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 #define QEMU_NBD_OPT_FORK          263
  52
  53 #define MBR_SIZE 512
  54
  55 static NBDExport *exp;
  56 static bool newproto;
  57 static int verbose;
  58 static char *srcpath;
  59 static SocketAddress *saddr;
  60 static int persistent = 0;
  61 static enum { RUNNING, TERMINATE, TERMINATING, TERMINATED } state;
  62 static int shared = 1;
  63 static int nb_fds;
  64 static QIOChannelSocket *server_ioc;
  65 static int server_watch = -1;
  66 static QCryptoTLSCreds *tlscreds;
  67
  68 static void usage(const char *name)
  69 {
  70     (printf) (
  71 "Usage: %s [OPTIONS] FILE\n"
  72 "QEMU Disk Network Block Device Server\n"
  73 "\n"
  74 "  -h, --help                display this help and exit\n"
  75 "  -V, --version             output version information and exit\n"
  76 "\n"
  77 "Connection properties:\n"
  78 "  -p, --port=PORT           port to listen on (default `%d')\n"
  79 "  -b, --bind=IFACE          interface to bind to (default `0.0.0.0')\n"
  80 "  -k, --socket=PATH         path to the unix socket\n"
  81 "                            (default '"SOCKET_PATH"')\n"
  82 "  -e, --shared=NUM          device can be shared by NUM clients (default '1')\n"
  83 "  -t, --persistent          don't exit on the last connection\n"
  84 "  -v, --verbose             display extra debugging information\n"
  85 "  -x, --export-name=NAME    expose export by name\n"
  86 "  -D, --description=TEXT    with -x, also export a human-readable description\n"
  87 "\n"
  88 "Exposing part of the image:\n"
  89 "  -o, --offset=OFFSET       offset into the image\n"
  90 "  -P, --partition=NUM       only expose partition NUM\n"
  91 "\n"
  92 "General purpose options:\n"
  93 "  --object type,id=ID,...   define an object such as 'secret' for providing\n"
  94 "                            passwords and/or encryption keys\n"
  95 "  -T, --trace [[enable=]<pattern>][,events=<file>][,file=<file>]\n"
  96 "                            specify tracing options\n"
  97 "  --fork                    fork off the server process and exit the parent\n"
  98 "                            once the server is running\n"
  99 #ifdef __linux__
 100 "Kernel NBD client support:\n"
 101 "  -c, --connect=DEV         connect FILE to the local NBD device DEV\n"
 102 "  -d, --disconnect          disconnect the specified device\n"
 103 "\n"
 104 #endif
 105 "\n"
 106 "Block device options:\n"
 107 "  -f, --format=FORMAT       set image format (raw, qcow2, ...)\n"
 108 "  -r, --read-only           export read-only\n"
 109 "  -s, --snapshot            use FILE as an external snapshot, create a temporary\n"
 110 "                            file with backing_file=FILE, redirect the write to\n"
 111 "                            the temporary one\n"
 112 "  -l, --load-snapshot=SNAPSHOT_PARAM\n"
 113 "                            load an internal snapshot inside FILE and export it\n"
 114 "                            as an read-only device, SNAPSHOT_PARAM format is\n"
 115 "                            'snapshot.id=[ID],snapshot.name=[NAME]', or\n"
 116 "                            '[ID_OR_NAME]'\n"
 117 "  -n, --nocache             disable host cache\n"
 118 "      --cache=MODE          set cache mode (none, writeback, ...)\n"
 119 "      --aio=MODE            set AIO mode (native or threads)\n"
 120 "      --discard=MODE        set discard mode (ignore, unmap)\n"
 121 "      --detect-zeroes=MODE  set detect-zeroes mode (off, on, unmap)\n"
 122 "      --image-opts          treat FILE as a full set of image options\n"
 123 "\n"
 124 "Report bugs to <qemu-devel@nongnu.org>\n"
 125     , name, NBD_DEFAULT_PORT, "DEVICE");
 126 }
 127
 128 static void version(const char *name)
 129 {
 130     printf(
 131 "%s version 0.0.1\n"
 132 "Written by Anthony Liguori.\n"
 133 "\n"
 134 "Copyright (C) 2006 Anthony Liguori <anthony@codemonkey.ws>.\n"
 135 "This is free software; see the source for copying conditions.  There is NO\n"
 136 "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
 137     , name);
 138 }
 139
 140 struct partition_record
