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             };
 225
 226             printf("  size:  %" PRIu64 "\n", list[i].size);
 227             printf("  flags: 0x%x (", list[i].flags);
 228             for (size_t bit = 0; bit < ARRAY_SIZE(flag_names); bit++) {
 229                 if (flag_names[bit] && (list[i].flags & (1 << bit))) {
 230                     printf(" %s", flag_names[bit]);
 231                 }
 232             }
 233             printf(" )\n");
 234         }
 235         if (list[i].min_block) {
 236             printf("  min block: %u\n", list[i].min_block);
 237             printf("  opt block: %u\n", list[i].opt_block);
 238             printf("  max block: %u\n", list[i].max_block);
 239         }
 240         if (list[i].n_contexts) {
 241             printf("  available meta contexts: %d\n", list[i].n_contexts);
 242             for (j = 0; j < list[i].n_contexts; j++) {
 243                 printf("   %s\n", list[i].contexts[j]);
 244             }
 245         }
 246     }
 247     nbd_free_export_list(list, rc);
 248
 249     ret = EXIT_SUCCESS;
 250  out:
 251     object_unref(OBJECT(sioc));
 252     return ret;
 253 }
 254
 255
 256 #if HAVE_NBD_DEVICE
 257 static void *show_parts(void *arg)
 258 {
 259     char *device = arg;
 260     int nbd;
 261
 262     /* linux just needs an open() to trigger
 263      * the partition table update
 264      * but remember to load the module with max_part != 0 :
 265      *     modprobe nbd max_part=63
 266      */
 267     nbd = open(device, O_RDWR);
 268     if (nbd >= 0) {
 269         close(nbd);
 270     }
 271     return NULL;
 272 }
 273
 274 struct NbdClientOpts {
 275     char *device;
 276     bool fork_process;
 277     bool verbose;
 278 };
 279
 280 static void *nbd_client_thread(void *arg)
 281 {
 282     struct NbdClientOpts *opts = arg;
 283     NBDExportInfo info = { .request_sizes = false, .name = g_strdup("") };
 284     QIOChannelSocket *sioc;
 285     int fd = -1;
 286     int ret = EXIT_FAILURE;
 287     pthread_t show_parts_thread;
 288     Error *local_error = NULL;
 289
 290     sioc = qio_channel_socket_new();
 291     if (qio_channel_socket_connect_sync(sioc,
 292                                         saddr,
 293                                         &local_error) < 0) {
 294         error_report_err(local_error);
 295         goto out;
 296     }
 297
 298     if (nbd_receive_negotiate(NULL, QIO_CHANNEL(sioc),
 299                               NULL, NULL, NULL, &info, &local_error) < 0) {
 300         if (local_error) {
 301             error_report_err(local_error);
 302         }
 303         goto out;
 304     }
 305
 306     fd = open(opts->device, O_RDWR);
 307     if (fd < 0) {
 308         /* Linux-only, we can use %m in printf.  */
 309         error_report("Failed to open %s: %m", opts->device);
 310         goto out;
 311     }
 312
 313     if (nbd_init(fd, sioc, &info, &local_error) < 0) {
 314         error_report_err(local_error);
 315         goto out;
 316     }
 317
 318     /* update partition table */
 319     pthread_create(&show_parts_thread, NULL, show_parts, opts->device);
 320
 321     if (opts->verbose && !opts->fork_process) {
 322         fprintf(stderr, "NBD device %s is now connected to %s\n",
 323                 opts->device, srcpath);
 324     } else {
 325         /* Close stderr so that the qemu-nbd process exits.  */
 326         if (dup2(STDOUT_FILENO, STDERR_FILENO) < 0) {
 327             error_report("Could not set stderr to /dev/null: %s",
 328                          strerror(errno));
 329             exit(EXIT_FAILURE);
 330         }
 331     }
 332
 333     if (nbd_client(fd) < 0) {
 334         goto out;
 335     }
 336
 337     ret = EXIT_SUCCESS;
 338
 339  out:
 340     if (fd >= 0) {
 341         close(fd);
 342     }
 343     object_unref(OBJECT(sioc));
 344     g_free(info.name);
 345     kill(getpid(), SIGTERM);
 346     return (void *) (intptr_t) ret;
 347 }
 348 #endif /* HAVE_NBD_DEVICE */
 349
 350 static int nbd_can_accept(void)
 351 {
 352     return state == RUNNING && (shared == 0 || nb_fds < shared);
 353 }
 354
 355 static void nbd_update_server_watch(void);
 356
 357 static void nbd_client_closed(NBDClient *client, bool negotiated)
 358 {
 359     nb_fds--;
 360     if (negotiated && nb_fds == 0 && !persistent && state == RUNNING) {
 361         state = TERMINATE;
 362     }
 363     nbd_update_server_watch();
