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-common.h"
  20 #include "block.h"
  21 #include "nbd.h"
  22
  23 #include <stdarg.h>
  24 #include <stdio.h>
  25 #include <getopt.h>
  26 #include <err.h>
  27 #include <sys/types.h>
  28 #include <sys/socket.h>
  29 #include <netinet/in.h>
  30 #include <netinet/tcp.h>
  31 #include <arpa/inet.h>
  32 #include <signal.h>
  33 #include <libgen.h>
  34 #include <pthread.h>
  35
  36 #define SOCKET_PATH    "/var/lock/qemu-nbd-%s"
  37
  38 static NBDExport *exp;
  39 static int verbose;
  40 static char *srcpath;
  41 static char *sockpath;
  42 static bool sigterm_reported;
  43 static bool nbd_started;
  44 static int shared = 1;
  45 static int nb_fds;
  46
  47 static void usage(const char *name)
  48 {
  49     (printf) (
  50 "Usage: %s [OPTIONS] FILE\n"
  51 "QEMU Disk Network Block Device Server\n"
  52 "\n"
  53 "  -h, --help           display this help and exit\n"
  54 "  -V, --version        output version information and exit\n"
  55 "\n"
  56 "Connection properties:\n"
  57 "  -p, --port=PORT      port to listen on (default `%d')\n"
  58 "  -b, --bind=IFACE     interface to bind to (default `0.0.0.0')\n"
  59 "  -k, --socket=PATH    path to the unix socket\n"
  60 "                       (default '"SOCKET_PATH"')\n"
  61 "  -e, --shared=NUM     device can be shared by NUM clients (default '1')\n"
  62 "  -t, --persistent     don't exit on the last connection\n"
  63 "  -v, --verbose        display extra debugging information\n"
  64 "\n"
  65 "Exposing part of the image:\n"
  66 "  -o, --offset=OFFSET  offset into the image\n"
  67 "  -P, --partition=NUM  only expose partition NUM\n"
  68 "\n"
  69 #ifdef __linux__
  70 "Kernel NBD client support:\n"
  71 "  -c, --connect=DEV    connect FILE to the local NBD device DEV\n"
  72 "  -d, --disconnect     disconnect the specified device\n"
  73 "\n"
  74 #endif
  75 "\n"
  76 "Block device options:\n"
  77 "  -r, --read-only      export read-only\n"
  78 "  -s, --snapshot       use snapshot file\n"
  79 "  -n, --nocache        disable host cache\n"
  80 "\n"
  81 "Report bugs to <qemu-devel@nongnu.org>\n"
  82     , name, NBD_DEFAULT_PORT, "DEVICE");
  83 }
  84
  85 static void version(const char *name)
  86 {
  87     printf(
  88 "%s version 0.0.1\n"
  89 "Written by Anthony Liguori.\n"
  90 "\n"
  91 "Copyright (C) 2006 Anthony Liguori <anthony@codemonkey.ws>.\n"
  92 "This is free software; see the source for copying conditions.  There is NO\n"
  93 "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
  94     , name);
  95 }
  96
  97 struct partition_record
  98 {
  99     uint8_t bootable;
 100     uint8_t start_head;
 101     uint32_t start_cylinder;
 102     uint8_t start_sector;
 103     uint8_t system;
 104     uint8_t end_head;
 105     uint8_t end_cylinder;
 106     uint8_t end_sector;
 107     uint32_t start_sector_abs;
 108     uint32_t nb_sectors_abs;
 109 };
 110
 111 static void read_partition(uint8_t *p, struct partition_record *r)
 112 {
 113     r->bootable = p[0];
 114     r->start_head = p[1];
 115     r->start_cylinder = p[3] | ((p[2] << 2) & 0x0300);
 116     r->start_sector = p[2] & 0x3f;
 117     r->system = p[4];
 118     r->end_head = p[5];
 119     r->end_cylinder = p[7] | ((p[6] << 2) & 0x300);
 120     r->end_sector = p[6] & 0x3f;
 121     r->start_sector_abs = p[8] | p[9] << 8 | p[10] << 16 | p[11] << 24;
 122     r->nb_sectors_abs = p[12] | p[13] << 8 | p[14] << 16 | p[15] << 24;
 123 }
 124
 125 static int find_partition(BlockDriverState *bs, int partition,
 126                           off_t *offset, off_t *size)
 127 {
 128     struct partition_record mbr[4];
 129     uint8_t data[512];
 130     int i;
 131     int ext_partnum = 4;
 132     int ret;
 133
 134     if ((ret = bdrv_read(bs, 0, data, 1)) < 0) {
 135         errno = -ret;
 136         err(EXIT_FAILURE, "error while reading");
 137     }
 138
 139     if (data[510] != 0x55 || data[511] != 0xaa) {
 140         return -EINVAL;
 141     }
 142
 143     for (i = 0; i < 4; i++) {
 144         read_partition(&data[446 + 16 * i], &mbr[i]);
 145
 146         if (!mbr[i].nb_sectors_abs)
 147             continue;
 148
 149         if (mbr[i].system == 0xF || mbr[i].system == 0x5) {
 150             struct partition_record ext[4];
 151             uint8_t data1[512];
 152             int j;
 153
 154             if ((ret = bdrv_read(bs, mbr[i].start_sector_abs, data1, 1)) < 0) {
 155                 errno = -ret;
 156                 err(EXIT_FAILURE, "error while reading");
