qemu-nbd.c

   1 /*
   2  *  Copyright (C) 2005  Anthony Liguori <anthony@codemonkey.ws>
   3  *
   4  *  Network Block Device
   5  *
   6  *  This program is free software; you can redistribute it and/or modify
   7  *  it under the terms of the GNU General Public License as published by
   8  *  the Free Software Foundation; under version 2 of the License.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  You should have received a copy of the GNU General Public License
  16  *  along with this program; if not, see <http://www.gnu.org/licenses/>.
  17  */
  18
  19 #include "qemu-common.h"
  20 #include "block_int.h"
  21 #include "nbd.h"
  22
  23 #include <stdarg.h>
  24 #include <stdio.h>
  25 #include <getopt.h>
  26 #include <err.h>
  27 #include <sys/types.h>
  28 #include <sys/socket.h>
  29 #include <netinet/in.h>
  30 #include <netinet/tcp.h>
  31 #include <arpa/inet.h>
  32 #include <signal.h>
  33 #include <libgen.h>
  34 #include <pthread.h>
  35
  36 #define SOCKET_PATH    "/var/lock/qemu-nbd-%s"
  37
  38 #define NBD_BUFFER_SIZE (1024*1024)
  39
  40 static int sigterm_wfd;
  41 static int verbose;
  42 static char *device;
  43 static char *srcpath;
  44 static char *sockpath;
  45
  46 static void usage(const char *name)
  47 {
  48     printf(
  49 "Usage: %s [OPTIONS] FILE\n"
  50 "QEMU Disk Network Block Device Server\n"
  51 "\n"
  52 "  -p, --port=PORT      port to listen on (default `%d')\n"
  53 "  -o, --offset=OFFSET  offset into the image\n"
  54 "  -b, --bind=IFACE     interface to bind to (default `0.0.0.0')\n"
  55 "  -k, --socket=PATH    path to the unix socket\n"
  56 "                       (default '"SOCKET_PATH"')\n"
  57 "  -r, --read-only      export read-only\n"
  58 "  -P, --partition=NUM  only expose partition NUM\n"
  59 "  -s, --snapshot       use snapshot file\n"
  60 "  -n, --nocache        disable host cache\n"
  61 "  -c, --connect=DEV    connect FILE to the local NBD device DEV\n"
  62 "  -d, --disconnect     disconnect the specified device\n"
  63 "  -e, --shared=NUM     device can be shared by NUM clients (default '1')\n"
  64 "  -t, --persistent     don't exit on the last connection\n"
  65 "  -v, --verbose        display extra debugging information\n"
  66 "  -h, --help           display this help and exit\n"
  67 "  -V, --version        output version information and exit\n"
  68 "\n"
  69 "Report bugs to <anthony@codemonkey.ws>\n"
  70     , name, NBD_DEFAULT_PORT, "DEVICE");
  71 }
  72
  73 static void version(const char *name)
  74 {
  75     printf(
  76 "%s version 0.0.1\n"
  77 "Written by Anthony Liguori.\n"
  78 "\n"
  79 "Copyright (C) 2006 Anthony Liguori <anthony@codemonkey.ws>.\n"
  80 "This is free software; see the source for copying conditions.  There is NO\n"
  81 "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
  82     , name);
  83 }
  84
  85 struct partition_record
  86 {
  87     uint8_t bootable;
  88     uint8_t start_head;
  89     uint32_t start_cylinder;
  90     uint8_t start_sector;
  91     uint8_t system;
  92     uint8_t end_head;
  93     uint8_t end_cylinder;
  94     uint8_t end_sector;
  95     uint32_t start_sector_abs;
  96     uint32_t nb_sectors_abs;
  97 };
  98
  99 static void read_partition(uint8_t *p, struct partition_record *r)
 100 {
 101     r->bootable = p[0];
 102     r->start_head = p[1];
 103     r->start_cylinder = p[3] | ((p[2] << 2) & 0x0300);
 104     r->start_sector = p[2] & 0x3f;
 105     r->system = p[4];
 106     r->end_head = p[5];
 107     r->end_cylinder = p[7] | ((p[6] << 2) & 0x300);
 108     r->end_sector = p[6] & 0x3f;
 109     r->start_sector_abs = p[8] | p[9] << 8 | p[10] << 16 | p[11] << 24;
 110     r->nb_sectors_abs = p[12] | p[13] << 8 | p[14] << 16 | p[15] << 24;
 111 }
 112
