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 "sysemu/block-backend.h"
  21 #include "block/block_int.h"
  22 #include "block/nbd.h"
  23 #include "qemu/main-loop.h"
  24 #include "qemu/sockets.h"
  25 #include "qemu/error-report.h"
  26 #include "block/snapshot.h"
  27 #include "qapi/util.h"
  28 #include "qapi/qmp/qstring.h"
  29
  30 #include <stdarg.h>
  31 #include <stdio.h>
  32 #include <getopt.h>
  33 #include <err.h>
  34 #include <sys/types.h>
  35 #include <sys/socket.h>
  36 #include <netinet/in.h>
  37 #include <netinet/tcp.h>
  38 #include <arpa/inet.h>
  39 #include <signal.h>
  40 #include <libgen.h>
  41 #include <pthread.h>
  42
  43 #define SOCKET_PATH                "/var/lock/qemu-nbd-%s"
  44 #define QEMU_NBD_OPT_CACHE         1
  45 #define QEMU_NBD_OPT_AIO           2
  46 #define QEMU_NBD_OPT_DISCARD       3
  47 #define QEMU_NBD_OPT_DETECT_ZEROES 4
  48
  49 static NBDExport *exp;
  50 static int verbose;
  51 static char *srcpath;
  52 static SocketAddress *saddr;
  53 static int persistent = 0;
  54 static enum { RUNNING, TERMINATE, TERMINATING, TERMINATED } state;
  55 static int shared = 1;
  56 static int nb_fds;
  57 static int server_fd;
  58
  59 static void usage(const char *name)
  60 {
  61     (printf) (
  62 "Usage: %s [OPTIONS] FILE\n"
  63 "QEMU Disk Network Block Device Server\n"
  64 "\n"
  65 "  -h, --help                display this help and exit\n"
  66 "  -V, --version             output version information and exit\n"
  67 "\n"
  68 "Connection properties:\n"
  69 "  -p, --port=PORT           port to listen on (default `%d')\n"
  70 "  -b, --bind=IFACE          interface to bind to (default `0.0.0.0')\n"
  71 "  -k, --socket=PATH         path to the unix socket\n"
  72 "                            (default '"SOCKET_PATH"')\n"
  73 "  -e, --shared=NUM          device can be shared by NUM clients (default '1')\n"
  74 "  -t, --persistent          don't exit on the last connection\n"
  75 "  -v, --verbose             display extra debugging information\n"
  76 "\n"
  77 "Exposing part of the image:\n"
  78 "  -o, --offset=OFFSET       offset into the image\n"
  79 "  -P, --partition=NUM       only expose partition NUM\n"
  80 "\n"
  81 #ifdef __linux__
  82 "Kernel NBD client support:\n"
  83 "  -c, --connect=DEV         connect FILE to the local NBD device DEV\n"
  84 "  -d, --disconnect          disconnect the specified device\n"
  85 "\n"
  86 #endif
  87 "\n"
  88 "Block device options:\n"
  89 "  -f, --format=FORMAT       set image format (raw, qcow2, ...)\n"
  90 "  -r, --read-only           export read-only\n"
  91 "  -s, --snapshot            use FILE as an external snapshot, create a temporary\n"
  92 "                            file with backing_file=FILE, redirect the write to\n"
  93 "                            the temporary one\n"
  94 "  -l, --load-snapshot=SNAPSHOT_PARAM\n"
  95 "                            load an internal snapshot inside FILE and export it\n"
  96 "                            as an read-only device, SNAPSHOT_PARAM format is\n"
  97 "                            'snapshot.id=[ID],snapshot.name=[NAME]', or\n"
  98 "                            '[ID_OR_NAME]'\n"
  99 "  -n, --nocache             disable host cache\n"
 100 "      --cache=MODE          set cache mode (none, writeback, ...)\n"
 101 "      --aio=MODE            set AIO mode (native or threads)\n"
 102 "      --discard=MODE        set discard mode (ignore, unmap)\n"
 103 "      --detect-zeroes=MODE  set detect-zeroes mode (off, on, unmap)\n"
 104 "\n"
 105 "Report bugs to <qemu-devel@nongnu.org>\n"
 106     , name, NBD_DEFAULT_PORT, "DEVICE");
 107 }
 108
 109 static void version(const char *name)