 141 {
 142     uint8_t bootable;
 143     uint8_t start_head;
 144     uint32_t start_cylinder;
 145     uint8_t start_sector;
 146     uint8_t system;
 147     uint8_t end_head;
 148     uint8_t end_cylinder;
 149     uint8_t end_sector;
 150     uint32_t start_sector_abs;
 151     uint32_t nb_sectors_abs;
 152 };
 153
 154 static void read_partition(uint8_t *p, struct partition_record *r)
 155 {
 156     r->bootable = p[0];
 157     r->start_head = p[1];
 158     r->start_cylinder = p[3] | ((p[2] << 2) & 0x0300);
 159     r->start_sector = p[2] & 0x3f;
 160     r->system = p[4];
 161     r->end_head = p[5];
 162     r->end_cylinder = p[7] | ((p[6] << 2) & 0x300);
 163     r->end_sector = p[6] & 0x3f;
 164
 165     r->start_sector_abs = ldl_le_p(p + 8);
 166     r->nb_sectors_abs   = ldl_le_p(p + 12);
 167 }
 168
 169 static int find_partition(BlockBackend *blk, int partition,
 170                           off_t *offset, off_t *size)
 171 {
 172     struct partition_record mbr[4];
 173     uint8_t data[MBR_SIZE];
 174     int i;
 175     int ext_partnum = 4;
 176     int ret;
 177
 178     ret = blk_pread(blk, 0, data, sizeof(data));
 179     if (ret < 0) {
 180         error_report("error while reading: %s", strerror(-ret));
 181         exit(EXIT_FAILURE);
 182     }
 183
 184     if (data[510] != 0x55 || data[511] != 0xaa) {
 185         return -EINVAL;
 186     }
 187
 188     for (i = 0; i < 4; i++) {
 189         read_partition(&data[446 + 16 * i], &mbr[i]);
 190
 191         if (!mbr[i].system || !mbr[i].nb_sectors_abs) {
 192             continue;
 193         }
 194
 195         if (mbr[i].system == 0xF || mbr[i].system == 0x5) {
 196             struct partition_record ext[4];
 197             uint8_t data1[MBR_SIZE];
 198             int j;
 199
 200             ret = blk_pread(blk, mbr[i].start_sector_abs * MBR_SIZE,
 201                             data1, sizeof(data1));
 202             if (ret < 0) {
 203                 error_report("error while reading: %s", strerror(-ret));
 204                 exit(EXIT_FAILURE);
 205             }
 206
 207             for (j = 0; j < 4; j++) {
 208                 read_partition(&data1[446 + 16 * j], &ext[j]);
 209                 if (!ext[j].system || !ext[j].nb_sectors_abs) {
 210                     continue;
 211                 }
 212
 213                 if ((ext_partnum + j + 1) == partition) {
 214                     *offset = (uint64_t)ext[j].start_sector_abs << 9;
 215                     *size = (uint64_t)ext[j].nb_sectors_abs << 9;
 216                     return 0;
 217                 }
 218             }
 219             ext_partnum += 4;
 220         } else if ((i + 1) == partition) {
 221             *offset = (uint64_t)mbr[i].start_sector_abs << 9;
 222             *size = (uint64_t)mbr[i].nb_sectors_abs << 9;
 223             return 0;
 224         }
 225     }
 226
 227     return -ENOENT;
 228 }
 229
 230 static void termsig_handler(int signum)
 231 {
 232     atomic_cmpxchg(&state, RUNNING, TERMINATE);
 233     qemu_notify_event();
 234 }
 235
 236
 237 static void *show_parts(void *arg)
 238 {
 239     char *device = arg;
 240     int nbd;
 241
 242     /* linux just needs an open() to trigger
 243      * the partition table update
 244      * but remember to load the module with max_part != 0 :
 245      *     modprobe nbd max_part=63
 246      */
 247     nbd = open(device, O_RDWR);
 248     if (nbd >= 0) {
 249         close(nbd);
 250     }
 251     return NULL;
 252 }
 253
 254 static void *nbd_client_thread(void *arg)
 255 {
 256     char *device = arg;
 257     off_t size;
 258     uint16_t nbdflags;
 259     QIOChannelSocket *sioc;
 260     int fd;
 261     int ret;
 262     pthread_t show_parts_thread;
 263     Error *local_error = NULL;
 264
 265     sioc = qio_channel_socket_new();
 266     if (qio_channel_socket_connect_sync(sioc,
 267                                         saddr,
 268                                         &local_error) < 0) {
 269         error_report_err(local_error);
 270         goto out;
 271     }
 272
