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 "qapi/error.h"
  25#include "qemu/cutils.h"
  26#include "sysemu/block-backend.h"
  27#include "block/block_int.h"
  28#include "block/nbd.h"
  29#include "qemu/main-loop.h"
  30#include "qemu/option.h"
  31#include "qemu/error-report.h"
  32#include "qemu/config-file.h"
  33#include "qemu/bswap.h"
  34#include "qemu/log.h"
  35#include "qemu/systemd.h"
  36#include "block/snapshot.h"
  37#include "qapi/qmp/qdict.h"
  38#include "qapi/qmp/qstring.h"
  39#include "qom/object_interfaces.h"
  40#include "io/channel-socket.h"
  41#include "io/net-listener.h"
  42#include "crypto/init.h"
  43#include "trace/control.h"
  44#include "qemu-version.h"
  45
  46#define SOCKET_PATH                "/var/lock/qemu-nbd-%s"
  47#define QEMU_NBD_OPT_CACHE         256
  48#define QEMU_NBD_OPT_AIO           257
  49#define QEMU_NBD_OPT_DISCARD       258
  50#define QEMU_NBD_OPT_DETECT_ZEROES 259
  51#define QEMU_NBD_OPT_OBJECT        260
  52#define QEMU_NBD_OPT_TLSCREDS      261
  53#define QEMU_NBD_OPT_IMAGE_OPTS    262
  54#define QEMU_NBD_OPT_FORK          263
  55
  56#define MBR_SIZE 512
  57
  58static NBDExport *exp;
  59static bool newproto;
  60static int verbose;
  61static char *srcpath;
  62static SocketAddress *saddr;
  63static int persistent = 0;
  64static enum { RUNNING, TERMINATE, TERMINATING, TERMINATED } state;
  65static int shared = 1;
  66static int nb_fds;
  67static QIONetListener *server;
  68static QCryptoTLSCreds *tlscreds;
  69
  70static void usage(const char *name)
  71{
  72    (printf) (
  73"Usage: %s [OPTIONS] FILE\n"
  74"QEMU Disk Network Block Device Server\n"
  75"\n"
  76"  -h, --help                display this help and exit\n"
  77"  -V, --version             output version information and exit\n"
  78"\n"
  79"Connection properties:\n"
  80"  -p, --port=PORT           port to listen on (default `%d')\n"
  81"  -b, --bind=IFACE          interface to bind to (default `0.0.0.0')\n"
  82"  -k, --socket=PATH         path to the unix socket\n"
  83"                            (default '"SOCKET_PATH"')\n"
  84"  -e, --shared=NUM          device can be shared by NUM clients (default '1')\n"
  85"  -t, --persistent          don't exit on the last connection\n"
  86"  -v, --verbose             display extra debugging information\n"
  87"  -x, --export-name=NAME    expose export by name\n"
  88"  -D, --description=TEXT    with -x, also export a human-readable description\n"
  89"\n"
  90"Exposing part of the image:\n"
  91"  -o, --offset=OFFSET       offset into the image\n"
  92"  -P, --partition=NUM       only expose partition NUM\n"
  93"\n"
  94"General purpose options:\n"
  95"  --object type,id=ID,...   define an object such as 'secret' for providing\n"
  96"                            passwords and/or encryption keys\n"
  97"  -T, --trace [[enable=]<pattern>][,events=<file>][,file=<file>]\n"
  98"                            specify tracing options\n"
  99"  --fork                    fork off the server process and exit the parent\n"
 100"                            once the server is running\n"
 101#ifdef __linux__
 102"Kernel NBD client support:\n"
 103"  -c, --connect=DEV         connect FILE to the local NBD device DEV\n"
 104"  -d, --disconnect          disconnect the specified device\n"
 105"\n"
 106#endif
 107"\n"
 108"Block device options:\n"
 109"  -f, --format=FORMAT       set image format (raw, qcow2, ...)\n"
 110"  -r, --read-only           export read-only\n"
 111"  -s, --snapshot            use FILE as an external snapshot, create a temporary\n"
 112"                            file with backing_file=FILE, redirect the write to\n"
 113"                            the temporary one\n"
 114"  -l, --load-snapshot=SNAPSHOT_PARAM\n"
 115"                            load an internal snapshot inside FILE and export it\n"
 116"                            as an read-only device, SNAPSHOT_PARAM format is\n"
 117"                            'snapshot.id=[ID],snapshot.name=[NAME]', or\n"
 118"                            '[ID_OR_NAME]'\n"
 119"  -n, --nocache             disable host cache\n"
 120"      --cache=MODE          set cache mode (none, writeback, ...)\n"
 121"      --aio=MODE            set AIO mode (native or threads)\n"