 110 {
 111     printf(
 112 "%s version 0.0.1\n"
 113 "Written by Anthony Liguori.\n"
 114 "\n"
 115 "Copyright (C) 2006 Anthony Liguori <anthony@codemonkey.ws>.\n"
 116 "This is free software; see the source for copying conditions.  There is NO\n"
 117 "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
 118     , name);
 119 }
 120
 121 struct partition_record
 122 {
 123     uint8_t bootable;
 124     uint8_t start_head;
 125     uint32_t start_cylinder;
 126     uint8_t start_sector;
 127     uint8_t system;
 128     uint8_t end_head;
 129     uint8_t end_cylinder;
 130     uint8_t end_sector;
 131     uint32_t start_sector_abs;
 132     uint32_t nb_sectors_abs;
 133 };
 134
 135 static void read_partition(uint8_t *p, struct partition_record *r)
 136 {
 137     r->bootable = p[0];
 138     r->start_head = p[1];
 139     r->start_cylinder = p[3] | ((p[2] << 2) & 0x0300);
 140     r->start_sector = p[2] & 0x3f;
 141     r->system = p[4];
 142     r->end_head = p[5];
 143     r->end_cylinder = p[7] | ((p[6] << 2) & 0x300);
 144     r->end_sector = p[6] & 0x3f;
 145
 146     r->start_sector_abs = le32_to_cpup((uint32_t *)(p +  8));
 147     r->nb_sectors_abs   = le32_to_cpup((uint32_t *)(p + 12));
 148 }
 149
 150 static int find_partition(BlockBackend *blk, int partition,
 151                           off_t *offset, off_t *size)
 152 {
 153     struct partition_record mbr[4];
 154     uint8_t data[512];
 155     int i;
 156     int ext_partnum = 4;
 157     int ret;
 158
 159     if ((ret = blk_read(blk, 0, data, 1)) < 0) {
 160         errno = -ret;
 161         err(EXIT_FAILURE, "error while reading");
 162     }
 163
 164     if (data[510] != 0x55 || data[511] != 0xaa) {
 165         return -EINVAL;
 166     }
 167
 168     for (i = 0; i < 4; i++) {
 169         read_partition(&data[446 + 16 * i], &mbr[i]);
 170
 171         if (!mbr[i].system || !mbr[i].nb_sectors_abs) {
 172             continue;
 173         }
 174
 175         if (mbr[i].system == 0xF || mbr[i].system == 0x5) {
 176             struct partition_record ext[4];
 177             uint8_t data1[512];
 178             int j;
 179
 180             if ((ret = blk_read(blk, mbr[i].start_sector_abs, data1, 1)) < 0) {
 181                 errno = -ret;
 182                 err(EXIT_FAILURE, "error while reading");
 183             }
 184
 185             for (j = 0; j < 4; j++) {
 186                 read_partition(&data1[446 + 16 * j], &ext[j]);
 187                 if (!ext[j].system || !ext[j].nb_sectors_abs) {
 188                     continue;
 189                 }
 190
 191                 if ((ext_partnum + j + 1) == partition) {
 192                     *offset = (uint64_t)ext[j].start_sector_abs << 9;
 193                     *size = (uint64_t)ext[j].nb_sectors_abs << 9;
 194                     return 0;
 195                 }
 196             }
 197             ext_partnum += 4;
 198         } else if ((i + 1) == partition) {
 199             *offset = (uint64_t)mbr[i].start_sector_abs << 9;
 200             *size = (uint64_t)mbr[i].nb_sectors_abs << 9;
 201             return 0;
 202         }
 203     }
 204
 205     return -ENOENT;
 206 }
 207
 208 static void termsig_handler(int signum)
 209 {
 210     state = TERMINATE;
 211     qemu_notify_event();
 212 }
 213
 214
 215 static void *show_parts(void *arg)
 216 {
 217     char *device = arg;
 218     int nbd;
 219
 220     /* linux just needs an open() to trigger
 221      * the partition table update
 222      * but remember to load the module with max_part != 0 :
 223      *     modprobe nbd max_part=63
 224      */
 225     nbd = open(device, O_RDWR);
