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/qmp/qstring.h"
  38 #include "qom/object_interfaces.h"
  39 #include "io/channel-socket.h"
  40 #include "crypto/init.h"
  41 #include "trace/control.h"
  42 #include "qemu-version.h"
  43
  44 #define SOCKET_PATH                "/var/lock/qemu-nbd-%s"
  45 #define QEMU_NBD_OPT_CACHE         256
  46 #define QEMU_NBD_OPT_AIO           257
  47 #define QEMU_NBD_OPT_DISCARD       258
  48 #define QEMU_NBD_OPT_DETECT_ZEROES 259
  49 #define QEMU_NBD_OPT_OBJECT        260
  50 #define QEMU_NBD_OPT_TLSCREDS      261
  51 #define QEMU_NBD_OPT_IMAGE_OPTS    262
  52 #define QEMU_NBD_OPT_FORK          263
  53
  54 #define MBR_SIZE 512
  55
  56 static NBDExport *exp;
  57 static bool newproto;
  58 static int verbose;
  59 static char *srcpath;
  60 static SocketAddress *saddr;
  61 static int persistent = 0;
  62 static enum { RUNNING, TERMINATE, TERMINATING, TERMINATED } state;
  63 static int shared = 1;
  64 static int nb_fds;
  65 static QIOChannelSocket *server_ioc;
  66 static int server_watch = -1;
  67 static QCryptoTLSCreds *tlscreds;
  68
  69 static void usage(const char *name)
  70 {
  71     (printf) (
  72 "Usage: %s [OPTIONS] FILE\n"
  73 "QEMU Disk Network Block Device Server\n"
  74 "\n"
  75 "  -h, --help                display this help and exit\n"
  76 "  -V, --version             output version information and exit\n"
  77 "\n"
  78 "Connection properties:\n"
  79 "  -p, --port=PORT           port to listen on (default `%d')\n"
  80 "  -b, --bind=IFACE          interface to bind to (default `0.0.0.0')\n"
  81 "  -k, --socket=PATH         path to the unix socket\n"
  82 "                            (default '"SOCKET_PATH"')\n"
  83 "  -e, --shared=NUM          device can be shared by NUM clients (default '1')\n"
  84 "  -t, --persistent          don't exit on the last connection\n"
  85 "  -v, --verbose             display extra debugging information\n"
  86 "  -x, --export-name=NAME    expose export by name\n"
  87 "  -D, --description=TEXT    with -x, also export a human-readable description\n"
  88 "\n"
  89 "Exposing part of the image:\n"
  90 "  -o, --offset=OFFSET       offset into the image\n"
  91 "  -P, --partition=NUM       only expose partition NUM\n"
  92 "\n"
  93 "General purpose options:\n"
  94 "  --object type,id=ID,...   define an object such as 'secret' for providing\n"
  95 "                            passwords and/or encryption keys\n"
  96 "  -T, --trace [[enable=]<pattern>][,events=<file>][,file=<file>]\n"
  97 "                            specify tracing options\n"
  98 "  --fork                    fork off the server process and exit the parent\n"
  99 "                            once the server is running\n"
 100 #ifdef __linux__
 101 "Kernel NBD client support:\n"
 102 "  -c, --connect=DEV         connect FILE to the local NBD device DEV\n"
 103 "  -d, --disconnect          disconnect the specified device\n"
 104 "\n"
 105 #endif
 106 "\n"
 107 "Block device options:\n"
 108 "  -f, --format=FORMAT       set image format (raw, qcow2, ...)\n"
 109 "  -r, --read-only           export read-only\n"
 110 "  -s, --snapshot            use FILE as an external snapshot, create a temporary\n"
 111 "                            file with backing_file=FILE, redirect the write to\n"
 112 "                            the temporary one\n"
 113 "  -l, --load-snapshot=SNAPSHOT_PARAM\n"
 114 "                            load an internal snapshot inside FILE and export it\n"
 115 "                            as an read-only device, SNAPSHOT_PARAM format is\n"
 116 "                            'snapshot.id=[ID],snapshot.name=[NAME]', or\n"
 117 "                            '[ID_OR_NAME]'\n"
 118 "  -n, --nocache             disable host cache\n"
 119 "      --cache=MODE          set cache mode (none, writeback, ...)\n"
 120 "      --aio=MODE            set AIO mode (native or threads)\n"
 121 "      --discard=MODE        set discard mode (ignore, unmap)\n"
 122 "      --detect-zeroes=MODE  set detect-zeroes mode (off, on, unmap)\n"
 123 "      --image-opts          treat FILE as a full set of image options\n"
 124 "\n"
 125 QEMU_HELP_BOTTOM "\n"
 126     , name, NBD_DEFAULT_PORT, "DEVICE");
 127 }
 128
 129 static void version(const char *name)
 130 {
 131     printf(
 132 "%s " QEMU_VERSION QEMU_PKGVERSION "\n"
 133 "Written by Anthony Liguori.\n"
 134 "\n"
 135 QEMU_COPYRIGHT "\n"
 136 "This is free software; see the source for copying conditions.  There is NO\n"