 113 static int find_partition(BlockDriverState *bs, int partition,
 114                           off_t *offset, off_t *size)
 115 {
 116     struct partition_record mbr[4];
 117     uint8_t data[512];
 118     int i;
 119     int ext_partnum = 4;
 120     int ret;
 121
 122     if ((ret = bdrv_read(bs, 0, data, 1)) < 0) {
 123         errno = -ret;
 124         err(EXIT_FAILURE, "error while reading");
 125     }
 126
 127     if (data[510] != 0x55 || data[511] != 0xaa) {
 128         errno = -EINVAL;
 129         return -1;
 130     }
 131
 132     for (i = 0; i < 4; i++) {
 133         read_partition(&data[446 + 16 * i], &mbr[i]);
 134
 135         if (!mbr[i].nb_sectors_abs)
 136             continue;
 137
 138         if (mbr[i].system == 0xF || mbr[i].system == 0x5) {
 139             struct partition_record ext[4];
 140             uint8_t data1[512];
 141             int j;
 142
 143             if ((ret = bdrv_read(bs, mbr[i].start_sector_abs, data1, 1)) < 0) {
 144                 errno = -ret;
 145                 err(EXIT_FAILURE, "error while reading");
 146             }
 147
 148             for (j = 0; j < 4; j++) {
 149                 read_partition(&data1[446 + 16 * j], &ext[j]);
 150                 if (!ext[j].nb_sectors_abs)
 151                     continue;
 152
 153                 if ((ext_partnum + j + 1) == partition) {
 154                     *offset = (uint64_t)ext[j].start_sector_abs << 9;
 155                     *size = (uint64_t)ext[j].nb_sectors_abs << 9;
 156                     return 0;
 157                 }
 158             }
 159             ext_partnum += 4;
 160         } else if ((i + 1) == partition) {
 161             *offset = (uint64_t)mbr[i].start_sector_abs << 9;
 162             *size = (uint64_t)mbr[i].nb_sectors_abs << 9;
 163             return 0;
 164         }
 165     }
 166
 167     errno = -ENOENT;
 168     return -1;
 169 }
 170
 171 static void termsig_handler(int signum)
 172 {
 173     static int sigterm_reported;
 174     if (!sigterm_reported) {
 175         sigterm_reported = (write(sigterm_wfd, "", 1) == 1);
 176     }
 177 }
 178
 179 static void *show_parts(void *arg)
 180 {
 181     int nbd;
 182
 183     /* linux just needs an open() to trigger
 184      * the partition table update
 185      * but remember to load the module with max_part != 0 :
 186      *     modprobe nbd max_part=63
 187      */
 188     nbd = open(device, O_RDWR);
 189     if (nbd != -1) {
 190         close(nbd);
 191     }
 192     return NULL;
 193 }
 194
 195 static void *nbd_client_thread(void *arg)
 196 {
 197     int fd = *(int *)arg;
 198     off_t size;
 199     size_t blocksize;
 200     uint32_t nbdflags;
 201     int sock;
 202     int ret;
 203     pthread_t show_parts_thread;
 204
 205     do {
 206         sock = unix_socket_outgoing(sockpath);
 207         if (sock == -1) {
 208             if (errno != ENOENT && errno != ECONNREFUSED) {
 209                 goto out;
 210             }
 211             sleep(1);  /* wait parent */
 212         }
 213     } while (sock == -1);
 214
 215     ret = nbd_receive_negotiate(sock, NULL, &nbdflags,
 216                                 &size, &blocksize);
 217     if (ret == -1) {
 218         goto out;
 219     }
 220
 221     ret = nbd_init(fd, sock, nbdflags, size, blocksize);
 222     if (ret == -1) {
 223         goto out;
 224     }
 225
 226     /* update partition table */
 227     pthread_create(&show_parts_thread, NULL, show_parts, NULL);
 228
 229     if (verbose) {
 230         fprintf(stderr, "NBD device %s is now connected to %s\n",
 231                 device, srcpath);
 232     } else {
 233         /* Close stderr so that the qemu-nbd process exits.  */
 234         dup2(STDOUT_FILENO, STDERR_FILENO);
