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
  34 #define SOCKET_PATH    "/var/lock/qemu-nbd-%s"
  35
  36 #define NBD_BUFFER_SIZE (1024*1024)
  37
  38 static int verbose;
  39
  40 static void usage(const char *name)
  41 {
  42     printf(
  43 "Usage: %s [OPTIONS] FILE\n"
  44 "QEMU Disk Network Block Device Server\n"
  45 "\n"
  46 "  -p, --port=PORT      port to listen on (default `1024')\n"
  47 "  -o, --offset=OFFSET  offset into the image\n"
  48 "  -b, --bind=IFACE     interface to bind to (default `0.0.0.0')\n"
  49 "  -k, --socket=PATH    path to the unix socket\n"
  50 "                       (default '"SOCKET_PATH"')\n"
  51 "  -r, --read-only      export read-only\n"
  52 "  -P, --partition=NUM  only expose partition NUM\n"
  53 "  -s, --snapshot       use snapshot file\n"
  54 "  -n, --nocache        disable host cache\n"
  55 "  -c, --connect=DEV    connect FILE to the local NBD device DEV\n"
  56 "  -d, --disconnect     disconnect the specified device\n"
  57 "  -e, --shared=NUM     device can be shared by NUM clients (default '1')\n"
  58 "  -t, --persistent     don't exit on the last connection\n"
  59 "  -v, --verbose        display extra debugging information\n"
  60 "  -h, --help           display this help and exit\n"
  61 "  -V, --version        output version information and exit\n"
  62 "\n"
  63 "Report bugs to <anthony@codemonkey.ws>\n"
  64     , name, "DEVICE");
  65 }
  66
  67 static void version(const char *name)
  68 {
  69     printf(
  70 "%s version 0.0.1\n"
  71 "Written by Anthony Liguori.\n"
  72 "\n"
  73 "Copyright (C) 2006 Anthony Liguori <anthony@codemonkey.ws>.\n"
  74 "This is free software; see the source for copying conditions.  There is NO\n"
  75 "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
  76     , name);
  77 }
  78
  79 struct partition_record
  80 {
  81     uint8_t bootable;
  82     uint8_t start_head;
  83     uint32_t start_cylinder;
  84     uint8_t start_sector;
  85     uint8_t system;
  86     uint8_t end_head;
  87     uint8_t end_cylinder;
  88     uint8_t end_sector;
  89     uint32_t start_sector_abs;
  90     uint32_t nb_sectors_abs;
  91 };
  92
  93 static void read_partition(uint8_t *p, struct partition_record *r)
  94 {
  95     r->bootable = p[0];
  96     r->start_head = p[1];
  97     r->start_cylinder = p[3] | ((p[2] << 2) & 0x0300);
  98     r->start_sector = p[2] & 0x3f;
  99     r->system = p[4];
 100     r->end_head = p[5];
 101     r->end_cylinder = p[7] | ((p[6] << 2) & 0x300);
 102     r->end_sector = p[6] & 0x3f;
 103     r->start_sector_abs = p[8] | p[9] << 8 | p[10] << 16 | p[11] << 24;
 104     r->nb_sectors_abs = p[12] | p[13] << 8 | p[14] << 16 | p[15] << 24;
 105 }
 106
 107 static int find_partition(BlockDriverState *bs, int partition,
 108                           off_t *offset, off_t *size)
 109 {
 110     struct partition_record mbr[4];
 111     uint8_t data[512];
 112     int i;
 113     int ext_partnum = 4;
 114
 115     if (bdrv_read(bs, 0, data, 1))
 116         errx(EINVAL, "error while reading");
 117
 118     if (data[510] != 0x55 || data[511] != 0xaa) {
 119         errno = -EINVAL;
 120         return -1;
 121     }
 122
 123     for (i = 0; i < 4; i++) {
 124         read_partition(&data[446 + 16 * i], &mbr[i]);
 125
 126         if (!mbr[i].nb_sectors_abs)
 127             continue;
 128
 129         if (mbr[i].system == 0xF || mbr[i].system == 0x5) {
 130             struct partition_record ext[4];
 131             uint8_t data1[512];
 132             int j;
 133
 134             if (bdrv_read(bs, mbr[i].start_sector_abs, data1, 1))
