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 #define QEMU_NBD_OPT_CACHE  1
  38 #define QEMU_NBD_OPT_AIO    2
  39
  40 static NBDExport *exp;
  41 static int verbose;
  42 static char *srcpath;
  43 static char *sockpath;
  44 static bool sigterm_reported;
  45 static bool nbd_started;
  46 static int shared = 1;
  47 static int nb_fds;
  48
  49 static void usage(const char *name)
  50 {
  51     (printf) (
  52 "Usage: %s [OPTIONS] FILE\n"
  53 "QEMU Disk Network Block Device Server\n"
  54 "\n"
  55 "  -h, --help           display this help and exit\n"
  56 "  -V, --version        output version information and exit\n"
  57 "\n"
  58 "Connection properties:\n"
  59 "  -p, --port=PORT      port to listen on (default `%d')\n"
  60 "  -b, --bind=IFACE     interface to bind to (default `0.0.0.0')\n"
  61 "  -k, --socket=PATH    path to the unix socket\n"
  62 "                       (default '"SOCKET_PATH"')\n"
  63 "  -e, --shared=NUM     device can be shared by NUM clients (default '1')\n"
  64 "  -t, --persistent     don't exit on the last connection\n"
  65 "  -v, --verbose        display extra debugging information\n"
  66 "\n"
  67 "Exposing part of the image:\n"
  68 "  -o, --offset=OFFSET  offset into the image\n"
  69 "  -P, --partition=NUM  only expose partition NUM\n"
  70 "\n"
  71 #ifdef __linux__
  72 "Kernel NBD client support:\n"
  73 "  -c, --connect=DEV    connect FILE to the local NBD device DEV\n"
  74 "  -d, --disconnect     disconnect the specified device\n"
  75 "\n"
  76 #endif
  77 "\n"
  78 "Block device options:\n"
  79 "  -r, --read-only      export read-only\n"
  80 "  -s, --snapshot       use snapshot file\n"
  81 "  -n, --nocache        disable host cache\n"
  82 "      --cache=MODE     set cache mode (none, writeback, ...)\n"
  83 #ifdef CONFIG_LINUX_AIO
  84 "      --aio=MODE       set AIO mode (native or threads)\n"
  85 #endif
  86 "\n"
  87 "Report bugs to <qemu-devel@nongnu.org>\n"
  88     , name, NBD_DEFAULT_PORT, "DEVICE");
  89 }
  90
  91 static void version(const char *name)
  92 {
  93     printf(
  94 "%s version 0.0.1\n"
  95 "Written by Anthony Liguori.\n"
  96 "\n"
  97 "Copyright (C) 2006 Anthony Liguori <anthony@codemonkey.ws>.\n"
  98 "This is free software; see the source for copying conditions.  There is NO\n"
  99 "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
 100     , name);
 101 }
 102
 103 struct partition_record
 104 {
 105     uint8_t bootable;
 106     uint8_t start_head;
 107     uint32_t start_cylinder;
 108     uint8_t start_sector;
 109     uint8_t system;
 110     uint8_t end_head;
 111     uint8_t end_cylinder;
 112     uint8_t end_sector;
 113     uint32_t start_sector_abs;
 114     uint32_t nb_sectors_abs;
 115 };
 116
 117 static void read_partition(uint8_t *p, struct partition_record *r)
 118 {
 119     r->bootable = p[0];
 120     r->start_head = p[1];
 121     r->start_cylinder = p[3] | ((p[2] << 2) & 0x0300);
 122     r->start_sector = p[2] & 0x3f;
 123     r->system = p[4];
 124     r->end_head = p[5];
 125     r->end_cylinder = p[7] | ((p[6] << 2) & 0x300);
 126     r->end_sector = p[6] & 0x3f;
 127     r->start_sector_abs = p[8] | p[9] << 8 | p[10] << 16 | p[11] << 24;
 128     r->nb_sectors_abs = p[12] | p[13] << 8 | p[14] << 16 | p[15] << 24;
 129 }
 130
 131 static int find_partition(BlockDriverState *bs, int partition,
 132                           off_t *offset, off_t *size)
 133 {
 134     struct partition_record mbr[4];
 135     uint8_t data[512];
 136     int i;
 137     int ext_partnum = 4;
 138     int ret;
 139
 140     if ((ret = bdrv_read(bs, 0, data, 1)) < 0) {
 141         errno = -ret;
