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 *export;
  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 "  -B, --bitmap=NAME         expose a persistent dirty bitmap\n"
  99 "\n"
 100 "General purpose options:\n"
 101 "  --object type,id=ID,...   define an object such as 'secret' for providing\n"
 102 "                            passwords and/or encryption keys\n"
 103 "  --tls-creds=ID            use id of an earlier --object to provide TLS\n"
 104 "  -T, --trace [[enable=]<pattern>][,events=<file>][,file=<file>]\n"
 105 "                            specify tracing options\n"
 106 "  --fork                    fork off the server process and exit the parent\n"
 107 "                            once the server is running\n"
 108 #if HAVE_NBD_DEVICE
 109 "\n"
 110 "Kernel NBD client support:\n"
 111 "  -c, --connect=DEV         connect FILE to the local NBD device DEV\n"
 112 "  -d, --disconnect          disconnect the specified device\n"
 113 #endif
 114 "\n"
 115 "Block device options:\n"
 116 "  -f, --format=FORMAT       set image format (raw, qcow2, ...)\n"
 117 "  -r, --read-only           export read-only\n"
 118 "  -s, --snapshot            use FILE as an external snapshot, create a temporary\n"
 119 "                            file with backing_file=FILE, redirect the write to\n"
 120 "                            the temporary one\n"
 121 "  -l, --load-snapshot=SNAPSHOT_PARAM\n"
 122 "                            load an internal snapshot inside FILE and export it\n"
 123 "                            as an read-only device, SNAPSHOT_PARAM format is\n"
 124 "                            'snapshot.id=[ID],snapshot.name=[NAME]', or\n"
 125 "                            '[ID_OR_NAME]'\n"
 126 "  -n, --nocache             disable host cache\n"
 127 "      --cache=MODE          set cache mode (none, writeback, ...)\n"
 128 "      --aio=MODE            set AIO mode (native or threads)\n"
 129 "      --discard=MODE        set discard mode (ignore, unmap)\n"
 130 "      --detect-zeroes=MODE  set detect-zeroes mode (off, on, unmap)\n"
 131 "      --image-opts          treat FILE as a full set of image options\n"
 132 "\n"
 133 QEMU_HELP_BOTTOM "\n"
 134     , name, NBD_DEFAULT_PORT, "DEVICE");
 135 }
 136
 137 static void version(const char *name)
 138 {
 139     printf(
 140 "%s " QEMU_FULL_VERSION "\n"
 141 "Written by Anthony Liguori.\n"
 142 "\n"
 143 QEMU_COPYRIGHT "\n"
 144 "This is free software; see the source for copying conditions.  There is NO\n"
 145 "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
 146     , name);
 147 }
 148
 149 struct partition_record
 150 {
 151     uint8_t bootable;
 152     uint8_t start_head;
 153     uint32_t start_cylinder;
 154     uint8_t start_sector;
 155     uint8_t system;
 156     uint8_t end_head;
 157     uint8_t end_cylinder;
 158     uint8_t end_sector;
 159     uint32_t start_sector_abs;
 160     uint32_t nb_sectors_abs;
 161 };
 162
 163 static void read_partition(uint8_t *p, struct partition_record *r)
 164 {
 165     r->bootable = p[0];
 166     r->start_head = p[1];
 167     r->start_cylinder = p[3] | ((p[2] << 2) & 0x0300);
 168     r->start_sector = p[2] & 0x3f;
 169     r->system = p[4];
 170     r->end_head = p[5];
 171     r->end_cylinder = p[7] | ((p[6] << 2) & 0x300);
 172     r->end_sector = p[6] & 0x3f;
 173
 174     r->start_sector_abs = ldl_le_p(p + 8);
 175     r->nb_sectors_abs   = ldl_le_p(p + 12);
 176 }
 177
 178 static int find_partition(BlockBackend *blk, int partition,
 179                           uint64_t *offset, uint64_t *size)
 180 {
 181     struct partition_record mbr[4];
 182     uint8_t data[MBR_SIZE];
 183     int i;
 184     int ext_partnum = 4;
 185     int ret;
 186
 187     ret = blk_pread(blk, 0, data, sizeof(data));
 188     if (ret < 0) {
 189         error_report("error while reading: %s", strerror(-ret));
 190         exit(EXIT_FAILURE);
 191     }
 192
 193     if (data[510] != 0x55 || data[511] != 0xaa) {
 194         return -EINVAL;
 195     }
 196
 197     for (i = 0; i < 4; i++) {
