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 "qemu-common.h"
  25 #include "qapi/error.h"
  26 #include "qemu/cutils.h"
  27 #include "sysemu/block-backend.h"
  28 #include "sysemu/runstate.h" /* for qemu_system_killed() prototype */
  29 #include "block/block_int.h"
  30 #include "block/nbd.h"
  31 #include "qemu/main-loop.h"
  32 #include "qemu/module.h"
  33 #include "qemu/option.h"
  34 #include "qemu/error-report.h"
  35 #include "qemu/config-file.h"
  36 #include "qemu/bswap.h"
  37 #include "qemu/log.h"
  38 #include "qemu/systemd.h"
  39 #include "block/snapshot.h"
  40 #include "qapi/qmp/qdict.h"
  41 #include "qapi/qmp/qstring.h"
  42 #include "qom/object_interfaces.h"
  43 #include "io/channel-socket.h"
  44 #include "io/net-listener.h"
  45 #include "crypto/init.h"
  46 #include "trace/control.h"
  47 #include "qemu-version.h"
  48
  49 #ifdef __linux__
  50 #define HAVE_NBD_DEVICE 1
  51 #else
  52 #define HAVE_NBD_DEVICE 0
  53 #endif
  54
  55 #define SOCKET_PATH                "/var/lock/qemu-nbd-%s"
  56 #define QEMU_NBD_OPT_CACHE         256
  57 #define QEMU_NBD_OPT_AIO           257
  58 #define QEMU_NBD_OPT_DISCARD       258
  59 #define QEMU_NBD_OPT_DETECT_ZEROES 259
  60 #define QEMU_NBD_OPT_OBJECT        260
  61 #define QEMU_NBD_OPT_TLSCREDS      261
  62 #define QEMU_NBD_OPT_IMAGE_OPTS    262
  63 #define QEMU_NBD_OPT_FORK          263
  64 #define QEMU_NBD_OPT_TLSAUTHZ      264
  65 #define QEMU_NBD_OPT_PID_FILE      265
  66
  67 #define MBR_SIZE 512
  68
  69 static int verbose;
  70 static char *srcpath;
  71 static SocketAddress *saddr;
  72 static int persistent = 0;
  73 static enum { RUNNING, TERMINATE, TERMINATED } state;
  74 static int shared = 1;
  75 static int nb_fds;
  76 static QIONetListener *server;
  77 static QCryptoTLSCreds *tlscreds;
  78 static const char *tlsauthz;
  79
  80 static void usage(const char *name)
  81 {
  82     (printf) (
  83 "Usage: %s [OPTIONS] FILE\n"
  84 "  or:  %s -L [OPTIONS]\n"
  85 "QEMU Disk Network Block Device Utility\n"
  86 "\n"
  87 "  -h, --help                display this help and exit\n"
  88 "  -V, --version             output version information and exit\n"
  89 "\n"
  90 "Connection properties:\n"
  91 "  -p, --port=PORT           port to listen on (default `%d')\n"
  92 "  -b, --bind=IFACE          interface to bind to (default `0.0.0.0')\n"
  93 "  -k, --socket=PATH         path to the unix socket\n"
  94 "                            (default '"SOCKET_PATH"')\n"
  95 "  -e, --shared=NUM          device can be shared by NUM clients (default '1')\n"
  96 "  -t, --persistent          don't exit on the last connection\n"
  97 "  -v, --verbose             display extra debugging information\n"
  98 "  -x, --export-name=NAME    expose export by name (default is empty string)\n"
  99 "  -D, --description=TEXT    export a human-readable description\n"
 100 "\n"
 101 "Exposing part of the image:\n"
 102 "  -o, --offset=OFFSET       offset into the image\n"
 103 "  -A, --allocation-depth    expose the allocation depth\n"
 104 "  -B, --bitmap=NAME         expose a persistent dirty bitmap\n"
 105 "\n"
 106 "General purpose options:\n"
 107 "  -L, --list                list exports available from another NBD server\n"
 108 "  --object type,id=ID,...   define an object such as 'secret' for providing\n"
 109 "                            passwords and/or encryption keys\n"
 110 "  --tls-creds=ID            use id of an earlier --object to provide TLS\n"
 111 "  --tls-authz=ID            use id of an earlier --object to provide\n"
 112 "                            authorization\n"
 113 "  -T, --trace [[enable=]<pattern>][,events=<file>][,file=<file>]\n"
 114 "                            specify tracing options\n"
 115 "  --fork                    fork off the server process and exit the parent\n"
 116 "                            once the server is running\n"
 117 "  --pid-file=PATH           store the server's process ID in the given file\n"
 118 #if HAVE_NBD_DEVICE
 119 "\n"
 120 "Kernel NBD client support:\n"
 121 "  -c, --connect=DEV         connect FILE to the local NBD device DEV\n"
 122 "  -d, --disconnect          disconnect the specified device\n"
 123 #endif
 124 "\n"
 125 "Block device options:\n"
 126 "  -f, --format=FORMAT       set image format (raw, qcow2, ...)\n"
 127 "  -r, --read-only           export read-only\n"
 128 "  -s, --snapshot            use FILE as an external snapshot, create a temporary\n"
 129 "                            file with backing_file=FILE, redirect the write to\n"
 130 "                            the temporary one\n"
 131 "  -l, --load-snapshot=SNAPSHOT_PARAM\n"
 132 "                            load an internal snapshot inside FILE and export it\n"
 133 "                            as an read-only device, SNAPSHOT_PARAM format is\n"
 134 "                            'snapshot.id=[ID],snapshot.name=[NAME]', or\n"
 135 "                            '[ID_OR_NAME]'\n"
