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 __linux__
  51#define HAVE_NBD_DEVICE 1
  52#else
  53#define HAVE_NBD_DEVICE 0
  54#endif
  55
  56#define SOCKET_PATH                "/var/lock/qemu-nbd-%s"
  57#define QEMU_NBD_OPT_CACHE         256
  58#define QEMU_NBD_OPT_AIO           257
  59#define QEMU_NBD_OPT_DISCARD       258
  60#define QEMU_NBD_OPT_DETECT_ZEROES 259
  61#define QEMU_NBD_OPT_OBJECT        260
  62#define QEMU_NBD_OPT_TLSCREDS      261
  63#define QEMU_NBD_OPT_IMAGE_OPTS    262
  64#define QEMU_NBD_OPT_FORK          263
  65#define QEMU_NBD_OPT_TLSAUTHZ      264
  66#define QEMU_NBD_OPT_PID_FILE      265
  67
  68#define MBR_SIZE 512
  69
  70static int verbose;
  71static char *srcpath;
  72static SocketAddress *saddr;
  73static int persistent = 0;
  74static enum { RUNNING, TERMINATE, TERMINATED } state;
  75static int shared = 1;
  76static int nb_fds;
  77static QIONetListener *server;
  78static QCryptoTLSCreds *tlscreds;
  79static const char *tlsauthz;
  80
  81static void usage(const char *name)
  82{
  83    (printf) (
  84"Usage: %s [OPTIONS] FILE\n"
  85"  or:  %s -L [OPTIONS]\n"
  86"QEMU Disk Network Block Device Utility\n"
  87"\n"
  88"  -h, --help                display this help and exit\n"
  89"  -V, --version             output version information and exit\n"
  90"\n"
  91"Connection properties:\n"
  92"  -p, --port=PORT           port to listen on (default `%d')\n"
  93"  -b, --bind=IFACE          interface to bind to (default `0.0.0.0')\n"
  94"  -k, --socket=PATH         path to the unix socket\n"
  95"                            (default '"SOCKET_PATH"')\n"
  96"  -e, --shared=NUM          device can be shared by NUM clients (default '1')\n"
  97"  -t, --persistent          don't exit on the last connection\n"
  98"  -v, --verbose             display extra debugging information\n"
  99"  -x, --export-name=NAME    expose export by name (default is empty string)\n"
 100"  -D, --description=TEXT    export a human-readable description\n"
 101"\n"
 102"Exposing part of the image:\n"
 103"  -o, --offset=OFFSET       offset into the image\n"
 104"  -A, --allocation-depth    expose the allocation depth\n"
 105"  -B, --bitmap=NAME         expose a persistent dirty bitmap\n"
 106"\n"
 107"General purpose options:\n"
 108"  -L, --list                list exports available from another NBD server\n"
 109"  --object type,id=ID,...   define an object such as 'secret' for providing\n"
 110"                            passwords and/or encryption keys\n"
 111"  --tls-creds=ID            use id of an earlier --object to provide TLS\n"
 112"  --tls-authz=ID            use id of an earlier --object to provide\n"
 113"                            authorization\n"
 114"  -T, --trace [[enable=]<pattern>][,events=<file>][,file=<file>]\n"
 115"                            specify tracing options\n"
 116"  --fork                    fork off the server process and exit the parent\n"
 117"                            once the server is running\n"
 118"  --pid-file=PATH           store the server's process ID in the given file\n"
 119#if HAVE_NBD_DEVICE
 120"\n"
 121"Kernel NBD client support:\n"
 122"  -c, --connect=DEV         connect FILE to the local NBD device DEV\n"
 123"  -d, --disconnect          disconnect the specified device\n"
 124#endif
 125"\n"
 126"Block device options:\n"
 127"  -f, --format=FORMAT       set image format (raw, qcow2, ...)\n"
 128"  -r, --read-only           export read-only\n"
 129"  -s, --snapshot            use FILE as an external snapshot, create a temporary\n"
 130"                            file with backing_file=FILE, redirect the write to\n"
 131"                            the temporary one\n"
 132"  -l, --load-snapshot=SNAPSHOT_PARAM\n"
 133"                            load an internal snapshot inside FILE and export it\n"
 134"                            as an read-only device, SNAPSHOT_PARAM format is\n"
 135"                            'snapshot.id=[ID],snapshot.name=[NAME]', or\n"
 136"                            '[ID_OR_NAME]'\n"
 137"  -n, --nocache             disable host cache\n"
 138"      --cache=MODE          set cache mode used to access the disk image, the\n"
 139"                            valid options are: 'none', 'writeback' (default),\n"
 140"                            'writethrough', 'directsync' and 'unsafe'\n"
 141"      --aio=MODE            set AIO mode (native, io_uring or threads)\n"
 142"      --discard=MODE        set discard mode (ignore, unmap)\n"
 143"      --detect-zeroes=MODE  set detect-zeroes mode (off, on, unmap)\n"
 144"      --image-opts          treat FILE as a full set of image options\n"
 145"\n"
 146QEMU_HELP_BOTTOM "\n"
 147    , name, name, NBD_DEFAULT_PORT, "DEVICE");
 148}
 149
 150static void version(const char *name)
 151{
 152    printf(