 142         err(EXIT_FAILURE, "error while reading");
 143     }
 144
 145     if (data[510] != 0x55 || data[511] != 0xaa) {
 146         return -EINVAL;
 147     }
 148
 149     for (i = 0; i < 4; i++) {
 150         read_partition(&data[446 + 16 * i], &mbr[i]);
 151
 152         if (!mbr[i].nb_sectors_abs)
 153             continue;
 154
 155         if (mbr[i].system == 0xF || mbr[i].system == 0x5) {
 156             struct partition_record ext[4];
 157             uint8_t data1[512];
 158             int j;
 159
 160             if ((ret = bdrv_read(bs, mbr[i].start_sector_abs, data1, 1)) < 0) {
 161                 errno = -ret;
 162                 err(EXIT_FAILURE, "error while reading");
 163             }
 164
 165             for (j = 0; j < 4; j++) {
 166                 read_partition(&data1[446 + 16 * j], &ext[j]);
 167                 if (!ext[j].nb_sectors_abs)
 168                     continue;
 169
 170                 if ((ext_partnum + j + 1) == partition) {
 171                     *offset = (uint64_t)ext[j].start_sector_abs << 9;
 172                     *size = (uint64_t)ext[j].nb_sectors_abs << 9;
 173                     return 0;
 174                 }
 175             }
 176             ext_partnum += 4;
 177         } else if ((i + 1) == partition) {
 178             *offset = (uint64_t)mbr[i].start_sector_abs << 9;
 179             *size = (uint64_t)mbr[i].nb_sectors_abs << 9;
 180             return 0;
 181         }
 182     }
 183
 184     return -ENOENT;
 185 }
 186
 187 static void termsig_handler(int signum)
 188 {
 189     sigterm_reported = true;
 190     qemu_notify_event();
 191 }
 192
 193 static void *show_parts(void *arg)
 194 {
 195     char *device = arg;
 196     int nbd;
 197
 198     /* linux just needs an open() to trigger
 199      * the partition table update
 200      * but remember to load the module with max_part != 0 :
 201      *     modprobe nbd max_part=63
 202      */
 203     nbd = open(device, O_RDWR);
 204     if (nbd >= 0) {
 205         close(nbd);
 206     }
 207     return NULL;
 208 }
 209
 210 static void *nbd_client_thread(void *arg)
 211 {
 212     char *device = arg;
 213     off_t size;
 214     size_t blocksize;
 215     uint32_t nbdflags;
 216     int fd, sock;
 217     int ret;
 218     pthread_t show_parts_thread;
 219
 220     sock = unix_socket_outgoing(sockpath);
 221     if (sock < 0) {
 222         goto out;
 223     }
 224
 225     ret = nbd_receive_negotiate(sock, NULL, &nbdflags,
 226                                 &size, &blocksize);
 227     if (ret < 0) {
 228         goto out;
 229     }
 230
 231     fd = open(device, O_RDWR);
 232     if (fd < 0) {
 233         /* Linux-only, we can use %m in printf.  */
 234         fprintf(stderr, "Failed to open %s: %m", device);
 235         goto out;
 236     }
 237
 238     ret = nbd_init(fd, sock, nbdflags, size, blocksize);
 239     if (ret < 0) {
 240         goto out;
 241     }
 242
 243     /* update partition table */
 244     pthread_create(&show_parts_thread, NULL, show_parts, device);
 245
 246     if (verbose) {
 247         fprintf(stderr, "NBD device %s is now connected to %s\n",
 248                 device, srcpath);
 249     } else {
 250         /* Close stderr so that the qemu-nbd process exits.  */
 251         dup2(STDOUT_FILENO, STDERR_FILENO);
 252     }
 253
 254     ret = nbd_client(fd);
 255     if (ret) {
 256         goto out;
 257     }
 258     close(fd);
 259     kill(getpid(), SIGTERM);
 260     return (void *) EXIT_SUCCESS;
 261
 262 out:
 263     kill(getpid(), SIGTERM);
 264     return (void *) EXIT_FAILURE;
 265 }
 266
