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             goto out;
 209         }
 210     } while (sock == -1);
 211
 212     ret = nbd_receive_negotiate(sock, NULL, &nbdflags,
 213                                 &size, &blocksize);
 214     if (ret == -1) {
 215         goto out;
 216     }
 217
 218     ret = nbd_init(fd, sock, nbdflags, size, blocksize);
 219     if (ret == -1) {
 220         goto out;
 221     }
 222
 223     /* update partition table */
 224     pthread_create(&show_parts_thread, NULL, show_parts, NULL);
 225
 226     if (verbose) {
 227         fprintf(stderr, "NBD device %s is now connected to %s\n",
 228                 device, srcpath);
 229     } else {
 230         /* Close stderr so that the qemu-nbd process exits.  */
 231         dup2(STDOUT_FILENO, STDERR_FILENO);
 232     }
 233
 234     ret = nbd_client(fd);
 235     if (ret) {
 236         goto out;
 237     }
 238     close(fd);
 239     kill(getpid(), SIGTERM);
 240     return (void *) EXIT_SUCCESS;
 241
 242 out:
 243     kill(getpid(), SIGTERM);
 244     return (void *) EXIT_FAILURE;
 245 }
 246
 247 int main(int argc, char **argv)
 248 {
 249     BlockDriverState *bs;
 250     off_t dev_offset = 0;
 251     off_t offset = 0;
 252     uint32_t nbdflags = 0;
 253     bool disconnect = false;
 254     const char *bindto = "0.0.0.0";
 255     int port = NBD_DEFAULT_PORT;
 256     struct sockaddr_in addr;
 257     socklen_t addr_len = sizeof(addr);
 258     off_t fd_size;
 259     const char *sopt = "hVb:o:p:rsnP:c:dvk:e:t";
 260     struct option lopt[] = {
 261         { "help", 0, NULL, 'h' },
 262         { "version", 0, NULL, 'V' },
 263         { "bind", 1, NULL, 'b' },
 264         { "port", 1, NULL, 'p' },
 265         { "socket", 1, NULL, 'k' },
 266         { "offset", 1, NULL, 'o' },
 267         { "read-only", 0, NULL, 'r' },
 268         { "partition", 1, NULL, 'P' },
 269         { "connect", 1, NULL, 'c' },
 270         { "disconnect", 0, NULL, 'd' },
 271         { "snapshot", 0, NULL, 's' },
 272         { "nocache", 0, NULL, 'n' },
 273         { "shared", 1, NULL, 'e' },
 274         { "persistent", 0, NULL, 't' },
 275         { "verbose", 0, NULL, 'v' },
 276         { NULL, 0, NULL, 0 }
 277     };
 278     int ch;
 279     int opt_ind = 0;
 280     int li;
 281     char *end;
 282     int flags = BDRV_O_RDWR;
 283     int partition = -1;
 284     int ret;
 285     int shared = 1;
 286     uint8_t *data;
 287     fd_set fds;
 288     int *sharing_fds;
 289     int fd;
 290     int i;
 291     int nb_fds = 0;
 292     int max_fd;
 293     int persistent = 0;
 294     pthread_t client_thread;
 295
 296     /* The client thread uses SIGTERM to interrupt the server.  A signal
 297      * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
 298      */
 299     struct sigaction sa_sigterm;
 300     int sigterm_fd[2];
 301     if (qemu_pipe(sigterm_fd) == -1) {
 302         err(EXIT_FAILURE, "Error setting up communication pipe");
 303     }
 304
 305     sigterm_wfd = sigterm_fd[1];
 306     memset(&sa_sigterm, 0, sizeof(sa_sigterm));
 307     sa_sigterm.sa_handler = termsig_handler;
 308     sigaction(SIGTERM, &sa_sigterm, NULL);
 309
 310     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 311         switch (ch) {
 312         case 's':
 313             flags |= BDRV_O_SNAPSHOT;
 314             break;
 315         case 'n':
 316             flags |= BDRV_O_NOCACHE | BDRV_O_CACHE_WB;
 317             break;
 318         case 'b':
 319             bindto = optarg;
 320             break;
 321         case 'p':
 322             li = strtol(optarg, &end, 0);
 323             if (*end) {
 324                 errx(EXIT_FAILURE, "Invalid port `%s'", optarg);
 325             }
 326             if (li < 1 || li > 65535) {
 327                 errx(EXIT_FAILURE, "Port out of range `%s'", optarg);
 328             }
 329             port = (uint16_t)li;
 330             break;
 331         case 'o':
 332                 dev_offset = strtoll (optarg, &end, 0);
 333             if (*end) {
 334                 errx(EXIT_FAILURE, "Invalid offset `%s'", optarg);
 335             }
 336             if (dev_offset < 0) {