 136 "  -n, --nocache             disable host cache\n"
 137 "      --cache=MODE          set cache mode (none, writeback, ...)\n"
 138 "      --aio=MODE            set AIO mode (native, io_uring or threads)\n"
 139 "      --discard=MODE        set discard mode (ignore, unmap)\n"
 140 "      --detect-zeroes=MODE  set detect-zeroes mode (off, on, unmap)\n"
 141 "      --image-opts          treat FILE as a full set of image options\n"
 142 "\n"
 143 QEMU_HELP_BOTTOM "\n"
 144     , name, name, NBD_DEFAULT_PORT, "DEVICE");
 145 }
 146
 147 static void version(const char *name)
 148 {
 149     printf(
 150 "%s " QEMU_FULL_VERSION "\n"
 151 "Written by Anthony Liguori.\n"
 152 "\n"
 153 QEMU_COPYRIGHT "\n"
 154 "This is free software; see the source for copying conditions.  There is NO\n"
 155 "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
 156     , name);
 157 }
 158
 159 #ifdef CONFIG_POSIX
 160 /*
 161  * The client thread uses SIGTERM to interrupt the server.  A signal
 162  * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
 163  */
 164 void qemu_system_killed(int signum, pid_t pid)
 165 {
 166     qatomic_cmpxchg(&state, RUNNING, TERMINATE);
 167     qemu_notify_event();
 168 }
 169 #endif /* CONFIG_POSIX */
 170
 171 static int qemu_nbd_client_list(SocketAddress *saddr, QCryptoTLSCreds *tls,
 172                                 const char *hostname)
 173 {
 174     int ret = EXIT_FAILURE;
 175     int rc;
 176     Error *err = NULL;
 177     QIOChannelSocket *sioc;
 178     NBDExportInfo *list;
 179     int i, j;
 180
 181     sioc = qio_channel_socket_new();
 182     if (qio_channel_socket_connect_sync(sioc, saddr, &err) < 0) {
 183         error_report_err(err);
 184         return EXIT_FAILURE;
 185     }
 186     rc = nbd_receive_export_list(QIO_CHANNEL(sioc), tls, hostname, &list,
 187                                  &err);
 188     if (rc < 0) {
 189         if (err) {
 190             error_report_err(err);
 191         }
 192         goto out;
 193     }
 194     printf("exports available: %d\n", rc);
 195     for (i = 0; i < rc; i++) {
 196         printf(" export: '%s'\n", list[i].name);
 197         if (list[i].description && *list[i].description) {
 198             printf("  description: %s\n", list[i].description);
 199         }
 200         if (list[i].flags & NBD_FLAG_HAS_FLAGS) {
 201             static const char *const flag_names[] = {
 202                 [NBD_FLAG_READ_ONLY_BIT]            = "readonly",
 203                 [NBD_FLAG_SEND_FLUSH_BIT]           = "flush",
 204                 [NBD_FLAG_SEND_FUA_BIT]             = "fua",
 205                 [NBD_FLAG_ROTATIONAL_BIT]           = "rotational",
 206                 [NBD_FLAG_SEND_TRIM_BIT]            = "trim",
 207                 [NBD_FLAG_SEND_WRITE_ZEROES_BIT]    = "zeroes",
 208                 [NBD_FLAG_SEND_DF_BIT]              = "df",
 209                 [NBD_FLAG_CAN_MULTI_CONN_BIT]       = "multi",
 210                 [NBD_FLAG_SEND_RESIZE_BIT]          = "resize",
 211                 [NBD_FLAG_SEND_CACHE_BIT]           = "cache",
 212                 [NBD_FLAG_SEND_FAST_ZERO_BIT]       = "fast-zero",
 213             };
 214
 215             printf("  size:  %" PRIu64 "\n", list[i].size);
 216             printf("  flags: 0x%x (", list[i].flags);
 217             for (size_t bit = 0; bit < ARRAY_SIZE(flag_names); bit++) {
 218                 if (flag_names[bit] && (list[i].flags & (1 << bit))) {
 219                     printf(" %s", flag_names[bit]);
 220                 }
 221             }
 222             printf(" )\n");
 223         }
 224         if (list[i].min_block) {
 225             printf("  min block: %u\n", list[i].min_block);
 226             printf("  opt block: %u\n", list[i].opt_block);
 227             printf("  max block: %u\n", list[i].max_block);
 228         }
 229         if (list[i].n_contexts) {
 230             printf("  available meta contexts: %d\n", list[i].n_contexts);
 231             for (j = 0; j < list[i].n_contexts; j++) {
 232                 printf("   %s\n", list[i].contexts[j]);
 233             }
 234         }
 235     }
 236     nbd_free_export_list(list, rc);
 237
 238     ret = EXIT_SUCCESS;
 239  out:
 240     object_unref(OBJECT(sioc));
 241     return ret;
 242 }
 243
 244
 245 #if HAVE_NBD_DEVICE
 246 static void *show_parts(void *arg)
 247 {
 248     char *device = arg;
 249     int nbd;
 250
 251     /* linux just needs an open() to trigger
 252      * the partition table update
 253      * but remember to load the module with max_part != 0 :
 254      *     modprobe nbd max_part=63
 255      */
 256     nbd = open(device, O_RDWR);
 257     if (nbd >= 0) {
 258         close(nbd);
 259     }
 260     return NULL;
 261 }
 262
 263 static void *nbd_client_thread(void *arg)
 264 {
 265     char *device = arg;
 266     NBDExportInfo info = { .request_sizes = false, .name = g_strdup("") };
