
515 ssize_t retval = write(fd, p, size);
517 *bytes_written += retval;
520 } else if (retval == 0) {
521 VLOG_WARN("write returned 0");
523 } else if (errno != EINTR) {