 364     nbd_client_put(client);
 365 }
 366
 367 static void nbd_accept(QIONetListener *listener, QIOChannelSocket *cioc,
 368                        gpointer opaque)
 369 {
 370     if (state >= TERMINATE) {
 371         return;
 372     }
 373
 374     nb_fds++;
 375     nbd_update_server_watch();
 376     nbd_client_new(cioc, tlscreds, tlsauthz, nbd_client_closed);
 377 }
 378
 379 static void nbd_update_server_watch(void)
 380 {
 381     if (nbd_can_accept()) {
 382         qio_net_listener_set_client_func(server, nbd_accept, NULL, NULL);
 383     } else {
 384         qio_net_listener_set_client_func(server, NULL, NULL, NULL);
 385     }
 386 }
 387
 388
 389 static SocketAddress *nbd_build_socket_address(const char *sockpath,
 390                                                const char *bindto,
 391                                                const char *port)
 392 {
 393     SocketAddress *saddr;
 394
 395     saddr = g_new0(SocketAddress, 1);
 396     if (sockpath) {
 397         saddr->type = SOCKET_ADDRESS_TYPE_UNIX;
 398         saddr->u.q_unix.path = g_strdup(sockpath);
 399     } else {
 400         InetSocketAddress *inet;
 401         saddr->type = SOCKET_ADDRESS_TYPE_INET;
 402         inet = &saddr->u.inet;
 403         inet->host = g_strdup(bindto);
 404         if (port) {
 405             inet->port = g_strdup(port);
 406         } else  {
 407             inet->port = g_strdup_printf("%d", NBD_DEFAULT_PORT);
 408         }
 409     }
 410
 411     return saddr;
 412 }
 413
 414
 415 static QemuOptsList file_opts = {
 416     .name = "file",
 417     .implied_opt_name = "file",
 418     .head = QTAILQ_HEAD_INITIALIZER(file_opts.head),
 419     .desc = {
 420         /* no elements => accept any params */
 421         { /* end of list */ }
 422     },
 423 };
 424
 425 static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, bool list,
 426                                           Error **errp)
 427 {
 428     Object *obj;
 429     QCryptoTLSCreds *creds;
 430
 431     obj = object_resolve_path_component(
 432         object_get_objects_root(), id);
 433     if (!obj) {
 434         error_setg(errp, "No TLS credentials with id '%s'",
 435                    id);
 436         return NULL;
 437     }
 438     creds = (QCryptoTLSCreds *)
 439         object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
 440     if (!creds) {
 441         error_setg(errp, "Object with id '%s' is not TLS credentials",
 442                    id);
 443         return NULL;
 444     }
 445
 446     if (!qcrypto_tls_creds_check_endpoint(creds,
 447                                           list
 448                                           ? QCRYPTO_TLS_CREDS_ENDPOINT_CLIENT
 449                                           : QCRYPTO_TLS_CREDS_ENDPOINT_SERVER,
 450                                           errp)) {
 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                                                    const char *selinux,
 476                                                    bool list)
 477 {
 478     if (device != NULL) {
 479         return "NBD device can't be set when using socket activation";
 480     }
 481
 482     if (sockpath != NULL) {
 483         return "Unix socket can't be set when using socket activation";
 484     }
 485
 486     if (address != NULL) {
 487         return "The interface can't be set when using socket activation";
 488     }
 489
 490     if (port != NULL) {
 491         return "TCP port number can't be set when using socket activation";
 492     }
 493
 494     if (selinux != NULL) {
 495         return "SELinux label can't be set when using socket activation";
 496     }
 497
 498     if (list) {
 499         return "List mode is incompatible with socket activation";
 500     }
 501
 502     return NULL;
 503 }
 504
 505 static void qemu_nbd_shutdown(void)
 506 {
 507     job_cancel_sync_all();
 508     blk_exp_close_all();
 509     bdrv_close_all();
 510 }
 511
 512 int main(int argc, char **argv)
 513 {
 514     BlockBackend *blk;
 515     BlockDriverState *bs;
 516     uint64_t dev_offset = 0;
 517     bool readonly = false;
 518     bool disconnect = false;
 519     const char *bindto = NULL;
 520     const char *port = NULL;
 521     char *sockpath = NULL;
 522     char *device = NULL;
 523     QemuOpts *sn_opts = NULL;
 524     const char *sn_id_or_name = NULL;
 525     const char *sopt = "hVb:o:p:rsnc:dvk:e:f:tl:x:T:D:AB:L";