 273     ret = nbd_receive_negotiate(QIO_CHANNEL(sioc), NULL, &nbdflags,
 274                                 NULL, NULL, NULL,
 275                                 &size, &local_error);
 276     if (ret < 0) {
 277         if (local_error) {
 278             error_report_err(local_error);
 279         }
 280         goto out_socket;
 281     }
 282
 283     fd = open(device, O_RDWR);
 284     if (fd < 0) {
 285         /* Linux-only, we can use %m in printf.  */
 286         error_report("Failed to open %s: %m", device);
 287         goto out_socket;
 288     }
 289
 290     ret = nbd_init(fd, sioc, nbdflags, size);
 291     if (ret < 0) {
 292         goto out_fd;
 293     }
 294
 295     /* update partition table */
 296     pthread_create(&show_parts_thread, NULL, show_parts, device);
 297
 298     if (verbose) {
 299         fprintf(stderr, "NBD device %s is now connected to %s\n",
 300                 device, srcpath);
 301     } else {
 302         /* Close stderr so that the qemu-nbd process exits.  */
 303         dup2(STDOUT_FILENO, STDERR_FILENO);
 304     }
 305
 306     ret = nbd_client(fd);
 307     if (ret) {
 308         goto out_fd;
 309     }
 310     close(fd);
 311     object_unref(OBJECT(sioc));
 312     kill(getpid(), SIGTERM);
 313     return (void *) EXIT_SUCCESS;
 314
 315 out_fd:
 316     close(fd);
 317 out_socket:
 318     object_unref(OBJECT(sioc));
 319 out:
 320     kill(getpid(), SIGTERM);
 321     return (void *) EXIT_FAILURE;
 322 }
 323
 324 static int nbd_can_accept(void)
 325 {
 326     return nb_fds < shared;
 327 }
 328
 329 static void nbd_export_closed(NBDExport *exp)
 330 {
 331     assert(state == TERMINATING);
 332     state = TERMINATED;
 333 }
 334
 335 static void nbd_update_server_watch(void);
 336
 337 static void nbd_client_closed(NBDClient *client)
 338 {
 339     nb_fds--;
 340     if (nb_fds == 0 && !persistent && state == RUNNING) {
 341         state = TERMINATE;
 342     }
 343     nbd_update_server_watch();
 344     nbd_client_put(client);
 345 }
 346
 347 static gboolean nbd_accept(QIOChannel *ioc, GIOCondition cond, gpointer opaque)
 348 {
 349     QIOChannelSocket *cioc;
 350
 351     cioc = qio_channel_socket_accept(QIO_CHANNEL_SOCKET(ioc),
 352                                      NULL);
 353     if (!cioc) {
 354         return TRUE;
 355     }
 356
 357     if (state >= TERMINATE) {
 358         object_unref(OBJECT(cioc));
 359         return TRUE;
 360     }
 361
 362     nb_fds++;
 363     nbd_update_server_watch();
 364     nbd_client_new(newproto ? NULL : exp, cioc,
 365                    tlscreds, NULL, nbd_client_closed);
 366     object_unref(OBJECT(cioc));
 367
 368     return TRUE;
 369 }
 370
 371 static void nbd_update_server_watch(void)
 372 {
 373     if (nbd_can_accept()) {
 374         if (server_watch == -1) {
 375             server_watch = qio_channel_add_watch(QIO_CHANNEL(server_ioc),
 376                                                  G_IO_IN,
 377                                                  nbd_accept,
 378                                                  NULL, NULL);
 379         }
 380     } else {
 381         if (server_watch != -1) {
 382             g_source_remove(server_watch);
 383             server_watch = -1;
 384         }
 385     }
 386 }
 387
 388
 389 static SocketAddress *nbd_build_socket_address(const char *sockpath,
 390                                                const char *bindto,
 391                                                const char *port)
 392 {
 393     SocketAddress *saddr;
 394
 395     saddr = g_new0(SocketAddress, 1);
 396     if (sockpath) {
 397         saddr->type = SOCKET_ADDRESS_KIND_UNIX;
 398         saddr->u.q_unix.data = g_new0(UnixSocketAddress, 1);
 399         saddr->u.q_unix.data->path = g_strdup(sockpath);
 400     } else {
 401         InetSocketAddress *inet;
 402         saddr->type = SOCKET_ADDRESS_KIND_INET;
 403         inet = saddr->u.inet.data = g_new0(InetSocketAddress, 1);
 404         inet->host = g_strdup(bindto);
 405         if (port) {
 406             inet->port = g_strdup(port);
 407         } else  {