 153"%s " QEMU_FULL_VERSION "\n"
 154"Written by Anthony Liguori.\n"
 155"\n"
 156QEMU_COPYRIGHT "\n"
 157"This is free software; see the source for copying conditions.  There is NO\n"
 158"warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
 159    , name);
 160}
 161
 162#ifdef CONFIG_POSIX
 163/*
 164 * The client thread uses SIGTERM to interrupt the server.  A signal
 165 * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
 166 */
 167void qemu_system_killed(int signum, pid_t pid)
 168{
 169    qatomic_cmpxchg(&state, RUNNING, TERMINATE);
 170    qemu_notify_event();
 171}
 172#endif /* CONFIG_POSIX */
 173
 174static int qemu_nbd_client_list(SocketAddress *saddr, QCryptoTLSCreds *tls,
 175                                const char *hostname)
 176{
 177    int ret = EXIT_FAILURE;
 178    int rc;
 179    Error *err = NULL;
 180    QIOChannelSocket *sioc;
 181    NBDExportInfo *list;
 182    int i, j;
 183
 184    sioc = qio_channel_socket_new();
 185    if (qio_channel_socket_connect_sync(sioc, saddr, &err) < 0) {
 186        error_report_err(err);
 187        goto out;
 188    }
 189    rc = nbd_receive_export_list(QIO_CHANNEL(sioc), tls, hostname, &list,
 190                                 &err);
 191    if (rc < 0) {
 192        if (err) {
 193            error_report_err(err);
 194        }
 195        goto out;
 196    }
 197    printf("exports available: %d\n", rc);
 198    for (i = 0; i < rc; i++) {
 199        printf(" export: '%s'\n", list[i].name);
 200        if (list[i].description && *list[i].description) {
 201            printf("  description: %s\n", list[i].description);
 202        }
 203        if (list[i].flags & NBD_FLAG_HAS_FLAGS) {
 204            static const char *const flag_names[] = {
 205                [NBD_FLAG_READ_ONLY_BIT]            = "readonly",
 206                [NBD_FLAG_SEND_FLUSH_BIT]           = "flush",
 207                [NBD_FLAG_SEND_FUA_BIT]             = "fua",
 208                [NBD_FLAG_ROTATIONAL_BIT]           = "rotational",
 209                [NBD_FLAG_SEND_TRIM_BIT]            = "trim",
 210                [NBD_FLAG_SEND_WRITE_ZEROES_BIT]    = "zeroes",
 211                [NBD_FLAG_SEND_DF_BIT]              = "df",
 212                [NBD_FLAG_CAN_MULTI_CONN_BIT]       = "multi",
 213                [NBD_FLAG_SEND_RESIZE_BIT]          = "resize",
 214                [NBD_FLAG_SEND_CACHE_BIT]           = "cache",
 215                [NBD_FLAG_SEND_FAST_ZERO_BIT]       = "fast-zero",
 216            };
 217
 218            printf("  size:  %" PRIu64 "\n", list[i].size);
 219            printf("  flags: 0x%x (", list[i].flags);
 220            for (size_t bit = 0; bit < ARRAY_SIZE(flag_names); bit++) {
 221                if (flag_names[bit] && (list[i].flags & (1 << bit))) {
 222                    printf(" %s", flag_names[bit]);
 223                }
 224            }
 225            printf(" )\n");
 226        }
 227        if (list[i].min_block) {
 228            printf("  min block: %u\n", list[i].min_block);
 229            printf("  opt block: %u\n", list[i].opt_block);
 230            printf("  max block: %u\n", list[i].max_block);
 231        }
 232        if (list[i].n_contexts) {
 233            printf("  available meta contexts: %d\n", list[i].n_contexts);
 234            for (j = 0; j < list[i].n_contexts; j++) {
 235                printf("   %s\n", list[i].contexts[j]);
 236            }
 237        }
 238    }
 239    nbd_free_export_list(list, rc);
 240
 241    ret = EXIT_SUCCESS;
 242 out:
 243    object_unref(OBJECT(sioc));
 244    return ret;
 245}
 246
 247
 248#if HAVE_NBD_DEVICE
 249static void *show_parts(void *arg)
 250{
 251    char *device = arg;
 252    int nbd;
 253
 254    /* linux just needs an open() to trigger
 255     * the partition table update
 256     * but remember to load the module with max_part != 0 :
 257     *     modprobe nbd max_part=63
 258     */
 259    nbd = open(device, O_RDWR);
 260    if (nbd >= 0) {
 261        close(nbd);
 262    }
 263    return NULL;
 264}
 265
 266static void *nbd_client_thread(void *arg)
 267{
 268    char *device = arg;
 269    NBDExportInfo info = { .request_sizes = false, .name = g_strdup("") };
 270    QIOChannelSocket *sioc;
 271    int fd = -1;
 272    int ret = EXIT_FAILURE;
 273    pthread_t show_parts_thread;
 274    Error *local_error = NULL;
 275
 276    sioc = qio_channel_socket_new();
 277    if (qio_channel_socket_connect_sync(sioc,
 278                                        saddr,
 279                                        &local_error) < 0) {
 280        error_report_err(local_error);
 281        goto out;
 282    }
 283
 284    if (nbd_receive_negotiate(NULL, QIO_CHANNEL(sioc),
 285                              NULL, NULL, NULL, &info, &local_error) < 0) {
 286        if (local_error) {
