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 <getopt.h>
  21 #include <libgen.h>
  22 #include <pthread.h>
  23
  24 #include "qapi/error.h"
  25 #include "qemu-common.h"
  26 #include "qemu/cutils.h"
  27 #include "sysemu/block-backend.h"
  28 #include "block/block_int.h"
  29 #include "block/nbd.h"
  30 #include "qemu/main-loop.h"
  31 #include "qemu/error-report.h"
  32 #include "qemu/config-file.h"
  33 #include "qemu/bswap.h"
  34 #include "qemu/log.h"
  35 #include "qemu/systemd.h"
  36 #include "block/snapshot.h"
  37 #include "qapi/qmp/qstring.h"
  38 #include "qom/object_interfaces.h"
  39 #include "io/channel-socket.h"
  40 #include "io/net-listener.h"
  41 #include "crypto/init.h"
  42 #include "trace/control.h"
  43 #include "qemu-version.h"
  44
  45 #define SOCKET_PATH                "/var/lock/qemu-nbd-%s"
  46 #define QEMU_NBD_OPT_CACHE         256
  47 #define QEMU_NBD_OPT_AIO           257
  48 #define QEMU_NBD_OPT_DISCARD       258
  49 #define QEMU_NBD_OPT_DETECT_ZEROES 259
  50 #define QEMU_NBD_OPT_OBJECT        260
  51 #define QEMU_NBD_OPT_TLSCREDS      261
  52 #define QEMU_NBD_OPT_IMAGE_OPTS    262
  53 #define QEMU_NBD_OPT_FORK          263
  54
  55 #define MBR_SIZE 512
  56
  57 static NBDExport *exp;
  58 static bool newproto;
  59 static int verbose;
  60 static char *srcpath;
  61 static SocketAddress *saddr;
  62 static int persistent = 0;
  63 static enum { RUNNING, TERMINATE, TERMINATING, TERMINATED } state;
  64 static int shared = 1;
  65 static int nb_fds;
  66 static QIONetListener *server;
  67 static QCryptoTLSCreds *tlscreds;
  68
  69 static void usage(const char *name)
  70 {
  71     (printf) (
  72 "Usage: %s [OPTIONS] FILE\n"
  73 "QEMU Disk Network Block Device Server\n"
  74 "\n"
  75 "  -h, --help                display this help and exit\n"
  76 "  -V, --version             output version information and exit\n"
  77 "\n"
  78 "Connection properties:\n"
  79 "  -p, --port=PORT           port to listen on (default `%d')\n"
  80 "  -b, --bind=IFACE          interface to bind to (default `0.0.0.0')\n"
  81 "  -k, --socket=PATH         path to the unix socket\n"
  82 "                            (default '"SOCKET_PATH"')\n"
  83 "  -e, --shared=NUM          device can be shared by NUM clients (default '1')\n"
  84 "  -t, --persistent          don't exit on the last connection\n"
  85 "  -v, --verbose             display extra debugging information\n"
  86 "  -x, --export-name=NAME    expose export by name\n"
  87 "  -D, --description=TEXT    with -x, also export a human-readable description\n"
  88 "\n"
  89 "Exposing part of the image:\n"
  90 "  -o, --offset=OFFSET       offset into the image\n"
  91 "  -P, --partition=NUM       only expose partition NUM\n"
  92 "\n"
  93 "General purpose options:\n"
  94 "  --object type,id=ID,...   define an object such as 'secret' for providing\n"
  95 "                            passwords and/or encryption keys\n"
  96 "  -T, --trace [[enable=]<pattern>][,events=<file>][,file=<file>]\n"
  97 "                            specify tracing options\n"
  98 "  --fork                    fork off the server process and exit the parent\n"
  99 "                            once the server is running\n"
 100 #ifdef __linux__
 101 "Kernel NBD client support:\n"
 102 "  -c, --connect=DEV         connect FILE to the local NBD device DEV\n"
 103 "  -d, --disconnect          disconnect the specified device\n"
 104 "\n"
 105 #endif
 106 "\n"
 107 "Block device options:\n"
 108 "  -f, --format=FORMAT       set image format (raw, qcow2, ...)\n"
 109 "  -r, --read-only           export read-only\n"
 110 "  -s, --snapshot            use FILE as an external snapshot, create a temporary\n"
 111 "                            file with backing_file=FILE, redirect the write to\n"
 112 "                            the temporary one\n"
 113 "  -l, --load-snapshot=SNAPSHOT_PARAM\n"
 114 "                            load an internal snapshot inside FILE and export it\n"
 115 "                            as an read-only device, SNAPSHOT_PARAM format is\n"
 116 "                            'snapshot.id=[ID],snapshot.name=[NAME]', or\n"
 117 "                            '[ID_OR_NAME]'\n"
 118 "  -n, --nocache             disable host cache\n"
 119 "      --cache=MODE          set cache mode (none, writeback, ...)\n"
 120 "      --aio=MODE            set AIO mode (native or threads)\n"
