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 "block/block_int.h"
  29 #include "block/nbd.h"
  30 #include "qemu/main-loop.h"
  31 #include "qemu/module.h"
  32 #include "qemu/option.h"
  33 #include "qemu/error-report.h"
  34 #include "qemu/config-file.h"
  35 #include "qemu/bswap.h"
  36 #include "qemu/log.h"
  37 #include "qemu/systemd.h"
  38 #include "block/snapshot.h"
  39 #include "qapi/qmp/qdict.h"
  40 #include "qapi/qmp/qstring.h"
  41 #include "qom/object_interfaces.h"
  42 #include "io/channel-socket.h"
  43 #include "io/net-listener.h"
  44 #include "crypto/init.h"
  45 #include "trace/control.h"
  46 #include "qemu-version.h"
  47
  48 #ifdef __linux__
  49 #define HAVE_NBD_DEVICE 1
  50 #else
  51 #define HAVE_NBD_DEVICE 0
  52 #endif
  53
  54 #define SOCKET_PATH                "/var/lock/qemu-nbd-%s"
  55 #define QEMU_NBD_OPT_CACHE         256
  56 #define QEMU_NBD_OPT_AIO           257
  57 #define QEMU_NBD_OPT_DISCARD       258
  58 #define QEMU_NBD_OPT_DETECT_ZEROES 259
  59 #define QEMU_NBD_OPT_OBJECT        260
  60 #define QEMU_NBD_OPT_TLSCREDS      261
  61 #define QEMU_NBD_OPT_IMAGE_OPTS    262
  62 #define QEMU_NBD_OPT_FORK          263
  63 #define QEMU_NBD_OPT_TLSAUTHZ      264
  64 #define QEMU_NBD_OPT_PID_FILE      265
  65
  66 #define MBR_SIZE 512
  67
  68 static NBDExport *export;
  69 static int verbose;
  70 static char *srcpath;
  71 static SocketAddress *saddr;
  72 static int persistent = 0;
  73 static enum { RUNNING, TERMINATE, TERMINATING, 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 "  -B, --bitmap=NAME         expose a persistent dirty bitmap\n"
 104 "\n"
 105 "General purpose options:\n"
 106 "  -L, --list                list exports available from another NBD server\n"
 107 "  --object type,id=ID,...   define an object such as 'secret' for providing\n"
 108 "                            passwords and/or encryption keys\n"
 109 "  --tls-creds=ID            use id of an earlier --object to provide TLS\n"
 110 "  --tls-authz=ID            use id of an earlier --object to provide\n"
 111 "                            authorization\n"
 112 "  -T, --trace [[enable=]<pattern>][,events=<file>][,file=<file>]\n"
 113 "                            specify tracing options\n"
 114 "  --fork                    fork off the server process and exit the parent\n"
 115 "                            once the server is running\n"
 116 "  --pid-file=PATH           store the server's process ID in the given file\n"
 117 #if HAVE_NBD_DEVICE
 118 "\n"
 119 "Kernel NBD client support:\n"
 120 "  -c, --connect=DEV         connect FILE to the local NBD device DEV\n"
 121 "  -d, --disconnect          disconnect the specified device\n"
 122 #endif
 123 "\n"
 124 "Block device options:\n"
 125 "  -f, --format=FORMAT       set image format (raw, qcow2, ...)\n"
 126 "  -r, --read-only           export read-only\n"
 127 "  -s, --snapshot            use FILE as an external snapshot, create a temporary\n"
 128 "                            file with backing_file=FILE, redirect the write to\n"
 129 "                            the temporary one\n"
 130 "  -l, --load-snapshot=SNAPSHOT_PARAM\n"
 131 "                            load an internal snapshot inside FILE and export it\n"
 132 "                            as an read-only device, SNAPSHOT_PARAM format is\n"
 133 "                            'snapshot.id=[ID],snapshot.name=[NAME]', or\n"
 134 "                            '[ID_OR_NAME]'\n"
 135 "  -n, --nocache             disable host cache\n"
 136 "      --cache=MODE          set cache mode (none, writeback, ...)\n"
 137 "      --aio=MODE            set AIO mode (native, io_uring or threads)\n"
 138 "      --discard=MODE        set discard mode (ignore, unmap)\n"
 139 "      --detect-zeroes=MODE  set detect-zeroes mode (off, on, unmap)\n"
 140 "      --image-opts          treat FILE as a full set of image options\n"
 141 "\n"
 142 QEMU_HELP_BOTTOM "\n"
 143     , name, name, NBD_DEFAULT_PORT, "DEVICE");
 144 }
 145
 146 static void version(const char *name)
 147 {
 148     printf(
 149 "%s " QEMU_FULL_VERSION "\n"
 150 "Written by Anthony Liguori.\n"
 151 "\n"
 152 QEMU_COPYRIGHT "\n"
 153 "This is free software; see the source for copying conditions.  There is NO\n"
 154 "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
 155     , name);
 156 }
 157
 158 #if HAVE_NBD_DEVICE
 159 static void termsig_handler(int signum)
 160 {
 161     qatomic_cmpxchg(&state, RUNNING, TERMINATE);
 162     qemu_notify_event();
 163 }
 164 #endif /* HAVE_NBD_DEVICE */
 165
 166 static int qemu_nbd_client_list(SocketAddress *saddr, QCryptoTLSCreds *tls,
 167                                 const char *hostname)
 168 {
 169     int ret = EXIT_FAILURE;
 170     int rc;
 171     Error *err = NULL;
 172     QIOChannelSocket *sioc;
 173     NBDExportInfo *list;
 174     int i, j;
 175
 176     sioc = qio_channel_socket_new();
