qemu-nbd.c

   1 /*
   2  *  Copyright (C) 2005  Anthony Liguori <anthony@codemonkey.ws>
   3  *
   4  *  Network Block Device
   5  *
   6  *  This program is free software; you can redistribute it and/or modify
   7  *  it under the terms of the GNU General Public License as published by
   8  *  the Free Software Foundation; under version 2 of the License.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  You should have received a copy of the GNU General Public License
  16  *  along with this program; if not, see <http://www.gnu.org/licenses/>.
  17  */
  18
  19 #include "qemu/osdep.h"
  20 #include <getopt.h>
  21 #include <libgen.h>
  22 #include <pthread.h>
  23
  24 #include "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 #ifdef __linux__
  47 #define HAVE_NBD_DEVICE 1
  48 #else
  49 #define HAVE_NBD_DEVICE 0
  50 #endif
  51
  52 #define SOCKET_PATH                "/var/lock/qemu-nbd-%s"
  53 #define QEMU_NBD_OPT_CACHE         256
  54 #define QEMU_NBD_OPT_AIO           257
  55 #define QEMU_NBD_OPT_DISCARD       258
  56 #define QEMU_NBD_OPT_DETECT_ZEROES 259
  57 #define QEMU_NBD_OPT_OBJECT        260
  58 #define QEMU_NBD_OPT_TLSCREDS      261
  59 #define QEMU_NBD_OPT_IMAGE_OPTS    262
  60 #define QEMU_NBD_OPT_FORK          263
  61
  62 #define MBR_SIZE 512
  63
  64 static NBDExport *exp;
  65 static int verbose;
  66 static char *srcpath;
  67 static SocketAddress *saddr;
  68 static int persistent = 0;
  69 static enum { RUNNING, TERMINATE, TERMINATING, TERMINATED } state;
  70 static int shared = 1;
  71 static int nb_fds;
  72 static QIONetListener *server;
  73 static QCryptoTLSCreds *tlscreds;
  74
  75 static void usage(const char *name)
  76 {
  77     (printf) (
  78 "Usage: %s [OPTIONS] FILE\n"
  79 "QEMU Disk Network Block Device Server\n"
  80 "\n"
  81 "  -h, --help                display this help and exit\n"
  82 "  -V, --version             output version information and exit\n"
  83 "\n"
  84 "Connection properties:\n"
  85 "  -p, --port=PORT           port to listen on (default `%d')\n"
  86 "  -b, --bind=IFACE          interface to bind to (default `0.0.0.0')\n"
  87 "  -k, --socket=PATH         path to the unix socket\n"
  88 "                            (default '"SOCKET_PATH"')\n"
  89 "  -e, --shared=NUM          device can be shared by NUM clients (default '1')\n"
  90 "  -t, --persistent          don't exit on the last connection\n"
  91 "  -v, --verbose             display extra debugging information\n"
  92 "  -x, --export-name=NAME    expose export by name (default is empty string)\n"
  93 "  -D, --description=TEXT    export a human-readable description\n"
  94 "\n"
  95 "Exposing part of the image:\n"
  96 "  -o, --offset=OFFSET       offset into the image\n"
  97 "  -P, --partition=NUM       only expose partition NUM\n"
  98 "\n"
  99 "General purpose options:\n"
 100 "  --object type,id=ID,...   define an object such as 'secret' for providing\n"
 101 "                            passwords and/or encryption keys\n"
 102 "  --tls-creds=ID            use id of an earlier --object to provide TLS\n"
 103 "  -T, --trace [[enable=]<pattern>][,events=<file>][,file=<file>]\n"
 104 "                            specify tracing options\n"
 105 "  --fork                    fork off the server process and exit the parent\n"
 106 "                            once the server is running\n"
 107 #if HAVE_NBD_DEVICE
 108 "\n"
 109 "Kernel NBD client support:\n"
 110 "  -c, --connect=DEV         connect FILE to the local NBD device DEV\n"
 111 "  -d, --disconnect          disconnect the specified device\n"
 112 #endif
 113 "\n"
 114 "Block device options:\n"
 115 "  -f, --format=FORMAT       set image format (raw, qcow2, ...)\n"
 116 "  -r, --read-only           export read-only\n"
 117 "  -s, --snapshot            use FILE as an external snapshot, create a temporary\n"
 118 "                            file with backing_file=FILE, redirect the write to\n"
 119 "                            the temporary one\n"
 120 "  -l, --load-snapshot=SNAPSHOT_PARAM\n"
 121 "                            load an internal snapshot inside FILE and export it\n"
 122 "                            as an read-only device, SNAPSHOT_PARAM format is\n"
 123 "                            'snapshot.id=[ID],snapshot.name=[NAME]', or\n"
 124 "                            '[ID_OR_NAME]'\n"
 125 "  -n, --nocache             disable host cache\n"
 126 "      --cache=MODE          set cache mode (none, writeback, ...)\n"