 121 "      --discard=MODE        set discard mode (ignore, unmap)\n"
 122 "      --detect-zeroes=MODE  set detect-zeroes mode (off, on, unmap)\n"
 123 "      --image-opts          treat FILE as a full set of image options\n"
 124 "\n"
 125 QEMU_HELP_BOTTOM "\n"
 126     , name, NBD_DEFAULT_PORT, "DEVICE");
 127 }
 128
 129 static void version(const char *name)
 130 {
 131     printf(
 132 "%s " QEMU_VERSION QEMU_PKGVERSION "\n"
 133 "Written by Anthony Liguori.\n"
 134 "\n"
 135 QEMU_COPYRIGHT "\n"
 136 "This is free software; see the source for copying conditions.  There is NO\n"
 137 "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
 138     , name);
 139 }
 140
 141 struct partition_record
 142 {
 143     uint8_t bootable;
 144     uint8_t start_head;
 145     uint32_t start_cylinder;
 146     uint8_t start_sector;
 147     uint8_t system;
 148     uint8_t end_head;
 149     uint8_t end_cylinder;
 150     uint8_t end_sector;
 151     uint32_t start_sector_abs;
 152     uint32_t nb_sectors_abs;
 153 };
 154
 155 static void read_partition(uint8_t *p, struct partition_record *r)
 156 {
 157     r->bootable = p[0];
 158     r->start_head = p[1];
 159     r->start_cylinder = p[3] | ((p[2] << 2) & 0x0300);
 160     r->start_sector = p[2] & 0x3f;
 161     r->system = p[4];
 162     r->end_head = p[5];
 163     r->end_cylinder = p[7] | ((p[6] << 2) & 0x300);
 164     r->end_sector = p[6] & 0x3f;
 165
 166     r->start_sector_abs = ldl_le_p(p + 8);
 167     r->nb_sectors_abs   = ldl_le_p(p + 12);
 168 }
 169
 170 static int find_partition(BlockBackend *blk, int partition,
 171                           off_t *offset, off_t *size)
 172 {
 173     struct partition_record mbr[4];
 174     uint8_t data[MBR_SIZE];
 175     int i;
 176     int ext_partnum = 4;
 177     int ret;
 178
 179     ret = blk_pread(blk, 0, data, sizeof(data));
 180     if (ret < 0) {
 181         error_report("error while reading: %s", strerror(-ret));
 182         exit(EXIT_FAILURE);
 183     }
 184
 185     if (data[510] != 0x55 || data[511] != 0xaa) {
 186         return -EINVAL;
 187     }
 188
 189     for (i = 0; i < 4; i++) {
 190         read_partition(&data[446 + 16 * i], &mbr[i]);
 191
 192         if (!mbr[i].system || !mbr[i].nb_sectors_abs) {
 193             continue;
 194         }
 195
 196         if (mbr[i].system == 0xF || mbr[i].system == 0x5) {
 197             struct partition_record ext[4];
 198             uint8_t data1[MBR_SIZE];
 199             int j;
 200
 201             ret = blk_pread(blk, mbr[i].start_sector_abs * MBR_SIZE,
 202                             data1, sizeof(data1));
 203             if (ret < 0) {
 204                 error_report("error while reading: %s", strerror(-ret));
 205                 exit(EXIT_FAILURE);
 206             }
 207
 208             for (j = 0; j < 4; j++) {
 209                 read_partition(&data1[446 + 16 * j], &ext[j]);
 210                 if (!ext[j].system || !ext[j].nb_sectors_abs) {
 211                     continue;
 212                 }
 213
 214                 if ((ext_partnum + j + 1) == partition) {
 215                     *offset = (uint64_t)ext[j].start_sector_abs << 9;
 216                     *size = (uint64_t)ext[j].nb_sectors_abs << 9;
 217                     return 0;
 218                 }
 219             }
 220             ext_partnum += 4;
 221         } else if ((i + 1) == partition) {
 222             *offset = (uint64_t)mbr[i].start_sector_abs << 9;
 223             *size = (uint64_t)mbr[i].nb_sectors_abs << 9;
 224             return 0;
 225         }
 226     }
 227
 228     return -ENOENT;
 229 }
 230
 231 static void termsig_handler(int signum)
 232 {
 233     atomic_cmpxchg(&state, RUNNING, TERMINATE);
 234     qemu_notify_event();
 235 }
 236
 237
 238 static void *show_parts(void *arg)
 239 {
 240     char *device = arg;
 241     int nbd;
 242
 243     /* linux just needs an open() to trigger
 244      * the partition table update
 245      * but remember to load the module with max_part != 0 :
 246      *     modprobe nbd max_part=63
 247      */
 248     nbd = open(device, O_RDWR);
 249     if (nbd >= 0) {
 250         close(nbd);
 251     }
 252     return NULL;
 253 }
 254
 255 static void *nbd_client_thread(void *arg)
 256 {
 257     char *device = arg;
 258     NBDExportInfo info = { .request_sizes = false, };