 526     struct option lopt[] = {
 527         { "help", no_argument, NULL, 'h' },
 528         { "version", no_argument, NULL, 'V' },
 529         { "bind", required_argument, NULL, 'b' },
 530         { "port", required_argument, NULL, 'p' },
 531         { "socket", required_argument, NULL, 'k' },
 532         { "offset", required_argument, NULL, 'o' },
 533         { "read-only", no_argument, NULL, 'r' },
 534         { "allocation-depth", no_argument, NULL, 'A' },
 535         { "bitmap", required_argument, NULL, 'B' },
 536         { "connect", required_argument, NULL, 'c' },
 537         { "disconnect", no_argument, NULL, 'd' },
 538         { "list", no_argument, NULL, 'L' },
 539         { "snapshot", no_argument, NULL, 's' },
 540         { "load-snapshot", required_argument, NULL, 'l' },
 541         { "nocache", no_argument, NULL, 'n' },
 542         { "cache", required_argument, NULL, QEMU_NBD_OPT_CACHE },
 543         { "aio", required_argument, NULL, QEMU_NBD_OPT_AIO },
 544         { "discard", required_argument, NULL, QEMU_NBD_OPT_DISCARD },
 545         { "detect-zeroes", required_argument, NULL,
 546           QEMU_NBD_OPT_DETECT_ZEROES },
 547         { "shared", required_argument, NULL, 'e' },
 548         { "format", required_argument, NULL, 'f' },
 549         { "persistent", no_argument, NULL, 't' },
 550         { "verbose", no_argument, NULL, 'v' },
 551         { "object", required_argument, NULL, QEMU_NBD_OPT_OBJECT },
 552         { "export-name", required_argument, NULL, 'x' },
 553         { "description", required_argument, NULL, 'D' },
 554         { "tls-creds", required_argument, NULL, QEMU_NBD_OPT_TLSCREDS },
 555         { "tls-hostname", required_argument, NULL, QEMU_NBD_OPT_TLSHOSTNAME },
 556         { "tls-authz", required_argument, NULL, QEMU_NBD_OPT_TLSAUTHZ },
 557         { "image-opts", no_argument, NULL, QEMU_NBD_OPT_IMAGE_OPTS },
 558         { "trace", required_argument, NULL, 'T' },
 559         { "fork", no_argument, NULL, QEMU_NBD_OPT_FORK },
 560         { "pid-file", required_argument, NULL, QEMU_NBD_OPT_PID_FILE },
 561         { "selinux-label", required_argument, NULL,
 562           QEMU_NBD_OPT_SELINUX_LABEL },
 563         { NULL, 0, NULL, 0 }
 564     };
 565     int ch;
 566     int opt_ind = 0;
 567     int flags = BDRV_O_RDWR;
 568     int ret = 0;
 569     bool seen_cache = false;
 570     bool seen_discard = false;
 571     bool seen_aio = false;
 572     pthread_t client_thread;
 573     const char *fmt = NULL;
 574     Error *local_err = NULL;
 575     BlockdevDetectZeroesOptions detect_zeroes = BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
 576     QDict *options = NULL;
 577     const char *export_name = NULL; /* defaults to "" later for server mode */
 578     const char *export_description = NULL;
 579     BlockDirtyBitmapOrStrList *bitmaps = NULL;
 580     bool alloc_depth = false;
 581     const char *tlscredsid = NULL;
 582     const char *tlshostname = NULL;
 583     bool imageOpts = false;
 584     bool writethrough = false; /* Client will flush as needed. */
 585     bool verbose = false;
 586     bool fork_process = false;
 587     bool list = false;
 588     unsigned socket_activation;
 589     const char *pid_file_name = NULL;
 590     const char *selinux_label = NULL;
 591     BlockExportOptions *export_opts;
 592
 593 #ifdef CONFIG_POSIX
 594     os_setup_early_signal_handling();
 595     os_setup_signal_handling();
 596 #endif
 597
 598     socket_init();
 599     error_init(argv[0]);
 600     module_call_init(MODULE_INIT_TRACE);
 601     qcrypto_init(&error_fatal);
 602
 603     module_call_init(MODULE_INIT_QOM);
 604     qemu_add_opts(&qemu_trace_opts);
 605     qemu_init_exec_dir(argv[0]);
 606
 607     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 608         switch (ch) {
 609         case 's':
 610             flags |= BDRV_O_SNAPSHOT;
 611             break;
 612         case 'n':
 613             optarg = (char *) "none";
 614             /* fallthrough */
 615         case QEMU_NBD_OPT_CACHE:
 616             if (seen_cache) {
 617                 error_report("-n and --cache can only be specified once");
 618                 exit(EXIT_FAILURE);
 619             }
 620             seen_cache = true;
 621             if (bdrv_parse_cache_mode(optarg, &flags, &writethrough) == -1) {
 622                 error_report("Invalid cache mode `%s'", optarg);
