qemu/qemu-nbd.c
<<
>>
Prefs
   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 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
  76static int verbose;
  77static char *srcpath;
  78static SocketAddress *saddr;
  79static int persistent = 0;
  80static enum { RUNNING, TERMINATE, TERMINATED } state;
  81static int shared = 1;
  82static int nb_fds;
  83static QIONetListener *server;
  84static QCryptoTLSCreds *tlscreds;
  85static const char *tlsauthz;
  86
  87static 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"
 155QEMU_HELP_BOTTOM "\n"
 156    , name, name, NBD_DEFAULT_PORT, "DEVICE");
 157}
 158
 159static void version(const char *name)
 160{
 161    printf(
 162"%s " QEMU_FULL_VERSION "\n"
 163"Written by Anthony Liguori.\n"
 164"\n"
 165QEMU_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 */
 176void 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
 183static 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
 258static 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
 275static void *nbd_client_thread(void *arg)
 276{
 277    char *device = arg;
 278    NBDExportInfo info = { .request_sizes = false, .name = g_strdup("") };
 279    QIOChannelSocket *sioc;
 280    int fd = -1;
 281    int ret = EXIT_FAILURE;
 282    pthread_t show_parts_thread;
 283    Error *local_error = NULL;
 284
 285    sioc = qio_channel_socket_new();
 286    if (qio_channel_socket_connect_sync(sioc,
 287                                        saddr,
 288                                        &local_error) < 0) {
 289        error_report_err(local_error);
 290        goto out;
 291    }
 292
 293    if (nbd_receive_negotiate(NULL, QIO_CHANNEL(sioc),
 294                              NULL, NULL, NULL, &info, &local_error) < 0) {
 295        if (local_error) {
 296            error_report_err(local_error);
 297        }
 298        goto out;
 299    }
 300
 301    fd = open(device, O_RDWR);
 302    if (fd < 0) {
 303        /* Linux-only, we can use %m in printf.  */
 304        error_report("Failed to open %s: %m", device);
 305        goto out;
 306    }
 307
 308    if (nbd_init(fd, sioc, &info, &local_error) < 0) {
 309        error_report_err(local_error);
 310        goto out;
 311    }
 312
 313    /* update partition table */
 314    pthread_create(&show_parts_thread, NULL, show_parts, device);
 315
 316    if (verbose) {
 317        fprintf(stderr, "NBD device %s is now connected to %s\n",
 318                device, srcpath);
 319    } else {
 320        /* Close stderr so that the qemu-nbd process exits.  */
 321        dup2(STDOUT_FILENO, STDERR_FILENO);
 322    }
 323
 324    if (nbd_client(fd) < 0) {
 325        goto out;
 326    }
 327
 328    ret = EXIT_SUCCESS;
 329
 330 out:
 331    if (fd >= 0) {
 332        close(fd);
 333    }
 334    object_unref(OBJECT(sioc));
 335    g_free(info.name);
 336    kill(getpid(), SIGTERM);
 337    return (void *) (intptr_t) ret;
 338}
 339#endif /* HAVE_NBD_DEVICE */
 340
 341static int nbd_can_accept(void)
 342{
 343    return state == RUNNING && (shared == 0 || nb_fds < shared);
 344}
 345
 346static void nbd_update_server_watch(void);
 347
 348static void nbd_client_closed(NBDClient *client, bool negotiated)
 349{
 350    nb_fds--;
 351    if (negotiated && nb_fds == 0 && !persistent && state == RUNNING) {
 352        state = TERMINATE;
 353    }
 354    nbd_update_server_watch();
 355    nbd_client_put(client);
 356}
 357
 358static void nbd_accept(QIONetListener *listener, QIOChannelSocket *cioc,
 359                       gpointer opaque)
 360{
 361    if (state >= TERMINATE) {
 362        return;
 363    }
 364
 365    nb_fds++;
 366    nbd_update_server_watch();
 367    nbd_client_new(cioc, tlscreds, tlsauthz, nbd_client_closed);
 368}
 369
 370static void nbd_update_server_watch(void)
 371{
 372    if (nbd_can_accept()) {
 373        qio_net_listener_set_client_func(server, nbd_accept, NULL, NULL);
 374    } else {
 375        qio_net_listener_set_client_func(server, NULL, NULL, NULL);
 376    }
 377}
 378
 379
 380static SocketAddress *nbd_build_socket_address(const char *sockpath,