 408             inet->port = g_strdup_printf("%d", NBD_DEFAULT_PORT);
 409         }
 410     }
 411
 412     return saddr;
 413 }
 414
 415
 416 static QemuOptsList file_opts = {
 417     .name = "file",
 418     .implied_opt_name = "file",
 419     .head = QTAILQ_HEAD_INITIALIZER(file_opts.head),
 420     .desc = {
 421         /* no elements => accept any params */
 422         { /* end of list */ }
 423     },
 424 };
 425
 426 static QemuOptsList qemu_object_opts = {
 427     .name = "object",
 428     .implied_opt_name = "qom-type",
 429     .head = QTAILQ_HEAD_INITIALIZER(qemu_object_opts.head),
 430     .desc = {
 431         { }
 432     },
 433 };
 434
 435
 436
 437 static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, Error **errp)
 438 {
 439     Object *obj;
 440     QCryptoTLSCreds *creds;
 441
 442     obj = object_resolve_path_component(
 443         object_get_objects_root(), id);
 444     if (!obj) {
 445         error_setg(errp, "No TLS credentials with id '%s'",
 446                    id);
 447         return NULL;
 448     }
 449     creds = (QCryptoTLSCreds *)
 450         object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
 451     if (!creds) {
 452         error_setg(errp, "Object with id '%s' is not TLS credentials",
 453                    id);
 454         return NULL;
 455     }
 456
 457     if (creds->endpoint != QCRYPTO_TLS_CREDS_ENDPOINT_SERVER) {
 458         error_setg(errp,
 459                    "Expecting TLS credentials with a server endpoint");
 460         return NULL;
 461     }
 462     object_ref(obj);
 463     return creds;
 464 }
 465
 466
 467 int main(int argc, char **argv)
 468 {
 469     BlockBackend *blk;
 470     BlockDriverState *bs;
 471     off_t dev_offset = 0;
 472     uint16_t nbdflags = 0;
 473     bool disconnect = false;
 474     const char *bindto = "0.0.0.0";
 475     const char *port = NULL;
 476     char *sockpath = NULL;
 477     char *device = NULL;
 478     off_t fd_size;
 479     QemuOpts *sn_opts = NULL;
 480     const char *sn_id_or_name = NULL;
 481     const char *sopt = "hVb:o:p:rsnP:c:dvk:e:f:tl:x:T:D:";
 482     struct option lopt[] = {
 483         { "help", no_argument, NULL, 'h' },
 484         { "version", no_argument, NULL, 'V' },
 485         { "bind", required_argument, NULL, 'b' },
 486         { "port", required_argument, NULL, 'p' },
 487         { "socket", required_argument, NULL, 'k' },
 488         { "offset", required_argument, NULL, 'o' },
 489         { "read-only", no_argument, NULL, 'r' },
 490         { "partition", required_argument, NULL, 'P' },
 491         { "connect", required_argument, NULL, 'c' },
 492         { "disconnect", no_argument, NULL, 'd' },
 493         { "snapshot", no_argument, NULL, 's' },
 494         { "load-snapshot", required_argument, NULL, 'l' },
 495         { "nocache", no_argument, NULL, 'n' },
 496         { "cache", required_argument, NULL, QEMU_NBD_OPT_CACHE },
 497         { "aio", required_argument, NULL, QEMU_NBD_OPT_AIO },
 498         { "discard", required_argument, NULL, QEMU_NBD_OPT_DISCARD },
 499         { "detect-zeroes", required_argument, NULL,
 500           QEMU_NBD_OPT_DETECT_ZEROES },
 501         { "shared", required_argument, NULL, 'e' },
 502         { "format", required_argument, NULL, 'f' },
 503         { "persistent", no_argument, NULL, 't' },
 504         { "verbose", no_argument, NULL, 'v' },
 505         { "object", required_argument, NULL, QEMU_NBD_OPT_OBJECT },
 506         { "export-name", required_argument, NULL, 'x' },
 507         { "description", required_argument, NULL, 'D' },
 508         { "tls-creds", required_argument, NULL, QEMU_NBD_OPT_TLSCREDS },
 509         { "image-opts", no_argument, NULL, QEMU_NBD_OPT_IMAGE_OPTS },
 510         { "trace", required_argument, NULL, 'T' },
 511         { "fork", no_argument, NULL, QEMU_NBD_OPT_FORK },
 512         { NULL, 0, NULL, 0 }
 513     };
 514     int ch;
 515     int opt_ind = 0;
 516     char *end;
 517     int flags = BDRV_O_RDWR;
 518     int partition = -1;
 519     int ret = 0;
 520     bool seen_cache = false;