 157             }
 158
 159             for (j = 0; j < 4; j++) {
 160                 read_partition(&data1[446 + 16 * j], &ext[j]);
 161                 if (!ext[j].nb_sectors_abs)
 162                     continue;
 163
 164                 if ((ext_partnum + j + 1) == partition) {
 165                     *offset = (uint64_t)ext[j].start_sector_abs << 9;
 166                     *size = (uint64_t)ext[j].nb_sectors_abs << 9;
 167                     return 0;
 168                 }
 169             }
 170             ext_partnum += 4;
 171         } else if ((i + 1) == partition) {
 172             *offset = (uint64_t)mbr[i].start_sector_abs << 9;
 173             *size = (uint64_t)mbr[i].nb_sectors_abs << 9;
 174             return 0;
 175         }
 176     }
 177
 178     return -ENOENT;
 179 }
 180
 181 static void termsig_handler(int signum)
 182 {
 183     sigterm_reported = true;
 184     qemu_notify_event();
 185 }
 186
 187 static void *show_parts(void *arg)
 188 {
 189     char *device = arg;
 190     int nbd;
 191
 192     /* linux just needs an open() to trigger
 193      * the partition table update
 194      * but remember to load the module with max_part != 0 :
 195      *     modprobe nbd max_part=63
 196      */
 197     nbd = open(device, O_RDWR);
 198     if (nbd >= 0) {
 199         close(nbd);
 200     }
 201     return NULL;
 202 }
 203
 204 static void *nbd_client_thread(void *arg)
 205 {
 206     char *device = arg;
 207     off_t size;
 208     size_t blocksize;
 209     uint32_t nbdflags;
 210     int fd, sock;
 211     int ret;
 212     pthread_t show_parts_thread;
 213
 214     sock = unix_socket_outgoing(sockpath);
 215     if (sock < 0) {
 216         goto out;
 217     }
 218
 219     ret = nbd_receive_negotiate(sock, NULL, &nbdflags,
 220                                 &size, &blocksize);
 221     if (ret < 0) {
 222         goto out;
 223     }
 224
 225     fd = open(device, O_RDWR);
 226     if (fd < 0) {
 227         /* Linux-only, we can use %m in printf.  */
 228         fprintf(stderr, "Failed to open %s: %m", device);
 229         goto out;
 230     }
 231
 232     ret = nbd_init(fd, sock, nbdflags, size, blocksize);
 233     if (ret < 0) {
 234         goto out;
 235     }
 236
 237     /* update partition table */
 238     pthread_create(&show_parts_thread, NULL, show_parts, device);
 239
 240     if (verbose) {
 241         fprintf(stderr, "NBD device %s is now connected to %s\n",
 242                 device, srcpath);
 243     } else {
 244         /* Close stderr so that the qemu-nbd process exits.  */
 245         dup2(STDOUT_FILENO, STDERR_FILENO);
 246     }
 247
 248     ret = nbd_client(fd);
 249     if (ret) {
 250         goto out;
 251     }
 252     close(fd);
 253     kill(getpid(), SIGTERM);
 254     return (void *) EXIT_SUCCESS;
 255
 256 out:
 257     kill(getpid(), SIGTERM);
 258     return (void *) EXIT_FAILURE;
 259 }
 260
 261 static int nbd_can_accept(void *opaque)
 262 {
 263     return nb_fds < shared;
 264 }
 265
 266 static void nbd_client_closed(NBDClient *client)
 267 {
 268     nb_fds--;
 269     qemu_notify_event();
 270 }
 271
 272 static void nbd_accept(void *opaque)
 273 {
 274     int server_fd = (uintptr_t) opaque;
 275     struct sockaddr_in addr;
 276     socklen_t addr_len = sizeof(addr);
 277
 278     int fd = accept(server_fd, (struct sockaddr *)&addr, &addr_len);
 279     nbd_started = true;
 280     if (fd >= 0 && nbd_client_new(exp, fd, nbd_client_closed)) {
 281         nb_fds++;
 282     }
 283 }
 284
 285 int main(int argc, char **argv)
 286 {
 287     BlockDriverState *bs;
 288     off_t dev_offset = 0;
 289     uint32_t nbdflags = 0;
 290     bool disconnect = false;
 291     const char *bindto = "0.0.0.0";
 292     char *device = NULL;
 293     int port = NBD_DEFAULT_PORT;
 294     off_t fd_size;
 295     const char *sopt = "hVb:o:p:rsnP:c:dvk:e:t";
 296     struct option lopt[] = {
 297         { "help", 0, NULL, 'h' },
 298         { "version", 0, NULL, 'V' },
 299         { "bind", 1, NULL, 'b' },
 300         { "port", 1, NULL, 'p' },
 301         { "socket", 1, NULL, 'k' },
 302         { "offset", 1, NULL, 'o' },
 303         { "read-only", 0, NULL, 'r' },
 304         { "partition", 1, NULL, 'P' },
 305         { "connect", 1, NULL, 'c' },
 306         { "disconnect", 0, NULL, 'd' },
 307         { "snapshot", 0, NULL, 's' },
 308         { "nocache", 0, NULL, 'n' },
 309         { "shared", 1, NULL, 'e' },
