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 "qemu/help-texts.h"
  25 #include "qapi/error.h"
  26 #include "qemu/cutils.h"
  27 #include "sysemu/block-backend.h"
  28 #include "sysemu/runstate.h" /* for qemu_system_killed() prototype */
  29 #include "block/block_int.h"
  30 #include "block/nbd.h"
  31 #include "qemu/main-loop.h"
  32 #include "qemu/module.h"
  33 #include "qemu/option.h"
  34 #include "qemu/error-report.h"
  35 #include "qemu/config-file.h"
  36 #include "qemu/bswap.h"
  37 #include "qemu/log.h"
  38 #include "qemu/systemd.h"
  39 #include "block/snapshot.h"
  40 #include "qapi/qmp/qdict.h"
  41 #include "qapi/qmp/qstring.h"
  42 #include "qom/object_interfaces.h"
  43 #include "io/channel-socket.h"
  44 #include "io/net-listener.h"
  45 #include "crypto/init.h"
  46 #include "crypto/tlscreds.h"
  47 #include "trace/control.h"
  48 #include "qemu-version.h"
  49
  50 #ifdef CONFIG_SELINUX
  51 #include <selinux/selinux.h>
  52 #endif
  53
  54 #ifdef __linux__
  55 #define HAVE_NBD_DEVICE 1
  56 #else
  57 #define HAVE_NBD_DEVICE 0
  58 #endif
  59
  60 #define SOCKET_PATH                "/var/lock/qemu-nbd-%s"
  61 #define QEMU_NBD_OPT_CACHE         256
  62 #define QEMU_NBD_OPT_AIO           257
  63 #define QEMU_NBD_OPT_DISCARD       258
  64 #define QEMU_NBD_OPT_DETECT_ZEROES 259
  65 #define QEMU_NBD_OPT_OBJECT        260
  66 #define QEMU_NBD_OPT_TLSCREDS      261
  67 #define QEMU_NBD_OPT_IMAGE_OPTS    262
  68 #define QEMU_NBD_OPT_FORK          263
  69 #define QEMU_NBD_OPT_TLSAUTHZ      264
  70 #define QEMU_NBD_OPT_PID_FILE      265
  71 #define QEMU_NBD_OPT_SELINUX_LABEL 266
  72 #define QEMU_NBD_OPT_TLSHOSTNAME   267
  73
  74 #define MBR_SIZE 512
  75
  76 static char *srcpath;
  77 static SocketAddress *saddr;
  78 static int persistent = 0;
  79 static enum { RUNNING, TERMINATE, TERMINATED } state;
  80 static int shared = 1;
  81 static int nb_fds;
  82 static QIONetListener *server;
  83 static QCryptoTLSCreds *tlscreds;
  84 static const char *tlsauthz;
  85
  86 static void usage(const char *name)
  87 {
  88     (printf) (
  89 "Usage: %s [OPTIONS] FILE\n"
  90 "  or:  %s -L [OPTIONS]\n"
  91 "QEMU Disk Network Block Device Utility\n"
  92 "\n"
  93 "  -h, --help                display this help and exit\n"
  94 "  -V, --version             output version information and exit\n"
  95 "\n"
  96 "Connection properties:\n"
  97 "  -p, --port=PORT           port to listen on (default `%d')\n"
  98 "  -b, --bind=IFACE          interface to bind to (default `0.0.0.0')\n"
  99 "  -k, --socket=PATH         path to the unix socket\n"
 100 "                            (default '"SOCKET_PATH"')\n"
 101 "  -e, --shared=NUM          device can be shared by NUM clients (default '1')\n"
 102 "  -t, --persistent          don't exit on the last connection\n"
 103 "  -v, --verbose             display extra debugging information\n"
 104 "  -x, --export-name=NAME    expose export by name (default is empty string)\n"
 105 "  -D, --description=TEXT    export a human-readable description\n"
 106 "\n"
 107 "Exposing part of the image:\n"
 108 "  -o, --offset=OFFSET       offset into the image\n"
 109 "  -A, --allocation-depth    expose the allocation depth\n"
 110 "  -B, --bitmap=NAME         expose a persistent dirty bitmap\n"
 111 "\n"
 112 "General purpose options:\n"
 113 "  -L, --list                list exports available from another NBD server\n"
 114 "  --object type,id=ID,...   define an object such as 'secret' for providing\n"
 115 "                            passwords and/or encryption keys\n"
 116 "  --tls-creds=ID            use id of an earlier --object to provide TLS\n"
 117 "  --tls-authz=ID            use id of an earlier --object to provide\n"
 118 "                            authorization\n"
 119 "  -T, --trace [[enable=]<pattern>][,events=<file>][,file=<file>]\n"
 120 "                            specify tracing options\n"
 121 "  --fork                    fork off the server process and exit the parent\n"
 122 "                            once the server is running\n"
 123 "  --pid-file=PATH           store the server's process ID in the given file\n"
 124 #ifdef CONFIG_SELINUX
 125 "  --selinux-label=LABEL     set SELinux process label on listening socket\n"
 126 #endif
 127 #if HAVE_NBD_DEVICE
 128 "\n"
 129 "Kernel NBD client support:\n"
 130 "  -c, --connect=DEV         connect FILE to the local NBD device DEV\n"
 131 "  -d, --disconnect          disconnect the specified device\n"
 132 #endif
 133 "\n"
 134 "Block device options:\n"
 135 "  -f, --format=FORMAT       set image format (raw, qcow2, ...)\n"
 136 "  -r, --read-only           export read-only\n"
 137 "  -s, --snapshot            use FILE as an external snapshot, create a temporary\n"