 122"      --discard=MODE        set discard mode (ignore, unmap)\n"
 123"      --detect-zeroes=MODE  set detect-zeroes mode (off, on, unmap)\n"
 124"      --image-opts          treat FILE as a full set of image options\n"
 125"\n"
 126QEMU_HELP_BOTTOM "\n"
 127    , name, NBD_DEFAULT_PORT, "DEVICE");
 128}
 129
 130static void version(const char *name)
 131{
 132    printf(
 133"%s " QEMU_FULL_VERSION "\n"
 134"Written by Anthony Liguori.\n"
 135"\n"
 136QEMU_COPYRIGHT "\n"
 137"This is free software; see the source for copying conditions.  There is NO\n"
 138"warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
 139    , name);
 140}
 141
 142struct partition_record
 143{
 144    uint8_t bootable;
 145    uint8_t start_head;
 146    uint32_t start_cylinder;
 147    uint8_t start_sector;
 148    uint8_t system;
 149    uint8_t end_head;
 150    uint8_t end_cylinder;
 151    uint8_t end_sector;
 152    uint32_t start_sector_abs;
 153    uint32_t nb_sectors_abs;
 154};
 155
 156static void read_partition(uint8_t *p, struct partition_record *r)
 157{
 158    r->bootable = p[0];
 159    r->start_head = p[1];
 160    r->start_cylinder = p[3] | ((p[2] << 2) & 0x0300);
 161    r->start_sector = p[2] & 0x3f;
 162    r->system = p[4];
 163    r->end_head = p[5];
 164    r->end_cylinder = p[7] | ((p[6] << 2) & 0x300);
 165    r->end_sector = p[6] & 0x3f;
 166
 167    r->start_sector_abs = ldl_le_p(p + 8);
 168    r->nb_sectors_abs   = ldl_le_p(p + 12);
 169}
 170
 171static int find_partition(BlockBackend *blk, int partition,
 172                          off_t *offset, off_t *size)
 173{
 174    struct partition_record mbr[4];
 175    uint8_t data[MBR_SIZE];
 176    int i;
 177    int ext_partnum = 4;
 178    int ret;
 179
 180    ret = blk_pread(blk, 0, data, sizeof(data));
 181    if (ret < 0) {
 182        error_report("error while reading: %s", strerror(-ret));
 183        exit(EXIT_FAILURE);
 184    }
 185
 186    if (data[510] != 0x55 || data[511] != 0xaa) {
 187        return -EINVAL;
 188    }
 189
 190    for (i = 0; i < 4; i++) {
 191        read_partition(&data[446 + 16 * i], &mbr[i]);
 192
 193        if (!mbr[i].system || !mbr[i].nb_sectors_abs) {
 194            continue;
 195        }
 196
 197        if (mbr[i].system == 0xF || mbr[i].system == 0x5) {
 198            struct partition_record ext[4];
 199            uint8_t data1[MBR_SIZE];
 200            int j;
 201
 202            ret = blk_pread(blk, mbr[i].start_sector_abs * MBR_SIZE,
 203                            data1, sizeof(data1));
 204            if (ret < 0) {
 205                error_report("error while reading: %s", strerror(-ret));
 206                exit(EXIT_FAILURE);
 207            }
 208
 209            for (j = 0; j < 4; j++) {
 210                read_partition(&data1[446 + 16 * j], &ext[j]);
 211                if (!ext[j].system || !ext[j].nb_sectors_abs) {
 212                    continue;
 213                }
 214
 215                if ((ext_partnum + j + 1) == partition) {
 216                    *offset = (uint64_t)ext[j].start_sector_abs << 9;
 217                    *size = (uint64_t)ext[j].nb_sectors_abs << 9;
 218                    return 0;
 219                }
 220            }
 221            ext_partnum += 4;
 222        } else if ((i + 1) == partition) {
 223            *offset = (uint64_t)mbr[i].start_sector_abs << 9;
 224            *size = (uint64_t)mbr[i].nb_sectors_abs << 9;
 225            return 0;
 226        }
 227    }
 228
 229    return -ENOENT;
 230}
 231
 232static void termsig_handler(int signum)
 233{
 234    atomic_cmpxchg(&state, RUNNING, TERMINATE);
 235    qemu_notify_event();
 236}
 237
 238
 239static void *show_parts(void *arg)
 240{
 241    char *device = arg;
 242    int nbd;
 243
 244    /* linux just needs an open() to trigger
 245     * the partition table update
 246     * but remember to load the module with max_part != 0 :
 247     *     modprobe nbd max_part=63
 248     */
 249    nbd = open(device, O_RDWR);
 250    if (nbd >= 0) {
 251        close(nbd);
 252    }
 253    return NULL;
 254}
 255
 256static void *nbd_client_thread(void *arg)
 257{
 258    char *device = arg;
 259    NBDExportInfo info = { .request_sizes = false, };
 260    QIOChannelSocket *sioc;
 261    int fd;
 262    int ret;
