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-common.h"
  26 #include "qemu/cutils.h"
  27 #include "sysemu/block-backend.h"
  28 #include "block/block_int.h"
  29 #include "block/nbd.h"
  30 #include "qemu/main-loop.h"
  31 #include "qemu/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/util.h"
  38 #include "qapi/qmp/qstring.h"
  39 #include "qom/object_interfaces.h"
  40 #include "io/channel-socket.h"
  41 #include "crypto/init.h"
  42 #include "trace/control.h"
  43 #include "qemu-version.h"
  44
  45 #define SOCKET_PATH                "/var/lock/qemu-nbd-%s"
  46 #define QEMU_NBD_OPT_CACHE         256
  47 #define QEMU_NBD_OPT_AIO           257
  48 #define QEMU_NBD_OPT_DISCARD       258
  49 #define QEMU_NBD_OPT_DETECT_ZEROES 259
  50 #define QEMU_NBD_OPT_OBJECT        260
  51 #define QEMU_NBD_OPT_TLSCREDS      261
  52 #define QEMU_NBD_OPT_IMAGE_OPTS    262
  53 #define QEMU_NBD_OPT_FORK          263
  54
  55 #define MBR_SIZE 512
  56
  57 static NBDExport *exp;
  58 static bool newproto;
  59 static int verbose;
  60 static char *srcpath;
  61 static SocketAddress *saddr;
  62 static int persistent = 0;
  63 static enum { RUNNING, TERMINATE, TERMINATING, TERMINATED } state;
  64 static int shared = 1;
  65 static int nb_fds;
  66 static QIOChannelSocket *server_ioc;
  67 static int server_watch = -1;
  68 static QCryptoTLSCreds *tlscreds;
  69
  70 static 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"
 126 QEMU_HELP_BOTTOM "\n"
 127     , name, NBD_DEFAULT_PORT, "DEVICE");
 128 }
 129
 130 static void version(const char *name)
 131 {
 132     printf(
 133 "%s " QEMU_VERSION QEMU_PKGVERSION "\n"
 134 "Written by Anthony Liguori.\n"
 135 "\n"
 136 QEMU_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
 142 struct 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
 156 static 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
 171 static 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
 232 static void termsig_handler(int signum)
 233 {
 234     atomic_cmpxchg(&state, RUNNING, TERMINATE);
 235     qemu_notify_event();
 236 }
 237
 238
 239 static 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
 256 static 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
 316 out_fd:
 317     close(fd);
 318 out_socket:
 319     object_unref(OBJECT(sioc));
 320 out:
 321     kill(getpid(), SIGTERM);
 322     return (void *) EXIT_FAILURE;
 323 }
 324
 325 static int nbd_can_accept(void)
 326 {
 327     return state == RUNNING && nb_fds < shared;
 328 }
 329
 330 static void nbd_export_closed(NBDExport *exp)
 331 {
 332     assert(state == TERMINATING);
 333     state = TERMINATED;
 334 }
 335
 336 static void nbd_update_server_watch(void);
 337
 338 static 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
 348 static gboolean nbd_accept(QIOChannel *ioc, GIOCondition cond, gpointer opaque)
 349 {
 350     QIOChannelSocket *cioc;
 351
 352     cioc = qio_channel_socket_accept(QIO_CHANNEL_SOCKET(ioc),
 353                                      NULL);
 354     if (!cioc) {
 355         return TRUE;
 356     }
 357
 358     if (state >= TERMINATE) {
 359         object_unref(OBJECT(cioc));
 360         return TRUE;
 361     }
 362
 363     nb_fds++;
 364     nbd_update_server_watch();
 365     nbd_client_new(newproto ? NULL : exp, cioc,
 366                    tlscreds, NULL, nbd_client_closed);