 177     if (qio_channel_socket_connect_sync(sioc, saddr, &err) < 0) {
 178         error_report_err(err);
 179         return EXIT_FAILURE;
 180     }
 181     rc = nbd_receive_export_list(QIO_CHANNEL(sioc), tls, hostname, &list,
 182                                  &err);
 183     if (rc < 0) {
 184         if (err) {
 185             error_report_err(err);
 186         }
 187         goto out;
 188     }
 189     printf("exports available: %d\n", rc);
 190     for (i = 0; i < rc; i++) {
 191         printf(" export: '%s'\n", list[i].name);
 192         if (list[i].description && *list[i].description) {
 193             printf("  description: %s\n", list[i].description);
 194         }
 195         if (list[i].flags & NBD_FLAG_HAS_FLAGS) {
 196             static const char *const flag_names[] = {
 197                 [NBD_FLAG_READ_ONLY_BIT]            = "readonly",
 198                 [NBD_FLAG_SEND_FLUSH_BIT]           = "flush",
 199                 [NBD_FLAG_SEND_FUA_BIT]             = "fua",
 200                 [NBD_FLAG_ROTATIONAL_BIT]           = "rotational",
 201                 [NBD_FLAG_SEND_TRIM_BIT]            = "trim",
 202                 [NBD_FLAG_SEND_WRITE_ZEROES_BIT]    = "zeroes",
 203                 [NBD_FLAG_SEND_DF_BIT]              = "df",
 204                 [NBD_FLAG_CAN_MULTI_CONN_BIT]       = "multi",
 205                 [NBD_FLAG_SEND_RESIZE_BIT]          = "resize",
 206                 [NBD_FLAG_SEND_CACHE_BIT]           = "cache",
 207                 [NBD_FLAG_SEND_FAST_ZERO_BIT]       = "fast-zero",
 208             };
 209
 210             printf("  size:  %" PRIu64 "\n", list[i].size);
 211             printf("  flags: 0x%x (", list[i].flags);
 212             for (size_t bit = 0; bit < ARRAY_SIZE(flag_names); bit++) {
 213                 if (flag_names[bit] && (list[i].flags & (1 << bit))) {
 214                     printf(" %s", flag_names[bit]);
 215                 }
 216             }
 217             printf(" )\n");
 218         }
 219         if (list[i].min_block) {
 220             printf("  min block: %u\n", list[i].min_block);
 221             printf("  opt block: %u\n", list[i].opt_block);
 222             printf("  max block: %u\n", list[i].max_block);
 223         }
 224         if (list[i].n_contexts) {
 225             printf("  available meta contexts: %d\n", list[i].n_contexts);
 226             for (j = 0; j < list[i].n_contexts; j++) {
 227                 printf("   %s\n", list[i].contexts[j]);
 228             }
 229         }
 230     }
 231     nbd_free_export_list(list, rc);
 232
 233     ret = EXIT_SUCCESS;
 234  out:
 235     object_unref(OBJECT(sioc));
 236     return ret;
 237 }
 238
 239
 240 #if HAVE_NBD_DEVICE
 241 static void *show_parts(void *arg)
 242 {
 243     char *device = arg;
 244     int nbd;
 245
 246     /* linux just needs an open() to trigger
 247      * the partition table update
 248      * but remember to load the module with max_part != 0 :
 249      *     modprobe nbd max_part=63
 250      */
 251     nbd = open(device, O_RDWR);
 252     if (nbd >= 0) {
 253         close(nbd);
 254     }
 255     return NULL;
 256 }
 257
 258 static void *nbd_client_thread(void *arg)
 259 {
 260     char *device = arg;
 261     NBDExportInfo info = { .request_sizes = false, .name = g_strdup("") };
 262     QIOChannelSocket *sioc;
 263     int fd;
 264     int ret;
 265     pthread_t show_parts_thread;
 266     Error *local_error = NULL;
 267
 268     sioc = qio_channel_socket_new();
 269     if (qio_channel_socket_connect_sync(sioc,
 270                                         saddr,
 271                                         &local_error) < 0) {
 272         error_report_err(local_error);
 273         goto out;
 274     }
 275
 276     ret = nbd_receive_negotiate(NULL, QIO_CHANNEL(sioc),
 277                                 NULL, NULL, NULL, &info, &local_error);
 278     if (ret < 0) {
 279         if (local_error) {
 280             error_report_err(local_error);
 281         }
 282         goto out_socket;
 283     }
 284
 285     fd = open(device, O_RDWR);
 286     if (fd < 0) {
 287         /* Linux-only, we can use %m in printf.  */
 288         error_report("Failed to open %s: %m", device);
 289         goto out_socket;
 290     }
 291
 292     ret = nbd_init(fd, sioc, &info, &local_error);
 293     if (ret < 0) {
 294         error_report_err(local_error);
 295         goto out_fd;
 296     }
 297
 298     /* update partition table */
 299     pthread_create(&show_parts_thread, NULL, show_parts, device);
 300
 301     if (verbose) {
 302         fprintf(stderr, "NBD device %s is now connected to %s\n",
 303                 device, srcpath);
 304     } else {
 305         /* Close stderr so that the qemu-nbd process exits.  */
 306         dup2(STDOUT_FILENO, STDERR_FILENO);
 307     }
 308
 309     ret = nbd_client(fd);
 310     if (ret) {