 267     QIOChannelSocket *sioc;
 268     int fd;
 269     int ret;
 270     pthread_t show_parts_thread;
 271     Error *local_error = NULL;
 272
 273     sioc = qio_channel_socket_new();
 274     if (qio_channel_socket_connect_sync(sioc,
 275                                         saddr,
 276                                         &local_error) < 0) {
 277         error_report_err(local_error);
 278         goto out;
 279     }
 280
 281     ret = nbd_receive_negotiate(NULL, QIO_CHANNEL(sioc),
 282                                 NULL, NULL, NULL, &info, &local_error);
 283     if (ret < 0) {
 284         if (local_error) {
 285             error_report_err(local_error);
 286         }
 287         goto out_socket;
 288     }
 289
 290     fd = open(device, O_RDWR);
 291     if (fd < 0) {
 292         /* Linux-only, we can use %m in printf.  */
 293         error_report("Failed to open %s: %m", device);
 294         goto out_socket;
 295     }
 296
 297     ret = nbd_init(fd, sioc, &info, &local_error);
 298     if (ret < 0) {
 299         error_report_err(local_error);
 300         goto out_fd;
 301     }
 302
 303     /* update partition table */
 304     pthread_create(&show_parts_thread, NULL, show_parts, device);
 305
 306     if (verbose) {
 307         fprintf(stderr, "NBD device %s is now connected to %s\n",
 308                 device, srcpath);
 309     } else {
 310         /* Close stderr so that the qemu-nbd process exits.  */
 311         dup2(STDOUT_FILENO, STDERR_FILENO);
 312     }
 313
 314     ret = nbd_client(fd);
 315     if (ret) {
 316         goto out_fd;
 317     }
 318     close(fd);
 319     object_unref(OBJECT(sioc));
 320     g_free(info.name);
 321     kill(getpid(), SIGTERM);
 322     return (void *) EXIT_SUCCESS;
 323
 324 out_fd:
 325     close(fd);
 326 out_socket:
 327     object_unref(OBJECT(sioc));
 328 out:
 329     g_free(info.name);
 330     kill(getpid(), SIGTERM);
 331     return (void *) EXIT_FAILURE;
 332 }
 333 #endif /* HAVE_NBD_DEVICE */
 334
 335 static int nbd_can_accept(void)
 336 {
 337     return state == RUNNING && nb_fds < shared;
 338 }
 339
 340 static void nbd_update_server_watch(void);
 341
 342 static void nbd_client_closed(NBDClient *client, bool negotiated)
 343 {
 344     nb_fds--;
 345     if (negotiated && nb_fds == 0 && !persistent && state == RUNNING) {
 346         state = TERMINATE;
 347     }
 348     nbd_update_server_watch();
 349     nbd_client_put(client);
 350 }
 351
 352 static void nbd_accept(QIONetListener *listener, QIOChannelSocket *cioc,
 353                        gpointer opaque)
 354 {
 355     if (state >= TERMINATE) {
 356         return;
 357     }
 358
 359     nb_fds++;
 360     nbd_update_server_watch();
 361     nbd_client_new(cioc, tlscreds, tlsauthz, nbd_client_closed);
 362 }
 363
 364 static void nbd_update_server_watch(void)
 365 {
 366     if (nbd_can_accept()) {
 367         qio_net_listener_set_client_func(server, nbd_accept, NULL, NULL);
 368     } else {
 369         qio_net_listener_set_client_func(server, NULL, NULL, NULL);
 370     }
 371 }
 372
 373
 374 static SocketAddress *nbd_build_socket_address(const char *sockpath,
 375                                                const char *bindto,
 376                                                const char *port)
 377 {
 378     SocketAddress *saddr;
 379
 380     saddr = g_new0(SocketAddress, 1);
 381     if (sockpath) {
 382         saddr->type = SOCKET_ADDRESS_TYPE_UNIX;
 383         saddr->u.q_unix.path = g_strdup(sockpath);
 384     } else {
 385         InetSocketAddress *inet;
 386         saddr->type = SOCKET_ADDRESS_TYPE_INET;
 387         inet = &saddr->u.inet;
 388         inet->host = g_strdup(bindto);
 389         if (port) {
 390             inet->port = g_strdup(port);
 391         } else  {
 392             inet->port = g_strdup_printf("%d", NBD_DEFAULT_PORT);
 393         }
 394     }
 395
 396     return saddr;
 397 }
 398
 399
 400 static QemuOptsList file_opts = {
 401     .name = "file",
 402     .implied_opt_name = "file",
 403     .head = QTAILQ_HEAD_INITIALIZER(file_opts.head),
 404     .desc = {
 405         /* no elements => accept any params */
 406         { /* end of list */ }
 407     },
 408 };
 409
 410 static QemuOptsList qemu_object_opts = {
 411     .name = "object",
 412     .implied_opt_name = "qom-type",
 413     .head = QTAILQ_HEAD_INITIALIZER(qemu_object_opts.head),
 414     .desc = {
 415         { }
 416     },
 417 };
 418
 419 static bool qemu_nbd_object_print_help(const char *type, QemuOpts *opts)
 420 {
 421     if (user_creatable_print_help(type, opts)) {
 422         exit(0);
 423     }
 424     return true;
 425 }
 426
 427
 428 static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, bool list,
 429                                           Error **errp)
 430 {
 431     Object *obj;
 432     QCryptoTLSCreds *creds;
 433
 434     obj = object_resolve_path_component(
