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/help-texts.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 CONFIG_SELINUX
  51 #include <selinux/selinux.h>
  52 #endif
  53
  54 #ifdef __linux__
  55 #define HAVE_NBD_DEVICE 1
  56 #else
  57 #define HAVE_NBD_DEVICE 0
  58 #endif
  59
  60 #define SOCKET_PATH                "/var/lock/qemu-nbd-%s"
  61 #define QEMU_NBD_OPT_CACHE         256
  62 #define QEMU_NBD_OPT_AIO           257
  63 #define QEMU_NBD_OPT_DISCARD       258
  64 #define QEMU_NBD_OPT_DETECT_ZEROES 259
  65 #define QEMU_NBD_OPT_OBJECT        260
  66 #define QEMU_NBD_OPT_TLSCREDS      261
  67 #define QEMU_NBD_OPT_IMAGE_OPTS    262
  68 #define QEMU_NBD_OPT_FORK          263
  69 #define QEMU_NBD_OPT_TLSAUTHZ      264
  70 #define QEMU_NBD_OPT_PID_FILE      265
  71 #define QEMU_NBD_OPT_SELINUX_LABEL 266
  72 #define QEMU_NBD_OPT_TLSHOSTNAME   267
  73
  74 #define MBR_SIZE 512
  75
  76 static int verbose;
  77 static char *srcpath;
  78 static SocketAddress *saddr;
  79 static int persistent = 0;
  80 static enum { RUNNING, TERMINATE, TERMINATED } state;
  81 static int shared = 1;
  82 static int nb_fds;
  83 static QIONetListener *server;
  84 static QCryptoTLSCreds *tlscreds;
  85 static const char *tlsauthz;
  86
  87 static void usage(const char *name)
  88 {
  89     (printf) (
  90 "Usage: %s [OPTIONS] FILE\n"
  91 "  or:  %s -L [OPTIONS]\n"
  92 "QEMU Disk Network Block Device Utility\n"
  93 "\n"
  94 "  -h, --help                display this help and exit\n"
  95 "  -V, --version             output version information and exit\n"
  96 "\n"
  97 "Connection properties:\n"
  98 "  -p, --port=PORT           port to listen on (default `%d')\n"
  99 "  -b, --bind=IFACE          interface to bind to (default `0.0.0.0')\n"
 100 "  -k, --socket=PATH         path to the unix socket\n"
 101 "                            (default '"SOCKET_PATH"')\n"
 102 "  -e, --shared=NUM          device can be shared by NUM clients (default '1')\n"
 103 "  -t, --persistent          don't exit on the last connection\n"
 104 "  -v, --verbose             display extra debugging information\n"
 105 "  -x, --export-name=NAME    expose export by name (default is empty string)\n"
 106 "  -D, --description=TEXT    export a human-readable description\n"
 107 "\n"
 108 "Exposing part of the image:\n"
 109 "  -o, --offset=OFFSET       offset into the image\n"
 110 "  -A, --allocation-depth    expose the allocation depth\n"
 111 "  -B, --bitmap=NAME         expose a persistent dirty bitmap\n"
 112 "\n"
 113 "General purpose options:\n"
 114 "  -L, --list                list exports available from another NBD server\n"
 115 "  --object type,id=ID,...   define an object such as 'secret' for providing\n"
 116 "                            passwords and/or encryption keys\n"
 117 "  --tls-creds=ID            use id of an earlier --object to provide TLS\n"
 118 "  --tls-authz=ID            use id of an earlier --object to provide\n"
 119 "                            authorization\n"
 120 "  -T, --trace [[enable=]<pattern>][,events=<file>][,file=<file>]\n"
 121 "                            specify tracing options\n"
 122 "  --fork                    fork off the server process and exit the parent\n"
 123 "                            once the server is running\n"
 124 "  --pid-file=PATH           store the server's process ID in the given file\n"
 125 #ifdef CONFIG_SELINUX
 126 "  --selinux-label=LABEL     set SELinux process label on listening socket\n"
 127 #endif
 128 #if HAVE_NBD_DEVICE
 129 "\n"
 130 "Kernel NBD client support:\n"
 131 "  -c, --connect=DEV         connect FILE to the local NBD device DEV\n"
 132 "  -d, --disconnect          disconnect the specified device\n"
 133 #endif
 134 "\n"
 135 "Block device options:\n"
 136 "  -f, --format=FORMAT       set image format (raw, qcow2, ...)\n"
 137 "  -r, --read-only           export read-only\n"
 138 "  -s, --snapshot            use FILE as an external snapshot, create a temporary\n"
 139 "                            file with backing_file=FILE, redirect the write to\n"
 140 "                            the temporary one\n"
 141 "  -l, --load-snapshot=SNAPSHOT_PARAM\n"
 142 "                            load an internal snapshot inside FILE and export it\n"
 143 "                            as an read-only device, SNAPSHOT_PARAM format is\n"
 144 "                            'snapshot.id=[ID],snapshot.name=[NAME]', or\n"
 145 "                            '[ID_OR_NAME]'\n"
 146 "  -n, --nocache             disable host cache\n"
 147 "      --cache=MODE          set cache mode used to access the disk image, the\n"
 148 "                            valid options are: 'none', 'writeback' (default),\n"
 149 "                            'writethrough', 'directsync' and 'unsafe'\n"
 150 "      --aio=MODE            set AIO mode (native, io_uring or threads)\n"
 151 "      --discard=MODE        set discard mode (ignore, unmap)\n"
 152 "      --detect-zeroes=MODE  set detect-zeroes mode (off, on, unmap)\n"
 153 "      --image-opts          treat FILE as a full set of image options\n"
