lib/socket-util.c

   1 /*
   2  * Copyright (c) 2008, 2009 Nicira Networks.
   3  *
   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:
   7  *
   8  *     http://www.apache.org/licenses/LICENSE-2.0
   9  *
  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.
  15  */
  16
  17 #include <config.h>
  18 #include "socket-util.h"
  19 #include <arpa/inet.h>
  20 #include <errno.h>
  21 #include <fcntl.h>
  22 #include <netdb.h>
  23 #include <poll.h>
  24 #include <stddef.h>
  25 #include <stdio.h>
  26 #include <stdlib.h>
  27 #include <string.h>
  28 #include <sys/resource.h>
  29 #include <sys/un.h>
  30 #include <unistd.h>
  31 #include "fatal-signal.h"
  32 #include "util.h"
  33
  34 #include "vlog.h"
  35 #define THIS_MODULE VLM_socket_util
  36
  37 /* Sets 'fd' to non-blocking mode.  Returns 0 if successful, otherwise a
  38  * positive errno value. */
  39 int
  40 set_nonblocking(int fd)
  41 {
  42     int flags = fcntl(fd, F_GETFL, 0);
  43     if (flags != -1) {
  44         if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) != -1) {
  45             return 0;
  46         } else {
  47             VLOG_ERR("fcntl(F_SETFL) failed: %s", strerror(errno));
  48             return errno;
  49         }
  50     } else {
  51         VLOG_ERR("fcntl(F_GETFL) failed: %s", strerror(errno));
  52         return errno;
  53     }
  54 }
  55
  56 /* Returns the maximum valid FD value, plus 1. */
  57 int
  58 get_max_fds(void)
  59 {
  60     static int max_fds = -1;
  61     if (max_fds < 0) {
  62         struct rlimit r;
  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) {
  67             max_fds = r.rlim_cur;
  68         } else {
  69             VLOG_WARN("failed to obtain fd limit, defaulting to 1024");
  70             max_fds = 1024;
  71         }
  72     }
  73     return max_fds;
  74 }
  75
  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. */
  79 int
  80 lookup_ip(const char *host_name, struct in_addr *addr)
  81 {
  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);
  85         return ENOENT;
  86     }
  87     return 0;
  88 }
  89
  90 /* Returns the error condition associated with socket 'fd' and resets the
  91  * socket's error status. */
  92 int
  93 get_socket_error(int fd)
  94 {
  95     int error;
  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);
  99         error = errno;
 100         VLOG_ERR_RL(&rl, "getsockopt(SO_ERROR): %s", strerror(error));
 101     }
 102     return error;
 103 }
 104
 105 int
 106 check_connection_completion(int fd)
 107 {
 108     struct pollfd pfd;
 109     int retval;
 110
 111     pfd.fd = fd;
 112     pfd.events = POLLOUT;
 113     do {
 114         retval = poll(&pfd, 1, 0);
 115     } while (retval < 0 && errno == EINTR);
 116     if (retval == 1) {
 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));
 121         return errno;
 122     } else {
 123         return EAGAIN;
 124     }
 125 }
 126
 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. */
 132 int
 133 drain_rcvbuf(int fd)
 134 {
 135     socklen_t rcvbuf_len;
 136     size_t rcvbuf;
 137
 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));
 142         return errno;
 143     }
 144     while (rcvbuf > 0) {
 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.
 150          *
 151          * On other Unix-like OSes, MSG_TRUNC has no effect in the flags
 152          * argument. */
 153 #ifdef __linux__
 154 #define BUFFER_SIZE 1
 155 #else
 156 #define BUFFER_SIZE 2048
 157 #endif
 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) {
 162             break;
 163         }
 164         rcvbuf -= n_bytes;
 165     }
 166     return 0;
 167 }
 168
 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.)*/
 172 void
 173 drain_fd(int fd, size_t n_packets)
 174 {
 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. */
 180         char buffer[128];
 181         if (read(fd, buffer, sizeof buffer) <= 0) {
 182             break;
 183         }
 184     }
 185 }
 186
 187 /* Stores in '*un' a sockaddr_un that refers to file 'name'.  Stores in
 188  * '*un_len' the size of the sockaddr_un. */
 189 static void
 190 make_sockaddr_un(const char *name, struct sockaddr_un* un, socklen_t *un_len)
 191 {
 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);
 197 }
 198
 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.
