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/osdep.h"
  20 #include "qapi/error.h"
  21 #include "qemu-common.h"
  22 #include "qemu/cutils.h"
  23 #include "sysemu/block-backend.h"
  24 #include "block/block_int.h"
  25 #include "block/nbd.h"
  26 #include "qemu/main-loop.h"
  27 #include "qemu/error-report.h"
  28 #include "qemu/config-file.h"
  29 #include "qemu/bswap.h"
  30 #include "qemu/log.h"
  31 #include "block/snapshot.h"
  32 #include "qapi/util.h"
  33 #include "qapi/qmp/qstring.h"
  34 #include "qom/object_interfaces.h"
  35 #include "io/channel-socket.h"
  36 #include "crypto/init.h"
  37 #include "trace/control.h"
  38
  39 #include <getopt.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         256
  45 #define QEMU_NBD_OPT_AIO           257
  46 #define QEMU_NBD_OPT_DISCARD       258
  47 #define QEMU_NBD_OPT_DETECT_ZEROES 259
  48 #define QEMU_NBD_OPT_OBJECT        260
  49 #define QEMU_NBD_OPT_TLSCREDS      261
  50 #define QEMU_NBD_OPT_IMAGE_OPTS    262
  51
  52 #define MBR_SIZE 512
  53
  54 static NBDExport *exp;
  55 static bool newproto;
  56 static int verbose;
  57 static char *srcpath;
  58 static SocketAddress *saddr;
  59 static int persistent = 0;
  60 static enum { RUNNING, TERMINATE, TERMINATING, TERMINATED } state;
  61 static int shared = 1;
  62 static int nb_fds;
  63 static QIOChannelSocket *server_ioc;
  64 static int server_watch = -1;
  65 static QCryptoTLSCreds *tlscreds;
  66
  67 static void usage(const char *name)
  68 {
  69     (printf) (
  70 "Usage: %s [OPTIONS] FILE\n"
  71 "QEMU Disk Network Block Device Server\n"
  72 "\n"
  73 "  -h, --help                display this help and exit\n"
  74 "  -V, --version             output version information and exit\n"
  75 "\n"
  76 "Connection properties:\n"
  77 "  -p, --port=PORT           port to listen on (default `%d')\n"
  78 "  -b, --bind=IFACE          interface to bind to (default `0.0.0.0')\n"
  79 "  -k, --socket=PATH         path to the unix socket\n"
  80 "                            (default '"SOCKET_PATH"')\n"
  81 "  -e, --shared=NUM          device can be shared by NUM clients (default '1')\n"
  82 "  -t, --persistent          don't exit on the last connection\n"
  83 "  -v, --verbose             display extra debugging information\n"
  84 "  -x, --export-name=NAME    expose export by name\n"
  85 "\n"
  86 "Exposing part of the image:\n"
  87 "  -o, --offset=OFFSET       offset into the image\n"
  88 "  -P, --partition=NUM       only expose partition NUM\n"
  89 "\n"
  90 "General purpose options:\n"
  91 "  --object type,id=ID,...   define an object such as 'secret' for providing\n"
  92 "                            passwords and/or encryption keys\n"
  93 "  -T, --trace [[enable=]<pattern>][,events=<file>][,file=<file>]\n"
  94 "                            specify tracing options\n"
  95 #ifdef __linux__
  96 "Kernel NBD client support:\n"
  97 "  -c, --connect=DEV         connect FILE to the local NBD device DEV\n"
  98 "  -d, --disconnect          disconnect the specified device\n"
  99 "\n"
 100 #endif
 101 "\n"
 102 "Block device options:\n"
 103 "  -f, --format=FORMAT       set image format (raw, qcow2, ...)\n"
 104 "  -r, --read-only           export read-only\n"
 105 "  -s, --snapshot            use FILE as an external snapshot, create a temporary\n"
 106 "                            file with backing_file=FILE, redirect the write to\n"
 107 "                            the temporary one\n"
 108 "  -l, --load-snapshot=SNAPSHOT_PARAM\n"
 109 "                            load an internal snapshot inside FILE and export it\n"
 110 "                            as an read-only device, SNAPSHOT_PARAM format is\n"
 111 "                            'snapshot.id=[ID],snapshot.name=[NAME]', or\n"
 112 "                            '[ID_OR_NAME]'\n"
 113 "  -n, --nocache             disable host cache\n"
 114 "      --cache=MODE          set cache mode (none, writeback, ...)\n"
 115 "      --aio=MODE            set AIO mode (native or threads)\n"
 116 "      --discard=MODE        set discard mode (ignore, unmap)\n"
