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     char *trace_file = NULL;
 585     bool fork_process = false;
 586     bool list = false;
 587     int old_stderr = -1;
 588     unsigned socket_activation;
 589     const char *pid_file_name = NULL;
 590     BlockExportOptions *export_opts;
 591
 592 #ifdef CONFIG_POSIX
 593     os_setup_early_signal_handling();
 594     os_setup_signal_handling();
 595 #endif
 596
 597     socket_init();
 598     error_init(argv[0]);
 599     module_call_init(MODULE_INIT_TRACE);
 600     qcrypto_init(&error_fatal);
 601
 602     module_call_init(MODULE_INIT_QOM);
 603     qemu_add_opts(&qemu_object_opts);
 604     qemu_add_opts(&qemu_trace_opts);
 605     qemu_init_exec_dir(argv[0]);
 606
 607     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 608         switch (ch) {
 609         case 's':
 610             flags |= BDRV_O_SNAPSHOT;
 611             break;
 612         case 'n':
 613             optarg = (char *) "none";
 614             /* fallthrough */
 615         case QEMU_NBD_OPT_CACHE:
 616             if (seen_cache) {
 617                 error_report("-n and --cache can only be specified once");
 618                 exit(EXIT_FAILURE);
 619             }
 620             seen_cache = true;
 621             if (bdrv_parse_cache_mode(optarg, &flags, &writethrough) == -1) {
 622                 error_report("Invalid cache mode `%s'", optarg);
 623                 exit(EXIT_FAILURE);
 624             }
 625             break;
 626         case QEMU_NBD_OPT_AIO:
 627             if (seen_aio) {
 628                 error_report("--aio can only be specified once");
 629                 exit(EXIT_FAILURE);
 630             }
 631             seen_aio = true;
 632             if (bdrv_parse_aio(optarg, &flags) < 0) {
 633                 error_report("Invalid aio mode '%s'", optarg);
 634                 exit(EXIT_FAILURE);
 635             }
 636             break;
 637         case QEMU_NBD_OPT_DISCARD:
 638             if (seen_discard) {
 639                 error_report("--discard can only be specified once");
 640                 exit(EXIT_FAILURE);
 641             }
 642             seen_discard = true;
 643             if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
 644                 error_report("Invalid discard mode `%s'", optarg);
 645                 exit(EXIT_FAILURE);
 646             }
 647             break;
 648         case QEMU_NBD_OPT_DETECT_ZEROES:
 649             detect_zeroes =
 650                 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
 651                                 optarg,
 652                                 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
 653                                 &local_err);
 654             if (local_err) {
 655                 error_reportf_err(local_err,
 656                                   "Failed to parse detect_zeroes mode: ");
 657                 exit(EXIT_FAILURE);
 658             }
 659             if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
 660                 !(flags & BDRV_O_UNMAP)) {
 661                 error_report("setting detect-zeroes to unmap is not allowed "
 662                              "without setting discard operation to unmap");
 663                 exit(EXIT_FAILURE);
 664             }
 665             break;
 666         case 'b':
 667             bindto = optarg;
 668             break;
 669         case 'p':
 670             port = optarg;
 671             break;
 672         case 'o':
 673             if (qemu_strtou64(optarg, NULL, 0, &dev_offset) < 0) {
 674                 error_report("Invalid offset '%s'", optarg);
 675                 exit(EXIT_FAILURE);
 676             }
 677             break;
 678         case 'l':
 679             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
 680                 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
 681                                                   optarg, false);
 682                 if (!sn_opts) {