 127 "      --aio=MODE            set AIO mode (native or threads)\n"
 128 "      --discard=MODE        set discard mode (ignore, unmap)\n"
 129 "      --detect-zeroes=MODE  set detect-zeroes mode (off, on, unmap)\n"
 130 "      --image-opts          treat FILE as a full set of image options\n"
 131 "\n"
 132 QEMU_HELP_BOTTOM "\n"
 133     , name, NBD_DEFAULT_PORT, "DEVICE");
 134 }
 135
 136 static void version(const char *name)
 137 {
 138     printf(
 139 "%s " QEMU_FULL_VERSION "\n"
 140 "Written by Anthony Liguori.\n"
 141 "\n"
 142 QEMU_COPYRIGHT "\n"
 143 "This is free software; see the source for copying conditions.  There is NO\n"
 144 "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
 145     , name);
 146 }
 147
 148 struct partition_record
 149 {
 150     uint8_t bootable;
 151     uint8_t start_head;
 152     uint32_t start_cylinder;
 153     uint8_t start_sector;
 154     uint8_t system;
 155     uint8_t end_head;
 156     uint8_t end_cylinder;
 157     uint8_t end_sector;
 158     uint32_t start_sector_abs;
 159     uint32_t nb_sectors_abs;
 160 };
 161
 162 static void read_partition(uint8_t *p, struct partition_record *r)
 163 {
 164     r->bootable = p[0];
 165     r->start_head = p[1];
 166     r->start_cylinder = p[3] | ((p[2] << 2) & 0x0300);
 167     r->start_sector = p[2] & 0x3f;
 168     r->system = p[4];
 169     r->end_head = p[5];
 170     r->end_cylinder = p[7] | ((p[6] << 2) & 0x300);
 171     r->end_sector = p[6] & 0x3f;
 172
 173     r->start_sector_abs = ldl_le_p(p + 8);
 174     r->nb_sectors_abs   = ldl_le_p(p + 12);
 175 }
 176
 177 static int find_partition(BlockBackend *blk, int partition,
 178                           off_t *offset, off_t *size)
 179 {
 180     struct partition_record mbr[4];
 181     uint8_t data[MBR_SIZE];
 182     int i;
 183     int ext_partnum = 4;
 184     int ret;
 185
 186     ret = blk_pread(blk, 0, data, sizeof(data));
 187     if (ret < 0) {
 188         error_report("error while reading: %s", strerror(-ret));
 189         exit(EXIT_FAILURE);
 190     }
 191
 192     if (data[510] != 0x55 || data[511] != 0xaa) {
 193         return -EINVAL;
 194     }
 195
 196     for (i = 0; i < 4; i++) {
 197         read_partition(&data[446 + 16 * i], &mbr[i]);
 198
 199         if (!mbr[i].system || !mbr[i].nb_sectors_abs) {
 200             continue;
 201         }
 202
 203         if (mbr[i].system == 0xF || mbr[i].system == 0x5) {
 204             struct partition_record ext[4];
 205             uint8_t data1[MBR_SIZE];
 206             int j;
 207
 208             ret = blk_pread(blk, mbr[i].start_sector_abs * MBR_SIZE,
 209                             data1, sizeof(data1));
 210             if (ret < 0) {
 211                 error_report("error while reading: %s", strerror(-ret));
 212                 exit(EXIT_FAILURE);
 213             }
 214
 215             for (j = 0; j < 4; j++) {
 216                 read_partition(&data1[446 + 16 * j], &ext[j]);
 217                 if (!ext[j].system || !ext[j].nb_sectors_abs) {
 218                     continue;
 219                 }
 220
 221                 if ((ext_partnum + j + 1) == partition) {
 222                     *offset = (uint64_t)ext[j].start_sector_abs << 9;
 223                     *size = (uint64_t)ext[j].nb_sectors_abs << 9;
 224                     return 0;
 225                 }
 226             }
 227             ext_partnum += 4;
 228         } else if ((i + 1) == partition) {
 229             *offset = (uint64_t)mbr[i].start_sector_abs << 9;
 230             *size = (uint64_t)mbr[i].nb_sectors_abs << 9;
 231             return 0;
 232         }
 233     }
 234
 235     return -ENOENT;
 236 }
 237
 238 static void termsig_handler(int signum)
 239 {
 240     atomic_cmpxchg(&state, RUNNING, TERMINATE);
 241     qemu_notify_event();
 242 }
 243
 244
 245 #if HAVE_NBD_DEVICE
 246 static void *show_parts(void *arg)
 247 {
 248     char *device = arg;
 249     int nbd;
 250
 251     /* linux just needs an open() to trigger
 252      * the partition table update
 253      * but remember to load the module with max_part != 0 :
 254      *     modprobe nbd max_part=63
 255      */
 256     nbd = open(device, O_RDWR);
 257     if (nbd >= 0) {
 258         close(nbd);
 259     }
 260     return NULL;
 261 }
 262
 263 static void *nbd_client_thread(void *arg)
 264 {
 265     char *device = arg;
 266     NBDExportInfo info = { .request_sizes = false, };
 267     QIOChannelSocket *sioc;
 268     int fd;
 269     int ret;
