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         goto out;
 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 = -1;
 269     int ret = EXIT_FAILURE;
 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     if (nbd_receive_negotiate(NULL, QIO_CHANNEL(sioc),
 282                               NULL, NULL, NULL, &info, &local_error) < 0) {
 283         if (local_error) {
 284             error_report_err(local_error);
 285         }
 286         goto out;
 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;
 294     }
 295
 296     if (nbd_init(fd, sioc, &info, &local_error) < 0) {
 297         error_report_err(local_error);
 298         goto out;
 299     }
 300
 301     /* update partition table */
 302     pthread_create(&show_parts_thread, NULL, show_parts, device);
 303
 304     if (verbose) {
 305         fprintf(stderr, "NBD device %s is now connected to %s\n",
 306                 device, srcpath);
 307     } else {
 308         /* Close stderr so that the qemu-nbd process exits.  */
 309         dup2(STDOUT_FILENO, STDERR_FILENO);
 310     }
 311
 312     if (nbd_client(fd) < 0) {
 313         goto out;
 314     }
 315
 316     ret = EXIT_SUCCESS;
 317
 318  out:
 319     if (fd >= 0) {
 320         close(fd);
 321     }
 322     object_unref(OBJECT(sioc));
 323     g_free(info.name);
 324     kill(getpid(), SIGTERM);
 325     return (void *) (intptr_t) ret;
 326 }
 327 #endif /* HAVE_NBD_DEVICE */
 328
 329 static int nbd_can_accept(void)
 330 {
 331     return state == RUNNING && (shared == 0 || nb_fds < shared);
 332 }
 333
 334 static void nbd_update_server_watch(void);
 335
 336 static void nbd_client_closed(NBDClient *client, bool negotiated)
 337 {
 338     nb_fds--;
 339     if (negotiated && nb_fds == 0 && !persistent && state == RUNNING) {
 340         state = TERMINATE;
 341     }
 342     nbd_update_server_watch();
 343     nbd_client_put(client);
 344 }
 345
 346 static void nbd_accept(QIONetListener *listener, QIOChannelSocket *cioc,
 347                        gpointer opaque)
 348 {
 349     if (state >= TERMINATE) {
 350         return;
 351     }
 352
 353     nb_fds++;
 354     nbd_update_server_watch();
 355     nbd_client_new(cioc, tlscreds, tlsauthz, nbd_client_closed);
 356 }
 357
 358 static void nbd_update_server_watch(void)
 359 {
 360     if (nbd_can_accept()) {
 361         qio_net_listener_set_client_func(server, nbd_accept, NULL, NULL);
 362     } else {
 363         qio_net_listener_set_client_func(server, NULL, NULL, NULL);
 364     }
 365 }
 366
 367
 368 static SocketAddress *nbd_build_socket_address(const char *sockpath,
 369                                                const char *bindto,
 370                                                const char *port)
 371 {
 372     SocketAddress *saddr;
 373
 374     saddr = g_new0(SocketAddress, 1);
 375     if (sockpath) {
 376         saddr->type = SOCKET_ADDRESS_TYPE_UNIX;
 377         saddr->u.q_unix.path = g_strdup(sockpath);
 378     } else {
 379         InetSocketAddress *inet;
 380         saddr->type = SOCKET_ADDRESS_TYPE_INET;
 381         inet = &saddr->u.inet;
 382         inet->host = g_strdup(bindto);
 383         if (port) {
 384             inet->port = g_strdup(port);
 385         } else  {
 386             inet->port = g_strdup_printf("%d", NBD_DEFAULT_PORT);
 387         }
 388     }
 389
 390     return saddr;
 391 }
 392
 393
 394 static QemuOptsList file_opts = {
 395     .name = "file",
 396     .implied_opt_name = "file",
 397     .head = QTAILQ_HEAD_INITIALIZER(file_opts.head),
 398     .desc = {
 399         /* no elements => accept any params */
 400         { /* end of list */ }
 401     },
 402 };
 403
 404 static QemuOptsList qemu_object_opts = {
 405     .name = "object",
 406     .implied_opt_name = "qom-type",
 407     .head = QTAILQ_HEAD_INITIALIZER(qemu_object_opts.head),
 408     .desc = {
 409         { }
 410     },
 411 };
 412
 413 static bool qemu_nbd_object_print_help(const char *type, QemuOpts *opts)