 367     object_unref(OBJECT(cioc));
 368
 369     return TRUE;
 370 }
 371
 372 static void nbd_update_server_watch(void)
 373 {
 374     if (nbd_can_accept()) {
 375         if (server_watch == -1) {
 376             server_watch = qio_channel_add_watch(QIO_CHANNEL(server_ioc),
 377                                                  G_IO_IN,
 378                                                  nbd_accept,
 379                                                  NULL, NULL);
 380         }
 381     } else {
 382         if (server_watch != -1) {
 383             g_source_remove(server_watch);
 384             server_watch = -1;
 385         }
 386     }
 387 }
 388
 389
 390 static SocketAddress *nbd_build_socket_address(const char *sockpath,
 391                                                const char *bindto,
 392                                                const char *port)
 393 {
 394     SocketAddress *saddr;
 395
 396     saddr = g_new0(SocketAddress, 1);
 397     if (sockpath) {
 398         saddr->type = SOCKET_ADDRESS_TYPE_UNIX;
 399         saddr->u.q_unix.path = g_strdup(sockpath);
 400     } else {
 401         InetSocketAddress *inet;
 402         saddr->type = SOCKET_ADDRESS_TYPE_INET;
 403         inet = &saddr->u.inet;
 404         inet->host = g_strdup(bindto);
 405         if (port) {
 406             inet->port = g_strdup(port);
 407         } else  {
 408             inet->port = g_strdup_printf("%d", NBD_DEFAULT_PORT);
 409         }
 410     }
 411
 412     return saddr;
 413 }
 414
 415
 416 static QemuOptsList file_opts = {
 417     .name = "file",
 418     .implied_opt_name = "file",
 419     .head = QTAILQ_HEAD_INITIALIZER(file_opts.head),
 420     .desc = {
 421         /* no elements => accept any params */
 422         { /* end of list */ }
 423     },
 424 };
 425
 426 static QemuOptsList qemu_object_opts = {
 427     .name = "object",
 428     .implied_opt_name = "qom-type",
 429     .head = QTAILQ_HEAD_INITIALIZER(qemu_object_opts.head),
 430     .desc = {
 431         { }
 432     },
 433 };
 434
 435
 436
 437 static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, Error **errp)
 438 {
 439     Object *obj;
 440     QCryptoTLSCreds *creds;
 441
 442     obj = object_resolve_path_component(
 443         object_get_objects_root(), id);
 444     if (!obj) {
 445         error_setg(errp, "No TLS credentials with id '%s'",
 446                    id);
 447         return NULL;
 448     }
 449     creds = (QCryptoTLSCreds *)
 450         object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
 451     if (!creds) {
 452         error_setg(errp, "Object with id '%s' is not TLS credentials",
 453                    id);
 454         return NULL;
 455     }
 456
 457     if (creds->endpoint != QCRYPTO_TLS_CREDS_ENDPOINT_SERVER) {
 458         error_setg(errp,
 459                    "Expecting TLS credentials with a server endpoint");
 460         return NULL;
 461     }
 462     object_ref(obj);
 463     return creds;
 464 }
 465
 466 static void setup_address_and_port(const char **address, const char **port)
 467 {
 468     if (*address == NULL) {
 469         *address = "0.0.0.0";
 470     }
 471
 472     if (*port == NULL) {
 473         *port = stringify(NBD_DEFAULT_PORT);
 474     }
 475 }
 476
 477 /*
 478  * Check socket parameters compatibility when socket activation is used.
