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 "block/snapshot.h"
  32 #include "qapi/util.h"
  33 #include "qapi/qmp/qstring.h"
  34 #include "qom/object_interfaces.h"
  35 #include "io/channel-socket.h"
  36 #include "crypto/init.h"
  37 #include "trace/control.h"
  38
  39 #include <getopt.h>
  40 #include <libgen.h>
  41 #include <pthread.h>
  42
  43 #define SOCKET_PATH                "/var/lock/qemu-nbd-%s"
  44 #define QEMU_NBD_OPT_CACHE         256
  45 #define QEMU_NBD_OPT_AIO           257
  46 #define QEMU_NBD_OPT_DISCARD       258
  47 #define QEMU_NBD_OPT_DETECT_ZEROES 259
  48 #define QEMU_NBD_OPT_OBJECT        260
  49 #define QEMU_NBD_OPT_TLSCREDS      261
  50 #define QEMU_NBD_OPT_IMAGE_OPTS    262
  51 #define QEMU_NBD_OPT_FORK          263
  52
  53 #define MBR_SIZE 512
  54
  55 static NBDExport *exp;
  56 static bool newproto;
  57 static int verbose;
  58 static char *srcpath;
  59 static SocketAddress *saddr;
  60 static int persistent = 0;
  61 static enum { RUNNING, TERMINATE, TERMINATING, TERMINATED } state;
  62 static int shared = 1;
  63 static int nb_fds;
  64 static QIOChannelSocket *server_ioc;
  65 static int server_watch = -1;
  66 static QCryptoTLSCreds *tlscreds;
  67
  68 static void usage(const char *name)
  69 {
  70     (printf) (
  71 "Usage: %s [OPTIONS] FILE\n"
  72 "QEMU Disk Network Block Device Server\n"
  73 "\n"
  74 "  -h, --help                display this help and exit\n"
  75 "  -V, --version             output version information and exit\n"
  76 "\n"
  77 "Connection properties:\n"
  78 "  -p, --port=PORT           port to listen on (default `%d')\n"
  79 "  -b, --bind=IFACE          interface to bind to (default `0.0.0.0')\n"
  80 "  -k, --socket=PATH         path to the unix socket\n"
  81 "                            (default '"SOCKET_PATH"')\n"
  82 "  -e, --shared=NUM          device can be shared by NUM clients (default '1')\n"
  83 "  -t, --persistent          don't exit on the last connection\n"
  84 "  -v, --verbose             display extra debugging information\n"
  85 "  -x, --export-name=NAME    expose export by name\n"
  86 "  -D, --description=TEXT    with -x, also export a human-readable description\n"
  87 "\n"
  88 "Exposing part of the image:\n"
  89 "  -o, --offset=OFFSET       offset into the image\n"
  90 "  -P, --partition=NUM       only expose partition NUM\n"
  91 "\n"
  92 "General purpose options:\n"
  93 "  --object type,id=ID,...   define an object such as 'secret' for providing\n"
  94 "                            passwords and/or encryption keys\n"
  95 "  -T, --trace [[enable=]<pattern>][,events=<file>][,file=<file>]\n"
  96 "                            specify tracing options\n"
  97 "  --fork                    fork off the server process and exit the parent\n"
  98 "                            once the server is running\n"
  99 #ifdef __linux__
 100 "Kernel NBD client support:\n"
 101 "  -c, --connect=DEV         connect FILE to the local NBD device DEV\n"
 102 "  -d, --disconnect          disconnect the specified device\n"
 103 "\n"
 104 #endif
 105 "\n"
 106 "Block device options:\n"
 107 "  -f, --format=FORMAT       set image format (raw, qcow2, ...)\n"
 108 "  -r, --read-only           export read-only\n"
 109 "  -s, --snapshot            use FILE as an external snapshot, create a temporary\n"
 110 "                            file with backing_file=FILE, redirect the write to\n"
 111 "                            the temporary one\n"
 112 "  -l, --load-snapshot=SNAPSHOT_PARAM\n"
 113 "                            load an internal snapshot inside FILE and export it\n"
 114 "                            as an read-only device, SNAPSHOT_PARAM format is\n"
 115 "                            'snapshot.id=[ID],snapshot.name=[NAME]', or\n"
 116 "                            '[ID_OR_NAME]'\n"
 117 "  -n, --nocache             disable host cache\n"
 118 "      --cache=MODE          set cache mode (none, writeback, ...)\n"
 119 "      --aio=MODE            set AIO mode (native or threads)\n"
 120 "      --discard=MODE        set discard mode (ignore, unmap)\n"
 121 "      --detect-zeroes=MODE  set detect-zeroes mode (off, on, unmap)\n"
 122 "      --image-opts          treat FILE as a full set of image options\n"
 123 "\n"
 124 "Report bugs to <qemu-devel@nongnu.org>\n"
 125     , name, NBD_DEFAULT_PORT, "DEVICE");
 126 }
 127
 128 static void version(const char *name)
 129 {
 130     printf(
 131 "%s version 0.0.1\n"
 132 "Written by Anthony Liguori.\n"
 133 "\n"
 134 "Copyright (C) 2006 Anthony Liguori <anthony@codemonkey.ws>.\n"
 135 "This is free software; see the source for copying conditions.  There is NO\n"
 136 "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
 137     , name);
 138 }
 139
 140 struct partition_record
 141 {
 142     uint8_t bootable;
 143     uint8_t start_head;
 144     uint32_t start_cylinder;
 145     uint8_t start_sector;
 146     uint8_t system;
 147     uint8_t end_head;
 148     uint8_t end_cylinder;
 149     uint8_t end_sector;