 623                 exit(EXIT_FAILURE);
 624             }
 625             break;
 626         case QEMU_NBD_OPT_AIO:
 627             if (seen_aio) {
 628                 error_report("--aio can only be specified once");
 629                 exit(EXIT_FAILURE);
 630             }
 631             seen_aio = true;
 632             if (bdrv_parse_aio(optarg, &flags) < 0) {
 633                 error_report("Invalid aio mode '%s'", optarg);
 634                 exit(EXIT_FAILURE);
 635             }
 636             break;
 637         case QEMU_NBD_OPT_DISCARD:
 638             if (seen_discard) {
 639                 error_report("--discard can only be specified once");
 640                 exit(EXIT_FAILURE);
 641             }
 642             seen_discard = true;
 643             if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
 644                 error_report("Invalid discard mode `%s'", optarg);
 645                 exit(EXIT_FAILURE);
 646             }
 647             break;
 648         case QEMU_NBD_OPT_DETECT_ZEROES:
 649             detect_zeroes =
 650                 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
 651                                 optarg,
 652                                 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
 653                                 &local_err);
 654             if (local_err) {
 655                 error_reportf_err(local_err,
 656                                   "Failed to parse detect_zeroes mode: ");
 657                 exit(EXIT_FAILURE);
 658             }
 659             if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
 660                 !(flags & BDRV_O_UNMAP)) {
 661                 error_report("setting detect-zeroes to unmap is not allowed "
 662                              "without setting discard operation to unmap");
 663                 exit(EXIT_FAILURE);
 664             }
 665             break;
 666         case 'b':
 667             bindto = optarg;
 668             break;
 669         case 'p':
 670             port = optarg;
 671             break;
 672         case 'o':
 673             if (qemu_strtou64(optarg, NULL, 0, &dev_offset) < 0) {
 674                 error_report("Invalid offset '%s'", optarg);
 675                 exit(EXIT_FAILURE);
 676             }
 677             break;
 678         case 'l':
 679             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
 680                 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
 681                                                   optarg, false);
 682                 if (!sn_opts) {
 683                     error_report("Failed in parsing snapshot param `%s'",
 684                                  optarg);
 685                     exit(EXIT_FAILURE);
 686                 }
 687             } else {
 688                 sn_id_or_name = optarg;
 689             }
 690             /* fall through */
 691         case 'r':
 692             readonly = true;
 693             flags &= ~BDRV_O_RDWR;
 694             break;
 695         case 'A':
 696             alloc_depth = true;
 697             break;
 698         case 'B':
 699             {
 700                 BlockDirtyBitmapOrStr *el = g_new(BlockDirtyBitmapOrStr, 1);
 701                 *el = (BlockDirtyBitmapOrStr) {
 702                     .type = QTYPE_QSTRING,
 703                     .u.local = g_strdup(optarg),
 704                 };
 705                 QAPI_LIST_PREPEND(bitmaps, el);
 706             }
 707             break;
 708         case 'k':
 709             sockpath = optarg;
 710             if (sockpath[0] != '/') {
 711                 error_report("socket path must be absolute");
 712                 exit(EXIT_FAILURE);
 713             }
 714             break;
 715         case 'd':
 716             disconnect = true;
 717             break;
 718         case 'c':
 719             device = optarg;
 720             break;
 721         case 'e':
 722             if (qemu_strtoi(optarg, NULL, 0, &shared) < 0 ||
 723                 shared < 0) {
 724                 error_report("Invalid shared device number '%s'", optarg);
 725                 exit(EXIT_FAILURE);
 726             }
 727             break;
 728         case 'f':
 729             fmt = optarg;
 730             break;
 731         case 't':
 732             persistent = 1;
 733             break;
 734         case 'x':
 735             export_name = optarg;