 479  */
 480 static const char *socket_activation_validate_opts(const char *device,
 481                                                    const char *sockpath,
 482                                                    const char *address,
 483                                                    const char *port)
 484 {
 485     if (device != NULL) {
 486         return "NBD device can't be set when using socket activation";
 487     }
 488
 489     if (sockpath != NULL) {
 490         return "Unix socket can't be set when using socket activation";
 491     }
 492
 493     if (address != NULL) {
 494         return "The interface can't be set when using socket activation";
 495     }
 496
 497     if (port != NULL) {
 498         return "TCP port number can't be set when using socket activation";
 499     }
 500
 501     return NULL;
 502 }
 503
 504 int main(int argc, char **argv)
 505 {
 506     BlockBackend *blk;
 507     BlockDriverState *bs;
 508     off_t dev_offset = 0;
 509     uint16_t nbdflags = 0;
 510     bool disconnect = false;
 511     const char *bindto = NULL;
 512     const char *port = NULL;
 513     char *sockpath = NULL;
 514     char *device = NULL;
 515     off_t fd_size;
 516     QemuOpts *sn_opts = NULL;
 517     const char *sn_id_or_name = NULL;
 518     const char *sopt = "hVb:o:p:rsnP:c:dvk:e:f:tl:x:T:D:";
 519     struct option lopt[] = {
 520         { "help", no_argument, NULL, 'h' },
 521         { "version", no_argument, NULL, 'V' },
 522         { "bind", required_argument, NULL, 'b' },
 523         { "port", required_argument, NULL, 'p' },
 524         { "socket", required_argument, NULL, 'k' },
 525         { "offset", required_argument, NULL, 'o' },
 526         { "read-only", no_argument, NULL, 'r' },
 527         { "partition", required_argument, NULL, 'P' },
 528         { "connect", required_argument, NULL, 'c' },
 529         { "disconnect", no_argument, NULL, 'd' },
 530         { "snapshot", no_argument, NULL, 's' },
 531         { "load-snapshot", required_argument, NULL, 'l' },
 532         { "nocache", no_argument, NULL, 'n' },
 533         { "cache", required_argument, NULL, QEMU_NBD_OPT_CACHE },
 534         { "aio", required_argument, NULL, QEMU_NBD_OPT_AIO },
 535         { "discard", required_argument, NULL, QEMU_NBD_OPT_DISCARD },
 536         { "detect-zeroes", required_argument, NULL,
 537           QEMU_NBD_OPT_DETECT_ZEROES },
 538         { "shared", required_argument, NULL, 'e' },
 539         { "format", required_argument, NULL, 'f' },
 540         { "persistent", no_argument, NULL, 't' },
 541         { "verbose", no_argument, NULL, 'v' },
 542         { "object", required_argument, NULL, QEMU_NBD_OPT_OBJECT },
 543         { "export-name", required_argument, NULL, 'x' },
 544         { "description", required_argument, NULL, 'D' },
 545         { "tls-creds", required_argument, NULL, QEMU_NBD_OPT_TLSCREDS },
 546         { "image-opts", no_argument, NULL, QEMU_NBD_OPT_IMAGE_OPTS },
 547         { "trace", required_argument, NULL, 'T' },
 548         { "fork", no_argument, NULL, QEMU_NBD_OPT_FORK },
 549         { NULL, 0, NULL, 0 }
 550     };
 551     int ch;
 552     int opt_ind = 0;
 553     char *end;
 554     int flags = BDRV_O_RDWR;
 555     int partition = -1;
 556     int ret = 0;
 557     bool seen_cache = false;
 558     bool seen_discard = false;
 559     bool seen_aio = false;
 560     pthread_t client_thread;
 561     const char *fmt = NULL;
 562     Error *local_err = NULL;
 563     BlockdevDetectZeroesOptions detect_zeroes = BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
 564     QDict *options = NULL;
 565     const char *export_name = NULL;
 566     const char *export_description = NULL;
 567     const char *tlscredsid = NULL;
 568     bool imageOpts = false;
 569     bool writethrough = true;
 570     char *trace_file = NULL;
 571     bool fork_process = false;
 572     int old_stderr = -1;
 573     unsigned socket_activation;
 574
 575     /* The client thread uses SIGTERM to interrupt the server.  A signal
 576      * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
 577      */
 578     struct sigaction sa_sigterm;
 579     memset(&sa_sigterm, 0, sizeof(sa_sigterm));
 580     sa_sigterm.sa_handler = termsig_handler;
 581     sigaction(SIGTERM, &sa_sigterm, NULL);
 582
 583 #ifdef CONFIG_POSIX
 584     signal(SIGPIPE, SIG_IGN);
 585 #endif
 586
 587     module_call_init(MODULE_INIT_TRACE);
 588     qcrypto_init(&error_fatal);
 589
 590     module_call_init(MODULE_INIT_QOM);
 591     qemu_add_opts(&qemu_object_opts);
 592     qemu_add_opts(&qemu_trace_opts);
 593     qemu_init_exec_dir(argv[0]);
 594
 595     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 596         switch (ch) {
 597         case 's':
 598             flags |= BDRV_O_SNAPSHOT;
 599             break;
 600         case 'n':
 601             optarg = (char *) "none";
 602             /* fallthrough */