 311         goto out_fd;
 312     }
 313     close(fd);
 314     object_unref(OBJECT(sioc));
 315     g_free(info.name);
 316     kill(getpid(), SIGTERM);
 317     return (void *) EXIT_SUCCESS;
 318
 319 out_fd:
 320     close(fd);
 321 out_socket:
 322     object_unref(OBJECT(sioc));
 323 out:
 324     g_free(info.name);
 325     kill(getpid(), SIGTERM);
 326     return (void *) EXIT_FAILURE;
 327 }
 328 #endif /* HAVE_NBD_DEVICE */
 329
 330 static int nbd_can_accept(void)
 331 {
 332     return state == RUNNING && nb_fds < shared;
 333 }
 334
 335 static void nbd_export_closed(NBDExport *export)
 336 {
 337     assert(state == TERMINATING);
 338     state = TERMINATED;
 339 }
 340
 341 static void nbd_update_server_watch(void);
 342
 343 static void nbd_client_closed(NBDClient *client, bool negotiated)
 344 {
 345     nb_fds--;
 346     if (negotiated && nb_fds == 0 && !persistent && state == RUNNING) {
 347         state = TERMINATE;
 348     }
 349     nbd_update_server_watch();
 350     nbd_client_put(client);
 351 }
 352
 353 static void nbd_accept(QIONetListener *listener, QIOChannelSocket *cioc,
 354                        gpointer opaque)
 355 {
 356     if (state >= TERMINATE) {
 357         return;
 358     }
 359
 360     nb_fds++;
 361     nbd_update_server_watch();
 362     nbd_client_new(cioc, tlscreds, tlsauthz, nbd_client_closed);
 363 }
 364
 365 static void nbd_update_server_watch(void)
 366 {
 367     if (nbd_can_accept()) {
 368         qio_net_listener_set_client_func(server, nbd_accept, NULL, NULL);
 369     } else {
 370         qio_net_listener_set_client_func(server, NULL, NULL, NULL);
 371     }
 372 }
 373
 374
 375 static SocketAddress *nbd_build_socket_address(const char *sockpath,
 376                                                const char *bindto,
 377                                                const char *port)
 378 {
 379     SocketAddress *saddr;
 380
 381     saddr = g_new0(SocketAddress, 1);
 382     if (sockpath) {
 383         saddr->type = SOCKET_ADDRESS_TYPE_UNIX;
 384         saddr->u.q_unix.path = g_strdup(sockpath);
 385     } else {
 386         InetSocketAddress *inet;
 387         saddr->type = SOCKET_ADDRESS_TYPE_INET;
 388         inet = &saddr->u.inet;
 389         inet->host = g_strdup(bindto);
 390         if (port) {
 391             inet->port = g_strdup(port);
 392         } else  {
 393             inet->port = g_strdup_printf("%d", NBD_DEFAULT_PORT);
 394         }
 395     }
 396
 397     return saddr;
 398 }
 399
 400
 401 static QemuOptsList file_opts = {
 402     .name = "file",
 403     .implied_opt_name = "file",
 404     .head = QTAILQ_HEAD_INITIALIZER(file_opts.head),
 405     .desc = {
 406         /* no elements => accept any params */
 407         { /* end of list */ }
 408     },
 409 };
 410
 411 static QemuOptsList qemu_object_opts = {
 412     .name = "object",
 413     .implied_opt_name = "qom-type",
 414     .head = QTAILQ_HEAD_INITIALIZER(qemu_object_opts.head),
 415     .desc = {
 416         { }
 417     },
 418 };
 419
 420 static bool qemu_nbd_object_print_help(const char *type, QemuOpts *opts)
 421 {
 422     if (user_creatable_print_help(type, opts)) {
 423         exit(0);
 424     }
 425     return true;
 426 }
 427
 428
 429 static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, bool list,
 430                                           Error **errp)
 431 {
 432     Object *obj;
 433     QCryptoTLSCreds *creds;
 434
 435     obj = object_resolve_path_component(
 436         object_get_objects_root(), id);
 437     if (!obj) {
 438         error_setg(errp, "No TLS credentials with id '%s'",
 439                    id);
 440         return NULL;
 441     }
 442     creds = (QCryptoTLSCreds *)
 443         object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
 444     if (!creds) {
 445         error_setg(errp, "Object with id '%s' is not TLS credentials",
 446                    id);
 447         return NULL;
 448     }
 449
 450     if (list) {
 451         if (creds->endpoint != QCRYPTO_TLS_CREDS_ENDPOINT_CLIENT) {
 452             error_setg(errp,
 453                        "Expecting TLS credentials with a client endpoint");
 454             return NULL;
 455         }
 456     } else {
 457         if (creds->endpoint != QCRYPTO_TLS_CREDS_ENDPOINT_SERVER) {
 458             error_setg(errp,
 459                        "Expecting TLS credentials with a server endpoint");
 460             return NULL;
 461         }
 462     }
 463     object_ref(obj);
 464     return creds;
 465 }
 466
 467 static void setup_address_and_port(const char **address, const char **port)
 468 {
 469     if (*address == NULL) {
 470         *address = "0.0.0.0";
 471     }
 472
 473     if (*port == NULL) {
 474         *port = stringify(NBD_DEFAULT_PORT);
 475     }
 476 }
 477
 478 /*
 479  * Check socket parameters compatibility when socket activation is used.