 287            error_report_err(local_error);
 288        }
 289        goto out;
 290    }
 291
 292    fd = open(device, O_RDWR);
 293    if (fd < 0) {
 294        /* Linux-only, we can use %m in printf.  */
 295        error_report("Failed to open %s: %m", device);
 296        goto out;
 297    }
 298
 299    if (nbd_init(fd, sioc, &info, &local_error) < 0) {
 300        error_report_err(local_error);
 301        goto out;
 302    }
 303
 304    /* update partition table */
 305    pthread_create(&show_parts_thread, NULL, show_parts, device);
 306
 307    if (verbose) {
 308        fprintf(stderr, "NBD device %s is now connected to %s\n",
 309                device, srcpath);
 310    } else {
 311        /* Close stderr so that the qemu-nbd process exits.  */
 312        dup2(STDOUT_FILENO, STDERR_FILENO);
 313    }
 314
 315    if (nbd_client(fd) < 0) {
 316        goto out;
 317    }
 318
 319    ret = EXIT_SUCCESS;
 320
 321 out:
 322    if (fd >= 0) {
 323        close(fd);
 324    }
 325    object_unref(OBJECT(sioc));
 326    g_free(info.name);
 327    kill(getpid(), SIGTERM);
 328    return (void *) (intptr_t) ret;
 329}
 330#endif /* HAVE_NBD_DEVICE */
 331
 332static int nbd_can_accept(void)
 333{
 334    return state == RUNNING && (shared == 0 || nb_fds < shared);
 335}
 336
 337static void nbd_update_server_watch(void);
 338
 339static void nbd_client_closed(NBDClient *client, bool negotiated)
 340{
 341    nb_fds--;
 342    if (negotiated && nb_fds == 0 && !persistent && state == RUNNING) {
 343        state = TERMINATE;
 344    }
 345    nbd_update_server_watch();
 346    nbd_client_put(client);
 347}
 348
 349static void nbd_accept(QIONetListener *listener, QIOChannelSocket *cioc,
 350                       gpointer opaque)
 351{
 352    if (state >= TERMINATE) {
 353        return;
 354    }
 355
 356    nb_fds++;
 357    nbd_update_server_watch();
 358    nbd_client_new(cioc, tlscreds, tlsauthz, nbd_client_closed);
 359}
 360
 361static void nbd_update_server_watch(void)
 362{
 363    if (nbd_can_accept()) {
 364        qio_net_listener_set_client_func(server, nbd_accept, NULL, NULL);
 365    } else {
 366        qio_net_listener_set_client_func(server, NULL, NULL, NULL);
 367    }
 368}
 369
 370
 371static SocketAddress *nbd_build_socket_address(const char *sockpath,
 372                                               const char *bindto,
 373                                               const char *port)
 374{
 375    SocketAddress *saddr;
 376
 377    saddr = g_new0(SocketAddress, 1);
 378    if (sockpath) {
 379        saddr->type = SOCKET_ADDRESS_TYPE_UNIX;
 380        saddr->u.q_unix.path = g_strdup(sockpath);
 381    } else {
 382        InetSocketAddress *inet;
 383        saddr->type = SOCKET_ADDRESS_TYPE_INET;
 384        inet = &saddr->u.inet;
 385        inet->host = g_strdup(bindto);
 386        if (port) {
 387            inet->port = g_strdup(port);
 388        } else  {
 389            inet->port = g_strdup_printf("%d", NBD_DEFAULT_PORT);
 390        }
 391    }
 392
 393    return saddr;
 394}
 395
 396
 397static QemuOptsList file_opts = {
 398    .name = "file",
 399    .implied_opt_name = "file",
 400    .head = QTAILQ_HEAD_INITIALIZER(file_opts.head),
 401    .desc = {
 402        /* no elements => accept any params */
 403        { /* end of list */ }
 404    },
 405};
 406
 407static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, bool list,
 408                                          Error **errp)
 409{
 410    Object *obj;
 411    QCryptoTLSCreds *creds;
 412
 413    obj = object_resolve_path_component(
 414        object_get_objects_root(), id);
 415    if (!obj) {
 416        error_setg(errp, "No TLS credentials with id '%s'",
 417                   id);
 418        return NULL;
 419    }
 420    creds = (QCryptoTLSCreds *)
 421        object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
 422    if (!creds) {
 423        error_setg(errp, "Object with id '%s' is not TLS credentials",
 424                   id);
 425        return NULL;
 426    }
 427
 428    if (!qcrypto_tls_creds_check_endpoint(creds,
 429                                          list
 430                                          ? QCRYPTO_TLS_CREDS_ENDPOINT_CLIENT
 431                                          : QCRYPTO_TLS_CREDS_ENDPOINT_SERVER,
 432                                          errp)) {
 433        return NULL;
 434    }
 435    object_ref(obj);
 436    return creds;
 437}
 438
 439static void setup_address_and_port(const char **address, const char **port)
 440{
 441    if (*address == NULL) {
 442        *address = "0.0.0.0";
 443    }
 444
 445    if (*port == NULL) {
 446        *port = stringify(NBD_DEFAULT_PORT);
 447    }
 448}
 449
 450/*
 451 * Check socket parameters compatibility when socket activation is used.