 204  *
 205  * Returns the socket's fd if successful, otherwise a negative errno value. */
 206 int
 207 make_unix_socket(int style, bool nonblock, bool passcred UNUSED,
 208                  const char *bind_path, const char *connect_path)
 209 {
 210     int error;
 211     int fd;
 212
 213     fd = socket(PF_UNIX, style, 0);
 214     if (fd < 0) {
 215         return -errno;
 216     }
 217
 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.)  */
 222     if (nonblock) {
 223         int flags = fcntl(fd, F_GETFL, 0);
 224         if (flags == -1) {
 225             goto error;
 226         }
 227         if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) == -1) {
 228             goto error;
 229         }
 230     }
 231
 232     if (bind_path) {
 233         struct sockaddr_un un;
 234         socklen_t un_len;
 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));
 238         }
 239         fatal_signal_add_file_to_unlink(bind_path);
 240         if (bind(fd, (struct sockaddr*) &un, un_len)
 241             || fchmod(fd, S_IRWXU)) {
 242             goto error;
 243         }
 244     }
 245
 246     if (connect_path) {
 247         struct sockaddr_un un;
 248         socklen_t un_len;
 249         make_sockaddr_un(connect_path, &un, &un_len);
 250         if (connect(fd, (struct sockaddr*) &un, un_len)
 251             && errno != EINPROGRESS) {
 252             goto error;
 253         }
 254     }
 255
 256 #ifdef SCM_CREDENTIALS
 257     if (passcred) {
 258         int enable = 1;
 259         if (setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &enable, sizeof(enable))) {
 260             goto error;
 261         }
 262     }
 263 #endif
 264
 265     return fd;
 266
 267 error:
 268     if (bind_path) {
 269         fatal_signal_remove_file_to_unlink(bind_path);
 270     }
 271     error = errno;
 272     close(fd);
 273     return -error;
 274 }
 275
 276 int
 277 get_unix_name_len(socklen_t sun_len)
 278 {
 279     return (sun_len >= offsetof(struct sockaddr_un, sun_path)
 280             ? sun_len - offsetof(struct sockaddr_un, sun_path)
 281             : 0);
 282 }
 283
 284 uint32_t
 285 guess_netmask(uint32_t ip)
 286 {
 287     ip = ntohl(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));                          /* ??? */
 292 }
 293
 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'.
 298  *
 299  * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
 300  *
 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'.
 305  *
 306  * If 'sinp' is non-null, then on success the target address is stored into
 307  * '*sinp'. */
 308 int
 309 inet_open_active(int style, const char *target_, uint16_t default_port,
 310                  struct sockaddr_in *sinp, int *fdp)
 311 {
 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;
 317     int fd = -1;
 318     int error;
 319
 320     /* Defaults. */
 321     memset(&sin, 0, sizeof sin);
 322     sin.sin_family = AF_INET;
 323     sin.sin_port = htons(default_port);
 324
 325     /* Tokenize. */
 326     host_name = strtok_r(target, ":", &save_ptr);
 327     port_string = strtok_r(NULL, ":", &save_ptr);
 328     if (!host_name) {
 329         ovs_error(0, "%s: bad peer name format", target_);
 330         error = EAFNOSUPPORT;
 331         goto exit;
 332     }
 333
 334     /* Look up IP, port. */
 335     error = lookup_ip(host_name, &sin.sin_addr);
 336     if (error) {
 337         goto exit;
 338     }
 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;
 344         goto exit;
 345     }
 346
 347     /* Create non-blocking socket. */
 348     fd = socket(AF_INET, style, 0);
 349     if (fd < 0) {
 350         VLOG_ERR("%s: socket: %s", target_, strerror(errno));
 351         error = errno;
 352         goto exit;
 353     }
 354     error = set_nonblocking(fd);
 355     if (error) {
 356         goto exit_close;
 357     }
 358
 359     /* Connect. */
 360     error = connect(fd, (struct sockaddr *) &sin, sizeof sin) == 0 ? 0 : errno;
 361     if (error == EINPROGRESS) {
 362         error = EAGAIN;
 363     } else if (error && error != EAGAIN) {
 364         goto exit_close;
 365     }
 366
 367     /* Success: error is 0 or EAGAIN. */
 368     goto exit;
 369
 370 exit_close:
 371     close(fd);
 372 exit:
 373     if (!error || error == EAGAIN) {
 374         if (sinp) {
 375             *sinp = sin;
 376         }
 377         *fdp = fd;
 378     } else {
 379         *fdp = -1;
 380     }
 381     free(target);
 382     return error;
 383 }
 384
 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.