 414 {
 415     if (user_creatable_print_help(type, opts)) {
 416         exit(0);
 417     }
 418     return true;
 419 }
 420
 421
 422 static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, bool list,
 423                                           Error **errp)
 424 {
 425     Object *obj;
 426     QCryptoTLSCreds *creds;
 427
 428     obj = object_resolve_path_component(
 429         object_get_objects_root(), id);
 430     if (!obj) {
 431         error_setg(errp, "No TLS credentials with id '%s'",
 432                    id);
 433         return NULL;
 434     }
 435     creds = (QCryptoTLSCreds *)
 436         object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
 437     if (!creds) {
 438         error_setg(errp, "Object with id '%s' is not TLS credentials",
 439                    id);
 440         return NULL;
 441     }
 442
 443     if (list) {
 444         if (creds->endpoint != QCRYPTO_TLS_CREDS_ENDPOINT_CLIENT) {
 445             error_setg(errp,
 446                        "Expecting TLS credentials with a client endpoint");
 447             return NULL;
 448         }
 449     } else {
 450         if (creds->endpoint != QCRYPTO_TLS_CREDS_ENDPOINT_SERVER) {
 451             error_setg(errp,
 452                        "Expecting TLS credentials with a server endpoint");
 453             return NULL;
 454         }
 455     }
 456     object_ref(obj);
 457     return creds;
 458 }
 459
 460 static void setup_address_and_port(const char **address, const char **port)
 461 {
 462     if (*address == NULL) {
 463         *address = "0.0.0.0";
 464     }
 465
 466     if (*port == NULL) {
 467         *port = stringify(NBD_DEFAULT_PORT);
 468     }
 469 }
 470
 471 /*
 472  * Check socket parameters compatibility when socket activation is used.
 473  */
 474 static const char *socket_activation_validate_opts(const char *device,
 475                                                    const char *sockpath,
 476                                                    const char *address,
 477                                                    const char *port,
 478                                                    bool list)
 479 {
 480     if (device != NULL) {
 481         return "NBD device can't be set when using socket activation";
 482     }
 483
 484     if (sockpath != NULL) {
 485         return "Unix socket can't be set when using socket activation";
 486     }
 487
 488     if (address != NULL) {
 489         return "The interface can't be set when using socket activation";
 490     }
 491
 492     if (port != NULL) {
 493         return "TCP port number can't be set when using socket activation";
 494     }
 495
 496     if (list) {
 497         return "List mode is incompatible with socket activation";
 498     }
 499
 500     return NULL;
 501 }
 502
 503 static void qemu_nbd_shutdown(void)
 504 {
 505     job_cancel_sync_all();
 506     blk_exp_close_all();
 507     bdrv_close_all();
 508 }
 509
 510 int main(int argc, char **argv)
 511 {
 512     BlockBackend *blk;
 513     BlockDriverState *bs;
 514     uint64_t dev_offset = 0;
 515     bool readonly = false;
 516     bool disconnect = false;
 517     const char *bindto = NULL;
 518     const char *port = NULL;
 519     char *sockpath = NULL;
 520     char *device = NULL;
 521     QemuOpts *sn_opts = NULL;
 522     const char *sn_id_or_name = NULL;
 523     const char *sopt = "hVb:o:p:rsnc:dvk:e:f:tl:x:T:D:AB:L";
 524     struct option lopt[] = {
 525         { "help", no_argument, NULL, 'h' },
 526         { "version", no_argument, NULL, 'V' },
 527         { "bind", required_argument, NULL, 'b' },
 528         { "port", required_argument, NULL, 'p' },
 529         { "socket", required_argument, NULL, 'k' },
 530         { "offset", required_argument, NULL, 'o' },
 531         { "read-only", no_argument, NULL, 'r' },
 532         { "allocation-depth", no_argument, NULL, 'A' },
 533         { "bitmap", required_argument, NULL, 'B' },