 198         read_partition(&data[446 + 16 * i], &mbr[i]);
 199
 200         if (!mbr[i].system || !mbr[i].nb_sectors_abs) {
 201             continue;
 202         }
 203
 204         if (mbr[i].system == 0xF || mbr[i].system == 0x5) {
 205             struct partition_record ext[4];
 206             uint8_t data1[MBR_SIZE];
 207             int j;
 208
 209             ret = blk_pread(blk, mbr[i].start_sector_abs * MBR_SIZE,
 210                             data1, sizeof(data1));
 211             if (ret < 0) {
 212                 error_report("error while reading: %s", strerror(-ret));
 213                 exit(EXIT_FAILURE);
 214             }
 215
 216             for (j = 0; j < 4; j++) {
 217                 read_partition(&data1[446 + 16 * j], &ext[j]);
 218                 if (!ext[j].system || !ext[j].nb_sectors_abs) {
 219                     continue;
 220                 }
 221
 222                 if ((ext_partnum + j + 1) == partition) {
 223                     *offset = (uint64_t)ext[j].start_sector_abs << 9;
 224                     *size = (uint64_t)ext[j].nb_sectors_abs << 9;
 225                     return 0;
 226                 }
 227             }
 228             ext_partnum += 4;
 229         } else if ((i + 1) == partition) {
 230             *offset = (uint64_t)mbr[i].start_sector_abs << 9;
 231             *size = (uint64_t)mbr[i].nb_sectors_abs << 9;
 232             return 0;
 233         }
 234     }
 235
 236     return -ENOENT;
 237 }
 238
 239 static void termsig_handler(int signum)
 240 {
 241     atomic_cmpxchg(&state, RUNNING, TERMINATE);
 242     qemu_notify_event();
 243 }
 244
 245
 246 #if HAVE_NBD_DEVICE
 247 static void *show_parts(void *arg)
 248 {
 249     char *device = arg;
 250     int nbd;
 251
 252     /* linux just needs an open() to trigger
 253      * the partition table update
 254      * but remember to load the module with max_part != 0 :
 255      *     modprobe nbd max_part=63
 256      */
 257     nbd = open(device, O_RDWR);
 258     if (nbd >= 0) {
 259         close(nbd);
 260     }
 261     return NULL;
 262 }
 263
 264 static void *nbd_client_thread(void *arg)
 265 {
 266     char *device = arg;
 267     NBDExportInfo info = { .request_sizes = false, .name = g_strdup("") };
 268     QIOChannelSocket *sioc;
 269     int fd;
 270     int ret;
 271     pthread_t show_parts_thread;
 272     Error *local_error = NULL;
 273
 274     sioc = qio_channel_socket_new();
 275     if (qio_channel_socket_connect_sync(sioc,
 276                                         saddr,
 277                                         &local_error) < 0) {
 278         error_report_err(local_error);
 279         goto out;
 280     }
 281
 282     ret = nbd_receive_negotiate(QIO_CHANNEL(sioc),
 283                                 NULL, NULL, NULL, &info, &local_error);
 284     if (ret < 0) {
 285         if (local_error) {
 286             error_report_err(local_error);
 287         }
 288         goto out_socket;
 289     }
 290
 291     fd = open(device, O_RDWR);
 292     if (fd < 0) {
 293         /* Linux-only, we can use %m in printf.  */
 294         error_report("Failed to open %s: %m", device);
 295         goto out_socket;
 296     }
 297
 298     ret = nbd_init(fd, sioc, &info, &local_error);
 299     if (ret < 0) {
 300         error_report_err(local_error);
 301         goto out_fd;
 302     }
 303
 304     /* update partition table */
 305     pthread_create(&show_parts_thread, NULL, show_parts, device);
 306
 307     if (verbose) {
 308         fprintf(stderr, "NBD device %s is now connected to %s\n",
 309                 device, srcpath);
 310     } else {
 311         /* Close stderr so that the qemu-nbd process exits.  */
 312         dup2(STDOUT_FILENO, STDERR_FILENO);
 313     }
 314
 315     ret = nbd_client(fd);
 316     if (ret) {
 317         goto out_fd;
 318     }
 319     close(fd);
 320     object_unref(OBJECT(sioc));
 321     g_free(info.name);
 322     kill(getpid(), SIGTERM);
 323     return (void *) EXIT_SUCCESS;
 324
 325 out_fd:
 326     close(fd);
 327 out_socket:
 328     object_unref(OBJECT(sioc));