 390  *
 391  * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
 392  *
 393  * For TCP, the socket will have SO_REUSEADDR turned on.
 394  *
 395  * On success, returns a non-negative file descriptor.  On failure, returns a
 396  * negative errno value. */
 397 int
 398 inet_open_passive(int style, const char *target_, uint16_t default_port)
 399 {
 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;
 405     int fd, error;
 406     unsigned int yes  = 1;
 407
 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);
 413
 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;
 421         goto exit;
 422     }
 423
 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);
 428         if (error) {
 429             goto exit;
 430         }
 431     }
 432
 433     /* Create non-blocking socket, set SO_REUSEADDR. */
 434     fd = socket(AF_INET, style, 0);
 435     if (fd < 0) {
 436         error = errno;
 437         VLOG_ERR("%s: socket: %s", target_, strerror(error));
 438         goto exit;
 439     }
 440     error = set_nonblocking(fd);
 441     if (error) {
 442         goto exit_close;
 443     }
 444     if (style == SOCK_STREAM
 445         && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) {
 446         error = errno;
 447         VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s", target_, strerror(error));
 448         goto exit_close;
 449     }
 450
 451     /* Bind. */
 452     if (bind(fd, (struct sockaddr *) &sin, sizeof sin) < 0) {
 453         error = errno;
 454         VLOG_ERR("%s: bind: %s", target_, strerror(error));
 455         goto exit_close;
 456     }
 457
 458     /* Listen. */
 459     if (listen(fd, 10) < 0) {
 460         error = errno;
 461         VLOG_ERR("%s: listen: %s", target_, strerror(error));
 462         goto exit_close;
 463     }
 464     error = 0;
 465     goto exit;
 466
 467 exit_close:
 468     close(fd);
 469 exit:
 470     free(target);
 471     return error ? -error : fd;
 472 }
 473
 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). */
 477 int
 478 get_null_fd(void)
 479 {
 480     static int null_fd = -1;
 481     if (null_fd < 0) {
 482         null_fd = open("/dev/null", O_RDWR);
 483         if (null_fd < 0) {
 484             int error = errno;
 485             VLOG_ERR("could not open /dev/null: %s", strerror(error));
 486             return -error;
 487         }
 488     }
 489     return null_fd;
 490 }
 491
 492 int
 493 read_fully(int fd, void *p_, size_t size, size_t *bytes_read)
 494 {
 495     uint8_t *p = p_;
 496
 497     *bytes_read = 0;
 498     while (size > 0) {
 499         ssize_t retval = read(fd, p, size);
 500         if (retval > 0) {
 501             *bytes_read += retval;
 502             size -= retval;
 503             p += retval;
 504         } else if (retval == 0) {
 505             return EOF;
 506         } else if (errno != EINTR) {
 507             return errno;
 508         }
 509     }
 510     return 0;
 511 }
 512
 513 int
 514 write_fully(int fd, const void *p_, size_t size, size_t *bytes_written)
 515 {
 516     const uint8_t *p = p_;
 517
 518     *bytes_written = 0;
 519     while (size > 0) {
 520         ssize_t retval = write(fd, p, size);
 521         if (retval > 0) {
 522             *bytes_written += retval;
 523             size -= retval;
 524             p += retval;
 525         } else if (retval == 0) {
 526             VLOG_WARN("write returned 0");
 527             return EPROTO;
 528         } else if (errno != EINTR) {
 529             return errno;
 530         }
 531     }
 532     return 0;
 533 }