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 char *sockpath;
  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 #ifdef CONFIG_LINUX_AIO
 102 "      --aio=MODE            set AIO mode (native or threads)\n"
 103 #endif
 104 "      --discard=MODE        set discard mode (ignore, unmap)\n"
 105 "      --detect-zeroes=MODE  set detect-zeroes mode (off, on, discard)\n"
 106 "\n"
 107 "Report bugs to <qemu-devel@nongnu.org>\n"
 108     , name, NBD_DEFAULT_PORT, "DEVICE");
 109 }
 110
 111 static void version(const char *name)
 112 {
 113     printf(
 114 "%s version 0.0.1\n"
 115 "Written by Anthony Liguori.\n"
 116 "\n"
 117 "Copyright (C) 2006 Anthony Liguori <anthony@codemonkey.ws>.\n"
 118 "This is free software; see the source for copying conditions.  There is NO\n"
 119 "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
 120     , name);
 121 }
 122
 123 struct partition_record
 124 {
 125     uint8_t bootable;
 126     uint8_t start_head;
 127     uint32_t start_cylinder;
 128     uint8_t start_sector;
 129     uint8_t system;
 130     uint8_t end_head;
 131     uint8_t end_cylinder;
 132     uint8_t end_sector;
 133     uint32_t start_sector_abs;
 134     uint32_t nb_sectors_abs;
 135 };
 136
 137 static void read_partition(uint8_t *p, struct partition_record *r)
 138 {
 139     r->bootable = p[0];
 140     r->start_head = p[1];
 141     r->start_cylinder = p[3] | ((p[2] << 2) & 0x0300);
 142     r->start_sector = p[2] & 0x3f;
 143     r->system = p[4];
 144     r->end_head = p[5];
 145     r->end_cylinder = p[7] | ((p[6] << 2) & 0x300);
 146     r->end_sector = p[6] & 0x3f;
 147
 148     r->start_sector_abs = le32_to_cpup((uint32_t *)(p +  8));
 149     r->nb_sectors_abs   = le32_to_cpup((uint32_t *)(p + 12));
 150 }
 151
 152 static int find_partition(BlockBackend *blk, int partition,
 153                           off_t *offset, off_t *size)
 154 {
 155     struct partition_record mbr[4];
 156     uint8_t data[512];
 157     int i;
 158     int ext_partnum = 4;
 159     int ret;
 160
 161     if ((ret = blk_read(blk, 0, data, 1)) < 0) {
 162         errno = -ret;
 163         err(EXIT_FAILURE, "error while reading");
 164     }
 165
 166     if (data[510] != 0x55 || data[511] != 0xaa) {
 167         return -EINVAL;
 168     }
 169
 170     for (i = 0; i < 4; i++) {
 171         read_partition(&data[446 + 16 * i], &mbr[i]);
 172
 173         if (!mbr[i].system || !mbr[i].nb_sectors_abs) {
 174             continue;
 175         }
 176
 177         if (mbr[i].system == 0xF || mbr[i].system == 0x5) {
 178             struct partition_record ext[4];
 179             uint8_t data1[512];
 180             int j;
 181
 182             if ((ret = blk_read(blk, mbr[i].start_sector_abs, data1, 1)) < 0) {
 183                 errno = -ret;
 184                 err(EXIT_FAILURE, "error while reading");
 185             }
 186
 187             for (j = 0; j < 4; j++) {
 188                 read_partition(&data1[446 + 16 * j], &ext[j]);
 189                 if (!ext[j].system || !ext[j].nb_sectors_abs) {
 190                     continue;
 191                 }
 192
 193                 if ((ext_partnum + j + 1) == partition) {
 194                     *offset = (uint64_t)ext[j].start_sector_abs << 9;
 195                     *size = (uint64_t)ext[j].nb_sectors_abs << 9;
 196                     return 0;
 197                 }
 198             }
 199             ext_partnum += 4;
 200         } else if ((i + 1) == partition) {
 201             *offset = (uint64_t)mbr[i].start_sector_abs << 9;
 202             *size = (uint64_t)mbr[i].nb_sectors_abs << 9;
 203             return 0;
 204         }
 205     }
 206
 207     return -ENOENT;
 208 }
 209
 210 static void termsig_handler(int signum)
 211 {
 212     state = TERMINATE;
 213     qemu_notify_event();
 214 }
 215
 216 static void combine_addr(char *buf, size_t len, const char* address,
 217                          uint16_t port)