 534         { "connect", required_argument, NULL, 'c' },
 535         { "disconnect", no_argument, NULL, 'd' },
 536         { "list", no_argument, NULL, 'L' },
 537         { "snapshot", no_argument, NULL, 's' },
 538         { "load-snapshot", required_argument, NULL, 'l' },
 539         { "nocache", no_argument, NULL, 'n' },
 540         { "cache", required_argument, NULL, QEMU_NBD_OPT_CACHE },
 541         { "aio", required_argument, NULL, QEMU_NBD_OPT_AIO },
 542         { "discard", required_argument, NULL, QEMU_NBD_OPT_DISCARD },
 543         { "detect-zeroes", required_argument, NULL,
 544           QEMU_NBD_OPT_DETECT_ZEROES },
 545         { "shared", required_argument, NULL, 'e' },
 546         { "format", required_argument, NULL, 'f' },
 547         { "persistent", no_argument, NULL, 't' },
 548         { "verbose", no_argument, NULL, 'v' },
 549         { "object", required_argument, NULL, QEMU_NBD_OPT_OBJECT },
 550         { "export-name", required_argument, NULL, 'x' },
 551         { "description", required_argument, NULL, 'D' },
 552         { "tls-creds", required_argument, NULL, QEMU_NBD_OPT_TLSCREDS },
 553         { "tls-authz", required_argument, NULL, QEMU_NBD_OPT_TLSAUTHZ },
 554         { "image-opts", no_argument, NULL, QEMU_NBD_OPT_IMAGE_OPTS },
 555         { "trace", required_argument, NULL, 'T' },
 556         { "fork", no_argument, NULL, QEMU_NBD_OPT_FORK },
 557         { "pid-file", required_argument, NULL, QEMU_NBD_OPT_PID_FILE },
 558         { NULL, 0, NULL, 0 }
 559     };
 560     int ch;
 561     int opt_ind = 0;
 562     int flags = BDRV_O_RDWR;
 563     int ret = 0;
 564     bool seen_cache = false;
 565     bool seen_discard = false;
 566     bool seen_aio = false;
 567     pthread_t client_thread;
 568     const char *fmt = NULL;
 569     Error *local_err = NULL;
 570     BlockdevDetectZeroesOptions detect_zeroes = BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
 571     QDict *options = NULL;
 572     const char *export_name = NULL; /* defaults to "" later for server mode */
 573     const char *export_description = NULL;
 574     strList *bitmaps = NULL;
 575     bool alloc_depth = false;
 576     const char *tlscredsid = NULL;
 577     bool imageOpts = false;
 578     bool writethrough = true;
 579     bool fork_process = false;
 580     bool list = false;
 581     int old_stderr = -1;
 582     unsigned socket_activation;
 583     const char *pid_file_name = NULL;
 584     BlockExportOptions *export_opts;
 585
 586 #ifdef CONFIG_POSIX
 587     os_setup_early_signal_handling();
 588     os_setup_signal_handling();
 589 #endif
 590
 591     socket_init();
 592     error_init(argv[0]);
 593     module_call_init(MODULE_INIT_TRACE);
 594     qcrypto_init(&error_fatal);
 595
 596     module_call_init(MODULE_INIT_QOM);
 597     qemu_add_opts(&qemu_object_opts);
 598     qemu_add_opts(&qemu_trace_opts);
 599     qemu_init_exec_dir(argv[0]);
 600
 601     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 602         switch (ch) {
 603         case 's':
 604             flags |= BDRV_O_SNAPSHOT;
 605             break;
 606         case 'n':
 607             optarg = (char *) "none";
 608             /* fallthrough */
 609         case QEMU_NBD_OPT_CACHE:
 610             if (seen_cache) {
 611                 error_report("-n and --cache can only be specified once");
 612                 exit(EXIT_FAILURE);
 613             }
 614             seen_cache = true;
 615             if (bdrv_parse_cache_mode(optarg, &flags, &writethrough) == -1) {
 616                 error_report("Invalid cache mode `%s'", optarg);
 617                 exit(EXIT_FAILURE);
 618             }
 619             break;
 620         case QEMU_NBD_OPT_AIO:
 621             if (seen_aio) {
 622                 error_report("--aio can only be specified once");
 623                 exit(EXIT_FAILURE);
 624             }
 625             seen_aio = true;
 626             if (bdrv_parse_aio(optarg, &flags) < 0) {