 138 "                            file with backing_file=FILE, redirect the write to\n"
 139 "                            the temporary one\n"
 140 "  -l, --load-snapshot=SNAPSHOT_PARAM\n"
 141 "                            load an internal snapshot inside FILE and export it\n"
 142 "                            as an read-only device, SNAPSHOT_PARAM format is\n"
 143 "                            'snapshot.id=[ID],snapshot.name=[NAME]', or\n"
 144 "                            '[ID_OR_NAME]'\n"
 145 "  -n, --nocache             disable host cache\n"
 146 "      --cache=MODE          set cache mode used to access the disk image, the\n"
 147 "                            valid options are: 'none', 'writeback' (default),\n"
 148 "                            'writethrough', 'directsync' and 'unsafe'\n"
 149 "      --aio=MODE            set AIO mode (native, io_uring or threads)\n"
 150 "      --discard=MODE        set discard mode (ignore, unmap)\n"
 151 "      --detect-zeroes=MODE  set detect-zeroes mode (off, on, unmap)\n"
 152 "      --image-opts          treat FILE as a full set of image options\n"
 153 "\n"
 154 QEMU_HELP_BOTTOM "\n"
 155     , name, name, NBD_DEFAULT_PORT, "DEVICE");
 156 }
 157
 158 static void version(const char *name)
 159 {
 160     printf(
 161 "%s " QEMU_FULL_VERSION "\n"
 162 "Written by Anthony Liguori.\n"
 163 "\n"
 164 QEMU_COPYRIGHT "\n"
 165 "This is free software; see the source for copying conditions.  There is NO\n"
 166 "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
 167     , name);
 168 }
 169
 170 #ifdef CONFIG_POSIX
 171 /*
 172  * The client thread uses SIGTERM to interrupt the server.  A signal
 173  * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
 174  */
 175 void qemu_system_killed(int signum, pid_t pid)
 176 {
 177     qatomic_cmpxchg(&state, RUNNING, TERMINATE);
 178     qemu_notify_event();
 179 }
 180 #endif /* CONFIG_POSIX */
 181
 182 static int qemu_nbd_client_list(SocketAddress *saddr, QCryptoTLSCreds *tls,
 183                                 const char *hostname)
 184 {
 185     int ret = EXIT_FAILURE;
 186     int rc;
 187     Error *err = NULL;
 188     QIOChannelSocket *sioc;
 189     NBDExportInfo *list;
 190     int i, j;
 191
 192     sioc = qio_channel_socket_new();
 193     if (qio_channel_socket_connect_sync(sioc, saddr, &err) < 0) {
 194         error_report_err(err);
 195         goto out;
 196     }
 197     rc = nbd_receive_export_list(QIO_CHANNEL(sioc), tls, hostname, &list,
 198                                  &err);
 199     if (rc < 0) {
 200         if (err) {
 201             error_report_err(err);
 202         }
 203         goto out;
 204     }
 205     printf("exports available: %d\n", rc);
 206     for (i = 0; i < rc; i++) {
 207         printf(" export: '%s'\n", list[i].name);
 208         if (list[i].description && *list[i].description) {
 209             printf("  description: %s\n", list[i].description);
 210         }
 211         if (list[i].flags & NBD_FLAG_HAS_FLAGS) {
 212             static const char *const flag_names[] = {
 213                 [NBD_FLAG_READ_ONLY_BIT]            = "readonly",
 214                 [NBD_FLAG_SEND_FLUSH_BIT]           = "flush",
 215                 [NBD_FLAG_SEND_FUA_BIT]             = "fua",
 216                 [NBD_FLAG_ROTATIONAL_BIT]           = "rotational",
 217                 [NBD_FLAG_SEND_TRIM_BIT]            = "trim",
 218                 [NBD_FLAG_SEND_WRITE_ZEROES_BIT]    = "zeroes",
 219                 [NBD_FLAG_SEND_DF_BIT]              = "df",
 220                 [NBD_FLAG_CAN_MULTI_CONN_BIT]       = "multi",
 221                 [NBD_FLAG_SEND_RESIZE_BIT]          = "resize",
 222                 [NBD_FLAG_SEND_CACHE_BIT]           = "cache",
 223                 [NBD_FLAG_SEND_FAST_ZERO_BIT]       = "fast-zero",
 224                 [NBD_FLAG_BLOCK_STAT_PAYLOAD_BIT]   = "block-status-payload",
 225             };
 226
 227             printf("  size:  %" PRIu64 "\n", list[i].size);
 228             printf("  flags: 0x%x (", list[i].flags);
 229             for (size_t bit = 0; bit < ARRAY_SIZE(flag_names); bit++) {
 230                 if (flag_names[bit] && (list[i].flags & (1 << bit))) {
 231                     printf(" %s", flag_names[bit]);
 232                 }
 233             }
 234             printf(" )\n");
 235         }
 236         if (list[i].min_block) {
 237             printf("  min block: %u\n", list[i].min_block);
 238             printf("  opt block: %u\n", list[i].opt_block);
 239             printf("  max block: %u\n", list[i].max_block);
 240         }
 241         printf("  transaction size: %s\n",
 242                list[i].mode >= NBD_MODE_EXTENDED ?