 218 {
 219     /* If the address-part contains a colon, it's an IPv6 IP so needs [] */
 220     if (strstr(address, ":")) {
 221         snprintf(buf, len, "[%s]:%u", address, port);
 222     } else {
 223         snprintf(buf, len, "%s:%u", address, port);
 224     }
 225 }
 226
 227 static int tcp_socket_incoming(const char *address, uint16_t port)
 228 {
 229     char address_and_port[128];
 230     Error *local_err = NULL;
 231
 232     combine_addr(address_and_port, 128, address, port);
 233     int fd = inet_listen(address_and_port, NULL, 0, SOCK_STREAM, 0, &local_err);
 234
 235     if (local_err != NULL) {
 236         error_report_err(local_err);
 237     }
 238     return fd;
 239 }
 240
 241 static int unix_socket_incoming(const char *path)
 242 {
 243     Error *local_err = NULL;
 244     int fd = unix_listen(path, NULL, 0, &local_err);
 245
 246     if (local_err != NULL) {
 247         error_report_err(local_err);
 248     }
 249     return fd;
 250 }
 251
 252 static int unix_socket_outgoing(const char *path)
 253 {
 254     Error *local_err = NULL;
 255     int fd = unix_connect(path, &local_err);
 256
 257     if (local_err != NULL) {
 258         error_report_err(local_err);
 259     }
 260     return fd;
 261 }
 262
 263 static void *show_parts(void *arg)
 264 {
 265     char *device = arg;
 266     int nbd;
 267
 268     /* linux just needs an open() to trigger
 269      * the partition table update
 270      * but remember to load the module with max_part != 0 :
 271      *     modprobe nbd max_part=63
 272      */
 273     nbd = open(device, O_RDWR);
 274     if (nbd >= 0) {
 275         close(nbd);
 276     }
 277     return NULL;
 278 }
 279
 280 static void *nbd_client_thread(void *arg)
 281 {
 282     char *device = arg;
 283     off_t size;
 284     uint32_t nbdflags;
 285     int fd, sock;
 286     int ret;
 287     pthread_t show_parts_thread;
 288     Error *local_error = NULL;
 289
 290     sock = unix_socket_outgoing(sockpath);
 291     if (sock < 0) {
 292         goto out;
 293     }
 294
 295     ret = nbd_receive_negotiate(sock, NULL, &nbdflags,
 296                                 &size, &local_error);
 297     if (ret < 0) {
 298         if (local_error) {
 299             fprintf(stderr, "%s\n", error_get_pretty(local_error));
 300             error_free(local_error);
 301         }
 302         goto out_socket;
 303     }
 304
 305     fd = open(device, O_RDWR);
 306     if (fd < 0) {
 307         /* Linux-only, we can use %m in printf.  */
 308         fprintf(stderr, "Failed to open %s: %m\n", device);
 309         goto out_socket;
 310     }
 311
 312     ret = nbd_init(fd, sock, nbdflags, size);
 313     if (ret < 0) {
 314         goto out_fd;
 315     }
 316
 317     /* update partition table */
 318     pthread_create(&show_parts_thread, NULL, show_parts, device);
 319
 320     if (verbose) {
 321         fprintf(stderr, "NBD device %s is now connected to %s\n",
 322                 device, srcpath);
 323     } else {
 324         /* Close stderr so that the qemu-nbd process exits.  */
 325         dup2(STDOUT_FILENO, STDERR_FILENO);
 326     }
 327
 328     ret = nbd_client(fd);
 329     if (ret) {
 330         goto out_fd;
 331     }
 332     close(fd);
 333     kill(getpid(), SIGTERM);
 334     return (void *) EXIT_SUCCESS;
 335
 336 out_fd:
 337     close(fd);
 338 out_socket:
 339     closesocket(sock);
 340 out:
 341     kill(getpid(), SIGTERM);
 342     return (void *) EXIT_FAILURE;
 343 }
 344
 345 static int nbd_can_accept(void)
 346 {
 347     return nb_fds < shared;
 348 }
 349
 350 static void nbd_export_closed(NBDExport *exp)
 351 {
 352     assert(state == TERMINATING);
 353     state = TERMINATED;
 354 }
 355
 356 static void nbd_update_server_fd_handler(int fd);
 357
 358 static void nbd_client_closed(NBDClient *client)
 359 {
 360     nb_fds--;
 361     if (nb_fds == 0 && !persistent && state == RUNNING) {