 310         { "persistent", 0, NULL, 't' },
 311         { "verbose", 0, NULL, 'v' },
 312         { NULL, 0, NULL, 0 }
 313     };
 314     int ch;
 315     int opt_ind = 0;
 316     int li;
 317     char *end;
 318     int flags = BDRV_O_RDWR;
 319     int partition = -1;
 320     int ret;
 321     int fd;
 322     int persistent = 0;
 323     pthread_t client_thread;
 324
 325     /* The client thread uses SIGTERM to interrupt the server.  A signal
 326      * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
 327      */
 328     struct sigaction sa_sigterm;
 329     memset(&sa_sigterm, 0, sizeof(sa_sigterm));
 330     sa_sigterm.sa_handler = termsig_handler;
 331     sigaction(SIGTERM, &sa_sigterm, NULL);
 332
 333     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 334         switch (ch) {
 335         case 's':
 336             flags |= BDRV_O_SNAPSHOT;
 337             break;
 338         case 'n':
 339             flags |= BDRV_O_NOCACHE | BDRV_O_CACHE_WB;
 340             break;
 341         case 'b':
 342             bindto = optarg;
 343             break;
 344         case 'p':
 345             li = strtol(optarg, &end, 0);
 346             if (*end) {
 347                 errx(EXIT_FAILURE, "Invalid port `%s'", optarg);
 348             }
 349             if (li < 1 || li > 65535) {
 350                 errx(EXIT_FAILURE, "Port out of range `%s'", optarg);
 351             }
 352             port = (uint16_t)li;
 353             break;
 354         case 'o':
 355                 dev_offset = strtoll (optarg, &end, 0);
 356             if (*end) {
 357                 errx(EXIT_FAILURE, "Invalid offset `%s'", optarg);
 358             }
 359             if (dev_offset < 0) {
 360                 errx(EXIT_FAILURE, "Offset must be positive `%s'", optarg);
 361             }
 362             break;
 363         case 'r':
 364             nbdflags |= NBD_FLAG_READ_ONLY;
 365             flags &= ~BDRV_O_RDWR;
 366             break;
 367         case 'P':
 368             partition = strtol(optarg, &end, 0);
 369             if (*end)
 370                 errx(EXIT_FAILURE, "Invalid partition `%s'", optarg);
 371             if (partition < 1 || partition > 8)
 372                 errx(EXIT_FAILURE, "Invalid partition %d", partition);
 373             break;
 374         case 'k':
 375             sockpath = optarg;
 376             if (sockpath[0] != '/')
 377                 errx(EXIT_FAILURE, "socket path must be absolute\n");
 378             break;
 379         case 'd':
 380             disconnect = true;
 381             break;
 382         case 'c':
 383             device = optarg;
 384             break;
 385         case 'e':
 386             shared = strtol(optarg, &end, 0);
 387             if (*end) {
 388                 errx(EXIT_FAILURE, "Invalid shared device number '%s'", optarg);
 389             }
 390             if (shared < 1) {
 391                 errx(EXIT_FAILURE, "Shared device number must be greater than 0\n");
 392             }
 393             break;
 394         case 't':
 395             persistent = 1;
 396             break;
 397         case 'v':
 398             verbose = 1;
 399             break;
 400         case 'V':
 401             version(argv[0]);
 402             exit(0);
 403             break;
 404         case 'h':
 405             usage(argv[0]);
 406             exit(0);
 407             break;
 408         case '?':
 409             errx(EXIT_FAILURE, "Try `%s --help' for more information.",
 410                  argv[0]);
 411         }
 412     }
 413
 414     if ((argc - optind) != 1) {
 415         errx(EXIT_FAILURE, "Invalid number of argument.\n"
 416              "Try `%s --help' for more information.",
 417              argv[0]);
 418     }
 419
 420     if (disconnect) {
 421         fd = open(argv[optind], O_RDWR);
 422         if (fd < 0) {
 423             err(EXIT_FAILURE, "Cannot open %s", argv[optind]);
 424         }
 425         nbd_disconnect(fd);
 426
 427         close(fd);
 428
 429         printf("%s disconnected\n", argv[optind]);
 430
 431         return 0;
 432     }
 433
 434     if (device && !verbose) {
 435         int stderr_fd[2];
 436         pid_t pid;
 437         int ret;
 438
 439         if (qemu_pipe(stderr_fd) < 0) {
 440             err(EXIT_FAILURE, "Error setting up communication pipe");
 441         }
 442
 443         /* Now daemonize, but keep a communication channel open to
 444          * print errors and exit with the proper status code.