 683                     error_report("Failed in parsing snapshot param `%s'",
 684                                  optarg);
 685                     exit(EXIT_FAILURE);
 686                 }
 687             } else {
 688                 sn_id_or_name = optarg;
 689             }
 690             /* fall through */
 691         case 'r':
 692             readonly = true;
 693             flags &= ~BDRV_O_RDWR;
 694             break;
 695         case 'A':
 696             alloc_depth = true;
 697             break;
 698         case 'B':
 699             QAPI_LIST_PREPEND(bitmaps, g_strdup(optarg));
 700             break;
 701         case 'k':
 702             sockpath = optarg;
 703             if (sockpath[0] != '/') {
 704                 error_report("socket path must be absolute");
 705                 exit(EXIT_FAILURE);
 706             }
 707             break;
 708         case 'd':
 709             disconnect = true;
 710             break;
 711         case 'c':
 712             device = optarg;
 713             break;
 714         case 'e':
 715             if (qemu_strtoi(optarg, NULL, 0, &shared) < 0 ||
 716                 shared < 1) {
 717                 error_report("Invalid shared device number '%s'", optarg);
 718                 exit(EXIT_FAILURE);
 719             }
 720             break;
 721         case 'f':
 722             fmt = optarg;
 723             break;
 724         case 't':
 725             persistent = 1;
 726             break;
 727         case 'x':
 728             export_name = optarg;
 729             if (strlen(export_name) > NBD_MAX_STRING_SIZE) {
 730                 error_report("export name '%s' too long", export_name);
 731                 exit(EXIT_FAILURE);
 732             }
 733             break;
 734         case 'D':
 735             export_description = optarg;
 736             if (strlen(export_description) > NBD_MAX_STRING_SIZE) {
 737                 error_report("export description '%s' too long",
 738                              export_description);
 739                 exit(EXIT_FAILURE);
 740             }
 741             break;
 742         case 'v':
 743             verbose = 1;
 744             break;
 745         case 'V':
 746             version(argv[0]);
 747             exit(0);
 748             break;
 749         case 'h':
 750             usage(argv[0]);
 751             exit(0);
 752             break;
 753         case '?':
 754             error_report("Try `%s --help' for more information.", argv[0]);
 755             exit(EXIT_FAILURE);
 756         case QEMU_NBD_OPT_OBJECT: {
 757             QemuOpts *opts;
 758             opts = qemu_opts_parse_noisily(&qemu_object_opts,
 759                                            optarg, true);
 760             if (!opts) {
 761                 exit(EXIT_FAILURE);
 762             }
 763         }   break;
 764         case QEMU_NBD_OPT_TLSCREDS:
 765             tlscredsid = optarg;
 766             break;
 767         case QEMU_NBD_OPT_IMAGE_OPTS:
 768             imageOpts = true;
 769             break;
 770         case 'T':
 771             g_free(trace_file);
 772             trace_file = trace_opt_parse(optarg);
 773             break;
 774         case QEMU_NBD_OPT_TLSAUTHZ:
 775             tlsauthz = optarg;
 776             break;
 777         case QEMU_NBD_OPT_FORK:
 778             fork_process = true;
 779             break;
 780         case 'L':
 781             list = true;
 782             break;
 783         case QEMU_NBD_OPT_PID_FILE:
 784             pid_file_name = optarg;
 785             break;
 786         }
 787     }
 788
 789     if (list) {
 790         if (argc != optind) {
 791             error_report("List mode is incompatible with a file name");
 792             exit(EXIT_FAILURE);
 793         }
 794         if (export_name || export_description || dev_offset ||
 795             device || disconnect || fmt || sn_id_or_name || bitmaps ||
 796             alloc_depth || seen_aio || seen_discard || seen_cache) {
 797             error_report("List mode is incompatible with per-device settings");