 154 "\n"
 155 QEMU_HELP_BOTTOM "\n"
 156     , name, name, NBD_DEFAULT_PORT, "DEVICE");
 157 }
 158
 159 static void version(const char *name)
 160 {
 161     printf(
 162 "%s " QEMU_FULL_VERSION "\n"
 163 "Written by Anthony Liguori.\n"
 164 "\n"
 165 QEMU_COPYRIGHT "\n"
 166 "This is free software; see the source for copying conditions.  There is NO\n"
 167 "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
 168     , name);
 169 }
 170
 171 #ifdef CONFIG_POSIX
 172 /*
 173  * The client thread uses SIGTERM to interrupt the server.  A signal
 174  * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
 175  */
 176 void qemu_system_killed(int signum, pid_t pid)
 177 {
 178     qatomic_cmpxchg(&state, RUNNING, TERMINATE);
 179     qemu_notify_event();
 180 }
 181 #endif /* CONFIG_POSIX */
 182
 183 static int qemu_nbd_client_list(SocketAddress *saddr, QCryptoTLSCreds *tls,
 184                                 const char *hostname)
 185 {
 186     int ret = EXIT_FAILURE;
 187     int rc;
 188     Error *err = NULL;
 189     QIOChannelSocket *sioc;
 190     NBDExportInfo *list;
 191     int i, j;
 192
 193     sioc = qio_channel_socket_new();
 194     if (qio_channel_socket_connect_sync(sioc, saddr, &err) < 0) {
 195         error_report_err(err);
 196         goto out;
 197     }
 198     rc = nbd_receive_export_list(QIO_CHANNEL(sioc), tls, hostname, &list,
 199                                  &err);
 200     if (rc < 0) {
 201         if (err) {
 202             error_report_err(err);
 203         }
 204         goto out;
 205     }
 206     printf("exports available: %d\n", rc);
 207     for (i = 0; i < rc; i++) {
 208         printf(" export: '%s'\n", list[i].name);
 209         if (list[i].description && *list[i].description) {
 210             printf("  description: %s\n", list[i].description);
 211         }
 212         if (list[i].flags & NBD_FLAG_HAS_FLAGS) {
 213             static const char *const flag_names[] = {
 214                 [NBD_FLAG_READ_ONLY_BIT]            = "readonly",
 215                 [NBD_FLAG_SEND_FLUSH_BIT]           = "flush",
 216                 [NBD_FLAG_SEND_FUA_BIT]             = "fua",
 217                 [NBD_FLAG_ROTATIONAL_BIT]           = "rotational",
 218                 [NBD_FLAG_SEND_TRIM_BIT]            = "trim",
 219                 [NBD_FLAG_SEND_WRITE_ZEROES_BIT]    = "zeroes",
 220                 [NBD_FLAG_SEND_DF_BIT]              = "df",
 221                 [NBD_FLAG_CAN_MULTI_CONN_BIT]       = "multi",
 222                 [NBD_FLAG_SEND_RESIZE_BIT]          = "resize",
 223                 [NBD_FLAG_SEND_CACHE_BIT]           = "cache",
 224                 [NBD_FLAG_SEND_FAST_ZERO_BIT]       = "fast-zero",
 225             };
 226
 227             printf("  size:  %" PRIu64 "\n", list[i].size);
 228             printf("  flags: 0x%x (", list[i].flags);
 229             for (size_t bit = 0; bit < ARRAY_SIZE(flag_names); bit++) {
 230                 if (flag_names[bit] && (list[i].flags & (1 << bit))) {
 231                     printf(" %s", flag_names[bit]);
 232                 }
 233             }
 234             printf(" )\n");
 235         }
 236         if (list[i].min_block) {
 237             printf("  min block: %u\n", list[i].min_block);
 238             printf("  opt block: %u\n", list[i].opt_block);
 239             printf("  max block: %u\n", list[i].max_block);
 240         }
 241         if (list[i].n_contexts) {
 242             printf("  available meta contexts: %d\n", list[i].n_contexts);
 243             for (j = 0; j < list[i].n_contexts; j++) {
 244                 printf("   %s\n", list[i].contexts[j]);
 245             }
 246         }
 247     }
 248     nbd_free_export_list(list, rc);
 249
 250     ret = EXIT_SUCCESS;
 251  out:
 252     object_unref(OBJECT(sioc));
 253     return ret;
 254 }
 255
 256
 257 #if HAVE_NBD_DEVICE
 258 static void *show_parts(void *arg)
 259 {
 260     char *device = arg;
 261     int nbd;
 262
 263     /* linux just needs an open() to trigger
 264      * the partition table update
 265      * but remember to load the module with max_part != 0 :
 266      *     modprobe nbd max_part=63
 267      */
 268     nbd = open(device, O_RDWR);
 269     if (nbd >= 0) {
 270         close(nbd);
 271     }
 272     return NULL;
 273 }
 274
 275 struct NbdClientOpts {
 276     char *device;
 277 };
 278
 279 static void *nbd_client_thread(void *arg)
 280 {
 281     struct NbdClientOpts *opts = arg;
 282     NBDExportInfo info = { .request_sizes = false, .name = g_strdup("") };
 283     QIOChannelSocket *sioc;
 284     int fd = -1;
 285     int ret = EXIT_FAILURE;
 286     pthread_t show_parts_thread;
 287     Error *local_error = NULL;
 288
 289     sioc = qio_channel_socket_new();
 290     if (qio_channel_socket_connect_sync(sioc,
 291                                         saddr,
 292                                         &local_error) < 0) {
 293         error_report_err(local_error);
 294         goto out;
 295     }
 296
 297     if (nbd_receive_negotiate(NULL, QIO_CHANNEL(sioc),