 263    pthread_t show_parts_thread;
 264    Error *local_error = NULL;
 265
 266    sioc = qio_channel_socket_new();
 267    if (qio_channel_socket_connect_sync(sioc,
 268                                        saddr,
 269                                        &local_error) < 0) {
 270        error_report_err(local_error);
 271        goto out;
 272    }
 273
 274    ret = nbd_receive_negotiate(QIO_CHANNEL(sioc), NULL,
 275                                NULL, NULL, NULL, &info, &local_error);
 276    if (ret < 0) {
 277        if (local_error) {
 278            error_report_err(local_error);
 279        }
 280        goto out_socket;
 281    }
 282
 283    fd = open(device, O_RDWR);
 284    if (fd < 0) {
 285        /* Linux-only, we can use %m in printf.  */
 286        error_report("Failed to open %s: %m", device);
 287        goto out_socket;
 288    }
 289
 290    ret = nbd_init(fd, sioc, &info, &local_error);
 291    if (ret < 0) {
 292        error_report_err(local_error);
 293        goto out_fd;
 294    }
 295
 296    /* update partition table */
 297    pthread_create(&show_parts_thread, NULL, show_parts, device);
 298
 299    if (verbose) {
 300        fprintf(stderr, "NBD device %s is now connected to %s\n",
 301                device, srcpath);
 302    } else {
 303        /* Close stderr so that the qemu-nbd process exits.  */
 304        dup2(STDOUT_FILENO, STDERR_FILENO);
 305    }
 306
 307    ret = nbd_client(fd);
 308    if (ret) {
 309        goto out_fd;
 310    }
 311    close(fd);
 312    object_unref(OBJECT(sioc));
 313    kill(getpid(), SIGTERM);
 314    return (void *) EXIT_SUCCESS;
 315
 316out_fd:
 317    close(fd);
 318out_socket:
 319    object_unref(OBJECT(sioc));
 320out:
 321    kill(getpid(), SIGTERM);
 322    return (void *) EXIT_FAILURE;
 323}
 324
 325static int nbd_can_accept(void)
 326{
 327    return state == RUNNING && nb_fds < shared;
 328}
 329
 330static void nbd_export_closed(NBDExport *exp)
 331{
 332    assert(state == TERMINATING);
 333    state = TERMINATED;
 334}
 335
 336static void nbd_update_server_watch(void);
 337
 338static void nbd_client_closed(NBDClient *client, bool negotiated)
 339{
 340    nb_fds--;
 341    if (negotiated && nb_fds == 0 && !persistent && state == RUNNING) {
 342        state = TERMINATE;
 343    }
 344    nbd_update_server_watch();
 345    nbd_client_put(client);
 346}
 347
 348static void nbd_accept(QIONetListener *listener, QIOChannelSocket *cioc,
 349                       gpointer opaque)
 350{
 351    if (state >= TERMINATE) {
 352        return;
 353    }
 354
 355    nb_fds++;
 356    nbd_update_server_watch();
 357    nbd_client_new(newproto ? NULL : exp, cioc,
 358                   tlscreds, NULL, 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 QemuOptsList qemu_object_opts = {
 408    .name = "object",
 409    .implied_opt_name = "qom-type",
 410    .head = QTAILQ_HEAD_INITIALIZER(qemu_object_opts.head),
 411    .desc = {
 412        { }
 413    },
 414};
 415
 416
 417
 418static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, Error **errp)
 419{
 420    Object *obj;
 421    QCryptoTLSCreds *creds;
 422
 423    obj = object_resolve_path_component(
 424        object_get_objects_root(), id);
 425    if (!obj) {
 426        error_setg(errp, "No TLS credentials with id '%s'",
 427                   id);
 428        return NULL;
 429    }
 430    creds = (QCryptoTLSCreds *)
 431        object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
 432    if (!creds) {
 433        error_setg(errp, "Object with id '%s' is not TLS credentials",
 434                   id);
 435        return NULL;
 436    }
 437
 438    if (creds->endpoint != QCRYPTO_TLS_CREDS_ENDPOINT_SERVER) {
 439        error_setg(errp,
 440                   "Expecting TLS credentials with a server endpoint");
 441        return NULL;
 442    }
 443    object_ref(obj);
 444    return creds;
 445}
 446
 447static void setup_address_and_port(const char **address, const char **port)
 448{
 449    if (*address == NULL) {
 450        *address = "0.0.0.0";
 451    }
 452
 453    if (*port == NULL) {
 454        *port = stringify(NBD_DEFAULT_PORT);
 455    }
 456}
 457
 458/*
 459 * Check socket parameters compatibility when socket activation is used.