 603         case QEMU_NBD_OPT_CACHE:
 604             if (seen_cache) {
 605                 error_report("-n and --cache can only be specified once");
 606                 exit(EXIT_FAILURE);
 607             }
 608             seen_cache = true;
 609             if (bdrv_parse_cache_mode(optarg, &flags, &writethrough) == -1) {
 610                 error_report("Invalid cache mode `%s'", optarg);
 611                 exit(EXIT_FAILURE);
 612             }
 613             break;
 614         case QEMU_NBD_OPT_AIO:
 615             if (seen_aio) {
 616                 error_report("--aio can only be specified once");
 617                 exit(EXIT_FAILURE);
 618             }
 619             seen_aio = true;
 620             if (!strcmp(optarg, "native")) {
 621                 flags |= BDRV_O_NATIVE_AIO;
 622             } else if (!strcmp(optarg, "threads")) {
 623                 /* this is the default */
 624             } else {
 625                error_report("invalid aio mode `%s'", optarg);
 626                exit(EXIT_FAILURE);
 627             }
 628             break;
 629         case QEMU_NBD_OPT_DISCARD:
 630             if (seen_discard) {
 631                 error_report("--discard can only be specified once");
 632                 exit(EXIT_FAILURE);
 633             }
 634             seen_discard = true;
 635             if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
 636                 error_report("Invalid discard mode `%s'", optarg);
 637                 exit(EXIT_FAILURE);
 638             }
 639             break;
 640         case QEMU_NBD_OPT_DETECT_ZEROES:
 641             detect_zeroes =
 642                 qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
 643                                 optarg,
 644                                 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
 645                                 &local_err);
 646             if (local_err) {
 647                 error_reportf_err(local_err,
 648                                   "Failed to parse detect_zeroes mode: ");
 649                 exit(EXIT_FAILURE);
 650             }
 651             if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
 652                 !(flags & BDRV_O_UNMAP)) {
 653                 error_report("setting detect-zeroes to unmap is not allowed "
 654                              "without setting discard operation to unmap");
 655                 exit(EXIT_FAILURE);
 656             }
 657             break;
 658         case 'b':
 659             bindto = optarg;
 660             break;
 661         case 'p':
 662             port = optarg;
 663             break;
 664         case 'o':
 665                 dev_offset = strtoll (optarg, &end, 0);
 666             if (*end) {
 667                 error_report("Invalid offset `%s'", optarg);
 668                 exit(EXIT_FAILURE);
 669             }
 670             if (dev_offset < 0) {
 671                 error_report("Offset must be positive `%s'", optarg);
 672                 exit(EXIT_FAILURE);
 673             }
 674             break;
 675         case 'l':
 676             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
 677                 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
 678                                                   optarg, false);
 679                 if (!sn_opts) {
 680                     error_report("Failed in parsing snapshot param `%s'",
 681                                  optarg);
 682                     exit(EXIT_FAILURE);
 683                 }
 684             } else {
 685                 sn_id_or_name = optarg;
 686             }
 687             /* fall through */
 688         case 'r':
 689             nbdflags |= NBD_FLAG_READ_ONLY;
 690             flags &= ~BDRV_O_RDWR;
 691             break;
 692         case 'P':
 693             partition = strtol(optarg, &end, 0);
 694             if (*end) {
 695                 error_report("Invalid partition `%s'", optarg);
 696                 exit(EXIT_FAILURE);
 697             }
 698             if (partition < 1 || partition > 8) {
 699                 error_report("Invalid partition %d", partition);
 700                 exit(EXIT_FAILURE);
 701             }
 702             break;
 703         case 'k':
 704             sockpath = optarg;
 705             if (sockpath[0] != '/') {
 706                 error_report("socket path must be absolute");
 707                 exit(EXIT_FAILURE);
 708             }
 709             break;
 710         case 'd':
 711             disconnect = true;
 712             break;
 713         case 'c':
 714             device = optarg;
 715             break;
 716         case 'e':
 717             shared = strtol(optarg, &end, 0);
 718             if (*end) {
 719                 error_report("Invalid shared device number '%s'", optarg);
 720                 exit(EXIT_FAILURE);
 721             }