 435         object_get_objects_root(), id);
 436     if (!obj) {
 437         error_setg(errp, "No TLS credentials with id '%s'",
 438                    id);
 439         return NULL;
 440     }
 441     creds = (QCryptoTLSCreds *)
 442         object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
 443     if (!creds) {
 444         error_setg(errp, "Object with id '%s' is not TLS credentials",
 445                    id);
 446         return NULL;
 447     }
 448
 449     if (list) {
 450         if (creds->endpoint != QCRYPTO_TLS_CREDS_ENDPOINT_CLIENT) {
 451             error_setg(errp,
 452                        "Expecting TLS credentials with a client endpoint");
 453             return NULL;
 454         }
 455     } else {
 456         if (creds->endpoint != QCRYPTO_TLS_CREDS_ENDPOINT_SERVER) {
 457             error_setg(errp,
 458                        "Expecting TLS credentials with a server endpoint");
 459             return NULL;
 460         }
 461     }
 462     object_ref(obj);
 463     return creds;
 464 }
 465
 466 static void setup_address_and_port(const char **address, const char **port)
 467 {
 468     if (*address == NULL) {
 469         *address = "0.0.0.0";
 470     }
 471
 472     if (*port == NULL) {
 473         *port = stringify(NBD_DEFAULT_PORT);
 474     }
 475 }
 476
 477 /*
 478  * Check socket parameters compatibility when socket activation is used.
 479  */
 480 static const char *socket_activation_validate_opts(const char *device,
 481                                                    const char *sockpath,
 482                                                    const char *address,
 483                                                    const char *port,
 484                                                    bool list)
 485 {
 486     if (device != NULL) {
 487         return "NBD device can't be set when using socket activation";
 488     }
 489
 490     if (sockpath != NULL) {
 491         return "Unix socket can't be set when using socket activation";
 492     }
 493
 494     if (address != NULL) {
 495         return "The interface can't be set when using socket activation";
 496     }
 497
 498     if (port != NULL) {
 499         return "TCP port number can't be set when using socket activation";
 500     }
 501
 502     if (list) {
 503         return "List mode is incompatible with socket activation";
 504     }
 505
 506     return NULL;
 507 }
 508
 509 static void qemu_nbd_shutdown(void)
 510 {
 511     job_cancel_sync_all();
 512     bdrv_close_all();
 513 }
 514
 515 int main(int argc, char **argv)
 516 {
 517     BlockBackend *blk;
 518     BlockDriverState *bs;
 519     uint64_t dev_offset = 0;
 520     bool readonly = false;
 521     bool disconnect = false;
 522     const char *bindto = NULL;
 523     const char *port = NULL;
 524     char *sockpath = NULL;
 525     char *device = NULL;
 526     QemuOpts *sn_opts = NULL;
 527     const char *sn_id_or_name = NULL;
 528     const char *sopt = "hVb:o:p:rsnc:dvk:e:f:tl:x:T:D:AB:L";
 529     struct option lopt[] = {
 530         { "help", no_argument, NULL, 'h' },
 531         { "version", no_argument, NULL, 'V' },
 532         { "bind", required_argument, NULL, 'b' },
 533         { "port", required_argument, NULL, 'p' },
 534         { "socket", required_argument, NULL, 'k' },
 535         { "offset", required_argument, NULL, 'o' },
 536         { "read-only", no_argument, NULL, 'r' },
 537         { "allocation-depth", no_argument, NULL, 'A' },
 538         { "bitmap", required_argument, NULL, 'B' },
 539         { "connect", required_argument, NULL, 'c' },
 540         { "disconnect", no_argument, NULL, 'd' },
 541         { "list", no_argument, NULL, 'L' },
 542         { "snapshot", no_argument, NULL, 's' },
 543         { "load-snapshot", required_argument, NULL, 'l' },
 544         { "nocache", no_argument, NULL, 'n' },
 545         { "cache", required_argument, NULL, QEMU_NBD_OPT_CACHE },
 546         { "aio", required_argument, NULL, QEMU_NBD_OPT_AIO },
 547         { "discard", required_argument, NULL, QEMU_NBD_OPT_DISCARD },
 548         { "detect-zeroes", required_argument, NULL,
 549           QEMU_NBD_OPT_DETECT_ZEROES },
 550         { "shared", required_argument, NULL, 'e' },
 551         { "format", required_argument, NULL, 'f' },
 552         { "persistent", no_argument, NULL, 't' },
 553         { "verbose", no_argument, NULL, 'v' },
 554         { "object", required_argument, NULL, QEMU_NBD_OPT_OBJECT },
 555         { "export-name", required_argument, NULL, 'x' },
 556         { "description", required_argument, NULL, 'D' },
 557         { "tls-creds", required_argument, NULL, QEMU_NBD_OPT_TLSCREDS },
 558         { "tls-authz", required_argument, NULL, QEMU_NBD_OPT_TLSAUTHZ },
 559         { "image-opts", no_argument, NULL, QEMU_NBD_OPT_IMAGE_OPTS },
 560         { "trace", required_argument, NULL, 'T' },