 362         state = TERMINATE;
 363     }
 364     nbd_update_server_fd_handler(server_fd);
 365     qemu_notify_event();
 366     nbd_client_put(client);
 367 }
 368
 369 static void nbd_accept(void *opaque)
 370 {
 371     struct sockaddr_in addr;
 372     socklen_t addr_len = sizeof(addr);
 373
 374     int fd = accept(server_fd, (struct sockaddr *)&addr, &addr_len);
 375     if (fd < 0) {
 376         perror("accept");
 377         return;
 378     }
 379
 380     if (state >= TERMINATE) {
 381         close(fd);
 382         return;
 383     }
 384
 385     if (nbd_client_new(exp, fd, nbd_client_closed)) {
 386         nb_fds++;
 387         nbd_update_server_fd_handler(server_fd);
 388     } else {
 389         shutdown(fd, 2);
 390         close(fd);
 391     }
 392 }
 393
 394 static void nbd_update_server_fd_handler(int fd)
 395 {
 396     if (nbd_can_accept()) {
 397         qemu_set_fd_handler(fd, nbd_accept, NULL, (void *)(uintptr_t)fd);
 398     } else {
 399         qemu_set_fd_handler(fd, NULL, NULL, NULL);
 400     }
 401 }
 402
 403 int main(int argc, char **argv)
 404 {
 405     BlockBackend *blk;
 406     BlockDriverState *bs;
 407     off_t dev_offset = 0;
 408     uint32_t nbdflags = 0;
 409     bool disconnect = false;
 410     const char *bindto = "0.0.0.0";
 411     char *device = NULL;
 412     int port = NBD_DEFAULT_PORT;
 413     off_t fd_size;
 414     QemuOpts *sn_opts = NULL;
 415     const char *sn_id_or_name = NULL;
 416     const char *sopt = "hVb:o:p:rsnP:c:dvk:e:f:tl:";
 417     struct option lopt[] = {
 418         { "help", 0, NULL, 'h' },
 419         { "version", 0, NULL, 'V' },
 420         { "bind", 1, NULL, 'b' },
 421         { "port", 1, NULL, 'p' },
 422         { "socket", 1, NULL, 'k' },
 423         { "offset", 1, NULL, 'o' },
 424         { "read-only", 0, NULL, 'r' },
 425         { "partition", 1, NULL, 'P' },
 426         { "connect", 1, NULL, 'c' },
 427         { "disconnect", 0, NULL, 'd' },
 428         { "snapshot", 0, NULL, 's' },
 429         { "load-snapshot", 1, NULL, 'l' },
 430         { "nocache", 0, NULL, 'n' },
 431         { "cache", 1, NULL, QEMU_NBD_OPT_CACHE },
 432 #ifdef CONFIG_LINUX_AIO
 433         { "aio", 1, NULL, QEMU_NBD_OPT_AIO },
 434 #endif
 435         { "discard", 1, NULL, QEMU_NBD_OPT_DISCARD },
 436         { "detect-zeroes", 1, NULL, QEMU_NBD_OPT_DETECT_ZEROES },
 437         { "shared", 1, NULL, 'e' },
 438         { "format", 1, NULL, 'f' },
 439         { "persistent", 0, NULL, 't' },
 440         { "verbose", 0, NULL, 'v' },
 441         { NULL, 0, NULL, 0 }
 442     };
 443     int ch;
 444     int opt_ind = 0;
 445     int li;
 446     char *end;
 447     int flags = BDRV_O_RDWR;
 448     int partition = -1;
 449     int ret = 0;
 450     int fd;
 451     bool seen_cache = false;
 452     bool seen_discard = false;
 453 #ifdef CONFIG_LINUX_AIO
 454     bool seen_aio = false;
 455 #endif
 456     pthread_t client_thread;
 457     const char *fmt = NULL;
 458     Error *local_err = NULL;
 459     BlockdevDetectZeroesOptions detect_zeroes = BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
 460     QDict *options = NULL;
 461
 462     /* The client thread uses SIGTERM to interrupt the server.  A signal
 463      * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
 464      */
 465     struct sigaction sa_sigterm;
 466     memset(&sa_sigterm, 0, sizeof(sa_sigterm));
 467     sa_sigterm.sa_handler = termsig_handler;
 468     sigaction(SIGTERM, &sa_sigterm, NULL);
 469     qemu_init_exec_dir(argv[0]);
 470
 471     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 472         switch (ch) {
 473         case 's':
 474             flags |= BDRV_O_SNAPSHOT;
 475             break;
 476         case 'n':
 477             optarg = (char *) "none";
 478             /* fallthrough */
 479         case QEMU_NBD_OPT_CACHE:
 480             if (seen_cache) {