 452 */
 453static const char *socket_activation_validate_opts(const char *device,
 454                                                   const char *sockpath,
 455                                                   const char *address,
 456                                                   const char *port,
 457                                                   bool list)
 458{
 459    if (device != NULL) {
 460        return "NBD device can't be set when using socket activation";
 461    }
 462
 463    if (sockpath != NULL) {
 464        return "Unix socket can't be set when using socket activation";
 465    }
 466
 467    if (address != NULL) {
 468        return "The interface can't be set when using socket activation";
 469    }
 470
 471    if (port != NULL) {
 472        return "TCP port number can't be set when using socket activation";
 473    }
 474
 475    if (list) {
 476        return "List mode is incompatible with socket activation";
 477    }
 478
 479    return NULL;
 480}
 481
 482static void qemu_nbd_shutdown(void)
 483{
 484    job_cancel_sync_all();
 485    blk_exp_close_all();
 486    bdrv_close_all();
 487}
 488
 489int main(int argc, char **argv)
 490{
 491    BlockBackend *blk;
 492    BlockDriverState *bs;
 493    uint64_t dev_offset = 0;
 494    bool readonly = false;
 495    bool disconnect = false;
 496    const char *bindto = NULL;
 497    const char *port = NULL;
 498    char *sockpath = NULL;
 499    char *device = NULL;
 500    QemuOpts *sn_opts = NULL;
 501    const char *sn_id_or_name = NULL;
 502    const char *sopt = "hVb:o:p:rsnc:dvk:e:f:tl:x:T:D:AB:L";
 503    struct option lopt[] = {
 504        { "help", no_argument, NULL, 'h' },
 505        { "version", no_argument, NULL, 'V' },
 506        { "bind", required_argument, NULL, 'b' },
 507        { "port", required_argument, NULL, 'p' },
 508        { "socket", required_argument, NULL, 'k' },
 509        { "offset", required_argument, NULL, 'o' },
 510        { "read-only", no_argument, NULL, 'r' },
 511        { "allocation-depth", no_argument, NULL, 'A' },
 512        { "bitmap", required_argument, NULL, 'B' },
 513        { "connect", required_argument, NULL, 'c' },
 514        { "disconnect", no_argument, NULL, 'd' },
 515        { "list", no_argument, NULL, 'L' },
 516        { "snapshot", no_argument, NULL, 's' },
 517        { "load-snapshot", required_argument, NULL, 'l' },
 518        { "nocache", no_argument, NULL, 'n' },
 519        { "cache", required_argument, NULL, QEMU_NBD_OPT_CACHE },
 520        { "aio", required_argument, NULL, QEMU_NBD_OPT_AIO },
 521        { "discard", required_argument, NULL, QEMU_NBD_OPT_DISCARD },
 522        { "detect-zeroes", required_argument, NULL,
 523          QEMU_NBD_OPT_DETECT_ZEROES },
 524        { "shared", required_argument, NULL, 'e' },
 525        { "format", required_argument, NULL, 'f' },
 526        { "persistent", no_argument, NULL, 't' },
 527        { "verbose", no_argument, NULL, 'v' },
 528        { "object", required_argument, NULL, QEMU_NBD_OPT_OBJECT },
 529        { "export-name", required_argument, NULL, 'x' },
 530        { "description", required_argument, NULL, 'D' },
 531        { "tls-creds", required_argument, NULL, QEMU_NBD_OPT_TLSCREDS },
 532        { "tls-authz", required_argument, NULL, QEMU_NBD_OPT_TLSAUTHZ },
 533        { "image-opts", no_argument, NULL, QEMU_NBD_OPT_IMAGE_OPTS },
 534        { "trace", required_argument, NULL, 'T' },
 535        { "fork", no_argument, NULL, QEMU_NBD_OPT_FORK },
 536        { "pid-file", required_argument, NULL, QEMU_NBD_OPT_PID_FILE },
 537        { NULL, 0, NULL, 0 }
 538    };
 539    int ch;
 540    int opt_ind = 0;
 541    int flags = BDRV_O_RDWR;
 542    int ret = 0;
 543    bool seen_cache = false;
 544    bool seen_discard = false;
 545    bool seen_aio = false;
 546    pthread_t client_thread;
 547    const char *fmt = NULL;
 548    Error *local_err = NULL;
 549    BlockdevDetectZeroesOptions detect_zeroes = BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
 550    QDict *options = NULL;
 551    const char *export_name = NULL; /* defaults to "" later for server mode */
 552    const char *export_description = NULL;
 553    strList *bitmaps = NULL;
 554    bool alloc_depth = false;
 555    const char *tlscredsid = NULL;
 556    bool imageOpts = false;
 557    bool writethrough = false; /* Client will flush as needed. */
 558    bool fork_process = false;
 559    bool list = false;
 560    int old_stderr = -1;
 561    unsigned socket_activation;
 562    const char *pid_file_name = NULL;
 563    BlockExportOptions *export_opts;
 564
 565#ifdef CONFIG_POSIX
 566    os_setup_early_signal_handling();
 567    os_setup_signal_handling();
 568#endif
 569
 570    socket_init();
 571    error_init(argv[0]);
 572    module_call_init(MODULE_INIT_TRACE);
 573    qcrypto_init(&error_fatal);
 574
 575    module_call_init(MODULE_INIT_QOM);
 576    qemu_add_opts(&qemu_trace_opts);
 577    qemu_init_exec_dir(argv[0]);
 578
 579    while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 580        switch (ch) {
 581        case 's':
 582            flags |= BDRV_O_SNAPSHOT;
 583            break;
 584        case 'n':
 585            optarg = (char *) "none";