 736             if (strlen(export_name) > NBD_MAX_STRING_SIZE) {
 737                 error_report("export name '%s' too long", export_name);
 738                 exit(EXIT_FAILURE);
 739             }
 740             break;
 741         case 'D':
 742             export_description = optarg;
 743             if (strlen(export_description) > NBD_MAX_STRING_SIZE) {
 744                 error_report("export description '%s' too long",
 745                              export_description);
 746                 exit(EXIT_FAILURE);
 747             }
 748             break;
 749         case 'v':
 750             verbose = true;
 751             break;
 752         case 'V':
 753             version(argv[0]);
 754             exit(0);
 755             break;
 756         case 'h':
 757             usage(argv[0]);
 758             exit(0);
 759             break;
 760         case '?':
 761             error_report("Try `%s --help' for more information.", argv[0]);
 762             exit(EXIT_FAILURE);
 763         case QEMU_NBD_OPT_OBJECT:
 764             user_creatable_process_cmdline(optarg);
 765             break;
 766         case QEMU_NBD_OPT_TLSCREDS:
 767             tlscredsid = optarg;
 768             break;
 769         case QEMU_NBD_OPT_TLSHOSTNAME:
 770             tlshostname = optarg;
 771             break;
 772         case QEMU_NBD_OPT_IMAGE_OPTS:
 773             imageOpts = true;
 774             break;
 775         case 'T':
 776             trace_opt_parse(optarg);
 777             break;
 778         case QEMU_NBD_OPT_TLSAUTHZ:
 779             tlsauthz = optarg;
 780             break;
 781         case QEMU_NBD_OPT_FORK:
 782             fork_process = true;
 783             break;
 784         case 'L':
 785             list = true;
 786             break;
 787         case QEMU_NBD_OPT_PID_FILE:
 788             pid_file_name = optarg;
 789             break;
 790         case QEMU_NBD_OPT_SELINUX_LABEL:
 791             selinux_label = optarg;
 792             break;
 793         }
 794     }
 795
 796     if (list) {
 797         if (argc != optind) {
 798             error_report("List mode is incompatible with a file name");
 799             exit(EXIT_FAILURE);
 800         }
 801         if (export_name || export_description || dev_offset ||
 802             device || disconnect || fmt || sn_id_or_name || bitmaps ||
 803             alloc_depth || seen_aio || seen_discard || seen_cache) {
 804             error_report("List mode is incompatible with per-device settings");
 805             exit(EXIT_FAILURE);
 806         }
 807         if (fork_process) {
 808             error_report("List mode is incompatible with forking");
 809             exit(EXIT_FAILURE);
 810         }
 811     } else if ((argc - optind) != 1) {
 812         error_report("Invalid number of arguments");
 813         error_printf("Try `%s --help' for more information.\n", argv[0]);
 814         exit(EXIT_FAILURE);
 815     } else if (!export_name) {
 816         export_name = "";
 817     }
 818
 819     if (!trace_init_backends()) {
 820         exit(1);
 821     }
 822     trace_init_file();
 823     qemu_set_log(LOG_TRACE, &error_fatal);
 824
 825     socket_activation = check_socket_activation();
 826     if (socket_activation == 0) {
 827         if (!sockpath) {
 828             setup_address_and_port(&bindto, &port);
 829         }
 830     } else {
 831         /* Using socket activation - check user didn't use -p etc. */
 832         const char *err_msg = socket_activation_validate_opts(device, sockpath,
 833                                                               bindto, port,
 834                                                               selinux_label,
 835                                                               list);
 836         if (err_msg != NULL) {
 837             error_report("%s", err_msg);
 838             exit(EXIT_FAILURE);
 839         }
 840
 841         /* qemu-nbd can only listen on a single socket.  */
 842         if (socket_activation > 1) {
 843             error_report("qemu-nbd does not support socket activation with %s > 1",
 844                          "LISTEN_FDS");
 845             exit(EXIT_FAILURE);
 846         }
 847     }
 848
 849     if (tlscredsid) {
 850         if (device) {
 851             error_report("TLS is not supported with a host device");
 852             exit(EXIT_FAILURE);
 853         }
 854         if (tlsauthz && list) {