 481                 errx(EXIT_FAILURE, "-n and --cache can only be specified once");
 482             }
 483             seen_cache = true;
 484             if (bdrv_parse_cache_flags(optarg, &flags) == -1) {
 485                 errx(EXIT_FAILURE, "Invalid cache mode `%s'", optarg);
 486             }
 487             break;
 488 #ifdef CONFIG_LINUX_AIO
 489         case QEMU_NBD_OPT_AIO:
 490             if (seen_aio) {
 491                 errx(EXIT_FAILURE, "--aio can only be specified once");
 492             }
 493             seen_aio = true;
 494             if (!strcmp(optarg, "native")) {
 495                 flags |= BDRV_O_NATIVE_AIO;
 496             } else if (!strcmp(optarg, "threads")) {
 497                 /* this is the default */
 498             } else {
 499                errx(EXIT_FAILURE, "invalid aio mode `%s'", optarg);
 500             }
 501             break;
 502 #endif
 503         case QEMU_NBD_OPT_DISCARD:
 504             if (seen_discard) {
 505                 errx(EXIT_FAILURE, "--discard can only be specified once");
 506             }
 507             seen_discard = true;
 508             if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
 509                 errx(EXIT_FAILURE, "Invalid discard mode `%s'", optarg);
 510             }
 511             break;
 512         case QEMU_NBD_OPT_DETECT_ZEROES:
 513             detect_zeroes =
 514                 qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
 515                                 optarg,
 516                                 BLOCKDEV_DETECT_ZEROES_OPTIONS_MAX,
 517                                 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
 518                                 &local_err);
 519             if (local_err) {
 520                 errx(EXIT_FAILURE, "Failed to parse detect_zeroes mode: %s",
 521                      error_get_pretty(local_err));
 522             }
 523             if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
 524                 !(flags & BDRV_O_UNMAP)) {
 525                 errx(EXIT_FAILURE, "setting detect-zeroes to unmap is not allowed "
 526                                    "without setting discard operation to unmap");
 527             }
 528             break;
 529         case 'b':
 530             bindto = optarg;
 531             break;
 532         case 'p':
 533             li = strtol(optarg, &end, 0);
 534             if (*end) {
 535                 errx(EXIT_FAILURE, "Invalid port `%s'", optarg);
 536             }
 537             if (li < 1 || li > 65535) {
 538                 errx(EXIT_FAILURE, "Port out of range `%s'", optarg);
 539             }
 540             port = (uint16_t)li;
 541             break;
 542         case 'o':
 543                 dev_offset = strtoll (optarg, &end, 0);
 544             if (*end) {
 545                 errx(EXIT_FAILURE, "Invalid offset `%s'", optarg);
 546             }
 547             if (dev_offset < 0) {
 548                 errx(EXIT_FAILURE, "Offset must be positive `%s'", optarg);
 549             }
 550             break;
 551         case 'l':
 552             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
 553                 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
 554                                                   optarg, false);
 555                 if (!sn_opts) {
 556                     errx(EXIT_FAILURE, "Failed in parsing snapshot param `%s'",
 557                          optarg);
 558                 }
 559             } else {
 560                 sn_id_or_name = optarg;
 561             }
 562             /* fall through */
 563         case 'r':
 564             nbdflags |= NBD_FLAG_READ_ONLY;
 565             flags &= ~BDRV_O_RDWR;
 566             break;
 567         case 'P':
 568             partition = strtol(optarg, &end, 0);
 569             if (*end) {
 570                 errx(EXIT_FAILURE, "Invalid partition `%s'", optarg);