 243                "64-bit" : "32-bit");
 244         if (list[i].n_contexts) {
 245             printf("  available meta contexts: %d\n", list[i].n_contexts);
 246             for (j = 0; j < list[i].n_contexts; j++) {
 247                 printf("   %s\n", list[i].contexts[j]);
 248             }
 249         }
 250     }
 251     nbd_free_export_list(list, rc);
 252
 253     ret = EXIT_SUCCESS;
 254  out:
 255     object_unref(OBJECT(sioc));
 256     return ret;
 257 }
 258
 259
 260 #if HAVE_NBD_DEVICE
 261 static void *show_parts(void *arg)
 262 {
 263     char *device = arg;
 264     int nbd;
 265
 266     /* linux just needs an open() to trigger
 267      * the partition table update
 268      * but remember to load the module with max_part != 0 :
 269      *     modprobe nbd max_part=63
 270      */
 271     nbd = open(device, O_RDWR);
 272     if (nbd >= 0) {
 273         close(nbd);
 274     }
 275     return NULL;
 276 }
 277
 278 struct NbdClientOpts {
 279     char *device;
 280     bool fork_process;
 281     bool verbose;
 282 };
 283
 284 static void *nbd_client_thread(void *arg)
 285 {
 286     struct NbdClientOpts *opts = arg;
 287     /* TODO: Revisit this if nbd.ko ever gains support for structured reply */
 288     NBDExportInfo info = { .request_sizes = false, .name = g_strdup(""),
 289                            .mode = NBD_MODE_SIMPLE };
 290     QIOChannelSocket *sioc;
 291     int fd = -1;
 292     int ret = EXIT_FAILURE;
 293     pthread_t show_parts_thread;
 294     Error *local_error = NULL;
 295
 296     sioc = qio_channel_socket_new();
 297     if (qio_channel_socket_connect_sync(sioc,
 298                                         saddr,
 299                                         &local_error) < 0) {
 300         error_report_err(local_error);
 301         goto out;
 302     }
 303
 304     if (nbd_receive_negotiate(NULL, QIO_CHANNEL(sioc),
 305                               NULL, NULL, NULL, &info, &local_error) < 0) {
 306         if (local_error) {
 307             error_report_err(local_error);
 308         }
 309         goto out;
 310     }
 311
 312     fd = open(opts->device, O_RDWR);
 313     if (fd < 0) {
 314         /* Linux-only, we can use %m in printf.  */
 315         error_report("Failed to open %s: %m", opts->device);
 316         goto out;
 317     }
 318
 319     if (nbd_init(fd, sioc, &info, &local_error) < 0) {
 320         error_report_err(local_error);
 321         goto out;
 322     }
 323
 324     /* update partition table */
 325     pthread_create(&show_parts_thread, NULL, show_parts, opts->device);
 326
 327     if (opts->verbose && !opts->fork_process) {
 328         fprintf(stderr, "NBD device %s is now connected to %s\n",
 329                 opts->device, srcpath);
 330     } else {
 331         /* Close stderr so that the qemu-nbd process exits.  */
 332         if (dup2(STDOUT_FILENO, STDERR_FILENO) < 0) {
 333             error_report("Could not set stderr to /dev/null: %s",
 334                          strerror(errno));
 335             exit(EXIT_FAILURE);
 336         }
 337     }
 338
 339     if (nbd_client(fd) < 0) {
 340         goto out;
 341     }
 342
 343     ret = EXIT_SUCCESS;
 344
 345  out:
 346     if (fd >= 0) {
 347         close(fd);
 348     }
 349     object_unref(OBJECT(sioc));
 350     g_free(info.name);
 351     kill(getpid(), SIGTERM);
 352     return (void *) (intptr_t) ret;
 353 }
 354 #endif /* HAVE_NBD_DEVICE */
 355
 356 static int nbd_can_accept(void)
 357 {
 358     return state == RUNNING && (shared == 0 || nb_fds < shared);
 359 }
 360
 361 static void nbd_update_server_watch(void);
 362
 363 static void nbd_client_closed(NBDClient *client, bool negotiated)
 364 {
 365     nb_fds--;
 366     if (negotiated && nb_fds == 0 && !persistent && state == RUNNING) {
 367         state = TERMINATE;
 368     }
 369     nbd_update_server_watch();
 370     nbd_client_put(client);
 371 }
 372
 373 static void nbd_accept(QIONetListener *listener, QIOChannelSocket *cioc,
 374                        gpointer opaque)
 375 {
 376     if (state >= TERMINATE) {
 377         return;
 378     }
 379
 380     nb_fds++;
 381     nbd_update_server_watch();
 382     nbd_client_new(cioc, tlscreds, tlsauthz, nbd_client_closed);
 383 }
 384
 385 static void nbd_update_server_watch(void)
 386 {
 387     if (nbd_can_accept()) {
 388         qio_net_listener_set_client_func(server, nbd_accept, NULL, NULL);
 389     } else {
 390         qio_net_listener_set_client_func(server, NULL, NULL, NULL);
 391     }
 392 }
 393
 394
 395 static SocketAddress *nbd_build_socket_address(const char *sockpath,
 396                                                const char *bindto,
 397                                                const char *port)
 398 {
 399     SocketAddress *saddr;
 400
 401     saddr = g_new0(SocketAddress, 1);
 402     if (sockpath) {
 403         saddr->type = SOCKET_ADDRESS_TYPE_UNIX;
 404         saddr->u.q_unix.path = g_strdup(sockpath);
 405     } else {
 406         InetSocketAddress *inet;
 407         saddr->type = SOCKET_ADDRESS_TYPE_INET;
 408         inet = &saddr->u.inet;
 409         inet->host = g_strdup(bindto);
 410         if (port) {
 411             inet->port = g_strdup(port);
 412         } else  {
 413             inet->port = g_strdup_printf("%d", NBD_DEFAULT_PORT);
 414         }
 415     }
 416
 417     return saddr;
 418 }
 419
 420
 421 static QemuOptsList file_opts = {
 422     .name = "file",
 423     .implied_opt_name = "file",
 424     .head = QTAILQ_HEAD_INITIALIZER(file_opts.head),
 425     .desc = {
 426         /* no elements => accept any params */
 427         { /* end of list */ }
 428     },
 429 };
 430
 431 static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, bool list,
 432                                           Error **errp)
 433 {
 434     Object *obj;
 435     QCryptoTLSCreds *creds;
 436
 437     obj = object_resolve_path_component(
 438         object_get_objects_root(), id);
 439     if (!obj) {
 440         error_setg(errp, "No TLS credentials with id '%s'",
 441                    id);
 442         return NULL;
 443     }
 444     creds = (QCryptoTLSCreds *)
 445         object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
 446     if (!creds) {
 447         error_setg(errp, "Object with id '%s' is not TLS credentials",
 448                    id);
 449         return NULL;
 450     }
 451
 452     if (!qcrypto_tls_creds_check_endpoint(creds,
 453                                           list
 454                                           ? QCRYPTO_TLS_CREDS_ENDPOINT_CLIENT
 455                                           : QCRYPTO_TLS_CREDS_ENDPOINT_SERVER,
 456                                           errp)) {
 457         return NULL;
 458     }
 459     object_ref(obj);
 460     return creds;
 461 }
 462
 463 static void setup_address_and_port(const char **address, const char **port)
 464 {
 465     if (*address == NULL) {
 466         *address = "0.0.0.0";
 467     }
 468
 469     if (*port == NULL) {
 470         *port = stringify(NBD_DEFAULT_PORT);
 471     }
 472 }
 473
 474 /*
 475  * Check socket parameters compatibility when socket activation is used.