 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,
 274                                 NULL, NULL, NULL, &info, &local_error);
 275     if (ret < 0) {
 276         if (local_error) {
 277             error_report_err(local_error);
 278         }
 279         goto out_socket;
 280     }
 281
 282     fd = open(device, O_RDWR);
 283     if (fd < 0) {
 284         /* Linux-only, we can use %m in printf.  */
 285         error_report("Failed to open %s: %m", device);
 286         goto out_socket;
 287     }
 288
 289     ret = nbd_init(fd, sioc, &info, &local_error);
 290     if (ret < 0) {
 291         error_report_err(local_error);
 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 state == RUNNING && 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, bool negotiated)
 338 {
 339     nb_fds--;
 340     if (negotiated && nb_fds == 0 && !persistent && state == RUNNING) {
 341         state = TERMINATE;
 342     }
 343     nbd_update_server_watch();
 344     nbd_client_put(client);
 345 }
 346
 347 static void nbd_accept(QIONetListener *listener, QIOChannelSocket *cioc,
 348                        gpointer opaque)
 349 {
 350     if (state >= TERMINATE) {
 351         return;
 352     }
 353
 354     nb_fds++;
 355     nbd_update_server_watch();
 356     nbd_client_new(newproto ? NULL : exp, cioc,
 357                    tlscreds, NULL, nbd_client_closed);
 358 }
 359
 360 static void nbd_update_server_watch(void)
 361 {
 362     if (nbd_can_accept()) {
 363         qio_net_listener_set_client_func(server, nbd_accept, NULL, NULL);
 364     } else {
 365         qio_net_listener_set_client_func(server, NULL, NULL, NULL);
 366     }
 367 }
 368
 369
 370 static SocketAddress *nbd_build_socket_address(const char *sockpath,
 371                                                const char *bindto,
 372                                                const char *port)
 373 {
 374     SocketAddress *saddr;
 375
 376     saddr = g_new0(SocketAddress, 1);
 377     if (sockpath) {
 378         saddr->type = SOCKET_ADDRESS_TYPE_UNIX;
 379         saddr->u.q_unix.path = g_strdup(sockpath);
 380     } else {
 381         InetSocketAddress *inet;
 382         saddr->type = SOCKET_ADDRESS_TYPE_INET;
 383         inet = &saddr->u.inet;
 384         inet->host = g_strdup(bindto);
 385         if (port) {
 386             inet->port = g_strdup(port);
 387         } else  {
 388             inet->port = g_strdup_printf("%d", NBD_DEFAULT_PORT);
 389         }
 390     }
 391
 392     return saddr;
 393 }
 394
 395
 396 static QemuOptsList file_opts = {
 397     .name = "file",
 398     .implied_opt_name = "file",
 399     .head = QTAILQ_HEAD_INITIALIZER(file_opts.head),
 400     .desc = {
 401         /* no elements => accept any params */
 402         { /* end of list */ }
 403     },
 404 };
 405
 406 static QemuOptsList qemu_object_opts = {
 407     .name = "object",
 408     .implied_opt_name = "qom-type",
 409     .head = QTAILQ_HEAD_INITIALIZER(qemu_object_opts.head),
 410     .desc = {
 411         { }
 412     },
 413 };
 414
 415
 416
 417 static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, Error **errp)
 418 {
 419     Object *obj;
 420     QCryptoTLSCreds *creds;
 421
 422     obj = object_resolve_path_component(
 423         object_get_objects_root(), id);
 424     if (!obj) {
 425         error_setg(errp, "No TLS credentials with id '%s'",
 426                    id);
 427         return NULL;
 428     }
 429     creds = (QCryptoTLSCreds *)
 430         object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
 431     if (!creds) {
 432         error_setg(errp, "Object with id '%s' is not TLS credentials",
 433                    id);
 434         return NULL;
 435     }
 436
 437     if (creds->endpoint != QCRYPTO_TLS_CREDS_ENDPOINT_SERVER) {
 438         error_setg(errp,
 439                    "Expecting TLS credentials with a server endpoint");
 440         return NULL;
 441     }
 442     object_ref(obj);
 443     return creds;
 444 }
 445
 446 static void setup_address_and_port(const char **address, const char **port)
 447 {
 448     if (*address == NULL) {
 449         *address = "0.0.0.0";
 450     }
 451
 452     if (*port == NULL) {
 453         *port = stringify(NBD_DEFAULT_PORT);
 454     }
 455 }
 456
 457 /*
 458  * Check socket parameters compatibility when socket activation is used.