 561         { "fork", no_argument, NULL, QEMU_NBD_OPT_FORK },
 562         { "pid-file", required_argument, NULL, QEMU_NBD_OPT_PID_FILE },
 563         { NULL, 0, NULL, 0 }
 564     };
 565     int ch;
 566     int opt_ind = 0;
 567     int flags = BDRV_O_RDWR;
 568     int ret = 0;
 569     bool seen_cache = false;
 570     bool seen_discard = false;
 571     bool seen_aio = false;
 572     pthread_t client_thread;
 573     const char *fmt = NULL;
 574     Error *local_err = NULL;
 575     BlockdevDetectZeroesOptions detect_zeroes = BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
 576     QDict *options = NULL;
 577     const char *export_name = NULL; /* defaults to "" later for server mode */
 578     const char *export_description = NULL;
 579     strList *bitmaps = NULL;
 580     bool alloc_depth = false;
 581     const char *tlscredsid = NULL;
 582     bool imageOpts = false;
 583     bool writethrough = true;
 584     bool fork_process = false;
 585     bool list = false;
 586     int old_stderr = -1;
 587     unsigned socket_activation;
 588     const char *pid_file_name = NULL;
 589     BlockExportOptions *export_opts;
 590
 591 #ifdef CONFIG_POSIX
 592     os_setup_early_signal_handling();
 593     os_setup_signal_handling();
 594 #endif
 595
 596     socket_init();
 597     error_init(argv[0]);
 598     module_call_init(MODULE_INIT_TRACE);
 599     qcrypto_init(&error_fatal);
 600
 601     module_call_init(MODULE_INIT_QOM);
 602     qemu_add_opts(&qemu_object_opts);
 603     qemu_add_opts(&qemu_trace_opts);
 604     qemu_init_exec_dir(argv[0]);
 605
 606     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 607         switch (ch) {
 608         case 's':
 609             flags |= BDRV_O_SNAPSHOT;
 610             break;
 611         case 'n':
 612             optarg = (char *) "none";
 613             /* fallthrough */
 614         case QEMU_NBD_OPT_CACHE:
 615             if (seen_cache) {
 616                 error_report("-n and --cache can only be specified once");
 617                 exit(EXIT_FAILURE);
 618             }
 619             seen_cache = true;
 620             if (bdrv_parse_cache_mode(optarg, &flags, &writethrough) == -1) {
 621                 error_report("Invalid cache mode `%s'", optarg);
 622                 exit(EXIT_FAILURE);
 623             }
 624             break;
 625         case QEMU_NBD_OPT_AIO:
 626             if (seen_aio) {
 627                 error_report("--aio can only be specified once");
 628                 exit(EXIT_FAILURE);
 629             }
 630             seen_aio = true;
 631             if (bdrv_parse_aio(optarg, &flags) < 0) {
 632                 error_report("Invalid aio mode '%s'", optarg);
 633                 exit(EXIT_FAILURE);
 634             }
 635             break;
 636         case QEMU_NBD_OPT_DISCARD:
 637             if (seen_discard) {
 638                 error_report("--discard can only be specified once");
 639                 exit(EXIT_FAILURE);
 640             }
 641             seen_discard = true;
 642             if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
 643                 error_report("Invalid discard mode `%s'", optarg);
 644                 exit(EXIT_FAILURE);
 645             }
 646             break;
 647         case QEMU_NBD_OPT_DETECT_ZEROES:
 648             detect_zeroes =
 649                 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
 650                                 optarg,
 651                                 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
 652                                 &local_err);
 653             if (local_err) {
 654                 error_reportf_err(local_err,
 655                                   "Failed to parse detect_zeroes mode: ");
 656                 exit(EXIT_FAILURE);
 657             }
 658             if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
 659                 !(flags & BDRV_O_UNMAP)) {
 660                 error_report("setting detect-zeroes to unmap is not allowed "
 661                              "without setting discard operation to unmap");
 662                 exit(EXIT_FAILURE);
 663             }
 664             break;
 665         case 'b':
 666             bindto = optarg;
 667             break;
 668         case 'p':
 669             port = optarg;
 670             break;
 671         case 'o':
 672             if (qemu_strtou64(optarg, NULL, 0, &dev_offset) < 0) {
 673                 error_report("Invalid offset '%s'", optarg);
 674                 exit(EXIT_FAILURE);
 675             }
 676             break;
 677         case 'l':
 678             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
 679                 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
 680                                                   optarg, false);
 681                 if (!sn_opts) {
 682                     error_report("Failed in parsing snapshot param `%s'",