 298                               NULL, NULL, NULL, &info, &local_error) < 0) {
 299         if (local_error) {
 300             error_report_err(local_error);
 301         }
 302         goto out;
 303     }
 304
 305     fd = open(opts->device, O_RDWR);
 306     if (fd < 0) {
 307         /* Linux-only, we can use %m in printf.  */
 308         error_report("Failed to open %s: %m", opts->device);
 309         goto out;
 310     }
 311
 312     if (nbd_init(fd, sioc, &info, &local_error) < 0) {
 313         error_report_err(local_error);
 314         goto out;
 315     }
 316
 317     /* update partition table */
 318     pthread_create(&show_parts_thread, NULL, show_parts, opts->device);
 319
 320     if (verbose) {
 321         fprintf(stderr, "NBD device %s is now connected to %s\n",
 322                 opts->device, srcpath);
 323     } else {
 324         /* Close stderr so that the qemu-nbd process exits.  */
 325         dup2(STDOUT_FILENO, STDERR_FILENO);
 326     }
 327
 328     if (nbd_client(fd) < 0) {
 329         goto out;
 330     }
 331
 332     ret = EXIT_SUCCESS;
 333
 334  out:
 335     if (fd >= 0) {
 336         close(fd);
 337     }
 338     object_unref(OBJECT(sioc));
 339     g_free(info.name);
 340     kill(getpid(), SIGTERM);
 341     return (void *) (intptr_t) ret;
 342 }
 343 #endif /* HAVE_NBD_DEVICE */
 344
 345 static int nbd_can_accept(void)
 346 {
 347     return state == RUNNING && (shared == 0 || nb_fds < shared);
 348 }
 349
 350 static void nbd_update_server_watch(void);
 351
 352 static void nbd_client_closed(NBDClient *client, bool negotiated)
 353 {
 354     nb_fds--;
 355     if (negotiated && nb_fds == 0 && !persistent && state == RUNNING) {
 356         state = TERMINATE;
 357     }
 358     nbd_update_server_watch();
 359     nbd_client_put(client);
 360 }
 361
 362 static void nbd_accept(QIONetListener *listener, QIOChannelSocket *cioc,
 363                        gpointer opaque)
 364 {
 365     if (state >= TERMINATE) {
 366         return;
 367     }
 368
 369     nb_fds++;
 370     nbd_update_server_watch();
 371     nbd_client_new(cioc, tlscreds, tlsauthz, nbd_client_closed);
 372 }
 373
 374 static void nbd_update_server_watch(void)
 375 {
 376     if (nbd_can_accept()) {
 377         qio_net_listener_set_client_func(server, nbd_accept, NULL, NULL);
 378     } else {
 379         qio_net_listener_set_client_func(server, NULL, NULL, NULL);
 380     }
 381 }
 382
 383
 384 static SocketAddress *nbd_build_socket_address(const char *sockpath,
 385                                                const char *bindto,
 386                                                const char *port)
 387 {
 388     SocketAddress *saddr;
 389
 390     saddr = g_new0(SocketAddress, 1);
 391     if (sockpath) {
 392         saddr->type = SOCKET_ADDRESS_TYPE_UNIX;
 393         saddr->u.q_unix.path = g_strdup(sockpath);
 394     } else {
 395         InetSocketAddress *inet;
 396         saddr->type = SOCKET_ADDRESS_TYPE_INET;
 397         inet = &saddr->u.inet;
 398         inet->host = g_strdup(bindto);
 399         if (port) {
 400             inet->port = g_strdup(port);
 401         } else  {
 402             inet->port = g_strdup_printf("%d", NBD_DEFAULT_PORT);
 403         }
 404     }
 405
 406     return saddr;
 407 }
 408
 409
 410 static QemuOptsList file_opts = {
 411     .name = "file",
 412     .implied_opt_name = "file",
 413     .head = QTAILQ_HEAD_INITIALIZER(file_opts.head),
 414     .desc = {
 415         /* no elements => accept any params */
 416         { /* end of list */ }
 417     },
 418 };
 419
 420 static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, bool list,
 421                                           Error **errp)
 422 {
 423     Object *obj;
 424     QCryptoTLSCreds *creds;
 425
 426     obj = object_resolve_path_component(
 427         object_get_objects_root(), id);
 428     if (!obj) {
 429         error_setg(errp, "No TLS credentials with id '%s'",
 430                    id);
 431         return NULL;
 432     }
 433     creds = (QCryptoTLSCreds *)
 434         object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
 435     if (!creds) {
 436         error_setg(errp, "Object with id '%s' is not TLS credentials",
 437                    id);
 438         return NULL;
 439     }
 440
 441     if (!qcrypto_tls_creds_check_endpoint(creds,
 442                                           list
 443                                           ? QCRYPTO_TLS_CREDS_ENDPOINT_CLIENT
 444                                           : QCRYPTO_TLS_CREDS_ENDPOINT_SERVER,
 445                                           errp)) {
 446         return NULL;
 447     }
 448     object_ref(obj);
 449     return creds;
 450 }
 451
 452 static void setup_address_and_port(const char **address, const char **port)
 453 {
 454     if (*address == NULL) {
 455         *address = "0.0.0.0";
 456     }
 457
 458     if (*port == NULL) {
 459         *port = stringify(NBD_DEFAULT_PORT);
 460     }
 461 }
 462
 463 /*
 464  * Check socket parameters compatibility when socket activation is used.