 150     uint32_t start_sector_abs;
 151     uint32_t nb_sectors_abs;
 152 };
 153
 154 static void read_partition(uint8_t *p, struct partition_record *r)
 155 {
 156     r->bootable = p[0];
 157     r->start_head = p[1];
 158     r->start_cylinder = p[3] | ((p[2] << 2) & 0x0300);
 159     r->start_sector = p[2] & 0x3f;
 160     r->system = p[4];
 161     r->end_head = p[5];
 162     r->end_cylinder = p[7] | ((p[6] << 2) & 0x300);
 163     r->end_sector = p[6] & 0x3f;
 164
 165     r->start_sector_abs = ldl_le_p(p + 8);
 166     r->nb_sectors_abs   = ldl_le_p(p + 12);
 167 }
 168
 169 static int find_partition(BlockBackend *blk, int partition,
 170                           off_t *offset, off_t *size)
 171 {
 172     struct partition_record mbr[4];
 173     uint8_t data[MBR_SIZE];
 174     int i;
 175     int ext_partnum = 4;
 176     int ret;
 177
 178     ret = blk_pread(blk, 0, data, sizeof(data));
 179     if (ret < 0) {
 180         error_report("error while reading: %s", strerror(-ret));
 181         exit(EXIT_FAILURE);
 182     }
 183
 184     if (data[510] != 0x55 || data[511] != 0xaa) {
 185         return -EINVAL;
 186     }
 187
 188     for (i = 0; i < 4; i++) {
 189         read_partition(&data[446 + 16 * i], &mbr[i]);
 190
 191         if (!mbr[i].system || !mbr[i].nb_sectors_abs) {
 192             continue;
 193         }
 194
 195         if (mbr[i].system == 0xF || mbr[i].system == 0x5) {
 196             struct partition_record ext[4];
 197             uint8_t data1[MBR_SIZE];
 198             int j;
 199
 200             ret = blk_pread(blk, mbr[i].start_sector_abs * MBR_SIZE,
 201                             data1, sizeof(data1));
 202             if (ret < 0) {
 203                 error_report("error while reading: %s", strerror(-ret));
 204                 exit(EXIT_FAILURE);
 205             }
 206
 207             for (j = 0; j < 4; j++) {
 208                 read_partition(&data1[446 + 16 * j], &ext[j]);
 209                 if (!ext[j].system || !ext[j].nb_sectors_abs) {
 210                     continue;
 211                 }
 212
 213                 if ((ext_partnum + j + 1) == partition) {
 214                     *offset = (uint64_t)ext[j].start_sector_abs << 9;
 215                     *size = (uint64_t)ext[j].nb_sectors_abs << 9;
 216                     return 0;
 217                 }
 218             }
 219             ext_partnum += 4;
 220         } else if ((i + 1) == partition) {
 221             *offset = (uint64_t)mbr[i].start_sector_abs << 9;
 222             *size = (uint64_t)mbr[i].nb_sectors_abs << 9;
 223             return 0;
 224         }
 225     }
 226
 227     return -ENOENT;
 228 }
 229
 230 static void termsig_handler(int signum)
 231 {
 232     atomic_cmpxchg(&state, RUNNING, TERMINATE);
 233     qemu_notify_event();
 234 }
 235
 236
 237 static void *show_parts(void *arg)
 238 {
 239     char *device = arg;
 240     int nbd;
 241
 242     /* linux just needs an open() to trigger
 243      * the partition table update
 244      * but remember to load the module with max_part != 0 :
 245      *     modprobe nbd max_part=63
 246      */
 247     nbd = open(device, O_RDWR);
 248     if (nbd >= 0) {
 249         close(nbd);
 250     }
 251     return NULL;
 252 }
 253
 254 static void *nbd_client_thread(void *arg)
 255 {
 256     char *device = arg;
 257     off_t size;
 258     uint16_t nbdflags;
 259     QIOChannelSocket *sioc;
 260     int fd;
 261     int ret;
 262     pthread_t show_parts_thread;
 263     Error *local_error = NULL;
 264
 265     sioc = qio_channel_socket_new();
 266     if (qio_channel_socket_connect_sync(sioc,
 267                                         saddr,
 268                                         &local_error) < 0) {
 269         error_report_err(local_error);
 270         goto out;
 271     }
 272
 273     ret = nbd_receive_negotiate(QIO_CHANNEL(sioc), NULL, &nbdflags,
 274                                 NULL, NULL, NULL,
 275                                 &size, &local_error);
 276     if (ret < 0) {
 277         if (local_error) {
 278             error_report_err(local_error);
 279         }
 280         goto out_socket;
 281     }
 282
 283     fd = open(device, O_RDWR);
 284     if (fd < 0) {
 285         /* Linux-only, we can use %m in printf.  */
 286         error_report("Failed to open %s: %m", device);
 287         goto out_socket;
 288     }
 289
 290     ret = nbd_init(fd, sioc, nbdflags, size);
 291     if (ret < 0) {
 292         goto out_fd;
 293     }
 294
 295     /* update partition table */
 296     pthread_create(&show_parts_thread, NULL, show_parts, device);
 297
 298     if (verbose) {
 299         fprintf(stderr, "NBD device %s is now connected to %s\n",
 300                 device, srcpath);
 301     } else {
 302         /* Close stderr so that the qemu-nbd process exits.  */
 303         dup2(STDOUT_FILENO, STDERR_FILENO);
 304     }
 305
 306     ret = nbd_client(fd);
 307     if (ret) {
 308         goto out_fd;
 309     }
 310     close(fd);