 722             if (shared < 1) {
 723                 error_report("Shared device number must be greater than 0");
 724                 exit(EXIT_FAILURE);
 725             }
 726             break;
 727         case 'f':
 728             fmt = optarg;
 729             break;
 730         case 't':
 731             persistent = 1;
 732             break;
 733         case 'x':
 734             export_name = optarg;
 735             break;
 736         case 'D':
 737             export_description = optarg;
 738             break;
 739         case 'v':
 740             verbose = 1;
 741             break;
 742         case 'V':
 743             version(argv[0]);
 744             exit(0);
 745             break;
 746         case 'h':
 747             usage(argv[0]);
 748             exit(0);
 749             break;
 750         case '?':
 751             error_report("Try `%s --help' for more information.", argv[0]);
 752             exit(EXIT_FAILURE);
 753         case QEMU_NBD_OPT_OBJECT: {
 754             QemuOpts *opts;
 755             opts = qemu_opts_parse_noisily(&qemu_object_opts,
 756                                            optarg, true);
 757             if (!opts) {
 758                 exit(EXIT_FAILURE);
 759             }
 760         }   break;
 761         case QEMU_NBD_OPT_TLSCREDS:
 762             tlscredsid = optarg;
 763             break;
 764         case QEMU_NBD_OPT_IMAGE_OPTS:
 765             imageOpts = true;
 766             break;
 767         case 'T':
 768             g_free(trace_file);
 769             trace_file = trace_opt_parse(optarg);
 770             break;
 771         case QEMU_NBD_OPT_FORK:
 772             fork_process = true;
 773             break;
 774         }
 775     }
 776
 777     if ((argc - optind) != 1) {
 778         error_report("Invalid number of arguments");
 779         error_printf("Try `%s --help' for more information.\n", argv[0]);
 780         exit(EXIT_FAILURE);
 781     }
 782
 783     if (qemu_opts_foreach(&qemu_object_opts,
 784                           user_creatable_add_opts_foreach,
 785                           NULL, NULL)) {
 786         exit(EXIT_FAILURE);
 787     }
 788
 789     if (!trace_init_backends()) {
 790         exit(1);
 791     }
 792     trace_init_file(trace_file);
 793     qemu_set_log(LOG_TRACE);
 794
 795     socket_activation = check_socket_activation();
 796     if (socket_activation == 0) {
 797         setup_address_and_port(&bindto, &port);
 798     } else {
 799         /* Using socket activation - check user didn't use -p etc. */
 800         const char *err_msg = socket_activation_validate_opts(device, sockpath,
 801                                                               bindto, port);
 802         if (err_msg != NULL) {
 803             error_report("%s", err_msg);
 804             exit(EXIT_FAILURE);
 805         }
 806
 807         /* qemu-nbd can only listen on a single socket.  */
 808         if (socket_activation > 1) {
 809             error_report("qemu-nbd does not support socket activation with %s > 1",
 810                          "LISTEN_FDS");
 811             exit(EXIT_FAILURE);
 812         }
 813     }
 814
 815     if (tlscredsid) {
 816         if (sockpath) {
 817             error_report("TLS is only supported with IPv4/IPv6");
 818             exit(EXIT_FAILURE);
 819         }
 820         if (device) {
 821             error_report("TLS is not supported with a host device");
 822             exit(EXIT_FAILURE);
 823         }
 824         if (!export_name) {
 825             /* Set the default NBD protocol export name, since
 826              * we *must* use new style protocol for TLS */
 827             export_name = "";
 828         }
 829         tlscreds = nbd_get_tls_creds(tlscredsid, &local_err);
 830         if (local_err) {
 831             error_report("Failed to get TLS creds %s",
 832                          error_get_pretty(local_err));
 833             exit(EXIT_FAILURE);
 834         }
 835     }
 836
 837     if (disconnect) {
 838         int nbdfd = open(argv[optind], O_RDWR);
 839         if (nbdfd < 0) {
 840             error_report("Cannot open %s: %s", argv[optind],
 841                          strerror(errno));
 842             exit(EXIT_FAILURE);
 843         }
 844         nbd_disconnect(nbdfd);
 845
 846         close(nbdfd);
 847
 848         printf("%s disconnected\n", argv[optind]);
 849
 850         return 0;
 851     }
 852
 853     if ((device && !verbose) || fork_process) {
 854         int stderr_fd[2];
 855         pid_t pid;
 856         int ret;
 857
 858         if (qemu_pipe(stderr_fd) < 0) {
 859             error_report("Error setting up communication pipe: %s",
 860                          strerror(errno));
 861             exit(EXIT_FAILURE);
 862         }
 863
 864         /* Now daemonize, but keep a communication channel open to
 865          * print errors and exit with the proper status code.