 226     if (nbd >= 0) {
 227         close(nbd);
 228     }
 229     return NULL;
 230 }
 231
 232 static void *nbd_client_thread(void *arg)
 233 {
 234     char *device = arg;
 235     off_t size;
 236     uint32_t nbdflags;
 237     int fd, sock;
 238     int ret;
 239     pthread_t show_parts_thread;
 240     Error *local_error = NULL;
 241
 242
 243     sock = socket_connect(saddr, &local_error, NULL, NULL);
 244     if (sock < 0) {
 245         error_report_err(local_error);
 246         goto out;
 247     }
 248
 249     ret = nbd_receive_negotiate(sock, NULL, &nbdflags,
 250                                 &size, &local_error);
 251     if (ret < 0) {
 252         if (local_error) {
 253             fprintf(stderr, "%s\n", error_get_pretty(local_error));
 254             error_free(local_error);
 255         }
 256         goto out_socket;
 257     }
 258
 259     fd = open(device, O_RDWR);
 260     if (fd < 0) {
 261         /* Linux-only, we can use %m in printf.  */
 262         fprintf(stderr, "Failed to open %s: %m\n", device);
 263         goto out_socket;
 264     }
 265
 266     ret = nbd_init(fd, sock, nbdflags, size);
 267     if (ret < 0) {
 268         goto out_fd;
 269     }
 270
 271     /* update partition table */
 272     pthread_create(&show_parts_thread, NULL, show_parts, device);
 273
 274     if (verbose) {
 275         fprintf(stderr, "NBD device %s is now connected to %s\n",
 276                 device, srcpath);
 277     } else {
 278         /* Close stderr so that the qemu-nbd process exits.  */
 279         dup2(STDOUT_FILENO, STDERR_FILENO);
 280     }
 281
 282     ret = nbd_client(fd);
 283     if (ret) {
 284         goto out_fd;
 285     }
 286     close(fd);
 287     kill(getpid(), SIGTERM);
 288     return (void *) EXIT_SUCCESS;
 289
 290 out_fd:
 291     close(fd);
 292 out_socket:
 293     closesocket(sock);
 294 out:
 295     kill(getpid(), SIGTERM);
 296     return (void *) EXIT_FAILURE;
 297 }
 298
 299 static int nbd_can_accept(void)
 300 {
 301     return nb_fds < shared;
 302 }
 303
 304 static void nbd_export_closed(NBDExport *exp)
 305 {
 306     assert(state == TERMINATING);
 307     state = TERMINATED;
 308 }
 309
 310 static void nbd_update_server_fd_handler(int fd);
 311
 312 static void nbd_client_closed(NBDClient *client)
 313 {
 314     nb_fds--;
 315     if (nb_fds == 0 && !persistent && state == RUNNING) {
 316         state = TERMINATE;
 317     }
 318     nbd_update_server_fd_handler(server_fd);
 319     nbd_client_put(client);
 320 }
 321
 322 static void nbd_accept(void *opaque)
 323 {
 324     struct sockaddr_in addr;
 325     socklen_t addr_len = sizeof(addr);
 326
 327     int fd = accept(server_fd, (struct sockaddr *)&addr, &addr_len);
 328     if (fd < 0) {
 329         perror("accept");
 330         return;
 331     }
 332
 333     if (state >= TERMINATE) {
 334         close(fd);
 335         return;
 336     }
 337
 338     if (nbd_client_new(exp, fd, nbd_client_closed)) {
 339         nb_fds++;
 340         nbd_update_server_fd_handler(server_fd);
 341     } else {
 342         shutdown(fd, 2);
 343         close(fd);
 344     }
 345 }
 346
 347 static void nbd_update_server_fd_handler(int fd)
 348 {
 349     if (nbd_can_accept()) {
 350         qemu_set_fd_handler(fd, nbd_accept, NULL, (void *)(uintptr_t)fd);
 351     } else {
 352         qemu_set_fd_handler(fd, NULL, NULL, NULL);
 353     }
 354 }
 355
 356
 357 static SocketAddress *nbd_build_socket_address(const char *sockpath,
 358                                                const char *bindto,
 359                                                const char *port)
 360 {
 361     SocketAddress *saddr;