 117 "      --detect-zeroes=MODE  set detect-zeroes mode (off, on, unmap)\n"
 118 "      --image-opts          treat FILE as a full set of image options\n"
 119 "\n"
 120 "Report bugs to <qemu-devel@nongnu.org>\n"
 121     , name, NBD_DEFAULT_PORT, "DEVICE");
 122 }
 123
 124 static void version(const char *name)
 125 {
 126     printf(
 127 "%s version 0.0.1\n"
 128 "Written by Anthony Liguori.\n"
 129 "\n"
 130 "Copyright (C) 2006 Anthony Liguori <anthony@codemonkey.ws>.\n"
 131 "This is free software; see the source for copying conditions.  There is NO\n"
 132 "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
 133     , name);
 134 }
 135
 136 struct partition_record
 137 {
 138     uint8_t bootable;
 139     uint8_t start_head;
 140     uint32_t start_cylinder;
 141     uint8_t start_sector;
 142     uint8_t system;
 143     uint8_t end_head;
 144     uint8_t end_cylinder;
 145     uint8_t end_sector;
 146     uint32_t start_sector_abs;
 147     uint32_t nb_sectors_abs;
 148 };
 149
 150 static void read_partition(uint8_t *p, struct partition_record *r)
 151 {
 152     r->bootable = p[0];
 153     r->start_head = p[1];
 154     r->start_cylinder = p[3] | ((p[2] << 2) & 0x0300);
 155     r->start_sector = p[2] & 0x3f;
 156     r->system = p[4];
 157     r->end_head = p[5];
 158     r->end_cylinder = p[7] | ((p[6] << 2) & 0x300);
 159     r->end_sector = p[6] & 0x3f;
 160
 161     r->start_sector_abs = ldl_le_p(p + 8);
 162     r->nb_sectors_abs   = ldl_le_p(p + 12);
 163 }
 164
 165 static int find_partition(BlockBackend *blk, int partition,
 166                           off_t *offset, off_t *size)
 167 {
 168     struct partition_record mbr[4];
 169     uint8_t data[MBR_SIZE];
 170     int i;
 171     int ext_partnum = 4;
 172     int ret;
 173
 174     ret = blk_pread(blk, 0, data, sizeof(data));
 175     if (ret < 0) {
 176         error_report("error while reading: %s", strerror(-ret));
 177         exit(EXIT_FAILURE);
 178     }
 179
 180     if (data[510] != 0x55 || data[511] != 0xaa) {
 181         return -EINVAL;
 182     }
 183
 184     for (i = 0; i < 4; i++) {
 185         read_partition(&data[446 + 16 * i], &mbr[i]);
 186
 187         if (!mbr[i].system || !mbr[i].nb_sectors_abs) {
 188             continue;
 189         }
 190
 191         if (mbr[i].system == 0xF || mbr[i].system == 0x5) {
 192             struct partition_record ext[4];
 193             uint8_t data1[MBR_SIZE];
 194             int j;
 195
 196             ret = blk_pread(blk, mbr[i].start_sector_abs * MBR_SIZE,
 197                             data1, sizeof(data1));
 198             if (ret < 0) {
 199                 error_report("error while reading: %s", strerror(-ret));
 200                 exit(EXIT_FAILURE);
 201             }
 202
 203             for (j = 0; j < 4; j++) {
 204                 read_partition(&data1[446 + 16 * j], &ext[j]);
 205                 if (!ext[j].system || !ext[j].nb_sectors_abs) {
 206                     continue;
 207                 }
 208
 209                 if ((ext_partnum + j + 1) == partition) {
 210                     *offset = (uint64_t)ext[j].start_sector_abs << 9;
 211                     *size = (uint64_t)ext[j].nb_sectors_abs << 9;
 212                     return 0;
 213                 }
 214             }
 215             ext_partnum += 4;
 216         } else if ((i + 1) == partition) {
 217             *offset = (uint64_t)mbr[i].start_sector_abs << 9;
 218             *size = (uint64_t)mbr[i].nb_sectors_abs << 9;
 219             return 0;
 220         }
 221     }
 222
 223     return -ENOENT;
 224 }
 225
 226 static void termsig_handler(int signum)
 227 {
 228     atomic_cmpxchg(&state, RUNNING, TERMINATE);
 229     qemu_notify_event();
 230 }
 231
 232
 233 static void *show_parts(void *arg)
 234 {
 235     char *device = arg;
 236     int nbd;
 237
 238     /* linux just needs an open() to trigger
 239      * the partition table update
 240      * but remember to load the module with max_part != 0 :
 241      *     modprobe nbd max_part=63
 242      */
 243     nbd = open(device, O_RDWR);
 244     if (nbd >= 0) {
 245         close(nbd);
 246     }