 311     object_unref(OBJECT(sioc));
 312     kill(getpid(), SIGTERM);
 313     return (void *) EXIT_SUCCESS;
 314
 315 out_fd:
 316     close(fd);
 317 out_socket:
 318     object_unref(OBJECT(sioc));
 319 out:
 320     kill(getpid(), SIGTERM);
 321     return (void *) EXIT_FAILURE;
 322 }
 323
 324 static int nbd_can_accept(void)
 325 {
 326     return nb_fds < shared;
 327 }
 328
 329 static void nbd_export_closed(NBDExport *exp)
 330 {
 331     assert(state == TERMINATING);
 332     state = TERMINATED;
 333 }
 334
 335 static void nbd_update_server_watch(void);
 336
 337 static void nbd_client_closed(NBDClient *client)
 338 {
 339     nb_fds--;
 340     if (nb_fds == 0 && !persistent && state == RUNNING) {
 341         state = TERMINATE;
 342     }
 343     nbd_update_server_watch();
 344     nbd_client_put(client);
 345 }
 346
 347 static gboolean nbd_accept(QIOChannel *ioc, GIOCondition cond, gpointer opaque)
 348 {
 349     QIOChannelSocket *cioc;
 350
 351     cioc = qio_channel_socket_accept(QIO_CHANNEL_SOCKET(ioc),
 352                                      NULL);
 353     if (!cioc) {
 354         return TRUE;
 355     }
 356
 357     if (state >= TERMINATE) {
 358         object_unref(OBJECT(cioc));
 359         return TRUE;
 360     }
 361
 362     nb_fds++;
 363     nbd_update_server_watch();
 364     nbd_client_new(newproto ? NULL : exp, cioc,
 365                    tlscreds, NULL, nbd_client_closed);
 366     object_unref(OBJECT(cioc));
 367
 368     return TRUE;
 369 }
 370
 371 static void nbd_update_server_watch(void)
 372 {
 373     if (nbd_can_accept()) {
 374         if (server_watch == -1) {
 375             server_watch = qio_channel_add_watch(QIO_CHANNEL(server_ioc),
 376                                                  G_IO_IN,
 377                                                  nbd_accept,
 378                                                  NULL, NULL);
 379         }
 380     } else {
 381         if (server_watch != -1) {
 382             g_source_remove(server_watch);
 383             server_watch = -1;
 384         }
 385     }
 386 }
 387
 388
 389 static SocketAddress *nbd_build_socket_address(const char *sockpath,
 390                                                const char *bindto,
 391                                                const char *port)
 392 {
 393     SocketAddress *saddr;
 394
 395     saddr = g_new0(SocketAddress, 1);
 396     if (sockpath) {
 397         saddr->type = SOCKET_ADDRESS_KIND_UNIX;
 398         saddr->u.q_unix.data = g_new0(UnixSocketAddress, 1);
 399         saddr->u.q_unix.data->path = g_strdup(sockpath);
 400     } else {
 401         InetSocketAddress *inet;
 402         saddr->type = SOCKET_ADDRESS_KIND_INET;
 403         inet = saddr->u.inet.data = g_new0(InetSocketAddress, 1);
 404         inet->host = g_strdup(bindto);
 405         if (port) {
 406             inet->port = g_strdup(port);
 407         } else  {
 408             inet->port = g_strdup_printf("%d", NBD_DEFAULT_PORT);
 409         }
 410     }
 411
 412     return saddr;
 413 }
 414
 415
 416 static QemuOptsList file_opts = {
 417     .name = "file",
 418     .implied_opt_name = "file",
 419     .head = QTAILQ_HEAD_INITIALIZER(file_opts.head),
 420     .desc = {
 421         /* no elements => accept any params */
 422         { /* end of list */ }
 423     },
 424 };
 425
 426 static QemuOptsList qemu_object_opts = {
 427     .name = "object",
 428     .implied_opt_name = "qom-type",
 429     .head = QTAILQ_HEAD_INITIALIZER(qemu_object_opts.head),
 430     .desc = {
 431         { }
 432     },
 433 };
 434
 435
 436
 437 static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, Error **errp)
 438 {
 439     Object *obj;
 440     QCryptoTLSCreds *creds;
 441
 442     obj = object_resolve_path_component(
 443         object_get_objects_root(), id);
 444     if (!obj) {
 445         error_setg(errp, "No TLS credentials with id '%s'",
 446                    id);
 447         return NULL;
 448     }
 449     creds = (QCryptoTLSCreds *)
 450         object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
 451     if (!creds) {
 452         error_setg(errp, "Object with id '%s' is not TLS credentials",
 453                    id);
 454         return NULL;
 455     }
 456
 457     if (creds->endpoint != QCRYPTO_TLS_CREDS_ENDPOINT_SERVER) {
 458         error_setg(errp,
 459                    "Expecting TLS credentials with a server endpoint");
 460         return NULL;
 461     }
 462     object_ref(obj);
 463     return creds;
 464 }
 465
 466 static void setup_address_and_port(const char **address, const char **port)
 467 {
 468     if (*address == NULL) {
 469         *address = "0.0.0.0";
 470     }
 471
 472     if (*port == NULL) {
 473         *port = stringify(NBD_DEFAULT_PORT);
 474     }
 475 }
 476
 477 #define FIRST_SOCKET_ACTIVATION_FD 3 /* defined by systemd ABI */
 478
 479 #ifndef _WIN32
 480 /*
 481  * Check if socket activation was requested via use of the
 482  * LISTEN_FDS and LISTEN_PID environment variables.