 362
 363     saddr = g_new0(SocketAddress, 1);
 364     if (sockpath) {
 365         saddr->type = SOCKET_ADDRESS_KIND_UNIX;
 366         saddr->u.q_unix = g_new0(UnixSocketAddress, 1);
 367         saddr->u.q_unix->path = g_strdup(sockpath);
 368     } else {
 369         saddr->type = SOCKET_ADDRESS_KIND_INET;
 370         saddr->u.inet = g_new0(InetSocketAddress, 1);
 371         saddr->u.inet->host = g_strdup(bindto);
 372         if (port) {
 373             saddr->u.inet->port = g_strdup(port);
 374         } else  {
 375             saddr->u.inet->port = g_strdup_printf("%d", NBD_DEFAULT_PORT);
 376         }
 377     }
 378
 379     return saddr;
 380 }
 381
 382
 383 int main(int argc, char **argv)
 384 {
 385     BlockBackend *blk;
 386     BlockDriverState *bs;
 387     off_t dev_offset = 0;
 388     uint32_t nbdflags = 0;
 389     bool disconnect = false;
 390     const char *bindto = "0.0.0.0";
 391     const char *port = NULL;
 392     char *sockpath = NULL;
 393     char *device = NULL;
 394     off_t fd_size;
 395     QemuOpts *sn_opts = NULL;
 396     const char *sn_id_or_name = NULL;
 397     const char *sopt = "hVb:o:p:rsnP:c:dvk:e:f:tl:";
 398     struct option lopt[] = {
 399         { "help", 0, NULL, 'h' },
 400         { "version", 0, NULL, 'V' },
 401         { "bind", 1, NULL, 'b' },
 402         { "port", 1, NULL, 'p' },
 403         { "socket", 1, NULL, 'k' },
 404         { "offset", 1, NULL, 'o' },
 405         { "read-only", 0, NULL, 'r' },
 406         { "partition", 1, NULL, 'P' },
 407         { "connect", 1, NULL, 'c' },
 408         { "disconnect", 0, NULL, 'd' },
 409         { "snapshot", 0, NULL, 's' },
 410         { "load-snapshot", 1, NULL, 'l' },
 411         { "nocache", 0, NULL, 'n' },
 412         { "cache", 1, NULL, QEMU_NBD_OPT_CACHE },
 413         { "aio", 1, NULL, QEMU_NBD_OPT_AIO },
 414         { "discard", 1, NULL, QEMU_NBD_OPT_DISCARD },
 415         { "detect-zeroes", 1, NULL, QEMU_NBD_OPT_DETECT_ZEROES },
 416         { "shared", 1, NULL, 'e' },
 417         { "format", 1, NULL, 'f' },
 418         { "persistent", 0, NULL, 't' },
 419         { "verbose", 0, NULL, 'v' },
 420         { NULL, 0, NULL, 0 }
 421     };
 422     int ch;
 423     int opt_ind = 0;
 424     char *end;
 425     int flags = BDRV_O_RDWR;
 426     int partition = -1;
 427     int ret = 0;
 428     int fd;
 429     bool seen_cache = false;
 430     bool seen_discard = false;
 431     bool seen_aio = false;
 432     pthread_t client_thread;
 433     const char *fmt = NULL;
 434     Error *local_err = NULL;
 435     BlockdevDetectZeroesOptions detect_zeroes = BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
 436     QDict *options = NULL;
 437
 438     /* The client thread uses SIGTERM to interrupt the server.  A signal
 439      * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
 440      */
 441     struct sigaction sa_sigterm;
 442     memset(&sa_sigterm, 0, sizeof(sa_sigterm));
 443     sa_sigterm.sa_handler = termsig_handler;
 444     sigaction(SIGTERM, &sa_sigterm, NULL);
 445     qemu_init_exec_dir(argv[0]);
 446
 447     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 448         switch (ch) {
 449         case 's':
 450             flags |= BDRV_O_SNAPSHOT;
 451             break;
 452         case 'n':
 453             optarg = (char *) "none";
 454             /* fallthrough */
 455         case QEMU_NBD_OPT_CACHE:
 456             if (seen_cache) {
 457                 errx(EXIT_FAILURE, "-n and --cache can only be specified once");
 458             }
 459             seen_cache = true;
 460             if (bdrv_parse_cache_flags(optarg, &flags) == -1) {