 683                                  optarg);
 684                     exit(EXIT_FAILURE);
 685                 }
 686             } else {
 687                 sn_id_or_name = optarg;
 688             }
 689             /* fall through */
 690         case 'r':
 691             readonly = true;
 692             flags &= ~BDRV_O_RDWR;
 693             break;
 694         case 'A':
 695             alloc_depth = true;
 696             break;
 697         case 'B':
 698             QAPI_LIST_PREPEND(bitmaps, g_strdup(optarg));
 699             break;
 700         case 'k':
 701             sockpath = optarg;
 702             if (sockpath[0] != '/') {
 703                 error_report("socket path must be absolute");
 704                 exit(EXIT_FAILURE);
 705             }
 706             break;
 707         case 'd':
 708             disconnect = true;
 709             break;
 710         case 'c':
 711             device = optarg;
 712             break;
 713         case 'e':
 714             if (qemu_strtoi(optarg, NULL, 0, &shared) < 0 ||
 715                 shared < 1) {
 716                 error_report("Invalid shared device number '%s'", optarg);
 717                 exit(EXIT_FAILURE);
 718             }
 719             break;
 720         case 'f':
 721             fmt = optarg;
 722             break;
 723         case 't':
 724             persistent = 1;
 725             break;
 726         case 'x':
 727             export_name = optarg;
 728             if (strlen(export_name) > NBD_MAX_STRING_SIZE) {
 729                 error_report("export name '%s' too long", export_name);
 730                 exit(EXIT_FAILURE);
 731             }
 732             break;
 733         case 'D':
 734             export_description = optarg;
 735             if (strlen(export_description) > NBD_MAX_STRING_SIZE) {
 736                 error_report("export description '%s' too long",
 737                              export_description);
 738                 exit(EXIT_FAILURE);
 739             }
 740             break;
 741         case 'v':
 742             verbose = 1;
 743             break;
 744         case 'V':
 745             version(argv[0]);
 746             exit(0);
 747             break;
 748         case 'h':
 749             usage(argv[0]);
 750             exit(0);
 751             break;
 752         case '?':
 753             error_report("Try `%s --help' for more information.", argv[0]);
 754             exit(EXIT_FAILURE);
 755         case QEMU_NBD_OPT_OBJECT: {
 756             QemuOpts *opts;
 757             opts = qemu_opts_parse_noisily(&qemu_object_opts,
 758                                            optarg, true);
 759             if (!opts) {
 760                 exit(EXIT_FAILURE);
 761             }
 762         }   break;
 763         case QEMU_NBD_OPT_TLSCREDS:
 764             tlscredsid = optarg;
 765             break;
 766         case QEMU_NBD_OPT_IMAGE_OPTS:
 767             imageOpts = true;
 768             break;
 769         case 'T':
 770             trace_opt_parse(optarg);
 771             break;
 772         case QEMU_NBD_OPT_TLSAUTHZ:
 773             tlsauthz = optarg;
 774             break;
 775         case QEMU_NBD_OPT_FORK:
 776             fork_process = true;
 777             break;
 778         case 'L':
 779             list = true;
 780             break;
 781         case QEMU_NBD_OPT_PID_FILE:
 782             pid_file_name = optarg;
 783             break;
 784         }
 785     }
 786
 787     if (list) {
 788         if (argc != optind) {
 789             error_report("List mode is incompatible with a file name");
 790             exit(EXIT_FAILURE);
 791         }
 792         if (export_name || export_description || dev_offset ||
 793             device || disconnect || fmt || sn_id_or_name || bitmaps ||
 794             alloc_depth || seen_aio || seen_discard || seen_cache) {
 795             error_report("List mode is incompatible with per-device settings");
 796             exit(EXIT_FAILURE);
 797         }
 798         if (fork_process) {
 799             error_report("List mode is incompatible with forking");
 800             exit(EXIT_FAILURE);
 801         }
 802     } else if ((argc - optind) != 1) {
 803         error_report("Invalid number of arguments");
 804         error_printf("Try `%s --help' for more information.\n", argv[0]);
 805         exit(EXIT_FAILURE);
 806     } else if (!export_name) {
 807         export_name = "";
 808     }
 809
 810     qemu_opts_foreach(&qemu_object_opts,
 811                       user_creatable_add_opts_foreach,
 812                       qemu_nbd_object_print_help, &error_fatal);
 813
 814     if (!trace_init_backends()) {
 815         exit(1);
 816     }
 817     trace_init_file();
 818     qemu_set_log(LOG_TRACE);
 819
 820     socket_activation = check_socket_activation();
 821     if (socket_activation == 0) {
 822         setup_address_and_port(&bindto, &port);