 247     return NULL;
 248 }
 249
 250 static void *nbd_client_thread(void *arg)
 251 {
 252     char *device = arg;
 253     off_t size;
 254     uint16_t nbdflags;
 255     QIOChannelSocket *sioc;
 256     int fd;
 257     int ret;
 258     pthread_t show_parts_thread;
 259     Error *local_error = NULL;
 260
 261     sioc = qio_channel_socket_new();
 262     if (qio_channel_socket_connect_sync(sioc,
 263                                         saddr,
 264                                         &local_error) < 0) {
 265         error_report_err(local_error);
 266         goto out;
 267     }
 268
 269     ret = nbd_receive_negotiate(QIO_CHANNEL(sioc), NULL, &nbdflags,
 270                                 NULL, NULL, NULL,
 271                                 &size, &local_error);
 272     if (ret < 0) {
 273         if (local_error) {
 274             error_report_err(local_error);
 275         }
 276         goto out_socket;
 277     }
 278
 279     fd = open(device, O_RDWR);
 280     if (fd < 0) {
 281         /* Linux-only, we can use %m in printf.  */
 282         error_report("Failed to open %s: %m", device);
 283         goto out_socket;
 284     }
 285
 286     ret = nbd_init(fd, sioc, nbdflags, size);
 287     if (ret < 0) {
 288         goto out_fd;
 289     }
 290
 291     /* update partition table */
 292     pthread_create(&show_parts_thread, NULL, show_parts, device);
 293
 294     if (verbose) {
 295         fprintf(stderr, "NBD device %s is now connected to %s\n",
 296                 device, srcpath);
 297     } else {
 298         /* Close stderr so that the qemu-nbd process exits.  */
 299         dup2(STDOUT_FILENO, STDERR_FILENO);
 300     }
 301
 302     ret = nbd_client(fd);
 303     if (ret) {
 304         goto out_fd;
 305     }
 306     close(fd);
 307     object_unref(OBJECT(sioc));
 308     kill(getpid(), SIGTERM);
 309     return (void *) EXIT_SUCCESS;
 310
 311 out_fd:
 312     close(fd);
 313 out_socket:
 314     object_unref(OBJECT(sioc));
 315 out:
 316     kill(getpid(), SIGTERM);
 317     return (void *) EXIT_FAILURE;
 318 }
 319
 320 static int nbd_can_accept(void)
 321 {
 322     return nb_fds < shared;
 323 }
 324
 325 static void nbd_export_closed(NBDExport *exp)
 326 {
 327     assert(state == TERMINATING);
 328     state = TERMINATED;
 329 }
 330
 331 static void nbd_update_server_watch(void);
 332
 333 static void nbd_client_closed(NBDClient *client)
 334 {
 335     nb_fds--;
 336     if (nb_fds == 0 && !persistent && state == RUNNING) {
 337         state = TERMINATE;
 338     }
 339     nbd_update_server_watch();
 340     nbd_client_put(client);
 341 }
 342
 343 static gboolean nbd_accept(QIOChannel *ioc, GIOCondition cond, gpointer opaque)
 344 {
 345     QIOChannelSocket *cioc;
 346
 347     cioc = qio_channel_socket_accept(QIO_CHANNEL_SOCKET(ioc),
 348                                      NULL);
 349     if (!cioc) {
 350         return TRUE;
 351     }
 352
 353     if (state >= TERMINATE) {
 354         object_unref(OBJECT(cioc));
 355         return TRUE;
 356     }
 357
 358     nb_fds++;
 359     nbd_update_server_watch();
 360     nbd_client_new(newproto ? NULL : exp, cioc,
 361                    tlscreds, NULL, nbd_client_closed);
 362     object_unref(OBJECT(cioc));
 363
 364     return TRUE;
 365 }
 366
 367 static void nbd_update_server_watch(void)
 368 {
 369     if (nbd_can_accept()) {
 370         if (server_watch == -1) {
 371             server_watch = qio_channel_add_watch(QIO_CHANNEL(server_ioc),
 372                                                  G_IO_IN,
 373                                                  nbd_accept,
 374                                                  NULL, NULL);
 375         }
 376     } else {
 377         if (server_watch != -1) {
 378             g_source_remove(server_watch);
 379             server_watch = -1;
 380         }
 381     }
 382 }
 383
 384
 385 static SocketAddress *nbd_build_socket_address(const char *sockpath,
 386                                                const char *bindto,
 387                                                const char *port)
 388 {