 460 */
 461static const char *socket_activation_validate_opts(const char *device,
 462                                                   const char *sockpath,
 463                                                   const char *address,
 464                                                   const char *port)
 465{
 466    if (device != NULL) {
 467        return "NBD device can't be set when using socket activation";
 468    }
 469
 470    if (sockpath != NULL) {
 471        return "Unix socket can't be set when using socket activation";
 472    }
 473
 474    if (address != NULL) {
 475        return "The interface can't be set when using socket activation";
 476    }
 477
 478    if (port != NULL) {
 479        return "TCP port number can't be set when using socket activation";
 480    }
 481
 482    return NULL;
 483}
 484
 485int main(int argc, char **argv)
 486{
 487    BlockBackend *blk;
 488    BlockDriverState *bs;
 489    off_t dev_offset = 0;
 490    uint16_t nbdflags = 0;
 491    bool disconnect = false;
 492    const char *bindto = NULL;
 493    const char *port = NULL;
 494    char *sockpath = NULL;
 495    char *device = NULL;
 496    off_t fd_size;
 497    QemuOpts *sn_opts = NULL;
 498    const char *sn_id_or_name = NULL;
 499    const char *sopt = "hVb:o:p:rsnP:c:dvk:e:f:tl:x:T:D:";
 500    struct option lopt[] = {
 501        { "help", no_argument, NULL, 'h' },
 502        { "version", no_argument, NULL, 'V' },
 503        { "bind", required_argument, NULL, 'b' },
 504        { "port", required_argument, NULL, 'p' },
 505        { "socket", required_argument, NULL, 'k' },
 506        { "offset", required_argument, NULL, 'o' },
 507        { "read-only", no_argument, NULL, 'r' },
 508        { "partition", required_argument, NULL, 'P' },
 509        { "connect", required_argument, NULL, 'c' },
 510        { "disconnect", no_argument, NULL, 'd' },
 511        { "snapshot", no_argument, NULL, 's' },
 512        { "load-snapshot", required_argument, NULL, 'l' },
 513        { "nocache", no_argument, NULL, 'n' },
 514        { "cache", required_argument, NULL, QEMU_NBD_OPT_CACHE },
 515        { "aio", required_argument, NULL, QEMU_NBD_OPT_AIO },
 516        { "discard", required_argument, NULL, QEMU_NBD_OPT_DISCARD },
 517        { "detect-zeroes", required_argument, NULL,
 518          QEMU_NBD_OPT_DETECT_ZEROES },
 519        { "shared", required_argument, NULL, 'e' },
 520        { "format", required_argument, NULL, 'f' },
 521        { "persistent", no_argument, NULL, 't' },
 522        { "verbose", no_argument, NULL, 'v' },
 523        { "object", required_argument, NULL, QEMU_NBD_OPT_OBJECT },
 524        { "export-name", required_argument, NULL, 'x' },
 525        { "description", required_argument, NULL, 'D' },
 526        { "tls-creds", required_argument, NULL, QEMU_NBD_OPT_TLSCREDS },
 527        { "image-opts", no_argument, NULL, QEMU_NBD_OPT_IMAGE_OPTS },
 528        { "trace", required_argument, NULL, 'T' },
 529        { "fork", no_argument, NULL, QEMU_NBD_OPT_FORK },
 530        { NULL, 0, NULL, 0 }
 531    };
 532    int ch;
 533    int opt_ind = 0;
 534    char *end;
 535    int flags = BDRV_O_RDWR;
 536    int partition = -1;
 537    int ret = 0;
 538    bool seen_cache = false;
 539    bool seen_discard = false;
 540    bool seen_aio = false;
 541    pthread_t client_thread;
 542    const char *fmt = NULL;
 543    Error *local_err = NULL;
 544    BlockdevDetectZeroesOptions detect_zeroes = BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
 545    QDict *options = NULL;
 546    const char *export_name = NULL;
 547    const char *export_description = NULL;
 548    const char *tlscredsid = NULL;
 549    bool imageOpts = false;
 550    bool writethrough = true;
 551    char *trace_file = NULL;
 552    bool fork_process = false;
 553    int old_stderr = -1;
 554    unsigned socket_activation;
 555
 556    /* The client thread uses SIGTERM to interrupt the server.  A signal
 557     * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
 558     */
 559    struct sigaction sa_sigterm;
 560    memset(&sa_sigterm, 0, sizeof(sa_sigterm));
 561    sa_sigterm.sa_handler = termsig_handler;
 562    sigaction(SIGTERM, &sa_sigterm, NULL);
 563
 564#ifdef CONFIG_POSIX
 565    signal(SIGPIPE, SIG_IGN);
 566#endif
 567
 568    module_call_init(MODULE_INIT_TRACE);
 569    qcrypto_init(&error_fatal);
 570
 571    module_call_init(MODULE_INIT_QOM);
 572    qemu_add_opts(&qemu_object_opts);
 573    qemu_add_opts(&qemu_trace_opts);
 574    qemu_init_exec_dir(argv[0]);
 575
 576    while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 577        switch (ch) {
 578        case 's':
 579            flags |= BDRV_O_SNAPSHOT;
 580            break;
 581        case 'n':
 582            optarg = (char *) "none";
 583            /* fallthrough */
