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 && 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     bdrv_close_all();
 507 }
 508
 509 int main(int argc, char **argv)
 510 {
 511     BlockBackend *blk;
 512     BlockDriverState *bs;
 513     uint64_t dev_offset = 0;
 514     bool readonly = false;
 515     bool disconnect = false;
 516     const char *bindto = NULL;
 517     const char *port = NULL;
 518     char *sockpath = NULL;
 519     char *device = NULL;
 520     QemuOpts *sn_opts = NULL;
 521     const char *sn_id_or_name = NULL;
 522     const char *sopt = "hVb:o:p:rsnc:dvk:e:f:tl:x:T:D:AB:L";
 523     struct option lopt[] = {
 524         { "help", no_argument, NULL, 'h' },
 525         { "version", no_argument, NULL, 'V' },
 526         { "bind", required_argument, NULL, 'b' },
 527         { "port", required_argument, NULL, 'p' },
 528         { "socket", required_argument, NULL, 'k' },
 529         { "offset", required_argument, NULL, 'o' },
 530         { "read-only", no_argument, NULL, 'r' },
 531         { "allocation-depth", no_argument, NULL, 'A' },
 532         { "bitmap", required_argument, NULL, 'B' },
 533         { "connect", required_argument, NULL, 'c' },
 534         { "disconnect", no_argument, NULL, 'd' },
 535         { "list", no_argument, NULL, 'L' },
 536         { "snapshot", no_argument, NULL, 's' },
 537         { "load-snapshot", required_argument, NULL, 'l' },
 538         { "nocache", no_argument, NULL, 'n' },
 539         { "cache", required_argument, NULL, QEMU_NBD_OPT_CACHE },
 540         { "aio", required_argument, NULL, QEMU_NBD_OPT_AIO },
 541         { "discard", required_argument, NULL, QEMU_NBD_OPT_DISCARD },
 542         { "detect-zeroes", required_argument, NULL,
 543           QEMU_NBD_OPT_DETECT_ZEROES },
 544         { "shared", required_argument, NULL, 'e' },
 545         { "format", required_argument, NULL, 'f' },
 546         { "persistent", no_argument, NULL, 't' },
 547         { "verbose", no_argument, NULL, 'v' },
 548         { "object", required_argument, NULL, QEMU_NBD_OPT_OBJECT },
 549         { "export-name", required_argument, NULL, 'x' },
 550         { "description", required_argument, NULL, 'D' },
 551         { "tls-creds", required_argument, NULL, QEMU_NBD_OPT_TLSCREDS },
 552         { "tls-authz", required_argument, NULL, QEMU_NBD_OPT_TLSAUTHZ },
 553         { "image-opts", no_argument, NULL, QEMU_NBD_OPT_IMAGE_OPTS },
 554         { "trace", required_argument, NULL, 'T' },
 555         { "fork", no_argument, NULL, QEMU_NBD_OPT_FORK },
 556         { "pid-file", required_argument, NULL, QEMU_NBD_OPT_PID_FILE },
 557         { NULL, 0, NULL, 0 }
 558     };
 559     int ch;
 560     int opt_ind = 0;
 561     int flags = BDRV_O_RDWR;
 562     int ret = 0;
 563     bool seen_cache = false;
 564     bool seen_discard = false;
 565     bool seen_aio = false;
 566     pthread_t client_thread;
 567     const char *fmt = NULL;
 568     Error *local_err = NULL;
 569     BlockdevDetectZeroesOptions detect_zeroes = BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
 570     QDict *options = NULL;
 571     const char *export_name = NULL; /* defaults to "" later for server mode */
 572     const char *export_description = NULL;
 573     strList *bitmaps = NULL;
 574     bool alloc_depth = false;
 575     const char *tlscredsid = NULL;
 576     bool imageOpts = false;
 577     bool writethrough = true;
 578     bool fork_process = false;
 579     bool list = false;
 580     int old_stderr = -1;
 581     unsigned socket_activation;
 582     const char *pid_file_name = NULL;
 583     BlockExportOptions *export_opts;
 584
 585 #ifdef CONFIG_POSIX
 586     os_setup_early_signal_handling();
 587     os_setup_signal_handling();
 588 #endif
 589
 590     socket_init();
 591     error_init(argv[0]);
 592     module_call_init(MODULE_INIT_TRACE);
 593     qcrypto_init(&error_fatal);
 594
 595     module_call_init(MODULE_INIT_QOM);
 596     qemu_add_opts(&qemu_object_opts);
 597     qemu_add_opts(&qemu_trace_opts);
 598     qemu_init_exec_dir(argv[0]);
 599
 600     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 601         switch (ch) {
 602         case 's':
 603             flags |= BDRV_O_SNAPSHOT;
 604             break;
 605         case 'n':
 606             optarg = (char *) "none";
