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 "qapi/error.h"
  21 #include "qemu-common.h"
  22 #include "qemu/cutils.h"
  23 #include "sysemu/block-backend.h"
  24 #include "block/block_int.h"
  25 #include "block/nbd.h"
  26 #include "qemu/main-loop.h"
  27 #include "qemu/error-report.h"
  28 #include "qemu/config-file.h"
  29 #include "qemu/bswap.h"
  30 #include "qemu/log.h"
  31 #include "qemu/systemd.h"
  32 #include "block/snapshot.h"
  33 #include "qapi/util.h"
  34 #include "qapi/qmp/qstring.h"
  35 #include "qom/object_interfaces.h"
  36 #include "io/channel-socket.h"
  37 #include "crypto/init.h"
  38 #include "trace/control.h"
  39
  40 #include <getopt.h>
  41 #include <libgen.h>
  42 #include <pthread.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 "Report bugs to <qemu-devel@nongnu.org>\n"
 126     , name, NBD_DEFAULT_PORT, "DEVICE");
 127 }
 128
 129 static void version(const char *name)
 130 {
 131     printf(
 132 "%s version 0.0.1\n"
 133 "Written by Anthony Liguori.\n"
 134 "\n"
 135 "Copyright (C) 2006 Anthony Liguori <anthony@codemonkey.ws>.\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     off_t size;
 259     uint16_t nbdflags;
 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, &nbdflags,
 275                                 NULL, NULL, NULL,
 276                                 &size, &local_error);
 277     if (ret < 0) {
 278         if (local_error) {
 279             error_report_err(local_error);
 280         }
 281         goto out_socket;
 282     }
 283
 284     fd = open(device, O_RDWR);
 285     if (fd < 0) {
 286         /* Linux-only, we can use %m in printf.  */
 287         error_report("Failed to open %s: %m", device);
 288         goto out_socket;
 289     }
 290
 291     ret = nbd_init(fd, sioc, nbdflags, size);
 292     if (ret < 0) {
 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 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)
 339 {
 340     nb_fds--;
 341     if (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_KIND_UNIX;
 399         saddr->u.q_unix.data = g_new0(UnixSocketAddress, 1);
 400         saddr->u.q_unix.data->path = g_strdup(sockpath);
 401     } else {
 402         InetSocketAddress *inet;
 403         saddr->type = SOCKET_ADDRESS_KIND_INET;
 404         inet = saddr->u.inet.data = g_new0(InetSocketAddress, 1);
 405         inet->host = g_strdup(bindto);
 406         if (port) {
 407             inet->port = g_strdup(port);
 408         } else  {
 409             inet->port = g_strdup_printf("%d", NBD_DEFAULT_PORT);
 410         }
 411     }
 412
 413     return saddr;
 414 }
 415
 416
 417 static QemuOptsList file_opts = {
 418     .name = "file",
 419     .implied_opt_name = "file",
 420     .head = QTAILQ_HEAD_INITIALIZER(file_opts.head),
 421     .desc = {
 422         /* no elements => accept any params */
 423         { /* end of list */ }
 424     },
 425 };
 426
 427 static QemuOptsList qemu_object_opts = {
 428     .name = "object",
 429     .implied_opt_name = "qom-type",
 430     .head = QTAILQ_HEAD_INITIALIZER(qemu_object_opts.head),
 431     .desc = {
 432         { }
 433     },
 434 };
 435
 436
 437
 438 static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, Error **errp)
 439 {
 440     Object *obj;
 441     QCryptoTLSCreds *creds;
 442
 443     obj = object_resolve_path_component(
 444         object_get_objects_root(), id);
 445     if (!obj) {
 446         error_setg(errp, "No TLS credentials with id '%s'",
 447                    id);
 448         return NULL;
 449     }
 450     creds = (QCryptoTLSCreds *)
 451         object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
 452     if (!creds) {
 453         error_setg(errp, "Object with id '%s' is not TLS credentials",
 454                    id);
 455         return NULL;
 456     }
 457
 458     if (creds->endpoint != QCRYPTO_TLS_CREDS_ENDPOINT_SERVER) {
 459         error_setg(errp,
 460                    "Expecting TLS credentials with a server endpoint");
 461         return NULL;
 462     }
 463     object_ref(obj);
 464     return creds;
 465 }
 466
 467 static void setup_address_and_port(const char **address, const char **port)
 468 {
 469     if (*address == NULL) {
 470         *address = "0.0.0.0";
 471     }
 472
 473     if (*port == NULL) {
 474         *port = stringify(NBD_DEFAULT_PORT);
 475     }
 476 }
 477
 478 /*
 479  * Check socket parameters compatibility when socket activation is used.