 584        case QEMU_NBD_OPT_CACHE:
 585            if (seen_cache) {
 586                error_report("-n and --cache can only be specified once");
 587                exit(EXIT_FAILURE);
 588            }
 589            seen_cache = true;
 590            if (bdrv_parse_cache_mode(optarg, &flags, &writethrough) == -1) {
 591                error_report("Invalid cache mode `%s'", optarg);
 592                exit(EXIT_FAILURE);
 593            }
 594            break;
 595        case QEMU_NBD_OPT_AIO:
 596            if (seen_aio) {
 597                error_report("--aio can only be specified once");
 598                exit(EXIT_FAILURE);
 599            }
 600            seen_aio = true;
 601            if (!strcmp(optarg, "native")) {
 602                flags |= BDRV_O_NATIVE_AIO;
 603            } else if (!strcmp(optarg, "threads")) {
 604                /* this is the default */
 605            } else {
 606               error_report("invalid aio mode `%s'", optarg);
 607               exit(EXIT_FAILURE);
 608            }
 609            break;
 610        case QEMU_NBD_OPT_DISCARD:
 611            if (seen_discard) {
 612                error_report("--discard can only be specified once");
 613                exit(EXIT_FAILURE);
 614            }
 615            seen_discard = true;
 616            if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
 617                error_report("Invalid discard mode `%s'", optarg);
 618                exit(EXIT_FAILURE);
 619            }
 620            break;
 621        case QEMU_NBD_OPT_DETECT_ZEROES:
 622            detect_zeroes =
 623                qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
 624                                optarg,
 625                                BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
 626                                &local_err);
 627            if (local_err) {
 628                error_reportf_err(local_err,
 629                                  "Failed to parse detect_zeroes mode: ");
 630                exit(EXIT_FAILURE);
 631            }
 632            if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
 633                !(flags & BDRV_O_UNMAP)) {
 634                error_report("setting detect-zeroes to unmap is not allowed "
 635                             "without setting discard operation to unmap");
 636                exit(EXIT_FAILURE);
 637            }
 638            break;
 639        case 'b':
 640            bindto = optarg;
 641            break;
 642        case 'p':
 643            port = optarg;
 644            break;
 645        case 'o':
 646                dev_offset = strtoll (optarg, &end, 0);
 647            if (*end) {
 648                error_report("Invalid offset `%s'", optarg);
 649                exit(EXIT_FAILURE);
 650            }
 651            if (dev_offset < 0) {
 652                error_report("Offset must be positive `%s'", optarg);
 653                exit(EXIT_FAILURE);
 654            }
 655            break;
 656        case 'l':
 657            if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
 658                sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
 659                                                  optarg, false);
 660                if (!sn_opts) {
 661                    error_report("Failed in parsing snapshot param `%s'",
 662                                 optarg);
 663                    exit(EXIT_FAILURE);
 664                }
 665            } else {
 666                sn_id_or_name = optarg;
 667            }
 668            /* fall through */
 669        case 'r':
 670            nbdflags |= NBD_FLAG_READ_ONLY;
 671            flags &= ~BDRV_O_RDWR;
 672            break;
 673        case 'P':
 674            partition = strtol(optarg, &end, 0);
 675            if (*end) {
 676                error_report("Invalid partition `%s'", optarg);
 677                exit(EXIT_FAILURE);
 678            }
 679            if (partition < 1 || partition > 8) {
 680                error_report("Invalid partition %d", partition);
 681                exit(EXIT_FAILURE);
 682            }
 683            break;
 684        case 'k':
 685            sockpath = optarg;
 686            if (sockpath[0] != '/') {
 687                error_report("socket path must be absolute");
 688                exit(EXIT_FAILURE);
 689            }
 690            break;
 691        case 'd':
 692            disconnect = true;
 693            break;
 694        case 'c':
 695            device = optarg;
 696            break;
 697        case 'e':
 698            shared = strtol(optarg, &end, 0);
 699            if (*end) {
 700                error_report("Invalid shared device number '%s'", optarg);
 701                exit(EXIT_FAILURE);
 702            }