 270     pthread_t show_parts_thread;
 271     Error *local_error = NULL;
 272
 273     sioc = qio_channel_socket_new();
 274     if (qio_channel_socket_connect_sync(sioc,
 275                                         saddr,
 276                                         &local_error) < 0) {
 277         error_report_err(local_error);
 278         goto out;
 279     }
 280
 281     ret = nbd_receive_negotiate(QIO_CHANNEL(sioc), NULL,
 282                                 NULL, NULL, NULL, &info, &local_error);
 283     if (ret < 0) {
 284         if (local_error) {
 285             error_report_err(local_error);
 286         }
 287         goto out_socket;
 288     }
 289
 290     fd = open(device, O_RDWR);
 291     if (fd < 0) {
 292         /* Linux-only, we can use %m in printf.  */
 293         error_report("Failed to open %s: %m", device);
 294         goto out_socket;
 295     }
 296
 297     ret = nbd_init(fd, sioc, &info, &local_error);
 298     if (ret < 0) {
 299         error_report_err(local_error);
 300         goto out_fd;
 301     }
 302
 303     /* update partition table */
 304     pthread_create(&show_parts_thread, NULL, show_parts, device);
 305
 306     if (verbose) {
 307         fprintf(stderr, "NBD device %s is now connected to %s\n",
 308                 device, srcpath);
 309     } else {
 310         /* Close stderr so that the qemu-nbd process exits.  */
 311         dup2(STDOUT_FILENO, STDERR_FILENO);
 312     }
 313
 314     ret = nbd_client(fd);
 315     if (ret) {
 316         goto out_fd;
 317     }
 318     close(fd);
 319     object_unref(OBJECT(sioc));
 320     kill(getpid(), SIGTERM);
 321     return (void *) EXIT_SUCCESS;
 322
 323 out_fd:
 324     close(fd);
 325 out_socket:
 326     object_unref(OBJECT(sioc));
 327 out:
 328     kill(getpid(), SIGTERM);
 329     return (void *) EXIT_FAILURE;
 330 }
 331 #endif /* HAVE_NBD_DEVICE */
 332
 333 static int nbd_can_accept(void)
 334 {
 335     return state == RUNNING && nb_fds < shared;
 336 }
 337
 338 static void nbd_export_closed(NBDExport *exp)
 339 {
 340     assert(state == TERMINATING);
 341     state = TERMINATED;
 342 }
 343
 344 static void nbd_update_server_watch(void);
 345
 346 static void nbd_client_closed(NBDClient *client, bool negotiated)
 347 {
 348     nb_fds--;
 349     if (negotiated && nb_fds == 0 && !persistent && state == RUNNING) {
 350         state = TERMINATE;
 351     }
 352     nbd_update_server_watch();
 353     nbd_client_put(client);
 354 }
 355
 356 static void nbd_accept(QIONetListener *listener, QIOChannelSocket *cioc,
 357                        gpointer opaque)
 358 {
 359     if (state >= TERMINATE) {
 360         return;
 361     }
 362
 363     nb_fds++;
 364     nbd_update_server_watch();
 365     nbd_client_new(cioc, tlscreds, NULL, nbd_client_closed);
 366 }
 367
 368 static void nbd_update_server_watch(void)
 369 {
 370     if (nbd_can_accept()) {
 371         qio_net_listener_set_client_func(server, nbd_accept, NULL, NULL);
 372     } else {
 373         qio_net_listener_set_client_func(server, NULL, NULL, NULL);
 374     }
 375 }
 376
 377
 378 static SocketAddress *nbd_build_socket_address(const char *sockpath,
 379                                                const char *bindto,
 380                                                const char *port)
 381 {
 382     SocketAddress *saddr;
 383
 384     saddr = g_new0(SocketAddress, 1);
 385     if (sockpath) {
 386         saddr->type = SOCKET_ADDRESS_TYPE_UNIX;
 387         saddr->u.q_unix.path = g_strdup(sockpath);
 388     } else {
 389         InetSocketAddress *inet;
 390         saddr->type = SOCKET_ADDRESS_TYPE_INET;
 391         inet = &saddr->u.inet;
 392         inet->host = g_strdup(bindto);
 393         if (port) {
 394             inet->port = g_strdup(port);
 395         } else  {
 396             inet->port = g_strdup_printf("%d", NBD_DEFAULT_PORT);
 397         }
 398     }
 399
 400     return saddr;
 401 }
 402
 403
 404 static QemuOptsList file_opts = {
 405     .name = "file",
 406     .implied_opt_name = "file",
 407     .head = QTAILQ_HEAD_INITIALIZER(file_opts.head),
 408     .desc = {
 409         /* no elements => accept any params */
 410         { /* end of list */ }
 411     },
 412 };
 413
 414 static QemuOptsList qemu_object_opts = {
 415     .name = "object",
 416     .implied_opt_name = "qom-type",
 417     .head = QTAILQ_HEAD_INITIALIZER(qemu_object_opts.head),
 418     .desc = {
 419         { }
 420     },
 421 };