 337                 errx(EXIT_FAILURE, "Offset must be positive `%s'", optarg);
 338             }
 339             break;
 340         case 'r':
 341             nbdflags |= NBD_FLAG_READ_ONLY;
 342             flags &= ~BDRV_O_RDWR;
 343             break;
 344         case 'P':
 345             partition = strtol(optarg, &end, 0);
 346             if (*end)
 347                 errx(EXIT_FAILURE, "Invalid partition `%s'", optarg);
 348             if (partition < 1 || partition > 8)
 349                 errx(EXIT_FAILURE, "Invalid partition %d", partition);
 350             break;
 351         case 'k':
 352             sockpath = optarg;
 353             if (sockpath[0] != '/')
 354                 errx(EXIT_FAILURE, "socket path must be absolute\n");
 355             break;
 356         case 'd':
 357             disconnect = true;
 358             break;
 359         case 'c':
 360             device = optarg;
 361             break;
 362         case 'e':
 363             shared = strtol(optarg, &end, 0);
 364             if (*end) {
 365                 errx(EXIT_FAILURE, "Invalid shared device number '%s'", optarg);
 366             }
 367             if (shared < 1) {
 368                 errx(EXIT_FAILURE, "Shared device number must be greater than 0\n");
 369             }
 370             break;
 371         case 't':
 372             persistent = 1;
 373             break;
 374         case 'v':
 375             verbose = 1;
 376             break;
 377         case 'V':
 378             version(argv[0]);
 379             exit(0);
 380             break;
 381         case 'h':
 382             usage(argv[0]);
 383             exit(0);
 384             break;
 385         case '?':
 386             errx(EXIT_FAILURE, "Try `%s --help' for more information.",
 387                  argv[0]);
 388         }
 389     }
 390
 391     if ((argc - optind) != 1) {
 392         errx(EXIT_FAILURE, "Invalid number of argument.\n"
 393              "Try `%s --help' for more information.",
 394              argv[0]);
 395     }
 396
 397     if (disconnect) {
 398         fd = open(argv[optind], O_RDWR);
 399         if (fd == -1)
 400             err(EXIT_FAILURE, "Cannot open %s", argv[optind]);
 401
 402         nbd_disconnect(fd);
 403
 404         close(fd);
 405
 406         printf("%s disconnected\n", argv[optind]);
 407
 408         return 0;
 409     }
 410
 411     if (device && !verbose) {
 412         int stderr_fd[2];
 413         pid_t pid;
 414         int ret;
 415
 416         if (qemu_pipe(stderr_fd) == -1) {
 417             err(EXIT_FAILURE, "Error setting up communication pipe");
 418         }
 419
 420         /* Now daemonize, but keep a communication channel open to
 421          * print errors and exit with the proper status code.
 422          */
 423         pid = fork();
 424         if (pid == 0) {
 425             close(stderr_fd[0]);
 426             ret = qemu_daemon(0, 0);
 427
 428             /* Temporarily redirect stderr to the parent's pipe...  */
 429             dup2(stderr_fd[1], STDERR_FILENO);
 430             if (ret == -1) {
 431                 err(EXIT_FAILURE, "Failed to daemonize");
 432             }
 433
 434             /* ... close the descriptor we inherited and go on.  */
 435             close(stderr_fd[1]);
 436         } else {
 437             bool errors = false;
 438             char *buf;
 439
 440             /* In the parent.  Print error messages from the child until
 441              * it closes the pipe.
 442              */
 443             close(stderr_fd[1]);
 444             buf = g_malloc(1024);
 445             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
 446                 errors = true;
 447                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
 448                 if (ret == -1) {
 449                     exit(EXIT_FAILURE);
 450                 }
 451             }
 452             if (ret == -1) {
 453                 err(EXIT_FAILURE, "Cannot read from daemon");
 454             }
 455
 456             /* Usually the daemon should not print any message.
 457              * Exit with zero status in that case.