 480  */
 481 static const char *socket_activation_validate_opts(const char *device,
 482                                                    const char *sockpath,
 483                                                    const char *address,
 484                                                    const char *port,
 485                                                    bool list)
 486 {
 487     if (device != NULL) {
 488         return "NBD device can't be set when using socket activation";
 489     }
 490
 491     if (sockpath != NULL) {
 492         return "Unix socket can't be set when using socket activation";
 493     }
 494
 495     if (address != NULL) {
 496         return "The interface can't be set when using socket activation";
 497     }
 498
 499     if (port != NULL) {
 500         return "TCP port number can't be set when using socket activation";
 501     }
 502
 503     if (list) {
 504         return "List mode is incompatible with socket activation";
 505     }
 506
 507     return NULL;
 508 }
 509
 510 static void qemu_nbd_shutdown(void)
 511 {
 512     job_cancel_sync_all();
 513     bdrv_close_all();
 514 }
 515
 516 int main(int argc, char **argv)
 517 {
 518     BlockBackend *blk;
 519     BlockDriverState *bs;
 520     uint64_t dev_offset = 0;
 521     bool readonly = false;
 522     bool disconnect = false;
 523     const char *bindto = NULL;
 524     const char *port = NULL;
 525     char *sockpath = NULL;
 526     char *device = NULL;
 527     int64_t fd_size;
 528     QemuOpts *sn_opts = NULL;
 529     const char *sn_id_or_name = NULL;
 530     const char *sopt = "hVb:o:p:rsnc:dvk:e:f:tl:x:T:D:B:L";
 531     struct option lopt[] = {
 532         { "help", no_argument, NULL, 'h' },
 533         { "version", no_argument, NULL, 'V' },
 534         { "bind", required_argument, NULL, 'b' },
 535         { "port", required_argument, NULL, 'p' },
 536         { "socket", required_argument, NULL, 'k' },
 537         { "offset", required_argument, NULL, 'o' },
 538         { "read-only", no_argument, NULL, 'r' },
 539         { "bitmap", required_argument, NULL, 'B' },
 540         { "connect", required_argument, NULL, 'c' },
 541         { "disconnect", no_argument, NULL, 'd' },
 542         { "list", no_argument, NULL, 'L' },
 543         { "snapshot", no_argument, NULL, 's' },
 544         { "load-snapshot", required_argument, NULL, 'l' },
 545         { "nocache", no_argument, NULL, 'n' },
 546         { "cache", required_argument, NULL, QEMU_NBD_OPT_CACHE },
 547         { "aio", required_argument, NULL, QEMU_NBD_OPT_AIO },
 548         { "discard", required_argument, NULL, QEMU_NBD_OPT_DISCARD },
 549         { "detect-zeroes", required_argument, NULL,
 550           QEMU_NBD_OPT_DETECT_ZEROES },
 551         { "shared", required_argument, NULL, 'e' },
 552         { "format", required_argument, NULL, 'f' },
 553         { "persistent", no_argument, NULL, 't' },
 554         { "verbose", no_argument, NULL, 'v' },
 555         { "object", required_argument, NULL, QEMU_NBD_OPT_OBJECT },
 556         { "export-name", required_argument, NULL, 'x' },
 557         { "description", required_argument, NULL, 'D' },
 558         { "tls-creds", required_argument, NULL, QEMU_NBD_OPT_TLSCREDS },
 559         { "tls-authz", required_argument, NULL, QEMU_NBD_OPT_TLSAUTHZ },
 560         { "image-opts", no_argument, NULL, QEMU_NBD_OPT_IMAGE_OPTS },
 561         { "trace", required_argument, NULL, 'T' },
 562         { "fork", no_argument, NULL, QEMU_NBD_OPT_FORK },
 563         { "pid-file", required_argument, NULL, QEMU_NBD_OPT_PID_FILE },
 564         { NULL, 0, NULL, 0 }
 565     };
 566     int ch;
 567     int opt_ind = 0;
 568     int flags = BDRV_O_RDWR;
 569     int ret = 0;
 570     bool seen_cache = false;
 571     bool seen_discard = false;
 572     bool seen_aio = false;
 573     pthread_t client_thread;
 574     const char *fmt = NULL;
 575     Error *local_err = NULL;
 576     BlockdevDetectZeroesOptions detect_zeroes = BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
 577     QDict *options = NULL;
 578     const char *export_name = NULL; /* defaults to "" later for server mode */
 579     const char *export_description = NULL;
 580     const char *bitmap = NULL;
 581     const char *tlscredsid = NULL;
 582     bool imageOpts = false;
 583     bool writethrough = true;
 584     char *trace_file = NULL;
 585     bool fork_process = false;
 586     bool list = false;
 587     int old_stderr = -1;
 588     unsigned socket_activation;
 589     const char *pid_file_name = NULL;
 590
 591 #if HAVE_NBD_DEVICE