 267 static int nbd_can_accept(void *opaque)
 268 {
 269     return nb_fds < shared;
 270 }
 271
 272 static void nbd_client_closed(NBDClient *client)
 273 {
 274     nb_fds--;
 275     qemu_notify_event();
 276 }
 277
 278 static void nbd_accept(void *opaque)
 279 {
 280     int server_fd = (uintptr_t) opaque;
 281     struct sockaddr_in addr;
 282     socklen_t addr_len = sizeof(addr);
 283
 284     int fd = accept(server_fd, (struct sockaddr *)&addr, &addr_len);
 285     nbd_started = true;
 286     if (fd >= 0 && nbd_client_new(exp, fd, nbd_client_closed)) {
 287         nb_fds++;
 288     }
 289 }
 290
 291 int main(int argc, char **argv)
 292 {
 293     BlockDriverState *bs;
 294     off_t dev_offset = 0;
 295     uint32_t nbdflags = 0;
 296     bool disconnect = false;
 297     const char *bindto = "0.0.0.0";
 298     char *device = NULL;
 299     int port = NBD_DEFAULT_PORT;
 300     off_t fd_size;
 301     const char *sopt = "hVb:o:p:rsnP:c:dvk:e:t";
 302     struct option lopt[] = {
 303         { "help", 0, NULL, 'h' },
 304         { "version", 0, NULL, 'V' },
 305         { "bind", 1, NULL, 'b' },
 306         { "port", 1, NULL, 'p' },
 307         { "socket", 1, NULL, 'k' },
 308         { "offset", 1, NULL, 'o' },
 309         { "read-only", 0, NULL, 'r' },
 310         { "partition", 1, NULL, 'P' },
 311         { "connect", 1, NULL, 'c' },
 312         { "disconnect", 0, NULL, 'd' },
 313         { "snapshot", 0, NULL, 's' },
 314         { "nocache", 0, NULL, 'n' },
 315         { "cache", 1, NULL, QEMU_NBD_OPT_CACHE },
 316 #ifdef CONFIG_LINUX_AIO
 317         { "aio", 1, NULL, QEMU_NBD_OPT_AIO },
 318 #endif
 319         { "shared", 1, NULL, 'e' },
 320         { "persistent", 0, NULL, 't' },
 321         { "verbose", 0, NULL, 'v' },
 322         { NULL, 0, NULL, 0 }
 323     };
 324     int ch;
 325     int opt_ind = 0;
 326     int li;
 327     char *end;
 328     int flags = BDRV_O_RDWR;
 329     int partition = -1;
 330     int ret;
 331     int fd;
 332     int persistent = 0;
 333     bool seen_cache = false;
 334 #ifdef CONFIG_LINUX_AIO
 335     bool seen_aio = false;
 336 #endif
 337     pthread_t client_thread;
 338
 339     /* The client thread uses SIGTERM to interrupt the server.  A signal
 340      * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
 341      */
 342     struct sigaction sa_sigterm;
 343     memset(&sa_sigterm, 0, sizeof(sa_sigterm));
 344     sa_sigterm.sa_handler = termsig_handler;
 345     sigaction(SIGTERM, &sa_sigterm, NULL);
 346
 347     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 348         switch (ch) {
 349         case 's':
 350             flags |= BDRV_O_SNAPSHOT;
 351             break;
 352         case 'n':
 353             optarg = (char *) "none";
 354             /* fallthrough */
 355         case QEMU_NBD_OPT_CACHE:
 356             if (seen_cache) {
 357                 errx(EXIT_FAILURE, "-n and --cache can only be specified once");
 358             }
 359             seen_cache = true;
 360             if (bdrv_parse_cache_flags(optarg, &flags) == -1) {
 361                 errx(EXIT_FAILURE, "Invalid cache mode `%s'", optarg);
 362             }
 363             break;
 364 #ifdef CONFIG_LINUX_AIO
 365         case QEMU_NBD_OPT_AIO:
 366             if (seen_aio) {
 367                 errx(EXIT_FAILURE, "--aio can only be specified once");
 368             }
 369             seen_aio = true;
 370             if (!strcmp(optarg, "native")) {
 371                 flags |= BDRV_O_NATIVE_AIO;
 372             } else if (!strcmp(optarg, "threads")) {
 373                 /* this is the default */
 374             } else {
