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, write to the Free Software
  17  *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  18  */
  19
  20 #include <qemu-common.h>
  21 #include "block_int.h"
  22 #include "nbd.h"
  23
  24 #include <stdarg.h>
  25 #include <stdio.h>
  26 #include <getopt.h>
  27 #include <err.h>
  28 #include <sys/types.h>
  29 #include <sys/socket.h>
  30 #include <netinet/in.h>
  31 #include <netinet/tcp.h>
  32 #include <arpa/inet.h>
  33 #include <signal.h>
  34
  35 #define SOCKET_PATH    "/var/lock/qemu-nbd-%s"
  36
  37 #define NBD_BUFFER_SIZE (1024*1024)
  38
  39 int verbose;
  40
  41 static void usage(const char *name)
  42 {
  43     printf(
  44 "Usage: %s [OPTIONS] FILE\n"
  45 "QEMU Disk Network Block Device Server\n"
  46 "\n"
  47 "  -p, --port=PORT      port to listen on (default `1024')\n"
  48 "  -o, --offset=OFFSET  offset into the image\n"
  49 "  -b, --bind=IFACE     interface to bind to (default `0.0.0.0')\n"
  50 "  -k, --socket=PATH    path to the unix socket\n"
  51 "                       (default '"SOCKET_PATH"')\n"
  52 "  -r, --read-only      export read-only\n"
  53 "  -P, --partition=NUM  only expose partition NUM\n"
  54 "  -s, --snapshot       use snapshot file\n"
  55 "  -n, --nocache        disable host cache\n"
  56 "  -c, --connect=DEV    connect FILE to the local NBD device DEV\n"
  57 "  -d, --disconnect     disconnect the specified device\n"
  58 "  -e, --shared=NUM     device can be shared by NUM clients (default '1')\n"
  59 "  -t, --persistent     don't exit on the last connection\n"
  60 "  -v, --verbose        display extra debugging information\n"
  61 "  -h, --help           display this help and exit\n"
  62 "  -V, --version        output version information and exit\n"
  63 "\n"
  64 "Report bugs to <anthony@codemonkey.ws>\n"
  65     , name, "DEVICE");
  66 }
  67
  68 static void version(const char *name)
  69 {
  70     printf(
  71 "%s version 0.0.1\n"
  72 "Written by Anthony Liguori.\n"
  73 "\n"
  74 "Copyright (C) 2006 Anthony Liguori <anthony@codemonkey.ws>.\n"
  75 "This is free software; see the source for copying conditions.  There is NO\n"
  76 "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
  77     , name);
  78 }
  79
  80 struct partition_record
  81 {
  82     uint8_t bootable;
  83     uint8_t start_head;
  84     uint32_t start_cylinder;
  85     uint8_t start_sector;
  86     uint8_t system;
  87     uint8_t end_head;
  88     uint8_t end_cylinder;
  89     uint8_t end_sector;
  90     uint32_t start_sector_abs;
  91     uint32_t nb_sectors_abs;
  92 };
  93
  94 static void read_partition(uint8_t *p, struct partition_record *r)
  95 {
  96     r->bootable = p[0];
  97     r->start_head = p[1];
  98     r->start_cylinder = p[3] | ((p[2] << 2) & 0x0300);
  99     r->start_sector = p[2] & 0x3f;
 100     r->system = p[4];
 101     r->end_head = p[5];
 102     r->end_cylinder = p[7] | ((p[6] << 2) & 0x300);
 103     r->end_sector = p[6] & 0x3f;
 104     r->start_sector_abs = p[8] | p[9] << 8 | p[10] << 16 | p[11] << 24;
 105     r->nb_sectors_abs = p[12] | p[13] << 8 | p[14] << 16 | p[15] << 24;
 106 }
 107
 108 static int find_partition(BlockDriverState *bs, int partition,
 109                           off_t *offset, off_t *size)
 110 {
 111     struct partition_record mbr[4];
 112     uint8_t data[512];
 113     int i;
 114     int ext_partnum = 4;
 115
 116     if (bdrv_read(bs, 0, data, 1))
 117         errx(EINVAL, "error while reading");
 118
 119     if (data[510] != 0x55 || data[511] != 0xaa) {
 120         errno = -EINVAL;
 121         return -1;
 122     }
 123
 124     for (i = 0; i < 4; i++) {
 125         read_partition(&data[446 + 16 * i], &mbr[i]);