 381                                               const char *bindto,
 382                                               const char *port)
 383{
 384    SocketAddress *saddr;
 385
 386    saddr = g_new0(SocketAddress, 1);
 387    if (sockpath) {
 388        saddr->type = SOCKET_ADDRESS_TYPE_UNIX;
 389        saddr->u.q_unix.path = g_strdup(sockpath);
 390    } else {
 391        InetSocketAddress *inet;
 392        saddr->type = SOCKET_ADDRESS_TYPE_INET;
 393        inet = &saddr->u.inet;
 394        inet->host = g_strdup(bindto);
 395        if (port) {
 396            inet->port = g_strdup(port);
 397        } else  {
 398            inet->port = g_strdup_printf("%d", NBD_DEFAULT_PORT);
 399        }
 400    }
 401
 402    return saddr;
 403}
 404
 405
 406static QemuOptsList file_opts = {
 407    .name = "file",
 408    .implied_opt_name = "file",
 409    .head = QTAILQ_HEAD_INITIALIZER(file_opts.head),
 410    .desc = {
 411        /* no elements => accept any params */
 412        { /* end of list */ }
 413    },
 414};
 415
 416static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, bool list,
 417                                          Error **errp)
 418{
 419    Object *obj;
 420    QCryptoTLSCreds *creds;
 421
 422    obj = object_resolve_path_component(
 423        object_get_objects_root(), id);
 424    if (!obj) {
 425        error_setg(errp, "No TLS credentials with id '%s'",
 426                   id);
 427        return NULL;
 428    }
 429    creds = (QCryptoTLSCreds *)
 430        object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
 431    if (!creds) {
 432        error_setg(errp, "Object with id '%s' is not TLS credentials",
 433                   id);
 434        return NULL;
 435    }
 436
 437    if (!qcrypto_tls_creds_check_endpoint(creds,
 438                                          list
 439                                          ? QCRYPTO_TLS_CREDS_ENDPOINT_CLIENT
 440                                          : QCRYPTO_TLS_CREDS_ENDPOINT_SERVER,
 441                                          errp)) {
 442        return NULL;
 443    }
 444    object_ref(obj);
 445    return creds;
 446}
 447
 448static void setup_address_and_port(const char **address, const char **port)
 449{
 450    if (*address == NULL) {
 451        *address = "0.0.0.0";
 452    }
 453
 454    if (*port == NULL) {
 455        *port = stringify(NBD_DEFAULT_PORT);
 456    }
 457}
 458
 459/*
 460 * Check socket parameters compatibility when socket activation is used.
 461 */
 462static const char *socket_activation_validate_opts(const char *device,
 463                                                   const char *sockpath,
 464                                                   const char *address,
 465                                                   const char *port,
 466                                                   const char *selinux,
 467                                                   bool list)
 468{
 469    if (device != NULL) {
 470        return "NBD device can't be set when using socket activation";
 471    }
 472
 473    if (sockpath != NULL) {
 474        return "Unix socket can't be set when using socket activation";
 475    }
 476
 477    if (address != NULL) {
 478        return "The interface can't be set when using socket activation";
 479    }
 480
 481    if (port != NULL) {
 482        return "TCP port number can't be set when using socket activation";
 483    }
 484
 485    if (selinux != NULL) {
 486        return "SELinux label can't be set when using socket activation";
 487    }
 488
 489    if (list) {
 490        return "List mode is incompatible with socket activation";
 491    }
 492
 493    return NULL;
 494}
 495
 496static void qemu_nbd_shutdown(void)
 497{
 498    job_cancel_sync_all();
 499    blk_exp_close_all();
 500    bdrv_close_all();
 501}
 502
 503int main(int argc, char **argv)
 504{
 505    BlockBackend *blk;
 506    BlockDriverState *bs;
 507    uint64_t dev_offset = 0;
 508    bool readonly = false;
 509    bool disconnect = false;
 510    const char *bindto = NULL;
 511    const char *port = NULL;
 512    char *sockpath = NULL;
 513    char *device = NULL;