 798             exit(EXIT_FAILURE);
 799         }
 800         if (fork_process) {
 801             error_report("List mode is incompatible with forking");
 802             exit(EXIT_FAILURE);
 803         }
 804     } else if ((argc - optind) != 1) {
 805         error_report("Invalid number of arguments");
 806         error_printf("Try `%s --help' for more information.\n", argv[0]);
 807         exit(EXIT_FAILURE);
 808     } else if (!export_name) {
 809         export_name = "";
 810     }
 811
 812     qemu_opts_foreach(&qemu_object_opts,
 813                       user_creatable_add_opts_foreach,
 814                       qemu_nbd_object_print_help, &error_fatal);
 815
 816     if (!trace_init_backends()) {
 817         exit(1);
 818     }
 819     trace_init_file(trace_file);
 820     qemu_set_log(LOG_TRACE);
 821
 822     socket_activation = check_socket_activation();
 823     if (socket_activation == 0) {
 824         setup_address_and_port(&bindto, &port);
 825     } else {
 826         /* Using socket activation - check user didn't use -p etc. */
 827         const char *err_msg = socket_activation_validate_opts(device, sockpath,
 828                                                               bindto, port,
 829                                                               list);
 830         if (err_msg != NULL) {
 831             error_report("%s", err_msg);
 832             exit(EXIT_FAILURE);
 833         }
 834
 835         /* qemu-nbd can only listen on a single socket.  */
 836         if (socket_activation > 1) {
 837             error_report("qemu-nbd does not support socket activation with %s > 1",
 838                          "LISTEN_FDS");
 839             exit(EXIT_FAILURE);
 840         }
 841     }
 842
 843     if (tlscredsid) {
 844         if (sockpath) {
 845             error_report("TLS is only supported with IPv4/IPv6");
 846             exit(EXIT_FAILURE);
 847         }
 848         if (device) {
 849             error_report("TLS is not supported with a host device");
 850             exit(EXIT_FAILURE);
 851         }
 852         if (tlsauthz && list) {
 853             error_report("TLS authorization is incompatible with export list");
 854             exit(EXIT_FAILURE);
 855         }
 856         tlscreds = nbd_get_tls_creds(tlscredsid, list, &local_err);
 857         if (local_err) {
 858             error_reportf_err(local_err, "Failed to get TLS creds: ");
 859             exit(EXIT_FAILURE);
 860         }
 861     } else {
 862         if (tlsauthz) {
 863             error_report("--tls-authz is not permitted without --tls-creds");
 864             exit(EXIT_FAILURE);
 865         }
 866     }
 867
 868     if (list) {
 869         saddr = nbd_build_socket_address(sockpath, bindto, port);
 870         return qemu_nbd_client_list(saddr, tlscreds, bindto);
 871     }
 872
 873 #if !HAVE_NBD_DEVICE
 874     if (disconnect || device) {
 875         error_report("Kernel /dev/nbdN support not available");
 876         exit(EXIT_FAILURE);
 877     }
 878 #else /* HAVE_NBD_DEVICE */
 879     if (disconnect) {
 880         int nbdfd = open(argv[optind], O_RDWR);
 881         if (nbdfd < 0) {
 882             error_report("Cannot open %s: %s", argv[optind],
 883                          strerror(errno));
 884             exit(EXIT_FAILURE);
 885         }
 886         nbd_disconnect(nbdfd);
 887
 888         close(nbdfd);
 889
 890         printf("%s disconnected\n", argv[optind]);
 891
 892         return 0;
 893     }
 894 #endif
 895
 896     if ((device && !verbose) || fork_process) {
 897 #ifndef WIN32
 898         int stderr_fd[2];
 899         pid_t pid;
 900         int ret;
 901
 902         if (qemu_pipe(stderr_fd) < 0) {
 903             error_report("Error setting up communication pipe: %s",
 904                          strerror(errno));
 905             exit(EXIT_FAILURE);
 906         }
 907
 908         /* Now daemonize, but keep a communication channel open to
 909          * print errors and exit with the proper status code.