 476  */
 477 static const char *socket_activation_validate_opts(const char *device,
 478                                                    const char *sockpath,
 479                                                    const char *address,
 480                                                    const char *port,
 481                                                    const char *selinux,
 482                                                    bool list)
 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     if (selinux != NULL) {
 501         return "SELinux label can't be set when using socket activation";
 502     }
 503
 504     if (list) {
 505         return "List mode is incompatible with socket activation";
 506     }
 507
 508     return NULL;
 509 }
 510
 511 static void qemu_nbd_shutdown(void)
 512 {
 513     job_cancel_sync_all();
 514     blk_exp_close_all();
 515     bdrv_close_all();
 516 }
 517
 518 int main(int argc, char **argv)
 519 {
 520     BlockBackend *blk;
 521     BlockDriverState *bs;
 522     uint64_t dev_offset = 0;
 523     bool readonly = false;
 524     bool disconnect = false;
 525     const char *bindto = NULL;
 526     const char *port = NULL;
 527     char *sockpath = NULL;
 528     char *device = NULL;
 529     QemuOpts *sn_opts = NULL;
 530     const char *sn_id_or_name = NULL;
 531     const char *sopt = "hVb:o:p:rsnc:dvk:e:f:tl:x:T:D:AB:L";
 532     struct option lopt[] = {
 533         { "help", no_argument, NULL, 'h' },
 534         { "version", no_argument, NULL, 'V' },
 535         { "bind", required_argument, NULL, 'b' },
 536         { "port", required_argument, NULL, 'p' },
 537         { "socket", required_argument, NULL, 'k' },
 538         { "offset", required_argument, NULL, 'o' },
 539         { "read-only", no_argument, NULL, 'r' },
 540         { "allocation-depth", no_argument, NULL, 'A' },
 541         { "bitmap", required_argument, NULL, 'B' },
 542         { "connect", required_argument, NULL, 'c' },
 543         { "disconnect", no_argument, NULL, 'd' },
 544         { "list", no_argument, NULL, 'L' },
 545         { "snapshot", no_argument, NULL, 's' },
 546         { "load-snapshot", required_argument, NULL, 'l' },
 547         { "nocache", no_argument, NULL, 'n' },
 548         { "cache", required_argument, NULL, QEMU_NBD_OPT_CACHE },
 549         { "aio", required_argument, NULL, QEMU_NBD_OPT_AIO },
 550         { "discard", required_argument, NULL, QEMU_NBD_OPT_DISCARD },
 551         { "detect-zeroes", required_argument, NULL,
 552           QEMU_NBD_OPT_DETECT_ZEROES },
 553         { "shared", required_argument, NULL, 'e' },
 554         { "format", required_argument, NULL, 'f' },
 555         { "persistent", no_argument, NULL, 't' },
 556         { "verbose", no_argument, NULL, 'v' },
 557         { "object", required_argument, NULL, QEMU_NBD_OPT_OBJECT },
 558         { "export-name", required_argument, NULL, 'x' },
 559         { "description", required_argument, NULL, 'D' },
 560         { "tls-creds", required_argument, NULL, QEMU_NBD_OPT_TLSCREDS },
 561         { "tls-hostname", required_argument, NULL, QEMU_NBD_OPT_TLSHOSTNAME },
 562         { "tls-authz", required_argument, NULL, QEMU_NBD_OPT_TLSAUTHZ },
 563         { "image-opts", no_argument, NULL, QEMU_NBD_OPT_IMAGE_OPTS },
 564         { "trace", required_argument, NULL, 'T' },
 565         { "fork", no_argument, NULL, QEMU_NBD_OPT_FORK },
 566         { "pid-file", required_argument, NULL, QEMU_NBD_OPT_PID_FILE },
 567         { "selinux-label", required_argument, NULL,
 568           QEMU_NBD_OPT_SELINUX_LABEL },
 569         { NULL, 0, NULL, 0 }
 570     };
 571     int ch;
 572     int opt_ind = 0;
 573     int flags = BDRV_O_RDWR;
 574     int ret = 0;
 575     bool seen_cache = false;
 576     bool seen_discard = false;
 577     bool seen_aio = false;
 578     pthread_t client_thread;
 579     const char *fmt = NULL;
 580     Error *local_err = NULL;
 581     BlockdevDetectZeroesOptions detect_zeroes = BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
 582     QDict *options = NULL;
 583     const char *export_name = NULL; /* defaults to "" later for server mode */
 584     const char *export_description = NULL;
 585     BlockDirtyBitmapOrStrList *bitmaps = NULL;
 586     bool alloc_depth = false;
 587     const char *tlscredsid = NULL;
 588     const char *tlshostname = NULL;
 589     bool imageOpts = false;
 590     bool writethrough = false; /* Client will flush as needed. */
 591     bool verbose = false;
 592     bool fork_process = false;
 593     bool list = false;
 594     unsigned socket_activation;
 595     const char *pid_file_name = NULL;
 596     const char *selinux_label = NULL;
 597     BlockExportOptions *export_opts;
 598 #if HAVE_NBD_DEVICE
 599     struct NbdClientOpts opts;
 600 #endif
 601
 602 #ifdef CONFIG_POSIX
 603     os_setup_early_signal_handling();
 604     os_setup_signal_handling();
 605 #endif
 606
 607     socket_init();
 608     error_init(argv[0]);
 609     module_call_init(MODULE_INIT_TRACE);
 610     qcrypto_init(&error_fatal);
 611
 612     module_call_init(MODULE_INIT_QOM);
 613     qemu_add_opts(&qemu_trace_opts);
 614     qemu_init_exec_dir(argv[0]);
 615
 616     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 617         switch (ch) {
 618         case 's':
 619             flags |= BDRV_O_SNAPSHOT;
 620             break;
 621         case 'n':
 622             optarg = (char *) "none";
 623             /* fallthrough */
 624         case QEMU_NBD_OPT_CACHE:
 625             if (seen_cache) {
 626                 error_report("-n and --cache can only be specified once");
 627                 exit(EXIT_FAILURE);
 628             }
 629             seen_cache = true;
 630             if (bdrv_parse_cache_mode(optarg, &flags, &writethrough) == -1) {
 631                 error_report("Invalid cache mode `%s'", optarg);