 521     bool seen_discard = false;
 522     bool seen_aio = false;
 523     pthread_t client_thread;
 524     const char *fmt = NULL;
 525     Error *local_err = NULL;
 526     BlockdevDetectZeroesOptions detect_zeroes = BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
 527     QDict *options = NULL;
 528     const char *export_name = NULL;
 529     const char *export_description = NULL;
 530     const char *tlscredsid = NULL;
 531     bool imageOpts = false;
 532     bool writethrough = true;
 533     char *trace_file = NULL;
 534     bool fork_process = false;
 535     int old_stderr = -1;
 536
 537     /* The client thread uses SIGTERM to interrupt the server.  A signal
 538      * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
 539      */
 540     struct sigaction sa_sigterm;
 541     memset(&sa_sigterm, 0, sizeof(sa_sigterm));
 542     sa_sigterm.sa_handler = termsig_handler;
 543     sigaction(SIGTERM, &sa_sigterm, NULL);
 544
 545     module_call_init(MODULE_INIT_TRACE);
 546     qcrypto_init(&error_fatal);
 547
 548     module_call_init(MODULE_INIT_QOM);
 549     qemu_add_opts(&qemu_object_opts);
 550     qemu_add_opts(&qemu_trace_opts);
 551     qemu_init_exec_dir(argv[0]);
 552
 553     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 554         switch (ch) {
 555         case 's':
 556             flags |= BDRV_O_SNAPSHOT;
 557             break;
 558         case 'n':
 559             optarg = (char *) "none";
 560             /* fallthrough */
 561         case QEMU_NBD_OPT_CACHE:
 562             if (seen_cache) {
 563                 error_report("-n and --cache can only be specified once");
 564                 exit(EXIT_FAILURE);
 565             }
 566             seen_cache = true;
 567             if (bdrv_parse_cache_mode(optarg, &flags, &writethrough) == -1) {
 568                 error_report("Invalid cache mode `%s'", optarg);
 569                 exit(EXIT_FAILURE);
 570             }
 571             break;
 572         case QEMU_NBD_OPT_AIO:
 573             if (seen_aio) {
 574                 error_report("--aio can only be specified once");
 575                 exit(EXIT_FAILURE);
 576             }
 577             seen_aio = true;
 578             if (!strcmp(optarg, "native")) {
 579                 flags |= BDRV_O_NATIVE_AIO;
 580             } else if (!strcmp(optarg, "threads")) {
 581                 /* this is the default */
 582             } else {
 583                error_report("invalid aio mode `%s'", optarg);
 584                exit(EXIT_FAILURE);
 585             }
 586             break;
 587         case QEMU_NBD_OPT_DISCARD:
 588             if (seen_discard) {
 589                 error_report("--discard can only be specified once");
 590                 exit(EXIT_FAILURE);
 591             }
 592             seen_discard = true;
 593             if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
 594                 error_report("Invalid discard mode `%s'", optarg);
 595                 exit(EXIT_FAILURE);
 596             }
 597             break;
 598         case QEMU_NBD_OPT_DETECT_ZEROES:
 599             detect_zeroes =
 600                 qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
 601                                 optarg,
 602                                 BLOCKDEV_DETECT_ZEROES_OPTIONS__MAX,
 603                                 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
 604                                 &local_err);
 605             if (local_err) {
 606                 error_reportf_err(local_err,
 607                                   "Failed to parse detect_zeroes mode: ");
 608                 exit(EXIT_FAILURE);
 609             }
 610             if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
 611                 !(flags & BDRV_O_UNMAP)) {
 612                 error_report("setting detect-zeroes to unmap is not allowed "
 613                              "without setting discard operation to unmap");
 614                 exit(EXIT_FAILURE);
 615             }
 616             break;
 617         case 'b':