 461                 errx(EXIT_FAILURE, "Invalid cache mode `%s'", optarg);
 462             }
 463             break;
 464         case QEMU_NBD_OPT_AIO:
 465             if (seen_aio) {
 466                 errx(EXIT_FAILURE, "--aio can only be specified once");
 467             }
 468             seen_aio = true;
 469             if (!strcmp(optarg, "native")) {
 470                 flags |= BDRV_O_NATIVE_AIO;
 471             } else if (!strcmp(optarg, "threads")) {
 472                 /* this is the default */
 473             } else {
 474                errx(EXIT_FAILURE, "invalid aio mode `%s'", optarg);
 475             }
 476             break;
 477         case QEMU_NBD_OPT_DISCARD:
 478             if (seen_discard) {
 479                 errx(EXIT_FAILURE, "--discard can only be specified once");
 480             }
 481             seen_discard = true;
 482             if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
 483                 errx(EXIT_FAILURE, "Invalid discard mode `%s'", optarg);
 484             }
 485             break;
 486         case QEMU_NBD_OPT_DETECT_ZEROES:
 487             detect_zeroes =
 488                 qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
 489                                 optarg,
 490                                 BLOCKDEV_DETECT_ZEROES_OPTIONS_MAX,
 491                                 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
 492                                 &local_err);
 493             if (local_err) {
 494                 errx(EXIT_FAILURE, "Failed to parse detect_zeroes mode: %s",
 495                      error_get_pretty(local_err));
 496             }
 497             if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
 498                 !(flags & BDRV_O_UNMAP)) {
 499                 errx(EXIT_FAILURE, "setting detect-zeroes to unmap is not allowed "
 500                                    "without setting discard operation to unmap");
 501             }
 502             break;
 503         case 'b':
 504             bindto = optarg;
 505             break;
 506         case 'p':
 507             port = optarg;
 508             break;
 509         case 'o':
 510                 dev_offset = strtoll (optarg, &end, 0);
 511             if (*end) {
 512                 errx(EXIT_FAILURE, "Invalid offset `%s'", optarg);
 513             }
 514             if (dev_offset < 0) {
 515                 errx(EXIT_FAILURE, "Offset must be positive `%s'", optarg);
 516             }
 517             break;
 518         case 'l':
 519             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
 520                 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
 521                                                   optarg, false);
 522                 if (!sn_opts) {
 523                     errx(EXIT_FAILURE, "Failed in parsing snapshot param `%s'",
 524                          optarg);
 525                 }
 526             } else {
 527                 sn_id_or_name = optarg;
 528             }
 529             /* fall through */
 530         case 'r':
 531             nbdflags |= NBD_FLAG_READ_ONLY;
 532             flags &= ~BDRV_O_RDWR;
 533             break;
 534         case 'P':
 535             partition = strtol(optarg, &end, 0);
 536             if (*end) {
 537                 errx(EXIT_FAILURE, "Invalid partition `%s'", optarg);
 538             }
 539             if (partition < 1 || partition > 8) {
 540                 errx(EXIT_FAILURE, "Invalid partition %d", partition);
 541             }
 542             break;
 543         case 'k':
 544             sockpath = optarg;
 545             if (sockpath[0] != '/') {