 823     } else {
 824         /* Using socket activation - check user didn't use -p etc. */
 825         const char *err_msg = socket_activation_validate_opts(device, sockpath,
 826                                                               bindto, port,
 827                                                               list);
 828         if (err_msg != NULL) {
 829             error_report("%s", err_msg);
 830             exit(EXIT_FAILURE);
 831         }
 832
 833         /* qemu-nbd can only listen on a single socket.  */
 834         if (socket_activation > 1) {
 835             error_report("qemu-nbd does not support socket activation with %s > 1",
 836                          "LISTEN_FDS");
 837             exit(EXIT_FAILURE);
 838         }
 839     }
 840
 841     if (tlscredsid) {
 842         if (sockpath) {
 843             error_report("TLS is only supported with IPv4/IPv6");
 844             exit(EXIT_FAILURE);
 845         }
 846         if (device) {
 847             error_report("TLS is not supported with a host device");
 848             exit(EXIT_FAILURE);
 849         }
 850         if (tlsauthz && list) {
 851             error_report("TLS authorization is incompatible with export list");
 852             exit(EXIT_FAILURE);
 853         }
 854         tlscreds = nbd_get_tls_creds(tlscredsid, list, &local_err);
 855         if (local_err) {
 856             error_reportf_err(local_err, "Failed to get TLS creds: ");
 857             exit(EXIT_FAILURE);
 858         }
 859     } else {
 860         if (tlsauthz) {
 861             error_report("--tls-authz is not permitted without --tls-creds");
 862             exit(EXIT_FAILURE);
 863         }
 864     }
 865
 866     if (list) {
 867         saddr = nbd_build_socket_address(sockpath, bindto, port);
 868         return qemu_nbd_client_list(saddr, tlscreds, bindto);
 869     }
 870
 871 #if !HAVE_NBD_DEVICE
 872     if (disconnect || device) {
 873         error_report("Kernel /dev/nbdN support not available");
 874         exit(EXIT_FAILURE);
 875     }
 876 #else /* HAVE_NBD_DEVICE */
 877     if (disconnect) {
 878         int nbdfd = open(argv[optind], O_RDWR);
 879         if (nbdfd < 0) {
 880             error_report("Cannot open %s: %s", argv[optind],
 881                          strerror(errno));
 882             exit(EXIT_FAILURE);
 883         }
 884         nbd_disconnect(nbdfd);
 885
 886         close(nbdfd);
 887
 888         printf("%s disconnected\n", argv[optind]);
 889
 890         return 0;
 891     }
 892 #endif
 893
 894     if ((device && !verbose) || fork_process) {
 895 #ifndef WIN32
 896         int stderr_fd[2];
 897         pid_t pid;
 898         int ret;
 899
 900         if (qemu_pipe(stderr_fd) < 0) {
 901             error_report("Error setting up communication pipe: %s",
 902                          strerror(errno));
 903             exit(EXIT_FAILURE);
 904         }
 905
 906         /* Now daemonize, but keep a communication channel open to
 907          * print errors and exit with the proper status code.
 908          */
 909         pid = fork();
 910         if (pid < 0) {
 911             error_report("Failed to fork: %s", strerror(errno));
 912             exit(EXIT_FAILURE);
 913         } else if (pid == 0) {
 914             close(stderr_fd[0]);
 915
 916             /* Remember parent's stderr if we will be restoring it. */
 917             if (fork_process) {
 918                 old_stderr = dup(STDERR_FILENO);
 919             }
 920
 921             ret = qemu_daemon(1, 0);
 922
 923             /* Temporarily redirect stderr to the parent's pipe...  */
 924             dup2(stderr_fd[1], STDERR_FILENO);
 925             if (ret < 0) {
 926                 error_report("Failed to daemonize: %s", strerror(errno));
 927                 exit(EXIT_FAILURE);
 928             }
 929
 930             /* ... close the descriptor we inherited and go on.  */
 931             close(stderr_fd[1]);
 932         } else {
 933             bool errors = false;
 934             char *buf;
 935
 936             /* In the parent.  Print error messages from the child until
 937              * it closes the pipe.
 938              */
 939             close(stderr_fd[1]);
 940             buf = g_malloc(1024);
 941             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
 942                 errors = true;
 943                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
 944                 if (ret < 0) {
 945                     exit(EXIT_FAILURE);
 946                 }
 947             }
 948             if (ret < 0) {
 949                 error_report("Cannot read from daemon: %s",
 950                              strerror(errno));
 951                 exit(EXIT_FAILURE);
 952             }
 953
 954             /* Usually the daemon should not print any message.
 955              * Exit with zero status in that case.