 137 "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
 138     , name);
 139 }
 140
 141 struct partition_record
 142 {
 143     uint8_t bootable;
 144     uint8_t start_head;
 145     uint32_t start_cylinder;
 146     uint8_t start_sector;
 147     uint8_t system;
 148     uint8_t end_head;
 149     uint8_t end_cylinder;
 150     uint8_t end_sector;
 151     uint32_t start_sector_abs;
 152     uint32_t nb_sectors_abs;
 153 };
 154
 155 static void read_partition(uint8_t *p, struct partition_record *r)
 156 {
 157     r->bootable = p[0];
 158     r->start_head = p[1];
 159     r->start_cylinder = p[3] | ((p[2] << 2) & 0x0300);
 160     r->start_sector = p[2] & 0x3f;
 161     r->system = p[4];
 162     r->end_head = p[5];
 163     r->end_cylinder = p[7] | ((p[6] << 2) & 0x300);
 164     r->end_sector = p[6] & 0x3f;
 165
 166     r->start_sector_abs = ldl_le_p(p + 8);
 167     r->nb_sectors_abs   = ldl_le_p(p + 12);
 168 }
 169
 170 static int find_partition(BlockBackend *blk, int partition,
 171                           off_t *offset, off_t *size)
 172 {
 173     struct partition_record mbr[4];
 174     uint8_t data[MBR_SIZE];
 175     int i;
 176     int ext_partnum = 4;
 177     int ret;
 178
 179     ret = blk_pread(blk, 0, data, sizeof(data));
 180     if (ret < 0) {
 181         error_report("error while reading: %s", strerror(-ret));
 182         exit(EXIT_FAILURE);
 183     }
 184
 185     if (data[510] != 0x55 || data[511] != 0xaa) {
 186         return -EINVAL;
 187     }
 188
 189     for (i = 0; i < 4; i++) {
 190         read_partition(&data[446 + 16 * i], &mbr[i]);
 191
 192         if (!mbr[i].system || !mbr[i].nb_sectors_abs) {
 193             continue;
 194         }
 195
 196         if (mbr[i].system == 0xF || mbr[i].system == 0x5) {
 197             struct partition_record ext[4];
 198             uint8_t data1[MBR_SIZE];
 199             int j;
 200
 201             ret = blk_pread(blk, mbr[i].start_sector_abs * MBR_SIZE,
 202                             data1, sizeof(data1));
 203             if (ret < 0) {
 204                 error_report("error while reading: %s", strerror(-ret));
 205                 exit(EXIT_FAILURE);
 206             }
 207
 208             for (j = 0; j < 4; j++) {
 209                 read_partition(&data1[446 + 16 * j], &ext[j]);
 210                 if (!ext[j].system || !ext[j].nb_sectors_abs) {
 211                     continue;
 212                 }
 213
 214                 if ((ext_partnum + j + 1) == partition) {
 215                     *offset = (uint64_t)ext[j].start_sector_abs << 9;
 216                     *size = (uint64_t)ext[j].nb_sectors_abs << 9;
 217                     return 0;
 218                 }
 219             }
 220             ext_partnum += 4;
 221         } else if ((i + 1) == partition) {
 222             *offset = (uint64_t)mbr[i].start_sector_abs << 9;
 223             *size = (uint64_t)mbr[i].nb_sectors_abs << 9;
 224             return 0;
 225         }
 226     }
 227
 228     return -ENOENT;
 229 }
 230
 231 static void termsig_handler(int signum)
 232 {
 233     atomic_cmpxchg(&state, RUNNING, TERMINATE);
 234     qemu_notify_event();
 235 }
 236
 237
 238 static void *show_parts(void *arg)
 239 {
 240     char *device = arg;
 241     int nbd;
 242
 243     /* linux just needs an open() to trigger
 244      * the partition table update
 245      * but remember to load the module with max_part != 0 :
 246      *     modprobe nbd max_part=63
 247      */
 248     nbd = open(device, O_RDWR);
 249     if (nbd >= 0) {
 250         close(nbd);
 251     }
 252     return NULL;
 253 }
 254
 255 static void *nbd_client_thread(void *arg)
 256 {
 257     char *device = arg;
 258     NBDExportInfo info = { .request_sizes = false, };
 259     QIOChannelSocket *sioc;
 260     int fd;
 261     int ret;
 262     pthread_t show_parts_thread;
 263     Error *local_error = NULL;
 264
 265     sioc = qio_channel_socket_new();
 266     if (qio_channel_socket_connect_sync(sioc,
 267                                         saddr,
 268                                         &local_error) < 0) {
 269         error_report_err(local_error);
 270         goto out;
 271     }
 272
 273     ret = nbd_receive_negotiate(QIO_CHANNEL(sioc), NULL,
 274                                 NULL, NULL, NULL, &info, &local_error);
 275     if (ret < 0) {
 276         if (local_error) {
 277             error_report_err(local_error);