 607             /* fallthrough */
 608         case QEMU_NBD_OPT_CACHE:
 609             if (seen_cache) {
 610                 error_report("-n and --cache can only be specified once");
 611                 exit(EXIT_FAILURE);
 612             }
 613             seen_cache = true;
 614             if (bdrv_parse_cache_mode(optarg, &flags, &writethrough) == -1) {
 615                 error_report("Invalid cache mode `%s'", optarg);
 616                 exit(EXIT_FAILURE);
 617             }
 618             break;
 619         case QEMU_NBD_OPT_AIO:
 620             if (seen_aio) {
 621                 error_report("--aio can only be specified once");
 622                 exit(EXIT_FAILURE);
 623             }
 624             seen_aio = true;
 625             if (bdrv_parse_aio(optarg, &flags) < 0) {
 626                 error_report("Invalid aio mode '%s'", optarg);
 627                 exit(EXIT_FAILURE);
 628             }
 629             break;
 630         case QEMU_NBD_OPT_DISCARD:
 631             if (seen_discard) {
 632                 error_report("--discard can only be specified once");
 633                 exit(EXIT_FAILURE);
 634             }
 635             seen_discard = true;
 636             if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
 637                 error_report("Invalid discard mode `%s'", optarg);
 638                 exit(EXIT_FAILURE);
 639             }
 640             break;
 641         case QEMU_NBD_OPT_DETECT_ZEROES:
 642             detect_zeroes =
 643                 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
 644                                 optarg,
 645                                 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
 646                                 &local_err);
 647             if (local_err) {
 648                 error_reportf_err(local_err,
 649                                   "Failed to parse detect_zeroes mode: ");
 650                 exit(EXIT_FAILURE);
 651             }
 652             if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
 653                 !(flags & BDRV_O_UNMAP)) {
 654                 error_report("setting detect-zeroes to unmap is not allowed "
 655                              "without setting discard operation to unmap");
 656                 exit(EXIT_FAILURE);
 657             }
 658             break;
 659         case 'b':
 660             bindto = optarg;
 661             break;
 662         case 'p':
 663             port = optarg;
 664             break;
 665         case 'o':
 666             if (qemu_strtou64(optarg, NULL, 0, &dev_offset) < 0) {
 667                 error_report("Invalid offset '%s'", optarg);
 668                 exit(EXIT_FAILURE);
 669             }
 670             break;
 671         case 'l':
 672             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
 673                 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
 674                                                   optarg, false);
 675                 if (!sn_opts) {
 676                     error_report("Failed in parsing snapshot param `%s'",
 677                                  optarg);
 678                     exit(EXIT_FAILURE);
 679                 }
 680             } else {
 681                 sn_id_or_name = optarg;
 682             }
 683             /* fall through */
 684         case 'r':
 685             readonly = true;
 686             flags &= ~BDRV_O_RDWR;
 687             break;
 688         case 'A':
 689             alloc_depth = true;
 690             break;
 691         case 'B':
 692             QAPI_LIST_PREPEND(bitmaps, g_strdup(optarg));
 693             break;
 694         case 'k':
 695             sockpath = optarg;
 696             if (sockpath[0] != '/') {
 697                 error_report("socket path must be absolute");
 698                 exit(EXIT_FAILURE);
 699             }
 700             break;
 701         case 'd':
 702             disconnect = true;
 703             break;
 704         case 'c':
 705             device = optarg;
 706             break;
 707         case 'e':
 708             if (qemu_strtoi(optarg, NULL, 0, &shared) < 0 ||
 709                 shared < 1) {
 710                 error_report("Invalid shared device number '%s'", optarg);
 711                 exit(EXIT_FAILURE);