 546                 errx(EXIT_FAILURE, "socket path must be absolute\n");
 547             }
 548             break;
 549         case 'd':
 550             disconnect = true;
 551             break;
 552         case 'c':
 553             device = optarg;
 554             break;
 555         case 'e':
 556             shared = strtol(optarg, &end, 0);
 557             if (*end) {
 558                 errx(EXIT_FAILURE, "Invalid shared device number '%s'", optarg);
 559             }
 560             if (shared < 1) {
 561                 errx(EXIT_FAILURE, "Shared device number must be greater than 0\n");
 562             }
 563             break;
 564         case 'f':
 565             fmt = optarg;
 566             break;
 567         case 't':
 568             persistent = 1;
 569             break;
 570         case 'v':
 571             verbose = 1;
 572             break;
 573         case 'V':
 574             version(argv[0]);
 575             exit(0);
 576             break;
 577         case 'h':
 578             usage(argv[0]);
 579             exit(0);
 580             break;
 581         case '?':
 582             errx(EXIT_FAILURE, "Try `%s --help' for more information.",
 583                  argv[0]);
 584         }
 585     }
 586
 587     if ((argc - optind) != 1) {
 588         errx(EXIT_FAILURE, "Invalid number of argument.\n"
 589              "Try `%s --help' for more information.",
 590              argv[0]);
 591     }
 592
 593     if (disconnect) {
 594         fd = open(argv[optind], O_RDWR);
 595         if (fd < 0) {
 596             err(EXIT_FAILURE, "Cannot open %s", argv[optind]);
 597         }
 598         nbd_disconnect(fd);
 599
 600         close(fd);
 601
 602         printf("%s disconnected\n", argv[optind]);
 603
 604         return 0;
 605     }
 606
 607     if (device && !verbose) {
 608         int stderr_fd[2];
 609         pid_t pid;
 610         int ret;
 611
 612         if (qemu_pipe(stderr_fd) < 0) {
 613             err(EXIT_FAILURE, "Error setting up communication pipe");
 614         }
 615
 616         /* Now daemonize, but keep a communication channel open to
 617          * print errors and exit with the proper status code.
 618          */
 619         pid = fork();
 620         if (pid < 0) {
 621             err(EXIT_FAILURE, "Failed to fork");
 622         } else if (pid == 0) {
 623             close(stderr_fd[0]);
 624             ret = qemu_daemon(1, 0);
 625
 626             /* Temporarily redirect stderr to the parent's pipe...  */
 627             dup2(stderr_fd[1], STDERR_FILENO);
 628             if (ret < 0) {
 629                 err(EXIT_FAILURE, "Failed to daemonize");
 630             }
 631
 632             /* ... close the descriptor we inherited and go on.  */
 633             close(stderr_fd[1]);
 634         } else {
 635             bool errors = false;
 636             char *buf;
 637
 638             /* In the parent.  Print error messages from the child until
 639              * it closes the pipe.
 640              */
 641             close(stderr_fd[1]);
 642             buf = g_malloc(1024);
 643             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
 644                 errors = true;
 645                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
 646                 if (ret < 0) {
 647                     exit(EXIT_FAILURE);
 648                 }
 649             }
 650             if (ret < 0) {
 651                 err(EXIT_FAILURE, "Cannot read from daemon");
 652             }
 653
 654             /* Usually the daemon should not print any message.
 655              * Exit with zero status in that case.