 910          */
 911         pid = fork();
 912         if (pid < 0) {
 913             error_report("Failed to fork: %s", strerror(errno));
 914             exit(EXIT_FAILURE);
 915         } else if (pid == 0) {
 916             close(stderr_fd[0]);
 917
 918             /* Remember parent's stderr if we will be restoring it. */
 919             if (fork_process) {
 920                 old_stderr = dup(STDERR_FILENO);
 921             }
 922
 923             ret = qemu_daemon(1, 0);
 924
 925             /* Temporarily redirect stderr to the parent's pipe...  */
 926             dup2(stderr_fd[1], STDERR_FILENO);
 927             if (ret < 0) {
 928                 error_report("Failed to daemonize: %s", strerror(errno));
 929                 exit(EXIT_FAILURE);
 930             }
 931
 932             /* ... close the descriptor we inherited and go on.  */
 933             close(stderr_fd[1]);
 934         } else {
 935             bool errors = false;
 936             char *buf;
 937
 938             /* In the parent.  Print error messages from the child until
 939              * it closes the pipe.
 940              */
 941             close(stderr_fd[1]);
 942             buf = g_malloc(1024);
 943             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
 944                 errors = true;
 945                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
 946                 if (ret < 0) {
 947                     exit(EXIT_FAILURE);
 948                 }
 949             }
 950             if (ret < 0) {
 951                 error_report("Cannot read from daemon: %s",
 952                              strerror(errno));
 953                 exit(EXIT_FAILURE);
 954             }
 955
 956             /* Usually the daemon should not print any message.
 957              * Exit with zero status in that case.
 958              */
 959             exit(errors);
 960         }
 961 #else /* WIN32 */
 962         error_report("Unable to fork into background on Windows hosts");
 963         exit(EXIT_FAILURE);
 964 #endif /* WIN32 */
 965     }
 966
 967     if (device != NULL && sockpath == NULL) {
 968         sockpath = g_malloc(128);
 969         snprintf(sockpath, 128, SOCKET_PATH, basename(device));
 970     }
 971
 972     server = qio_net_listener_new();
 973     if (socket_activation == 0) {
 974         saddr = nbd_build_socket_address(sockpath, bindto, port);
 975         if (qio_net_listener_open_sync(server, saddr, 1, &local_err) < 0) {
 976             object_unref(OBJECT(server));
 977             error_report_err(local_err);
 978             exit(EXIT_FAILURE);
 979         }
 980     } else {
 981         size_t i;
 982         /* See comment in check_socket_activation above. */
 983         for (i = 0; i < socket_activation; i++) {
 984             QIOChannelSocket *sioc;
 985             sioc = qio_channel_socket_new_fd(FIRST_SOCKET_ACTIVATION_FD + i,
 986                                              &local_err);
 987             if (sioc == NULL) {
 988                 object_unref(OBJECT(server));
 989                 error_reportf_err(local_err,
 990                                   "Failed to use socket activation: ");
 991                 exit(EXIT_FAILURE);
 992             }
 993             qio_net_listener_add(server, sioc);
 994             object_unref(OBJECT(sioc));
 995         }
 996     }
 997
 998     if (qemu_init_main_loop(&local_err)) {
 999         error_report_err(local_err);
1000         exit(EXIT_FAILURE);
1001     }
1002     bdrv_init();
1003     atexit(qemu_nbd_shutdown);
1004
1005     srcpath = argv[optind];
1006     if (imageOpts) {
1007         QemuOpts *opts;
1008         if (fmt) {
1009             error_report("--image-opts and -f are mutually exclusive");
1010             exit(EXIT_FAILURE);
1011         }
1012         opts = qemu_opts_parse_noisily(&file_opts, srcpath, true);
1013         if (!opts) {
1014             qemu_opts_reset(&file_opts);
1015             exit(EXIT_FAILURE);
1016         }
1017         options = qemu_opts_to_qdict(opts, NULL);
1018         qemu_opts_reset(&file_opts);
1019         blk = blk_new_open(NULL, NULL, options, flags, &local_err);
1020     } else {
1021         if (fmt) {
1022             options = qdict_new();