 278         }
 279         goto out_socket;
 280     }
 281
 282     fd = open(device, O_RDWR);
 283     if (fd < 0) {
 284         /* Linux-only, we can use %m in printf.  */
 285         error_report("Failed to open %s: %m", device);
 286         goto out_socket;
 287     }
 288
 289     ret = nbd_init(fd, sioc, &info, &local_error);
 290     if (ret < 0) {
 291         error_report_err(local_error);
 292         goto out_fd;
 293     }
 294
 295     /* update partition table */
 296     pthread_create(&show_parts_thread, NULL, show_parts, device);
 297
 298     if (verbose) {
 299         fprintf(stderr, "NBD device %s is now connected to %s\n",
 300                 device, srcpath);
 301     } else {
 302         /* Close stderr so that the qemu-nbd process exits.  */
 303         dup2(STDOUT_FILENO, STDERR_FILENO);
 304     }
 305
 306     ret = nbd_client(fd);
 307     if (ret) {
 308         goto out_fd;
 309     }
 310     close(fd);
 311     object_unref(OBJECT(sioc));
 312     kill(getpid(), SIGTERM);
 313     return (void *) EXIT_SUCCESS;
 314
 315 out_fd:
 316     close(fd);
 317 out_socket:
 318     object_unref(OBJECT(sioc));
 319 out:
 320     kill(getpid(), SIGTERM);
 321     return (void *) EXIT_FAILURE;
 322 }
 323
 324 static int nbd_can_accept(void)
 325 {
 326     return state == RUNNING && nb_fds < shared;
 327 }
 328
 329 static void nbd_export_closed(NBDExport *exp)
 330 {
 331     assert(state == TERMINATING);
 332     state = TERMINATED;
 333 }
 334
 335 static void nbd_update_server_watch(void);
 336
 337 static void nbd_client_closed(NBDClient *client, bool negotiated)
 338 {
 339     nb_fds--;
 340     if (negotiated && nb_fds == 0 && !persistent && state == RUNNING) {
 341         state = TERMINATE;
 342     }
 343     nbd_update_server_watch();
 344     nbd_client_put(client);
 345 }
 346
 347 static gboolean nbd_accept(QIOChannel *ioc, GIOCondition cond, gpointer opaque)
 348 {
 349     QIOChannelSocket *cioc;
 350
 351     cioc = qio_channel_socket_accept(QIO_CHANNEL_SOCKET(ioc),
 352                                      NULL);
 353     if (!cioc) {
 354         return TRUE;
 355     }
 356
 357     if (state >= TERMINATE) {
 358         object_unref(OBJECT(cioc));
 359         return TRUE;
 360     }
 361
 362     nb_fds++;
 363     nbd_update_server_watch();
 364     nbd_client_new(newproto ? NULL : exp, cioc,
 365                    tlscreds, NULL, nbd_client_closed);
 366     object_unref(OBJECT(cioc));
 367
 368     return TRUE;
 369 }
 370
 371 static void nbd_update_server_watch(void)
 372 {
 373     if (nbd_can_accept()) {
 374         if (server_watch == -1) {
 375             server_watch = qio_channel_add_watch(QIO_CHANNEL(server_ioc),
 376                                                  G_IO_IN,
 377                                                  nbd_accept,
 378                                                  NULL, NULL);
 379         }
 380     } else {
 381         if (server_watch != -1) {
 382             g_source_remove(server_watch);
 383             server_watch = -1;
 384         }
 385     }
 386 }
 387
 388
 389 static SocketAddress *nbd_build_socket_address(const char *sockpath,
 390                                                const char *bindto,
 391                                                const char *port)
 392 {
 393     SocketAddress *saddr;
 394
 395     saddr = g_new0(SocketAddress, 1);
 396     if (sockpath) {
 397         saddr->type = SOCKET_ADDRESS_TYPE_UNIX;
 398         saddr->u.q_unix.path = g_strdup(sockpath);
 399     } else {
 400         InetSocketAddress *inet;
 401         saddr->type = SOCKET_ADDRESS_TYPE_INET;
 402         inet = &saddr->u.inet;
 403         inet->host = g_strdup(bindto);
 404         if (port) {
 405             inet->port = g_strdup(port);
 406         } else  {
 407             inet->port = g_strdup_printf("%d", NBD_DEFAULT_PORT);
 408         }
 409     }
 410
 411     return saddr;
 412 }
 413
 414
 415 static QemuOptsList file_opts = {
 416     .name = "file",
 417     .implied_opt_name = "file",
 418     .head = QTAILQ_HEAD_INITIALIZER(file_opts.head),
 419     .desc = {
 420         /* no elements => accept any params */
 421         { /* end of list */ }
 422     },
 423 };
 424
 425 static QemuOptsList qemu_object_opts = {
 426     .name = "object",
 427     .implied_opt_name = "qom-type",
 428     .head = QTAILQ_HEAD_INITIALIZER(qemu_object_opts.head),
 429     .desc = {