 632                 exit(EXIT_FAILURE);
 633             }
 634             break;
 635         case QEMU_NBD_OPT_AIO:
 636             if (seen_aio) {
 637                 error_report("--aio can only be specified once");
 638                 exit(EXIT_FAILURE);
 639             }
 640             seen_aio = true;
 641             if (bdrv_parse_aio(optarg, &flags) < 0) {
 642                 error_report("Invalid aio mode '%s'", optarg);
 643                 exit(EXIT_FAILURE);
 644             }
 645             break;
 646         case QEMU_NBD_OPT_DISCARD:
 647             if (seen_discard) {
 648                 error_report("--discard can only be specified once");
 649                 exit(EXIT_FAILURE);
 650             }
 651             seen_discard = true;
 652             if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
 653                 error_report("Invalid discard mode `%s'", optarg);
 654                 exit(EXIT_FAILURE);
 655             }
 656             break;
 657         case QEMU_NBD_OPT_DETECT_ZEROES:
 658             detect_zeroes =
 659                 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
 660                                 optarg,
 661                                 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
 662                                 &local_err);
 663             if (local_err) {
 664                 error_reportf_err(local_err,
 665                                   "Failed to parse detect_zeroes mode: ");
 666                 exit(EXIT_FAILURE);
 667             }
 668             if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
 669                 !(flags & BDRV_O_UNMAP)) {
 670                 error_report("setting detect-zeroes to unmap is not allowed "
 671                              "without setting discard operation to unmap");
 672                 exit(EXIT_FAILURE);
 673             }
 674             break;
 675         case 'b':
 676             bindto = optarg;
 677             break;
 678         case 'p':
 679             port = optarg;
 680             break;
 681         case 'o':
 682             if (qemu_strtou64(optarg, NULL, 0, &dev_offset) < 0) {
 683                 error_report("Invalid offset '%s'", optarg);
 684                 exit(EXIT_FAILURE);
 685             }
 686             break;
 687         case 'l':
 688             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
 689                 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
 690                                                   optarg, false);
 691                 if (!sn_opts) {
 692                     error_report("Failed in parsing snapshot param `%s'",
 693                                  optarg);
 694                     exit(EXIT_FAILURE);
 695                 }
 696             } else {
 697                 sn_id_or_name = optarg;
 698             }
 699             /* fall through */
 700         case 'r':
 701             readonly = true;
 702             flags &= ~BDRV_O_RDWR;
 703             break;
 704         case 'A':
 705             alloc_depth = true;
 706             break;
 707         case 'B':
 708             {
 709                 BlockDirtyBitmapOrStr *el = g_new(BlockDirtyBitmapOrStr, 1);
 710                 *el = (BlockDirtyBitmapOrStr) {
 711                     .type = QTYPE_QSTRING,
 712                     .u.local = g_strdup(optarg),
 713                 };
 714                 QAPI_LIST_PREPEND(bitmaps, el);
 715             }
 716             break;
 717         case 'k':
 718             sockpath = optarg;
 719             if (sockpath[0] != '/') {
 720                 error_report("socket path must be absolute");
 721                 exit(EXIT_FAILURE);
 722             }
 723             break;
 724         case 'd':
 725             disconnect = true;
 726             break;
 727         case 'c':
 728             device = optarg;
 729             break;
 730         case 'e':
 731             if (qemu_strtoi(optarg, NULL, 0, &shared) < 0 ||
 732                 shared < 0) {
 733                 error_report("Invalid shared device number '%s'", optarg);
 734                 exit(EXIT_FAILURE);
 735             }
 736             break;
 737         case 'f':
 738             fmt = optarg;
 739             break;
 740         case 't':
 741             persistent = 1;
 742             break;
 743         case 'x':
 744             export_name = optarg;
 745             if (strlen(export_name) > NBD_MAX_STRING_SIZE) {
 746                 error_report("export name '%s' too long", export_name);
 747                 exit(EXIT_FAILURE);
 748             }
 749             break;
 750         case 'D':
 751             export_description = optarg;
 752             if (strlen(export_description) > NBD_MAX_STRING_SIZE) {
 753                 error_report("export description '%s' too long",
 754                              export_description);
 755                 exit(EXIT_FAILURE);
 756             }
 757             break;
 758         case 'v':
 759             verbose = true;
 760             break;
 761         case 'V':
 762             version(argv[0]);
 763             exit(0);
 764             break;
 765         case 'h':
 766             usage(argv[0]);
 767             exit(0);
 768             break;
 769         case '?':
 770             error_report("Try `%s --help' for more information.", argv[0]);
 771             exit(EXIT_FAILURE);
 772         case QEMU_NBD_OPT_OBJECT:
 773             user_creatable_process_cmdline(optarg);
 774             break;
 775         case QEMU_NBD_OPT_TLSCREDS:
 776             tlscredsid = optarg;
 777             break;
 778         case QEMU_NBD_OPT_TLSHOSTNAME:
 779             tlshostname = optarg;
 780             break;
 781         case QEMU_NBD_OPT_IMAGE_OPTS:
 782             imageOpts = true;
 783             break;
 784         case 'T':
 785             trace_opt_parse(optarg);
 786             break;
 787         case QEMU_NBD_OPT_TLSAUTHZ:
 788             tlsauthz = optarg;
 789             break;
 790         case QEMU_NBD_OPT_FORK:
 791             fork_process = true;
 792             break;
 793         case 'L':
 794             list = true;
 795             break;
 796         case QEMU_NBD_OPT_PID_FILE:
 797             pid_file_name = optarg;
 798             break;
 799         case QEMU_NBD_OPT_SELINUX_LABEL:
 800             selinux_label = optarg;
 801             break;
 802         }
 803     }
 804
 805     if (list) {
 806         if (argc != optind) {
 807             error_report("List mode is incompatible with a file name");
 808             exit(EXIT_FAILURE);
 809         }
 810         if (export_name || export_description || dev_offset ||
 811             device || disconnect || fmt || sn_id_or_name || bitmaps ||
 812             alloc_depth || seen_aio || seen_discard || seen_cache) {
 813             error_report("List mode is incompatible with per-device settings");
 814             exit(EXIT_FAILURE);
 815         }
 816         if (fork_process) {
 817             error_report("List mode is incompatible with forking");
 818             exit(EXIT_FAILURE);
 819         }
 820     } else if ((argc - optind) != 1) {
 821         error_report("Invalid number of arguments");
 822         error_printf("Try `%s --help' for more information.\n", argv[0]);
 823         exit(EXIT_FAILURE);
 824     } else if (!export_name) {
 825         export_name = "";
 826     }
 827
 828     if (!trace_init_backends()) {
 829         exit(1);
 830     }
 831     trace_init_file();
 832     qemu_set_log(LOG_TRACE, &error_fatal);
 833
 834     socket_activation = check_socket_activation();
 835     if (socket_activation == 0) {
 836         if (!sockpath) {
 837             setup_address_and_port(&bindto, &port);
 838         }
 839     } else {
 840         /* Using socket activation - check user didn't use -p etc. */
 841         const char *err_msg = socket_activation_validate_opts(device, sockpath,
 842                                                               bindto, port,
 843                                                               selinux_label,
 844                                                               list);
 845         if (err_msg != NULL) {
 846             error_report("%s", err_msg);
 847             exit(EXIT_FAILURE);
 848         }
 849
 850         /* qemu-nbd can only listen on a single socket.  */
 851         if (socket_activation > 1) {
 852             error_report("qemu-nbd does not support socket activation with %s > 1",
 853                          "LISTEN_FDS");
 854             exit(EXIT_FAILURE);
 855         }
 856     }
 857
 858     if (tlscredsid) {
 859         if (device) {
 860             error_report("TLS is not supported with a host device");
 861             exit(EXIT_FAILURE);
 862         }
 863         if (tlsauthz && list) {
 864             error_report("TLS authorization is incompatible with export list");
 865             exit(EXIT_FAILURE);
 866         }
 867         if (tlshostname && !list) {
 868             error_report("TLS hostname is only supported with export list");
 869             exit(EXIT_FAILURE);
 870         }
 871         tlscreds = nbd_get_tls_creds(tlscredsid, list, &local_err);
 872         if (local_err) {
 873             error_reportf_err(local_err, "Failed to get TLS creds: ");
 874             exit(EXIT_FAILURE);
 875         }
 876     } else {
 877         if (tlsauthz) {
 878             error_report("--tls-authz is not permitted without --tls-creds");
 879             exit(EXIT_FAILURE);
 880         }
 881         if (tlshostname) {
 882             error_report("--tls-hostname is not permitted without --tls-creds");
 883             exit(EXIT_FAILURE);
 884         }
 885     }
 886
 887     if (selinux_label) {
 888 #ifdef CONFIG_SELINUX
 889         if (sockpath == NULL && device == NULL) {
 890             error_report("--selinux-label is not permitted without --socket");
 891             exit(EXIT_FAILURE);
 892         }
 893 #else
 894         error_report("SELinux support not enabled in this binary");
 895         exit(EXIT_FAILURE);
 896 #endif
 897     }
 898
 899     if (list) {
 900         saddr = nbd_build_socket_address(sockpath, bindto, port);
 901         return qemu_nbd_client_list(saddr, tlscreds,
 902                                     tlshostname ? tlshostname : bindto);
 903     }
 904
 905 #if !HAVE_NBD_DEVICE
 906     if (disconnect || device) {
 907         error_report("Kernel /dev/nbdN support not available");
 908         exit(EXIT_FAILURE);
 909     }
 910 #else /* HAVE_NBD_DEVICE */
 911     if (disconnect) {
 912         int nbdfd = open(argv[optind], O_RDWR);
 913         if (nbdfd < 0) {
 914             error_report("Cannot open %s: %s", argv[optind],
 915                          strerror(errno));
 916             exit(EXIT_FAILURE);
 917         }
 918         nbd_disconnect(nbdfd);
 919
 920         close(nbdfd);
 921
 922         printf("%s disconnected\n", argv[optind]);
 923
 924         return 0;
 925     }
 926 #endif
 927
 928     if ((device && !verbose) || fork_process) {
 929 #ifndef WIN32
 930         g_autoptr(GError) err = NULL;
 931         int stderr_fd[2];
 932         pid_t pid;
 933         int ret;
 934
 935         if (!g_unix_open_pipe(stderr_fd, FD_CLOEXEC, &err)) {
 936             error_report("Error setting up communication pipe: %s",
 937                          err->message);
 938             exit(EXIT_FAILURE);
 939         }
 940
 941         /* Now daemonize, but keep a communication channel open to
 942          * print errors and exit with the proper status code.