 618             bindto = optarg;
 619             break;
 620         case 'p':
 621             port = optarg;
 622             break;
 623         case 'o':
 624                 dev_offset = strtoll (optarg, &end, 0);
 625             if (*end) {
 626                 error_report("Invalid offset `%s'", optarg);
 627                 exit(EXIT_FAILURE);
 628             }
 629             if (dev_offset < 0) {
 630                 error_report("Offset must be positive `%s'", optarg);
 631                 exit(EXIT_FAILURE);
 632             }
 633             break;
 634         case 'l':
 635             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
 636                 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
 637                                                   optarg, false);
 638                 if (!sn_opts) {
 639                     error_report("Failed in parsing snapshot param `%s'",
 640                                  optarg);
 641                     exit(EXIT_FAILURE);
 642                 }
 643             } else {
 644                 sn_id_or_name = optarg;
 645             }
 646             /* fall through */
 647         case 'r':
 648             nbdflags |= NBD_FLAG_READ_ONLY;
 649             flags &= ~BDRV_O_RDWR;
 650             break;
 651         case 'P':
 652             partition = strtol(optarg, &end, 0);
 653             if (*end) {
 654                 error_report("Invalid partition `%s'", optarg);
 655                 exit(EXIT_FAILURE);
 656             }
 657             if (partition < 1 || partition > 8) {
 658                 error_report("Invalid partition %d", partition);
 659                 exit(EXIT_FAILURE);
 660             }
 661             break;
 662         case 'k':
 663             sockpath = optarg;
 664             if (sockpath[0] != '/') {
 665                 error_report("socket path must be absolute");
 666                 exit(EXIT_FAILURE);
 667             }
 668             break;
 669         case 'd':
 670             disconnect = true;
 671             break;
 672         case 'c':
 673             device = optarg;
 674             break;
 675         case 'e':
 676             shared = strtol(optarg, &end, 0);
 677             if (*end) {
 678                 error_report("Invalid shared device number '%s'", optarg);
 679                 exit(EXIT_FAILURE);
 680             }
 681             if (shared < 1) {
 682                 error_report("Shared device number must be greater than 0");
 683                 exit(EXIT_FAILURE);
 684             }
 685             break;
 686         case 'f':
 687             fmt = optarg;
 688             break;
 689         case 't':
 690             persistent = 1;
 691             break;
 692         case 'x':
 693             export_name = optarg;
 694             break;
 695         case 'D':
 696             export_description = optarg;
 697             break;
 698         case 'v':
 699             verbose = 1;
 700             break;
 701         case 'V':
 702             version(argv[0]);
 703             exit(0);
 704             break;
 705         case 'h':
 706             usage(argv[0]);
 707             exit(0);
 708             break;
 709         case '?':
 710             error_report("Try `%s --help' for more information.", argv[0]);
 711             exit(EXIT_FAILURE);
 712         case QEMU_NBD_OPT_OBJECT: {
 713             QemuOpts *opts;
 714             opts = qemu_opts_parse_noisily(&qemu_object_opts,
 715                                            optarg, true);
 716             if (!opts) {
 717                 exit(EXIT_FAILURE);
 718             }
 719         }   break;
 720         case QEMU_NBD_OPT_TLSCREDS:
 721             tlscredsid = optarg;
 722             break;
 723         case QEMU_NBD_OPT_IMAGE_OPTS:
 724             imageOpts = true;
 725             break;
 726         case 'T':
 727             g_free(trace_file);
 728             trace_file = trace_opt_parse(optarg);
 729             break;
 730         case QEMU_NBD_OPT_FORK:
 731             fork_process = true;