 483  *
 484  * Returns 0 if no socket activation, or the number of FDs.
 485  */
 486 static unsigned int check_socket_activation(void)
 487 {
 488     const char *s;
 489     unsigned long pid;
 490     unsigned long nr_fds;
 491     unsigned int i;
 492     int fd;
 493     int err;
 494
 495     s = getenv("LISTEN_PID");
 496     if (s == NULL) {
 497         return 0;
 498     }
 499     err = qemu_strtoul(s, NULL, 10, &pid);
 500     if (err) {
 501         if (verbose) {
 502             fprintf(stderr, "malformed %s environment variable (ignored)\n",
 503                     "LISTEN_PID");
 504         }
 505         return 0;
 506     }
 507     if (pid != getpid()) {
 508         if (verbose) {
 509             fprintf(stderr, "%s was not for us (ignored)\n",
 510                     "LISTEN_PID");
 511         }
 512         return 0;
 513     }
 514
 515     s = getenv("LISTEN_FDS");
 516     if (s == NULL) {
 517         return 0;
 518     }
 519     err = qemu_strtoul(s, NULL, 10, &nr_fds);
 520     if (err) {
 521         if (verbose) {
 522             fprintf(stderr, "malformed %s environment variable (ignored)\n",
 523                     "LISTEN_FDS");
 524         }
 525         return 0;
 526     }
 527     assert(nr_fds <= UINT_MAX);
 528
 529     /* A limitation of current qemu-nbd is that it can only listen on
 530      * a single socket.  When that limitation is lifted, we can change
 531      * this function to allow LISTEN_FDS > 1, and remove the assertion
 532      * in the main function below.
 533      */
 534     if (nr_fds > 1) {
 535         error_report("qemu-nbd does not support socket activation with %s > 1",
 536                      "LISTEN_FDS");
 537         exit(EXIT_FAILURE);
 538     }
 539
 540     /* So these are not passed to any child processes we might start. */
 541     unsetenv("LISTEN_FDS");
 542     unsetenv("LISTEN_PID");
 543
 544     /* So the file descriptors don't leak into child processes. */
 545     for (i = 0; i < nr_fds; ++i) {
 546         fd = FIRST_SOCKET_ACTIVATION_FD + i;
 547         if (fcntl(fd, F_SETFD, FD_CLOEXEC) == -1) {
 548             /* If we cannot set FD_CLOEXEC then it probably means the file
 549              * descriptor is invalid, so socket activation has gone wrong
 550              * and we should exit.
 551              */
 552             error_report("Socket activation failed: "
 553                          "invalid file descriptor fd = %d: %m",
 554                          fd);
 555             exit(EXIT_FAILURE);
 556         }
 557     }
 558
 559     return (unsigned int) nr_fds;
 560 }
 561
 562 #else /* !_WIN32 */
 563 static unsigned int check_socket_activation(void)
 564 {
 565     return 0;
 566 }
 567 #endif
 568
 569 /*
 570  * Check socket parameters compatibility when socket activation is used.