 855             error_report("TLS authorization is incompatible with export list");
 856             exit(EXIT_FAILURE);
 857         }
 858         if (tlshostname && !list) {
 859             error_report("TLS hostname is only supported with export list");
 860             exit(EXIT_FAILURE);
 861         }
 862         tlscreds = nbd_get_tls_creds(tlscredsid, list, &local_err);
 863         if (local_err) {
 864             error_reportf_err(local_err, "Failed to get TLS creds: ");
 865             exit(EXIT_FAILURE);
 866         }
 867     } else {
 868         if (tlsauthz) {
 869             error_report("--tls-authz is not permitted without --tls-creds");
 870             exit(EXIT_FAILURE);
 871         }
 872         if (tlshostname) {
 873             error_report("--tls-hostname is not permitted without --tls-creds");
 874             exit(EXIT_FAILURE);
 875         }
 876     }
 877
 878     if (selinux_label) {
 879 #ifdef CONFIG_SELINUX
 880         if (sockpath == NULL && device == NULL) {
 881             error_report("--selinux-label is not permitted without --socket");
 882             exit(EXIT_FAILURE);
 883         }
 884 #else
 885         error_report("SELinux support not enabled in this binary");
 886         exit(EXIT_FAILURE);
 887 #endif
 888     }
 889
 890     if (list) {
 891         saddr = nbd_build_socket_address(sockpath, bindto, port);
 892         return qemu_nbd_client_list(saddr, tlscreds,
 893                                     tlshostname ? tlshostname : bindto);
 894     }
 895
 896 #if !HAVE_NBD_DEVICE
 897     if (disconnect || device) {
 898         error_report("Kernel /dev/nbdN support not available");
 899         exit(EXIT_FAILURE);
 900     }
 901 #else /* HAVE_NBD_DEVICE */
 902     if (disconnect) {
 903         int nbdfd = open(argv[optind], O_RDWR);
 904         if (nbdfd < 0) {
 905             error_report("Cannot open %s: %s", argv[optind],
 906                          strerror(errno));
 907             exit(EXIT_FAILURE);
 908         }
 909         nbd_disconnect(nbdfd);
 910
 911         close(nbdfd);
 912
 913         printf("%s disconnected\n", argv[optind]);
 914
 915         return 0;
 916     }
 917 #endif
 918
 919     if ((device && !verbose) || fork_process) {
 920 #ifndef WIN32
 921         g_autoptr(GError) err = NULL;
 922         int stderr_fd[2];
 923         pid_t pid;
 924         int ret;
 925
 926         if (!g_unix_open_pipe(stderr_fd, FD_CLOEXEC, &err)) {
 927             error_report("Error setting up communication pipe: %s",
 928                          err->message);
 929             exit(EXIT_FAILURE);
 930         }
 931
 932         /* Now daemonize, but keep a communication channel open to
 933          * print errors and exit with the proper status code.
 934          */
 935         pid = fork();
 936         if (pid < 0) {
 937             error_report("Failed to fork: %s", strerror(errno));
 938             exit(EXIT_FAILURE);
 939         } else if (pid == 0) {
 940             int saved_errno;
 941
 942             close(stderr_fd[0]);
 943
 944             ret = qemu_daemon(1, 0);
 945             saved_errno = errno;    /* dup2 will overwrite error below */
 946
 947             /* Temporarily redirect stderr to the parent's pipe...  */
 948             if (dup2(stderr_fd[1], STDERR_FILENO) < 0) {
 949                 char str[256];
 950                 snprintf(str, sizeof(str),
 951                          "%s: Failed to link stderr to the pipe: %s\n",
 952                          g_get_prgname(), strerror(errno));
 953                 /*
 954                  * We are unable to use error_report() here as we need to get
 955                  * stderr pointed to the parent's pipe. Write to that pipe
 956                  * manually.
 957                  */
 958                 ret = write(stderr_fd[1], str, strlen(str));
 959                 exit(EXIT_FAILURE);
 960             }
 961
 962             if (ret < 0) {
 963                 error_report("Failed to daemonize: %s", strerror(saved_errno));
 964                 exit(EXIT_FAILURE);
 965             }
 966
 967             /* ... close the descriptor we inherited and go on.  */
 968             close(stderr_fd[1]);
 969         } else {
 970             bool errors = false;
 971             char *buf;
 972
 973             /* In the parent.  Print error messages from the child until
 974              * it closes the pipe.