 480  */
 481 static const char *socket_activation_validate_opts(const char *device,
 482                                                    const char *sockpath,
 483                                                    const char *address,
 484                                                    const char *port)
 485 {
 486     if (device != NULL) {
 487         return "NBD device can't be set when using socket activation";
 488     }
 489
 490     if (sockpath != NULL) {
 491         return "Unix socket can't be set when using socket activation";
 492     }
 493
 494     if (address != NULL) {
 495         return "The interface can't be set when using socket activation";
 496     }
 497
 498     if (port != NULL) {
 499         return "TCP port number can't be set when using socket activation";
 500     }
 501
 502     return NULL;
 503 }
 504
 505 int main(int argc, char **argv)
 506 {
 507     BlockBackend *blk;
 508     BlockDriverState *bs;
 509     off_t dev_offset = 0;
 510     uint16_t nbdflags = 0;
 511     bool disconnect = false;
 512     const char *bindto = NULL;
 513     const char *port = NULL;
 514     char *sockpath = NULL;
 515     char *device = NULL;
 516     off_t fd_size;
 517     QemuOpts *sn_opts = NULL;
 518     const char *sn_id_or_name = NULL;
 519     const char *sopt = "hVb:o:p:rsnP:c:dvk:e:f:tl:x:T:D:";
 520     struct option lopt[] = {
 521         { "help", no_argument, NULL, 'h' },
 522         { "version", no_argument, NULL, 'V' },
 523         { "bind", required_argument, NULL, 'b' },
 524         { "port", required_argument, NULL, 'p' },
 525         { "socket", required_argument, NULL, 'k' },
 526         { "offset", required_argument, NULL, 'o' },
 527         { "read-only", no_argument, NULL, 'r' },
 528         { "partition", required_argument, NULL, 'P' },
 529         { "connect", required_argument, NULL, 'c' },
 530         { "disconnect", no_argument, NULL, 'd' },
 531         { "snapshot", no_argument, NULL, 's' },
 532         { "load-snapshot", required_argument, NULL, 'l' },
 533         { "nocache", no_argument, NULL, 'n' },
 534         { "cache", required_argument, NULL, QEMU_NBD_OPT_CACHE },
 535         { "aio", required_argument, NULL, QEMU_NBD_OPT_AIO },
 536         { "discard", required_argument, NULL, QEMU_NBD_OPT_DISCARD },
 537         { "detect-zeroes", required_argument, NULL,
 538           QEMU_NBD_OPT_DETECT_ZEROES },
 539         { "shared", required_argument, NULL, 'e' },
 540         { "format", required_argument, NULL, 'f' },
 541         { "persistent", no_argument, NULL, 't' },
 542         { "verbose", no_argument, NULL, 'v' },
 543         { "object", required_argument, NULL, QEMU_NBD_OPT_OBJECT },
 544         { "export-name", required_argument, NULL, 'x' },
 545         { "description", required_argument, NULL, 'D' },
 546         { "tls-creds", required_argument, NULL, QEMU_NBD_OPT_TLSCREDS },
 547         { "image-opts", no_argument, NULL, QEMU_NBD_OPT_IMAGE_OPTS },
 548         { "trace", required_argument, NULL, 'T' },
 549         { "fork", no_argument, NULL, QEMU_NBD_OPT_FORK },
 550         { NULL, 0, NULL, 0 }
 551     };
 552     int ch;
 553     int opt_ind = 0;
 554     char *end;
 555     int flags = BDRV_O_RDWR;
 556     int partition = -1;
 557     int ret = 0;
 558     bool seen_cache = false;
 559     bool seen_discard = false;
 560     bool seen_aio = false;
 561     pthread_t client_thread;
 562     const char *fmt = NULL;
 563     Error *local_err = NULL;
 564     BlockdevDetectZeroesOptions detect_zeroes = BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
 565     QDict *options = NULL;
 566     const char *export_name = NULL;
 567     const char *export_description = NULL;
 568     const char *tlscredsid = NULL;
 569     bool imageOpts = false;
 570     bool writethrough = true;
 571     char *trace_file = NULL;
 572     bool fork_process = false;
 573     int old_stderr = -1;
 574     unsigned socket_activation;
 575
 576     /* The client thread uses SIGTERM to interrupt the server.  A signal
 577      * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
 578      */
 579     struct sigaction sa_sigterm;
 580     memset(&sa_sigterm, 0, sizeof(sa_sigterm));
 581     sa_sigterm.sa_handler = termsig_handler;
 582     sigaction(SIGTERM, &sa_sigterm, NULL);
 583
 584     module_call_init(MODULE_INIT_TRACE);
 585     qcrypto_init(&error_fatal);
 586
 587     module_call_init(MODULE_INIT_QOM);
 588     qemu_add_opts(&qemu_object_opts);
 589     qemu_add_opts(&qemu_trace_opts);
 590     qemu_init_exec_dir(argv[0]);
 591
 592     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 593         switch (ch) {
 594         case 's':
 595             flags |= BDRV_O_SNAPSHOT;
 596             break;
 597         case 'n':
 598             optarg = (char *) "none";
 599             /* fallthrough */
 600         case QEMU_NBD_OPT_CACHE:
 601             if (seen_cache) {