 712             }
 713             break;
 714         case 'f':
 715             fmt = optarg;
 716             break;
 717         case 't':
 718             persistent = 1;
 719             break;
 720         case 'x':
 721             export_name = optarg;
 722             if (strlen(export_name) > NBD_MAX_STRING_SIZE) {
 723                 error_report("export name '%s' too long", export_name);
 724                 exit(EXIT_FAILURE);
 725             }
 726             break;
 727         case 'D':
 728             export_description = optarg;
 729             if (strlen(export_description) > NBD_MAX_STRING_SIZE) {
 730                 error_report("export description '%s' too long",
 731                              export_description);
 732                 exit(EXIT_FAILURE);
 733             }
 734             break;
 735         case 'v':
 736             verbose = 1;
 737             break;
 738         case 'V':
 739             version(argv[0]);
 740             exit(0);
 741             break;
 742         case 'h':
 743             usage(argv[0]);
 744             exit(0);
 745             break;
 746         case '?':
 747             error_report("Try `%s --help' for more information.", argv[0]);
 748             exit(EXIT_FAILURE);
 749         case QEMU_NBD_OPT_OBJECT: {
 750             QemuOpts *opts;
 751             opts = qemu_opts_parse_noisily(&qemu_object_opts,
 752                                            optarg, true);
 753             if (!opts) {
 754                 exit(EXIT_FAILURE);
 755             }
 756         }   break;
 757         case QEMU_NBD_OPT_TLSCREDS:
 758             tlscredsid = optarg;
 759             break;
 760         case QEMU_NBD_OPT_IMAGE_OPTS:
 761             imageOpts = true;
 762             break;
 763         case 'T':
 764             trace_opt_parse(optarg);
 765             break;
 766         case QEMU_NBD_OPT_TLSAUTHZ:
 767             tlsauthz = optarg;
 768             break;
 769         case QEMU_NBD_OPT_FORK:
 770             fork_process = true;
 771             break;
 772         case 'L':
 773             list = true;
 774             break;
 775         case QEMU_NBD_OPT_PID_FILE:
 776             pid_file_name = optarg;
 777             break;
 778         }
 779     }
 780
 781     if (list) {
 782         if (argc != optind) {
 783             error_report("List mode is incompatible with a file name");
 784             exit(EXIT_FAILURE);
 785         }
 786         if (export_name || export_description || dev_offset ||
 787             device || disconnect || fmt || sn_id_or_name || bitmaps ||
 788             alloc_depth || seen_aio || seen_discard || seen_cache) {
 789             error_report("List mode is incompatible with per-device settings");
 790             exit(EXIT_FAILURE);
 791         }
 792         if (fork_process) {
 793             error_report("List mode is incompatible with forking");
 794             exit(EXIT_FAILURE);
 795         }
 796     } else if ((argc - optind) != 1) {
 797         error_report("Invalid number of arguments");
 798         error_printf("Try `%s --help' for more information.\n", argv[0]);
 799         exit(EXIT_FAILURE);
 800     } else if (!export_name) {
 801         export_name = "";
 802     }
 803
 804     qemu_opts_foreach(&qemu_object_opts,
 805                       user_creatable_add_opts_foreach,
 806                       qemu_nbd_object_print_help, &error_fatal);
 807
 808     if (!trace_init_backends()) {
 809         exit(1);
 810     }
 811     trace_init_file();
 812     qemu_set_log(LOG_TRACE);
 813
 814     socket_activation = check_socket_activation();
 815     if (socket_activation == 0) {
 816         setup_address_and_port(&bindto, &port);
 817     } else {
 818         /* Using socket activation - check user didn't use -p etc. */
 819         const char *err_msg = socket_activation_validate_opts(device, sockpath,
 820                                                               bindto, port,
 821                                                               list);
 822         if (err_msg != NULL) {
 823             error_report("%s", err_msg);
 824             exit(EXIT_FAILURE);
 825         }
 826
 827         /* qemu-nbd can only listen on a single socket.  */
 828         if (socket_activation > 1) {