 586            /* fallthrough */
 587        case QEMU_NBD_OPT_CACHE:
 588            if (seen_cache) {
 589                error_report("-n and --cache can only be specified once");
 590                exit(EXIT_FAILURE);
 591            }
 592            seen_cache = true;
 593            if (bdrv_parse_cache_mode(optarg, &flags, &writethrough) == -1) {
 594                error_report("Invalid cache mode `%s'", optarg);
 595                exit(EXIT_FAILURE);
 596            }
 597            break;
 598        case QEMU_NBD_OPT_AIO:
 599            if (seen_aio) {
 600                error_report("--aio can only be specified once");
 601                exit(EXIT_FAILURE);
 602            }
 603            seen_aio = true;
 604            if (bdrv_parse_aio(optarg, &flags) < 0) {
 605                error_report("Invalid aio mode '%s'", optarg);
 606                exit(EXIT_FAILURE);
 607            }
 608            break;
 609        case QEMU_NBD_OPT_DISCARD:
 610            if (seen_discard) {
 611                error_report("--discard can only be specified once");
 612                exit(EXIT_FAILURE);
 613            }
 614            seen_discard = true;
 615            if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
 616                error_report("Invalid discard mode `%s'", optarg);
 617                exit(EXIT_FAILURE);
 618            }
 619            break;
 620        case QEMU_NBD_OPT_DETECT_ZEROES:
 621            detect_zeroes =
 622                qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
 623                                optarg,
 624                                BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
 625                                &local_err);
 626            if (local_err) {
 627                error_reportf_err(local_err,
 628                                  "Failed to parse detect_zeroes mode: ");
 629                exit(EXIT_FAILURE);
 630            }
 631            if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
 632                !(flags & BDRV_O_UNMAP)) {
 633                error_report("setting detect-zeroes to unmap is not allowed "
 634                             "without setting discard operation to unmap");
 635                exit(EXIT_FAILURE);
 636            }
 637            break;
 638        case 'b':
 639            bindto = optarg;
 640            break;
 641        case 'p':
 642            port = optarg;
 643            break;
 644        case 'o':
 645            if (qemu_strtou64(optarg, NULL, 0, &dev_offset) < 0) {
 646                error_report("Invalid offset '%s'", optarg);
 647                exit(EXIT_FAILURE);
 648            }
 649            break;
 650        case 'l':
 651            if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
 652                sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
 653                                                  optarg, false);
 654                if (!sn_opts) {
 655                    error_report("Failed in parsing snapshot param `%s'",
 656                                 optarg);
 657                    exit(EXIT_FAILURE);
 658                }
 659            } else {
 660                sn_id_or_name = optarg;
 661            }
 662            /* fall through */
 663        case 'r':
 664            readonly = true;
 665            flags &= ~BDRV_O_RDWR;
 666            break;
 667        case 'A':
 668            alloc_depth = true;
 669            break;
 670        case 'B':
 671            QAPI_LIST_PREPEND(bitmaps, g_strdup(optarg));
 672            break;
 673        case 'k':
 674            sockpath = optarg;
 675            if (sockpath[0] != '/') {
 676                error_report("socket path must be absolute");
 677                exit(EXIT_FAILURE);
 678            }
 679            break;
 680        case 'd':
 681            disconnect = true;
 682            break;
 683        case 'c':
 684            device = optarg;
 685            break;
 686        case 'e':
 687            if (qemu_strtoi(optarg, NULL, 0, &shared) < 0 ||
 688                shared < 0) {
 689                error_report("Invalid shared device number '%s'", optarg);
 690                exit(EXIT_FAILURE);
 691            }
 692            break;
 693        case 'f':
 694            fmt = optarg;
 695            break;
 696        case 't':
 697            persistent = 1;
 698            break;
 699        case 'x':
 700            export_name = optarg;
 701            if (strlen(export_name) > NBD_MAX_STRING_SIZE) {
 702                error_report("export name '%s' too long", export_name);
 703                exit(EXIT_FAILURE);
 704            }
 705            break;
 706        case 'D':
 707            export_description = optarg;
 708            if (strlen(export_description) > NBD_MAX_STRING_SIZE) {
 709                error_report("export description '%s' too long",
 710                             export_description);
 711                exit(EXIT_FAILURE);
 712            }
 713            break;
 714        case 'v':
 715            verbose = 1;
 716            break;
 717        case 'V':
 718            version(argv[0]);
 719            exit(0);
 720            break;
 721        case 'h':