 602                 error_report("-n and --cache can only be specified once");
 603                 exit(EXIT_FAILURE);
 604             }
 605             seen_cache = true;
 606             if (bdrv_parse_cache_mode(optarg, &flags, &writethrough) == -1) {
 607                 error_report("Invalid cache mode `%s'", optarg);
 608                 exit(EXIT_FAILURE);
 609             }
 610             break;
 611         case QEMU_NBD_OPT_AIO:
 612             if (seen_aio) {
 613                 error_report("--aio can only be specified once");
 614                 exit(EXIT_FAILURE);
 615             }
 616             seen_aio = true;
 617             if (!strcmp(optarg, "native")) {
 618                 flags |= BDRV_O_NATIVE_AIO;
 619             } else if (!strcmp(optarg, "threads")) {
 620                 /* this is the default */
 621             } else {
 622                error_report("invalid aio mode `%s'", optarg);
 623                exit(EXIT_FAILURE);
 624             }
 625             break;
 626         case QEMU_NBD_OPT_DISCARD:
 627             if (seen_discard) {
 628                 error_report("--discard can only be specified once");
 629                 exit(EXIT_FAILURE);
 630             }
 631             seen_discard = true;
 632             if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
 633                 error_report("Invalid discard mode `%s'", optarg);
 634                 exit(EXIT_FAILURE);
 635             }
 636             break;
 637         case QEMU_NBD_OPT_DETECT_ZEROES:
 638             detect_zeroes =
 639                 qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
 640                                 optarg,
 641                                 BLOCKDEV_DETECT_ZEROES_OPTIONS__MAX,
 642                                 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
 643                                 &local_err);
 644             if (local_err) {
 645                 error_reportf_err(local_err,
 646                                   "Failed to parse detect_zeroes mode: ");
 647                 exit(EXIT_FAILURE);
 648             }
 649             if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
 650                 !(flags & BDRV_O_UNMAP)) {
 651                 error_report("setting detect-zeroes to unmap is not allowed "
 652                              "without setting discard operation to unmap");
 653                 exit(EXIT_FAILURE);
 654             }
 655             break;
 656         case 'b':
 657             bindto = optarg;
 658             break;
 659         case 'p':
 660             port = optarg;
 661             break;
 662         case 'o':
 663                 dev_offset = strtoll (optarg, &end, 0);
 664             if (*end) {
 665                 error_report("Invalid offset `%s'", optarg);
 666                 exit(EXIT_FAILURE);
 667             }
 668             if (dev_offset < 0) {
 669                 error_report("Offset must be positive `%s'", optarg);
 670                 exit(EXIT_FAILURE);
 671             }
 672             break;
 673         case 'l':
 674             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
 675                 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
 676                                                   optarg, false);
 677                 if (!sn_opts) {
 678                     error_report("Failed in parsing snapshot param `%s'",
 679                                  optarg);
 680                     exit(EXIT_FAILURE);
 681                 }
 682             } else {
 683                 sn_id_or_name = optarg;
 684             }
 685             /* fall through */
 686         case 'r':
 687             nbdflags |= NBD_FLAG_READ_ONLY;
 688             flags &= ~BDRV_O_RDWR;
 689             break;
 690         case 'P':
 691             partition = strtol(optarg, &end, 0);
 692             if (*end) {
 693                 error_report("Invalid partition `%s'", optarg);
 694                 exit(EXIT_FAILURE);
 695             }
 696             if (partition < 1 || partition > 8) {
 697                 error_report("Invalid partition %d", partition);
 698                 exit(EXIT_FAILURE);
 699             }
 700             break;
 701         case 'k':
 702             sockpath = optarg;
 703             if (sockpath[0] != '/') {
 704                 error_report("socket path must be absolute");
 705                 exit(EXIT_FAILURE);
 706             }
 707             break;
 708         case 'd':
 709             disconnect = true;
 710             break;
 711         case 'c':
 712             device = optarg;
 713             break;
 714         case 'e':
 715             shared = strtol(optarg, &end, 0);
 716             if (*end) {
 717                 error_report("Invalid shared device number '%s'", optarg);
 718                 exit(EXIT_FAILURE);
 719             }