 866          */
 867         pid = fork();
 868         if (pid < 0) {
 869             error_report("Failed to fork: %s", strerror(errno));
 870             exit(EXIT_FAILURE);
 871         } else if (pid == 0) {
 872             close(stderr_fd[0]);
 873             ret = qemu_daemon(1, 0);
 874
 875             /* Temporarily redirect stderr to the parent's pipe...  */
 876             old_stderr = dup(STDERR_FILENO);
 877             dup2(stderr_fd[1], STDERR_FILENO);
 878             if (ret < 0) {
 879                 error_report("Failed to daemonize: %s", strerror(errno));
 880                 exit(EXIT_FAILURE);
 881             }
 882
 883             /* ... close the descriptor we inherited and go on.  */
 884             close(stderr_fd[1]);
 885         } else {
 886             bool errors = false;
 887             char *buf;
 888
 889             /* In the parent.  Print error messages from the child until
 890              * it closes the pipe.
 891              */
 892             close(stderr_fd[1]);
 893             buf = g_malloc(1024);
 894             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
 895                 errors = true;
 896                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
 897                 if (ret < 0) {
 898                     exit(EXIT_FAILURE);
 899                 }
 900             }
 901             if (ret < 0) {
 902                 error_report("Cannot read from daemon: %s",
 903                              strerror(errno));
 904                 exit(EXIT_FAILURE);
 905             }
 906
 907             /* Usually the daemon should not print any message.
 908              * Exit with zero status in that case.
 909              */
 910             exit(errors);
 911         }
 912     }
 913
 914     if (device != NULL && sockpath == NULL) {
 915         sockpath = g_malloc(128);
 916         snprintf(sockpath, 128, SOCKET_PATH, basename(device));
 917     }
 918
 919     if (socket_activation == 0) {
 920         server_ioc = qio_channel_socket_new();
 921         saddr = nbd_build_socket_address(sockpath, bindto, port);
 922         if (qio_channel_socket_listen_sync(server_ioc, saddr, &local_err) < 0) {
 923             object_unref(OBJECT(server_ioc));
 924             error_report_err(local_err);
 925             return 1;
 926         }
 927     } else {
 928         /* See comment in check_socket_activation above. */
 929         assert(socket_activation == 1);
 930         server_ioc = qio_channel_socket_new_fd(FIRST_SOCKET_ACTIVATION_FD,
 931                                                &local_err);
 932         if (server_ioc == NULL) {
 933             error_report("Failed to use socket activation: %s",
 934                          error_get_pretty(local_err));
 935             exit(EXIT_FAILURE);
 936         }
 937     }
 938
 939     if (qemu_init_main_loop(&local_err)) {
 940         error_report_err(local_err);
 941         exit(EXIT_FAILURE);
 942     }
 943     bdrv_init();
 944     atexit(bdrv_close_all);
 945
 946     srcpath = argv[optind];
 947     if (imageOpts) {
 948         QemuOpts *opts;
 949         if (fmt) {
 950             error_report("--image-opts and -f are mutually exclusive");
 951             exit(EXIT_FAILURE);
 952         }
 953         opts = qemu_opts_parse_noisily(&file_opts, srcpath, true);
 954         if (!opts) {
 955             qemu_opts_reset(&file_opts);
 956             exit(EXIT_FAILURE);
 957         }
 958         options = qemu_opts_to_qdict(opts, NULL);
 959         qemu_opts_reset(&file_opts);
 960         blk = blk_new_open(NULL, NULL, options, flags, &local_err);
 961     } else {
 962         if (fmt) {
 963             options = qdict_new();
 964             qdict_put_str(options, "driver", fmt);
 965         }
 966         blk = blk_new_open(srcpath, NULL, options, flags, &local_err);
 967     }
 968