 956              */
 957             exit(errors);
 958         }
 959 #else /* WIN32 */
 960         error_report("Unable to fork into background on Windows hosts");
 961         exit(EXIT_FAILURE);
 962 #endif /* WIN32 */
 963     }
 964
 965     if (device != NULL && sockpath == NULL) {
 966         sockpath = g_malloc(128);
 967         snprintf(sockpath, 128, SOCKET_PATH, basename(device));
 968     }
 969
 970     server = qio_net_listener_new();
 971     if (socket_activation == 0) {
 972         saddr = nbd_build_socket_address(sockpath, bindto, port);
 973         if (qio_net_listener_open_sync(server, saddr, 1, &local_err) < 0) {
 974             object_unref(OBJECT(server));
 975             error_report_err(local_err);
 976             exit(EXIT_FAILURE);
 977         }
 978     } else {
 979         size_t i;
 980         /* See comment in check_socket_activation above. */
 981         for (i = 0; i < socket_activation; i++) {
 982             QIOChannelSocket *sioc;
 983             sioc = qio_channel_socket_new_fd(FIRST_SOCKET_ACTIVATION_FD + i,
 984                                              &local_err);
 985             if (sioc == NULL) {
 986                 object_unref(OBJECT(server));
 987                 error_reportf_err(local_err,
 988                                   "Failed to use socket activation: ");
 989                 exit(EXIT_FAILURE);
 990             }
 991             qio_net_listener_add(server, sioc);
 992             object_unref(OBJECT(sioc));
 993         }
 994     }
 995
 996     if (qemu_init_main_loop(&local_err)) {
 997         error_report_err(local_err);
 998         exit(EXIT_FAILURE);
 999     }
1000     bdrv_init();
1001     atexit(qemu_nbd_shutdown);
1002
1003     srcpath = argv[optind];
1004     if (imageOpts) {
1005         QemuOpts *opts;
1006         if (fmt) {
1007             error_report("--image-opts and -f are mutually exclusive");
1008             exit(EXIT_FAILURE);
1009         }
1010         opts = qemu_opts_parse_noisily(&file_opts, srcpath, true);
1011         if (!opts) {
1012             qemu_opts_reset(&file_opts);
1013             exit(EXIT_FAILURE);
1014         }
1015         options = qemu_opts_to_qdict(opts, NULL);
1016         qemu_opts_reset(&file_opts);
1017         blk = blk_new_open(NULL, NULL, options, flags, &local_err);
1018     } else {
1019         if (fmt) {
1020             options = qdict_new();
1021             qdict_put_str(options, "driver", fmt);
1022         }
1023         blk = blk_new_open(srcpath, NULL, options, flags, &local_err);
1024     }
1025
1026     if (!blk) {
1027         error_reportf_err(local_err, "Failed to blk_new_open '%s': ",
1028                           argv[optind]);
1029         exit(EXIT_FAILURE);
1030     }
1031     bs = blk_bs(blk);
1032
1033     if (dev_offset) {
1034         QDict *raw_opts = qdict_new();
1035         qdict_put_str(raw_opts, "driver", "raw");
1036         qdict_put_str(raw_opts, "file", bs->node_name);
1037         qdict_put_int(raw_opts, "offset", dev_offset);
1038         bs = bdrv_open(NULL, NULL, raw_opts, flags, &error_fatal);
1039         blk_remove_bs(blk);
1040         blk_insert_bs(blk, bs, &error_fatal);
1041         bdrv_unref(bs);
1042     }
1043
1044     blk_set_enable_write_cache(blk, !writethrough);
1045
1046     if (sn_opts) {
1047         ret = bdrv_snapshot_load_tmp(bs,
1048                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
1049                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
1050                                      &local_err);
1051     } else if (sn_id_or_name) {
1052         ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
1053                                                    &local_err);
1054     }
1055     if (ret < 0) {
1056         error_reportf_err(local_err, "Failed to load snapshot: ");
1057         exit(EXIT_FAILURE);
1058     }
1059
1060     bs->detect_zeroes = detect_zeroes;
1061
1062     nbd_server_is_qemu_nbd(true);
1063
1064     export_opts = g_new(BlockExportOptions, 1);
1065     *export_opts = (BlockExportOptions) {
1066         .type               = BLOCK_EXPORT_TYPE_NBD,
1067         .id                 = g_strdup("qemu-nbd-export"),
1068         .node_name          = g_strdup(bdrv_get_node_name(bs)),
1069         .has_writethrough   = true,
1070         .writethrough       = writethrough,
1071         .has_writable       = true,
1072         .writable           = !readonly,
1073         .u.nbd = {
1074             .has_name             = true,
1075             .name                 = g_strdup(export_name),
1076             .has_description      = !!export_description,
1077             .description          = g_strdup(export_description),
1078             .has_bitmaps          = !!bitmaps,
1079             .bitmaps              = bitmaps,
1080             .has_allocation_depth = alloc_depth,
1081             .allocation_depth     = alloc_depth,
1082         },
1083     };
1084     blk_exp_add(export_opts, &error_fatal);
1085     qapi_free_BlockExportOptions(export_opts);
1086
1087     if (device) {
1088 #if HAVE_NBD_DEVICE
1089         int ret;
1090
1091         ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
1092         if (ret != 0) {
1093             error_report("Failed to create client thread: %s", strerror(ret));
1094             exit(EXIT_FAILURE);
1095         }
1096 #endif
1097     } else {
1098         /* Shut up GCC warnings.  */
1099         memset(&client_thread, 0, sizeof(client_thread));
1100     }
1101
1102     nbd_update_server_watch();
1103
1104     if (pid_file_name) {
1105         qemu_write_pidfile(pid_file_name, &error_fatal);
1106     }
1107
1108     /* now when the initialization is (almost) complete, chdir("/")
1109      * to free any busy filesystems */
1110     if (chdir("/") < 0) {
1111         error_report("Could not chdir to root directory: %s",
1112                      strerror(errno));
1113         exit(EXIT_FAILURE);
1114     }
1115
1116     if (fork_process) {
1117         dup2(old_stderr, STDERR_FILENO);
1118         close(old_stderr);
1119     }
1120
1121     state = RUNNING;
1122     do {
1123         main_loop_wait(false);
1124         if (state == TERMINATE) {
1125             blk_exp_close_all();
1126             state = TERMINATED;
1127         }
1128     } while (state != TERMINATED);
1129
1130     blk_unref(blk);
1131     if (sockpath) {
1132         unlink(sockpath);
1133     }
1134
1135     qemu_opts_del(sn_opts);
1136
1137     if (device) {
1138         void *ret;
1139         pthread_join(client_thread, &ret);
1140         exit(ret != NULL);
1141     } else {
1142         exit(EXIT_SUCCESS);
1143     }
1144 }