 720             if (shared < 1) {
 721                 error_report("Shared device number must be greater than 0");
 722                 exit(EXIT_FAILURE);
 723             }
 724             break;
 725         case 'f':
 726             fmt = optarg;
 727             break;
 728         case 't':
 729             persistent = 1;
 730             break;
 731         case 'x':
 732             export_name = optarg;
 733             break;
 734         case 'D':
 735             export_description = optarg;
 736             break;
 737         case 'v':
 738             verbose = 1;
 739             break;
 740         case 'V':
 741             version(argv[0]);
 742             exit(0);
 743             break;
 744         case 'h':
 745             usage(argv[0]);
 746             exit(0);
 747             break;
 748         case '?':
 749             error_report("Try `%s --help' for more information.", argv[0]);
 750             exit(EXIT_FAILURE);
 751         case QEMU_NBD_OPT_OBJECT: {
 752             QemuOpts *opts;
 753             opts = qemu_opts_parse_noisily(&qemu_object_opts,
 754                                            optarg, true);
 755             if (!opts) {
 756                 exit(EXIT_FAILURE);
 757             }
 758         }   break;
 759         case QEMU_NBD_OPT_TLSCREDS:
 760             tlscredsid = optarg;
 761             break;
 762         case QEMU_NBD_OPT_IMAGE_OPTS:
 763             imageOpts = true;
 764             break;
 765         case 'T':
 766             g_free(trace_file);
 767             trace_file = trace_opt_parse(optarg);
 768             break;
 769         case QEMU_NBD_OPT_FORK:
 770             fork_process = true;
 771             break;
 772         }
 773     }
 774
 775     if ((argc - optind) != 1) {
 776         error_report("Invalid number of arguments");
 777         error_printf("Try `%s --help' for more information.\n", argv[0]);
 778         exit(EXIT_FAILURE);
 779     }
 780
 781     if (qemu_opts_foreach(&qemu_object_opts,
 782                           user_creatable_add_opts_foreach,
 783                           NULL, NULL)) {
 784         exit(EXIT_FAILURE);
 785     }
 786
 787     if (!trace_init_backends()) {
 788         exit(1);
 789     }
 790     trace_init_file(trace_file);
 791     qemu_set_log(LOG_TRACE);
 792
 793     socket_activation = check_socket_activation();
 794     if (socket_activation == 0) {
 795         setup_address_and_port(&bindto, &port);
 796     } else {
 797         /* Using socket activation - check user didn't use -p etc. */
 798         const char *err_msg = socket_activation_validate_opts(device, sockpath,
 799                                                               bindto, port);
 800         if (err_msg != NULL) {
 801             error_report("%s", err_msg);
 802             exit(EXIT_FAILURE);
 803         }
 804
 805         /* qemu-nbd can only listen on a single socket.  */
 806         if (socket_activation > 1) {
 807             error_report("qemu-nbd does not support socket activation with %s > 1",
 808                          "LISTEN_FDS");
 809             exit(EXIT_FAILURE);
 810         }
 811     }
 812
 813     if (tlscredsid) {
 814         if (sockpath) {
 815             error_report("TLS is only supported with IPv4/IPv6");
 816             exit(EXIT_FAILURE);
 817         }
 818         if (device) {
 819             error_report("TLS is not supported with a host device");
 820             exit(EXIT_FAILURE);
 821         }
 822         if (!export_name) {
 823             /* Set the default NBD protocol export name, since
 824              * we *must* use new style protocol for TLS */
 825             export_name = "";
 826         }
 827         tlscreds = nbd_get_tls_creds(tlscredsid, &local_err);
 828         if (local_err) {
 829             error_report("Failed to get TLS creds %s",
 830                          error_get_pretty(local_err));
 831             exit(EXIT_FAILURE);
 832         }
 833     }
 834
 835     if (disconnect) {
 836         int nbdfd = open(argv[optind], O_RDWR);
 837         if (nbdfd < 0) {
 838             error_report("Cannot open %s: %s", argv[optind],
 839                          strerror(errno));
 840             exit(EXIT_FAILURE);
 841         }
 842         nbd_disconnect(nbdfd);
 843
 844         close(nbdfd);
 845
 846         printf("%s disconnected\n", argv[optind]);
 847
 848         return 0;
 849     }
 850
 851     if ((device && !verbose) || fork_process) {
 852         int stderr_fd[2];
 853         pid_t pid;
 854         int ret;
 855
 856         if (qemu_pipe(stderr_fd) < 0) {
 857             error_report("Error setting up communication pipe: %s",
 858                          strerror(errno));
 859             exit(EXIT_FAILURE);
 860         }
 861
 862         /* Now daemonize, but keep a communication channel open to
 863          * print errors and exit with the proper status code.