 514    QemuOpts *sn_opts = NULL;
 515    const char *sn_id_or_name = NULL;
 516    const char *sopt = "hVb:o:p:rsnc:dvk:e:f:tl:x:T:D:AB:L";
 517    struct option lopt[] = {
 518        { "help", no_argument, NULL, 'h' },
 519        { "version", no_argument, NULL, 'V' },
 520        { "bind", required_argument, NULL, 'b' },
 521        { "port", required_argument, NULL, 'p' },
 522        { "socket", required_argument, NULL, 'k' },
 523        { "offset", required_argument, NULL, 'o' },
 524        { "read-only", no_argument, NULL, 'r' },
 525        { "allocation-depth", no_argument, NULL, 'A' },
 526        { "bitmap", required_argument, NULL, 'B' },
 527        { "connect", required_argument, NULL, 'c' },
 528        { "disconnect", no_argument, NULL, 'd' },
 529        { "list", no_argument, NULL, 'L' },
 530        { "snapshot", no_argument, NULL, 's' },
 531        { "load-snapshot", required_argument, NULL, 'l' },
 532        { "nocache", no_argument, NULL, 'n' },
 533        { "cache", required_argument, NULL, QEMU_NBD_OPT_CACHE },
 534        { "aio", required_argument, NULL, QEMU_NBD_OPT_AIO },
 535        { "discard", required_argument, NULL, QEMU_NBD_OPT_DISCARD },
 536        { "detect-zeroes", required_argument, NULL,
 537          QEMU_NBD_OPT_DETECT_ZEROES },
 538        { "shared", required_argument, NULL, 'e' },
 539        { "format", required_argument, NULL, 'f' },
 540        { "persistent", no_argument, NULL, 't' },
 541        { "verbose", no_argument, NULL, 'v' },
 542        { "object", required_argument, NULL, QEMU_NBD_OPT_OBJECT },
 543        { "export-name", required_argument, NULL, 'x' },
 544        { "description", required_argument, NULL, 'D' },
 545        { "tls-creds", required_argument, NULL, QEMU_NBD_OPT_TLSCREDS },
 546        { "tls-hostname", required_argument, NULL, QEMU_NBD_OPT_TLSHOSTNAME },
 547        { "tls-authz", required_argument, NULL, QEMU_NBD_OPT_TLSAUTHZ },
 548        { "image-opts", no_argument, NULL, QEMU_NBD_OPT_IMAGE_OPTS },
 549        { "trace", required_argument, NULL, 'T' },
 550        { "fork", no_argument, NULL, QEMU_NBD_OPT_FORK },
 551        { "pid-file", required_argument, NULL, QEMU_NBD_OPT_PID_FILE },
 552        { "selinux-label", required_argument, NULL,
 553          QEMU_NBD_OPT_SELINUX_LABEL },
 554        { NULL, 0, NULL, 0 }
 555    };
 556    int ch;
 557    int opt_ind = 0;
 558    int flags = BDRV_O_RDWR;
 559    int ret = 0;
 560    bool seen_cache = false;
 561    bool seen_discard = false;
 562    bool seen_aio = false;
 563    pthread_t client_thread;
 564    const char *fmt = NULL;
 565    Error *local_err = NULL;
 566    BlockdevDetectZeroesOptions detect_zeroes = BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
 567    QDict *options = NULL;
 568    const char *export_name = NULL; /* defaults to "" later for server mode */
 569    const char *export_description = NULL;
 570    strList *bitmaps = NULL;
 571    bool alloc_depth = false;
 572    const char *tlscredsid = NULL;
 573    const char *tlshostname = NULL;
 574    bool imageOpts = false;
 575    bool writethrough = false; /* Client will flush as needed. */
 576    bool fork_process = false;
 577    bool list = false;
 578    int old_stderr = -1;
 579    unsigned socket_activation;
 580    const char *pid_file_name = NULL;
 581    const char *selinux_label = NULL;
 582    BlockExportOptions *export_opts;
 583
 584#ifdef CONFIG_POSIX
 585    os_setup_early_signal_handling();
 586    os_setup_signal_handling();
 587#endif
 588
 589    socket_init();
 590    error_init(argv[0]);
 591    module_call_init(MODULE_INIT_TRACE);
 592    qcrypto_init(&error_fatal);
 593
 594    module_call_init(MODULE_INIT_QOM);
 595    qemu_add_opts(&qemu_trace_opts);
 596    qemu_init_exec_dir(argv[0]);
 597
 598    while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 599        switch (ch) {
 600        case 's':
 601            flags |= BDRV_O_SNAPSHOT;
 602            break;
 603        case 'n':
 604            optarg = (char *) "none";
 605            /* fallthrough */
 606        case QEMU_NBD_OPT_CACHE:
 607            if (seen_cache) {