 571  */
 572 static const char *socket_activation_validate_opts(const char *device,
 573                                                    const char *sockpath,
 574                                                    const char *address,
 575                                                    const char *port)
 576 {
 577     if (device != NULL) {
 578         return "NBD device can't be set when using socket activation";
 579     }
 580
 581     if (sockpath != NULL) {
 582         return "Unix socket can't be set when using socket activation";
 583     }
 584
 585     if (address != NULL) {
 586         return "The interface can't be set when using socket activation";
 587     }
 588
 589     if (port != NULL) {
 590         return "TCP port number can't be set when using socket activation";
 591     }
 592
 593     return NULL;
 594 }
 595
 596 int main(int argc, char **argv)
 597 {
 598     BlockBackend *blk;
 599     BlockDriverState *bs;
 600     off_t dev_offset = 0;
 601     uint16_t nbdflags = 0;
 602     bool disconnect = false;
 603     const char *bindto = NULL;
 604     const char *port = NULL;
 605     char *sockpath = NULL;
 606     char *device = NULL;
 607     off_t fd_size;
 608     QemuOpts *sn_opts = NULL;
 609     const char *sn_id_or_name = NULL;
 610     const char *sopt = "hVb:o:p:rsnP:c:dvk:e:f:tl:x:T:D:";
 611     struct option lopt[] = {
 612         { "help", no_argument, NULL, 'h' },
 613         { "version", no_argument, NULL, 'V' },
 614         { "bind", required_argument, NULL, 'b' },
 615         { "port", required_argument, NULL, 'p' },
 616         { "socket", required_argument, NULL, 'k' },
 617         { "offset", required_argument, NULL, 'o' },
 618         { "read-only", no_argument, NULL, 'r' },
 619         { "partition", required_argument, NULL, 'P' },
 620         { "connect", required_argument, NULL, 'c' },
 621         { "disconnect", no_argument, NULL, 'd' },
 622         { "snapshot", no_argument, NULL, 's' },
 623         { "load-snapshot", required_argument, NULL, 'l' },
 624         { "nocache", no_argument, NULL, 'n' },
 625         { "cache", required_argument, NULL, QEMU_NBD_OPT_CACHE },
 626         { "aio", required_argument, NULL, QEMU_NBD_OPT_AIO },
 627         { "discard", required_argument, NULL, QEMU_NBD_OPT_DISCARD },
 628         { "detect-zeroes", required_argument, NULL,
 629           QEMU_NBD_OPT_DETECT_ZEROES },
 630         { "shared", required_argument, NULL, 'e' },
 631         { "format", required_argument, NULL, 'f' },
 632         { "persistent", no_argument, NULL, 't' },
 633         { "verbose", no_argument, NULL, 'v' },
 634         { "object", required_argument, NULL, QEMU_NBD_OPT_OBJECT },
 635         { "export-name", required_argument, NULL, 'x' },
 636         { "description", required_argument, NULL, 'D' },
 637         { "tls-creds", required_argument, NULL, QEMU_NBD_OPT_TLSCREDS },
 638         { "image-opts", no_argument, NULL, QEMU_NBD_OPT_IMAGE_OPTS },
 639         { "trace", required_argument, NULL, 'T' },
 640         { "fork", no_argument, NULL, QEMU_NBD_OPT_FORK },
 641         { NULL, 0, NULL, 0 }
 642     };
 643     int ch;
 644     int opt_ind = 0;
 645     char *end;
 646     int flags = BDRV_O_RDWR;
 647     int partition = -1;
 648     int ret = 0;
 649     bool seen_cache = false;
 650     bool seen_discard = false;
 651     bool seen_aio = false;
 652     pthread_t client_thread;
 653     const char *fmt = NULL;
 654     Error *local_err = NULL;
 655     BlockdevDetectZeroesOptions detect_zeroes = BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
 656     QDict *options = NULL;
 657     const char *export_name = NULL;
 658     const char *export_description = NULL;
 659     const char *tlscredsid = NULL;
 660     bool imageOpts = false;
 661     bool writethrough = true;
 662     char *trace_file = NULL;
 663     bool fork_process = false;
 664     int old_stderr = -1;
 665     unsigned socket_activation;
 666
 667     /* The client thread uses SIGTERM to interrupt the server.  A signal
 668      * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
 669      */
 670     struct sigaction sa_sigterm;
 671     memset(&sa_sigterm, 0, sizeof(sa_sigterm));
 672     sa_sigterm.sa_handler = termsig_handler;
 673     sigaction(SIGTERM, &sa_sigterm, NULL);
 674
 675     module_call_init(MODULE_INIT_TRACE);
 676     qcrypto_init(&error_fatal);
 677
 678     module_call_init(MODULE_INIT_QOM);
 679     qemu_add_opts(&qemu_object_opts);
 680     qemu_add_opts(&qemu_trace_opts);
 681     qemu_init_exec_dir(argv[0]);
 682
 683     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 684         switch (ch) {
 685         case 's':
 686             flags |= BDRV_O_SNAPSHOT;
 687             break;
 688         case 'n':
 689             optarg = (char *) "none";
 690             /* fallthrough */
 691         case QEMU_NBD_OPT_CACHE:
 692             if (seen_cache) {
 693                 error_report("-n and --cache can only be specified once");
 694                 exit(EXIT_FAILURE);
 695             }
 696             seen_cache = true;
 697             if (bdrv_parse_cache_mode(optarg, &flags, &writethrough) == -1) {
 698                 error_report("Invalid cache mode `%s'", optarg);
 699                 exit(EXIT_FAILURE);
 700             }
 701             break;
 702         case QEMU_NBD_OPT_AIO:
 703             if (seen_aio) {