 459  */
 460 static const char *socket_activation_validate_opts(const char *device,
 461                                                    const char *sockpath,
 462                                                    const char *address,
 463                                                    const char *port)
 464 {
 465     if (device != NULL) {
 466         return "NBD device can't be set when using socket activation";
 467     }
 468
 469     if (sockpath != NULL) {
 470         return "Unix socket can't be set when using socket activation";
 471     }
 472
 473     if (address != NULL) {
 474         return "The interface can't be set when using socket activation";
 475     }
 476
 477     if (port != NULL) {
 478         return "TCP port number can't be set when using socket activation";
 479     }
 480
 481     return NULL;
 482 }
 483
 484 int main(int argc, char **argv)
 485 {
 486     BlockBackend *blk;
 487     BlockDriverState *bs;
 488     off_t dev_offset = 0;
 489     uint16_t nbdflags = 0;
 490     bool disconnect = false;
 491     const char *bindto = NULL;
 492     const char *port = NULL;
 493     char *sockpath = NULL;
 494     char *device = NULL;
 495     off_t fd_size;
 496     QemuOpts *sn_opts = NULL;
 497     const char *sn_id_or_name = NULL;
 498     const char *sopt = "hVb:o:p:rsnP:c:dvk:e:f:tl:x:T:D:";
 499     struct option lopt[] = {
 500         { "help", no_argument, NULL, 'h' },
 501         { "version", no_argument, NULL, 'V' },
 502         { "bind", required_argument, NULL, 'b' },
 503         { "port", required_argument, NULL, 'p' },
 504         { "socket", required_argument, NULL, 'k' },
 505         { "offset", required_argument, NULL, 'o' },
 506         { "read-only", no_argument, NULL, 'r' },
 507         { "partition", required_argument, NULL, 'P' },
 508         { "connect", required_argument, NULL, 'c' },
 509         { "disconnect", no_argument, NULL, 'd' },
 510         { "snapshot", no_argument, NULL, 's' },
 511         { "load-snapshot", required_argument, NULL, 'l' },
 512         { "nocache", no_argument, NULL, 'n' },
 513         { "cache", required_argument, NULL, QEMU_NBD_OPT_CACHE },
 514         { "aio", required_argument, NULL, QEMU_NBD_OPT_AIO },
 515         { "discard", required_argument, NULL, QEMU_NBD_OPT_DISCARD },
 516         { "detect-zeroes", required_argument, NULL,
 517           QEMU_NBD_OPT_DETECT_ZEROES },
 518         { "shared", required_argument, NULL, 'e' },
 519         { "format", required_argument, NULL, 'f' },
 520         { "persistent", no_argument, NULL, 't' },
 521         { "verbose", no_argument, NULL, 'v' },
 522         { "object", required_argument, NULL, QEMU_NBD_OPT_OBJECT },
 523         { "export-name", required_argument, NULL, 'x' },
 524         { "description", required_argument, NULL, 'D' },
 525         { "tls-creds", required_argument, NULL, QEMU_NBD_OPT_TLSCREDS },
 526         { "image-opts", no_argument, NULL, QEMU_NBD_OPT_IMAGE_OPTS },
 527         { "trace", required_argument, NULL, 'T' },
 528         { "fork", no_argument, NULL, QEMU_NBD_OPT_FORK },
 529         { NULL, 0, NULL, 0 }
 530     };
 531     int ch;
 532     int opt_ind = 0;
 533     char *end;
 534     int flags = BDRV_O_RDWR;
 535     int partition = -1;
 536     int ret = 0;
 537     bool seen_cache = false;
 538     bool seen_discard = false;
 539     bool seen_aio = false;
 540     pthread_t client_thread;
 541     const char *fmt = NULL;
 542     Error *local_err = NULL;
 543     BlockdevDetectZeroesOptions detect_zeroes = BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
 544     QDict *options = NULL;
 545     const char *export_name = NULL;
 546     const char *export_description = NULL;
 547     const char *tlscredsid = NULL;
 548     bool imageOpts = false;
 549     bool writethrough = true;
 550     char *trace_file = NULL;
 551     bool fork_process = false;
 552     int old_stderr = -1;
 553     unsigned socket_activation;
 554
 555     /* The client thread uses SIGTERM to interrupt the server.  A signal
 556      * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
 557      */
 558     struct sigaction sa_sigterm;
 559     memset(&sa_sigterm, 0, sizeof(sa_sigterm));
 560     sa_sigterm.sa_handler = termsig_handler;
 561     sigaction(SIGTERM, &sa_sigterm, NULL);
 562
 563 #ifdef CONFIG_POSIX
 564     signal(SIGPIPE, SIG_IGN);
 565 #endif
 566
 567     module_call_init(MODULE_INIT_TRACE);
 568     qcrypto_init(&error_fatal);
 569
 570     module_call_init(MODULE_INIT_QOM);
 571     qemu_add_opts(&qemu_object_opts);
 572     qemu_add_opts(&qemu_trace_opts);
 573     qemu_init_exec_dir(argv[0]);
 574
 575     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 576         switch (ch) {
 577         case 's':
 578             flags |= BDRV_O_SNAPSHOT;
 579             break;
 580         case 'n':
 581             optarg = (char *) "none";
 582             /* fallthrough */