 608                error_report("-n and --cache can only be specified once");
 609                exit(EXIT_FAILURE);
 610            }
 611            seen_cache = true;
 612            if (bdrv_parse_cache_mode(optarg, &flags, &writethrough) == -1) {
 613                error_report("Invalid cache mode `%s'", optarg);
 614                exit(EXIT_FAILURE);
 615            }
 616            break;
 617        case QEMU_NBD_OPT_AIO:
 618            if (seen_aio) {
 619                error_report("--aio can only be specified once");
 620                exit(EXIT_FAILURE);
 621            }
 622            seen_aio = true;
 623            if (bdrv_parse_aio(optarg, &flags) < 0) {
 624                error_report("Invalid aio mode '%s'", optarg);
 625                exit(EXIT_FAILURE);
 626            }
 627            break;
 628        case QEMU_NBD_OPT_DISCARD:
 629            if (seen_discard) {
 630                error_report("--discard can only be specified once");
 631                exit(EXIT_FAILURE);
 632            }
 633            seen_discard = true;
 634            if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
 635                error_report("Invalid discard mode `%s'", optarg);
 636                exit(EXIT_FAILURE);
 637            }
 638            break;
 639        case QEMU_NBD_OPT_DETECT_ZEROES:
 640            detect_zeroes =
 641                qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
 642                                optarg,
 643                                BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
 644                                &local_err);
 645            if (local_err) {
 646                error_reportf_err(local_err,
 647                                  "Failed to parse detect_zeroes mode: ");
 648                exit(EXIT_FAILURE);
 649            }
 650            if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
 651                !(flags & BDRV_O_UNMAP)) {
 652                error_report("setting detect-zeroes to unmap is not allowed "
 653                             "without setting discard operation to unmap");
 654                exit(EXIT_FAILURE);
 655            }
 656            break;
 657        case 'b':
 658            bindto = optarg;
 659            break;
 660        case 'p':
 661            port = optarg;
 662            break;
 663        case 'o':
 664            if (qemu_strtou64(optarg, NULL, 0, &dev_offset) < 0) {
 665                error_report("Invalid offset '%s'", optarg);
 666                exit(EXIT_FAILURE);
 667            }
 668            break;
 669        case 'l':
 670            if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
 671                sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
 672                                                  optarg, false);
 673                if (!sn_opts) {
 674                    error_report("Failed in parsing snapshot param `%s'",
 675                                 optarg);
 676                    exit(EXIT_FAILURE);
 677                }
 678            } else {
 679                sn_id_or_name = optarg;
 680            }
 681            /* fall through */
 682        case 'r':
 683            readonly = true;
 684            flags &= ~BDRV_O_RDWR;
 685            break;
 686        case 'A':
 687            alloc_depth = true;
 688            break;
 689        case 'B':
 690            QAPI_LIST_PREPEND(bitmaps, g_strdup(optarg));
 691            break;
 692        case 'k':
 693            sockpath = optarg;
 694            if (sockpath[0] != '/') {
 695                error_report("socket path must be absolute");
 696                exit(EXIT_FAILURE);
 697            }
 698            break;
 699        case 'd':
 700            disconnect = true;
 701            break;
 702        case 'c':
 703            device = optarg;
 704            break;
 705        case 'e':
 706            if (qemu_strtoi(optarg, NULL, 0, &shared) < 0 ||
 707                shared < 0) {
 708                error_report("Invalid shared device number '%s'", optarg);
 709                exit(EXIT_FAILURE);
 710            }
 711            break;