 627                 error_report("Invalid aio mode '%s'", optarg);
 628                 exit(EXIT_FAILURE);
 629             }
 630             break;
 631         case QEMU_NBD_OPT_DISCARD:
 632             if (seen_discard) {
 633                 error_report("--discard can only be specified once");
 634                 exit(EXIT_FAILURE);
 635             }
 636             seen_discard = true;
 637             if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
 638                 error_report("Invalid discard mode `%s'", optarg);
 639                 exit(EXIT_FAILURE);
 640             }
 641             break;
 642         case QEMU_NBD_OPT_DETECT_ZEROES:
 643             detect_zeroes =
 644                 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
 645                                 optarg,
 646                                 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
 647                                 &local_err);
 648             if (local_err) {
 649                 error_reportf_err(local_err,
 650                                   "Failed to parse detect_zeroes mode: ");
 651                 exit(EXIT_FAILURE);
 652             }
 653             if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
 654                 !(flags & BDRV_O_UNMAP)) {
 655                 error_report("setting detect-zeroes to unmap is not allowed "
 656                              "without setting discard operation to unmap");
 657                 exit(EXIT_FAILURE);
 658             }
 659             break;
 660         case 'b':
 661             bindto = optarg;
 662             break;
 663         case 'p':
 664             port = optarg;
 665             break;
 666         case 'o':
 667             if (qemu_strtou64(optarg, NULL, 0, &dev_offset) < 0) {
 668                 error_report("Invalid offset '%s'", optarg);
 669                 exit(EXIT_FAILURE);
 670             }
 671             break;
 672         case 'l':
 673             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
 674                 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
 675                                                   optarg, false);
 676                 if (!sn_opts) {
 677                     error_report("Failed in parsing snapshot param `%s'",
 678                                  optarg);
 679                     exit(EXIT_FAILURE);
 680                 }
 681             } else {
 682                 sn_id_or_name = optarg;
 683             }
 684             /* fall through */
 685         case 'r':
 686             readonly = true;
 687             flags &= ~BDRV_O_RDWR;
 688             break;
 689         case 'A':
 690             alloc_depth = true;
 691             break;
 692         case 'B':
 693             QAPI_LIST_PREPEND(bitmaps, g_strdup(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 < 0) {
 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             trace_opt_parse(optarg);
 766             break;
 767         case QEMU_NBD_OPT_TLSAUTHZ:
 768             tlsauthz = optarg;
 769             break;
 770         case QEMU_NBD_OPT_FORK:
 771             fork_process = true;
 772             break;
 773         case 'L':
 774             list = true;
 775             break;
 776         case QEMU_NBD_OPT_PID_FILE:
 777             pid_file_name = optarg;
 778             break;
 779         }
 780     }
 781
 782     if (list) {
 783         if (argc != optind) {
 784             error_report("List mode is incompatible with a file name");
 785             exit(EXIT_FAILURE);
 786         }
 787         if (export_name || export_description || dev_offset ||
 788             device || disconnect || fmt || sn_id_or_name || bitmaps ||
 789             alloc_depth || seen_aio || seen_discard || seen_cache) {
 790             error_report("List mode is incompatible with per-device settings");
 791             exit(EXIT_FAILURE);
 792         }