 235     }
 236
 237     ret = nbd_client(fd);
 238     if (ret) {
 239         goto out;
 240     }
 241     close(fd);
 242     kill(getpid(), SIGTERM);
 243     return (void *) EXIT_SUCCESS;
 244
 245 out:
 246     kill(getpid(), SIGTERM);
 247     return (void *) EXIT_FAILURE;
 248 }
 249
 250 int main(int argc, char **argv)
 251 {
 252     BlockDriverState *bs;
 253     off_t dev_offset = 0;
 254     off_t offset = 0;
 255     uint32_t nbdflags = 0;
 256     bool disconnect = false;
 257     const char *bindto = "0.0.0.0";
 258     int port = NBD_DEFAULT_PORT;
 259     struct sockaddr_in addr;
 260     socklen_t addr_len = sizeof(addr);
 261     off_t fd_size;
 262     const char *sopt = "hVb:o:p:rsnP:c:dvk:e:t";
 263     struct option lopt[] = {
 264         { "help", 0, NULL, 'h' },
 265         { "version", 0, NULL, 'V' },
 266         { "bind", 1, NULL, 'b' },
 267         { "port", 1, NULL, 'p' },
 268         { "socket", 1, NULL, 'k' },
 269         { "offset", 1, NULL, 'o' },
 270         { "read-only", 0, NULL, 'r' },
 271         { "partition", 1, NULL, 'P' },
 272         { "connect", 1, NULL, 'c' },
 273         { "disconnect", 0, NULL, 'd' },
 274         { "snapshot", 0, NULL, 's' },
 275         { "nocache", 0, NULL, 'n' },
 276         { "shared", 1, NULL, 'e' },
 277         { "persistent", 0, NULL, 't' },
 278         { "verbose", 0, NULL, 'v' },
 279         { NULL, 0, NULL, 0 }
 280     };
 281     int ch;
 282     int opt_ind = 0;
 283     int li;
 284     char *end;
 285     int flags = BDRV_O_RDWR;
 286     int partition = -1;
 287     int ret;
 288     int shared = 1;
 289     uint8_t *data;
 290     fd_set fds;
 291     int *sharing_fds;
 292     int fd;
 293     int i;
 294     int nb_fds = 0;
 295     int max_fd;
 296     int persistent = 0;
 297     pthread_t client_thread;
 298
 299     /* The client thread uses SIGTERM to interrupt the server.  A signal
 300      * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
 301      */
 302     struct sigaction sa_sigterm;
 303     int sigterm_fd[2];
 304     if (qemu_pipe(sigterm_fd) == -1) {
 305         err(EXIT_FAILURE, "Error setting up communication pipe");
 306     }
 307
 308     sigterm_wfd = sigterm_fd[1];
 309     memset(&sa_sigterm, 0, sizeof(sa_sigterm));
 310     sa_sigterm.sa_handler = termsig_handler;
 311     sigaction(SIGTERM, &sa_sigterm, NULL);
 312
 313     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 314         switch (ch) {
 315         case 's':
 316             flags |= BDRV_O_SNAPSHOT;
 317             break;
 318         case 'n':
 319             flags |= BDRV_O_NOCACHE | BDRV_O_CACHE_WB;
 320             break;
 321         case 'b':
 322             bindto = optarg;
 323             break;
 324         case 'p':
 325             li = strtol(optarg, &end, 0);
 326             if (*end) {
 327                 errx(EXIT_FAILURE, "Invalid port `%s'", optarg);
 328             }
 329             if (li < 1 || li > 65535) {
 330                 errx(EXIT_FAILURE, "Port out of range `%s'", optarg);
 331             }
 332             port = (uint16_t)li;
 333             break;
 334         case 'o':
 335                 dev_offset = strtoll (optarg, &end, 0);
 336             if (*end) {
 337                 errx(EXIT_FAILURE, "Invalid offset `%s'", optarg);
 338             }
 339             if (dev_offset < 0) {
 340                 errx(EXIT_FAILURE, "Offset must be positive `%s'", optarg);
 341             }
 342             break;
 343         case 'r':
 344             nbdflags |= NBD_FLAG_READ_ONLY;
 345             flags &= ~BDRV_O_RDWR;
 346             break;
 347         case 'P':
 348             partition = strtol(optarg, &end, 0);
 349             if (*end)
 350                 errx(EXIT_FAILURE, "Invalid partition `%s'", optarg);