 571             }
 572             if (partition < 1 || partition > 8) {
 573                 errx(EXIT_FAILURE, "Invalid partition %d", partition);
 574             }
 575             break;
 576         case 'k':
 577             sockpath = optarg;
 578             if (sockpath[0] != '/') {
 579                 errx(EXIT_FAILURE, "socket path must be absolute\n");
 580             }
 581             break;
 582         case 'd':
 583             disconnect = true;
 584             break;
 585         case 'c':
 586             device = optarg;
 587             break;
 588         case 'e':
 589             shared = strtol(optarg, &end, 0);
 590             if (*end) {
 591                 errx(EXIT_FAILURE, "Invalid shared device number '%s'", optarg);
 592             }
 593             if (shared < 1) {
 594                 errx(EXIT_FAILURE, "Shared device number must be greater than 0\n");
 595             }
 596             break;
 597         case 'f':
 598             fmt = optarg;
 599             break;
 600         case 't':
 601             persistent = 1;
 602             break;
 603         case 'v':
 604             verbose = 1;
 605             break;
 606         case 'V':
 607             version(argv[0]);
 608             exit(0);
 609             break;
 610         case 'h':
 611             usage(argv[0]);
 612             exit(0);
 613             break;
 614         case '?':
 615             errx(EXIT_FAILURE, "Try `%s --help' for more information.",
 616                  argv[0]);
 617         }
 618     }
 619
 620     if ((argc - optind) != 1) {
 621         errx(EXIT_FAILURE, "Invalid number of argument.\n"
 622              "Try `%s --help' for more information.",
 623              argv[0]);
 624     }
 625
 626     if (disconnect) {
 627         fd = open(argv[optind], O_RDWR);
 628         if (fd < 0) {
 629             err(EXIT_FAILURE, "Cannot open %s", argv[optind]);
 630         }
 631         nbd_disconnect(fd);
 632
 633         close(fd);
 634
 635         printf("%s disconnected\n", argv[optind]);
 636
 637         return 0;
 638     }
 639
 640     if (device && !verbose) {
 641         int stderr_fd[2];
 642         pid_t pid;
 643         int ret;
 644
 645         if (qemu_pipe(stderr_fd) < 0) {
 646             err(EXIT_FAILURE, "Error setting up communication pipe");
 647         }
 648
 649         /* Now daemonize, but keep a communication channel open to
 650          * print errors and exit with the proper status code.
 651          */
 652         pid = fork();
 653         if (pid < 0) {
 654             err(EXIT_FAILURE, "Failed to fork");
 655         } else if (pid == 0) {
 656             close(stderr_fd[0]);
 657             ret = qemu_daemon(1, 0);
 658
 659             /* Temporarily redirect stderr to the parent's pipe...  */
 660             dup2(stderr_fd[1], STDERR_FILENO);
 661             if (ret < 0) {
 662                 err(EXIT_FAILURE, "Failed to daemonize");
 663             }
 664
 665             /* ... close the descriptor we inherited and go on.  */
 666             close(stderr_fd[1]);
 667         } else {
 668             bool errors = false;
 669             char *buf;
 670
 671             /* In the parent.  Print error messages from the child until
 672              * it closes the pipe.
 673              */
 674             close(stderr_fd[1]);
 675             buf = g_malloc(1024);
 676             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
 677                 errors = true;
 678                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
 679                 if (ret < 0) {
 680                     exit(EXIT_FAILURE);
 681                 }
 682             }
 683             if (ret < 0) {
 684                 err(EXIT_FAILURE, "Cannot read from daemon");
 685             }
 686
 687             /* Usually the daemon should not print any message.