 126
 127         if (!mbr[i].nb_sectors_abs)
 128             continue;
 129
 130         if (mbr[i].system == 0xF || mbr[i].system == 0x5) {
 131             struct partition_record ext[4];
 132             uint8_t data1[512];
 133             int j;
 134
 135             if (bdrv_read(bs, mbr[i].start_sector_abs, data1, 1))
 136                 errx(EINVAL, "error while reading");
 137
 138             for (j = 0; j < 4; j++) {
 139                 read_partition(&data1[446 + 16 * j], &ext[j]);
 140                 if (!ext[j].nb_sectors_abs)
 141                     continue;
 142
 143                 if ((ext_partnum + j + 1) == partition) {
 144                     *offset = (uint64_t)ext[j].start_sector_abs << 9;
 145                     *size = (uint64_t)ext[j].nb_sectors_abs << 9;
 146                     return 0;
 147                 }
 148             }
 149             ext_partnum += 4;
 150         } else if ((i + 1) == partition) {
 151             *offset = (uint64_t)mbr[i].start_sector_abs << 9;
 152             *size = (uint64_t)mbr[i].nb_sectors_abs << 9;
 153             return 0;
 154         }
 155     }
 156
 157     errno = -ENOENT;
 158     return -1;
 159 }
 160
 161 static void show_parts(const char *device)
 162 {
 163     if (fork() == 0) {
 164         int nbd;
 165
 166         /* linux just needs an open() to trigger
 167          * the partition table update
 168          * but remember to load the module with max_part != 0 :
 169          *     modprobe nbd max_part=63
 170          */
 171         nbd = open(device, O_RDWR);
 172         if (nbd != -1)
 173               close(nbd);
 174         exit(0);
 175     }
 176 }
 177
 178 int main(int argc, char **argv)
 179 {
 180     BlockDriverState *bs;
 181     off_t dev_offset = 0;
 182     off_t offset = 0;
 183     bool readonly = false;
 184     bool disconnect = false;
 185     const char *bindto = "0.0.0.0";
 186     int port = 1024;
 187     struct sockaddr_in addr;
 188     socklen_t addr_len = sizeof(addr);
 189     off_t fd_size;
 190     char *device = NULL;
 191     char *socket = NULL;
 192     char sockpath[128];
 193     const char *sopt = "hVbo:p:rsnP:c:dvk:e:t";
 194     struct option lopt[] = {
 195         { "help", 0, 0, 'h' },
 196         { "version", 0, 0, 'V' },
 197         { "bind", 1, 0, 'b' },
 198         { "port", 1, 0, 'p' },
 199         { "socket", 1, 0, 'k' },
 200         { "offset", 1, 0, 'o' },
 201         { "read-only", 0, 0, 'r' },
 202         { "partition", 1, 0, 'P' },
 203         { "connect", 1, 0, 'c' },
 204         { "disconnect", 0, 0, 'd' },
 205         { "snapshot", 0, 0, 's' },
 206         { "nocache", 0, 0, 'n' },
 207         { "shared", 1, 0, 'e' },
 208         { "persistent", 0, 0, 't' },
 209         { "verbose", 0, 0, 'v' },
 210         { NULL, 0, 0, 0 }
 211     };
 212     int ch;
 213     int opt_ind = 0;
 214     int li;
 215     char *end;
 216     int flags = 0;
 217     int partition = -1;
 218     int ret;
 219     int shared = 1;
 220     uint8_t *data;
 221     fd_set fds;
 222     int *sharing_fds;
 223     int fd;
 224     int i;
 225     int nb_fds = 0;
 226     int max_fd;
 227     int persistent = 0;
 228
 229     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 230         switch (ch) {
 231         case 's':
 232             flags |= BDRV_O_SNAPSHOT;
 233             break;
 234         case 'n':
 235             flags |= BDRV_O_DIRECT;
 236             break;
 237         case 'b':
 238             bindto = optarg;
 239             break;
 240         case 'p':
 241             li = strtol(optarg, &end, 0);
 242             if (*end) {
 243                 errx(EINVAL, "Invalid port `%s'", optarg);