 583         case QEMU_NBD_OPT_CACHE:
 584             if (seen_cache) {
 585                 error_report("-n and --cache can only be specified once");
 586                 exit(EXIT_FAILURE);
 587             }
 588             seen_cache = true;
 589             if (bdrv_parse_cache_mode(optarg, &flags, &writethrough) == -1) {
 590                 error_report("Invalid cache mode `%s'", optarg);
 591                 exit(EXIT_FAILURE);
 592             }
 593             break;
 594         case QEMU_NBD_OPT_AIO:
 595             if (seen_aio) {
 596                 error_report("--aio can only be specified once");
 597                 exit(EXIT_FAILURE);
 598             }
 599             seen_aio = true;
 600             if (!strcmp(optarg, "native")) {
 601                 flags |= BDRV_O_NATIVE_AIO;
 602             } else if (!strcmp(optarg, "threads")) {
 603                 /* this is the default */
 604             } else {
 605                error_report("invalid aio mode `%s'", optarg);
 606                exit(EXIT_FAILURE);
 607             }
 608             break;
 609         case QEMU_NBD_OPT_DISCARD:
 610             if (seen_discard) {
 611                 error_report("--discard can only be specified once");
 612                 exit(EXIT_FAILURE);
 613             }
 614             seen_discard = true;
 615             if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
 616                 error_report("Invalid discard mode `%s'", optarg);
 617                 exit(EXIT_FAILURE);
 618             }
 619             break;
 620         case QEMU_NBD_OPT_DETECT_ZEROES:
 621             detect_zeroes =
 622                 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
 623                                 optarg,
 624                                 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
 625                                 &local_err);
 626             if (local_err) {
 627                 error_reportf_err(local_err,
 628                                   "Failed to parse detect_zeroes mode: ");
 629                 exit(EXIT_FAILURE);
 630             }
 631             if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
 632                 !(flags & BDRV_O_UNMAP)) {
 633                 error_report("setting detect-zeroes to unmap is not allowed "
 634                              "without setting discard operation to unmap");
 635                 exit(EXIT_FAILURE);
 636             }
 637             break;
 638         case 'b':
 639             bindto = optarg;
 640             break;
 641         case 'p':
 642             port = optarg;
 643             break;
 644         case 'o':
 645                 dev_offset = strtoll (optarg, &end, 0);
 646             if (*end) {
 647                 error_report("Invalid offset `%s'", optarg);
 648                 exit(EXIT_FAILURE);
 649             }
 650             if (dev_offset < 0) {
 651                 error_report("Offset must be positive `%s'", optarg);
 652                 exit(EXIT_FAILURE);
 653             }
 654             break;
 655         case 'l':
 656             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
 657                 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
 658                                                   optarg, false);
 659                 if (!sn_opts) {
 660                     error_report("Failed in parsing snapshot param `%s'",
 661                                  optarg);
 662                     exit(EXIT_FAILURE);
 663                 }
 664             } else {
 665                 sn_id_or_name = optarg;
 666             }
 667             /* fall through */
 668         case 'r':
 669             nbdflags |= NBD_FLAG_READ_ONLY;
 670             flags &= ~BDRV_O_RDWR;
 671             break;
 672         case 'P':
 673             partition = strtol(optarg, &end, 0);
 674             if (*end) {
 675                 error_report("Invalid partition `%s'", optarg);
 676                 exit(EXIT_FAILURE);
 677             }
 678             if (partition < 1 || partition > 8) {
 679                 error_report("Invalid partition %d", partition);
 680                 exit(EXIT_FAILURE);
 681             }
 682             break;
 683         case 'k':
 684             sockpath = optarg;
 685             if (sockpath[0] != '/') {
 686                 error_report("socket path must be absolute");
 687                 exit(EXIT_FAILURE);
 688             }
 689             break;
 690         case 'd':
 691             disconnect = true;
 692             break;
 693         case 'c':
 694             device = optarg;
 695             break;
 696         case 'e':
 697             shared = strtol(optarg, &end, 0);
 698             if (*end) {
 699                 error_report("Invalid shared device number '%s'", optarg);
 700                 exit(EXIT_FAILURE);
 701             }