 975              */
 976             close(stderr_fd[1]);
 977             buf = g_malloc(1024);
 978             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
 979                 errors = true;
 980                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
 981                 if (ret < 0) {
 982                     exit(EXIT_FAILURE);
 983                 }
 984             }
 985             if (ret < 0) {
 986                 error_report("Cannot read from daemon: %s",
 987                              strerror(errno));
 988                 exit(EXIT_FAILURE);
 989             }
 990
 991             /* Usually the daemon should not print any message.
 992              * Exit with zero status in that case.
 993              */
 994             exit(errors);
 995         }
 996 #else /* WIN32 */
 997         error_report("Unable to fork into background on Windows hosts");
 998         exit(EXIT_FAILURE);
 999 #endif /* WIN32 */
1000     }
1001
1002     if (device != NULL && sockpath == NULL) {
1003         sockpath = g_malloc(128);
1004         snprintf(sockpath, 128, SOCKET_PATH, basename(device));
1005     }
1006
1007     server = qio_net_listener_new();
1008     if (socket_activation == 0) {
1009         int backlog;
1010
1011         if (persistent || shared == 0) {
1012             backlog = SOMAXCONN;
1013         } else {
1014             backlog = MIN(shared, SOMAXCONN);
1015         }
1016 #ifdef CONFIG_SELINUX
1017         if (selinux_label && setsockcreatecon_raw(selinux_label) == -1) {
1018             error_report("Cannot set SELinux socket create context to %s: %s",
1019                          selinux_label, strerror(errno));
1020             exit(EXIT_FAILURE);
1021         }
1022 #endif
1023         saddr = nbd_build_socket_address(sockpath, bindto, port);
1024         if (qio_net_listener_open_sync(server, saddr, backlog,
1025                                        &local_err) < 0) {
1026             object_unref(OBJECT(server));
1027             error_report_err(local_err);
1028             exit(EXIT_FAILURE);
1029         }
1030 #ifdef CONFIG_SELINUX
1031         if (selinux_label && setsockcreatecon_raw(NULL) == -1) {
1032             error_report("Cannot clear SELinux socket create context: %s",
1033                          strerror(errno));
1034             exit(EXIT_FAILURE);
1035         }
1036 #endif
1037     } else {
1038         size_t i;
1039         /* See comment in check_socket_activation above. */
1040         for (i = 0; i < socket_activation; i++) {
1041             QIOChannelSocket *sioc;
1042             sioc = qio_channel_socket_new_fd(FIRST_SOCKET_ACTIVATION_FD + i,
1043                                              &local_err);
1044             if (sioc == NULL) {
1045                 object_unref(OBJECT(server));
1046                 error_reportf_err(local_err,
1047                                   "Failed to use socket activation: ");
1048                 exit(EXIT_FAILURE);
1049             }
1050             qio_net_listener_add(server, sioc);
1051             object_unref(OBJECT(sioc));
1052         }
1053     }
1054
1055     qemu_init_main_loop(&error_fatal);
1056     bdrv_init();
1057     atexit(qemu_nbd_shutdown);
1058
1059     srcpath = argv[optind];
1060     if (imageOpts) {
1061         QemuOpts *opts;
1062         if (fmt) {
1063             error_report("--image-opts and -f are mutually exclusive");
1064             exit(EXIT_FAILURE);
1065         }
1066         opts = qemu_opts_parse_noisily(&file_opts, srcpath, true);
1067         if (!opts) {
1068             qemu_opts_reset(&file_opts);
1069             exit(EXIT_FAILURE);
1070         }
1071         options = qemu_opts_to_qdict(opts, NULL);
1072         qemu_opts_reset(&file_opts);
1073         blk = blk_new_open(NULL, NULL, options, flags, &local_err);
1074     } else {
1075         if (fmt) {
1076             options = qdict_new();
1077             qdict_put_str(options, "driver", fmt);
1078         }
1079         blk = blk_new_open(srcpath, NULL, options, flags, &local_err);
1080     }
1081
1082     if (!blk) {
1083         error_reportf_err(local_err, "Failed to blk_new_open '%s': ",
1084                           argv[optind]);
1085         exit(EXIT_FAILURE);
1086     }