 458              */
 459             exit(errors);
 460         }
 461     }
 462
 463     if (device) {
 464         /* Open before spawning new threads.  In the future, we may
 465          * drop privileges after opening.
 466          */
 467         fd = open(device, O_RDWR);
 468         if (fd == -1) {
 469             err(EXIT_FAILURE, "Failed to open %s", device);
 470         }
 471
 472         if (sockpath == NULL) {
 473             sockpath = g_malloc(128);
 474             snprintf(sockpath, 128, SOCKET_PATH, basename(device));
 475         }
 476     }
 477
 478     bdrv_init();
 479     atexit(bdrv_close_all);
 480
 481     bs = bdrv_new("hda");
 482     srcpath = argv[optind];
 483     if ((ret = bdrv_open(bs, srcpath, flags, NULL)) < 0) {
 484         errno = -ret;
 485         err(EXIT_FAILURE, "Failed to bdrv_open '%s'", argv[optind]);
 486     }
 487
 488     fd_size = bs->total_sectors * 512;
 489
 490     if (partition != -1 &&
 491         find_partition(bs, partition, &dev_offset, &fd_size)) {
 492         err(EXIT_FAILURE, "Could not find partition %d", partition);
 493     }
 494
 495     sharing_fds = g_malloc((shared + 1) * sizeof(int));
 496
 497     if (sockpath) {
 498         sharing_fds[0] = unix_socket_incoming(sockpath);
 499     } else {
 500         sharing_fds[0] = tcp_socket_incoming(bindto, port);
 501     }
 502
 503     if (sharing_fds[0] == -1)
 504         return 1;
 505
 506     if (device) {
 507         int ret;
 508
 509         ret = pthread_create(&client_thread, NULL, nbd_client_thread, &fd);
 510         if (ret != 0) {
 511             errx(EXIT_FAILURE, "Failed to create client thread: %s",
 512                  strerror(ret));
 513         }
 514     } else {
 515         /* Shut up GCC warnings.  */
 516         memset(&client_thread, 0, sizeof(client_thread));
 517     }
 518
 519     max_fd = sharing_fds[0];
 520     nb_fds++;
 521
 522     data = qemu_blockalign(bs, NBD_BUFFER_SIZE);
 523     if (data == NULL) {
 524         errx(EXIT_FAILURE, "Cannot allocate data buffer");
 525     }
 526
 527     do {
 528         FD_ZERO(&fds);
 529         FD_SET(sigterm_fd[0], &fds);
 530         for (i = 0; i < nb_fds; i++)
 531             FD_SET(sharing_fds[i], &fds);
 532
 533         do {
 534             ret = select(max_fd + 1, &fds, NULL, NULL, NULL);
 535         } while (ret == -1 && errno == EINTR);
 536         if (ret == -1 || FD_ISSET(sigterm_fd[0], &fds)) {
 537             break;
 538         }
 539
 540         if (FD_ISSET(sharing_fds[0], &fds))
 541             ret--;
 542         for (i = 1; i < nb_fds && ret; i++) {
 543             if (FD_ISSET(sharing_fds[i], &fds)) {
 544                 if (nbd_trip(bs, sharing_fds[i], fd_size, dev_offset,
 545                     &offset, nbdflags, data, NBD_BUFFER_SIZE) != 0) {
 546                     close(sharing_fds[i]);
 547                     nb_fds--;
 548                     sharing_fds[i] = sharing_fds[nb_fds];
 549                     i--;
 550                 }
 551                 ret--;
 552             }
 553         }
 554         /* new connection ? */
 555         if (FD_ISSET(sharing_fds[0], &fds)) {
 556             if (nb_fds < shared + 1) {
 557                 sharing_fds[nb_fds] = accept(sharing_fds[0],
 558                                              (struct sockaddr *)&addr,
 559                                              &addr_len);
 560                 if (sharing_fds[nb_fds] != -1 &&
 561                     nbd_negotiate(sharing_fds[nb_fds], fd_size, nbdflags) != -1) {
 562                         if (sharing_fds[nb_fds] > max_fd)
 563                             max_fd = sharing_fds[nb_fds];
 564                         nb_fds++;
 565                 }
 566             }
 567         }
 568     } while (persistent || nb_fds > 1);
 569     qemu_vfree(data);
 570
 571     close(sharing_fds[0]);
 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 }