 732             break;
 733         }
 734     }
 735
 736     if ((argc - optind) != 1) {
 737         error_report("Invalid number of arguments");
 738         error_printf("Try `%s --help' for more information.\n", argv[0]);
 739         exit(EXIT_FAILURE);
 740     }
 741
 742     if (qemu_opts_foreach(&qemu_object_opts,
 743                           user_creatable_add_opts_foreach,
 744                           NULL, NULL)) {
 745         exit(EXIT_FAILURE);
 746     }
 747
 748     if (!trace_init_backends()) {
 749         exit(1);
 750     }
 751     trace_init_file(trace_file);
 752     qemu_set_log(LOG_TRACE);
 753
 754     if (tlscredsid) {
 755         if (sockpath) {
 756             error_report("TLS is only supported with IPv4/IPv6");
 757             exit(EXIT_FAILURE);
 758         }
 759         if (device) {
 760             error_report("TLS is not supported with a host device");
 761             exit(EXIT_FAILURE);
 762         }
 763         if (!export_name) {
 764             /* Set the default NBD protocol export name, since
 765              * we *must* use new style protocol for TLS */
 766             export_name = "";
 767         }
 768         tlscreds = nbd_get_tls_creds(tlscredsid, &local_err);
 769         if (local_err) {
 770             error_report("Failed to get TLS creds %s",
 771                          error_get_pretty(local_err));
 772             exit(EXIT_FAILURE);
 773         }
 774     }
 775
 776     if (disconnect) {
 777         int nbdfd = open(argv[optind], O_RDWR);
 778         if (nbdfd < 0) {
 779             error_report("Cannot open %s: %s", argv[optind],
 780                          strerror(errno));
 781             exit(EXIT_FAILURE);
 782         }
 783         nbd_disconnect(nbdfd);
 784
 785         close(nbdfd);
 786
 787         printf("%s disconnected\n", argv[optind]);
 788
 789         return 0;
 790     }
 791
 792     if ((device && !verbose) || fork_process) {
 793         int stderr_fd[2];
 794         pid_t pid;
 795         int ret;
 796
 797         if (qemu_pipe(stderr_fd) < 0) {
 798             error_report("Error setting up communication pipe: %s",
 799                          strerror(errno));
 800             exit(EXIT_FAILURE);
 801         }
 802
 803         /* Now daemonize, but keep a communication channel open to
 804          * print errors and exit with the proper status code.
 805          */
 806         pid = fork();
 807         if (pid < 0) {
 808             error_report("Failed to fork: %s", strerror(errno));
 809             exit(EXIT_FAILURE);
 810         } else if (pid == 0) {
 811             close(stderr_fd[0]);
 812             ret = qemu_daemon(1, 0);
 813
 814             /* Temporarily redirect stderr to the parent's pipe...  */
 815             old_stderr = dup(STDERR_FILENO);
 816             dup2(stderr_fd[1], STDERR_FILENO);
 817             if (ret < 0) {
 818                 error_report("Failed to daemonize: %s", strerror(errno));
 819                 exit(EXIT_FAILURE);
 820             }
 821
 822             /* ... close the descriptor we inherited and go on.  */
 823             close(stderr_fd[1]);
 824         } else {
 825             bool errors = false;
 826             char *buf;
 827
 828             /* In the parent.  Print error messages from the child until
 829              * it closes the pipe.
 830              */
 831             close(stderr_fd[1]);
 832             buf = g_malloc(1024);
 833             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
 834                 errors = true;
 835                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
 836                 if (ret < 0) {
 837                     exit(EXIT_FAILURE);
 838                 }
 839             }
 840             if (ret < 0) {
 841                 error_report("Cannot read from daemon: %s",
 842                              strerror(errno));
 843                 exit(EXIT_FAILURE);
 844             }
 845
 846             /* Usually the daemon should not print any message.
 847              * Exit with zero status in that case.