 702             if (shared < 1) {
 703                 error_report("Shared device number must be greater than 0");
 704                 exit(EXIT_FAILURE);
 705             }
 706             break;
 707         case 'f':
 708             fmt = optarg;
 709             break;
 710         case 't':
 711             persistent = 1;
 712             break;
 713         case 'x':
 714             export_name = optarg;
 715             break;
 716         case 'D':
 717             export_description = optarg;
 718             break;
 719         case 'v':
 720             verbose = 1;
 721             break;
 722         case 'V':
 723             version(argv[0]);
 724             exit(0);
 725             break;
 726         case 'h':
 727             usage(argv[0]);
 728             exit(0);
 729             break;
 730         case '?':
 731             error_report("Try `%s --help' for more information.", argv[0]);
 732             exit(EXIT_FAILURE);
 733         case QEMU_NBD_OPT_OBJECT: {
 734             QemuOpts *opts;
 735             opts = qemu_opts_parse_noisily(&qemu_object_opts,
 736                                            optarg, true);
 737             if (!opts) {
 738                 exit(EXIT_FAILURE);
 739             }
 740         }   break;
 741         case QEMU_NBD_OPT_TLSCREDS:
 742             tlscredsid = optarg;
 743             break;
 744         case QEMU_NBD_OPT_IMAGE_OPTS:
 745             imageOpts = true;
 746             break;
 747         case 'T':
 748             g_free(trace_file);
 749             trace_file = trace_opt_parse(optarg);
 750             break;
 751         case QEMU_NBD_OPT_FORK:
 752             fork_process = true;
 753             break;
 754         }
 755     }
 756
 757     if ((argc - optind) != 1) {
 758         error_report("Invalid number of arguments");
 759         error_printf("Try `%s --help' for more information.\n", argv[0]);
 760         exit(EXIT_FAILURE);
 761     }
 762
 763     if (qemu_opts_foreach(&qemu_object_opts,
 764                           user_creatable_add_opts_foreach,
 765                           NULL, NULL)) {
 766         exit(EXIT_FAILURE);
 767     }
 768
 769     if (!trace_init_backends()) {
 770         exit(1);
 771     }
 772     trace_init_file(trace_file);
 773     qemu_set_log(LOG_TRACE);
 774
 775     socket_activation = check_socket_activation();
 776     if (socket_activation == 0) {
 777         setup_address_and_port(&bindto, &port);
 778     } else {
 779         /* Using socket activation - check user didn't use -p etc. */
 780         const char *err_msg = socket_activation_validate_opts(device, sockpath,
 781                                                               bindto, port);
 782         if (err_msg != NULL) {
 783             error_report("%s", err_msg);
 784             exit(EXIT_FAILURE);
 785         }
 786
 787         /* qemu-nbd can only listen on a single socket.  */
 788         if (socket_activation > 1) {
 789             error_report("qemu-nbd does not support socket activation with %s > 1",
 790                          "LISTEN_FDS");
 791             exit(EXIT_FAILURE);
 792         }
 793     }
 794
 795     if (tlscredsid) {
 796         if (sockpath) {
 797             error_report("TLS is only supported with IPv4/IPv6");
 798             exit(EXIT_FAILURE);
 799         }
 800         if (device) {
 801             error_report("TLS is not supported with a host device");
 802             exit(EXIT_FAILURE);
 803         }
 804         if (!export_name) {
 805             /* Set the default NBD protocol export name, since
 806              * we *must* use new style protocol for TLS */
 807             export_name = "";
 808         }
 809         tlscreds = nbd_get_tls_creds(tlscredsid, &local_err);
 810         if (local_err) {
 811             error_report("Failed to get TLS creds %s",
 812                          error_get_pretty(local_err));
 813             exit(EXIT_FAILURE);
 814         }
 815     }
 816
 817     if (disconnect) {
 818         int nbdfd = open(argv[optind], O_RDWR);
 819         if (nbdfd < 0) {
 820             error_report("Cannot open %s: %s", argv[optind],
 821                          strerror(errno));
 822             exit(EXIT_FAILURE);
 823         }
 824         nbd_disconnect(nbdfd);
 825
 826         close(nbdfd);
 827
 828         printf("%s disconnected\n", argv[optind]);
 829
 830         return 0;
 831     }
 832
 833     if ((device && !verbose) || fork_process) {
 834         int stderr_fd[2];
 835         pid_t pid;
 836         int ret;
 837
 838         if (qemu_pipe(stderr_fd) < 0) {
 839             error_report("Error setting up communication pipe: %s",
 840                          strerror(errno));
 841             exit(EXIT_FAILURE);
 842         }
 843
 844         /* Now daemonize, but keep a communication channel open to
 845          * print errors and exit with the proper status code.