 592     /* The client thread uses SIGTERM to interrupt the server.  A signal
 593      * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
 594      */
 595     struct sigaction sa_sigterm;
 596     memset(&sa_sigterm, 0, sizeof(sa_sigterm));
 597     sa_sigterm.sa_handler = termsig_handler;
 598     sigaction(SIGTERM, &sa_sigterm, NULL);
 599 #endif /* HAVE_NBD_DEVICE */
 600
 601 #ifdef CONFIG_POSIX
 602     signal(SIGPIPE, SIG_IGN);
 603 #endif
 604
 605     socket_init();
 606     error_init(argv[0]);
 607     module_call_init(MODULE_INIT_TRACE);
 608     qcrypto_init(&error_fatal);
 609
 610     module_call_init(MODULE_INIT_QOM);
 611     qemu_add_opts(&qemu_object_opts);
 612     qemu_add_opts(&qemu_trace_opts);
 613     qemu_init_exec_dir(argv[0]);
 614
 615     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 616         switch (ch) {
 617         case 's':
 618             flags |= BDRV_O_SNAPSHOT;
 619             break;
 620         case 'n':
 621             optarg = (char *) "none";
 622             /* fallthrough */
 623         case QEMU_NBD_OPT_CACHE:
 624             if (seen_cache) {
 625                 error_report("-n and --cache can only be specified once");
 626                 exit(EXIT_FAILURE);
 627             }
 628             seen_cache = true;
 629             if (bdrv_parse_cache_mode(optarg, &flags, &writethrough) == -1) {
 630                 error_report("Invalid cache mode `%s'", optarg);
 631                 exit(EXIT_FAILURE);
 632             }
 633             break;
 634         case QEMU_NBD_OPT_AIO:
 635             if (seen_aio) {
 636                 error_report("--aio can only be specified once");
 637                 exit(EXIT_FAILURE);
 638             }
 639             seen_aio = true;
 640             if (bdrv_parse_aio(optarg, &flags) < 0) {
 641                 error_report("Invalid aio mode '%s'", optarg);
 642                 exit(EXIT_FAILURE);
 643             }
 644             break;
 645         case QEMU_NBD_OPT_DISCARD:
 646             if (seen_discard) {
 647                 error_report("--discard can only be specified once");
 648                 exit(EXIT_FAILURE);
 649             }
 650             seen_discard = true;
 651             if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
 652                 error_report("Invalid discard mode `%s'", optarg);
 653                 exit(EXIT_FAILURE);
 654             }
 655             break;
 656         case QEMU_NBD_OPT_DETECT_ZEROES:
 657             detect_zeroes =
 658                 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
 659                                 optarg,
 660                                 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
 661                                 &local_err);
 662             if (local_err) {
 663                 error_reportf_err(local_err,
 664                                   "Failed to parse detect_zeroes mode: ");
 665                 exit(EXIT_FAILURE);
 666             }
 667             if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
 668                 !(flags & BDRV_O_UNMAP)) {
 669                 error_report("setting detect-zeroes to unmap is not allowed "
 670                              "without setting discard operation to unmap");
 671                 exit(EXIT_FAILURE);
 672             }
 673             break;
 674         case 'b':
 675             bindto = optarg;
 676             break;
 677         case 'p':
 678             port = optarg;
 679             break;
 680         case 'o':
 681             if (qemu_strtou64(optarg, NULL, 0, &dev_offset) < 0) {
 682                 error_report("Invalid offset '%s'", optarg);
 683                 exit(EXIT_FAILURE);
 684             }
 685             break;
 686         case 'l':
 687             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
 688                 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
 689                                                   optarg, false);
 690                 if (!sn_opts) {
 691                     error_report("Failed in parsing snapshot param `%s'",
 692                                  optarg);
 693                     exit(EXIT_FAILURE);
 694                 }
 695             } else {
 696                 sn_id_or_name = optarg;
 697             }
 698             /* fall through */
 699         case 'r':
 700             readonly = true;
 701             flags &= ~BDRV_O_RDWR;
 702             break;
 703         case 'B':
 704             bitmap = optarg;
 705             break;
 706         case 'k':
 707             sockpath = optarg;
 708             if (sockpath[0] != '/') {