 722            usage(argv[0]);
 723            exit(0);
 724            break;
 725        case '?':
 726            error_report("Try `%s --help' for more information.", argv[0]);
 727            exit(EXIT_FAILURE);
 728        case QEMU_NBD_OPT_OBJECT:
 729            user_creatable_process_cmdline(optarg);
 730            break;
 731        case QEMU_NBD_OPT_TLSCREDS:
 732            tlscredsid = optarg;
 733            break;
 734        case QEMU_NBD_OPT_IMAGE_OPTS:
 735            imageOpts = true;
 736            break;
 737        case 'T':
 738            trace_opt_parse(optarg);
 739            break;
 740        case QEMU_NBD_OPT_TLSAUTHZ:
 741            tlsauthz = optarg;
 742            break;
 743        case QEMU_NBD_OPT_FORK:
 744            fork_process = true;
 745            break;
 746        case 'L':
 747            list = true;
 748            break;
 749        case QEMU_NBD_OPT_PID_FILE:
 750            pid_file_name = optarg;
 751            break;
 752        }
 753    }
 754
 755    if (list) {
 756        if (argc != optind) {
 757            error_report("List mode is incompatible with a file name");
 758            exit(EXIT_FAILURE);
 759        }
 760        if (export_name || export_description || dev_offset ||
 761            device || disconnect || fmt || sn_id_or_name || bitmaps ||
 762            alloc_depth || seen_aio || seen_discard || seen_cache) {
 763            error_report("List mode is incompatible with per-device settings");
 764            exit(EXIT_FAILURE);
 765        }
 766        if (fork_process) {
 767            error_report("List mode is incompatible with forking");
 768            exit(EXIT_FAILURE);
 769        }
 770    } else if ((argc - optind) != 1) {
 771        error_report("Invalid number of arguments");
 772        error_printf("Try `%s --help' for more information.\n", argv[0]);
 773        exit(EXIT_FAILURE);
 774    } else if (!export_name) {
 775        export_name = "";
 776    }
 777
 778    if (!trace_init_backends()) {
 779        exit(1);
 780    }
 781    trace_init_file();
 782    qemu_set_log(LOG_TRACE);
 783
 784    socket_activation = check_socket_activation();
 785    if (socket_activation == 0) {
 786        setup_address_and_port(&bindto, &port);
 787    } else {
 788        /* Using socket activation - check user didn't use -p etc. */
 789        const char *err_msg = socket_activation_validate_opts(device, sockpath,
 790                                                              bindto, port,
 791                                                              list);
 792        if (err_msg != NULL) {
 793            error_report("%s", err_msg);
 794            exit(EXIT_FAILURE);
 795        }
 796
 797        /* qemu-nbd can only listen on a single socket.  */
 798        if (socket_activation > 1) {
 799            error_report("qemu-nbd does not support socket activation with %s > 1",
 800                         "LISTEN_FDS");
 801            exit(EXIT_FAILURE);
 802        }
 803    }
 804
 805    if (tlscredsid) {
 806        if (sockpath) {
 807            error_report("TLS is only supported with IPv4/IPv6");
 808            exit(EXIT_FAILURE);
 809        }
 810        if (device) {
 811            error_report("TLS is not supported with a host device");
 812            exit(EXIT_FAILURE);
 813        }
 814        if (tlsauthz && list) {
 815            error_report("TLS authorization is incompatible with export list");
 816            exit(EXIT_FAILURE);
 817        }
 818        tlscreds = nbd_get_tls_creds(tlscredsid, list, &local_err);
 819        if (local_err) {
 820            error_reportf_err(local_err, "Failed to get TLS creds: ");
 821            exit(EXIT_FAILURE);
 822        }
 823    } else {
 824        if (tlsauthz) {
 825            error_report("--tls-authz is not permitted without --tls-creds");
 826            exit(EXIT_FAILURE);
 827        }
 828    }
 829
 830    if (list) {
 831        saddr = nbd_build_socket_address(sockpath, bindto, port);
 832        return qemu_nbd_client_list(saddr, tlscreds, bindto);
 833    }
 834
 835#if !HAVE_NBD_DEVICE
 836    if (disconnect || device) {
 837        error_report("Kernel /dev/nbdN support not available");
 838        exit(EXIT_FAILURE);
 839    }
 840#else /* HAVE_NBD_DEVICE */
 841    if (disconnect) {
 842        int nbdfd = open(argv[optind], O_RDWR);
 843        if (nbdfd < 0) {
 844            error_report("Cannot open %s: %s", argv[optind],
 845                         strerror(errno));
 846            exit(EXIT_FAILURE);
 847        }
 848        nbd_disconnect(nbdfd);
 849
 850        close(nbdfd);
 851
 852        printf("%s disconnected\n", argv[optind]);
 853
 854        return 0;
 855    }
 856#endif
 857
 858    if ((device && !verbose) || fork_process) {
 859#ifndef WIN32
 860        int stderr_fd[2];
 861        pid_t pid;
 862        int ret;
 863
 864        if (qemu_pipe(stderr_fd) < 0) {
 865            error_report("Error setting up communication pipe: %s",
 866                         strerror(errno));
 867            exit(EXIT_FAILURE);
 868        }
 869
 870        /* Now daemonize, but keep a communication channel open to
 871         * print errors and exit with the proper status code.