1087     bs = blk_bs(blk);
1088
1089     if (dev_offset) {
1090         QDict *raw_opts = qdict_new();
1091         qdict_put_str(raw_opts, "driver", "raw");
1092         qdict_put_str(raw_opts, "file", bs->node_name);
1093         qdict_put_int(raw_opts, "offset", dev_offset);
1094
1095         aio_context_acquire(qemu_get_aio_context());
1096         bs = bdrv_open(NULL, NULL, raw_opts, flags, &error_fatal);
1097         aio_context_release(qemu_get_aio_context());
1098
1099         blk_remove_bs(blk);
1100         blk_insert_bs(blk, bs, &error_fatal);
1101         bdrv_unref(bs);
1102     }
1103
1104     blk_set_enable_write_cache(blk, !writethrough);
1105
1106     if (sn_opts) {
1107         ret = bdrv_snapshot_load_tmp(bs,
1108                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
1109                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
1110                                      &local_err);
1111     } else if (sn_id_or_name) {
1112         ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
1113                                                    &local_err);
1114     }
1115     if (ret < 0) {
1116         error_reportf_err(local_err, "Failed to load snapshot: ");
1117         exit(EXIT_FAILURE);
1118     }
1119
1120     bs->detect_zeroes = detect_zeroes;
1121
1122     nbd_server_is_qemu_nbd(shared);
1123
1124     export_opts = g_new(BlockExportOptions, 1);
1125     *export_opts = (BlockExportOptions) {
1126         .type               = BLOCK_EXPORT_TYPE_NBD,
1127         .id                 = g_strdup("qemu-nbd-export"),
1128         .node_name          = g_strdup(bdrv_get_node_name(bs)),
1129         .has_writethrough   = true,
1130         .writethrough       = writethrough,
1131         .has_writable       = true,
1132         .writable           = !readonly,
1133         .u.nbd = {
1134             .name                 = g_strdup(export_name),
1135             .description          = g_strdup(export_description),
1136             .has_bitmaps          = !!bitmaps,
1137             .bitmaps              = bitmaps,
1138             .has_allocation_depth = alloc_depth,
1139             .allocation_depth     = alloc_depth,
1140         },
1141     };
1142     blk_exp_add(export_opts, &error_fatal);
1143     qapi_free_BlockExportOptions(export_opts);
1144
1145     if (device) {
1146 #if HAVE_NBD_DEVICE
1147         int ret;
1148         struct NbdClientOpts opts = {
1149             .device = device,
1150             .fork_process = fork_process,
1151             .verbose = verbose,
1152         };
1153
1154         ret = pthread_create(&client_thread, NULL, nbd_client_thread, &opts);
1155         if (ret != 0) {
1156             error_report("Failed to create client thread: %s", strerror(ret));
1157             exit(EXIT_FAILURE);
1158         }
1159 #endif
1160     } else {
1161         /* Shut up GCC warnings.  */
1162         memset(&client_thread, 0, sizeof(client_thread));
1163     }
1164
1165     nbd_update_server_watch();
1166
1167     if (pid_file_name) {
1168         qemu_write_pidfile(pid_file_name, &error_fatal);
1169     }
1170
1171     /* now when the initialization is (almost) complete, chdir("/")
1172      * to free any busy filesystems */
1173     if (chdir("/") < 0) {
1174         error_report("Could not chdir to root directory: %s",
1175                      strerror(errno));
1176         exit(EXIT_FAILURE);
1177     }
1178
1179     if (fork_process) {
1180         if (dup2(STDOUT_FILENO, STDERR_FILENO) < 0) {
1181             error_report("Could not set stderr to /dev/null: %s",
1182                          strerror(errno));
1183             exit(EXIT_FAILURE);
1184         }
1185     }
1186
1187     state = RUNNING;
1188     do {
1189         main_loop_wait(false);
1190         if (state == TERMINATE) {
1191             blk_exp_close_all();
1192             state = TERMINATED;
1193         }
1194     } while (state != TERMINATED);
1195
1196     blk_unref(blk);
1197     if (sockpath) {
1198         unlink(sockpath);
1199     }
1200
1201     qemu_opts_del(sn_opts);
1202
1203     if (device) {
1204         void *ret;
1205         pthread_join(client_thread, &ret);
1206         exit(ret != NULL);
1207     } else {
1208         exit(EXIT_SUCCESS);
1209     }
1210 }