 135                 errx(EINVAL, "error while reading");
 136
 137             for (j = 0; j < 4; j++) {
 138                 read_partition(&data1[446 + 16 * j], &ext[j]);
 139                 if (!ext[j].nb_sectors_abs)
 140                     continue;
 141
 142                 if ((ext_partnum + j + 1) == partition) {
 143                     *offset = (uint64_t)ext[j].start_sector_abs << 9;
 144                     *size = (uint64_t)ext[j].nb_sectors_abs << 9;
 145                     return 0;
 146                 }
 147             }
 148             ext_partnum += 4;
 149         } else if ((i + 1) == partition) {
 150             *offset = (uint64_t)mbr[i].start_sector_abs << 9;
 151             *size = (uint64_t)mbr[i].nb_sectors_abs << 9;
 152             return 0;
 153         }
 154     }
 155
 156     errno = -ENOENT;
 157     return -1;
 158 }
 159
 160 static void show_parts(const char *device)
 161 {
 162     if (fork() == 0) {
 163         int nbd;
 164
 165         /* linux just needs an open() to trigger
 166          * the partition table update
 167          * but remember to load the module with max_part != 0 :
 168          *     modprobe nbd max_part=63
 169          */
 170         nbd = open(device, O_RDWR);
 171         if (nbd != -1)
 172               close(nbd);
 173         exit(0);
 174     }
 175 }
 176
 177 int main(int argc, char **argv)
 178 {
 179     BlockDriverState *bs;
 180     off_t dev_offset = 0;
 181     off_t offset = 0;
 182     bool readonly = false;
 183     bool disconnect = false;
 184     const char *bindto = "0.0.0.0";
 185     int port = 1024;
 186     struct sockaddr_in addr;
 187     socklen_t addr_len = sizeof(addr);
 188     off_t fd_size;
 189     char *device = NULL;
 190     char *socket = NULL;
 191     char sockpath[128];
 192     const char *sopt = "hVb:o:p:rsnP:c:dvk:e:t";
 193     struct option lopt[] = {
 194         { "help", 0, NULL, 'h' },
 195         { "version", 0, NULL, 'V' },
 196         { "bind", 1, NULL, 'b' },
 197         { "port", 1, NULL, 'p' },
 198         { "socket", 1, NULL, 'k' },
 199         { "offset", 1, NULL, 'o' },
 200         { "read-only", 0, NULL, 'r' },
 201         { "partition", 1, NULL, 'P' },
 202         { "connect", 1, NULL, 'c' },
 203         { "disconnect", 0, NULL, 'd' },
 204         { "snapshot", 0, NULL, 's' },
 205         { "nocache", 0, NULL, 'n' },
 206         { "shared", 1, NULL, 'e' },
 207         { "persistent", 0, NULL, 't' },
 208         { "verbose", 0, NULL, 'v' },
 209         { NULL, 0, NULL, 0 }
 210     };
 211     int ch;
 212     int opt_ind = 0;
 213     int li;
 214     char *end;
 215     int flags = 0;
 216     int partition = -1;
 217     int ret;
 218     int shared = 1;
 219     uint8_t *data;
 220     fd_set fds;
 221     int *sharing_fds;
 222     int fd;
 223     int i;
 224     int nb_fds = 0;
 225     int max_fd;
 226     int persistent = 0;
 227
 228     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 229         switch (ch) {
 230         case 's':
 231             flags |= BDRV_O_SNAPSHOT;
 232             break;
 233         case 'n':
 234             flags |= BDRV_O_NOCACHE;
 235             break;
 236         case 'b':
 237             bindto = optarg;
 238             break;
 239         case 'p':
 240             li = strtol(optarg, &end, 0);
 241             if (*end) {
 242                 errx(EINVAL, "Invalid port `%s'", optarg);
 243             }
 244             if (li < 1 || li > 65535) {
 245                 errx(EINVAL, "Port out of range `%s'", optarg);
 246             }
 247             port = (uint16_t)li;
 248             break;