 244             }
 245             if (li < 1 || li > 65535) {
 246                 errx(EINVAL, "Port out of range `%s'", optarg);
 247             }
 248             port = (uint16_t)li;
 249             break;
 250         case 'o':
 251                 dev_offset = strtoll (optarg, &end, 0);
 252             if (*end) {
 253                 errx(EINVAL, "Invalid offset `%s'", optarg);
 254             }
 255             if (dev_offset < 0) {
 256                 errx(EINVAL, "Offset must be positive `%s'", optarg);
 257             }
 258             break;
 259         case 'r':
 260             readonly = true;
 261             break;
 262         case 'P':
 263             partition = strtol(optarg, &end, 0);
 264             if (*end)
 265                 errx(EINVAL, "Invalid partition `%s'", optarg);
 266             if (partition < 1 || partition > 8)
 267                 errx(EINVAL, "Invalid partition %d", partition);
 268             break;
 269         case 'k':
 270             socket = optarg;
 271             if (socket[0] != '/')
 272                 errx(EINVAL, "socket path must be absolute\n");
 273             break;
 274         case 'd':
 275             disconnect = true;
 276             break;
 277         case 'c':
 278             device = optarg;
 279             break;
 280         case 'e':
 281             shared = strtol(optarg, &end, 0);
 282             if (*end) {
 283                 errx(EINVAL, "Invalid shared device number '%s'", optarg);
 284             }
 285             if (shared < 1) {
 286                 errx(EINVAL, "Shared device number must be greater than 0\n");
 287             }
 288             break;
 289         case 't':
 290             persistent = 1;
 291             break;
 292         case 'v':
 293             verbose = 1;
 294             break;
 295         case 'V':
 296             version(argv[0]);
 297             exit(0);
 298             break;
 299         case 'h':
 300             usage(argv[0]);
 301             exit(0);
 302             break;
 303         case '?':
 304             errx(EINVAL, "Try `%s --help' for more information.",
 305                  argv[0]);
 306         }
 307     }
 308
 309     if ((argc - optind) != 1) {
 310         errx(EINVAL, "Invalid number of argument.\n"
 311              "Try `%s --help' for more information.",
 312              argv[0]);
 313     }
 314
 315     if (disconnect) {
 316         fd = open(argv[optind], O_RDWR);
 317         if (fd == -1)
 318             errx(errno, "Cannot open %s", argv[optind]);
 319
 320         nbd_disconnect(fd);
 321
 322         close(fd);
 323
 324         printf("%s disconnected\n", argv[optind]);
 325
 326         return 0;
 327     }
 328
 329     bdrv_init();
 330
 331     bs = bdrv_new("hda");
 332     if (bs == NULL)
 333         return 1;
 334
 335     if (bdrv_open(bs, argv[optind], flags) == -1)
 336         return 1;
 337
 338     fd_size = bs->total_sectors * 512;
 339
 340     if (partition != -1 &&
 341         find_partition(bs, partition, &dev_offset, &fd_size))
 342         errx(errno, "Could not find partition %d", partition);
 343
 344     if (device) {
 345         pid_t pid;
 346         int sock;
 347
 348         if (!verbose)
 349             daemon(0, 0);       /* detach client and server */
 350
 351         if (socket == NULL) {
 352             sprintf(sockpath, SOCKET_PATH, basename(device));
 353             socket = sockpath;
 354         }
 355
 356         pid = fork();
 357         if (pid < 0)
 358             return 1;
 359         if (pid != 0) {
 360             off_t size;
 361             size_t blocksize;
 362
 363             ret = 0;
 364             bdrv_close(bs);