 704                 error_report("--aio can only be specified once");
 705                 exit(EXIT_FAILURE);
 706             }
 707             seen_aio = true;
 708             if (!strcmp(optarg, "native")) {
 709                 flags |= BDRV_O_NATIVE_AIO;
 710             } else if (!strcmp(optarg, "threads")) {
 711                 /* this is the default */
 712             } else {
 713                error_report("invalid aio mode `%s'", optarg);
 714                exit(EXIT_FAILURE);
 715             }
 716             break;
 717         case QEMU_NBD_OPT_DISCARD:
 718             if (seen_discard) {
 719                 error_report("--discard can only be specified once");
 720                 exit(EXIT_FAILURE);
 721             }
 722             seen_discard = true;
 723             if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
 724                 error_report("Invalid discard mode `%s'", optarg);
 725                 exit(EXIT_FAILURE);
 726             }
 727             break;
 728         case QEMU_NBD_OPT_DETECT_ZEROES:
 729             detect_zeroes =
 730                 qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
 731                                 optarg,
 732                                 BLOCKDEV_DETECT_ZEROES_OPTIONS__MAX,
 733                                 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
 734                                 &local_err);
 735             if (local_err) {
 736                 error_reportf_err(local_err,
 737                                   "Failed to parse detect_zeroes mode: ");
 738                 exit(EXIT_FAILURE);
 739             }
 740             if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
 741                 !(flags & BDRV_O_UNMAP)) {
 742                 error_report("setting detect-zeroes to unmap is not allowed "
 743                              "without setting discard operation to unmap");
 744                 exit(EXIT_FAILURE);
 745             }
 746             break;
 747         case 'b':
 748             bindto = optarg;
 749             break;
 750         case 'p':
 751             port = optarg;
 752             break;
 753         case 'o':
 754                 dev_offset = strtoll (optarg, &end, 0);
 755             if (*end) {
 756                 error_report("Invalid offset `%s'", optarg);
 757                 exit(EXIT_FAILURE);
 758             }
 759             if (dev_offset < 0) {
 760                 error_report("Offset must be positive `%s'", optarg);
 761                 exit(EXIT_FAILURE);
 762             }
 763             break;
 764         case 'l':
 765             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
 766                 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
 767                                                   optarg, false);
 768                 if (!sn_opts) {
 769                     error_report("Failed in parsing snapshot param `%s'",
 770                                  optarg);
 771                     exit(EXIT_FAILURE);
 772                 }
 773             } else {
 774                 sn_id_or_name = optarg;
 775             }
 776             /* fall through */
 777         case 'r':
 778             nbdflags |= NBD_FLAG_READ_ONLY;
 779             flags &= ~BDRV_O_RDWR;
 780             break;
 781         case 'P':
 782             partition = strtol(optarg, &end, 0);
 783             if (*end) {
 784                 error_report("Invalid partition `%s'", optarg);
 785                 exit(EXIT_FAILURE);
 786             }
 787             if (partition < 1 || partition > 8) {
 788                 error_report("Invalid partition %d", partition);
 789                 exit(EXIT_FAILURE);
 790             }
 791             break;
 792         case 'k':
 793             sockpath = optarg;
 794             if (sockpath[0] != '/') {
 795                 error_report("socket path must be absolute");
 796                 exit(EXIT_FAILURE);
 797             }
 798             break;
 799         case 'd':
 800             disconnect = true;
 801             break;
 802         case 'c':
 803             device = optarg;
 804             break;
 805         case 'e':
 806             shared = strtol(optarg, &end, 0);
 807             if (*end) {
 808                 error_report("Invalid shared device number '%s'", optarg);
 809                 exit(EXIT_FAILURE);
 810             }
 811             if (shared < 1) {
 812                 error_report("Shared device number must be greater than 0");
 813                 exit(EXIT_FAILURE);
 814             }
 815             break;
 816         case 'f':
 817             fmt = optarg;
 818             break;
 819         case 't':
 820             persistent = 1;
 821             break;
 822         case 'x':
 823             export_name = optarg;
 824             break;
 825         case 'D':
 826             export_description = optarg;
 827             break;
 828         case 'v':
 829             verbose = 1;
 830             break;
 831         case 'V':
 832             version(argv[0]);
 833             exit(0);
 834             break;
 835         case 'h':
 836             usage(argv[0]);
 837             exit(0);
 838             break;
 839         case '?':
 840             error_report("Try `%s --help' for more information.", argv[0]);
 841             exit(EXIT_FAILURE);