 389     SocketAddress *saddr;
 390
 391     saddr = g_new0(SocketAddress, 1);
 392     if (sockpath) {
 393         saddr->type = SOCKET_ADDRESS_KIND_UNIX;
 394         saddr->u.q_unix.data = g_new0(UnixSocketAddress, 1);
 395         saddr->u.q_unix.data->path = g_strdup(sockpath);
 396     } else {
 397         InetSocketAddress *inet;
 398         saddr->type = SOCKET_ADDRESS_KIND_INET;
 399         inet = saddr->u.inet.data = g_new0(InetSocketAddress, 1);
 400         inet->host = g_strdup(bindto);
 401         if (port) {
 402             inet->port = g_strdup(port);
 403         } else  {
 404             inet->port = g_strdup_printf("%d", NBD_DEFAULT_PORT);
 405         }
 406     }
 407
 408     return saddr;
 409 }
 410
 411
 412 static QemuOptsList file_opts = {
 413     .name = "file",
 414     .implied_opt_name = "file",
 415     .head = QTAILQ_HEAD_INITIALIZER(file_opts.head),
 416     .desc = {
 417         /* no elements => accept any params */
 418         { /* end of list */ }
 419     },
 420 };
 421
 422 static QemuOptsList qemu_object_opts = {
 423     .name = "object",
 424     .implied_opt_name = "qom-type",
 425     .head = QTAILQ_HEAD_INITIALIZER(qemu_object_opts.head),
 426     .desc = {
 427         { }
 428     },
 429 };
 430
 431
 432
 433 static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, Error **errp)
 434 {
 435     Object *obj;
 436     QCryptoTLSCreds *creds;
 437
 438     obj = object_resolve_path_component(
 439         object_get_objects_root(), id);
 440     if (!obj) {
 441         error_setg(errp, "No TLS credentials with id '%s'",
 442                    id);
 443         return NULL;
 444     }
 445     creds = (QCryptoTLSCreds *)
 446         object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
 447     if (!creds) {
 448         error_setg(errp, "Object with id '%s' is not TLS credentials",
 449                    id);
 450         return NULL;
 451     }
 452
 453     if (creds->endpoint != QCRYPTO_TLS_CREDS_ENDPOINT_SERVER) {
 454         error_setg(errp,
 455                    "Expecting TLS credentials with a server endpoint");
 456         return NULL;
 457     }
 458     object_ref(obj);
 459     return creds;
 460 }
 461
 462
 463 int main(int argc, char **argv)
 464 {
 465     BlockBackend *blk;
 466     BlockDriverState *bs;
 467     off_t dev_offset = 0;
 468     uint16_t nbdflags = 0;
 469     bool disconnect = false;
 470     const char *bindto = "0.0.0.0";
 471     const char *port = NULL;
 472     char *sockpath = NULL;
 473     char *device = NULL;
 474     off_t fd_size;
 475     QemuOpts *sn_opts = NULL;
 476     const char *sn_id_or_name = NULL;
 477     const char *sopt = "hVb:o:p:rsnP:c:dvk:e:f:tl:x:T:";
 478     struct option lopt[] = {
 479         { "help", no_argument, NULL, 'h' },
 480         { "version", no_argument, NULL, 'V' },
 481         { "bind", required_argument, NULL, 'b' },
 482         { "port", required_argument, NULL, 'p' },
 483         { "socket", required_argument, NULL, 'k' },
 484         { "offset", required_argument, NULL, 'o' },
 485         { "read-only", no_argument, NULL, 'r' },
 486         { "partition", required_argument, NULL, 'P' },
 487         { "connect", required_argument, NULL, 'c' },
 488         { "disconnect", no_argument, NULL, 'd' },
 489         { "snapshot", no_argument, NULL, 's' },
 490         { "load-snapshot", required_argument, NULL, 'l' },
 491         { "nocache", no_argument, NULL, 'n' },
 492         { "cache", required_argument, NULL, QEMU_NBD_OPT_CACHE },
 493         { "aio", required_argument, NULL, QEMU_NBD_OPT_AIO },
 494         { "discard", required_argument, NULL, QEMU_NBD_OPT_DISCARD },
 495         { "detect-zeroes", required_argument, NULL,
 496           QEMU_NBD_OPT_DETECT_ZEROES },
 497         { "shared", required_argument, NULL, 'e' },
 498         { "format", required_argument, NULL, 'f' },
 499         { "persistent", no_argument, NULL, 't' },
 500         { "verbose", no_argument, NULL, 'v' },
 501         { "object", required_argument, NULL, QEMU_NBD_OPT_OBJECT },
 502         { "export-name", required_argument, NULL, 'x' },