 848              */
 849             exit(errors);
 850         }
 851     }
 852
 853     if (device != NULL && sockpath == NULL) {
 854         sockpath = g_malloc(128);
 855         snprintf(sockpath, 128, SOCKET_PATH, basename(device));
 856     }
 857
 858     saddr = nbd_build_socket_address(sockpath, bindto, port);
 859
 860     if (qemu_init_main_loop(&local_err)) {
 861         error_report_err(local_err);
 862         exit(EXIT_FAILURE);
 863     }
 864     bdrv_init();
 865     atexit(bdrv_close_all);
 866
 867     srcpath = argv[optind];
 868     if (imageOpts) {
 869         QemuOpts *opts;
 870         if (fmt) {
 871             error_report("--image-opts and -f are mutually exclusive");
 872             exit(EXIT_FAILURE);
 873         }
 874         opts = qemu_opts_parse_noisily(&file_opts, srcpath, true);
 875         if (!opts) {
 876             qemu_opts_reset(&file_opts);
 877             exit(EXIT_FAILURE);
 878         }
 879         options = qemu_opts_to_qdict(opts, NULL);
 880         qemu_opts_reset(&file_opts);
 881         blk = blk_new_open(NULL, NULL, options, flags, &local_err);
 882     } else {
 883         if (fmt) {
 884             options = qdict_new();
 885             qdict_put(options, "driver", qstring_from_str(fmt));
 886         }
 887         blk = blk_new_open(srcpath, NULL, options, flags, &local_err);
 888     }
 889
 890     if (!blk) {
 891         error_reportf_err(local_err, "Failed to blk_new_open '%s': ",
 892                           argv[optind]);
 893         exit(EXIT_FAILURE);
 894     }
 895     bs = blk_bs(blk);
 896
 897     blk_set_enable_write_cache(blk, !writethrough);
 898
 899     if (sn_opts) {
 900         ret = bdrv_snapshot_load_tmp(bs,
 901                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
 902                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
 903                                      &local_err);
 904     } else if (sn_id_or_name) {
 905         ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
 906                                                    &local_err);
 907     }
 908     if (ret < 0) {
 909         error_reportf_err(local_err, "Failed to load snapshot: ");
 910         exit(EXIT_FAILURE);
 911     }
 912
 913     bs->detect_zeroes = detect_zeroes;
 914     fd_size = blk_getlength(blk);
 915     if (fd_size < 0) {
 916         error_report("Failed to determine the image length: %s",
 917                      strerror(-fd_size));
 918         exit(EXIT_FAILURE);
 919     }
 920
 921     if (dev_offset >= fd_size) {
 922         error_report("Offset (%lld) has to be smaller than the image size "
 923                      "(%lld)",
 924                      (long long int)dev_offset, (long long int)fd_size);
 925         exit(EXIT_FAILURE);
 926     }
 927     fd_size -= dev_offset;
 928
 929     if (partition != -1) {
 930         ret = find_partition(blk, partition, &dev_offset, &fd_size);
 931         if (ret < 0) {
 932             error_report("Could not find partition %d: %s", partition,
 933                          strerror(-ret));
 934             exit(EXIT_FAILURE);
 935         }
 936     }
 937
 938     exp = nbd_export_new(bs, dev_offset, fd_size, nbdflags, nbd_export_closed,
 939                          writethrough, NULL, &local_err);
 940     if (!exp) {
 941         error_report_err(local_err);
 942         exit(EXIT_FAILURE);
 943     }
 944     if (export_name) {
 945         nbd_export_set_name(exp, export_name);
 946         nbd_export_set_description(exp, export_description);
 947         newproto = true;
 948     } else if (export_description) {
 949         error_report("Export description requires an export name");
 950         exit(EXIT_FAILURE);
 951     }
 952
 953     server_ioc = qio_channel_socket_new();
 954     if (qio_channel_socket_listen_sync(server_ioc, saddr, &local_err) < 0) {
 955         object_unref(OBJECT(server_ioc));
 956         error_report_err(local_err);
 957         return 1;
 958     }
 959
 960     if (device) {
 961         int ret;
 962
 963         ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
 964         if (ret != 0) {
 965             error_report("Failed to create client thread: %s", strerror(ret));
 966             exit(EXIT_FAILURE);
 967         }
 968     } else {
 969         /* Shut up GCC warnings.  */
 970         memset(&client_thread, 0, sizeof(client_thread));
 971     }
 972
 973     nbd_update_server_watch();
 974
 975     /* now when the initialization is (almost) complete, chdir("/")
 976      * to free any busy filesystems */
 977     if (chdir("/") < 0) {
 978         error_report("Could not chdir to root directory: %s",
 979                      strerror(errno));
 980         exit(EXIT_FAILURE);
 981     }
 982
 983     if (fork_process) {
 984         dup2(old_stderr, STDERR_FILENO);
 985         close(old_stderr);
 986     }
 987
 988     state = RUNNING;
 989     do {
 990         main_loop_wait(false);
 991         if (state == TERMINATE) {
 992             state = TERMINATING;
 993             nbd_export_close(exp);
 994             nbd_export_put(exp);
 995             exp = NULL;
 996         }
 997     } while (state != TERMINATED);
 998
 999     blk_unref(blk);
1000     if (sockpath) {
1001         unlink(sockpath);
1002     }
1003
1004     qemu_opts_del(sn_opts);
1005
1006     if (device) {
1007         void *ret;
1008         pthread_join(client_thread, &ret);
1009         exit(ret != NULL);
1010     } else {
1011         exit(EXIT_SUCCESS);
1012     }
1013 }