 688              * Exit with zero status in that case.
 689              */
 690             exit(errors);
 691         }
 692     }
 693
 694     if (device != NULL && sockpath == NULL) {
 695         sockpath = g_malloc(128);
 696         snprintf(sockpath, 128, SOCKET_PATH, basename(device));
 697     }
 698
 699     if (qemu_init_main_loop(&local_err)) {
 700         error_report_err(local_err);
 701         exit(EXIT_FAILURE);
 702     }
 703     bdrv_init();
 704     atexit(bdrv_close_all);
 705
 706     if (fmt) {
 707         options = qdict_new();
 708         qdict_put(options, "driver", qstring_from_str(fmt));
 709     }
 710
 711     srcpath = argv[optind];
 712     blk = blk_new_open("hda", srcpath, NULL, options, flags, &local_err);
 713     if (!blk) {
 714         errx(EXIT_FAILURE, "Failed to blk_new_open '%s': %s", argv[optind],
 715              error_get_pretty(local_err));
 716     }
 717     bs = blk_bs(blk);
 718
 719     if (sn_opts) {
 720         ret = bdrv_snapshot_load_tmp(bs,
 721                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
 722                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
 723                                      &local_err);
 724     } else if (sn_id_or_name) {
 725         ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
 726                                                    &local_err);
 727     }
 728     if (ret < 0) {
 729         errno = -ret;
 730         err(EXIT_FAILURE,
 731             "Failed to load snapshot: %s",
 732             error_get_pretty(local_err));
 733     }
 734
 735     bs->detect_zeroes = detect_zeroes;
 736     fd_size = blk_getlength(blk);
 737     if (fd_size < 0) {
 738         errx(EXIT_FAILURE, "Failed to determine the image length: %s",
 739              strerror(-fd_size));
 740     }
 741
 742     if (partition != -1) {
 743         ret = find_partition(blk, partition, &dev_offset, &fd_size);
 744         if (ret < 0) {
 745             errno = -ret;
 746             err(EXIT_FAILURE, "Could not find partition %d", partition);
 747         }
 748     }
 749
 750     exp = nbd_export_new(blk, dev_offset, fd_size, nbdflags, nbd_export_closed,
 751                          &local_err);
 752     if (!exp) {
 753         errx(EXIT_FAILURE, "%s", error_get_pretty(local_err));
 754     }
 755
 756     if (sockpath) {
 757         fd = unix_socket_incoming(sockpath);
 758     } else {
 759         fd = tcp_socket_incoming(bindto, port);
 760     }
 761
 762     if (fd < 0) {
 763         return 1;
 764     }
 765
 766     if (device) {
 767         int ret;
 768
 769         ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
 770         if (ret != 0) {
 771             errx(EXIT_FAILURE, "Failed to create client thread: %s",
 772                  strerror(ret));
 773         }
 774     } else {
 775         /* Shut up GCC warnings.  */
 776         memset(&client_thread, 0, sizeof(client_thread));
 777     }
 778
 779     server_fd = fd;
 780     nbd_update_server_fd_handler(fd);
 781
 782     /* now when the initialization is (almost) complete, chdir("/")
 783      * to free any busy filesystems */
 784     if (chdir("/") < 0) {
 785         err(EXIT_FAILURE, "Could not chdir to root directory");
 786     }
 787
 788     state = RUNNING;
 789     do {
 790         main_loop_wait(false);
 791         if (state == TERMINATE) {
 792             state = TERMINATING;
 793             nbd_export_close(exp);
 794             nbd_export_put(exp);
 795             exp = NULL;
 796         }
 797     } while (state != TERMINATED);
 798
 799     blk_unref(blk);
 800     if (sockpath) {
 801         unlink(sockpath);
 802     }
 803
 804     qemu_opts_del(sn_opts);
 805
 806     if (device) {
 807         void *ret;
 808         pthread_join(client_thread, &ret);
 809         exit(ret != NULL);
 810     } else {
 811         exit(EXIT_SUCCESS);
 812     }
 813 }