 503         { "tls-creds", required_argument, NULL, QEMU_NBD_OPT_TLSCREDS },
 504         { "image-opts", no_argument, NULL, QEMU_NBD_OPT_IMAGE_OPTS },
 505         { "trace", required_argument, NULL, 'T' },
 506         { NULL, 0, NULL, 0 }
 507     };
 508     int ch;
 509     int opt_ind = 0;
 510     char *end;
 511     int flags = BDRV_O_RDWR;
 512     int partition = -1;
 513     int ret = 0;
 514     bool seen_cache = false;
 515     bool seen_discard = false;
 516     bool seen_aio = false;
 517     pthread_t client_thread;
 518     const char *fmt = NULL;
 519     Error *local_err = NULL;
 520     BlockdevDetectZeroesOptions detect_zeroes = BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
 521     QDict *options = NULL;
 522     const char *export_name = NULL;
 523     const char *tlscredsid = NULL;
 524     bool imageOpts = false;
 525     bool writethrough = true;
 526     char *trace_file = NULL;
 527
 528     /* The client thread uses SIGTERM to interrupt the server.  A signal
 529      * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
 530      */
 531     struct sigaction sa_sigterm;
 532     memset(&sa_sigterm, 0, sizeof(sa_sigterm));
 533     sa_sigterm.sa_handler = termsig_handler;
 534     sigaction(SIGTERM, &sa_sigterm, NULL);
 535
 536     module_call_init(MODULE_INIT_TRACE);
 537     qcrypto_init(&error_fatal);
 538
 539     module_call_init(MODULE_INIT_QOM);
 540     qemu_add_opts(&qemu_object_opts);
 541     qemu_add_opts(&qemu_trace_opts);
 542     qemu_init_exec_dir(argv[0]);
 543
 544     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 545         switch (ch) {
 546         case 's':
 547             flags |= BDRV_O_SNAPSHOT;
 548             break;
 549         case 'n':
 550             optarg = (char *) "none";
 551             /* fallthrough */
 552         case QEMU_NBD_OPT_CACHE:
 553             if (seen_cache) {
 554                 error_report("-n and --cache can only be specified once");
 555                 exit(EXIT_FAILURE);
 556             }
 557             seen_cache = true;
 558             if (bdrv_parse_cache_mode(optarg, &flags, &writethrough) == -1) {
 559                 error_report("Invalid cache mode `%s'", optarg);
 560                 exit(EXIT_FAILURE);
 561             }
 562             break;
 563         case QEMU_NBD_OPT_AIO:
 564             if (seen_aio) {
 565                 error_report("--aio can only be specified once");
 566                 exit(EXIT_FAILURE);
 567             }
 568             seen_aio = true;
 569             if (!strcmp(optarg, "native")) {
 570                 flags |= BDRV_O_NATIVE_AIO;
 571             } else if (!strcmp(optarg, "threads")) {
 572                 /* this is the default */
 573             } else {
 574                error_report("invalid aio mode `%s'", optarg);
 575                exit(EXIT_FAILURE);
 576             }
 577             break;
 578         case QEMU_NBD_OPT_DISCARD:
 579             if (seen_discard) {
 580                 error_report("--discard can only be specified once");
 581                 exit(EXIT_FAILURE);
 582             }
 583             seen_discard = true;
 584             if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
 585                 error_report("Invalid discard mode `%s'", optarg);
 586                 exit(EXIT_FAILURE);
 587             }
 588             break;
 589         case QEMU_NBD_OPT_DETECT_ZEROES:
 590             detect_zeroes =
 591                 qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
 592                                 optarg,
 593                                 BLOCKDEV_DETECT_ZEROES_OPTIONS__MAX,
 594                                 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
 595                                 &local_err);
 596             if (local_err) {
 597                 error_reportf_err(local_err,
 598                                   "Failed to parse detect_zeroes mode: ");
 599                 exit(EXIT_FAILURE);
 600             }
 601             if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