 872         */
 873        pid = fork();
 874        if (pid < 0) {
 875            error_report("Failed to fork: %s", strerror(errno));
 876            exit(EXIT_FAILURE);
 877        } else if (pid == 0) {
 878            close(stderr_fd[0]);
 879
 880            /* Remember parent's stderr if we will be restoring it. */
 881            if (fork_process) {
 882                old_stderr = dup(STDERR_FILENO);
 883            }
 884
 885            ret = qemu_daemon(1, 0);
 886
 887            /* Temporarily redirect stderr to the parent's pipe...  */
 888            dup2(stderr_fd[1], STDERR_FILENO);
 889            if (ret < 0) {
 890                error_report("Failed to daemonize: %s", strerror(errno));
 891                exit(EXIT_FAILURE);
 892            }
 893
 894            /* ... close the descriptor we inherited and go on.  */
 895            close(stderr_fd[1]);
 896        } else {
 897            bool errors = false;
 898            char *buf;
 899
 900            /* In the parent.  Print error messages from the child until
 901             * it closes the pipe.
 902             */
 903            close(stderr_fd[1]);
 904            buf = g_malloc(1024);
 905            while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
 906                errors = true;
 907                ret = qemu_write_full(STDERR_FILENO, buf, ret);
 908                if (ret < 0) {
 909                    exit(EXIT_FAILURE);
 910                }
 911            }
 912            if (ret < 0) {
 913                error_report("Cannot read from daemon: %s",
 914                             strerror(errno));
 915                exit(EXIT_FAILURE);
 916            }
 917
 918            /* Usually the daemon should not print any message.
 919             * Exit with zero status in that case.
 920             */
 921            exit(errors);
 922        }
 923#else /* WIN32 */
 924        error_report("Unable to fork into background on Windows hosts");
 925        exit(EXIT_FAILURE);
 926#endif /* WIN32 */
 927    }
 928
 929    if (device != NULL && sockpath == NULL) {
 930        sockpath = g_malloc(128);
 931        snprintf(sockpath, 128, SOCKET_PATH, basename(device));
 932    }
 933
 934    server = qio_net_listener_new();
 935    if (socket_activation == 0) {
 936        int backlog;
 937
 938        if (persistent || shared == 0) {
 939            backlog = SOMAXCONN;
 940        } else {
 941            backlog = MIN(shared, SOMAXCONN);
 942        }
 943        saddr = nbd_build_socket_address(sockpath, bindto, port);
 944        if (qio_net_listener_open_sync(server, saddr, backlog,
 945                                       &local_err) < 0) {
 946            object_unref(OBJECT(server));
 947            error_report_err(local_err);
 948            exit(EXIT_FAILURE);
 949        }
 950    } else {
 951        size_t i;
 952        /* See comment in check_socket_activation above. */
 953        for (i = 0; i < socket_activation; i++) {
 954            QIOChannelSocket *sioc;
 955            sioc = qio_channel_socket_new_fd(FIRST_SOCKET_ACTIVATION_FD + i,
 956                                             &local_err);
 957            if (sioc == NULL) {
 958                object_unref(OBJECT(server));
 959                error_reportf_err(local_err,
 960                                  "Failed to use socket activation: ");
 961                exit(EXIT_FAILURE);
 962            }
 963            qio_net_listener_add(server, sioc);
 964            object_unref(OBJECT(sioc));
 965        }
 966    }
 967
 968    if (qemu_init_main_loop(&local_err)) {
 969        error_report_err(local_err);
 970        exit(EXIT_FAILURE);
 971    }
 972    bdrv_init();
 973    atexit(qemu_nbd_shutdown);
 974
 975    srcpath = argv[optind];
 976    if (imageOpts) {
 977        QemuOpts *opts;
 978        if (fmt) {
 979            error_report("--image-opts and -f are mutually exclusive");
 980            exit(EXIT_FAILURE);
 981        }
 982        opts = qemu_opts_parse_noisily(&file_opts, srcpath, true);
 983        if (!opts) {
 984            qemu_opts_reset(&file_opts);
 985            exit(EXIT_FAILURE);
 986        }
 987        options = qemu_opts_to_qdict(opts, NULL);
 988        qemu_opts_reset(&file_opts);
 989        blk = blk_new_open(NULL, NULL, options, flags, &local_err);