 709                 error_report("socket path must be absolute");
 710                 exit(EXIT_FAILURE);
 711             }
 712             break;
 713         case 'd':
 714             disconnect = true;
 715             break;
 716         case 'c':
 717             device = optarg;
 718             break;
 719         case 'e':
 720             if (qemu_strtoi(optarg, NULL, 0, &shared) < 0 ||
 721                 shared < 1) {
 722                 error_report("Invalid shared device number '%s'", optarg);
 723                 exit(EXIT_FAILURE);
 724             }
 725             break;
 726         case 'f':
 727             fmt = optarg;
 728             break;
 729         case 't':
 730             persistent = 1;
 731             break;
 732         case 'x':
 733             export_name = optarg;
 734             if (strlen(export_name) > NBD_MAX_STRING_SIZE) {
 735                 error_report("export name '%s' too long", export_name);
 736                 exit(EXIT_FAILURE);
 737             }
 738             break;
 739         case 'D':
 740             export_description = optarg;
 741             if (strlen(export_description) > NBD_MAX_STRING_SIZE) {
 742                 error_report("export description '%s' too long",
 743                              export_description);
 744                 exit(EXIT_FAILURE);
 745             }
 746             break;
 747         case 'v':
 748             verbose = 1;
 749             break;
 750         case 'V':
 751             version(argv[0]);
 752             exit(0);
 753             break;
 754         case 'h':
 755             usage(argv[0]);
 756             exit(0);
 757             break;
 758         case '?':
 759             error_report("Try `%s --help' for more information.", argv[0]);
 760             exit(EXIT_FAILURE);
 761         case QEMU_NBD_OPT_OBJECT: {
 762             QemuOpts *opts;
 763             opts = qemu_opts_parse_noisily(&qemu_object_opts,
 764                                            optarg, true);
 765             if (!opts) {
 766                 exit(EXIT_FAILURE);
 767             }
 768         }   break;
 769         case QEMU_NBD_OPT_TLSCREDS:
 770             tlscredsid = optarg;
 771             break;
 772         case QEMU_NBD_OPT_IMAGE_OPTS:
 773             imageOpts = true;
 774             break;
 775         case 'T':
 776             g_free(trace_file);
 777             trace_file = trace_opt_parse(optarg);
 778             break;
 779         case QEMU_NBD_OPT_TLSAUTHZ:
 780             tlsauthz = optarg;
 781             break;
 782         case QEMU_NBD_OPT_FORK:
 783             fork_process = true;
 784             break;
 785         case 'L':
 786             list = true;
 787             break;
 788         case QEMU_NBD_OPT_PID_FILE:
 789             pid_file_name = optarg;
 790             break;
 791         }
 792     }
 793
 794     if (list) {
 795         if (argc != optind) {
 796             error_report("List mode is incompatible with a file name");
 797             exit(EXIT_FAILURE);
 798         }
 799         if (export_name || export_description || dev_offset ||
 800             device || disconnect || fmt || sn_id_or_name || bitmap ||
 801             seen_aio || seen_discard || seen_cache) {
 802             error_report("List mode is incompatible with per-device settings");
 803             exit(EXIT_FAILURE);
 804         }
 805         if (fork_process) {
 806             error_report("List mode is incompatible with forking");
 807             exit(EXIT_FAILURE);
 808         }
 809     } else if ((argc - optind) != 1) {
 810         error_report("Invalid number of arguments");
 811         error_printf("Try `%s --help' for more information.\n", argv[0]);
 812         exit(EXIT_FAILURE);
 813     } else if (!export_name) {
 814         export_name = "";
 815     }
 816
 817     qemu_opts_foreach(&qemu_object_opts,
 818                       user_creatable_add_opts_foreach,
 819                       qemu_nbd_object_print_help, &error_fatal);
 820
 821     if (!trace_init_backends()) {
 822         exit(1);
 823     }
 824     trace_init_file(trace_file);
 825     qemu_set_log(LOG_TRACE);
 826
 827     socket_activation = check_socket_activation();
 828     if (socket_activation == 0) {
 829         setup_address_and_port(&bindto, &port);
 830     } else {
 831         /* Using socket activation - check user didn't use -p etc. */
 832         const char *err_msg = socket_activation_validate_opts(device, sockpath,
 833                                                               bindto, port,