 703            if (shared < 1) {
 704                error_report("Shared device number must be greater than 0");
 705                exit(EXIT_FAILURE);
 706            }
 707            break;
 708        case 'f':
 709            fmt = optarg;
 710            break;
 711        case 't':
 712            persistent = 1;
 713            break;
 714        case 'x':
 715            export_name = optarg;
 716            break;
 717        case 'D':
 718            export_description = optarg;
 719            break;
 720        case 'v':
 721            verbose = 1;
 722            break;
 723        case 'V':
 724            version(argv[0]);
 725            exit(0);
 726            break;
 727        case 'h':
 728            usage(argv[0]);
 729            exit(0);
 730            break;
 731        case '?':
 732            error_report("Try `%s --help' for more information.", argv[0]);
 733            exit(EXIT_FAILURE);
 734        case QEMU_NBD_OPT_OBJECT: {
 735            QemuOpts *opts;
 736            opts = qemu_opts_parse_noisily(&qemu_object_opts,
 737                                           optarg, true);
 738            if (!opts) {
 739                exit(EXIT_FAILURE);
 740            }
 741        }   break;
 742        case QEMU_NBD_OPT_TLSCREDS:
 743            tlscredsid = optarg;
 744            break;
 745        case QEMU_NBD_OPT_IMAGE_OPTS:
 746            imageOpts = true;
 747            break;
 748        case 'T':
 749            g_free(trace_file);
 750            trace_file = trace_opt_parse(optarg);
 751            break;
 752        case QEMU_NBD_OPT_FORK:
 753            fork_process = true;
 754            break;
 755        }
 756    }
 757
 758    if ((argc - optind) != 1) {
 759        error_report("Invalid number of arguments");
 760        error_printf("Try `%s --help' for more information.\n", argv[0]);
 761        exit(EXIT_FAILURE);
 762    }
 763
 764    if (qemu_opts_foreach(&qemu_object_opts,
 765                          user_creatable_add_opts_foreach,
 766                          NULL, NULL)) {
 767        exit(EXIT_FAILURE);
 768    }
 769
 770    if (!trace_init_backends()) {
 771        exit(1);
 772    }
 773    trace_init_file(trace_file);
 774    qemu_set_log(LOG_TRACE);
 775
 776    socket_activation = check_socket_activation();
 777    if (socket_activation == 0) {
 778        setup_address_and_port(&bindto, &port);
 779    } else {
 780        /* Using socket activation - check user didn't use -p etc. */
 781        const char *err_msg = socket_activation_validate_opts(device, sockpath,
 782                                                              bindto, port);
 783        if (err_msg != NULL) {
 784            error_report("%s", err_msg);
 785            exit(EXIT_FAILURE);
 786        }
 787
 788        /* qemu-nbd can only listen on a single socket.  */
 789        if (socket_activation > 1) {
 790            error_report("qemu-nbd does not support socket activation with %s > 1",
 791                         "LISTEN_FDS");
 792            exit(EXIT_FAILURE);
 793        }
 794    }
 795
 796    if (tlscredsid) {
 797        if (sockpath) {
 798            error_report("TLS is only supported with IPv4/IPv6");
 799            exit(EXIT_FAILURE);
 800        }
 801        if (device) {
 802            error_report("TLS is not supported with a host device");
 803            exit(EXIT_FAILURE);
 804        }
 805        if (!export_name) {
 806            /* Set the default NBD protocol export name, since
 807             * we *must* use new style protocol for TLS */
 808            export_name = "";
 809        }
 810        tlscreds = nbd_get_tls_creds(tlscredsid, &local_err);
 811        if (local_err) {
 812            error_report("Failed to get TLS creds %s",
 813                         error_get_pretty(local_err));
 814            exit(EXIT_FAILURE);
 815        }
 816    }
 817
 818    if (disconnect) {
 819        int nbdfd = open(argv[optind], O_RDWR);
 820        if (nbdfd < 0) {
 821            error_report("Cannot open %s: %s", argv[optind],
 822                         strerror(errno));
 823            exit(EXIT_FAILURE);
 824        }
 825        nbd_disconnect(nbdfd);
 826
 827        close(nbdfd);
 828
 829        printf("%s disconnected\n", argv[optind]);
 830
 831        return 0;
 832    }
 833
 834    if ((device && !verbose) || fork_process) {
 835        int stderr_fd[2];
 836        pid_t pid;
 837        int ret;
 838
 839        if (qemu_pipe(stderr_fd) < 0) {
 840            error_report("Error setting up communication pipe: %s",
 841                         strerror(errno));
 842            exit(EXIT_FAILURE);
 843        }
 844
 845        /* Now daemonize, but keep a communication channel open to
 846         * print errors and exit with the proper status code.