 969     if (!blk) {
 970         error_reportf_err(local_err, "Failed to blk_new_open '%s': ",
 971                           argv[optind]);
 972         exit(EXIT_FAILURE);
 973     }
 974     bs = blk_bs(blk);
 975
 976     blk_set_enable_write_cache(blk, !writethrough);
 977
 978     if (sn_opts) {
 979         ret = bdrv_snapshot_load_tmp(bs,
 980                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
 981                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
 982                                      &local_err);
 983     } else if (sn_id_or_name) {
 984         ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
 985                                                    &local_err);
 986     }
 987     if (ret < 0) {
 988         error_reportf_err(local_err, "Failed to load snapshot: ");
 989         exit(EXIT_FAILURE);
 990     }
 991
 992     bs->detect_zeroes = detect_zeroes;
 993     fd_size = blk_getlength(blk);
 994     if (fd_size < 0) {
 995         error_report("Failed to determine the image length: %s",
 996                      strerror(-fd_size));
 997         exit(EXIT_FAILURE);
 998     }
 999
1000     if (dev_offset >= fd_size) {
1001         error_report("Offset (%lld) has to be smaller than the image size "
1002                      "(%lld)",
1003                      (long long int)dev_offset, (long long int)fd_size);
1004         exit(EXIT_FAILURE);
1005     }
1006     fd_size -= dev_offset;
1007
1008     if (partition != -1) {
1009         ret = find_partition(blk, partition, &dev_offset, &fd_size);
1010         if (ret < 0) {
1011             error_report("Could not find partition %d: %s", partition,
1012                          strerror(-ret));
1013             exit(EXIT_FAILURE);
1014         }
1015     }
1016
1017     exp = nbd_export_new(bs, dev_offset, fd_size, nbdflags, nbd_export_closed,
1018                          writethrough, NULL, &local_err);
1019     if (!exp) {
1020         error_report_err(local_err);
1021         exit(EXIT_FAILURE);
1022     }
1023     if (export_name) {
1024         nbd_export_set_name(exp, export_name);
1025         nbd_export_set_description(exp, export_description);
1026         newproto = true;
1027     } else if (export_description) {
1028         error_report("Export description requires an export name");
1029         exit(EXIT_FAILURE);
1030     }
1031
1032     if (device) {
1033         int ret;
1034
1035         ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
1036         if (ret != 0) {
1037             error_report("Failed to create client thread: %s", strerror(ret));
1038             exit(EXIT_FAILURE);
1039         }
1040     } else {
1041         /* Shut up GCC warnings.  */
1042         memset(&client_thread, 0, sizeof(client_thread));
1043     }
1044
1045     nbd_update_server_watch();
1046
1047     /* now when the initialization is (almost) complete, chdir("/")
1048      * to free any busy filesystems */
1049     if (chdir("/") < 0) {
1050         error_report("Could not chdir to root directory: %s",
1051                      strerror(errno));
1052         exit(EXIT_FAILURE);
1053     }
1054
1055     if (fork_process) {
1056         dup2(old_stderr, STDERR_FILENO);
1057         close(old_stderr);
1058     }
1059
1060     state = RUNNING;
1061     do {
1062         main_loop_wait(false);
1063         if (state == TERMINATE) {
1064             state = TERMINATING;
1065             nbd_export_close(exp);
1066             nbd_export_put(exp);
1067             exp = NULL;
1068         }
1069     } while (state != TERMINATED);
1070
1071     blk_unref(blk);
1072     if (sockpath) {
1073         unlink(sockpath);
1074     }
1075
1076     qemu_opts_del(sn_opts);
1077
1078     if (device) {
1079         void *ret;
1080         pthread_join(client_thread, &ret);
1081         exit(ret != NULL);
1082     } else {
1083         exit(EXIT_SUCCESS);
1084     }
1085 }