 834                                                               list);
 835         if (err_msg != NULL) {
 836             error_report("%s", err_msg);
 837             exit(EXIT_FAILURE);
 838         }
 839
 840         /* qemu-nbd can only listen on a single socket.  */
 841         if (socket_activation > 1) {
 842             error_report("qemu-nbd does not support socket activation with %s > 1",
 843                          "LISTEN_FDS");
 844             exit(EXIT_FAILURE);
 845         }
 846     }
 847
 848     if (tlscredsid) {
 849         if (sockpath) {
 850             error_report("TLS is only supported with IPv4/IPv6");
 851             exit(EXIT_FAILURE);
 852         }
 853         if (device) {
 854             error_report("TLS is not supported with a host device");
 855             exit(EXIT_FAILURE);
 856         }
 857         if (tlsauthz && list) {
 858             error_report("TLS authorization is incompatible with export list");
 859             exit(EXIT_FAILURE);
 860         }
 861         tlscreds = nbd_get_tls_creds(tlscredsid, list, &local_err);
 862         if (local_err) {
 863             error_reportf_err(local_err, "Failed to get TLS creds: ");
 864             exit(EXIT_FAILURE);
 865         }
 866     } else {
 867         if (tlsauthz) {
 868             error_report("--tls-authz is not permitted without --tls-creds");
 869             exit(EXIT_FAILURE);
 870         }
 871     }
 872
 873     if (list) {
 874         saddr = nbd_build_socket_address(sockpath, bindto, port);
 875         return qemu_nbd_client_list(saddr, tlscreds, bindto);
 876     }
 877
 878 #if !HAVE_NBD_DEVICE
 879     if (disconnect || device) {
 880         error_report("Kernel /dev/nbdN support not available");
 881         exit(EXIT_FAILURE);
 882     }
 883 #else /* HAVE_NBD_DEVICE */
 884     if (disconnect) {
 885         int nbdfd = open(argv[optind], O_RDWR);
 886         if (nbdfd < 0) {
 887             error_report("Cannot open %s: %s", argv[optind],
 888                          strerror(errno));
 889             exit(EXIT_FAILURE);
 890         }
 891         nbd_disconnect(nbdfd);
 892
 893         close(nbdfd);
 894
 895         printf("%s disconnected\n", argv[optind]);
 896
 897         return 0;
 898     }
 899 #endif
 900
 901     if ((device && !verbose) || fork_process) {
 902 #ifndef WIN32
 903         int stderr_fd[2];
 904         pid_t pid;
 905         int ret;
 906
 907         if (qemu_pipe(stderr_fd) < 0) {
 908             error_report("Error setting up communication pipe: %s",
 909                          strerror(errno));
 910             exit(EXIT_FAILURE);
 911         }
 912
 913         /* Now daemonize, but keep a communication channel open to
 914          * print errors and exit with the proper status code.
 915          */
 916         pid = fork();
 917         if (pid < 0) {
 918             error_report("Failed to fork: %s", strerror(errno));
 919             exit(EXIT_FAILURE);
 920         } else if (pid == 0) {
 921             close(stderr_fd[0]);
 922
 923             /* Remember parent's stderr if we will be restoring it. */
 924             if (fork_process) {
 925                 old_stderr = dup(STDERR_FILENO);
 926             }
 927
 928             ret = qemu_daemon(1, 0);
 929
 930             /* Temporarily redirect stderr to the parent's pipe...  */
 931             dup2(stderr_fd[1], STDERR_FILENO);
 932             if (ret < 0) {
 933                 error_report("Failed to daemonize: %s", strerror(errno));
 934                 exit(EXIT_FAILURE);
 935             }
 936
 937             /* ... close the descriptor we inherited and go on.  */
 938             close(stderr_fd[1]);
 939         } else {
 940             bool errors = false;
 941             char *buf;
 942
 943             /* In the parent.  Print error messages from the child until
 944              * it closes the pipe.
 945              */
 946             close(stderr_fd[1]);
 947             buf = g_malloc(1024);
 948             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
 949                 errors = true;
 950                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
 951                 if (ret < 0) {
 952                     exit(EXIT_FAILURE);
 953                 }
 954             }
 955             if (ret < 0) {
 956                 error_report("Cannot read from daemon: %s",
 957                              strerror(errno));
 958                 exit(EXIT_FAILURE);
 959             }
 960
 961             /* Usually the daemon should not print any message.
 962              * Exit with zero status in that case.