 430         { }
 431     },
 432 };
 433
 434
 435
 436 static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, Error **errp)
 437 {
 438     Object *obj;
 439     QCryptoTLSCreds *creds;
 440
 441     obj = object_resolve_path_component(
 442         object_get_objects_root(), id);
 443     if (!obj) {
 444         error_setg(errp, "No TLS credentials with id '%s'",
 445                    id);
 446         return NULL;
 447     }
 448     creds = (QCryptoTLSCreds *)
 449         object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
 450     if (!creds) {
 451         error_setg(errp, "Object with id '%s' is not TLS credentials",
 452                    id);
 453         return NULL;
 454     }
 455
 456     if (creds->endpoint != QCRYPTO_TLS_CREDS_ENDPOINT_SERVER) {
 457         error_setg(errp,
 458                    "Expecting TLS credentials with a server endpoint");
 459         return NULL;
 460     }
 461     object_ref(obj);
 462     return creds;
 463 }
 464
 465 static void setup_address_and_port(const char **address, const char **port)
 466 {
 467     if (*address == NULL) {
 468         *address = "0.0.0.0";
 469     }
 470
 471     if (*port == NULL) {
 472         *port = stringify(NBD_DEFAULT_PORT);
 473     }
 474 }
 475
 476 /*
 477  * Check socket parameters compatibility when socket activation is used.
 478  */
 479 static const char *socket_activation_validate_opts(const char *device,
 480                                                    const char *sockpath,
 481                                                    const char *address,
 482                                                    const char *port)
 483 {
 484     if (device != NULL) {
 485         return "NBD device can't be set when using socket activation";
 486     }
 487
 488     if (sockpath != NULL) {
 489         return "Unix socket can't be set when using socket activation";
 490     }
 491
 492     if (address != NULL) {
 493         return "The interface can't be set when using socket activation";
 494     }
 495
 496     if (port != NULL) {
 497         return "TCP port number can't be set when using socket activation";
 498     }
 499
 500     return NULL;
 501 }
 502
 503 int main(int argc, char **argv)
 504 {
 505     BlockBackend *blk;
 506     BlockDriverState *bs;
 507     off_t dev_offset = 0;
 508     uint16_t nbdflags = 0;
 509     bool disconnect = false;
 510     const char *bindto = NULL;
 511     const char *port = NULL;
 512     char *sockpath = NULL;
 513     char *device = NULL;
 514     off_t fd_size;
 515     QemuOpts *sn_opts = NULL;
 516     const char *sn_id_or_name = NULL;
 517     const char *sopt = "hVb:o:p:rsnP:c:dvk:e:f:tl:x:T:D:";
 518     struct option lopt[] = {
 519         { "help", no_argument, NULL, 'h' },
 520         { "version", no_argument, NULL, 'V' },
 521         { "bind", required_argument, NULL, 'b' },
 522         { "port", required_argument, NULL, 'p' },
 523         { "socket", required_argument, NULL, 'k' },
 524         { "offset", required_argument, NULL, 'o' },
 525         { "read-only", no_argument, NULL, 'r' },
 526         { "partition", required_argument, NULL, 'P' },
 527         { "connect", required_argument, NULL, 'c' },
 528         { "disconnect", no_argument, NULL, 'd' },
 529         { "snapshot", no_argument, NULL, 's' },
 530         { "load-snapshot", required_argument, NULL, 'l' },
 531         { "nocache", no_argument, NULL, 'n' },
 532         { "cache", required_argument, NULL, QEMU_NBD_OPT_CACHE },
 533         { "aio", required_argument, NULL, QEMU_NBD_OPT_AIO },
 534         { "discard", required_argument, NULL, QEMU_NBD_OPT_DISCARD },
 535         { "detect-zeroes", required_argument, NULL,
 536           QEMU_NBD_OPT_DETECT_ZEROES },
 537         { "shared", required_argument, NULL, 'e' },
 538         { "format", required_argument, NULL, 'f' },
 539         { "persistent", no_argument, NULL, 't' },
 540         { "verbose", no_argument, NULL, 'v' },
 541         { "object", required_argument, NULL, QEMU_NBD_OPT_OBJECT },
 542         { "export-name", required_argument, NULL, 'x' },
 543         { "description", required_argument, NULL, 'D' },
 544         { "tls-creds", required_argument, NULL, QEMU_NBD_OPT_TLSCREDS },
 545         { "image-opts", no_argument, NULL, QEMU_NBD_OPT_IMAGE_OPTS },
 546         { "trace", required_argument, NULL, 'T' },
 547         { "fork", no_argument, NULL, QEMU_NBD_OPT_FORK },
 548         { NULL, 0, NULL, 0 }