 375                errx(EXIT_FAILURE, "invalid aio mode `%s'", optarg);
 376             }
 377             break;
 378 #endif
 379         case 'b':
 380             bindto = optarg;
 381             break;
 382         case 'p':
 383             li = strtol(optarg, &end, 0);
 384             if (*end) {
 385                 errx(EXIT_FAILURE, "Invalid port `%s'", optarg);
 386             }
 387             if (li < 1 || li > 65535) {
 388                 errx(EXIT_FAILURE, "Port out of range `%s'", optarg);
 389             }
 390             port = (uint16_t)li;
 391             break;
 392         case 'o':
 393                 dev_offset = strtoll (optarg, &end, 0);
 394             if (*end) {
 395                 errx(EXIT_FAILURE, "Invalid offset `%s'", optarg);
 396             }
 397             if (dev_offset < 0) {
 398                 errx(EXIT_FAILURE, "Offset must be positive `%s'", optarg);
 399             }
 400             break;
 401         case 'r':
 402             nbdflags |= NBD_FLAG_READ_ONLY;
 403             flags &= ~BDRV_O_RDWR;
 404             break;
 405         case 'P':
 406             partition = strtol(optarg, &end, 0);
 407             if (*end)
 408                 errx(EXIT_FAILURE, "Invalid partition `%s'", optarg);
 409             if (partition < 1 || partition > 8)
 410                 errx(EXIT_FAILURE, "Invalid partition %d", partition);
 411             break;
 412         case 'k':
 413             sockpath = optarg;
 414             if (sockpath[0] != '/')
 415                 errx(EXIT_FAILURE, "socket path must be absolute\n");
 416             break;
 417         case 'd':
 418             disconnect = true;
 419             break;
 420         case 'c':
 421             device = optarg;
 422             break;
 423         case 'e':
 424             shared = strtol(optarg, &end, 0);
 425             if (*end) {
 426                 errx(EXIT_FAILURE, "Invalid shared device number '%s'", optarg);
 427             }
 428             if (shared < 1) {
 429                 errx(EXIT_FAILURE, "Shared device number must be greater than 0\n");
 430             }
 431             break;
 432         case 't':
 433             persistent = 1;
 434             break;
 435         case 'v':
 436             verbose = 1;
 437             break;
 438         case 'V':
 439             version(argv[0]);
 440             exit(0);
 441             break;
 442         case 'h':
 443             usage(argv[0]);
 444             exit(0);
 445             break;
 446         case '?':
 447             errx(EXIT_FAILURE, "Try `%s --help' for more information.",
 448                  argv[0]);
 449         }
 450     }
 451
 452     if ((argc - optind) != 1) {
 453         errx(EXIT_FAILURE, "Invalid number of argument.\n"
 454              "Try `%s --help' for more information.",
 455              argv[0]);
 456     }
 457
 458     if (disconnect) {
 459         fd = open(argv[optind], O_RDWR);
 460         if (fd < 0) {
 461             err(EXIT_FAILURE, "Cannot open %s", argv[optind]);
 462         }
 463         nbd_disconnect(fd);
 464
 465         close(fd);
 466
 467         printf("%s disconnected\n", argv[optind]);
 468
 469         return 0;
 470     }
 471
 472     if (device && !verbose) {
 473         int stderr_fd[2];
 474         pid_t pid;
 475         int ret;
 476
 477         if (qemu_pipe(stderr_fd) < 0) {
 478             err(EXIT_FAILURE, "Error setting up communication pipe");
 479         }
 480
 481         /* Now daemonize, but keep a communication channel open to
 482          * print errors and exit with the proper status code.