 465  */
 466 static const char *socket_activation_validate_opts(const char *device,
 467                                                    const char *sockpath,
 468                                                    const char *address,
 469                                                    const char *port,
 470                                                    const char *selinux,
 471                                                    bool list)
 472 {
 473     if (device != NULL) {
 474         return "NBD device can't be set when using socket activation";
 475     }
 476
 477     if (sockpath != NULL) {
 478         return "Unix socket can't be set when using socket activation";
 479     }
 480
 481     if (address != NULL) {
 482         return "The interface can't be set when using socket activation";
 483     }
 484
 485     if (port != NULL) {
 486         return "TCP port number can't be set when using socket activation";
 487     }
 488
 489     if (selinux != NULL) {
 490         return "SELinux label can't be set when using socket activation";
 491     }
 492
 493     if (list) {
 494         return "List mode is incompatible with socket activation";
 495     }
 496
 497     return NULL;
 498 }
 499
 500 static void qemu_nbd_shutdown(void)
 501 {
 502     job_cancel_sync_all();
 503     blk_exp_close_all();
 504     bdrv_close_all();
 505 }
 506
 507 int main(int argc, char **argv)
 508 {
 509     BlockBackend *blk;
 510     BlockDriverState *bs;
 511     uint64_t dev_offset = 0;
 512     bool readonly = false;
 513     bool disconnect = false;
 514     const char *bindto = NULL;
 515     const char *port = NULL;
 516     char *sockpath = NULL;
 517     char *device = NULL;
 518     QemuOpts *sn_opts = NULL;
 519     const char *sn_id_or_name = NULL;
 520     const char *sopt = "hVb:o:p:rsnc:dvk:e:f:tl:x:T:D:AB:L";
 521     struct option lopt[] = {
 522         { "help", no_argument, NULL, 'h' },
 523         { "version", no_argument, NULL, 'V' },
 524         { "bind", required_argument, NULL, 'b' },
 525         { "port", required_argument, NULL, 'p' },
 526         { "socket", required_argument, NULL, 'k' },
 527         { "offset", required_argument, NULL, 'o' },
 528         { "read-only", no_argument, NULL, 'r' },
 529         { "allocation-depth", no_argument, NULL, 'A' },
 530         { "bitmap", required_argument, NULL, 'B' },
 531         { "connect", required_argument, NULL, 'c' },
 532         { "disconnect", no_argument, NULL, 'd' },
 533         { "list", no_argument, NULL, 'L' },
 534         { "snapshot", no_argument, NULL, 's' },
 535         { "load-snapshot", required_argument, NULL, 'l' },
 536         { "nocache", no_argument, NULL, 'n' },
 537         { "cache", required_argument, NULL, QEMU_NBD_OPT_CACHE },
 538         { "aio", required_argument, NULL, QEMU_NBD_OPT_AIO },
 539         { "discard", required_argument, NULL, QEMU_NBD_OPT_DISCARD },
 540         { "detect-zeroes", required_argument, NULL,
 541           QEMU_NBD_OPT_DETECT_ZEROES },
 542         { "shared", required_argument, NULL, 'e' },
 543         { "format", required_argument, NULL, 'f' },
 544         { "persistent", no_argument, NULL, 't' },
 545         { "verbose", no_argument, NULL, 'v' },
 546         { "object", required_argument, NULL, QEMU_NBD_OPT_OBJECT },
 547         { "export-name", required_argument, NULL, 'x' },
 548         { "description", required_argument, NULL, 'D' },
 549         { "tls-creds", required_argument, NULL, QEMU_NBD_OPT_TLSCREDS },
 550         { "tls-hostname", required_argument, NULL, QEMU_NBD_OPT_TLSHOSTNAME },
 551         { "tls-authz", required_argument, NULL, QEMU_NBD_OPT_TLSAUTHZ },
 552         { "image-opts", no_argument, NULL, QEMU_NBD_OPT_IMAGE_OPTS },
 553         { "trace", required_argument, NULL, 'T' },
 554         { "fork", no_argument, NULL, QEMU_NBD_OPT_FORK },
 555         { "pid-file", required_argument, NULL, QEMU_NBD_OPT_PID_FILE },
 556         { "selinux-label", required_argument, NULL,
 557           QEMU_NBD_OPT_SELINUX_LABEL },
 558         { NULL, 0, NULL, 0 }
 559     };
 560     int ch;
 561     int opt_ind = 0;
 562     int flags = BDRV_O_RDWR;
 563     int ret = 0;
 564     bool seen_cache = false;
 565     bool seen_discard = false;
 566     bool seen_aio = false;
 567     pthread_t client_thread;
 568     const char *fmt = NULL;
 569     Error *local_err = NULL;
 570     BlockdevDetectZeroesOptions detect_zeroes = BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
 571     QDict *options = NULL;
 572     const char *export_name = NULL; /* defaults to "" later for server mode */
 573     const char *export_description = NULL;
 574     BlockDirtyBitmapOrStrList *bitmaps = NULL;
 575     bool alloc_depth = false;
 576     const char *tlscredsid = NULL;
 577     const char *tlshostname = NULL;
 578     bool imageOpts = false;
 579     bool writethrough = false; /* Client will flush as needed. */
 580     bool fork_process = false;
 581     bool list = false;
 582     int old_stderr = -1;
 583     unsigned socket_activation;
 584     const char *pid_file_name = NULL;
 585     const char *selinux_label = NULL;
 586     BlockExportOptions *export_opts;
 587
 588 #ifdef CONFIG_POSIX
 589     os_setup_early_signal_handling();
 590     os_setup_signal_handling();
 591 #endif
 592
 593     socket_init();
 594     error_init(argv[0]);
 595     module_call_init(MODULE_INIT_TRACE);
 596     qcrypto_init(&error_fatal);
 597
 598     module_call_init(MODULE_INIT_QOM);
 599     qemu_add_opts(&qemu_trace_opts);
 600     qemu_init_exec_dir(argv[0]);
 601
 602     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 603         switch (ch) {
 604         case 's':
 605             flags |= BDRV_O_SNAPSHOT;
 606             break;
 607         case 'n':
 608             optarg = (char *) "none";
 609             /* fallthrough */
 610         case QEMU_NBD_OPT_CACHE:
 611             if (seen_cache) {
 612                 error_report("-n and --cache can only be specified once");
 613                 exit(EXIT_FAILURE);
 614             }
 615             seen_cache = true;
 616             if (bdrv_parse_cache_mode(optarg, &flags, &writethrough) == -1) {
 617                 error_report("Invalid cache mode `%s'", optarg);
 618                 exit(EXIT_FAILURE);
 619             }
 620             break;
 621         case QEMU_NBD_OPT_AIO:
 622             if (seen_aio) {
 623                 error_report("--aio can only be specified once");
 624                 exit(EXIT_FAILURE);
 625             }
 626             seen_aio = true;
 627             if (bdrv_parse_aio(optarg, &flags) < 0) {
 628                 error_report("Invalid aio mode '%s'", optarg);
 629                 exit(EXIT_FAILURE);
 630             }
 631             break;
 632         case QEMU_NBD_OPT_DISCARD:
 633             if (seen_discard) {
 634                 error_report("--discard can only be specified once");
 635                 exit(EXIT_FAILURE);
 636             }
 637             seen_discard = true;
 638             if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
 639                 error_report("Invalid discard mode `%s'", optarg);
 640                 exit(EXIT_FAILURE);
 641             }
 642             break;
 643         case QEMU_NBD_OPT_DETECT_ZEROES:
 644             detect_zeroes =
 645                 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
 646                                 optarg,
 647                                 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
 648                                 &local_err);
 649             if (local_err) {
 650                 error_reportf_err(local_err,
 651                                   "Failed to parse detect_zeroes mode: ");
 652                 exit(EXIT_FAILURE);
 653             }
 654             if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
 655                 !(flags & BDRV_O_UNMAP)) {
 656                 error_report("setting detect-zeroes to unmap is not allowed "
 657                              "without setting discard operation to unmap");
 658                 exit(EXIT_FAILURE);
 659             }
 660             break;
 661         case 'b':
 662             bindto = optarg;
 663             break;
 664         case 'p':
 665             port = optarg;
 666             break;
 667         case 'o':
 668             if (qemu_strtou64(optarg, NULL, 0, &dev_offset) < 0) {
 669                 error_report("Invalid offset '%s'", optarg);
 670                 exit(EXIT_FAILURE);
 671             }
 672             break;
 673         case 'l':
 674             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
 675                 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
 676                                                   optarg, false);
 677                 if (!sn_opts) {
 678                     error_report("Failed in parsing snapshot param `%s'",
 679                                  optarg);
 680                     exit(EXIT_FAILURE);
 681                 }
 682             } else {
 683                 sn_id_or_name = optarg;
 684             }
 685             /* fall through */
 686         case 'r':
 687             readonly = true;
 688             flags &= ~BDRV_O_RDWR;
 689             break;
 690         case 'A':
 691             alloc_depth = true;
 692             break;
 693         case 'B':
 694             {
 695                 BlockDirtyBitmapOrStr *el = g_new(BlockDirtyBitmapOrStr, 1);
 696                 *el = (BlockDirtyBitmapOrStr) {
 697                     .type = QTYPE_QSTRING,
 698                     .u.local = g_strdup(optarg),
 699                 };
 700                 QAPI_LIST_PREPEND(bitmaps, el);
 701             }
 702             break;
 703         case 'k':
 704             sockpath = optarg;
 705             if (sockpath[0] != '/') {
 706                 error_report("socket path must be absolute");
 707                 exit(EXIT_FAILURE);
 708             }
 709             break;
 710         case 'd':
 711             disconnect = true;
 712             break;
 713         case 'c':
 714             device = optarg;
 715             break;
 716         case 'e':
 717             if (qemu_strtoi(optarg, NULL, 0, &shared) < 0 ||
 718                 shared < 0) {
 719                 error_report("Invalid shared device number '%s'", optarg);
 720                 exit(EXIT_FAILURE);
 721             }
 722             break;
 723         case 'f':
 724             fmt = optarg;
 725             break;
 726         case 't':
 727             persistent = 1;
 728             break;
 729         case 'x':
 730             export_name = optarg;
 731             if (strlen(export_name) > NBD_MAX_STRING_SIZE) {
 732                 error_report("export name '%s' too long", export_name);
 733                 exit(EXIT_FAILURE);
 734             }
 735             break;
 736         case 'D':
 737             export_description = optarg;
 738             if (strlen(export_description) > NBD_MAX_STRING_SIZE) {
 739                 error_report("export description '%s' too long",
 740                              export_description);
 741                 exit(EXIT_FAILURE);
 742             }
 743             break;
 744         case 'v':
 745             verbose = 1;
 746             break;
 747         case 'V':
 748             version(argv[0]);
 749             exit(0);
 750             break;
 751         case 'h':
 752             usage(argv[0]);
 753             exit(0);
 754             break;
 755         case '?':
 756             error_report("Try `%s --help' for more information.", argv[0]);
 757             exit(EXIT_FAILURE);
 758         case QEMU_NBD_OPT_OBJECT:
 759             user_creatable_process_cmdline(optarg);
 760             break;