 864          */
 865         pid = fork();
 866         if (pid < 0) {
 867             error_report("Failed to fork: %s", strerror(errno));
 868             exit(EXIT_FAILURE);
 869         } else if (pid == 0) {
 870             close(stderr_fd[0]);
 871             ret = qemu_daemon(1, 0);
 872
 873             /* Temporarily redirect stderr to the parent's pipe...  */
 874             old_stderr = dup(STDERR_FILENO);
 875             dup2(stderr_fd[1], STDERR_FILENO);
 876             if (ret < 0) {
 877                 error_report("Failed to daemonize: %s", strerror(errno));
 878                 exit(EXIT_FAILURE);
 879             }
 880
 881             /* ... close the descriptor we inherited and go on.  */
 882             close(stderr_fd[1]);
 883         } else {
 884             bool errors = false;
 885             char *buf;
 886
 887             /* In the parent.  Print error messages from the child until
 888              * it closes the pipe.
 889              */
 890             close(stderr_fd[1]);
 891             buf = g_malloc(1024);
 892             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
 893                 errors = true;
 894                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
 895                 if (ret < 0) {
 896                     exit(EXIT_FAILURE);
 897                 }
 898             }
 899             if (ret < 0) {
 900                 error_report("Cannot read from daemon: %s",
 901                              strerror(errno));
 902                 exit(EXIT_FAILURE);
 903             }
 904
 905             /* Usually the daemon should not print any message.
 906              * Exit with zero status in that case.
 907              */
 908             exit(errors);
 909         }
 910     }
 911
 912     if (device != NULL && sockpath == NULL) {
 913         sockpath = g_malloc(128);
 914         snprintf(sockpath, 128, SOCKET_PATH, basename(device));
 915     }
 916
 917     if (socket_activation == 0) {
 918         server_ioc = qio_channel_socket_new();
 919         saddr = nbd_build_socket_address(sockpath, bindto, port);
 920         if (qio_channel_socket_listen_sync(server_ioc, saddr, &local_err) < 0) {
 921             object_unref(OBJECT(server_ioc));
 922             error_report_err(local_err);
 923             return 1;
 924         }
 925     } else {
 926         /* See comment in check_socket_activation above. */
 927         assert(socket_activation == 1);
 928         server_ioc = qio_channel_socket_new_fd(FIRST_SOCKET_ACTIVATION_FD,
 929                                                &local_err);
 930         if (server_ioc == NULL) {
 931             error_report("Failed to use socket activation: %s",
 932                          error_get_pretty(local_err));
 933             exit(EXIT_FAILURE);
 934         }
 935     }
 936
 937     if (qemu_init_main_loop(&local_err)) {
 938         error_report_err(local_err);
 939         exit(EXIT_FAILURE);
 940     }
 941     bdrv_init();
 942     atexit(bdrv_close_all);
 943
 944     srcpath = argv[optind];
 945     if (imageOpts) {
 946         QemuOpts *opts;
 947         if (fmt) {
 948             error_report("--image-opts and -f are mutually exclusive");
 949             exit(EXIT_FAILURE);
 950         }
 951         opts = qemu_opts_parse_noisily(&file_opts, srcpath, true);
 952         if (!opts) {
 953             qemu_opts_reset(&file_opts);
 954             exit(EXIT_FAILURE);
 955         }
 956         options = qemu_opts_to_qdict(opts, NULL);
 957         qemu_opts_reset(&file_opts);
 958         blk = blk_new_open(NULL, NULL, options, flags, &local_err);
 959     } else {
 960         if (fmt) {
 961             options = qdict_new();
 962             qdict_put(options, "driver", qstring_from_str(fmt));
 963         }
 964         blk = blk_new_open(srcpath, NULL, options, flags, &local_err);
 965     }
 966