 351             if (partition < 1 || partition > 8)
 352                 errx(EXIT_FAILURE, "Invalid partition %d", partition);
 353             break;
 354         case 'k':
 355             sockpath = optarg;
 356             if (sockpath[0] != '/')
 357                 errx(EXIT_FAILURE, "socket path must be absolute\n");
 358             break;
 359         case 'd':
 360             disconnect = true;
 361             break;
 362         case 'c':
 363             device = optarg;
 364             break;
 365         case 'e':
 366             shared = strtol(optarg, &end, 0);
 367             if (*end) {
 368                 errx(EXIT_FAILURE, "Invalid shared device number '%s'", optarg);
 369             }
 370             if (shared < 1) {
 371                 errx(EXIT_FAILURE, "Shared device number must be greater than 0\n");
 372             }
 373             break;
 374         case 't':
 375             persistent = 1;
 376             break;
 377         case 'v':
 378             verbose = 1;
 379             break;
 380         case 'V':
 381             version(argv[0]);
 382             exit(0);
 383             break;
 384         case 'h':
 385             usage(argv[0]);
 386             exit(0);
 387             break;
 388         case '?':
 389             errx(EXIT_FAILURE, "Try `%s --help' for more information.",
 390                  argv[0]);
 391         }
 392     }
 393
 394     if ((argc - optind) != 1) {
 395         errx(EXIT_FAILURE, "Invalid number of argument.\n"
 396              "Try `%s --help' for more information.",
 397              argv[0]);
 398     }
 399
 400     if (disconnect) {
 401         fd = open(argv[optind], O_RDWR);
 402         if (fd == -1)
 403             err(EXIT_FAILURE, "Cannot open %s", argv[optind]);
 404
 405         nbd_disconnect(fd);
 406
 407         close(fd);
 408
 409         printf("%s disconnected\n", argv[optind]);
 410
 411         return 0;
 412     }
 413
 414     if (device && !verbose) {
 415         int stderr_fd[2];
 416         pid_t pid;
 417         int ret;
 418
 419         if (qemu_pipe(stderr_fd) == -1) {
 420             err(EXIT_FAILURE, "Error setting up communication pipe");
 421         }
 422
 423         /* Now daemonize, but keep a communication channel open to
 424          * print errors and exit with the proper status code.
 425          */
 426         pid = fork();
 427         if (pid == 0) {
 428             close(stderr_fd[0]);
 429             ret = qemu_daemon(0, 0);
 430
 431             /* Temporarily redirect stderr to the parent's pipe...  */
 432             dup2(stderr_fd[1], STDERR_FILENO);
 433             if (ret == -1) {
 434                 err(EXIT_FAILURE, "Failed to daemonize");
 435             }
 436
 437             /* ... close the descriptor we inherited and go on.  */
 438             close(stderr_fd[1]);
 439         } else {
 440             bool errors = false;
 441             char *buf;
 442
 443             /* In the parent.  Print error messages from the child until
 444              * it closes the pipe.
 445              */
 446             close(stderr_fd[1]);
 447             buf = g_malloc(1024);
 448             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
 449                 errors = true;
 450                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
 451                 if (ret == -1) {
 452                     exit(EXIT_FAILURE);
 453                 }
 454             }
 455             if (ret == -1) {
 456                 err(EXIT_FAILURE, "Cannot read from daemon");
 457             }
 458
 459             /* Usually the daemon should not print any message.
 460              * Exit with zero status in that case.