 422
 423
 424
 425 static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, Error **errp)
 426 {
 427     Object *obj;
 428     QCryptoTLSCreds *creds;
 429
 430     obj = object_resolve_path_component(
 431         object_get_objects_root(), id);
 432     if (!obj) {
 433         error_setg(errp, "No TLS credentials with id '%s'",
 434                    id);
 435         return NULL;
 436     }
 437     creds = (QCryptoTLSCreds *)
 438         object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
 439     if (!creds) {
 440         error_setg(errp, "Object with id '%s' is not TLS credentials",
 441                    id);
 442         return NULL;
 443     }
 444
 445     if (creds->endpoint != QCRYPTO_TLS_CREDS_ENDPOINT_SERVER) {
 446         error_setg(errp,
 447                    "Expecting TLS credentials with a server endpoint");
 448         return NULL;
 449     }
 450     object_ref(obj);
 451     return creds;
 452 }
 453
 454 static void setup_address_and_port(const char **address, const char **port)
 455 {
 456     if (*address == NULL) {
 457         *address = "0.0.0.0";
 458     }
 459
 460     if (*port == NULL) {
 461         *port = stringify(NBD_DEFAULT_PORT);
 462     }
 463 }
 464
 465 /*
 466  * Check socket parameters compatibility when socket activation is used.
 467  */
 468 static const char *socket_activation_validate_opts(const char *device,
 469                                                    const char *sockpath,
 470                                                    const char *address,
 471                                                    const char *port)
 472 {
 473     if (device != NULL) {
 474         return "NBD device can't be set when using socket activation";
 475     }
 476
 477     if (sockpath != NULL) {
 478         return "Unix socket can't be set when using socket activation";
 479     }
 480
 481     if (address != NULL) {
 482         return "The interface can't be set when using socket activation";
 483     }
 484
 485     if (port != NULL) {
 486         return "TCP port number can't be set when using socket activation";
 487     }
 488
 489     return NULL;
 490 }
 491
 492 static void qemu_nbd_shutdown(void)
 493 {
 494     job_cancel_sync_all();
 495     bdrv_close_all();
 496 }
 497
 498 int main(int argc, char **argv)
 499 {
 500     BlockBackend *blk;
 501     BlockDriverState *bs;
 502     off_t dev_offset = 0;
 503     uint16_t nbdflags = 0;
 504     bool disconnect = false;
 505     const char *bindto = NULL;
 506     const char *port = NULL;
 507     char *sockpath = NULL;
 508     char *device = NULL;
 509     off_t fd_size;
 510     QemuOpts *sn_opts = NULL;
 511     const char *sn_id_or_name = NULL;
 512     const char *sopt = "hVb:o:p:rsnP:c:dvk:e:f:tl:x:T:D:";
 513     struct option lopt[] = {
 514         { "help", no_argument, NULL, 'h' },
 515         { "version", no_argument, NULL, 'V' },
 516         { "bind", required_argument, NULL, 'b' },
 517         { "port", required_argument, NULL, 'p' },
 518         { "socket", required_argument, NULL, 'k' },
 519         { "offset", required_argument, NULL, 'o' },
 520         { "read-only", no_argument, NULL, 'r' },
 521         { "partition", required_argument, NULL, 'P' },
 522         { "connect", required_argument, NULL, 'c' },
 523         { "disconnect", no_argument, NULL, 'd' },
 524         { "snapshot", no_argument, NULL, 's' },
 525         { "load-snapshot", required_argument, NULL, 'l' },
 526         { "nocache", no_argument, NULL, 'n' },
 527         { "cache", required_argument, NULL, QEMU_NBD_OPT_CACHE },
 528         { "aio", required_argument, NULL, QEMU_NBD_OPT_AIO },
 529         { "discard", required_argument, NULL, QEMU_NBD_OPT_DISCARD },
 530         { "detect-zeroes", required_argument, NULL,
 531           QEMU_NBD_OPT_DETECT_ZEROES },
 532         { "shared", required_argument, NULL, 'e' },
 533         { "format", required_argument, NULL, 'f' },
 534         { "persistent", no_argument, NULL, 't' },
 535         { "verbose", no_argument, NULL, 'v' },
 536         { "object", required_argument, NULL, QEMU_NBD_OPT_OBJECT },
 537         { "export-name", required_argument, NULL, 'x' },
 538         { "description", required_argument, NULL, 'D' },
 539         { "tls-creds", required_argument, NULL, QEMU_NBD_OPT_TLSCREDS },
 540         { "image-opts", no_argument, NULL, QEMU_NBD_OPT_IMAGE_OPTS },
 541         { "trace", required_argument, NULL, 'T' },