 549     };
 550     int ch;
 551     int opt_ind = 0;
 552     char *end;
 553     int flags = BDRV_O_RDWR;
 554     int partition = -1;
 555     int ret = 0;
 556     bool seen_cache = false;
 557     bool seen_discard = false;
 558     bool seen_aio = false;
 559     pthread_t client_thread;
 560     const char *fmt = NULL;
 561     Error *local_err = NULL;
 562     BlockdevDetectZeroesOptions detect_zeroes = BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
 563     QDict *options = NULL;
 564     const char *export_name = NULL;
 565     const char *export_description = NULL;
 566     const char *tlscredsid = NULL;
 567     bool imageOpts = false;
 568     bool writethrough = true;
 569     char *trace_file = NULL;
 570     bool fork_process = false;
 571     int old_stderr = -1;
 572     unsigned socket_activation;
 573
 574     /* The client thread uses SIGTERM to interrupt the server.  A signal
 575      * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
 576      */
 577     struct sigaction sa_sigterm;
 578     memset(&sa_sigterm, 0, sizeof(sa_sigterm));
 579     sa_sigterm.sa_handler = termsig_handler;
 580     sigaction(SIGTERM, &sa_sigterm, NULL);
 581
 582 #ifdef CONFIG_POSIX
 583     signal(SIGPIPE, SIG_IGN);
 584 #endif
 585
 586     module_call_init(MODULE_INIT_TRACE);
 587     qcrypto_init(&error_fatal);
 588
 589     module_call_init(MODULE_INIT_QOM);
 590     qemu_add_opts(&qemu_object_opts);
 591     qemu_add_opts(&qemu_trace_opts);
 592     qemu_init_exec_dir(argv[0]);
 593
 594     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 595         switch (ch) {
 596         case 's':
 597             flags |= BDRV_O_SNAPSHOT;
 598             break;
 599         case 'n':
 600             optarg = (char *) "none";
 601             /* fallthrough */
 602         case QEMU_NBD_OPT_CACHE:
 603             if (seen_cache) {
 604                 error_report("-n and --cache can only be specified once");
 605                 exit(EXIT_FAILURE);
 606             }
 607             seen_cache = true;
 608             if (bdrv_parse_cache_mode(optarg, &flags, &writethrough) == -1) {
 609                 error_report("Invalid cache mode `%s'", optarg);
 610                 exit(EXIT_FAILURE);
 611             }
 612             break;
 613         case QEMU_NBD_OPT_AIO:
 614             if (seen_aio) {
 615                 error_report("--aio can only be specified once");
 616                 exit(EXIT_FAILURE);
 617             }
 618             seen_aio = true;
 619             if (!strcmp(optarg, "native")) {
 620                 flags |= BDRV_O_NATIVE_AIO;
 621             } else if (!strcmp(optarg, "threads")) {
 622                 /* this is the default */
 623             } else {
 624                error_report("invalid aio mode `%s'", optarg);
 625                exit(EXIT_FAILURE);
 626             }
 627             break;
 628         case QEMU_NBD_OPT_DISCARD:
 629             if (seen_discard) {
 630                 error_report("--discard can only be specified once");
 631                 exit(EXIT_FAILURE);
 632             }
 633             seen_discard = true;
 634             if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
 635                 error_report("Invalid discard mode `%s'", optarg);
 636                 exit(EXIT_FAILURE);
 637             }
 638             break;
 639         case QEMU_NBD_OPT_DETECT_ZEROES:
 640             detect_zeroes =
 641                 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
 642                                 optarg,
 643                                 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
 644                                 &local_err);
 645             if (local_err) {
 646                 error_reportf_err(local_err,
 647                                   "Failed to parse detect_zeroes mode: ");
 648                 exit(EXIT_FAILURE);
 649             }
 650             if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
 651                 !(flags & BDRV_O_UNMAP)) {
 652                 error_report("setting detect-zeroes to unmap is not allowed "
 653                              "without setting discard operation to unmap");
 654                 exit(EXIT_FAILURE);
 655             }
 656             break;
 657         case 'b':
 658             bindto = optarg;
 659             break;
 660         case 'p':
 661             port = optarg;
 662             break;