 761         case QEMU_NBD_OPT_TLSCREDS:
 762             tlscredsid = optarg;
 763             break;
 764         case QEMU_NBD_OPT_TLSHOSTNAME:
 765             tlshostname = optarg;
 766             break;
 767         case QEMU_NBD_OPT_IMAGE_OPTS:
 768             imageOpts = true;
 769             break;
 770         case 'T':
 771             trace_opt_parse(optarg);
 772             break;
 773         case QEMU_NBD_OPT_TLSAUTHZ:
 774             tlsauthz = optarg;
 775             break;
 776         case QEMU_NBD_OPT_FORK:
 777             fork_process = true;
 778             break;
 779         case 'L':
 780             list = true;
 781             break;
 782         case QEMU_NBD_OPT_PID_FILE:
 783             pid_file_name = optarg;
 784             break;
 785         case QEMU_NBD_OPT_SELINUX_LABEL:
 786             selinux_label = optarg;
 787             break;
 788         }
 789     }
 790
 791     if (list) {
 792         if (argc != optind) {
 793             error_report("List mode is incompatible with a file name");
 794             exit(EXIT_FAILURE);
 795         }
 796         if (export_name || export_description || dev_offset ||
 797             device || disconnect || fmt || sn_id_or_name || bitmaps ||
 798             alloc_depth || seen_aio || seen_discard || seen_cache) {
 799             error_report("List mode is incompatible with per-device settings");
 800             exit(EXIT_FAILURE);
 801         }
 802         if (fork_process) {
 803             error_report("List mode is incompatible with forking");
 804             exit(EXIT_FAILURE);
 805         }
 806     } else if ((argc - optind) != 1) {
 807         error_report("Invalid number of arguments");
 808         error_printf("Try `%s --help' for more information.\n", argv[0]);
 809         exit(EXIT_FAILURE);
 810     } else if (!export_name) {
 811         export_name = "";
 812     }
 813
 814     if (!trace_init_backends()) {
 815         exit(1);
 816     }
 817     trace_init_file();
 818     qemu_set_log(LOG_TRACE, &error_fatal);
 819
 820     socket_activation = check_socket_activation();
 821     if (socket_activation == 0) {
 822         if (!sockpath) {
 823             setup_address_and_port(&bindto, &port);
 824         }
 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                                                               selinux_label,
 830                                                               list);
 831         if (err_msg != NULL) {
 832             error_report("%s", err_msg);
 833             exit(EXIT_FAILURE);
 834         }
 835
 836         /* qemu-nbd can only listen on a single socket.  */
 837         if (socket_activation > 1) {
 838             error_report("qemu-nbd does not support socket activation with %s > 1",
 839                          "LISTEN_FDS");
 840             exit(EXIT_FAILURE);
 841         }
 842     }
 843
 844     if (tlscredsid) {
 845         if (device) {
 846             error_report("TLS is not supported with a host device");
 847             exit(EXIT_FAILURE);
 848         }
 849         if (tlsauthz && list) {
 850             error_report("TLS authorization is incompatible with export list");
 851             exit(EXIT_FAILURE);
 852         }
 853         if (tlshostname && !list) {
 854             error_report("TLS hostname is only supported with export list");
 855             exit(EXIT_FAILURE);
 856         }
 857         tlscreds = nbd_get_tls_creds(tlscredsid, list, &local_err);
 858         if (local_err) {
 859             error_reportf_err(local_err, "Failed to get TLS creds: ");
 860             exit(EXIT_FAILURE);
 861         }
 862     } else {
 863         if (tlsauthz) {
 864             error_report("--tls-authz is not permitted without --tls-creds");
 865             exit(EXIT_FAILURE);
 866         }
 867         if (tlshostname) {
 868             error_report("--tls-hostname is not permitted without --tls-creds");
 869             exit(EXIT_FAILURE);
 870         }
 871     }
 872
 873     if (selinux_label) {
 874 #ifdef CONFIG_SELINUX
 875         if (sockpath == NULL && device == NULL) {
 876             error_report("--selinux-label is not permitted without --socket");
 877             exit(EXIT_FAILURE);
 878         }
 879 #else
 880         error_report("SELinux support not enabled in this binary");
 881         exit(EXIT_FAILURE);
 882 #endif
 883     }
 884
 885     if (list) {
 886         saddr = nbd_build_socket_address(sockpath, bindto, port);
 887         return qemu_nbd_client_list(saddr, tlscreds,
 888                                     tlshostname ? tlshostname : bindto);
 889     }
 890
 891 #if !HAVE_NBD_DEVICE
 892     if (disconnect || device) {
 893         error_report("Kernel /dev/nbdN support not available");
 894         exit(EXIT_FAILURE);
 895     }
 896 #else /* HAVE_NBD_DEVICE */
 897     if (disconnect) {
 898         int nbdfd = open(argv[optind], O_RDWR);
 899         if (nbdfd < 0) {
 900             error_report("Cannot open %s: %s", argv[optind],
 901                          strerror(errno));
 902             exit(EXIT_FAILURE);
 903         }
 904         nbd_disconnect(nbdfd);
 905
 906         close(nbdfd);
 907
 908         printf("%s disconnected\n", argv[optind]);
 909
 910         return 0;
 911     }
 912 #endif
 913
 914     if ((device && !verbose) || fork_process) {
 915 #ifndef WIN32
 916         g_autoptr(GError) err = NULL;
 917         int stderr_fd[2];
 918         pid_t pid;
 919         int ret;
 920
 921         if (!g_unix_open_pipe(stderr_fd, FD_CLOEXEC, &err)) {
 922             error_report("Error setting up communication pipe: %s",
 923                          err->message);
 924             exit(EXIT_FAILURE);
 925         }
 926
 927         /* Now daemonize, but keep a communication channel open to
 928          * print errors and exit with the proper status code.
