lib/socket-util.c

   1 /*
   2  * Copyright (c) 2008, 2009, 2010 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 OVS_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             printf("connect failed with %s\n", strerror(errno));
 253             goto error;
 254         }
 255     }
 256
 257 #ifdef SCM_CREDENTIALS
 258     if (passcred) {
 259         int enable = 1;
 260         if (setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &enable, sizeof(enable))) {
 261             goto error;
 262         }
 263     }
 264 #endif
 265
 266     return fd;
 267
 268 error:
 269     error = errno == EAGAIN ? EPROTO : errno;
 270     if (bind_path) {
 271         fatal_signal_remove_file_to_unlink(bind_path);
 272     }
 273     close(fd);
 274     return -error;
 275 }
 276
 277 int
 278 get_unix_name_len(socklen_t sun_len)
 279 {
 280     return (sun_len >= offsetof(struct sockaddr_un, sun_path)
 281             ? sun_len - offsetof(struct sockaddr_un, sun_path)
 282             : 0);
 283 }
 284
 285 uint32_t
 286 guess_netmask(uint32_t ip)
 287 {
 288     ip = ntohl(ip);
 289     return ((ip >> 31) == 0 ? htonl(0xff000000)   /* Class A */
 290             : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */
 291             : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */
 292             : htonl(0));                          /* ??? */
 293 }
 294
 295 /* Opens a non-blocking IPv4 socket of the specified 'style' and connects to
 296  * 'target', which should be a string in the format "<host>[:<port>]".  <host>
 297  * is required.  If 'default_port' is nonzero then <port> is optional and
 298  * defaults to 'default_port'.
 299  *
 300  * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
 301  *
 302  * On success, returns 0 (indicating connection complete) or EAGAIN (indicating
 303  * connection in progress), in which case the new file descriptor is stored
 304  * into '*fdp'.  On failure, returns a positive errno value other than EAGAIN
 305  * and stores -1 into '*fdp'.
 306  *
 307  * If 'sinp' is non-null, then on success the target address is stored into
 308  * '*sinp'. */
 309 int
 310 inet_open_active(int style, const char *target_, uint16_t default_port,
 311                  struct sockaddr_in *sinp, int *fdp)
 312 {
 313     char *target = xstrdup(target_);
 314     char *save_ptr = NULL;
 315     const char *host_name;
 316     const char *port_string;
 317     struct sockaddr_in sin;
 318     int fd = -1;
 319     int error;
 320
 321     /* Defaults. */
 322     memset(&sin, 0, sizeof sin);
 323     sin.sin_family = AF_INET;
 324     sin.sin_port = htons(default_port);
 325
 326     /* Tokenize. */
 327     host_name = strtok_r(target, ":", &save_ptr);
 328     port_string = strtok_r(NULL, ":", &save_ptr);
 329     if (!host_name) {
 330         ovs_error(0, "%s: bad peer name format", target_);
 331         error = EAFNOSUPPORT;
 332         goto exit;
 333     }
 334
 335     /* Look up IP, port. */
 336     error = lookup_ip(host_name, &sin.sin_addr);
 337     if (error) {
 338         goto exit;
 339     }
 340     if (port_string && atoi(port_string)) {
 341         sin.sin_port = htons(atoi(port_string));
 342     } else if (!default_port) {
 343         VLOG_ERR("%s: port number must be specified", target_);
 344         error = EAFNOSUPPORT;
 345         goto exit;
 346     }
 347
 348     /* Create non-blocking socket. */
 349     fd = socket(AF_INET, style, 0);
 350     if (fd < 0) {
 351         VLOG_ERR("%s: socket: %s", target_, strerror(errno));
 352         error = errno;
 353         goto exit;
 354     }
 355     error = set_nonblocking(fd);
 356     if (error) {
 357         goto exit_close;
 358     }
 359
 360     /* Connect. */
 361     error = connect(fd, (struct sockaddr *) &sin, sizeof sin) == 0 ? 0 : errno;
 362     if (error == EINPROGRESS) {
 363         error = EAGAIN;
 364     } else if (error && error != EAGAIN) {
 365         goto exit_close;
 366     }
 367
 368     /* Success: error is 0 or EAGAIN. */
 369     goto exit;
 370
 371 exit_close:
 372     close(fd);
 373 exit:
 374     if (!error || error == EAGAIN) {
 375         if (sinp) {
 376             *sinp = sin;
 377         }
 378         *fdp = fd;
 379     } else {
 380         *fdp = -1;
 381     }
 382     free(target);
 383     return error;
 384 }
 385
 386 /* Opens a non-blocking IPv4 socket of the specified 'style', binds to
 387  * 'target', and listens for incoming connections.  'target' should be a string
 388  * in the format "[<port>][:<ip>]":
 389  *
 390  *      - If 'default_port' is -1, then <port> is required.  Otherwise, if
 391  *        <port> is omitted, then 'default_port' is used instead.
 392  *
 393  *      - If <port> (or 'default_port', if used) is 0, then no port is bound
 394  *        and the TCP/IP stack will select a port.
 395  *
 396  *      - If <ip> is omitted then the IP address is wildcarded.
 397  *
 398  * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
 399  *
 400  * For TCP, the socket will have SO_REUSEADDR turned on.
 401  *
 402  * On success, returns a non-negative file descriptor.  On failure, returns a
 403  * negative errno value.