 542         { "fork", no_argument, NULL, QEMU_NBD_OPT_FORK },
 543         { NULL, 0, NULL, 0 }
 544     };
 545     int ch;
 546     int opt_ind = 0;
 547     char *end;
 548     int flags = BDRV_O_RDWR;
 549     int partition = -1;
 550     int ret = 0;
 551     bool seen_cache = false;
 552     bool seen_discard = false;
 553     bool seen_aio = false;
 554     pthread_t client_thread;
 555     const char *fmt = NULL;
 556     Error *local_err = NULL;
 557     BlockdevDetectZeroesOptions detect_zeroes = BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
 558     QDict *options = NULL;
 559     const char *export_name = ""; /* Default export name */
 560     const char *export_description = NULL;
 561     const char *tlscredsid = NULL;
 562     bool imageOpts = false;
 563     bool writethrough = true;
 564     char *trace_file = NULL;
 565     bool fork_process = false;
 566     int old_stderr = -1;
 567     unsigned socket_activation;
 568
 569     /* The client thread uses SIGTERM to interrupt the server.  A signal
 570      * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
 571      */
 572     struct sigaction sa_sigterm;
 573     memset(&sa_sigterm, 0, sizeof(sa_sigterm));
 574     sa_sigterm.sa_handler = termsig_handler;
 575     sigaction(SIGTERM, &sa_sigterm, NULL);
 576
 577 #ifdef CONFIG_POSIX
 578     signal(SIGPIPE, SIG_IGN);
 579 #endif
 580
 581     module_call_init(MODULE_INIT_TRACE);
 582     error_set_progname(argv[0]);
 583     qcrypto_init(&error_fatal);
 584
 585     module_call_init(MODULE_INIT_QOM);
 586     qemu_add_opts(&qemu_object_opts);
 587     qemu_add_opts(&qemu_trace_opts);
 588     qemu_init_exec_dir(argv[0]);
 589
 590     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 591         switch (ch) {
 592         case 's':
 593             flags |= BDRV_O_SNAPSHOT;
 594             break;
 595         case 'n':
 596             optarg = (char *) "none";
 597             /* fallthrough */
 598         case QEMU_NBD_OPT_CACHE:
 599             if (seen_cache) {
 600                 error_report("-n and --cache can only be specified once");
 601                 exit(EXIT_FAILURE);
 602             }
 603             seen_cache = true;
 604             if (bdrv_parse_cache_mode(optarg, &flags, &writethrough) == -1) {
 605                 error_report("Invalid cache mode `%s'", optarg);
 606                 exit(EXIT_FAILURE);
 607             }
 608             break;
 609         case QEMU_NBD_OPT_AIO:
 610             if (seen_aio) {
 611                 error_report("--aio can only be specified once");
 612                 exit(EXIT_FAILURE);
 613             }
 614             seen_aio = true;
 615             if (!strcmp(optarg, "native")) {
 616                 flags |= BDRV_O_NATIVE_AIO;
 617             } else if (!strcmp(optarg, "threads")) {
 618                 /* this is the default */
 619             } else {
 620                error_report("invalid aio mode `%s'", optarg);
 621                exit(EXIT_FAILURE);
 622             }
 623             break;
 624         case QEMU_NBD_OPT_DISCARD:
 625             if (seen_discard) {
 626                 error_report("--discard can only be specified once");
 627                 exit(EXIT_FAILURE);
 628             }
 629             seen_discard = true;
 630             if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
 631                 error_report("Invalid discard mode `%s'", optarg);
 632                 exit(EXIT_FAILURE);
 633             }
 634             break;
 635         case QEMU_NBD_OPT_DETECT_ZEROES:
 636             detect_zeroes =
 637                 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
 638                                 optarg,
 639                                 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
 640                                 &local_err);
 641             if (local_err) {
 642                 error_reportf_err(local_err,
 643                                   "Failed to parse detect_zeroes mode: ");
 644                 exit(EXIT_FAILURE);
 645             }
 646             if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
 647                 !(flags & BDRV_O_UNMAP)) {
 648                 error_report("setting detect-zeroes to unmap is not allowed "
 649                              "without setting discard operation to unmap");
 650                 exit(EXIT_FAILURE);
 651             }
 652             break;
 653         case 'b':
 654             bindto = optarg;