 329 out:
 330     g_free(info.name);
 331     kill(getpid(), SIGTERM);
 332     return (void *) EXIT_FAILURE;
 333 }
 334 #endif /* HAVE_NBD_DEVICE */
 335
 336 static int nbd_can_accept(void)
 337 {
 338     return state == RUNNING && nb_fds < shared;
 339 }
 340
 341 static void nbd_export_closed(NBDExport *export)
 342 {
 343     assert(state == TERMINATING);
 344     state = TERMINATED;
 345 }
 346
 347 static void nbd_update_server_watch(void);
 348
 349 static void nbd_client_closed(NBDClient *client, bool negotiated)
 350 {
 351     nb_fds--;
 352     if (negotiated && nb_fds == 0 && !persistent && state == RUNNING) {
 353         state = TERMINATE;
 354     }
 355     nbd_update_server_watch();
 356     nbd_client_put(client);
 357 }
 358
 359 static void nbd_accept(QIONetListener *listener, QIOChannelSocket *cioc,
 360                        gpointer opaque)
 361 {
 362     if (state >= TERMINATE) {
 363         return;
 364     }
 365
 366     nb_fds++;
 367     nbd_update_server_watch();
 368     nbd_client_new(cioc, tlscreds, NULL, nbd_client_closed);
 369 }
 370
 371 static void nbd_update_server_watch(void)
 372 {
 373     if (nbd_can_accept()) {
 374         qio_net_listener_set_client_func(server, nbd_accept, NULL, NULL);
 375     } else {
 376         qio_net_listener_set_client_func(server, NULL, NULL, NULL);
 377     }
 378 }
 379
 380
 381 static SocketAddress *nbd_build_socket_address(const char *sockpath,
 382                                                const char *bindto,
 383                                                const char *port)
 384 {
 385     SocketAddress *saddr;
 386
 387     saddr = g_new0(SocketAddress, 1);
 388     if (sockpath) {
 389         saddr->type = SOCKET_ADDRESS_TYPE_UNIX;
 390         saddr->u.q_unix.path = g_strdup(sockpath);
 391     } else {
 392         InetSocketAddress *inet;
 393         saddr->type = SOCKET_ADDRESS_TYPE_INET;
 394         inet = &saddr->u.inet;
 395         inet->host = g_strdup(bindto);
 396         if (port) {
 397             inet->port = g_strdup(port);
 398         } else  {
 399             inet->port = g_strdup_printf("%d", NBD_DEFAULT_PORT);
 400         }
 401     }
 402
 403     return saddr;
 404 }
 405
 406
 407 static QemuOptsList file_opts = {
 408     .name = "file",
 409     .implied_opt_name = "file",
 410     .head = QTAILQ_HEAD_INITIALIZER(file_opts.head),
 411     .desc = {
 412         /* no elements => accept any params */
 413         { /* end of list */ }
 414     },
 415 };
 416
 417 static QemuOptsList qemu_object_opts = {
 418     .name = "object",
 419     .implied_opt_name = "qom-type",
 420     .head = QTAILQ_HEAD_INITIALIZER(qemu_object_opts.head),
 421     .desc = {
 422         { }
 423     },
 424 };
 425
 426
 427
 428 static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, Error **errp)
 429 {
 430     Object *obj;
 431     QCryptoTLSCreds *creds;
 432
 433     obj = object_resolve_path_component(
 434         object_get_objects_root(), id);
 435     if (!obj) {
 436         error_setg(errp, "No TLS credentials with id '%s'",
 437                    id);
 438         return NULL;
 439     }
 440     creds = (QCryptoTLSCreds *)
 441         object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
 442     if (!creds) {
 443         error_setg(errp, "Object with id '%s' is not TLS credentials",
 444                    id);
 445         return NULL;
 446     }
 447
 448     if (creds->endpoint != QCRYPTO_TLS_CREDS_ENDPOINT_SERVER) {
 449         error_setg(errp,
 450                    "Expecting TLS credentials with a server endpoint");
 451         return NULL;
 452     }
 453     object_ref(obj);
 454     return creds;
 455 }
 456
 457 static void setup_address_and_port(const char **address, const char **port)
 458 {
 459     if (*address == NULL) {
 460         *address = "0.0.0.0";
 461     }
 462
 463     if (*port == NULL) {
 464         *port = stringify(NBD_DEFAULT_PORT);
 465     }
 466 }
 467
 468 /*
 469  * Check socket parameters compatibility when socket activation is used.