1023             qdict_put_str(options, "driver", fmt);
1024         }
1025         blk = blk_new_open(srcpath, NULL, options, flags, &local_err);
1026     }
1027
1028     if (!blk) {
1029         error_reportf_err(local_err, "Failed to blk_new_open '%s': ",
1030                           argv[optind]);
1031         exit(EXIT_FAILURE);
1032     }
1033     bs = blk_bs(blk);
1034
1035     if (dev_offset) {
1036         QDict *raw_opts = qdict_new();
1037         qdict_put_str(raw_opts, "driver", "raw");
1038         qdict_put_str(raw_opts, "file", bs->node_name);
1039         qdict_put_int(raw_opts, "offset", dev_offset);
1040         bs = bdrv_open(NULL, NULL, raw_opts, flags, &error_fatal);
1041         blk_remove_bs(blk);
1042         blk_insert_bs(blk, bs, &error_fatal);
1043         bdrv_unref(bs);
1044     }
1045
1046     blk_set_enable_write_cache(blk, !writethrough);
1047
1048     if (sn_opts) {
1049         ret = bdrv_snapshot_load_tmp(bs,
1050                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
1051                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
1052                                      &local_err);
1053     } else if (sn_id_or_name) {
1054         ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
1055                                                    &local_err);
1056     }
1057     if (ret < 0) {
1058         error_reportf_err(local_err, "Failed to load snapshot: ");
1059         exit(EXIT_FAILURE);
1060     }
1061
1062     bs->detect_zeroes = detect_zeroes;
1063
1064     nbd_server_is_qemu_nbd(true);
1065
1066     export_opts = g_new(BlockExportOptions, 1);
1067     *export_opts = (BlockExportOptions) {
1068         .type               = BLOCK_EXPORT_TYPE_NBD,
1069         .id                 = g_strdup("qemu-nbd-export"),
1070         .node_name          = g_strdup(bdrv_get_node_name(bs)),
1071         .has_writethrough   = true,
1072         .writethrough       = writethrough,
1073         .has_writable       = true,
1074         .writable           = !readonly,
1075         .u.nbd = {
1076             .has_name             = true,
1077             .name                 = g_strdup(export_name),
1078             .has_description      = !!export_description,
1079             .description          = g_strdup(export_description),
1080             .has_bitmaps          = !!bitmaps,
1081             .bitmaps              = bitmaps,
1082             .has_allocation_depth = alloc_depth,
1083             .allocation_depth     = alloc_depth,
1084         },
1085     };
1086     blk_exp_add(export_opts, &error_fatal);
1087     qapi_free_BlockExportOptions(export_opts);
1088
1089     if (device) {
1090 #if HAVE_NBD_DEVICE
1091         int ret;
1092
1093         ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
1094         if (ret != 0) {
1095             error_report("Failed to create client thread: %s", strerror(ret));
1096             exit(EXIT_FAILURE);
1097         }
1098 #endif
1099     } else {
1100         /* Shut up GCC warnings.  */
1101         memset(&client_thread, 0, sizeof(client_thread));
1102     }
1103
1104     nbd_update_server_watch();
1105
1106     if (pid_file_name) {
1107         qemu_write_pidfile(pid_file_name, &error_fatal);
1108     }
1109
1110     /* now when the initialization is (almost) complete, chdir("/")
1111      * to free any busy filesystems */
1112     if (chdir("/") < 0) {
1113         error_report("Could not chdir to root directory: %s",
1114                      strerror(errno));
1115         exit(EXIT_FAILURE);
1116     }
1117
1118     if (fork_process) {
1119         dup2(old_stderr, STDERR_FILENO);
1120         close(old_stderr);
1121     }
1122
1123     state = RUNNING;
1124     do {
1125         main_loop_wait(false);
1126         if (state == TERMINATE) {
1127             blk_exp_close_all();
1128             state = TERMINATED;
1129         }
1130     } while (state != TERMINATED);
1131
1132     blk_unref(blk);
1133     if (sockpath) {
1134         unlink(sockpath);
1135     }
1136
1137     qemu_opts_del(sn_opts);
1138
1139     if (device) {
1140         void *ret;
1141         pthread_join(client_thread, &ret);
1142         exit(ret != NULL);
1143     } else {
1144         exit(EXIT_SUCCESS);
1145     }
1146 }