 663         case 'o':
 664                 dev_offset = strtoll (optarg, &end, 0);
 665             if (*end) {
 666                 error_report("Invalid offset `%s'", optarg);
 667                 exit(EXIT_FAILURE);
 668             }
 669             if (dev_offset < 0) {
 670                 error_report("Offset must be positive `%s'", optarg);
 671                 exit(EXIT_FAILURE);
 672             }
 673             break;
 674         case 'l':
 675             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
 676                 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
 677                                                   optarg, false);
 678                 if (!sn_opts) {
 679                     error_report("Failed in parsing snapshot param `%s'",
 680                                  optarg);
 681                     exit(EXIT_FAILURE);
 682                 }
 683             } else {
 684                 sn_id_or_name = optarg;
 685             }
 686             /* fall through */
 687         case 'r':
 688             nbdflags |= NBD_FLAG_READ_ONLY;
 689             flags &= ~BDRV_O_RDWR;
 690             break;
 691         case 'P':
 692             partition = strtol(optarg, &end, 0);
 693             if (*end) {
 694                 error_report("Invalid partition `%s'", optarg);
 695                 exit(EXIT_FAILURE);
 696             }
 697             if (partition < 1 || partition > 8) {
 698                 error_report("Invalid partition %d", partition);
 699                 exit(EXIT_FAILURE);
 700             }
 701             break;
 702         case 'k':
 703             sockpath = optarg;
 704             if (sockpath[0] != '/') {
 705                 error_report("socket path must be absolute");
 706                 exit(EXIT_FAILURE);
 707             }
 708             break;
 709         case 'd':
 710             disconnect = true;
 711             break;
 712         case 'c':
 713             device = optarg;
 714             break;
 715         case 'e':
 716             shared = strtol(optarg, &end, 0);
 717             if (*end) {
 718                 error_report("Invalid shared device number '%s'", optarg);
 719                 exit(EXIT_FAILURE);
 720             }
 721             if (shared < 1) {
 722                 error_report("Shared device number must be greater than 0");
 723                 exit(EXIT_FAILURE);
 724             }
 725             break;
 726         case 'f':
 727             fmt = optarg;
 728             break;
 729         case 't':
 730             persistent = 1;
 731             break;
 732         case 'x':
 733             export_name = optarg;
 734             break;
 735         case 'D':
 736             export_description = optarg;
 737             break;
 738         case 'v':
 739             verbose = 1;
 740             break;
 741         case 'V':
 742             version(argv[0]);
 743             exit(0);
 744             break;
 745         case 'h':
 746             usage(argv[0]);
 747             exit(0);
 748             break;
 749         case '?':
 750             error_report("Try `%s --help' for more information.", argv[0]);
 751             exit(EXIT_FAILURE);
 752         case QEMU_NBD_OPT_OBJECT: {
 753             QemuOpts *opts;
 754             opts = qemu_opts_parse_noisily(&qemu_object_opts,
 755                                            optarg, true);
 756             if (!opts) {
 757                 exit(EXIT_FAILURE);
 758             }
 759         }   break;
 760         case QEMU_NBD_OPT_TLSCREDS:
 761             tlscredsid = optarg;
 762             break;
 763         case QEMU_NBD_OPT_IMAGE_OPTS:
 764             imageOpts = true;
 765             break;
 766         case 'T':
 767             g_free(trace_file);
 768             trace_file = trace_opt_parse(optarg);
 769             break;
 770         case QEMU_NBD_OPT_FORK:
 771             fork_process = true;
 772             break;
 773         }
 774     }
 775
 776     if ((argc - optind) != 1) {
 777         error_report("Invalid number of arguments");
 778         error_printf("Try `%s --help' for more information.\n", argv[0]);
 779         exit(EXIT_FAILURE);
 780     }
 781
 782     if (qemu_opts_foreach(&qemu_object_opts,
 783                           user_creatable_add_opts_foreach,
 784                           NULL, NULL)) {
 785         exit(EXIT_FAILURE);
 786     }
 787