 470  */
 471 static const char *socket_activation_validate_opts(const char *device,
 472                                                    const char *sockpath,
 473                                                    const char *address,
 474                                                    const char *port)
 475 {
 476     if (device != NULL) {
 477         return "NBD device can't be set when using socket activation";
 478     }
 479
 480     if (sockpath != NULL) {
 481         return "Unix socket can't be set when using socket activation";
 482     }
 483
 484     if (address != NULL) {
 485         return "The interface can't be set when using socket activation";
 486     }
 487
 488     if (port != NULL) {
 489         return "TCP port number can't be set when using socket activation";
 490     }
 491
 492     return NULL;
 493 }
 494
 495 static void qemu_nbd_shutdown(void)
 496 {
 497     job_cancel_sync_all();
 498     bdrv_close_all();
 499 }
 500
 501 int main(int argc, char **argv)
 502 {
 503     BlockBackend *blk;
 504     BlockDriverState *bs;
 505     uint64_t dev_offset = 0;
 506     uint16_t nbdflags = 0;
 507     bool disconnect = false;
 508     const char *bindto = NULL;
 509     const char *port = NULL;
 510     char *sockpath = NULL;
 511     char *device = NULL;
 512     int64_t fd_size;
 513     QemuOpts *sn_opts = NULL;
 514     const char *sn_id_or_name = NULL;
 515     const char *sopt = "hVb:o:p:rsnP:c:dvk:e:f:tl:x:T:D:B:";
 516     struct option lopt[] = {
 517         { "help", no_argument, NULL, 'h' },
 518         { "version", no_argument, NULL, 'V' },
 519         { "bind", required_argument, NULL, 'b' },
 520         { "port", required_argument, NULL, 'p' },
 521         { "socket", required_argument, NULL, 'k' },
 522         { "offset", required_argument, NULL, 'o' },
 523         { "read-only", no_argument, NULL, 'r' },
 524         { "partition", required_argument, NULL, 'P' },
 525         { "bitmap", required_argument, NULL, 'B' },
 526         { "connect", required_argument, NULL, 'c' },
 527         { "disconnect", no_argument, NULL, 'd' },
 528         { "snapshot", no_argument, NULL, 's' },
 529         { "load-snapshot", required_argument, NULL, 'l' },
 530         { "nocache", no_argument, NULL, 'n' },
 531         { "cache", required_argument, NULL, QEMU_NBD_OPT_CACHE },
 532         { "aio", required_argument, NULL, QEMU_NBD_OPT_AIO },
 533         { "discard", required_argument, NULL, QEMU_NBD_OPT_DISCARD },
 534         { "detect-zeroes", required_argument, NULL,
 535           QEMU_NBD_OPT_DETECT_ZEROES },
 536         { "shared", required_argument, NULL, 'e' },
 537         { "format", required_argument, NULL, 'f' },
 538         { "persistent", no_argument, NULL, 't' },
 539         { "verbose", no_argument, NULL, 'v' },
 540         { "object", required_argument, NULL, QEMU_NBD_OPT_OBJECT },
 541         { "export-name", required_argument, NULL, 'x' },
 542         { "description", required_argument, NULL, 'D' },
 543         { "tls-creds", required_argument, NULL, QEMU_NBD_OPT_TLSCREDS },
 544         { "image-opts", no_argument, NULL, QEMU_NBD_OPT_IMAGE_OPTS },
 545         { "trace", required_argument, NULL, 'T' },
 546         { "fork", no_argument, NULL, QEMU_NBD_OPT_FORK },
 547         { NULL, 0, NULL, 0 }
 548     };
 549     int ch;
 550     int opt_ind = 0;
 551     int flags = BDRV_O_RDWR;
 552     int partition = 0;
 553     int ret = 0;
 554     bool seen_cache = false;
 555     bool seen_discard = false;
 556     bool seen_aio = false;
 557     pthread_t client_thread;
 558     const char *fmt = NULL;
 559     Error *local_err = NULL;
 560     BlockdevDetectZeroesOptions detect_zeroes = BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
 561     QDict *options = NULL;
 562     const char *export_name = ""; /* Default export name */