 929          */
 930         pid = fork();
 931         if (pid < 0) {
 932             error_report("Failed to fork: %s", strerror(errno));
 933             exit(EXIT_FAILURE);
 934         } else if (pid == 0) {
 935             close(stderr_fd[0]);
 936
 937             /* Remember parent's stderr if we will be restoring it. */
 938             if (fork_process) {
 939                 old_stderr = dup(STDERR_FILENO);
 940             }
 941
 942             ret = qemu_daemon(1, 0);
 943
 944             /* Temporarily redirect stderr to the parent's pipe...  */
 945             dup2(stderr_fd[1], STDERR_FILENO);
 946             if (ret < 0) {
 947                 error_report("Failed to daemonize: %s", strerror(errno));
 948                 exit(EXIT_FAILURE);
 949             }
 950
 951             /* ... close the descriptor we inherited and go on.  */
 952             close(stderr_fd[1]);
 953         } else {
 954             bool errors = false;
 955             char *buf;
 956
 957             /* In the parent.  Print error messages from the child until
 958              * it closes the pipe.
 959              */
 960             close(stderr_fd[1]);
 961             buf = g_malloc(1024);
 962             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
 963                 errors = true;
 964                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
 965                 if (ret < 0) {
 966                     exit(EXIT_FAILURE);
 967                 }
 968             }
 969             if (ret < 0) {
 970                 error_report("Cannot read from daemon: %s",
 971                              strerror(errno));
 972                 exit(EXIT_FAILURE);
 973             }
 974
 975             /* Usually the daemon should not print any message.
 976              * Exit with zero status in that case.
 977              */
 978             exit(errors);
 979         }
 980 #else /* WIN32 */
 981         error_report("Unable to fork into background on Windows hosts");
 982         exit(EXIT_FAILURE);
 983 #endif /* WIN32 */
 984     }
 985
 986     if (device != NULL && sockpath == NULL) {
 987         sockpath = g_malloc(128);
 988         snprintf(sockpath, 128, SOCKET_PATH, basename(device));
 989     }
 990
 991     server = qio_net_listener_new();
 992     if (socket_activation == 0) {
 993         int backlog;
 994
 995         if (persistent || shared == 0) {
 996             backlog = SOMAXCONN;
 997         } else {
 998             backlog = MIN(shared, SOMAXCONN);
 999         }
1000 #ifdef CONFIG_SELINUX
1001         if (selinux_label && setsockcreatecon_raw(selinux_label) == -1) {
1002             error_report("Cannot set SELinux socket create context to %s: %s",
1003                          selinux_label, strerror(errno));
1004             exit(EXIT_FAILURE);
1005         }
1006 #endif
1007         saddr = nbd_build_socket_address(sockpath, bindto, port);
1008         if (qio_net_listener_open_sync(server, saddr, backlog,
1009                                        &local_err) < 0) {
1010             object_unref(OBJECT(server));
1011             error_report_err(local_err);
1012             exit(EXIT_FAILURE);
1013         }
1014 #ifdef CONFIG_SELINUX
1015         if (selinux_label && setsockcreatecon_raw(NULL) == -1) {
1016             error_report("Cannot clear SELinux socket create context: %s",
1017                          strerror(errno));
1018             exit(EXIT_FAILURE);
1019         }
1020 #endif
1021     } else {
1022         size_t i;
1023         /* See comment in check_socket_activation above. */
1024         for (i = 0; i < socket_activation; i++) {
1025             QIOChannelSocket *sioc;
1026             sioc = qio_channel_socket_new_fd(FIRST_SOCKET_ACTIVATION_FD + i,
1027                                              &local_err);
1028             if (sioc == NULL) {
1029                 object_unref(OBJECT(server));
1030                 error_reportf_err(local_err,
1031                                   "Failed to use socket activation: ");
1032                 exit(EXIT_FAILURE);
1033             }
1034             qio_net_listener_add(server, sioc);
1035             object_unref(OBJECT(sioc));
1036         }
1037     }
1038
1039     qemu_init_main_loop(&error_fatal);
1040     bdrv_init();
1041     atexit(qemu_nbd_shutdown);
1042
1043     srcpath = argv[optind];
1044     if (imageOpts) {
1045         QemuOpts *opts;
1046         if (fmt) {
1047             error_report("--image-opts and -f are mutually exclusive");
1048             exit(EXIT_FAILURE);
1049         }
1050         opts = qemu_opts_parse_noisily(&file_opts, srcpath, true);
1051         if (!opts) {
1052             qemu_opts_reset(&file_opts);