 656              */
 657             exit(errors);
 658         }
 659     }
 660
 661     if (device != NULL && sockpath == NULL) {
 662         sockpath = g_malloc(128);
 663         snprintf(sockpath, 128, SOCKET_PATH, basename(device));
 664     }
 665
 666     saddr = nbd_build_socket_address(sockpath, bindto, port);
 667
 668     if (qemu_init_main_loop(&local_err)) {
 669         error_report_err(local_err);
 670         exit(EXIT_FAILURE);
 671     }
 672     bdrv_init();
 673     atexit(bdrv_close_all);
 674
 675     if (fmt) {
 676         options = qdict_new();
 677         qdict_put(options, "driver", qstring_from_str(fmt));
 678     }
 679
 680     srcpath = argv[optind];
 681     blk = blk_new_open("hda", srcpath, NULL, options, flags, &local_err);
 682     if (!blk) {
 683         errx(EXIT_FAILURE, "Failed to blk_new_open '%s': %s", argv[optind],
 684              error_get_pretty(local_err));
 685     }
 686     bs = blk_bs(blk);
 687
 688     if (sn_opts) {
 689         ret = bdrv_snapshot_load_tmp(bs,
 690                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
 691                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
 692                                      &local_err);
 693     } else if (sn_id_or_name) {
 694         ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
 695                                                    &local_err);
 696     }
 697     if (ret < 0) {
 698         errno = -ret;
 699         err(EXIT_FAILURE,
 700             "Failed to load snapshot: %s",
 701             error_get_pretty(local_err));
 702     }
 703
 704     bs->detect_zeroes = detect_zeroes;
 705     fd_size = blk_getlength(blk);
 706     if (fd_size < 0) {
 707         errx(EXIT_FAILURE, "Failed to determine the image length: %s",
 708              strerror(-fd_size));
 709     }
 710
 711     if (partition != -1) {
 712         ret = find_partition(blk, partition, &dev_offset, &fd_size);
 713         if (ret < 0) {
 714             errno = -ret;
 715             err(EXIT_FAILURE, "Could not find partition %d", partition);
 716         }
 717     }
 718
 719     exp = nbd_export_new(blk, dev_offset, fd_size, nbdflags, nbd_export_closed,
 720                          &local_err);
 721     if (!exp) {
 722         errx(EXIT_FAILURE, "%s", error_get_pretty(local_err));
 723     }
 724
 725     fd = socket_listen(saddr, &local_err);
 726     if (fd < 0) {
 727         error_report_err(local_err);
 728         return 1;
 729     }
 730
 731     if (device) {
 732         int ret;
 733
 734         ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
 735         if (ret != 0) {
 736             errx(EXIT_FAILURE, "Failed to create client thread: %s",
 737                  strerror(ret));
 738         }
 739     } else {
 740         /* Shut up GCC warnings.  */
 741         memset(&client_thread, 0, sizeof(client_thread));
 742     }
 743
 744     server_fd = fd;
 745     nbd_update_server_fd_handler(fd);
 746
 747     /* now when the initialization is (almost) complete, chdir("/")
 748      * to free any busy filesystems */
 749     if (chdir("/") < 0) {
 750         err(EXIT_FAILURE, "Could not chdir to root directory");
 751     }
 752
 753     state = RUNNING;
 754     do {
 755         main_loop_wait(false);
 756         if (state == TERMINATE) {
 757             state = TERMINATING;
 758             nbd_export_close(exp);
 759             nbd_export_put(exp);
 760             exp = NULL;
 761         }
 762     } while (state != TERMINATED);
 763
 764     blk_unref(blk);
 765     if (sockpath) {
 766         unlink(sockpath);
 767     }
 768
 769     qemu_opts_del(sn_opts);
 770
 771     if (device) {
 772         void *ret;
 773         pthread_join(client_thread, &ret);
 774         exit(ret != NULL);
 775     } else {
 776         exit(EXIT_SUCCESS);
 777     }
 778 }