 846          */
 847         pid = fork();
 848         if (pid < 0) {
 849             error_report("Failed to fork: %s", strerror(errno));
 850             exit(EXIT_FAILURE);
 851         } else if (pid == 0) {
 852             close(stderr_fd[0]);
 853             ret = qemu_daemon(1, 0);
 854
 855             /* Temporarily redirect stderr to the parent's pipe...  */
 856             old_stderr = dup(STDERR_FILENO);
 857             dup2(stderr_fd[1], STDERR_FILENO);
 858             if (ret < 0) {
 859                 error_report("Failed to daemonize: %s", strerror(errno));
 860                 exit(EXIT_FAILURE);
 861             }
 862
 863             /* ... close the descriptor we inherited and go on.  */
 864             close(stderr_fd[1]);
 865         } else {
 866             bool errors = false;
 867             char *buf;
 868
 869             /* In the parent.  Print error messages from the child until
 870              * it closes the pipe.
 871              */
 872             close(stderr_fd[1]);
 873             buf = g_malloc(1024);
 874             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
 875                 errors = true;
 876                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
 877                 if (ret < 0) {
 878                     exit(EXIT_FAILURE);
 879                 }
 880             }
 881             if (ret < 0) {
 882                 error_report("Cannot read from daemon: %s",
 883                              strerror(errno));
 884                 exit(EXIT_FAILURE);
 885             }
 886
 887             /* Usually the daemon should not print any message.
 888              * Exit with zero status in that case.
 889              */
 890             exit(errors);
 891         }
 892     }
 893
 894     if (device != NULL && sockpath == NULL) {
 895         sockpath = g_malloc(128);
 896         snprintf(sockpath, 128, SOCKET_PATH, basename(device));
 897     }
 898
 899     server = qio_net_listener_new();
 900     if (socket_activation == 0) {
 901         saddr = nbd_build_socket_address(sockpath, bindto, port);
 902         if (qio_net_listener_open_sync(server, saddr, &local_err) < 0) {
 903             object_unref(OBJECT(server));
 904             error_report_err(local_err);
 905             exit(EXIT_FAILURE);
 906         }
 907     } else {
 908         size_t i;
 909         /* See comment in check_socket_activation above. */
 910         for (i = 0; i < socket_activation; i++) {
 911             QIOChannelSocket *sioc;
 912             sioc = qio_channel_socket_new_fd(FIRST_SOCKET_ACTIVATION_FD + i,
 913                                              &local_err);
 914             if (sioc == NULL) {
 915                 object_unref(OBJECT(server));
 916                 error_report("Failed to use socket activation: %s",
 917                              error_get_pretty(local_err));
 918                 exit(EXIT_FAILURE);
 919             }
 920             qio_net_listener_add(server, sioc);
 921             object_unref(OBJECT(sioc));
 922         }
 923     }
 924
 925     if (qemu_init_main_loop(&local_err)) {
 926         error_report_err(local_err);
 927         exit(EXIT_FAILURE);
 928     }
 929     bdrv_init();
 930     atexit(bdrv_close_all);
 931
 932     srcpath = argv[optind];
 933     if (imageOpts) {
 934         QemuOpts *opts;
 935         if (fmt) {
 936             error_report("--image-opts and -f are mutually exclusive");
 937             exit(EXIT_FAILURE);
 938         }
 939         opts = qemu_opts_parse_noisily(&file_opts, srcpath, true);
 940         if (!opts) {
 941             qemu_opts_reset(&file_opts);
 942             exit(EXIT_FAILURE);
 943         }
 944         options = qemu_opts_to_qdict(opts, NULL);
 945         qemu_opts_reset(&file_opts);
 946         blk = blk_new_open(NULL, NULL, options, flags, &local_err);
 947     } else {
 948         if (fmt) {
 949             options = qdict_new();
 950             qdict_put_str(options, "driver", fmt);