 563     const char *export_description = NULL;
 564     const char *bitmap = NULL;
 565     const char *tlscredsid = NULL;
 566     bool imageOpts = false;
 567     bool writethrough = true;
 568     char *trace_file = NULL;
 569     bool fork_process = false;
 570     int old_stderr = -1;
 571     unsigned socket_activation;
 572
 573     /* The client thread uses SIGTERM to interrupt the server.  A signal
 574      * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
 575      */
 576     struct sigaction sa_sigterm;
 577     memset(&sa_sigterm, 0, sizeof(sa_sigterm));
 578     sa_sigterm.sa_handler = termsig_handler;
 579     sigaction(SIGTERM, &sa_sigterm, NULL);
 580
 581 #ifdef CONFIG_POSIX
 582     signal(SIGPIPE, SIG_IGN);
 583 #endif
 584
 585     module_call_init(MODULE_INIT_TRACE);
 586     error_set_progname(argv[0]);
 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             if (qemu_strtou64(optarg, NULL, 0, &dev_offset) < 0) {
 665                 error_report("Invalid offset '%s'", optarg);
 666                 exit(EXIT_FAILURE);
 667             }
 668             break;
 669         case 'l':
 670             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
 671                 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
 672                                                   optarg, false);
 673                 if (!sn_opts) {
 674                     error_report("Failed in parsing snapshot param `%s'",
 675                                  optarg);
 676                     exit(EXIT_FAILURE);
 677                 }
 678             } else {
 679                 sn_id_or_name = optarg;
 680             }
 681             /* fall through */
 682         case 'r':
 683             nbdflags |= NBD_FLAG_READ_ONLY;
 684             flags &= ~BDRV_O_RDWR;
 685             break;
 686         case 'P':
 687             if (qemu_strtoi(optarg, NULL, 0, &partition) < 0 ||
 688                 partition < 1 || partition > 8) {
 689                 error_report("Invalid partition '%s'", optarg);
 690                 exit(EXIT_FAILURE);
 691             }
 692             break;
 693         case 'B':
 694             bitmap = optarg;
 695             break;
 696         case 'k':
 697             sockpath = optarg;
 698             if (sockpath[0] != '/') {
 699                 error_report("socket path must be absolute");
 700                 exit(EXIT_FAILURE);
 701             }
 702             break;
 703         case 'd':
 704             disconnect = true;
 705             break;
 706         case 'c':
 707             device = optarg;
 708             break;
 709         case 'e':
 710             if (qemu_strtoi(optarg, NULL, 0, &shared) < 0 ||
 711                 shared < 1) {
 712                 error_report("Invalid shared device number '%s'", optarg);
 713                 exit(EXIT_FAILURE);
 714             }
 715             break;
 716         case 'f':
 717             fmt = optarg;
 718             break;
 719         case 't':
 720             persistent = 1;
 721             break;
 722         case 'x':
 723             export_name = optarg;
 724             break;
 725         case 'D':
 726             export_description = optarg;
 727             break;
 728         case 'v':
 729             verbose = 1;
 730             break;
 731         case 'V':
 732             version(argv[0]);
 733             exit(0);
 734             break;
 735         case 'h':
 736             usage(argv[0]);
 737             exit(0);
 738             break;
 739         case '?':
 740             error_report("Try `%s --help' for more information.", argv[0]);
 741             exit(EXIT_FAILURE);
 742         case QEMU_NBD_OPT_OBJECT: {
 743             QemuOpts *opts;
 744             opts = qemu_opts_parse_noisily(&qemu_object_opts,
