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