 445          */
 446         pid = fork();
 447         if (pid == 0) {
 448             close(stderr_fd[0]);
 449             ret = qemu_daemon(1, 0);
 450
 451             /* Temporarily redirect stderr to the parent's pipe...  */
 452             dup2(stderr_fd[1], STDERR_FILENO);
 453             if (ret < 0) {
 454                 err(EXIT_FAILURE, "Failed to daemonize");
 455             }
 456
 457             /* ... close the descriptor we inherited and go on.  */
 458             close(stderr_fd[1]);
 459         } else {
 460             bool errors = false;
 461             char *buf;
 462
 463             /* In the parent.  Print error messages from the child until
 464              * it closes the pipe.
 465              */
 466             close(stderr_fd[1]);
 467             buf = g_malloc(1024);
 468             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
 469                 errors = true;
 470                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
 471                 if (ret < 0) {
 472                     exit(EXIT_FAILURE);
 473                 }
 474             }
 475             if (ret < 0) {
 476                 err(EXIT_FAILURE, "Cannot read from daemon");
 477             }
 478
 479             /* Usually the daemon should not print any message.
 480              * Exit with zero status in that case.
 481              */
 482             exit(errors);
 483         }
 484     }
 485
 486     if (device != NULL && sockpath == NULL) {
 487         sockpath = g_malloc(128);
 488         snprintf(sockpath, 128, SOCKET_PATH, basename(device));
 489     }
 490
 491     bdrv_init();
 492     atexit(bdrv_close_all);
 493
 494     bs = bdrv_new("hda");
 495     srcpath = argv[optind];
 496     if ((ret = bdrv_open(bs, srcpath, flags, NULL)) < 0) {
 497         errno = -ret;
 498         err(EXIT_FAILURE, "Failed to bdrv_open '%s'", argv[optind]);
 499     }
 500
 501     fd_size = bdrv_getlength(bs);
 502
 503     if (partition != -1) {
 504         ret = find_partition(bs, partition, &dev_offset, &fd_size);
 505         if (ret < 0) {
 506             errno = -ret;
 507             err(EXIT_FAILURE, "Could not find partition %d", partition);
 508         }
 509     }
 510
 511     exp = nbd_export_new(bs, dev_offset, fd_size, nbdflags);
 512
 513     if (sockpath) {
 514         fd = unix_socket_incoming(sockpath);
 515     } else {
 516         fd = tcp_socket_incoming(bindto, port);
 517     }
 518
 519     if (fd < 0) {
 520         return 1;
 521     }
 522
 523     if (device) {
 524         int ret;
 525
 526         ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
 527         if (ret != 0) {
 528             errx(EXIT_FAILURE, "Failed to create client thread: %s",
 529                  strerror(ret));
 530         }
 531     } else {
 532         /* Shut up GCC warnings.  */
 533         memset(&client_thread, 0, sizeof(client_thread));
 534     }
 535
 536     qemu_init_main_loop();
 537     qemu_set_fd_handler2(fd, nbd_can_accept, nbd_accept, NULL,
 538                          (void *)(uintptr_t)fd);
 539
 540     /* now when the initialization is (almost) complete, chdir("/")
 541      * to free any busy filesystems */
 542     if (chdir("/") < 0) {
 543         err(EXIT_FAILURE, "Could not chdir to root directory");
 544     }
 545
 546     do {
 547         main_loop_wait(false);
 548     } while (!sigterm_reported && (persistent || !nbd_started || nb_fds > 0));
 549
 550     nbd_export_close(exp);
 551     if (sockpath) {
 552         unlink(sockpath);
 553     }
 554
 555     if (device) {
 556         void *ret;
 557         pthread_join(client_thread, &ret);
 558         exit(ret != NULL);
 559     } else {
 560         exit(EXIT_SUCCESS);
 561     }
 562 }