 602                 !(flags & BDRV_O_UNMAP)) {
 603                 error_report("setting detect-zeroes to unmap is not allowed "
 604                              "without setting discard operation to unmap");
 605                 exit(EXIT_FAILURE);
 606             }
 607             break;
 608         case 'b':
 609             bindto = optarg;
 610             break;
 611         case 'p':
 612             port = optarg;
 613             break;
 614         case 'o':
 615                 dev_offset = strtoll (optarg, &end, 0);
 616             if (*end) {
 617                 error_report("Invalid offset `%s'", optarg);
 618                 exit(EXIT_FAILURE);
 619             }
 620             if (dev_offset < 0) {
 621                 error_report("Offset must be positive `%s'", optarg);
 622                 exit(EXIT_FAILURE);
 623             }
 624             break;
 625         case 'l':
 626             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
 627                 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
 628                                                   optarg, false);
 629                 if (!sn_opts) {
 630                     error_report("Failed in parsing snapshot param `%s'",
 631                                  optarg);
 632                     exit(EXIT_FAILURE);
 633                 }
 634             } else {
 635                 sn_id_or_name = optarg;
 636             }
 637             /* fall through */
 638         case 'r':
 639             nbdflags |= NBD_FLAG_READ_ONLY;
 640             flags &= ~BDRV_O_RDWR;
 641             break;
 642         case 'P':
 643             partition = strtol(optarg, &end, 0);
 644             if (*end) {
 645                 error_report("Invalid partition `%s'", optarg);
 646                 exit(EXIT_FAILURE);
 647             }
 648             if (partition < 1 || partition > 8) {
 649                 error_report("Invalid partition %d", partition);
 650                 exit(EXIT_FAILURE);
 651             }
 652             break;
 653         case 'k':
 654             sockpath = optarg;
 655             if (sockpath[0] != '/') {
 656                 error_report("socket path must be absolute");
 657                 exit(EXIT_FAILURE);
 658             }
 659             break;
 660         case 'd':
 661             disconnect = true;
 662             break;
 663         case 'c':
 664             device = optarg;
 665             break;
 666         case 'e':
 667             shared = strtol(optarg, &end, 0);
 668             if (*end) {
 669                 error_report("Invalid shared device number '%s'", optarg);
 670                 exit(EXIT_FAILURE);
 671             }
 672             if (shared < 1) {
 673                 error_report("Shared device number must be greater than 0");
 674                 exit(EXIT_FAILURE);
 675             }
 676             break;
 677         case 'f':
 678             fmt = optarg;
 679             break;
 680         case 't':
 681             persistent = 1;
 682             break;
 683         case 'x':
 684             export_name = optarg;
 685             break;
 686         case 'v':
 687             verbose = 1;
 688             break;
 689         case 'V':
 690             version(argv[0]);
 691             exit(0);
 692             break;
 693         case 'h':
 694             usage(argv[0]);
 695             exit(0);
 696             break;
 697         case '?':
 698             error_report("Try `%s --help' for more information.", argv[0]);
 699             exit(EXIT_FAILURE);
 700         case QEMU_NBD_OPT_OBJECT: {
 701             QemuOpts *opts;
 702             opts = qemu_opts_parse_noisily(&qemu_object_opts,
 703                                            optarg, true);
 704             if (!opts) {
 705                 exit(EXIT_FAILURE);
 706             }
 707         }   break;
 708         case QEMU_NBD_OPT_TLSCREDS:
 709             tlscredsid = optarg;
 710             break;
 711         case QEMU_NBD_OPT_IMAGE_OPTS:
 712             imageOpts = true;
 713             break;
 714         case 'T':