 842         case QEMU_NBD_OPT_OBJECT: {
 843             QemuOpts *opts;
 844             opts = qemu_opts_parse_noisily(&qemu_object_opts,
 845                                            optarg, true);
 846             if (!opts) {
 847                 exit(EXIT_FAILURE);
 848             }
 849         }   break;
 850         case QEMU_NBD_OPT_TLSCREDS:
 851             tlscredsid = optarg;
 852             break;
 853         case QEMU_NBD_OPT_IMAGE_OPTS:
 854             imageOpts = true;
 855             break;
 856         case 'T':
 857             g_free(trace_file);
 858             trace_file = trace_opt_parse(optarg);
 859             break;
 860         case QEMU_NBD_OPT_FORK:
 861             fork_process = true;
 862             break;
 863         }
 864     }
 865
 866     if ((argc - optind) != 1) {
 867         error_report("Invalid number of arguments");
 868         error_printf("Try `%s --help' for more information.\n", argv[0]);
 869         exit(EXIT_FAILURE);
 870     }
 871
 872     if (qemu_opts_foreach(&qemu_object_opts,
 873                           user_creatable_add_opts_foreach,
 874                           NULL, NULL)) {
 875         exit(EXIT_FAILURE);
 876     }
 877
 878     if (!trace_init_backends()) {
 879         exit(1);
 880     }
 881     trace_init_file(trace_file);
 882     qemu_set_log(LOG_TRACE);
 883
 884     socket_activation = check_socket_activation();
 885     if (socket_activation == 0) {
 886         setup_address_and_port(&bindto, &port);
 887     } else {
 888         /* Using socket activation - check user didn't use -p etc. */
 889         const char *err_msg = socket_activation_validate_opts(device, sockpath,
 890                                                               bindto, port);
 891         if (err_msg != NULL) {
 892             error_report("%s", err_msg);
 893             exit(EXIT_FAILURE);
 894         }
 895     }
 896
 897     if (tlscredsid) {
 898         if (sockpath) {
 899             error_report("TLS is only supported with IPv4/IPv6");
 900             exit(EXIT_FAILURE);
 901         }
 902         if (device) {
 903             error_report("TLS is not supported with a host device");
 904             exit(EXIT_FAILURE);
 905         }
 906         if (!export_name) {
 907             /* Set the default NBD protocol export name, since
 908              * we *must* use new style protocol for TLS */
 909             export_name = "";
 910         }
 911         tlscreds = nbd_get_tls_creds(tlscredsid, &local_err);
 912         if (local_err) {
 913             error_report("Failed to get TLS creds %s",
 914                          error_get_pretty(local_err));
 915             exit(EXIT_FAILURE);
 916         }
 917     }
 918
 919     if (disconnect) {
 920         int nbdfd = open(argv[optind], O_RDWR);
 921         if (nbdfd < 0) {
 922             error_report("Cannot open %s: %s", argv[optind],
 923                          strerror(errno));
 924             exit(EXIT_FAILURE);
 925         }
 926         nbd_disconnect(nbdfd);
 927
 928         close(nbdfd);
 929
 930         printf("%s disconnected\n", argv[optind]);
 931
 932         return 0;
 933     }
 934
 935     if ((device && !verbose) || fork_process) {
 936         int stderr_fd[2];
 937         pid_t pid;
 938         int ret;
 939
 940         if (qemu_pipe(stderr_fd) < 0) {
 941             error_report("Error setting up communication pipe: %s",
 942                          strerror(errno));
 943             exit(EXIT_FAILURE);
 944         }
 945
 946         /* Now daemonize, but keep a communication channel open to
 947          * print errors and exit with the proper status code.
 948          */
 949         pid = fork();
 950         if (pid < 0) {
 951             error_report("Failed to fork: %s", strerror(errno));
 952             exit(EXIT_FAILURE);
 953         } else if (pid == 0) {
 954             close(stderr_fd[0]);
 955             ret = qemu_daemon(1, 0);
 956
 957             /* Temporarily redirect stderr to the parent's pipe...  */
 958             old_stderr = dup(STDERR_FILENO);
 959             dup2(stderr_fd[1], STDERR_FILENO);
 960             if (ret < 0) {
 961                 error_report("Failed to daemonize: %s", strerror(errno));
 962                 exit(EXIT_FAILURE);
 963             }
 964
 965             /* ... close the descriptor we inherited and go on.  */
 966             close(stderr_fd[1]);
 967         } else {
 968             bool errors = false;
 969             char *buf;
 970
 971             /* In the parent.  Print error messages from the child until
 972              * it closes the pipe.
 973              */
 974             close(stderr_fd[1]);
 975             buf = g_malloc(1024);
 976             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
 977                 errors = true;
 978                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
 979                 if (ret < 0) {
 980                     exit(EXIT_FAILURE);
 981                 }
 982             }
 983             if (ret < 0) {
 984                 error_report("Cannot read from daemon: %s",
 985                              strerror(errno));
 986                 exit(EXIT_FAILURE);
 987             }
 988
 989             /* Usually the daemon should not print any message.
 990              * Exit with zero status in that case.