 365
 366             do {
 367                 sock = unix_socket_outgoing(socket);
 368                 if (sock == -1) {
 369                     if (errno != ENOENT && errno != ECONNREFUSED)
 370                         goto out;
 371                     sleep(1);   /* wait children */
 372                 }
 373             } while (sock == -1);
 374
 375             fd = open(device, O_RDWR);
 376             if (fd == -1) {
 377                 ret = 1;
 378                 goto out;
 379             }
 380
 381             ret = nbd_receive_negotiate(sock, &size, &blocksize);
 382             if (ret == -1) {
 383                 ret = 1;
 384                 goto out;
 385             }
 386
 387             ret = nbd_init(fd, sock, size, blocksize);
 388             if (ret == -1) {
 389                 ret = 1;
 390                 goto out;
 391             }
 392
 393             printf("NBD device %s is now connected to file %s\n",
 394                     device, argv[optind]);
 395
 396             /* update partition table */
 397
 398             show_parts(device);
 399
 400             nbd_client(fd, sock);
 401             close(fd);
 402  out:
 403             kill(pid, SIGTERM);
 404             unlink(socket);
 405
 406             return ret;
 407         }
 408         /* children */
 409     }
 410
 411     sharing_fds = qemu_malloc((shared + 1) * sizeof(int));
 412     if (sharing_fds == NULL)
 413         errx(ENOMEM, "Cannot allocate sharing fds");
 414
 415     if (socket) {
 416         sharing_fds[0] = unix_socket_incoming(socket);
 417     } else {
 418         sharing_fds[0] = tcp_socket_incoming(bindto, port);
 419     }
 420
 421     if (sharing_fds[0] == -1)
 422         return 1;
 423     max_fd = sharing_fds[0];
 424     nb_fds++;
 425
 426     data = qemu_memalign(512, NBD_BUFFER_SIZE);
 427     if (data == NULL)
 428         errx(ENOMEM, "Cannot allocate data buffer");
 429
 430     do {
 431
 432         FD_ZERO(&fds);
 433         for (i = 0; i < nb_fds; i++)
 434             FD_SET(sharing_fds[i], &fds);
 435
 436         ret = select(max_fd + 1, &fds, NULL, NULL, NULL);
 437         if (ret == -1)
 438             break;
 439
 440         if (FD_ISSET(sharing_fds[0], &fds))
 441             ret--;
 442         for (i = 1; i < nb_fds && ret; i++) {
 443             if (FD_ISSET(sharing_fds[i], &fds)) {
 444                 if (nbd_trip(bs, sharing_fds[i], fd_size, dev_offset,
 445                     &offset, readonly, data, NBD_BUFFER_SIZE) != 0) {
 446                     close(sharing_fds[i]);
 447                     nb_fds--;
 448                     sharing_fds[i] = sharing_fds[nb_fds];
 449                     i--;
 450                 }
 451                 ret--;
 452             }
 453         }
 454         /* new connection ? */
 455         if (FD_ISSET(sharing_fds[0], &fds)) {
 456             if (nb_fds < shared + 1) {
 457                 sharing_fds[nb_fds] = accept(sharing_fds[0],
 458                                              (struct sockaddr *)&addr,
 459                                              &addr_len);
 460                 if (sharing_fds[nb_fds] != -1 &&
 461                     nbd_negotiate(sharing_fds[nb_fds], fd_size) != -1) {
 462                         if (sharing_fds[nb_fds] > max_fd)
 463                             max_fd = sharing_fds[nb_fds];
 464                         nb_fds++;
 465                 }
 466             }
 467         }
 468     } while (persistent || nb_fds > 1);
 469     qemu_free(data);
 470
 471     close(sharing_fds[0]);
 472     bdrv_close(bs);
 473     qemu_free(sharing_fds);
 474     if (socket)
 475         unlink(socket);
 476
 477     return 0;
 478 }