 829             error_report("qemu-nbd does not support socket activation with %s > 1",
 830                          "LISTEN_FDS");
 831             exit(EXIT_FAILURE);
 832         }
 833     }
 834
 835     if (tlscredsid) {
 836         if (sockpath) {
 837             error_report("TLS is only supported with IPv4/IPv6");
 838             exit(EXIT_FAILURE);
 839         }
 840         if (device) {
 841             error_report("TLS is not supported with a host device");
 842             exit(EXIT_FAILURE);
 843         }
 844         if (tlsauthz && list) {
 845             error_report("TLS authorization is incompatible with export list");
 846             exit(EXIT_FAILURE);
 847         }
 848         tlscreds = nbd_get_tls_creds(tlscredsid, list, &local_err);
 849         if (local_err) {
 850             error_reportf_err(local_err, "Failed to get TLS creds: ");
 851             exit(EXIT_FAILURE);
 852         }
 853     } else {
 854         if (tlsauthz) {
 855             error_report("--tls-authz is not permitted without --tls-creds");
 856             exit(EXIT_FAILURE);
 857         }
 858     }
 859
 860     if (list) {
 861         saddr = nbd_build_socket_address(sockpath, bindto, port);
 862         return qemu_nbd_client_list(saddr, tlscreds, bindto);
 863     }
 864
 865 #if !HAVE_NBD_DEVICE
 866     if (disconnect || device) {
 867         error_report("Kernel /dev/nbdN support not available");
 868         exit(EXIT_FAILURE);
 869     }
 870 #else /* HAVE_NBD_DEVICE */
 871     if (disconnect) {
 872         int nbdfd = open(argv[optind], O_RDWR);
 873         if (nbdfd < 0) {
 874             error_report("Cannot open %s: %s", argv[optind],
 875                          strerror(errno));
 876             exit(EXIT_FAILURE);
 877         }
 878         nbd_disconnect(nbdfd);
 879
 880         close(nbdfd);
 881
 882         printf("%s disconnected\n", argv[optind]);
 883
 884         return 0;
 885     }
 886 #endif
 887
 888     if ((device && !verbose) || fork_process) {
 889 #ifndef WIN32
 890         int stderr_fd[2];
 891         pid_t pid;
 892         int ret;
 893
 894         if (qemu_pipe(stderr_fd) < 0) {
 895             error_report("Error setting up communication pipe: %s",
 896                          strerror(errno));
 897             exit(EXIT_FAILURE);
 898         }
 899
 900         /* Now daemonize, but keep a communication channel open to
 901          * print errors and exit with the proper status code.
 902          */
 903         pid = fork();
 904         if (pid < 0) {
 905             error_report("Failed to fork: %s", strerror(errno));
 906             exit(EXIT_FAILURE);
 907         } else if (pid == 0) {
 908             close(stderr_fd[0]);
 909
 910             /* Remember parent's stderr if we will be restoring it. */
 911             if (fork_process) {
 912                 old_stderr = dup(STDERR_FILENO);
 913             }
 914
 915             ret = qemu_daemon(1, 0);
 916
 917             /* Temporarily redirect stderr to the parent's pipe...  */
 918             dup2(stderr_fd[1], STDERR_FILENO);
 919             if (ret < 0) {
 920                 error_report("Failed to daemonize: %s", strerror(errno));
 921                 exit(EXIT_FAILURE);
 922             }
 923
 924             /* ... close the descriptor we inherited and go on.  */
 925             close(stderr_fd[1]);
 926         } else {
 927             bool errors = false;
 928             char *buf;
 929
 930             /* In the parent.  Print error messages from the child until
 931              * it closes the pipe.
 932              */
 933             close(stderr_fd[1]);
 934             buf = g_malloc(1024);
 935             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
 936                 errors = true;
 937                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
 938                 if (ret < 0) {
 939                     exit(EXIT_FAILURE);
 940                 }
 941             }
 942             if (ret < 0) {
 943                 error_report("Cannot read from daemon: %s",
 944                              strerror(errno));
 945                 exit(EXIT_FAILURE);
 946             }
 947
 948             /* Usually the daemon should not print any message.
 949              * Exit with zero status in that case.