 943          */
 944         pid = fork();
 945         if (pid < 0) {
 946             error_report("Failed to fork: %s", strerror(errno));
 947             exit(EXIT_FAILURE);
 948         } else if (pid == 0) {
 949             int saved_errno;
 950
 951             close(stderr_fd[0]);
 952
 953             ret = qemu_daemon(1, 0);
 954             saved_errno = errno;    /* dup2 will overwrite error below */
 955
 956             /* Temporarily redirect stderr to the parent's pipe...  */
 957             if (dup2(stderr_fd[1], STDERR_FILENO) < 0) {
 958                 char str[256];
 959                 snprintf(str, sizeof(str),
 960                          "%s: Failed to link stderr to the pipe: %s\n",
 961                          g_get_prgname(), strerror(errno));
 962                 /*
 963                  * We are unable to use error_report() here as we need to get
 964                  * stderr pointed to the parent's pipe. Write to that pipe
 965                  * manually.
 966                  */
 967                 ret = write(stderr_fd[1], str, strlen(str));
 968                 exit(EXIT_FAILURE);
 969             }
 970
 971             if (ret < 0) {
 972                 error_report("Failed to daemonize: %s", strerror(saved_errno));
 973                 exit(EXIT_FAILURE);
 974             }
 975
 976             /* ... close the descriptor we inherited and go on.  */
 977             close(stderr_fd[1]);
 978         } else {
 979             bool errors = false;
 980             char *buf;
 981
 982             /* In the parent.  Print error messages from the child until
 983              * it closes the pipe.
 984              */
 985             close(stderr_fd[1]);
 986             buf = g_malloc(1024);
 987             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
 988                 errors = true;
 989                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
 990                 if (ret < 0) {
 991                     exit(EXIT_FAILURE);
 992                 }
 993             }
 994             if (ret < 0) {
 995                 error_report("Cannot read from daemon: %s",
 996                              strerror(errno));
 997                 exit(EXIT_FAILURE);
 998             }
 999
1000             /* Usually the daemon should not print any message.
1001              * Exit with zero status in that case.
1002              */
1003             exit(errors);
1004         }
1005 #else /* WIN32 */
1006         error_report("Unable to fork into background on Windows hosts");
1007         exit(EXIT_FAILURE);
1008 #endif /* WIN32 */
1009     }
1010
1011     if (device != NULL && sockpath == NULL) {
1012         sockpath = g_malloc(128);
1013         snprintf(sockpath, 128, SOCKET_PATH, basename(device));
1014     }
1015
1016     server = qio_net_listener_new();
1017     if (socket_activation == 0) {
1018         int backlog;
1019
1020         if (persistent || shared == 0) {
1021             backlog = SOMAXCONN;
1022         } else {
1023             backlog = MIN(shared, SOMAXCONN);
1024         }
1025 #ifdef CONFIG_SELINUX
1026         if (selinux_label && setsockcreatecon_raw(selinux_label) == -1) {
1027             error_report("Cannot set SELinux socket create context to %s: %s",
1028                          selinux_label, strerror(errno));
1029             exit(EXIT_FAILURE);
1030         }
1031 #endif
1032         saddr = nbd_build_socket_address(sockpath, bindto, port);
1033         if (qio_net_listener_open_sync(server, saddr, backlog,
1034                                        &local_err) < 0) {
1035             object_unref(OBJECT(server));
1036             error_report_err(local_err);
1037             exit(EXIT_FAILURE);
1038         }
1039 #ifdef CONFIG_SELINUX
1040         if (selinux_label && setsockcreatecon_raw(NULL) == -1) {
1041             error_report("Cannot clear SELinux socket create context: %s",
1042                          strerror(errno));
1043             exit(EXIT_FAILURE);
1044         }
1045 #endif
1046     } else {
1047         size_t i;
1048         /* See comment in check_socket_activation above. */
1049         for (i = 0; i < socket_activation; i++) {
1050             QIOChannelSocket *sioc;
1051             sioc = qio_channel_socket_new_fd(FIRST_SOCKET_ACTIVATION_FD + i,
1052                                              &local_err);
1053             if (sioc == NULL) {
1054                 object_unref(OBJECT(server));
1055                 error_reportf_err(local_err,
1056                                   "Failed to use socket activation: ");
1057                 exit(EXIT_FAILURE);
1058             }
1059             qio_net_listener_add(server, sioc);
1060             object_unref(OBJECT(sioc));
1061         }
1062     }
1063
1064     qemu_init_main_loop(&error_fatal);
1065     bdrv_init();
1066     atexit(qemu_nbd_shutdown);
1067
1068     srcpath = argv[optind];
1069     if (imageOpts) {
1070         QemuOpts *opts;
1071         if (fmt) {
1072             error_report("--image-opts and -f are mutually exclusive");
1073             exit(EXIT_FAILURE);