 712        case 'f':
 713            fmt = optarg;
 714            break;
 715        case 't':
 716            persistent = 1;
 717            break;
 718        case 'x':
 719            export_name = optarg;
 720            if (strlen(export_name) > NBD_MAX_STRING_SIZE) {
 721                error_report("export name '%s' too long", export_name);
 722                exit(EXIT_FAILURE);
 723            }
 724            break;
 725        case 'D':
 726            export_description = optarg;
 727            if (strlen(export_description) > NBD_MAX_STRING_SIZE) {
 728                error_report("export description '%s' too long",
 729                             export_description);
 730                exit(EXIT_FAILURE);
 731            }
 732            break;
 733        case 'v':
 734            verbose = 1;
 735            break;
 736        case 'V':
 737            version(argv[0]);
 738            exit(0);
 739            break;
 740        case 'h':
 741            usage(argv[0]);
 742            exit(0);
 743            break;
 744        case '?':
 745            error_report("Try `%s --help' for more information.", argv[0]);
 746            exit(EXIT_FAILURE);
 747        case QEMU_NBD_OPT_OBJECT:
 748            user_creatable_process_cmdline(optarg);
 749            break;
 750        case QEMU_NBD_OPT_TLSCREDS:
 751            tlscredsid = optarg;
 752            break;
 753        case QEMU_NBD_OPT_TLSHOSTNAME:
 754            tlshostname = optarg;
 755            break;
 756        case QEMU_NBD_OPT_IMAGE_OPTS:
 757            imageOpts = true;
 758            break;
 759        case 'T':
 760            trace_opt_parse(optarg);
 761            break;
 762        case QEMU_NBD_OPT_TLSAUTHZ:
 763            tlsauthz = optarg;
 764            break;
 765        case QEMU_NBD_OPT_FORK:
 766            fork_process = true;
 767            break;
 768        case 'L':
 769            list = true;
 770            break;
 771        case QEMU_NBD_OPT_PID_FILE:
 772            pid_file_name = optarg;
 773            break;
 774        case QEMU_NBD_OPT_SELINUX_LABEL:
 775            selinux_label = optarg;
 776            break;
 777        }
 778    }
 779
 780    if (list) {
 781        if (argc != optind) {
 782            error_report("List mode is incompatible with a file name");
 783            exit(EXIT_FAILURE);
 784        }
 785        if (export_name || export_description || dev_offset ||
 786            device || disconnect || fmt || sn_id_or_name || bitmaps ||
 787            alloc_depth || seen_aio || seen_discard || seen_cache) {
 788            error_report("List mode is incompatible with per-device settings");
 789            exit(EXIT_FAILURE);
 790        }
 791        if (fork_process) {
 792            error_report("List mode is incompatible with forking");
 793            exit(EXIT_FAILURE);
 794        }
 795    } else if ((argc - optind) != 1) {
 796        error_report("Invalid number of arguments");
 797        error_printf("Try `%s --help' for more information.\n", argv[0]);
 798        exit(EXIT_FAILURE);
 799    } else if (!export_name) {
 800        export_name = "";
 801    }
 802
 803    if (!trace_init_backends()) {
 804        exit(1);
 805    }
 806    trace_init_file();
 807    qemu_set_log(LOG_TRACE);
 808
 809    socket_activation = check_socket_activation();
 810    if (socket_activation == 0) {
 811        if (!sockpath) {
 812            setup_address_and_port(&bindto, &port);
 813        }
 814    } else {
 815        /* Using socket activation - check user didn't use -p etc. */
 816        const char *err_msg = socket_activation_validate_opts(device, sockpath,
 817                                                              bindto, port,
 818                                                              selinux_label,
 819                                                              list);
 820        if (err_msg != NULL) {
 821            error_report("%s", err_msg);
 822            exit(EXIT_FAILURE);
 823        }
 824
 825        /* qemu-nbd can only listen on a single socket.  */
 826        if (socket_activation > 1) {