 967     if (!blk) {
 968         error_reportf_err(local_err, "Failed to blk_new_open '%s': ",
 969                           argv[optind]);
 970         exit(EXIT_FAILURE);
 971     }
 972     bs = blk_bs(blk);
 973
 974     blk_set_enable_write_cache(blk, !writethrough);
 975
 976     if (sn_opts) {
 977         ret = bdrv_snapshot_load_tmp(bs,
 978                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
 979                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
 980                                      &local_err);
 981     } else if (sn_id_or_name) {
 982         ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
 983                                                    &local_err);
 984     }
 985     if (ret < 0) {
 986         error_reportf_err(local_err, "Failed to load snapshot: ");
 987         exit(EXIT_FAILURE);
 988     }
 989
 990     bs->detect_zeroes = detect_zeroes;
 991     fd_size = blk_getlength(blk);
 992     if (fd_size < 0) {
 993         error_report("Failed to determine the image length: %s",
 994                      strerror(-fd_size));
 995         exit(EXIT_FAILURE);
 996     }
 997
 998     if (dev_offset >= fd_size) {
 999         error_report("Offset (%lld) has to be smaller than the image size "
1000                      "(%lld)",
1001                      (long long int)dev_offset, (long long int)fd_size);
1002         exit(EXIT_FAILURE);
1003     }
1004     fd_size -= dev_offset;
1005
1006     if (partition != -1) {
1007         ret = find_partition(blk, partition, &dev_offset, &fd_size);
1008         if (ret < 0) {
1009             error_report("Could not find partition %d: %s", partition,
1010                          strerror(-ret));
1011             exit(EXIT_FAILURE);
1012         }
1013     }
1014
1015     exp = nbd_export_new(bs, dev_offset, fd_size, nbdflags, nbd_export_closed,
1016                          writethrough, NULL, &local_err);
1017     if (!exp) {
1018         error_report_err(local_err);
1019         exit(EXIT_FAILURE);
1020     }
1021     if (export_name) {
1022         nbd_export_set_name(exp, export_name);
1023         nbd_export_set_description(exp, export_description);
1024         newproto = true;
1025     } else if (export_description) {
1026         error_report("Export description requires an export name");
1027         exit(EXIT_FAILURE);
1028     }
1029
1030     if (device) {
1031         int ret;
1032
1033         ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
1034         if (ret != 0) {
1035             error_report("Failed to create client thread: %s", strerror(ret));
1036             exit(EXIT_FAILURE);
1037         }
1038     } else {
1039         /* Shut up GCC warnings.  */
1040         memset(&client_thread, 0, sizeof(client_thread));
1041     }
1042
1043     nbd_update_server_watch();
1044
1045     /* now when the initialization is (almost) complete, chdir("/")
1046      * to free any busy filesystems */
1047     if (chdir("/") < 0) {
1048         error_report("Could not chdir to root directory: %s",
1049                      strerror(errno));
1050         exit(EXIT_FAILURE);
1051     }
1052
1053     if (fork_process) {
1054         dup2(old_stderr, STDERR_FILENO);
1055         close(old_stderr);
1056     }
1057
1058     state = RUNNING;
1059     do {
1060         main_loop_wait(false);
1061         if (state == TERMINATE) {
1062             state = TERMINATING;
1063             nbd_export_close(exp);
1064             nbd_export_put(exp);
1065             exp = NULL;
1066         }
1067     } while (state != TERMINATED);
1068
1069     blk_unref(blk);
1070     if (sockpath) {
1071         unlink(sockpath);
1072     }
1073
1074     qemu_opts_del(sn_opts);
1075
1076     if (device) {
1077         void *ret;
1078         pthread_join(client_thread, &ret);
1079         exit(ret != NULL);
1080     } else {
1081         exit(EXIT_SUCCESS);
1082     }
1083 }