 655             break;
 656         case 'p':
 657             port = optarg;
 658             break;
 659         case 'o':
 660                 dev_offset = strtoll (optarg, &end, 0);
 661             if (*end) {
 662                 error_report("Invalid offset `%s'", optarg);
 663                 exit(EXIT_FAILURE);
 664             }
 665             if (dev_offset < 0) {
 666                 error_report("Offset must be positive `%s'", optarg);
 667                 exit(EXIT_FAILURE);
 668             }
 669             break;
 670         case 'l':
 671             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
 672                 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
 673                                                   optarg, false);
 674                 if (!sn_opts) {
 675                     error_report("Failed in parsing snapshot param `%s'",
 676                                  optarg);
 677                     exit(EXIT_FAILURE);
 678                 }
 679             } else {
 680                 sn_id_or_name = optarg;
 681             }
 682             /* fall through */
 683         case 'r':
 684             nbdflags |= NBD_FLAG_READ_ONLY;
 685             flags &= ~BDRV_O_RDWR;
 686             break;
 687         case 'P':
 688             partition = strtol(optarg, &end, 0);
 689             if (*end) {
 690                 error_report("Invalid partition `%s'", optarg);
 691                 exit(EXIT_FAILURE);
 692             }
 693             if (partition < 1 || partition > 8) {
 694                 error_report("Invalid partition %d", partition);
 695                 exit(EXIT_FAILURE);
 696             }
 697             break;
 698         case 'k':
 699             sockpath = optarg;
 700             if (sockpath[0] != '/') {
 701                 error_report("socket path must be absolute");
 702                 exit(EXIT_FAILURE);
 703             }
 704             break;
 705         case 'd':
 706             disconnect = true;
 707             break;
 708         case 'c':
 709             device = optarg;
 710             break;
 711         case 'e':
 712             shared = strtol(optarg, &end, 0);
 713             if (*end) {
 714                 error_report("Invalid shared device number '%s'", optarg);
 715                 exit(EXIT_FAILURE);
 716             }
 717             if (shared < 1) {
 718                 error_report("Shared device number must be greater than 0");
 719                 exit(EXIT_FAILURE);
 720             }
 721             break;
 722         case 'f':
 723             fmt = optarg;
 724             break;
 725         case 't':
 726             persistent = 1;
 727             break;
 728         case 'x':
 729             export_name = optarg;
 730             break;
 731         case 'D':
 732             export_description = optarg;
 733             break;
 734         case 'v':
 735             verbose = 1;
 736             break;
 737         case 'V':
 738             version(argv[0]);
 739             exit(0);
 740             break;
 741         case 'h':
 742             usage(argv[0]);
 743             exit(0);
 744             break;
 745         case '?':
 746             error_report("Try `%s --help' for more information.", argv[0]);
 747             exit(EXIT_FAILURE);
 748         case QEMU_NBD_OPT_OBJECT: {
 749             QemuOpts *opts;
 750             opts = qemu_opts_parse_noisily(&qemu_object_opts,
 751                                            optarg, true);
 752             if (!opts) {
 753                 exit(EXIT_FAILURE);
 754             }
 755         }   break;
 756         case QEMU_NBD_OPT_TLSCREDS:
 757             tlscredsid = optarg;
 758             break;
 759         case QEMU_NBD_OPT_IMAGE_OPTS:
 760             imageOpts = true;
 761             break;
 762         case 'T':
 763             g_free(trace_file);
 764             trace_file = trace_opt_parse(optarg);
 765             break;
 766         case QEMU_NBD_OPT_FORK:
 767             fork_process = true;
 768             break;
 769         }
 770     }
 771
 772     if ((argc - optind) != 1) {
 773         error_report("Invalid number of arguments");
 774         error_printf("Try `%s --help' for more information.\n", argv[0]);
 775         exit(EXIT_FAILURE);
 776     }
 777
 778     qemu_opts_foreach(&qemu_object_opts,
 779                       user_creatable_add_opts_foreach,
 780                       NULL, &error_fatal);