 847         */
 848        pid = fork();
 849        if (pid < 0) {
 850            error_report("Failed to fork: %s", strerror(errno));
 851            exit(EXIT_FAILURE);
 852        } else if (pid == 0) {
 853            close(stderr_fd[0]);
 854            ret = qemu_daemon(1, 0);
 855
 856            /* Temporarily redirect stderr to the parent's pipe...  */
 857            old_stderr = dup(STDERR_FILENO);
 858            dup2(stderr_fd[1], STDERR_FILENO);
 859            if (ret < 0) {
 860                error_report("Failed to daemonize: %s", strerror(errno));
 861                exit(EXIT_FAILURE);
 862            }
 863
 864            /* ... close the descriptor we inherited and go on.  */
 865            close(stderr_fd[1]);
 866        } else {
 867            bool errors = false;
 868            char *buf;
 869
 870            /* In the parent.  Print error messages from the child until
 871             * it closes the pipe.
 872             */
 873            close(stderr_fd[1]);
 874            buf = g_malloc(1024);
 875            while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
 876                errors = true;
 877                ret = qemu_write_full(STDERR_FILENO, buf, ret);
 878                if (ret < 0) {
 879                    exit(EXIT_FAILURE);
 880                }
 881            }
 882            if (ret < 0) {
 883                error_report("Cannot read from daemon: %s",
 884                             strerror(errno));
 885                exit(EXIT_FAILURE);
 886            }
 887
 888            /* Usually the daemon should not print any message.
 889             * Exit with zero status in that case.
 890             */
 891            exit(errors);
 892        }
 893    }
 894
 895    if (device != NULL && sockpath == NULL) {
 896        sockpath = g_malloc(128);
 897        snprintf(sockpath, 128, SOCKET_PATH, basename(device));
 898    }
 899
 900    server = qio_net_listener_new();
 901    if (socket_activation == 0) {
 902        saddr = nbd_build_socket_address(sockpath, bindto, port);
 903        if (qio_net_listener_open_sync(server, saddr, &local_err) < 0) {
 904            object_unref(OBJECT(server));
 905            error_report_err(local_err);
 906            exit(EXIT_FAILURE);
 907        }
 908    } else {
 909        size_t i;
 910        /* See comment in check_socket_activation above. */
 911        for (i = 0; i < socket_activation; i++) {
 912            QIOChannelSocket *sioc;
 913            sioc = qio_channel_socket_new_fd(FIRST_SOCKET_ACTIVATION_FD + i,
 914                                             &local_err);
 915            if (sioc == NULL) {
 916                object_unref(OBJECT(server));
 917                error_report("Failed to use socket activation: %s",
 918                             error_get_pretty(local_err));
 919                exit(EXIT_FAILURE);
 920            }
 921            qio_net_listener_add(server, sioc);
 922            object_unref(OBJECT(sioc));
 923        }
 924    }
 925
 926    if (qemu_init_main_loop(&local_err)) {
 927        error_report_err(local_err);
 928        exit(EXIT_FAILURE);
 929    }
 930    bdrv_init();
 931    atexit(bdrv_close_all);
 932
 933    srcpath = argv[optind];
 934    if (imageOpts) {
 935        QemuOpts *opts;
 936        if (fmt) {
 937            error_report("--image-opts and -f are mutually exclusive");
 938            exit(EXIT_FAILURE);
 939        }
 940        opts = qemu_opts_parse_noisily(&file_opts, srcpath, true);
 941        if (!opts) {
 942            qemu_opts_reset(&file_opts);
 943            exit(EXIT_FAILURE);
 944        }
 945        options = qemu_opts_to_qdict(opts, NULL);
 946        qemu_opts_reset(&file_opts);
 947        blk = blk_new_open(NULL, NULL, options, flags, &local_err);
 948    } else {
 949        if (fmt) {
 950            options = qdict_new();
 951            qdict_put_str(options, "driver", fmt);
 952        }