 827            error_report("qemu-nbd does not support socket activation with %s > 1",
 828                         "LISTEN_FDS");
 829            exit(EXIT_FAILURE);
 830        }
 831    }
 832
 833    if (tlscredsid) {
 834        if (device) {
 835            error_report("TLS is not supported with a host device");
 836            exit(EXIT_FAILURE);
 837        }
 838        if (tlsauthz && list) {
 839            error_report("TLS authorization is incompatible with export list");
 840            exit(EXIT_FAILURE);
 841        }
 842        if (tlshostname && !list) {
 843            error_report("TLS hostname is only supported with export list");
 844            exit(EXIT_FAILURE);
 845        }
 846        tlscreds = nbd_get_tls_creds(tlscredsid, list, &local_err);
 847        if (local_err) {
 848            error_reportf_err(local_err, "Failed to get TLS creds: ");
 849            exit(EXIT_FAILURE);
 850        }
 851    } else {
 852        if (tlsauthz) {
 853            error_report("--tls-authz is not permitted without --tls-creds");
 854            exit(EXIT_FAILURE);
 855        }
 856        if (tlshostname) {
 857            error_report("--tls-hostname is not permitted without --tls-creds");
 858            exit(EXIT_FAILURE);
 859        }
 860    }
 861
 862    if (selinux_label) {
 863#ifdef CONFIG_SELINUX
 864        if (sockpath == NULL && device == NULL) {
 865            error_report("--selinux-label is not permitted without --socket");
 866            exit(EXIT_FAILURE);
 867        }
 868#else
 869        error_report("SELinux support not enabled in this binary");
 870        exit(EXIT_FAILURE);
 871#endif
 872    }
 873
 874    if (list) {
 875        saddr = nbd_build_socket_address(sockpath, bindto, port);
 876        return qemu_nbd_client_list(saddr, tlscreds,
 877                                    tlshostname ? tlshostname : bindto);
 878    }
 879
 880#if !HAVE_NBD_DEVICE
 881    if (disconnect || device) {
 882        error_report("Kernel /dev/nbdN support not available");
 883        exit(EXIT_FAILURE);
 884    }
 885#else /* HAVE_NBD_DEVICE */
 886    if (disconnect) {
 887        int nbdfd = open(argv[optind], O_RDWR);
 888        if (nbdfd < 0) {
 889            error_report("Cannot open %s: %s", argv[optind],
 890                         strerror(errno));
 891            exit(EXIT_FAILURE);
 892        }
 893        nbd_disconnect(nbdfd);
 894
 895        close(nbdfd);
 896
 897        printf("%s disconnected\n", argv[optind]);
 898
 899        return 0;
 900    }
 901#endif
 902
 903    if ((device && !verbose) || fork_process) {
 904#ifndef WIN32
 905        int stderr_fd[2];
 906        pid_t pid;
 907        int ret;
 908
 909        if (qemu_pipe(stderr_fd) < 0) {
 910            error_report("Error setting up communication pipe: %s",
 911                         strerror(errno));
 912            exit(EXIT_FAILURE);
 913        }
 914
 915        /* Now daemonize, but keep a communication channel open to
 916         * print errors and exit with the proper status code.
 917         */
 918        pid = fork();
 919        if (pid < 0) {
 920            error_report("Failed to fork: %s", strerror(errno));
 921            exit(EXIT_FAILURE);
 922        } else if (pid == 0) {
 923            close(stderr_fd[0]);
 924
 925            /* Remember parent's stderr if we will be restoring it. */
 926            if (fork_process) {
 927                old_stderr = dup(STDERR_FILENO);
 928            }
 929
 930            ret = qemu_daemon(1, 0);
 931
 932            /* Temporarily redirect stderr to the parent's pipe...  */
 933            dup2(stderr_fd[1], STDERR_FILENO);
 934            if (ret < 0) {
 935                error_report("Failed to daemonize: %s", strerror(errno));
 936                exit(EXIT_FAILURE);
 937            }
 938
 939            /* ... close the descriptor we inherited and go on.  */
 940            close(stderr_fd[1]);
 941        } else {
 942            bool errors = false;
 943            char *buf;
 944
 945            /* In the parent.  Print error messages from the child until
 946             * it closes the pipe.