 991              */
 992             exit(errors);
 993         }
 994     }
 995
 996     if (device != NULL && sockpath == NULL) {
 997         sockpath = g_malloc(128);
 998         snprintf(sockpath, 128, SOCKET_PATH, basename(device));
 999     }
1000
1001     if (socket_activation == 0) {
1002         server_ioc = qio_channel_socket_new();
1003         saddr = nbd_build_socket_address(sockpath, bindto, port);
1004         if (qio_channel_socket_listen_sync(server_ioc, saddr, &local_err) < 0) {
1005             object_unref(OBJECT(server_ioc));
1006             error_report_err(local_err);
1007             return 1;
1008         }
1009     } else {
1010         /* See comment in check_socket_activation above. */
1011         assert(socket_activation == 1);
1012         server_ioc = qio_channel_socket_new_fd(FIRST_SOCKET_ACTIVATION_FD,
1013                                                &local_err);
1014         if (server_ioc == NULL) {
1015             error_report("Failed to use socket activation: %s",
1016                          error_get_pretty(local_err));
1017             exit(EXIT_FAILURE);
1018         }
1019     }
1020
1021     if (qemu_init_main_loop(&local_err)) {
1022         error_report_err(local_err);
1023         exit(EXIT_FAILURE);
1024     }
1025     bdrv_init();
1026     atexit(bdrv_close_all);
1027
1028     srcpath = argv[optind];
1029     if (imageOpts) {
1030         QemuOpts *opts;
1031         if (fmt) {
1032             error_report("--image-opts and -f are mutually exclusive");
1033             exit(EXIT_FAILURE);
1034         }
1035         opts = qemu_opts_parse_noisily(&file_opts, srcpath, true);
1036         if (!opts) {
1037             qemu_opts_reset(&file_opts);
1038             exit(EXIT_FAILURE);
1039         }
1040         options = qemu_opts_to_qdict(opts, NULL);
1041         qemu_opts_reset(&file_opts);
1042         blk = blk_new_open(NULL, NULL, options, flags, &local_err);
1043     } else {
1044         if (fmt) {
1045             options = qdict_new();
1046             qdict_put(options, "driver", qstring_from_str(fmt));
1047         }
1048         blk = blk_new_open(srcpath, NULL, options, flags, &local_err);
1049     }
1050
1051     if (!blk) {
1052         error_reportf_err(local_err, "Failed to blk_new_open '%s': ",
1053                           argv[optind]);
1054         exit(EXIT_FAILURE);
1055     }
1056     bs = blk_bs(blk);
1057
1058     blk_set_enable_write_cache(blk, !writethrough);
1059
1060     if (sn_opts) {
1061         ret = bdrv_snapshot_load_tmp(bs,
1062                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
1063                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
1064                                      &local_err);
1065     } else if (sn_id_or_name) {
1066         ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
1067                                                    &local_err);
1068     }
1069     if (ret < 0) {
1070         error_reportf_err(local_err, "Failed to load snapshot: ");
1071         exit(EXIT_FAILURE);
1072     }
1073
1074     bs->detect_zeroes = detect_zeroes;
1075     fd_size = blk_getlength(blk);
1076     if (fd_size < 0) {
1077         error_report("Failed to determine the image length: %s",
1078                      strerror(-fd_size));
1079         exit(EXIT_FAILURE);
1080     }
1081
1082     if (dev_offset >= fd_size) {
1083         error_report("Offset (%lld) has to be smaller than the image size "
1084                      "(%lld)",
1085                      (long long int)dev_offset, (long long int)fd_size);
1086         exit(EXIT_FAILURE);
1087     }
1088     fd_size -= dev_offset;
1089
1090     if (partition != -1) {
1091         ret = find_partition(blk, partition, &dev_offset, &fd_size);
1092         if (ret < 0) {
1093             error_report("Could not find partition %d: %s", partition,
1094                          strerror(-ret));
1095             exit(EXIT_FAILURE);
1096         }
1097     }
1098
1099     exp = nbd_export_new(bs, dev_offset, fd_size, nbdflags, nbd_export_closed,
1100                          writethrough, NULL, &local_err);
1101     if (!exp) {
1102         error_report_err(local_err);
1103         exit(EXIT_FAILURE);
1104     }
1105     if (export_name) {
1106         nbd_export_set_name(exp, export_name);
1107         nbd_export_set_description(exp, export_description);
1108         newproto = true;
1109     } else if (export_description) {
1110         error_report("Export description requires an export name");
1111         exit(EXIT_FAILURE);
1112     }
1113
1114     if (device) {
1115         int ret;
1116
1117         ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
1118         if (ret != 0) {
1119             error_report("Failed to create client thread: %s", strerror(ret));
1120             exit(EXIT_FAILURE);
1121         }
1122     } else {
1123         /* Shut up GCC warnings.  */
1124         memset(&client_thread, 0, sizeof(client_thread));
1125     }
1126
1127     nbd_update_server_watch();
1128
1129     /* now when the initialization is (almost) complete, chdir("/")
1130      * to free any busy filesystems */
1131     if (chdir("/") < 0) {
1132         error_report("Could not chdir to root directory: %s",
1133                      strerror(errno));
1134         exit(EXIT_FAILURE);
1135     }
1136
1137     if (fork_process) {
1138         dup2(old_stderr, STDERR_FILENO);
1139         close(old_stderr);
1140     }
1141
1142     state = RUNNING;
1143     do {
1144         main_loop_wait(false);
1145         if (state == TERMINATE) {
1146             state = TERMINATING;
1147             nbd_export_close(exp);
1148             nbd_export_put(exp);
1149             exp = NULL;
1150         }
1151     } while (state != TERMINATED);
1152
1153     blk_unref(blk);
1154     if (sockpath) {
1155         unlink(sockpath);
1156     }
1157
1158     qemu_opts_del(sn_opts);
1159
1160     if (device) {
1161         void *ret;
1162         pthread_join(client_thread, &ret);
1163         exit(ret != NULL);
1164     } else {
1165         exit(EXIT_SUCCESS);
1166     }
1167 }