 788     if (!trace_init_backends()) {
 789         exit(1);
 790     }
 791     trace_init_file(trace_file);
 792     qemu_set_log(LOG_TRACE);
 793
 794     socket_activation = check_socket_activation();
 795     if (socket_activation == 0) {
 796         setup_address_and_port(&bindto, &port);
 797     } else {
 798         /* Using socket activation - check user didn't use -p etc. */
 799         const char *err_msg = socket_activation_validate_opts(device, sockpath,
 800                                                               bindto, port);
 801         if (err_msg != NULL) {
 802             error_report("%s", err_msg);
 803             exit(EXIT_FAILURE);
 804         }
 805
 806         /* qemu-nbd can only listen on a single socket.  */
 807         if (socket_activation > 1) {
 808             error_report("qemu-nbd does not support socket activation with %s > 1",
 809                          "LISTEN_FDS");
 810             exit(EXIT_FAILURE);
 811         }
 812     }
 813
 814     if (tlscredsid) {
 815         if (sockpath) {
 816             error_report("TLS is only supported with IPv4/IPv6");
 817             exit(EXIT_FAILURE);
 818         }
 819         if (device) {
 820             error_report("TLS is not supported with a host device");
 821             exit(EXIT_FAILURE);
 822         }
 823         if (!export_name) {
 824             /* Set the default NBD protocol export name, since
 825              * we *must* use new style protocol for TLS */
 826             export_name = "";
 827         }
 828         tlscreds = nbd_get_tls_creds(tlscredsid, &local_err);
 829         if (local_err) {
 830             error_report("Failed to get TLS creds %s",
 831                          error_get_pretty(local_err));
 832             exit(EXIT_FAILURE);
 833         }
 834     }
 835
 836     if (disconnect) {
 837         int nbdfd = open(argv[optind], O_RDWR);
 838         if (nbdfd < 0) {
 839             error_report("Cannot open %s: %s", argv[optind],
 840                          strerror(errno));
 841             exit(EXIT_FAILURE);
 842         }
 843         nbd_disconnect(nbdfd);
 844
 845         close(nbdfd);
 846
 847         printf("%s disconnected\n", argv[optind]);
 848
 849         return 0;
 850     }
 851
 852     if ((device && !verbose) || fork_process) {
 853         int stderr_fd[2];
 854         pid_t pid;
 855         int ret;
 856
 857         if (qemu_pipe(stderr_fd) < 0) {
 858             error_report("Error setting up communication pipe: %s",
 859                          strerror(errno));
 860             exit(EXIT_FAILURE);
 861         }
 862
 863         /* Now daemonize, but keep a communication channel open to
 864          * print errors and exit with the proper status code.
 865          */
 866         pid = fork();
 867         if (pid < 0) {
 868             error_report("Failed to fork: %s", strerror(errno));
 869             exit(EXIT_FAILURE);
 870         } else if (pid == 0) {
 871             close(stderr_fd[0]);
 872             ret = qemu_daemon(1, 0);
 873
 874             /* Temporarily redirect stderr to the parent's pipe...  */
 875             old_stderr = dup(STDERR_FILENO);
 876             dup2(stderr_fd[1], STDERR_FILENO);
 877             if (ret < 0) {
 878                 error_report("Failed to daemonize: %s", strerror(errno));
 879                 exit(EXIT_FAILURE);
 880             }
 881
 882             /* ... close the descriptor we inherited and go on.  */
 883             close(stderr_fd[1]);
 884         } else {
 885             bool errors = false;
 886             char *buf;
 887
 888             /* In the parent.  Print error messages from the child until
 889              * it closes the pipe.
 890              */
 891             close(stderr_fd[1]);
 892             buf = g_malloc(1024);
 893             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
 894                 errors = true;
 895                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
 896                 if (ret < 0) {
 897                     exit(EXIT_FAILURE);
 898                 }
 899             }
 900             if (ret < 0) {
 901                 error_report("Cannot read from daemon: %s",
 902                              strerror(errno));
 903                 exit(EXIT_FAILURE);
 904             }
 905
 906             /* Usually the daemon should not print any message.
 907              * Exit with zero status in that case.