 781
 782     if (!trace_init_backends()) {
 783         exit(1);
 784     }
 785     trace_init_file(trace_file);
 786     qemu_set_log(LOG_TRACE);
 787
 788     socket_activation = check_socket_activation();
 789     if (socket_activation == 0) {
 790         setup_address_and_port(&bindto, &port);
 791     } else {
 792         /* Using socket activation - check user didn't use -p etc. */
 793         const char *err_msg = socket_activation_validate_opts(device, sockpath,
 794                                                               bindto, port);
 795         if (err_msg != NULL) {
 796             error_report("%s", err_msg);
 797             exit(EXIT_FAILURE);
 798         }
 799
 800         /* qemu-nbd can only listen on a single socket.  */
 801         if (socket_activation > 1) {
 802             error_report("qemu-nbd does not support socket activation with %s > 1",
 803                          "LISTEN_FDS");
 804             exit(EXIT_FAILURE);
 805         }
 806     }
 807
 808     if (tlscredsid) {
 809         if (sockpath) {
 810             error_report("TLS is only supported with IPv4/IPv6");
 811             exit(EXIT_FAILURE);
 812         }
 813         if (device) {
 814             error_report("TLS is not supported with a host device");
 815             exit(EXIT_FAILURE);
 816         }
 817         tlscreds = nbd_get_tls_creds(tlscredsid, &local_err);
 818         if (local_err) {
 819             error_report("Failed to get TLS creds %s",
 820                          error_get_pretty(local_err));
 821             exit(EXIT_FAILURE);
 822         }
 823     }
 824
 825 #if !HAVE_NBD_DEVICE
 826     if (disconnect || device) {
 827         error_report("Kernel /dev/nbdN support not available");
 828         exit(EXIT_FAILURE);
 829     }
 830 #else /* HAVE_NBD_DEVICE */
 831     if (disconnect) {
 832         int nbdfd = open(argv[optind], O_RDWR);
 833         if (nbdfd < 0) {
 834             error_report("Cannot open %s: %s", argv[optind],
 835                          strerror(errno));
 836             exit(EXIT_FAILURE);
 837         }
 838         nbd_disconnect(nbdfd);
 839
 840         close(nbdfd);
 841
 842         printf("%s disconnected\n", argv[optind]);
 843
 844         return 0;
 845     }
 846 #endif
 847
 848     if ((device && !verbose) || fork_process) {
 849         int stderr_fd[2];
 850         pid_t pid;
 851         int ret;
 852
 853         if (qemu_pipe(stderr_fd) < 0) {
 854             error_report("Error setting up communication pipe: %s",
 855                          strerror(errno));
 856             exit(EXIT_FAILURE);
 857         }
 858
 859         /* Now daemonize, but keep a communication channel open to
 860          * print errors and exit with the proper status code.
 861          */
 862         pid = fork();
 863         if (pid < 0) {
 864             error_report("Failed to fork: %s", strerror(errno));
 865             exit(EXIT_FAILURE);
 866         } else if (pid == 0) {
 867             close(stderr_fd[0]);
 868             ret = qemu_daemon(1, 0);
 869
 870             /* Temporarily redirect stderr to the parent's pipe...  */
 871             old_stderr = dup(STDERR_FILENO);
 872             dup2(stderr_fd[1], STDERR_FILENO);
 873             if (ret < 0) {
 874                 error_report("Failed to daemonize: %s", strerror(errno));
 875                 exit(EXIT_FAILURE);
 876             }
 877
 878             /* ... close the descriptor we inherited and go on.  */
 879             close(stderr_fd[1]);
 880         } else {
 881             bool errors = false;
 882             char *buf;
 883
 884             /* In the parent.  Print error messages from the child until
 885              * it closes the pipe.
 886              */
 887             close(stderr_fd[1]);
 888             buf = g_malloc(1024);
 889             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
 890                 errors = true;
 891                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
 892                 if (ret < 0) {
 893                     exit(EXIT_FAILURE);
 894                 }
 895             }
 896             if (ret < 0) {
 897                 error_report("Cannot read from daemon: %s",
 898                              strerror(errno));
 899                 exit(EXIT_FAILURE);
 900             }
 901
 902             /* Usually the daemon should not print any message.
 903              * Exit with zero status in that case.