 990    } else {
 991        if (fmt) {
 992            options = qdict_new();
 993            qdict_put_str(options, "driver", fmt);
 994        }
 995        blk = blk_new_open(srcpath, NULL, options, flags, &local_err);
 996    }
 997
 998    if (!blk) {
 999        error_reportf_err(local_err, "Failed to blk_new_open '%s': ",
1000                          argv[optind]);
1001        exit(EXIT_FAILURE);
1002    }
1003    bs = blk_bs(blk);
1004
1005    if (dev_offset) {
1006        QDict *raw_opts = qdict_new();
1007        qdict_put_str(raw_opts, "driver", "raw");
1008        qdict_put_str(raw_opts, "file", bs->node_name);
1009        qdict_put_int(raw_opts, "offset", dev_offset);
1010        bs = bdrv_open(NULL, NULL, raw_opts, flags, &error_fatal);
1011        blk_remove_bs(blk);
1012        blk_insert_bs(blk, bs, &error_fatal);
1013        bdrv_unref(bs);
1014    }
1015
1016    blk_set_enable_write_cache(blk, !writethrough);
1017
1018    if (sn_opts) {
1019        ret = bdrv_snapshot_load_tmp(bs,
1020                                     qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
1021                                     qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
1022                                     &local_err);
1023    } else if (sn_id_or_name) {
1024        ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
1025                                                   &local_err);
1026    }
1027    if (ret < 0) {
1028        error_reportf_err(local_err, "Failed to load snapshot: ");
1029        exit(EXIT_FAILURE);
1030    }
1031
1032    bs->detect_zeroes = detect_zeroes;
1033
1034    nbd_server_is_qemu_nbd(true);
1035
1036    export_opts = g_new(BlockExportOptions, 1);
1037    *export_opts = (BlockExportOptions) {
1038        .type               = BLOCK_EXPORT_TYPE_NBD,
1039        .id                 = g_strdup("qemu-nbd-export"),
1040        .node_name          = g_strdup(bdrv_get_node_name(bs)),
1041        .has_writethrough   = true,
1042        .writethrough       = writethrough,
1043        .has_writable       = true,
1044        .writable           = !readonly,
1045        .u.nbd = {
1046            .has_name             = true,
1047            .name                 = g_strdup(export_name),
1048            .has_description      = !!export_description,
1049            .description          = g_strdup(export_description),
1050            .has_bitmaps          = !!bitmaps,
1051            .bitmaps              = bitmaps,
1052            .has_allocation_depth = alloc_depth,
1053            .allocation_depth     = alloc_depth,
1054        },
1055    };
1056    blk_exp_add(export_opts, &error_fatal);
1057    qapi_free_BlockExportOptions(export_opts);
1058
1059    if (device) {
1060#if HAVE_NBD_DEVICE
1061        int ret;
1062
1063        ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
1064        if (ret != 0) {
1065            error_report("Failed to create client thread: %s", strerror(ret));
1066            exit(EXIT_FAILURE);
1067        }
1068#endif
1069    } else {
1070        /* Shut up GCC warnings.  */
1071        memset(&client_thread, 0, sizeof(client_thread));
1072    }
1073
1074    nbd_update_server_watch();
1075
1076    if (pid_file_name) {
1077        qemu_write_pidfile(pid_file_name, &error_fatal);
1078    }
1079
1080    /* now when the initialization is (almost) complete, chdir("/")
1081     * to free any busy filesystems */
1082    if (chdir("/") < 0) {
1083        error_report("Could not chdir to root directory: %s",
1084                     strerror(errno));
1085        exit(EXIT_FAILURE);
1086    }
1087
1088    if (fork_process) {
1089        dup2(old_stderr, STDERR_FILENO);
1090        close(old_stderr);
1091    }
1092
1093    state = RUNNING;
1094    do {
1095        main_loop_wait(false);
1096        if (state == TERMINATE) {
1097            blk_exp_close_all();
1098            state = TERMINATED;
1099        }
1100    } while (state != TERMINATED);
1101
1102    blk_unref(blk);
1103    if (sockpath) {
1104        unlink(sockpath);
1105    }
1106
1107    qemu_opts_del(sn_opts);
1108
1109    if (device) {
1110        void *ret;
1111        pthread_join(client_thread, &ret);
1112        exit(ret != NULL);
1113    } else {
1114        exit(EXIT_SUCCESS);
1115    }
1116}
1117