 249         case 'o':
 250                 dev_offset = strtoll (optarg, &end, 0);
 251             if (*end) {
 252                 errx(EINVAL, "Invalid offset `%s'", optarg);
 253             }
 254             if (dev_offset < 0) {
 255                 errx(EINVAL, "Offset must be positive `%s'", optarg);
 256             }
 257             break;
 258         case 'r':
 259             readonly = true;
 260             break;
 261         case 'P':
 262             partition = strtol(optarg, &end, 0);
 263             if (*end)
 264                 errx(EINVAL, "Invalid partition `%s'", optarg);
 265             if (partition < 1 || partition > 8)
 266                 errx(EINVAL, "Invalid partition %d", partition);
 267             break;
 268         case 'k':
 269             socket = optarg;
 270             if (socket[0] != '/')
 271                 errx(EINVAL, "socket path must be absolute\n");
 272             break;
 273         case 'd':
 274             disconnect = true;
 275             break;
 276         case 'c':
 277             device = optarg;
 278             break;
 279         case 'e':
 280             shared = strtol(optarg, &end, 0);
 281             if (*end) {
 282                 errx(EINVAL, "Invalid shared device number '%s'", optarg);
 283             }
 284             if (shared < 1) {
 285                 errx(EINVAL, "Shared device number must be greater than 0\n");
 286             }
 287             break;
 288         case 't':
 289             persistent = 1;
 290             break;
 291         case 'v':
 292             verbose = 1;
 293             break;
 294         case 'V':
 295             version(argv[0]);
 296             exit(0);
 297             break;
 298         case 'h':
 299             usage(argv[0]);
 300             exit(0);
 301             break;
 302         case '?':
 303             errx(EINVAL, "Try `%s --help' for more information.",
 304                  argv[0]);
 305         }
 306     }
 307
 308     if ((argc - optind) != 1) {
 309         errx(EINVAL, "Invalid number of argument.\n"
 310              "Try `%s --help' for more information.",
 311              argv[0]);
 312     }
 313
 314     if (disconnect) {
 315         fd = open(argv[optind], O_RDWR);
 316         if (fd == -1)
 317             errx(errno, "Cannot open %s", argv[optind]);
 318
 319         nbd_disconnect(fd);
 320
 321         close(fd);
 322
 323         printf("%s disconnected\n", argv[optind]);
 324
 325         return 0;
 326     }
 327
 328     bdrv_init();
 329
 330     bs = bdrv_new("hda");
 331     if (bs == NULL)
 332         return 1;
 333
 334     if (bdrv_open(bs, argv[optind], flags) == -1)
 335         return 1;
 336
 337     fd_size = bs->total_sectors * 512;
 338
 339     if (partition != -1 &&
 340         find_partition(bs, partition, &dev_offset, &fd_size))
 341         errx(errno, "Could not find partition %d", partition);
 342
 343     if (device) {
 344         pid_t pid;
 345         int sock;
 346
 347         if (!verbose) {
 348             /* detach client and server */
 349             if (daemon(0, 0) == -1) {
 350                 errx(errno, "Failed to daemonize");
 351             }
 352         }
 353
 354         if (socket == NULL) {
 355             sprintf(sockpath, SOCKET_PATH, basename(device));
 356             socket = sockpath;
 357         }
 358
 359         pid = fork();
 360         if (pid < 0)
 361             return 1;
 362         if (pid != 0) {
 363             off_t size;
 364             size_t blocksize;
 365
 366             ret = 0;
 367             bdrv_close(bs);
