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