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