 715             g_free(trace_file);
 716             trace_file = trace_opt_parse(optarg);
 717             break;
 718         }
 719     }
 720
 721     if ((argc - optind) != 1) {
 722         error_report("Invalid number of arguments");
 723         error_printf("Try `%s --help' for more information.\n", argv[0]);
 724         exit(EXIT_FAILURE);
 725     }
 726
 727     if (qemu_opts_foreach(&qemu_object_opts,
 728                           user_creatable_add_opts_foreach,
 729                           NULL, NULL)) {
 730         exit(EXIT_FAILURE);
 731     }
 732
 733     if (!trace_init_backends()) {
 734         exit(1);
 735     }
 736     trace_init_file(trace_file);
 737     qemu_set_log(LOG_TRACE);
 738
 739     if (tlscredsid) {
 740         if (sockpath) {
 741             error_report("TLS is only supported with IPv4/IPv6");
 742             exit(EXIT_FAILURE);
 743         }
 744         if (device) {
 745             error_report("TLS is not supported with a host device");
 746             exit(EXIT_FAILURE);
 747         }
 748         if (!export_name) {
 749             /* Set the default NBD protocol export name, since
 750              * we *must* use new style protocol for TLS */
 751             export_name = "";
 752         }
 753         tlscreds = nbd_get_tls_creds(tlscredsid, &local_err);
 754         if (local_err) {
 755             error_report("Failed to get TLS creds %s",
 756                          error_get_pretty(local_err));
 757             exit(EXIT_FAILURE);
 758         }
 759     }
 760
 761     if (disconnect) {
 762         int nbdfd = open(argv[optind], O_RDWR);
 763         if (nbdfd < 0) {
 764             error_report("Cannot open %s: %s", argv[optind],
 765                          strerror(errno));
 766             exit(EXIT_FAILURE);
 767         }
 768         nbd_disconnect(nbdfd);
 769
 770         close(nbdfd);
 771
 772         printf("%s disconnected\n", argv[optind]);
 773
 774         return 0;
 775     }
 776
 777     if (device && !verbose) {
 778         int stderr_fd[2];
 779         pid_t pid;
 780         int ret;
 781
 782         if (qemu_pipe(stderr_fd) < 0) {
 783             error_report("Error setting up communication pipe: %s",
 784                          strerror(errno));
 785             exit(EXIT_FAILURE);
 786         }
 787
 788         /* Now daemonize, but keep a communication channel open to
 789          * print errors and exit with the proper status code.
 790          */
 791         pid = fork();
 792         if (pid < 0) {
 793             error_report("Failed to fork: %s", strerror(errno));
 794             exit(EXIT_FAILURE);
 795         } else if (pid == 0) {
 796             close(stderr_fd[0]);
 797             ret = qemu_daemon(1, 0);
 798
 799             /* Temporarily redirect stderr to the parent's pipe...  */
 800             dup2(stderr_fd[1], STDERR_FILENO);
 801             if (ret < 0) {
 802                 error_report("Failed to daemonize: %s", strerror(errno));
 803                 exit(EXIT_FAILURE);
 804             }
 805
 806             /* ... close the descriptor we inherited and go on.  */
 807             close(stderr_fd[1]);
 808         } else {
 809             bool errors = false;
 810             char *buf;
 811
 812             /* In the parent.  Print error messages from the child until
 813              * it closes the pipe.
 814              */
 815             close(stderr_fd[1]);
 816             buf = g_malloc(1024);
 817             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
 818                 errors = true;
 819                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
 820                 if (ret < 0) {
 821                     exit(EXIT_FAILURE);
 822                 }
 823             }
 824             if (ret < 0) {
 825                 error_report("Cannot read from daemon: %s",
 826                              strerror(errno));
 827                 exit(EXIT_FAILURE);
 828             }
 829
 830             /* Usually the daemon should not print any message.
 831              * Exit with zero status in that case.