 950              */
 951             exit(errors);
 952         }
 953 #else /* WIN32 */
 954         error_report("Unable to fork into background on Windows hosts");
 955         exit(EXIT_FAILURE);
 956 #endif /* WIN32 */
 957     }
 958
 959     if (device != NULL && sockpath == NULL) {
 960         sockpath = g_malloc(128);
 961         snprintf(sockpath, 128, SOCKET_PATH, basename(device));
 962     }
 963
 964     server = qio_net_listener_new();
 965     if (socket_activation == 0) {
 966         saddr = nbd_build_socket_address(sockpath, bindto, port);
 967         if (qio_net_listener_open_sync(server, saddr, 1, &local_err) < 0) {
 968             object_unref(OBJECT(server));
 969             error_report_err(local_err);
 970             exit(EXIT_FAILURE);
 971         }
 972     } else {
 973         size_t i;
 974         /* See comment in check_socket_activation above. */
 975         for (i = 0; i < socket_activation; i++) {
 976             QIOChannelSocket *sioc;
 977             sioc = qio_channel_socket_new_fd(FIRST_SOCKET_ACTIVATION_FD + i,
 978                                              &local_err);
 979             if (sioc == NULL) {
 980                 object_unref(OBJECT(server));
 981                 error_reportf_err(local_err,
 982                                   "Failed to use socket activation: ");
 983                 exit(EXIT_FAILURE);
 984             }
 985             qio_net_listener_add(server, sioc);
 986             object_unref(OBJECT(sioc));
 987         }
 988     }
 989
 990     if (qemu_init_main_loop(&local_err)) {
 991         error_report_err(local_err);
 992         exit(EXIT_FAILURE);
 993     }
 994     bdrv_init();
 995     atexit(qemu_nbd_shutdown);
 996
 997     srcpath = argv[optind];
 998     if (imageOpts) {
 999         QemuOpts *opts;
1000         if (fmt) {
1001             error_report("--image-opts and -f are mutually exclusive");
1002             exit(EXIT_FAILURE);
1003         }
1004         opts = qemu_opts_parse_noisily(&file_opts, srcpath, true);
1005         if (!opts) {
1006             qemu_opts_reset(&file_opts);
1007             exit(EXIT_FAILURE);
1008         }
1009         options = qemu_opts_to_qdict(opts, NULL);
1010         qemu_opts_reset(&file_opts);
1011         blk = blk_new_open(NULL, NULL, options, flags, &local_err);
1012     } else {
1013         if (fmt) {
1014             options = qdict_new();
1015             qdict_put_str(options, "driver", fmt);
1016         }
1017         blk = blk_new_open(srcpath, NULL, options, flags, &local_err);
1018     }
1019
1020     if (!blk) {
1021         error_reportf_err(local_err, "Failed to blk_new_open '%s': ",
1022                           argv[optind]);
1023         exit(EXIT_FAILURE);
1024     }
1025     bs = blk_bs(blk);
1026
1027     if (dev_offset) {
1028         QDict *raw_opts = qdict_new();
1029         qdict_put_str(raw_opts, "driver", "raw");
1030         qdict_put_str(raw_opts, "file", bs->node_name);
1031         qdict_put_int(raw_opts, "offset", dev_offset);
1032         bs = bdrv_open(NULL, NULL, raw_opts, flags, &error_fatal);
1033         blk_remove_bs(blk);
1034         blk_insert_bs(blk, bs, &error_fatal);
1035         bdrv_unref(bs);
1036     }
1037
1038     blk_set_enable_write_cache(blk, !writethrough);
1039
1040     if (sn_opts) {
1041         ret = bdrv_snapshot_load_tmp(bs,
1042                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
1043                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
1044                                      &local_err);
1045     } else if (sn_id_or_name) {
1046         ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
1047                                                    &local_err);
1048     }
1049     if (ret < 0) {
1050         error_reportf_err(local_err, "Failed to load snapshot: ");
1051         exit(EXIT_FAILURE);
1052     }
1053
1054     bs->detect_zeroes = detect_zeroes;
1055
1056     nbd_server_is_qemu_nbd(true);
1057
1058     export_opts = g_new(BlockExportOptions, 1);
1059     *export_opts = (BlockExportOptions) {
1060         .type               = BLOCK_EXPORT_TYPE_NBD,
1061         .id                 = g_strdup("qemu-nbd-export"),
1062         .node_name          = g_strdup(bdrv_get_node_name(bs)),
1063         .has_writethrough   = true,
1064         .writethrough       = writethrough,
1065         .has_writable       = true,
1066         .writable           = !readonly,
1067         .u.nbd = {
1068             .has_name             = true,
1069             .name                 = g_strdup(export_name),
1070             .has_description      = !!export_description,
1071             .description          = g_strdup(export_description),
1072             .has_bitmaps          = !!bitmaps,
1073             .bitmaps              = bitmaps,
1074             .has_allocation_depth = alloc_depth,
1075             .allocation_depth     = alloc_depth,
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 }