 793         if (fork_process) {
 794             error_report("List mode is incompatible with forking");
 795             exit(EXIT_FAILURE);
 796         }
 797     } else if ((argc - optind) != 1) {
 798         error_report("Invalid number of arguments");
 799         error_printf("Try `%s --help' for more information.\n", argv[0]);
 800         exit(EXIT_FAILURE);
 801     } else if (!export_name) {
 802         export_name = "";
 803     }
 804
 805     qemu_opts_foreach(&qemu_object_opts,
 806                       user_creatable_add_opts_foreach,
 807                       qemu_nbd_object_print_help, &error_fatal);
 808
 809     if (!trace_init_backends()) {
 810         exit(1);
 811     }
 812     trace_init_file();
 813     qemu_set_log(LOG_TRACE);
 814
 815     socket_activation = check_socket_activation();
 816     if (socket_activation == 0) {
 817         setup_address_and_port(&bindto, &port);
 818     } else {
 819         /* Using socket activation - check user didn't use -p etc. */
 820         const char *err_msg = socket_activation_validate_opts(device, sockpath,
 821                                                               bindto, port,
 822                                                               list);
 823         if (err_msg != NULL) {
 824             error_report("%s", err_msg);
 825             exit(EXIT_FAILURE);
 826         }
 827
 828         /* qemu-nbd can only listen on a single socket.  */
 829         if (socket_activation > 1) {
 830             error_report("qemu-nbd does not support socket activation with %s > 1",
 831                          "LISTEN_FDS");
 832             exit(EXIT_FAILURE);
 833         }
 834     }
 835
 836     if (tlscredsid) {
 837         if (sockpath) {
 838             error_report("TLS is only supported with IPv4/IPv6");
 839             exit(EXIT_FAILURE);
 840         }
 841         if (device) {
 842             error_report("TLS is not supported with a host device");
 843             exit(EXIT_FAILURE);
 844         }
 845         if (tlsauthz && list) {
 846             error_report("TLS authorization is incompatible with export list");
 847             exit(EXIT_FAILURE);
 848         }
 849         tlscreds = nbd_get_tls_creds(tlscredsid, list, &local_err);
 850         if (local_err) {
 851             error_reportf_err(local_err, "Failed to get TLS creds: ");
 852             exit(EXIT_FAILURE);
 853         }
 854     } else {
 855         if (tlsauthz) {
 856             error_report("--tls-authz is not permitted without --tls-creds");
 857             exit(EXIT_FAILURE);
 858         }
 859     }
 860
 861     if (list) {
 862         saddr = nbd_build_socket_address(sockpath, bindto, port);
 863         return qemu_nbd_client_list(saddr, tlscreds, bindto);
 864     }
 865
 866 #if !HAVE_NBD_DEVICE
 867     if (disconnect || device) {
 868         error_report("Kernel /dev/nbdN support not available");
 869         exit(EXIT_FAILURE);
 870     }
 871 #else /* HAVE_NBD_DEVICE */
 872     if (disconnect) {
 873         int nbdfd = open(argv[optind], O_RDWR);
 874         if (nbdfd < 0) {
 875             error_report("Cannot open %s: %s", argv[optind],
 876                          strerror(errno));
 877             exit(EXIT_FAILURE);
 878         }
 879         nbd_disconnect(nbdfd);
 880
 881         close(nbdfd);
 882
 883         printf("%s disconnected\n", argv[optind]);
 884
 885         return 0;
 886     }
 887 #endif
 888
 889     if ((device && !verbose) || fork_process) {
 890 #ifndef WIN32
 891         int stderr_fd[2];
 892         pid_t pid;
 893         int ret;
 894
 895         if (qemu_pipe(stderr_fd) < 0) {
 896             error_report("Error setting up communication pipe: %s",
 897                          strerror(errno));
 898             exit(EXIT_FAILURE);
 899         }
 900
 901         /* Now daemonize, but keep a communication channel open to
 902          * print errors and exit with the proper status code.