 745                                            optarg, true);
 746             if (!opts) {
 747                 exit(EXIT_FAILURE);
 748             }
 749         }   break;
 750         case QEMU_NBD_OPT_TLSCREDS:
 751             tlscredsid = optarg;
 752             break;
 753         case QEMU_NBD_OPT_IMAGE_OPTS:
 754             imageOpts = true;
 755             break;
 756         case 'T':
 757             g_free(trace_file);
 758             trace_file = trace_opt_parse(optarg);
 759             break;
 760         case QEMU_NBD_OPT_FORK:
 761             fork_process = true;
 762             break;
 763         }
 764     }
 765
 766     if ((argc - optind) != 1) {
 767         error_report("Invalid number of arguments");
 768         error_printf("Try `%s --help' for more information.\n", argv[0]);
 769         exit(EXIT_FAILURE);
 770     }
 771
 772     qemu_opts_foreach(&qemu_object_opts,
 773                       user_creatable_add_opts_foreach,
 774                       NULL, &error_fatal);
 775
 776     if (!trace_init_backends()) {
 777         exit(1);
 778     }
 779     trace_init_file(trace_file);
 780     qemu_set_log(LOG_TRACE);
 781
 782     socket_activation = check_socket_activation();
 783     if (socket_activation == 0) {
 784         setup_address_and_port(&bindto, &port);
 785     } else {
 786         /* Using socket activation - check user didn't use -p etc. */
 787         const char *err_msg = socket_activation_validate_opts(device, sockpath,
 788                                                               bindto, port);
 789         if (err_msg != NULL) {
 790             error_report("%s", err_msg);
 791             exit(EXIT_FAILURE);
 792         }
 793
 794         /* qemu-nbd can only listen on a single socket.  */
 795         if (socket_activation > 1) {
 796             error_report("qemu-nbd does not support socket activation with %s > 1",
 797                          "LISTEN_FDS");
 798             exit(EXIT_FAILURE);
 799         }
 800     }
 801
 802     if (tlscredsid) {
 803         if (sockpath) {
 804             error_report("TLS is only supported with IPv4/IPv6");
 805             exit(EXIT_FAILURE);
 806         }
 807         if (device) {
 808             error_report("TLS is not supported with a host device");
 809             exit(EXIT_FAILURE);
 810         }
 811         tlscreds = nbd_get_tls_creds(tlscredsid, &local_err);
 812         if (local_err) {
 813             error_report("Failed to get TLS creds %s",
 814                          error_get_pretty(local_err));
 815             exit(EXIT_FAILURE);
 816         }
 817     }
 818
 819 #if !HAVE_NBD_DEVICE
 820     if (disconnect || device) {
 821         error_report("Kernel /dev/nbdN support not available");
 822         exit(EXIT_FAILURE);
 823     }
 824 #else /* HAVE_NBD_DEVICE */
 825     if (disconnect) {
 826         int nbdfd = open(argv[optind], O_RDWR);
 827         if (nbdfd < 0) {
 828             error_report("Cannot open %s: %s", argv[optind],
 829                          strerror(errno));
 830             exit(EXIT_FAILURE);
 831         }
 832         nbd_disconnect(nbdfd);
 833
 834         close(nbdfd);
 835
 836         printf("%s disconnected\n", argv[optind]);
 837
 838         return 0;
 839     }
 840 #endif
 841
 842     if ((device && !verbose) || fork_process) {
 843         int stderr_fd[2];
 844         pid_t pid;
 845         int ret;
 846
 847         if (qemu_pipe(stderr_fd) < 0) {
 848             error_report("Error setting up communication pipe: %s",
 849                          strerror(errno));
 850             exit(EXIT_FAILURE);
 851         }
 852
 853         /* Now daemonize, but keep a communication channel open to
 854          * print errors and exit with the proper status code.