 963              */
 964             exit(errors);
 965         }
 966 #else /* WIN32 */
 967         error_report("Unable to fork into background on Windows hosts");
 968         exit(EXIT_FAILURE);
 969 #endif /* WIN32 */
 970     }
 971
 972     if (device != NULL && sockpath == NULL) {
 973         sockpath = g_malloc(128);
 974         snprintf(sockpath, 128, SOCKET_PATH, basename(device));
 975     }
 976
 977     server = qio_net_listener_new();
 978     if (socket_activation == 0) {
 979         saddr = nbd_build_socket_address(sockpath, bindto, port);
 980         if (qio_net_listener_open_sync(server, saddr, 1, &local_err) < 0) {
 981             object_unref(OBJECT(server));
 982             error_report_err(local_err);
 983             exit(EXIT_FAILURE);
 984         }
 985     } else {
 986         size_t i;
 987         /* See comment in check_socket_activation above. */
 988         for (i = 0; i < socket_activation; i++) {
 989             QIOChannelSocket *sioc;
 990             sioc = qio_channel_socket_new_fd(FIRST_SOCKET_ACTIVATION_FD + i,
 991                                              &local_err);
 992             if (sioc == NULL) {
 993                 object_unref(OBJECT(server));
 994                 error_reportf_err(local_err,
 995                                   "Failed to use socket activation: ");
 996                 exit(EXIT_FAILURE);
 997             }
 998             qio_net_listener_add(server, sioc);
 999             object_unref(OBJECT(sioc));
1000         }
1001     }
1002
1003     if (qemu_init_main_loop(&local_err)) {
1004         error_report_err(local_err);
1005         exit(EXIT_FAILURE);
1006     }
1007     bdrv_init();
1008     atexit(qemu_nbd_shutdown);
1009
1010     srcpath = argv[optind];
1011     if (imageOpts) {
1012         QemuOpts *opts;
1013         if (fmt) {
1014             error_report("--image-opts and -f are mutually exclusive");
1015             exit(EXIT_FAILURE);
1016         }
1017         opts = qemu_opts_parse_noisily(&file_opts, srcpath, true);
1018         if (!opts) {
1019             qemu_opts_reset(&file_opts);
1020             exit(EXIT_FAILURE);
1021         }
1022         options = qemu_opts_to_qdict(opts, NULL);
1023         qemu_opts_reset(&file_opts);
1024         blk = blk_new_open(NULL, NULL, options, flags, &local_err);
1025     } else {
1026         if (fmt) {
1027             options = qdict_new();
1028             qdict_put_str(options, "driver", fmt);
1029         }
1030         blk = blk_new_open(srcpath, NULL, options, flags, &local_err);
1031     }
1032
1033     if (!blk) {
1034         error_reportf_err(local_err, "Failed to blk_new_open '%s': ",
1035                           argv[optind]);
1036         exit(EXIT_FAILURE);
1037     }
1038     bs = blk_bs(blk);
1039
1040     blk_set_enable_write_cache(blk, !writethrough);
1041
1042     if (sn_opts) {
1043         ret = bdrv_snapshot_load_tmp(bs,
1044                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
1045                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
1046                                      &local_err);
1047     } else if (sn_id_or_name) {
1048         ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
1049                                                    &local_err);
1050     }
1051     if (ret < 0) {
1052         error_reportf_err(local_err, "Failed to load snapshot: ");
1053         exit(EXIT_FAILURE);
1054     }
1055
1056     bs->detect_zeroes = detect_zeroes;
1057     fd_size = blk_getlength(blk);
1058     if (fd_size < 0) {
1059         error_report("Failed to determine the image length: %s",
1060                      strerror(-fd_size));
1061         exit(EXIT_FAILURE);
1062     }
1063
1064     if (dev_offset >= fd_size) {
1065         error_report("Offset (%" PRIu64 ") has to be smaller than the image "
1066                      "size (%" PRId64 ")", dev_offset, fd_size);
1067         exit(EXIT_FAILURE);
1068     }
1069     fd_size -= dev_offset;
1070
1071     export = nbd_export_new(bs, dev_offset, fd_size, export_name,
1072                             export_description, bitmap, readonly, shared > 1,
1073                             nbd_export_closed, writethrough, NULL,
1074                             &error_fatal);
1075
1076     if (device) {
1077 #if HAVE_NBD_DEVICE
1078         int ret;
1079
1080         ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
1081         if (ret != 0) {
1082             error_report("Failed to create client thread: %s", strerror(ret));
1083             exit(EXIT_FAILURE);
1084         }
1085 #endif
1086     } else {
1087         /* Shut up GCC warnings.  */
1088         memset(&client_thread, 0, sizeof(client_thread));
1089     }
1090
1091     nbd_update_server_watch();
1092
1093     if (pid_file_name) {
1094         qemu_write_pidfile(pid_file_name, &error_fatal);
1095     }
1096
1097     /* now when the initialization is (almost) complete, chdir("/")
1098      * to free any busy filesystems */
1099     if (chdir("/") < 0) {
1100         error_report("Could not chdir to root directory: %s",
1101                      strerror(errno));
1102         exit(EXIT_FAILURE);
1103     }
1104
1105     if (fork_process) {
1106         dup2(old_stderr, STDERR_FILENO);
1107         close(old_stderr);
1108     }
1109
1110     state = RUNNING;
1111     do {
1112         main_loop_wait(false);
1113         if (state == TERMINATE) {
1114             state = TERMINATING;
1115             nbd_export_close(export);
1116             nbd_export_put(export);
1117             export = NULL;
1118         }
1119     } while (state != TERMINATED);
1120
1121     blk_unref(blk);
1122     if (sockpath) {
1123         unlink(sockpath);
1124     }
1125
1126     qemu_opts_del(sn_opts);
1127
1128     if (device) {
1129         void *ret;
1130         pthread_join(client_thread, &ret);
1131         exit(ret != NULL);
1132     } else {
1133         exit(EXIT_SUCCESS);
1134     }
1135 }