 947             */
 948            close(stderr_fd[1]);
 949            buf = g_malloc(1024);
 950            while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
 951                errors = true;
 952                ret = qemu_write_full(STDERR_FILENO, buf, ret);
 953                if (ret < 0) {
 954                    exit(EXIT_FAILURE);
 955                }
 956            }
 957            if (ret < 0) {
 958                error_report("Cannot read from daemon: %s",
 959                             strerror(errno));
 960                exit(EXIT_FAILURE);
 961            }
 962
 963            /* Usually the daemon should not print any message.
 964             * Exit with zero status in that case.
 965             */
 966            exit(errors);
 967        }
 968#else /* WIN32 */
 969        error_report("Unable to fork into background on Windows hosts");
 970        exit(EXIT_FAILURE);
 971#endif /* WIN32 */
 972    }
 973
 974    if (device != NULL && sockpath == NULL) {
 975        sockpath = g_malloc(128);
 976        snprintf(sockpath, 128, SOCKET_PATH, basename(device));
 977    }
 978
 979    server = qio_net_listener_new();
 980    if (socket_activation == 0) {
 981        int backlog;
 982
 983        if (persistent || shared == 0) {
 984            backlog = SOMAXCONN;
 985        } else {
 986            backlog = MIN(shared, SOMAXCONN);
 987        }
 988#ifdef CONFIG_SELINUX
 989        if (selinux_label && setsockcreatecon_raw(selinux_label) == -1) {
 990            error_report("Cannot set SELinux socket create context to %s: %s",
 991                         selinux_label, strerror(errno));
 992            exit(EXIT_FAILURE);
 993        }
 994#endif
 995        saddr = nbd_build_socket_address(sockpath, bindto, port);
 996        if (qio_net_listener_open_sync(server, saddr, backlog,
 997                                       &local_err) < 0) {
 998            object_unref(OBJECT(server));
 999            error_report_err(local_err);
1000            exit(EXIT_FAILURE);
1001        }
1002#ifdef CONFIG_SELINUX
1003        if (selinux_label && setsockcreatecon_raw(NULL) == -1) {
1004            error_report("Cannot clear SELinux socket create context: %s",
1005                         strerror(errno));
1006            exit(EXIT_FAILURE);
1007        }
1008#endif
1009    } else {
1010        size_t i;
1011        /* See comment in check_socket_activation above. */
1012        for (i = 0; i < socket_activation; i++) {
1013            QIOChannelSocket *sioc;
1014            sioc = qio_channel_socket_new_fd(FIRST_SOCKET_ACTIVATION_FD + i,
1015                                             &local_err);
1016            if (sioc == NULL) {
1017                object_unref(OBJECT(server));
1018                error_reportf_err(local_err,
1019                                  "Failed to use socket activation: ");
1020                exit(EXIT_FAILURE);
1021            }
1022            qio_net_listener_add(server, sioc);
1023            object_unref(OBJECT(sioc));
1024        }
1025    }
1026
1027    qemu_init_main_loop(&error_fatal);
1028    bdrv_init();
1029    atexit(qemu_nbd_shutdown);
1030
1031    srcpath = argv[optind];
1032    if (imageOpts) {
1033        QemuOpts *opts;
1034        if (fmt) {
1035            error_report("--image-opts and -f are mutually exclusive");
1036            exit(EXIT_FAILURE);
1037        }
1038        opts = qemu_opts_parse_noisily(&file_opts, srcpath, true);
1039        if (!opts) {
1040            qemu_opts_reset(&file_opts);
1041            exit(EXIT_FAILURE);
1042        }
1043        options = qemu_opts_to_qdict(opts, NULL);
1044        qemu_opts_reset(&file_opts);
1045        blk = blk_new_open(NULL, NULL, options, flags, &local_err);
1046    } else {
1047        if (fmt) {
1048            options = qdict_new();
1049            qdict_put_str(options, "driver", fmt);
1050        }
1051        blk = blk_new_open(srcpath, NULL, options, flags, &local_err);
1052    }
1053
1054    if (!blk) {