1053             exit(EXIT_FAILURE);
1054         }
1055         options = qemu_opts_to_qdict(opts, NULL);
1056         qemu_opts_reset(&file_opts);
1057         blk = blk_new_open(NULL, NULL, options, flags, &local_err);
1058     } else {
1059         if (fmt) {
1060             options = qdict_new();
1061             qdict_put_str(options, "driver", fmt);
1062         }
1063         blk = blk_new_open(srcpath, NULL, options, flags, &local_err);
1064     }
1065
1066     if (!blk) {
1067         error_reportf_err(local_err, "Failed to blk_new_open '%s': ",
1068                           argv[optind]);
1069         exit(EXIT_FAILURE);
1070     }
1071     bs = blk_bs(blk);
1072
1073     if (dev_offset) {
1074         QDict *raw_opts = qdict_new();
1075         qdict_put_str(raw_opts, "driver", "raw");
1076         qdict_put_str(raw_opts, "file", bs->node_name);
1077         qdict_put_int(raw_opts, "offset", dev_offset);
1078
1079         aio_context_acquire(qemu_get_aio_context());
1080         bs = bdrv_open(NULL, NULL, raw_opts, flags, &error_fatal);
1081         aio_context_release(qemu_get_aio_context());
1082
1083         blk_remove_bs(blk);
1084         blk_insert_bs(blk, bs, &error_fatal);
1085         bdrv_unref(bs);
1086     }
1087
1088     blk_set_enable_write_cache(blk, !writethrough);
1089
1090     if (sn_opts) {
1091         ret = bdrv_snapshot_load_tmp(bs,
1092                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
1093                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
1094                                      &local_err);
1095     } else if (sn_id_or_name) {
1096         ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
1097                                                    &local_err);
1098     }
1099     if (ret < 0) {
1100         error_reportf_err(local_err, "Failed to load snapshot: ");
1101         exit(EXIT_FAILURE);
1102     }
1103
1104     bs->detect_zeroes = detect_zeroes;
1105
1106     nbd_server_is_qemu_nbd(shared);
1107
1108     export_opts = g_new(BlockExportOptions, 1);
1109     *export_opts = (BlockExportOptions) {
1110         .type               = BLOCK_EXPORT_TYPE_NBD,
1111         .id                 = g_strdup("qemu-nbd-export"),
1112         .node_name          = g_strdup(bdrv_get_node_name(bs)),
1113         .has_writethrough   = true,
1114         .writethrough       = writethrough,
1115         .has_writable       = true,
1116         .writable           = !readonly,
1117         .u.nbd = {
1118             .name                 = g_strdup(export_name),
1119             .description          = g_strdup(export_description),
1120             .has_bitmaps          = !!bitmaps,
1121             .bitmaps              = bitmaps,
1122             .has_allocation_depth = alloc_depth,
1123             .allocation_depth     = alloc_depth,
1124         },
1125     };
1126     blk_exp_add(export_opts, &error_fatal);
1127     qapi_free_BlockExportOptions(export_opts);
1128
1129     if (device) {
1130 #if HAVE_NBD_DEVICE
1131         int ret;
1132         struct NbdClientOpts opts = {
1133             .device = device,
1134         };
1135
1136         ret = pthread_create(&client_thread, NULL, nbd_client_thread, &opts);
1137         if (ret != 0) {
1138             error_report("Failed to create client thread: %s", strerror(ret));
1139             exit(EXIT_FAILURE);
1140         }
1141 #endif
1142     } else {
1143         /* Shut up GCC warnings.  */
1144         memset(&client_thread, 0, sizeof(client_thread));
1145     }
1146
1147     nbd_update_server_watch();
1148
1149     if (pid_file_name) {
1150         qemu_write_pidfile(pid_file_name, &error_fatal);
1151     }
1152
1153     /* now when the initialization is (almost) complete, chdir("/")
1154      * to free any busy filesystems */
1155     if (chdir("/") < 0) {
1156         error_report("Could not chdir to root directory: %s",
1157                      strerror(errno));
1158         exit(EXIT_FAILURE);
1159     }
1160
1161     if (fork_process) {
1162         dup2(old_stderr, STDERR_FILENO);
1163         close(old_stderr);
1164     }
1165
1166     state = RUNNING;
1167     do {
1168         main_loop_wait(false);
1169         if (state == TERMINATE) {
1170             blk_exp_close_all();
1171             state = TERMINATED;
1172         }
1173     } while (state != TERMINATED);
1174
1175     blk_unref(blk);
1176     if (sockpath) {
1177         unlink(sockpath);
1178     }
1179
1180     qemu_opts_del(sn_opts);
1181
1182     if (device) {
1183         void *ret;
1184         pthread_join(client_thread, &ret);
1185         exit(ret != NULL);
1186     } else {
1187         exit(EXIT_SUCCESS);
1188     }
1189 }