 908              */
 909             exit(errors);
 910         }
 911     }
 912
 913     if (device != NULL && sockpath == NULL) {
 914         sockpath = g_malloc(128);
 915         snprintf(sockpath, 128, SOCKET_PATH, basename(device));
 916     }
 917
 918     if (socket_activation == 0) {
 919         server_ioc = qio_channel_socket_new();
 920         saddr = nbd_build_socket_address(sockpath, bindto, port);
 921         if (qio_channel_socket_listen_sync(server_ioc, saddr, &local_err) < 0) {
 922             object_unref(OBJECT(server_ioc));
 923             error_report_err(local_err);
 924             return 1;
 925         }
 926     } else {
 927         /* See comment in check_socket_activation above. */
 928         assert(socket_activation == 1);
 929         server_ioc = qio_channel_socket_new_fd(FIRST_SOCKET_ACTIVATION_FD,
 930                                                &local_err);
 931         if (server_ioc == NULL) {
 932             error_report("Failed to use socket activation: %s",
 933                          error_get_pretty(local_err));
 934             exit(EXIT_FAILURE);
 935         }
 936     }
 937
 938     if (qemu_init_main_loop(&local_err)) {
 939         error_report_err(local_err);
 940         exit(EXIT_FAILURE);
 941     }
 942     bdrv_init();
 943     atexit(bdrv_close_all);
 944
 945     srcpath = argv[optind];
 946     if (imageOpts) {
 947         QemuOpts *opts;
 948         if (fmt) {
 949             error_report("--image-opts and -f are mutually exclusive");
 950             exit(EXIT_FAILURE);
 951         }
 952         opts = qemu_opts_parse_noisily(&file_opts, srcpath, true);
 953         if (!opts) {
 954             qemu_opts_reset(&file_opts);
 955             exit(EXIT_FAILURE);
 956         }
 957         options = qemu_opts_to_qdict(opts, NULL);
 958         qemu_opts_reset(&file_opts);
 959         blk = blk_new_open(NULL, NULL, options, flags, &local_err);
 960     } else {
 961         if (fmt) {
 962             options = qdict_new();
 963             qdict_put_str(options, "driver", fmt);
 964         }
 965         blk = blk_new_open(srcpath, NULL, options, flags, &local_err);
 966     }
 967
 968     if (!blk) {
 969         error_reportf_err(local_err, "Failed to blk_new_open '%s': ",
 970                           argv[optind]);
 971         exit(EXIT_FAILURE);
 972     }
 973     bs = blk_bs(blk);
 974
 975     blk_set_enable_write_cache(blk, !writethrough);
 976
 977     if (sn_opts) {
 978         ret = bdrv_snapshot_load_tmp(bs,
 979                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
 980                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
 981                                      &local_err);
 982     } else if (sn_id_or_name) {
 983         ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
 984                                                    &local_err);
 985     }
 986     if (ret < 0) {
 987         error_reportf_err(local_err, "Failed to load snapshot: ");
 988         exit(EXIT_FAILURE);
 989     }
 990
 991     bs->detect_zeroes = detect_zeroes;
 992     fd_size = blk_getlength(blk);
 993     if (fd_size < 0) {
 994         error_report("Failed to determine the image length: %s",
 995                      strerror(-fd_size));
 996         exit(EXIT_FAILURE);
 997     }
 998
 999     if (dev_offset >= fd_size) {
1000         error_report("Offset (%lld) has to be smaller than the image size "
1001                      "(%lld)",
1002                      (long long int)dev_offset, (long long int)fd_size);
1003         exit(EXIT_FAILURE);
1004     }
1005     fd_size -= dev_offset;
1006
1007     if (partition != -1) {
1008         ret = find_partition(blk, partition, &dev_offset, &fd_size);
1009         if (ret < 0) {
1010             error_report("Could not find partition %d: %s", partition,
1011                          strerror(-ret));
1012             exit(EXIT_FAILURE);
1013         }
1014     }
1015
1016     exp = nbd_export_new(bs, dev_offset, fd_size, nbdflags, nbd_export_closed,
1017                          writethrough, NULL, &local_err);
1018     if (!exp) {
1019         error_report_err(local_err);
1020         exit(EXIT_FAILURE);
1021     }
1022     if (export_name) {
1023         nbd_export_set_name(exp, export_name);
1024         nbd_export_set_description(exp, export_description);
1025         newproto = true;
1026     } else if (export_description) {
1027         error_report("Export description requires an export name");
1028         exit(EXIT_FAILURE);
1029     }
1030
1031     if (device) {
1032         int ret;
1033
1034         ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
1035         if (ret != 0) {
1036             error_report("Failed to create client thread: %s", strerror(ret));
1037             exit(EXIT_FAILURE);
1038         }
1039     } else {
1040         /* Shut up GCC warnings.  */
1041         memset(&client_thread, 0, sizeof(client_thread));
1042     }
1043
1044     nbd_update_server_watch();
1045
1046     /* now when the initialization is (almost) complete, chdir("/")
1047      * to free any busy filesystems */
1048     if (chdir("/") < 0) {
1049         error_report("Could not chdir to root directory: %s",
1050                      strerror(errno));
1051         exit(EXIT_FAILURE);
1052     }
1053
1054     if (fork_process) {
1055         dup2(old_stderr, STDERR_FILENO);
1056         close(old_stderr);
1057     }
1058
1059     state = RUNNING;
1060     do {
1061         main_loop_wait(false);
1062         if (state == TERMINATE) {
1063             state = TERMINATING;
1064             nbd_export_close(exp);
1065             nbd_export_put(exp);
1066             exp = NULL;
1067         }
1068     } while (state != TERMINATED);
1069
1070     blk_unref(blk);
1071     if (sockpath) {
1072         unlink(sockpath);
1073     }
1074
1075     qemu_opts_del(sn_opts);
1076
1077     if (device) {
1078         void *ret;
1079         pthread_join(client_thread, &ret);
1080         exit(ret != NULL);
1081     } else {
1082         exit(EXIT_SUCCESS);
1083     }
1084 }