 483          */
 484         pid = fork();
 485         if (pid == 0) {
 486             close(stderr_fd[0]);
 487             ret = qemu_daemon(1, 0);
 488
 489             /* Temporarily redirect stderr to the parent's pipe...  */
 490             dup2(stderr_fd[1], STDERR_FILENO);
 491             if (ret < 0) {
 492                 err(EXIT_FAILURE, "Failed to daemonize");
 493             }
 494
 495             /* ... close the descriptor we inherited and go on.  */
 496             close(stderr_fd[1]);
 497         } else {
 498             bool errors = false;
 499             char *buf;
 500
 501             /* In the parent.  Print error messages from the child until
 502              * it closes the pipe.
 503              */
 504             close(stderr_fd[1]);
 505             buf = g_malloc(1024);
 506             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
 507                 errors = true;
 508                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
 509                 if (ret < 0) {
 510                     exit(EXIT_FAILURE);
 511                 }
 512             }
 513             if (ret < 0) {
 514                 err(EXIT_FAILURE, "Cannot read from daemon");
 515             }
 516
 517             /* Usually the daemon should not print any message.
 518              * Exit with zero status in that case.
 519              */
 520             exit(errors);
 521         }
 522     }
 523
 524     if (device != NULL && sockpath == NULL) {
 525         sockpath = g_malloc(128);
 526         snprintf(sockpath, 128, SOCKET_PATH, basename(device));
 527     }
 528
 529     bdrv_init();
 530     atexit(bdrv_close_all);
 531
 532     bs = bdrv_new("hda");
 533     srcpath = argv[optind];
 534     if ((ret = bdrv_open(bs, srcpath, flags, NULL)) < 0) {
 535         errno = -ret;
 536         err(EXIT_FAILURE, "Failed to bdrv_open '%s'", argv[optind]);
 537     }
 538
 539     fd_size = bdrv_getlength(bs);
 540
 541     if (partition != -1) {
 542         ret = find_partition(bs, partition, &dev_offset, &fd_size);
 543         if (ret < 0) {
 544             errno = -ret;
 545             err(EXIT_FAILURE, "Could not find partition %d", partition);
 546         }
 547     }
 548
 549     exp = nbd_export_new(bs, dev_offset, fd_size, nbdflags);
 550
 551     if (sockpath) {
 552         fd = unix_socket_incoming(sockpath);
 553     } else {
 554         fd = tcp_socket_incoming(bindto, port);
 555     }
 556
 557     if (fd < 0) {
 558         return 1;
 559     }
 560
 561     if (device) {
 562         int ret;
 563
 564         ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
 565         if (ret != 0) {
 566             errx(EXIT_FAILURE, "Failed to create client thread: %s",
 567                  strerror(ret));
 568         }
 569     } else {
 570         /* Shut up GCC warnings.  */
 571         memset(&client_thread, 0, sizeof(client_thread));
 572     }
 573
 574     qemu_init_main_loop();
 575     qemu_set_fd_handler2(fd, nbd_can_accept, nbd_accept, NULL,
 576                          (void *)(uintptr_t)fd);
 577
 578     /* now when the initialization is (almost) complete, chdir("/")
 579      * to free any busy filesystems */
 580     if (chdir("/") < 0) {
 581         err(EXIT_FAILURE, "Could not chdir to root directory");
 582     }
 583
 584     do {
 585         main_loop_wait(false);
 586     } while (!sigterm_reported && (persistent || !nbd_started || nb_fds > 0));
 587
 588     nbd_export_close(exp);
 589     if (sockpath) {
 590         unlink(sockpath);
 591     }
 592
 593     if (device) {
 594         void *ret;
 595         pthread_join(client_thread, &ret);
 596         exit(ret != NULL);
 597     } else {
 598         exit(EXIT_SUCCESS);
 599     }
 600 }