 368
 369             do {
 370                 sock = unix_socket_outgoing(socket);
 371                 if (sock == -1) {
 372                     if (errno != ENOENT && errno != ECONNREFUSED)
 373                         goto out;
 374                     sleep(1);   /* wait children */
 375                 }
 376             } while (sock == -1);
 377
 378             fd = open(device, O_RDWR);
 379             if (fd == -1) {
 380                 ret = 1;
 381                 goto out;
 382             }
 383
 384             ret = nbd_receive_negotiate(sock, &size, &blocksize);
 385             if (ret == -1) {
 386                 ret = 1;
 387                 goto out;
 388             }
 389
 390             ret = nbd_init(fd, sock, size, blocksize);
 391             if (ret == -1) {
 392                 ret = 1;
 393                 goto out;
 394             }
 395
 396             printf("NBD device %s is now connected to file %s\n",
 397                     device, argv[optind]);
 398
 399             /* update partition table */
 400
 401             show_parts(device);
 402
 403             nbd_client(fd, sock);
 404             close(fd);
 405  out:
 406             kill(pid, SIGTERM);
 407             unlink(socket);
 408
 409             return ret;
 410         }
 411         /* children */
 412     }
 413
 414     sharing_fds = qemu_malloc((shared + 1) * sizeof(int));
 415
 416     if (socket) {
 417         sharing_fds[0] = unix_socket_incoming(socket);
 418     } else {
 419         sharing_fds[0] = tcp_socket_incoming(bindto, port);
 420     }
 421
 422     if (sharing_fds[0] == -1)
 423         return 1;
 424     max_fd = sharing_fds[0];
 425     nb_fds++;
 426
 427     data = qemu_memalign(512, NBD_BUFFER_SIZE);
 428     if (data == NULL)
 429         errx(ENOMEM, "Cannot allocate data buffer");
 430
 431     do {
 432
 433         FD_ZERO(&fds);
 434         for (i = 0; i < nb_fds; i++)
 435             FD_SET(sharing_fds[i], &fds);
 436
 437         ret = select(max_fd + 1, &fds, NULL, NULL, NULL);
 438         if (ret == -1)
 439             break;
 440
 441         if (FD_ISSET(sharing_fds[0], &fds))
 442             ret--;
 443         for (i = 1; i < nb_fds && ret; i++) {
 444             if (FD_ISSET(sharing_fds[i], &fds)) {
 445                 if (nbd_trip(bs, sharing_fds[i], fd_size, dev_offset,
 446                     &offset, readonly, data, NBD_BUFFER_SIZE) != 0) {
 447                     close(sharing_fds[i]);
 448                     nb_fds--;
 449                     sharing_fds[i] = sharing_fds[nb_fds];
 450                     i--;
 451                 }
 452                 ret--;
 453             }
 454         }
 455         /* new connection ? */
 456         if (FD_ISSET(sharing_fds[0], &fds)) {
 457             if (nb_fds < shared + 1) {
 458                 sharing_fds[nb_fds] = accept(sharing_fds[0],
 459                                              (struct sockaddr *)&addr,
 460                                              &addr_len);
 461                 if (sharing_fds[nb_fds] != -1 &&
 462                     nbd_negotiate(sharing_fds[nb_fds], fd_size) != -1) {
 463                         if (sharing_fds[nb_fds] > max_fd)
 464                             max_fd = sharing_fds[nb_fds];
 465                         nb_fds++;
 466                 }
 467             }
 468         }
 469     } while (persistent || nb_fds > 1);
 470     qemu_free(data);
 471
 472     close(sharing_fds[0]);
 473     bdrv_close(bs);
 474     qemu_free(sharing_fds);
 475     if (socket)
 476         unlink(socket);
 477
 478     return 0;
 479 }