1074         }
1075         opts = qemu_opts_parse_noisily(&file_opts, srcpath, true);
1076         if (!opts) {
1077             qemu_opts_reset(&file_opts);
1078             exit(EXIT_FAILURE);
1079         }
1080         options = qemu_opts_to_qdict(opts, NULL);
1081         qemu_opts_reset(&file_opts);
1082         blk = blk_new_open(NULL, NULL, options, flags, &local_err);
1083     } else {
1084         if (fmt) {
1085             options = qdict_new();
1086             qdict_put_str(options, "driver", fmt);
1087         }
1088         blk = blk_new_open(srcpath, NULL, options, flags, &local_err);
1089     }
1090
1091     if (!blk) {
1092         error_reportf_err(local_err, "Failed to blk_new_open '%s': ",
1093                           argv[optind]);
1094         exit(EXIT_FAILURE);
1095     }
1096     bs = blk_bs(blk);
1097
1098     if (dev_offset) {
1099         QDict *raw_opts = qdict_new();
1100         qdict_put_str(raw_opts, "driver", "raw");
1101         qdict_put_str(raw_opts, "file", bs->node_name);
1102         qdict_put_int(raw_opts, "offset", dev_offset);
1103
1104         aio_context_acquire(qemu_get_aio_context());
1105         bs = bdrv_open(NULL, NULL, raw_opts, flags, &error_fatal);
1106         aio_context_release(qemu_get_aio_context());
1107
1108         blk_remove_bs(blk);
1109         blk_insert_bs(blk, bs, &error_fatal);
1110         bdrv_unref(bs);
1111     }
1112
1113     blk_set_enable_write_cache(blk, !writethrough);
1114
1115     if (sn_opts) {
1116         ret = bdrv_snapshot_load_tmp(bs,
1117                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
1118                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
1119                                      &local_err);
1120     } else if (sn_id_or_name) {
1121         ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
1122                                                    &local_err);
1123     }
1124     if (ret < 0) {
1125         error_reportf_err(local_err, "Failed to load snapshot: ");
1126         exit(EXIT_FAILURE);
1127     }
1128
1129     bs->detect_zeroes = detect_zeroes;
1130
1131     nbd_server_is_qemu_nbd(shared);
1132
1133     export_opts = g_new(BlockExportOptions, 1);
1134     *export_opts = (BlockExportOptions) {
1135         .type               = BLOCK_EXPORT_TYPE_NBD,
1136         .id                 = g_strdup("qemu-nbd-export"),
1137         .node_name          = g_strdup(bdrv_get_node_name(bs)),
1138         .has_writethrough   = true,
1139         .writethrough       = writethrough,
1140         .has_writable       = true,
1141         .writable           = !readonly,
1142         .u.nbd = {
1143             .name                 = g_strdup(export_name),
1144             .description          = g_strdup(export_description),
1145             .has_bitmaps          = !!bitmaps,
1146             .bitmaps              = bitmaps,
1147             .has_allocation_depth = alloc_depth,
1148             .allocation_depth     = alloc_depth,
1149         },
1150     };
1151     blk_exp_add(export_opts, &error_fatal);
1152     qapi_free_BlockExportOptions(export_opts);
1153
1154     if (device) {
1155 #if HAVE_NBD_DEVICE
1156         int ret;
1157         opts = (struct NbdClientOpts) {
1158             .device = device,
1159             .fork_process = fork_process,
1160             .verbose = verbose,
1161         };
1162
1163         ret = pthread_create(&client_thread, NULL, nbd_client_thread, &opts);
1164         if (ret != 0) {
1165             error_report("Failed to create client thread: %s", strerror(ret));
1166             exit(EXIT_FAILURE);
1167         }
1168 #endif
1169     } else {
1170         /* Shut up GCC warnings.  */
1171         memset(&client_thread, 0, sizeof(client_thread));
1172     }
1173
1174     nbd_update_server_watch();
1175
1176     if (pid_file_name) {
1177         qemu_write_pidfile(pid_file_name, &error_fatal);
1178     }
1179
1180     /* now when the initialization is (almost) complete, chdir("/")
1181      * to free any busy filesystems */
1182     if (chdir("/") < 0) {
1183         error_report("Could not chdir to root directory: %s",
1184                      strerror(errno));
1185         exit(EXIT_FAILURE);
1186     }
1187
1188     if (fork_process) {
1189         if (dup2(STDOUT_FILENO, STDERR_FILENO) < 0) {
1190             error_report("Could not set stderr to /dev/null: %s",
1191                          strerror(errno));
1192             exit(EXIT_FAILURE);
1193         }
1194     }
1195
1196     state = RUNNING;
1197     do {
1198         main_loop_wait(false);
1199         if (state == TERMINATE) {
1200             blk_exp_close_all();
1201             state = TERMINATED;
1202         }
1203     } while (state != TERMINATED);
1204
1205     blk_unref(blk);
1206     if (sockpath) {
1207         unlink(sockpath);
1208     }
1209
1210     qemu_opts_del(sn_opts);
1211
1212     if (device) {
1213         void *ret;
1214         pthread_join(client_thread, &ret);
1215         exit(ret != NULL);
1216     } else {
1217         exit(EXIT_SUCCESS);
1218     }
1219 }