1055        error_reportf_err(local_err, "Failed to blk_new_open '%s': ",
1056                          argv[optind]);
1057        exit(EXIT_FAILURE);
1058    }
1059    bs = blk_bs(blk);
1060
1061    if (dev_offset) {
1062        QDict *raw_opts = qdict_new();
1063        qdict_put_str(raw_opts, "driver", "raw");
1064        qdict_put_str(raw_opts, "file", bs->node_name);
1065        qdict_put_int(raw_opts, "offset", dev_offset);
1066        bs = bdrv_open(NULL, NULL, raw_opts, flags, &error_fatal);
1067        blk_remove_bs(blk);
1068        blk_insert_bs(blk, bs, &error_fatal);
1069        bdrv_unref(bs);
1070    }
1071
1072    blk_set_enable_write_cache(blk, !writethrough);
1073
1074    if (sn_opts) {
1075        ret = bdrv_snapshot_load_tmp(bs,
1076                                     qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
1077                                     qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
1078                                     &local_err);
1079    } else if (sn_id_or_name) {
1080        ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
1081                                                   &local_err);
1082    }
1083    if (ret < 0) {
1084        error_reportf_err(local_err, "Failed to load snapshot: ");
1085        exit(EXIT_FAILURE);
1086    }
1087
1088    bs->detect_zeroes = detect_zeroes;
1089
1090    nbd_server_is_qemu_nbd(true);
1091
1092    export_opts = g_new(BlockExportOptions, 1);
1093    *export_opts = (BlockExportOptions) {
1094        .type               = BLOCK_EXPORT_TYPE_NBD,
1095        .id                 = g_strdup("qemu-nbd-export"),
1096        .node_name          = g_strdup(bdrv_get_node_name(bs)),
1097        .has_writethrough   = true,
1098        .writethrough       = writethrough,
1099        .has_writable       = true,
1100        .writable           = !readonly,
1101        .u.nbd = {
1102            .has_name             = true,
1103            .name                 = g_strdup(export_name),
1104            .has_description      = !!export_description,
1105            .description          = g_strdup(export_description),
1106            .has_bitmaps          = !!bitmaps,
1107            .bitmaps              = bitmaps,
1108            .has_allocation_depth = alloc_depth,
1109            .allocation_depth     = alloc_depth,
1110        },
1111    };
1112    blk_exp_add(export_opts, &error_fatal);
1113    qapi_free_BlockExportOptions(export_opts);
1114
1115    if (device) {
1116#if HAVE_NBD_DEVICE
1117        int ret;
1118
1119        ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
1120        if (ret != 0) {
1121            error_report("Failed to create client thread: %s", strerror(ret));
1122            exit(EXIT_FAILURE);
1123        }
1124#endif
1125    } else {
1126        /* Shut up GCC warnings.  */
1127        memset(&client_thread, 0, sizeof(client_thread));
1128    }
1129
1130    nbd_update_server_watch();
1131
1132    if (pid_file_name) {
1133        qemu_write_pidfile(pid_file_name, &error_fatal);
1134    }
1135
1136    /* now when the initialization is (almost) complete, chdir("/")
1137     * to free any busy filesystems */
1138    if (chdir("/") < 0) {
1139        error_report("Could not chdir to root directory: %s",
1140                     strerror(errno));
1141        exit(EXIT_FAILURE);
1142    }
1143
1144    if (fork_process) {
1145        dup2(old_stderr, STDERR_FILENO);
1146        close(old_stderr);
1147    }
1148
1149    state = RUNNING;
1150    do {
1151        main_loop_wait(false);
1152        if (state == TERMINATE) {
1153            blk_exp_close_all();
1154            state = TERMINATED;
1155        }
1156    } while (state != TERMINATED);
1157
1158    blk_unref(blk);
1159    if (sockpath) {
1160        unlink(sockpath);
1161    }
1162
1163    qemu_opts_del(sn_opts);
1164
1165    if (device) {
1166        void *ret;
1167        pthread_join(client_thread, &ret);
1168        exit(ret != NULL);
1169    } else {
1170        exit(EXIT_SUCCESS);
1171    }
1172}
1173