 461              */
 462             exit(errors);
 463         }
 464     }
 465
 466     bdrv_init();
 467
 468     bs = bdrv_new("hda");
 469
 470     srcpath = argv[optind];
 471     if ((ret = bdrv_open(bs, srcpath, flags, NULL)) < 0) {
 472         errno = -ret;
 473         err(EXIT_FAILURE, "Failed to bdrv_open '%s'", srcpath);
 474     }
 475
 476     fd_size = bs->total_sectors * 512;
 477
 478     if (partition != -1 &&
 479         find_partition(bs, partition, &dev_offset, &fd_size))
 480         err(EXIT_FAILURE, "Could not find partition %d", partition);
 481
 482     if (device) {
 483         int ret;
 484
 485         /* Open before spawning new threads.  In the future, we may
 486          * drop privileges after opening.
 487          */
 488         fd = open(device, O_RDWR);
 489         if (fd == -1) {
 490             err(EXIT_FAILURE, "Failed to open %s", device);
 491         }
 492
 493         if (sockpath == NULL) {
 494             sockpath = g_malloc(128);
 495             snprintf(sockpath, 128, SOCKET_PATH, basename(device));
 496         }
 497
 498         ret = pthread_create(&client_thread, NULL, nbd_client_thread, &fd);
 499         if (ret != 0) {
 500             errx(EXIT_FAILURE, "Failed to create client thread: %s",
 501                  strerror(ret));
 502         }
 503     } else {
 504         /* Shut up GCC warnings.  */
 505         memset(&client_thread, 0, sizeof(client_thread));
 506     }
 507
 508     sharing_fds = g_malloc((shared + 1) * sizeof(int));
 509
 510     if (sockpath) {
 511         sharing_fds[0] = unix_socket_incoming(sockpath);
 512     } else {
 513         sharing_fds[0] = tcp_socket_incoming(bindto, port);
 514     }
 515
 516     if (sharing_fds[0] == -1)
 517         return 1;
 518     max_fd = sharing_fds[0];
 519     nb_fds++;
 520
 521     data = qemu_blockalign(bs, NBD_BUFFER_SIZE);
 522     if (data == NULL) {
 523         errx(EXIT_FAILURE, "Cannot allocate data buffer");
 524     }
 525
 526     do {
 527         FD_ZERO(&fds);
 528         FD_SET(sigterm_fd[0], &fds);
 529         for (i = 0; i < nb_fds; i++)
 530             FD_SET(sharing_fds[i], &fds);
 531
 532         do {
 533             ret = select(max_fd + 1, &fds, NULL, NULL, NULL);
 534         } while (ret == -1 && errno == EINTR);
 535         if (ret == -1 || FD_ISSET(sigterm_fd[0], &fds)) {
 536             break;
 537         }
 538
 539         if (FD_ISSET(sharing_fds[0], &fds))
 540             ret--;
 541         for (i = 1; i < nb_fds && ret; i++) {
 542             if (FD_ISSET(sharing_fds[i], &fds)) {
 543                 if (nbd_trip(bs, sharing_fds[i], fd_size, dev_offset,
 544                     &offset, nbdflags, data, NBD_BUFFER_SIZE) != 0) {
 545                     close(sharing_fds[i]);
 546                     nb_fds--;
 547                     sharing_fds[i] = sharing_fds[nb_fds];
 548                     i--;
 549                 }
 550                 ret--;
 551             }
 552         }
 553         /* new connection ? */
 554         if (FD_ISSET(sharing_fds[0], &fds)) {
 555             if (nb_fds < shared + 1) {
 556                 sharing_fds[nb_fds] = accept(sharing_fds[0],
 557                                              (struct sockaddr *)&addr,
 558                                              &addr_len);
 559                 if (sharing_fds[nb_fds] != -1 &&
 560                     nbd_negotiate(sharing_fds[nb_fds], fd_size, nbdflags) != -1) {
 561                         if (sharing_fds[nb_fds] > max_fd)
 562                             max_fd = sharing_fds[nb_fds];
 563                         nb_fds++;
 564                 }
 565             }
 566         }
 567     } while (persistent || nb_fds > 1);
 568     qemu_vfree(data);
 569
 570     close(sharing_fds[0]);
 571     bdrv_close(bs);
 572     g_free(sharing_fds);
 573     if (sockpath) {
 574         unlink(sockpath);
 575     }
 576
 577     if (device) {
 578         void *ret;
 579         pthread_join(client_thread, &ret);
 580         exit(ret != NULL);
 581     } else {
 582         exit(EXIT_SUCCESS);
 583     }
 584 }