 953        blk = blk_new_open(srcpath, NULL, options, flags, &local_err);
 954    }
 955
 956    if (!blk) {
 957        error_reportf_err(local_err, "Failed to blk_new_open '%s': ",
 958                          argv[optind]);
 959        exit(EXIT_FAILURE);
 960    }
 961    bs = blk_bs(blk);
 962
 963    blk_set_enable_write_cache(blk, !writethrough);
 964
 965    if (sn_opts) {
 966        ret = bdrv_snapshot_load_tmp(bs,
 967                                     qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
 968                                     qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
 969                                     &local_err);
 970    } else if (sn_id_or_name) {
 971        ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
 972                                                   &local_err);
 973    }
 974    if (ret < 0) {
 975        error_reportf_err(local_err, "Failed to load snapshot: ");
 976        exit(EXIT_FAILURE);
 977    }
 978
 979    bs->detect_zeroes = detect_zeroes;
 980    fd_size = blk_getlength(blk);
 981    if (fd_size < 0) {
 982        error_report("Failed to determine the image length: %s",
 983                     strerror(-fd_size));
 984        exit(EXIT_FAILURE);
 985    }
 986
 987    if (dev_offset >= fd_size) {
 988        error_report("Offset (%lld) has to be smaller than the image size "
 989                     "(%lld)",
 990                     (long long int)dev_offset, (long long int)fd_size);
 991        exit(EXIT_FAILURE);
 992    }
 993    fd_size -= dev_offset;
 994
 995    if (partition != -1) {
 996        ret = find_partition(blk, partition, &dev_offset, &fd_size);
 997        if (ret < 0) {
 998            error_report("Could not find partition %d: %s", partition,
 999                         strerror(-ret));
1000            exit(EXIT_FAILURE);
1001        }
1002    }
1003
1004    exp = nbd_export_new(bs, dev_offset, fd_size, nbdflags, nbd_export_closed,
1005                         writethrough, NULL, &local_err);
1006    if (!exp) {
1007        error_report_err(local_err);
1008        exit(EXIT_FAILURE);
1009    }
1010    if (export_name) {
1011        nbd_export_set_name(exp, export_name);
1012        nbd_export_set_description(exp, export_description);
1013        newproto = true;
1014    } else if (export_description) {
1015        error_report("Export description requires an export name");
1016        exit(EXIT_FAILURE);
1017    }
1018
1019    if (device) {
1020        int ret;
1021
1022        ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
1023        if (ret != 0) {
1024            error_report("Failed to create client thread: %s", strerror(ret));
1025            exit(EXIT_FAILURE);
1026        }
1027    } else {
1028        /* Shut up GCC warnings.  */
1029        memset(&client_thread, 0, sizeof(client_thread));
1030    }
1031
1032    nbd_update_server_watch();
1033
1034    /* now when the initialization is (almost) complete, chdir("/")
1035     * to free any busy filesystems */
1036    if (chdir("/") < 0) {
1037        error_report("Could not chdir to root directory: %s",
1038                     strerror(errno));
1039        exit(EXIT_FAILURE);
1040    }
1041
1042    if (fork_process) {
1043        dup2(old_stderr, STDERR_FILENO);
1044        close(old_stderr);
1045    }
1046
1047    state = RUNNING;
1048    do {
1049        main_loop_wait(false);
1050        if (state == TERMINATE) {
1051            state = TERMINATING;
1052            nbd_export_close(exp);
1053            nbd_export_put(exp);
1054            exp = NULL;
1055        }
1056    } while (state != TERMINATED);
1057
1058    blk_unref(blk);
1059    if (sockpath) {
1060        unlink(sockpath);
1061    }
1062
1063    qemu_opts_del(sn_opts);
1064
1065    if (device) {
1066        void *ret;
1067        pthread_join(client_thread, &ret);
1068        exit(ret != NULL);
1069    } else {
1070        exit(EXIT_SUCCESS);
1071    }
1072}
1073