 855          */
 856         pid = fork();
 857         if (pid < 0) {
 858             error_report("Failed to fork: %s", strerror(errno));
 859             exit(EXIT_FAILURE);
 860         } else if (pid == 0) {
 861             close(stderr_fd[0]);
 862             ret = qemu_daemon(1, 0);
 863
 864             /* Temporarily redirect stderr to the parent's pipe...  */
 865             old_stderr = dup(STDERR_FILENO);
 866             dup2(stderr_fd[1], STDERR_FILENO);
 867             if (ret < 0) {
 868                 error_report("Failed to daemonize: %s", strerror(errno));
 869                 exit(EXIT_FAILURE);
 870             }
 871
 872             /* ... close the descriptor we inherited and go on.  */
 873             close(stderr_fd[1]);
 874         } else {
 875             bool errors = false;
 876             char *buf;
 877
 878             /* In the parent.  Print error messages from the child until
 879              * it closes the pipe.
 880              */
 881             close(stderr_fd[1]);
 882             buf = g_malloc(1024);
 883             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
 884                 errors = true;
 885                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
 886                 if (ret < 0) {
 887                     exit(EXIT_FAILURE);
 888                 }
 889             }
 890             if (ret < 0) {
 891                 error_report("Cannot read from daemon: %s",
 892                              strerror(errno));
 893                 exit(EXIT_FAILURE);
 894             }
 895
 896             /* Usually the daemon should not print any message.
 897              * Exit with zero status in that case.
 898              */
 899             exit(errors);
 900         }
 901     }
 902
 903     if (device != NULL && sockpath == NULL) {
 904         sockpath = g_malloc(128);
 905         snprintf(sockpath, 128, SOCKET_PATH, basename(device));
 906     }
 907
 908     server = qio_net_listener_new();
 909     if (socket_activation == 0) {
 910         saddr = nbd_build_socket_address(sockpath, bindto, port);
 911         if (qio_net_listener_open_sync(server, saddr, &local_err) < 0) {
 912             object_unref(OBJECT(server));
 913             error_report_err(local_err);
 914             exit(EXIT_FAILURE);
 915         }
 916     } else {
 917         size_t i;
 918         /* See comment in check_socket_activation above. */
 919         for (i = 0; i < socket_activation; i++) {
 920             QIOChannelSocket *sioc;
 921             sioc = qio_channel_socket_new_fd(FIRST_SOCKET_ACTIVATION_FD + i,
 922                                              &local_err);
 923             if (sioc == NULL) {
 924                 object_unref(OBJECT(server));
 925                 error_report("Failed to use socket activation: %s",
 926                              error_get_pretty(local_err));
 927                 exit(EXIT_FAILURE);
 928             }
 929             qio_net_listener_add(server, sioc);
 930             object_unref(OBJECT(sioc));
 931         }
 932     }
 933
 934     if (qemu_init_main_loop(&local_err)) {
 935         error_report_err(local_err);
 936         exit(EXIT_FAILURE);
 937     }
 938     bdrv_init();
 939     atexit(qemu_nbd_shutdown);
 940
 941     srcpath = argv[optind];
 942     if (imageOpts) {
 943         QemuOpts *opts;
 944         if (fmt) {
 945             error_report("--image-opts and -f are mutually exclusive");
 946             exit(EXIT_FAILURE);
 947         }
 948         opts = qemu_opts_parse_noisily(&file_opts, srcpath, true);
 949         if (!opts) {
 950             qemu_opts_reset(&file_opts);
 951             exit(EXIT_FAILURE);
 952         }
 953         options = qemu_opts_to_qdict(opts, NULL);
 954         qemu_opts_reset(&file_opts);
 955         blk = blk_new_open(NULL, NULL, options, flags, &local_err);
 956     } else {
 957         if (fmt) {
 958             options = qdict_new();
 959             qdict_put_str(options, "driver", fmt);
 960         }
 961         blk = blk_new_open(srcpath, NULL, options, flags, &local_err);
 962     }
 963
 964     if (!blk) {
 965         error_reportf_err(local_err, "Failed to blk_new_open '%s': ",