 404  *
 405  * If 'sinp' is non-null, then on success the bound address is stored into
 406  * '*sinp'. */
 407 int
 408 inet_open_passive(int style, const char *target_, int default_port,
 409                   struct sockaddr_in *sinp)
 410 {
 411     char *target = xstrdup(target_);
 412     char *string_ptr = target;
 413     struct sockaddr_in sin;
 414     const char *host_name;
 415     const char *port_string;
 416     int fd, error, port;
 417     unsigned int yes  = 1;
 418
 419     /* Address defaults. */
 420     memset(&sin, 0, sizeof sin);
 421     sin.sin_family = AF_INET;
 422     sin.sin_addr.s_addr = htonl(INADDR_ANY);
 423     sin.sin_port = htons(default_port);
 424
 425     /* Parse optional port number. */
 426     port_string = strsep(&string_ptr, ":");
 427     if (port_string && str_to_int(port_string, 10, &port)) {
 428         sin.sin_port = htons(port);
 429     } else if (default_port < 0) {
 430         VLOG_ERR("%s: port number must be specified", target_);
 431         error = EAFNOSUPPORT;
 432         goto exit;
 433     }
 434
 435     /* Parse optional bind IP. */
 436     host_name = strsep(&string_ptr, ":");
 437     if (host_name && host_name[0]) {
 438         error = lookup_ip(host_name, &sin.sin_addr);
 439         if (error) {
 440             goto exit;
 441         }
 442     }
 443
 444     /* Create non-blocking socket, set SO_REUSEADDR. */
 445     fd = socket(AF_INET, style, 0);
 446     if (fd < 0) {
 447         error = errno;
 448         VLOG_ERR("%s: socket: %s", target_, strerror(error));
 449         goto exit;
 450     }
 451     error = set_nonblocking(fd);
 452     if (error) {
 453         goto exit_close;
 454     }
 455     if (style == SOCK_STREAM
 456         && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) {
 457         error = errno;
 458         VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s", target_, strerror(error));
 459         goto exit_close;
 460     }
 461
 462     /* Bind. */
 463     if (bind(fd, (struct sockaddr *) &sin, sizeof sin) < 0) {
 464         error = errno;
 465         VLOG_ERR("%s: bind: %s", target_, strerror(error));
 466         goto exit_close;
 467     }
 468
 469     /* Listen. */
 470     if (listen(fd, 10) < 0) {
 471         error = errno;
 472         VLOG_ERR("%s: listen: %s", target_, strerror(error));
 473         goto exit_close;
 474     }
 475
 476     if (sinp) {
 477         socklen_t sin_len = sizeof sin;
 478         if (getsockname(fd, (struct sockaddr *) &sin, &sin_len) < 0){
 479             error = errno;
 480             VLOG_ERR("%s: getsockname: %s", target_, strerror(error));
 481             goto exit_close;
 482         }
 483         if (sin.sin_family != AF_INET || sin_len != sizeof sin) {
 484             VLOG_ERR("%s: getsockname: invalid socket name", target_);
 485             goto exit_close;
 486         }
 487         *sinp = sin;
 488     }
 489
 490     error = 0;
 491     goto exit;
 492
 493 exit_close:
 494     close(fd);
 495 exit:
 496     free(target);
 497     return error ? -error : fd;
 498 }
 499
 500 /* Returns a readable and writable fd for /dev/null, if successful, otherwise
 501  * a negative errno value.  The caller must not close the returned fd (because
 502  * the same fd will be handed out to subsequent callers). */
 503 int
 504 get_null_fd(void)
 505 {
 506     static int null_fd = -1;
 507     if (null_fd < 0) {
 508         null_fd = open("/dev/null", O_RDWR);
 509         if (null_fd < 0) {
 510             int error = errno;
 511             VLOG_ERR("could not open /dev/null: %s", strerror(error));
 512             return -error;
 513         }
 514     }
 515     return null_fd;
 516 }
 517
 518 int
 519 read_fully(int fd, void *p_, size_t size, size_t *bytes_read)
 520 {
 521     uint8_t *p = p_;
 522
 523     *bytes_read = 0;
 524     while (size > 0) {
 525         ssize_t retval = read(fd, p, size);
 526         if (retval > 0) {
 527             *bytes_read += retval;
 528             size -= retval;
 529             p += retval;
 530         } else if (retval == 0) {
 531             return EOF;
 532         } else if (errno != EINTR) {
 533             return errno;
 534         }
 535     }
 536     return 0;
 537 }
 538
 539 int
 540 write_fully(int fd, const void *p_, size_t size, size_t *bytes_written)
 541 {
 542     const uint8_t *p = p_;
 543
 544     *bytes_written = 0;
 545     while (size > 0) {
 546         ssize_t retval = write(fd, p, size);
 547         if (retval > 0) {
 548             *bytes_written += retval;
 549             size -= retval;
 550             p += retval;
 551         } else if (retval == 0) {
 552             VLOG_WARN("write returned 0");
 553             return EPROTO;
 554         } else if (errno != EINTR) {
 555             return errno;
 556         }
 557     }
 558     return 0;
 559 }
 560
 561 /* Given file name 'file_name', fsyncs the directory in which it is contained.
 562  * Returns 0 if successful, otherwise a positive errno value. */
 563 int
 564 fsync_parent_dir(const char *file_name)
 565 {
 566     int error = 0;
 567     char *dir;
 568     int fd;
 569
 570     dir = dir_name(file_name);
 571     fd = open(dir, O_RDONLY);
 572     if (fd >= 0) {
 573         if (fsync(fd)) {
 574             if (errno == EINVAL || errno == EROFS) {
 575                 /* This directory does not support synchronization.  Not
 576                  * really an error. */
 577             } else {
 578                 error = errno;
 579                 VLOG_ERR("%s: fsync failed (%s)", dir, strerror(error));
 580             }
 581         }
 582         close(fd);
 583     } else {
 584         error = errno;
 585         VLOG_ERR("%s: open failed (%s)", dir, strerror(error));
 586     }
 587     free(dir);
 588
 589     return error;
 590 }