 903          */
 904         pid = fork();
 905         if (pid < 0) {
 906             error_report("Failed to fork: %s", strerror(errno));
 907             exit(EXIT_FAILURE);
 908         } else if (pid == 0) {
 909             close(stderr_fd[0]);
 910
 911             /* Remember parent's stderr if we will be restoring it. */
 912             if (fork_process) {
 913                 old_stderr = dup(STDERR_FILENO);
 914             }
 915
 916             ret = qemu_daemon(1, 0);
 917
 918             /* Temporarily redirect stderr to the parent's pipe...  */
 919             dup2(stderr_fd[1], STDERR_FILENO);
 920             if (ret < 0) {
 921                 error_report("Failed to daemonize: %s", strerror(errno));
 922                 exit(EXIT_FAILURE);
 923             }
 924
 925             /* ... close the descriptor we inherited and go on.  */
 926             close(stderr_fd[1]);
 927         } else {
 928             bool errors = false;
 929             char *buf;
 930
 931             /* In the parent.  Print error messages from the child until
 932              * it closes the pipe.
 933              */
 934             close(stderr_fd[1]);
 935             buf = g_malloc(1024);
 936             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
 937                 errors = true;
 938                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
 939                 if (ret < 0) {
 940                     exit(EXIT_FAILURE);
 941                 }
 942             }
 943             if (ret < 0) {
 944                 error_report("Cannot read from daemon: %s",
 945                              strerror(errno));
 946                 exit(EXIT_FAILURE);
 947             }
 948
 949             /* Usually the daemon should not print any message.
 950              * Exit with zero status in that case.
 951              */
 952             exit(errors);
 953         }
 954 #else /* WIN32 */
 955         error_report("Unable to fork into background on Windows hosts");
 956         exit(EXIT_FAILURE);
 957 #endif /* WIN32 */
 958     }
 959
 960     if (device != NULL && sockpath == NULL) {
 961         sockpath = g_malloc(128);
 962         snprintf(sockpath, 128, SOCKET_PATH, basename(device));
 963     }
 964
 965     server = qio_net_listener_new();
 966     if (socket_activation == 0) {
 967         int backlog;
 968
 969         if (persistent || shared == 0) {
 970             backlog = SOMAXCONN;
 971         } else {
 972             backlog = MIN(shared, SOMAXCONN);
 973         }
 974         saddr = nbd_build_socket_address(sockpath, bindto, port);
 975         if (qio_net_listener_open_sync(server, saddr, backlog,
 976                                        &local_err) < 0) {
 977             object_unref(OBJECT(server));
 978             error_report_err(local_err);
 979             exit(EXIT_FAILURE);
 980         }
 981     } else {
 982         size_t i;
 983         /* See comment in check_socket_activation above. */
 984         for (i = 0; i < socket_activation; i++) {
 985             QIOChannelSocket *sioc;
 986             sioc = qio_channel_socket_new_fd(FIRST_SOCKET_ACTIVATION_FD + i,
 987                                              &local_err);
 988             if (sioc == NULL) {
 989                 object_unref(OBJECT(server));
 990                 error_reportf_err(local_err,
 991                                   "Failed to use socket activation: ");
 992                 exit(EXIT_FAILURE);
 993             }
 994             qio_net_listener_add(server, sioc);
 995             object_unref(OBJECT(sioc));
 996         }
 997     }
 998
 999     if (qemu_init_main_loop(&local_err)) {
1000         error_report_err(local_err);
1001         exit(EXIT_FAILURE);
1002     }
1003     bdrv_init();
1004     atexit(qemu_nbd_shutdown);
1005
1006     srcpath = argv[optind];
1007     if (imageOpts) {
1008         QemuOpts *opts;
1009         if (fmt) {
1010             error_report("--image-opts and -f are mutually exclusive");
1011             exit(EXIT_FAILURE);
1012         }
1013         opts = qemu_opts_parse_noisily(&file_opts, srcpath, true);
1014         if (!opts) {
1015             qemu_opts_reset(&file_opts);
1016             exit(EXIT_FAILURE);
1017         }
1018         options = qemu_opts_to_qdict(opts, NULL);
1019         qemu_opts_reset(&file_opts);