 832              */
 833             exit(errors);
 834         }
 835     }
 836
 837     if (device != NULL && sockpath == NULL) {
 838         sockpath = g_malloc(128);
 839         snprintf(sockpath, 128, SOCKET_PATH, basename(device));
 840     }
 841
 842     saddr = nbd_build_socket_address(sockpath, bindto, port);
 843
 844     if (qemu_init_main_loop(&local_err)) {
 845         error_report_err(local_err);
 846         exit(EXIT_FAILURE);
 847     }
 848     bdrv_init();
 849     atexit(bdrv_close_all);
 850
 851     srcpath = argv[optind];
 852     if (imageOpts) {
 853         QemuOpts *opts;
 854         if (fmt) {
 855             error_report("--image-opts and -f are mutually exclusive");
 856             exit(EXIT_FAILURE);
 857         }
 858         opts = qemu_opts_parse_noisily(&file_opts, srcpath, true);
 859         if (!opts) {
 860             qemu_opts_reset(&file_opts);
 861             exit(EXIT_FAILURE);
 862         }
 863         options = qemu_opts_to_qdict(opts, NULL);
 864         qemu_opts_reset(&file_opts);
 865         blk = blk_new_open(NULL, NULL, options, flags, &local_err);
 866     } else {
 867         if (fmt) {
 868             options = qdict_new();
 869             qdict_put(options, "driver", qstring_from_str(fmt));
 870         }
 871         blk = blk_new_open(srcpath, NULL, options, flags, &local_err);
 872     }
 873
 874     if (!blk) {
 875         error_reportf_err(local_err, "Failed to blk_new_open '%s': ",
 876                           argv[optind]);
 877         exit(EXIT_FAILURE);
 878     }
 879     bs = blk_bs(blk);
 880
 881     blk_set_enable_write_cache(blk, !writethrough);
 882
 883     if (sn_opts) {
 884         ret = bdrv_snapshot_load_tmp(bs,
 885                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
 886                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
 887                                      &local_err);
 888     } else if (sn_id_or_name) {
 889         ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
 890                                                    &local_err);
 891     }
 892     if (ret < 0) {
 893         error_reportf_err(local_err, "Failed to load snapshot: ");
 894         exit(EXIT_FAILURE);
 895     }
 896
 897     bs->detect_zeroes = detect_zeroes;
 898     fd_size = blk_getlength(blk);
 899     if (fd_size < 0) {
 900         error_report("Failed to determine the image length: %s",
 901                      strerror(-fd_size));
 902         exit(EXIT_FAILURE);
 903     }
 904
 905     if (dev_offset >= fd_size) {
 906         error_report("Offset (%lld) has to be smaller than the image size "
 907                      "(%lld)",
 908                      (long long int)dev_offset, (long long int)fd_size);
 909         exit(EXIT_FAILURE);
 910     }
 911     fd_size -= dev_offset;
 912
 913     if (partition != -1) {
 914         ret = find_partition(blk, partition, &dev_offset, &fd_size);
 915         if (ret < 0) {
 916             error_report("Could not find partition %d: %s", partition,
 917                          strerror(-ret));
 918             exit(EXIT_FAILURE);
 919         }
 920     }
 921
 922     exp = nbd_export_new(bs, dev_offset, fd_size, nbdflags, nbd_export_closed,
 923                          writethrough, NULL, &local_err);
 924     if (!exp) {
 925         error_report_err(local_err);
 926         exit(EXIT_FAILURE);
 927     }
 928     if (export_name) {
 929         nbd_export_set_name(exp, export_name);
 930         newproto = true;
 931     }
 932
 933     server_ioc = qio_channel_socket_new();
 934     if (qio_channel_socket_listen_sync(server_ioc, saddr, &local_err) < 0) {
 935         object_unref(OBJECT(server_ioc));
 936         error_report_err(local_err);
 937         return 1;
 938     }
 939
 940     if (device) {
 941         int ret;
 942
 943         ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
 944         if (ret != 0) {
 945             error_report("Failed to create client thread: %s", strerror(ret));
 946             exit(EXIT_FAILURE);
 947         }
 948     } else {
 949         /* Shut up GCC warnings.  */
 950         memset(&client_thread, 0, sizeof(client_thread));
 951     }
 952
 953     nbd_update_server_watch();
 954
 955     /* now when the initialization is (almost) complete, chdir("/")
 956      * to free any busy filesystems */
 957     if (chdir("/") < 0) {
 958         error_report("Could not chdir to root directory: %s",
 959                      strerror(errno));
 960         exit(EXIT_FAILURE);
 961     }
 962
 963     state = RUNNING;
 964     do {
 965         main_loop_wait(false);
 966         if (state == TERMINATE) {
 967             state = TERMINATING;
 968             nbd_export_close(exp);
 969             nbd_export_put(exp);
 970             exp = NULL;
 971         }
 972     } while (state != TERMINATED);
 973
 974     blk_unref(blk);
 975     if (sockpath) {
 976         unlink(sockpath);
 977     }
 978
 979     qemu_opts_del(sn_opts);
 980
 981     if (device) {
 982         void *ret;
 983         pthread_join(client_thread, &ret);
 984         exit(ret != NULL);
 985     } else {
 986         exit(EXIT_SUCCESS);
 987     }
 988 }