 951         }
 952         blk = blk_new_open(srcpath, NULL, options, flags, &local_err);
 953     }
 954
 955     if (!blk) {
 956         error_reportf_err(local_err, "Failed to blk_new_open '%s': ",
 957                           argv[optind]);
 958         exit(EXIT_FAILURE);
 959     }
 960     bs = blk_bs(blk);
 961
 962     blk_set_enable_write_cache(blk, !writethrough);
 963
 964     if (sn_opts) {
 965         ret = bdrv_snapshot_load_tmp(bs,
 966                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
 967                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
 968                                      &local_err);
 969     } else if (sn_id_or_name) {
 970         ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
 971                                                    &local_err);
 972     }
 973     if (ret < 0) {
 974         error_reportf_err(local_err, "Failed to load snapshot: ");
 975         exit(EXIT_FAILURE);
 976     }
 977
 978     bs->detect_zeroes = detect_zeroes;
 979     fd_size = blk_getlength(blk);
 980     if (fd_size < 0) {
 981         error_report("Failed to determine the image length: %s",
 982                      strerror(-fd_size));
 983         exit(EXIT_FAILURE);
 984     }
 985
 986     if (dev_offset >= fd_size) {
 987         error_report("Offset (%lld) has to be smaller than the image size "
 988                      "(%lld)",
 989                      (long long int)dev_offset, (long long int)fd_size);
 990         exit(EXIT_FAILURE);
 991     }
 992     fd_size -= dev_offset;
 993
 994     if (partition != -1) {
 995         ret = find_partition(blk, partition, &dev_offset, &fd_size);
 996         if (ret < 0) {
 997             error_report("Could not find partition %d: %s", partition,
 998                          strerror(-ret));
 999             exit(EXIT_FAILURE);
1000         }
1001     }
1002
1003     exp = nbd_export_new(bs, dev_offset, fd_size, nbdflags, nbd_export_closed,
1004                          writethrough, NULL, &local_err);
1005     if (!exp) {
1006         error_report_err(local_err);
1007         exit(EXIT_FAILURE);
1008     }
1009     if (export_name) {
1010         nbd_export_set_name(exp, export_name);
1011         nbd_export_set_description(exp, export_description);
1012         newproto = true;
1013     } else if (export_description) {
1014         error_report("Export description requires an export name");
1015         exit(EXIT_FAILURE);
1016     }
1017
1018     if (device) {
1019         int ret;
1020
1021         ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
1022         if (ret != 0) {
1023             error_report("Failed to create client thread: %s", strerror(ret));
1024             exit(EXIT_FAILURE);
1025         }
1026     } else {
1027         /* Shut up GCC warnings.  */
1028         memset(&client_thread, 0, sizeof(client_thread));
1029     }
1030
1031     nbd_update_server_watch();
1032
1033     /* now when the initialization is (almost) complete, chdir("/")
1034      * to free any busy filesystems */
1035     if (chdir("/") < 0) {
1036         error_report("Could not chdir to root directory: %s",
1037                      strerror(errno));
1038         exit(EXIT_FAILURE);
1039     }
1040
1041     if (fork_process) {
1042         dup2(old_stderr, STDERR_FILENO);
1043         close(old_stderr);
1044     }
1045
1046     state = RUNNING;
1047     do {
1048         main_loop_wait(false);
1049         if (state == TERMINATE) {
1050             state = TERMINATING;
1051             nbd_export_close(exp);
1052             nbd_export_put(exp);
1053             exp = NULL;
1054         }
1055     } while (state != TERMINATED);
1056
1057     blk_unref(blk);
1058     if (sockpath) {
1059         unlink(sockpath);
1060     }
1061
1062     qemu_opts_del(sn_opts);
1063
1064     if (device) {
1065         void *ret;
1066         pthread_join(client_thread, &ret);
1067         exit(ret != NULL);
1068     } else {
1069         exit(EXIT_SUCCESS);
1070     }
1071 }