1020         blk = blk_new_open(NULL, NULL, options, flags, &local_err);
1021     } else {
1022         if (fmt) {
1023             options = qdict_new();
1024             qdict_put_str(options, "driver", fmt);
1025         }
1026         blk = blk_new_open(srcpath, NULL, options, flags, &local_err);
1027     }
1028
1029     if (!blk) {
1030         error_reportf_err(local_err, "Failed to blk_new_open '%s': ",
1031                           argv[optind]);
1032         exit(EXIT_FAILURE);
1033     }
1034     bs = blk_bs(blk);
1035
1036     if (dev_offset) {
1037         QDict *raw_opts = qdict_new();
1038         qdict_put_str(raw_opts, "driver", "raw");
1039         qdict_put_str(raw_opts, "file", bs->node_name);
1040         qdict_put_int(raw_opts, "offset", dev_offset);
1041         bs = bdrv_open(NULL, NULL, raw_opts, flags, &error_fatal);
1042         blk_remove_bs(blk);
1043         blk_insert_bs(blk, bs, &error_fatal);
1044         bdrv_unref(bs);
1045     }
1046
1047     blk_set_enable_write_cache(blk, !writethrough);
1048
1049     if (sn_opts) {
1050         ret = bdrv_snapshot_load_tmp(bs,
1051                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
1052                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
1053                                      &local_err);
1054     } else if (sn_id_or_name) {
1055         ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
1056                                                    &local_err);
1057     }
1058     if (ret < 0) {
1059         error_reportf_err(local_err, "Failed to load snapshot: ");
1060         exit(EXIT_FAILURE);
1061     }
1062
1063     bs->detect_zeroes = detect_zeroes;
1064
1065     nbd_server_is_qemu_nbd(true);
1066
1067     export_opts = g_new(BlockExportOptions, 1);
1068     *export_opts = (BlockExportOptions) {
1069         .type               = BLOCK_EXPORT_TYPE_NBD,
1070         .id                 = g_strdup("qemu-nbd-export"),
1071         .node_name          = g_strdup(bdrv_get_node_name(bs)),
1072         .has_writethrough   = true,
1073         .writethrough       = writethrough,
1074         .has_writable       = true,
1075         .writable           = !readonly,
1076         .u.nbd = {
1077             .has_name             = true,
1078             .name                 = g_strdup(export_name),
1079             .has_description      = !!export_description,
1080             .description          = g_strdup(export_description),
1081             .has_bitmaps          = !!bitmaps,
1082             .bitmaps              = bitmaps,
1083             .has_allocation_depth = alloc_depth,
1084             .allocation_depth     = alloc_depth,
1085         },
1086     };
1087     blk_exp_add(export_opts, &error_fatal);
1088     qapi_free_BlockExportOptions(export_opts);
1089
1090     if (device) {
1091 #if HAVE_NBD_DEVICE
1092         int ret;
1093
1094         ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
1095         if (ret != 0) {
1096             error_report("Failed to create client thread: %s", strerror(ret));
1097             exit(EXIT_FAILURE);
1098         }
1099 #endif
1100     } else {
1101         /* Shut up GCC warnings.  */
1102         memset(&client_thread, 0, sizeof(client_thread));
1103     }
1104
1105     nbd_update_server_watch();
1106
1107     if (pid_file_name) {
1108         qemu_write_pidfile(pid_file_name, &error_fatal);
1109     }
1110
1111     /* now when the initialization is (almost) complete, chdir("/")
1112      * to free any busy filesystems */
1113     if (chdir("/") < 0) {
1114         error_report("Could not chdir to root directory: %s",
1115                      strerror(errno));
1116         exit(EXIT_FAILURE);
1117     }
1118
1119     if (fork_process) {
1120         dup2(old_stderr, STDERR_FILENO);
1121         close(old_stderr);
1122     }
1123
1124     state = RUNNING;
1125     do {
1126         main_loop_wait(false);
1127         if (state == TERMINATE) {
1128             blk_exp_close_all();
1129             state = TERMINATED;
1130         }
1131     } while (state != TERMINATED);
1132
1133     blk_unref(blk);
1134     if (sockpath) {
1135         unlink(sockpath);
1136     }
1137
1138     qemu_opts_del(sn_opts);
1139
1140     if (device) {
1141         void *ret;
1142         pthread_join(client_thread, &ret);
1143         exit(ret != NULL);
1144     } else {
1145         exit(EXIT_SUCCESS);
1146     }
1147 }