 966                           argv[optind]);
 967         exit(EXIT_FAILURE);
 968     }
 969     bs = blk_bs(blk);
 970
 971     blk_set_enable_write_cache(blk, !writethrough);
 972
 973     if (sn_opts) {
 974         ret = bdrv_snapshot_load_tmp(bs,
 975                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
 976                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
 977                                      &local_err);
 978     } else if (sn_id_or_name) {
 979         ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
 980                                                    &local_err);
 981     }
 982     if (ret < 0) {
 983         error_reportf_err(local_err, "Failed to load snapshot: ");
 984         exit(EXIT_FAILURE);
 985     }
 986
 987     bs->detect_zeroes = detect_zeroes;
 988     fd_size = blk_getlength(blk);
 989     if (fd_size < 0) {
 990         error_report("Failed to determine the image length: %s",
 991                      strerror(-fd_size));
 992         exit(EXIT_FAILURE);
 993     }
 994
 995     if (dev_offset >= fd_size) {
 996         error_report("Offset (%" PRIu64 ") has to be smaller than the image "
 997                      "size (%" PRId64 ")", dev_offset, fd_size);
 998         exit(EXIT_FAILURE);
 999     }
1000     fd_size -= dev_offset;
1001
1002     if (partition) {
1003         uint64_t limit;
1004
1005         if (dev_offset) {
1006             error_report("Cannot request partition and offset together");
1007             exit(EXIT_FAILURE);
1008         }
1009         ret = find_partition(blk, partition, &dev_offset, &limit);
1010         if (ret < 0) {
1011             error_report("Could not find partition %d: %s", partition,
1012                          strerror(-ret));
1013             exit(EXIT_FAILURE);
1014         }
1015         /*
1016          * MBR partition limits are (32-bit << 9); this assert lets
1017          * the compiler know that we can't overflow 64 bits.
1018          */
1019         assert(dev_offset + limit >= dev_offset);
1020         if (dev_offset + limit > fd_size) {
1021             error_report("Discovered partition %d at offset %" PRIu64
1022                          " size %" PRIu64 ", but size exceeds file length %"
1023                          PRId64, partition, dev_offset, limit, fd_size);
1024             exit(EXIT_FAILURE);
1025         }
1026         fd_size = limit;
1027     }
1028
1029     export = nbd_export_new(bs, dev_offset, fd_size, export_name,
1030                             export_description, bitmap, nbdflags,
1031                             nbd_export_closed, writethrough, NULL,
1032                             &error_fatal);
1033
1034     if (device) {
1035 #if HAVE_NBD_DEVICE
1036         int ret;
1037
1038         ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
1039         if (ret != 0) {
1040             error_report("Failed to create client thread: %s", strerror(ret));
1041             exit(EXIT_FAILURE);
1042         }
1043 #endif
1044     } else {
1045         /* Shut up GCC warnings.  */
1046         memset(&client_thread, 0, sizeof(client_thread));
1047     }
1048
1049     nbd_update_server_watch();
1050
1051     /* now when the initialization is (almost) complete, chdir("/")
1052      * to free any busy filesystems */
1053     if (chdir("/") < 0) {
1054         error_report("Could not chdir to root directory: %s",
1055                      strerror(errno));
1056         exit(EXIT_FAILURE);
1057     }
1058
1059     if (fork_process) {
1060         dup2(old_stderr, STDERR_FILENO);
1061         close(old_stderr);
1062     }
1063
1064     state = RUNNING;
1065     do {
1066         main_loop_wait(false);
1067         if (state == TERMINATE) {
1068             state = TERMINATING;
1069             nbd_export_close(export);
1070             nbd_export_put(export);
1071             export = NULL;
1072         }
1073     } while (state != TERMINATED);
1074
1075     blk_unref(blk);
1076     if (sockpath) {
1077         unlink(sockpath);
1078     }
1079
1080     qemu_opts_del(sn_opts);
1081
1082     if (device) {
1083         void *ret;
1084         pthread_join(client_thread, &ret);
1085         exit(ret != NULL);
1086     } else {
1087         exit(EXIT_SUCCESS);
1088     }
1089 }