 904              */
 905             exit(errors);
 906         }
 907     }
 908
 909     if (device != NULL && sockpath == NULL) {
 910         sockpath = g_malloc(128);
 911         snprintf(sockpath, 128, SOCKET_PATH, basename(device));
 912     }
 913
 914     server = qio_net_listener_new();
 915     if (socket_activation == 0) {
 916         saddr = nbd_build_socket_address(sockpath, bindto, port);
 917         if (qio_net_listener_open_sync(server, saddr, &local_err) < 0) {
 918             object_unref(OBJECT(server));
 919             error_report_err(local_err);
 920             exit(EXIT_FAILURE);
 921         }
 922     } else {
 923         size_t i;
 924         /* See comment in check_socket_activation above. */
 925         for (i = 0; i < socket_activation; i++) {
 926             QIOChannelSocket *sioc;
 927             sioc = qio_channel_socket_new_fd(FIRST_SOCKET_ACTIVATION_FD + i,
 928                                              &local_err);
 929             if (sioc == NULL) {
 930                 object_unref(OBJECT(server));
 931                 error_report("Failed to use socket activation: %s",
 932                              error_get_pretty(local_err));
 933                 exit(EXIT_FAILURE);
 934             }
 935             qio_net_listener_add(server, sioc);
 936             object_unref(OBJECT(sioc));
 937         }
 938     }
 939
 940     if (qemu_init_main_loop(&local_err)) {
 941         error_report_err(local_err);
 942         exit(EXIT_FAILURE);
 943     }
 944     bdrv_init();
 945     atexit(qemu_nbd_shutdown);
 946
 947     srcpath = argv[optind];
 948     if (imageOpts) {
 949         QemuOpts *opts;
 950         if (fmt) {
 951             error_report("--image-opts and -f are mutually exclusive");
 952             exit(EXIT_FAILURE);
 953         }
 954         opts = qemu_opts_parse_noisily(&file_opts, srcpath, true);
 955         if (!opts) {
 956             qemu_opts_reset(&file_opts);
 957             exit(EXIT_FAILURE);
 958         }
 959         options = qemu_opts_to_qdict(opts, NULL);
 960         qemu_opts_reset(&file_opts);
 961         blk = blk_new_open(NULL, NULL, options, flags, &local_err);
 962     } else {
 963         if (fmt) {
 964             options = qdict_new();
 965             qdict_put_str(options, "driver", fmt);
 966         }
 967         blk = blk_new_open(srcpath, NULL, options, flags, &local_err);
 968     }
 969
 970     if (!blk) {
 971         error_reportf_err(local_err, "Failed to blk_new_open '%s': ",
 972                           argv[optind]);
 973         exit(EXIT_FAILURE);
 974     }
 975     bs = blk_bs(blk);
 976
 977     blk_set_enable_write_cache(blk, !writethrough);
 978
 979     if (sn_opts) {
 980         ret = bdrv_snapshot_load_tmp(bs,
 981                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
 982                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
 983                                      &local_err);
 984     } else if (sn_id_or_name) {
 985         ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
 986                                                    &local_err);
 987     }
 988     if (ret < 0) {
 989         error_reportf_err(local_err, "Failed to load snapshot: ");
 990         exit(EXIT_FAILURE);
 991     }
 992
 993     bs->detect_zeroes = detect_zeroes;
 994     fd_size = blk_getlength(blk);
 995     if (fd_size < 0) {
 996         error_report("Failed to determine the image length: %s",
 997                      strerror(-fd_size));
 998         exit(EXIT_FAILURE);
 999     }
1000
1001     if (dev_offset >= fd_size) {
1002         error_report("Offset (%lld) has to be smaller than the image size "
1003                      "(%lld)",
1004                      (long long int)dev_offset, (long long int)fd_size);
1005         exit(EXIT_FAILURE);
1006     }
1007     fd_size -= dev_offset;
1008
1009     if (partition != -1) {
1010         ret = find_partition(blk, partition, &dev_offset, &fd_size);
1011         if (ret < 0) {
1012             error_report("Could not find partition %d: %s", partition,
1013                          strerror(-ret));
1014             exit(EXIT_FAILURE);
1015         }
1016     }
1017
1018     exp = nbd_export_new(bs, dev_offset, fd_size, nbdflags, nbd_export_closed,
1019                          writethrough, NULL, &error_fatal);
1020     nbd_export_set_name(exp, export_name);
1021     nbd_export_set_description(exp, export_description);
1022
1023     if (device) {
1024 #if HAVE_NBD_DEVICE
1025         int ret;
1026
1027         ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
1028         if (ret != 0) {
1029             error_report("Failed to create client thread: %s", strerror(ret));
1030             exit(EXIT_FAILURE);
1031         }
1032 #endif
1033     } else {
1034         /* Shut up GCC warnings.  */
1035         memset(&client_thread, 0, sizeof(client_thread));
1036     }
1037
1038     nbd_update_server_watch();
1039
1040     /* now when the initialization is (almost) complete, chdir("/")
1041      * to free any busy filesystems */
1042     if (chdir("/") < 0) {
1043         error_report("Could not chdir to root directory: %s",
1044                      strerror(errno));
1045         exit(EXIT_FAILURE);
1046     }
1047
1048     if (fork_process) {
1049         dup2(old_stderr, STDERR_FILENO);
1050         close(old_stderr);
1051     }
1052
1053     state = RUNNING;
1054     do {
1055         main_loop_wait(false);
1056         if (state == TERMINATE) {
1057             state = TERMINATING;
1058             nbd_export_close(exp);
1059             nbd_export_put(exp);
1060             exp = NULL;
1061         }
1062     } while (state != TERMINATED);
1063
1064     blk_unref(blk);
1065     if (sockpath) {
1066         unlink(sockpath);
1067     }
1068
1069     qemu_opts_del(sn_opts);
1070
1071     if (device) {
1072         void *ret;
1073         pthread_join(client_thread, &ret);
1074         exit(ret != NULL);
1075     } else {
1076         exit(EXIT_SUCCESS);
1077     }
1078 }