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 int persistent = 0;
  45 static enum { RUNNING, TERMINATE, TERMINATING, TERMINATED } state;
  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     state = TERMINATE;
 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_export_closed(NBDExport *exp)
 273 {
 274     assert(state == TERMINATING);
 275     state = TERMINATED;
 276 }
 277
 278 static void nbd_client_closed(NBDClient *client)
 279 {
 280     nb_fds--;
 281     if (nb_fds == 0 && !persistent && state == RUNNING) {
 282         state = TERMINATE;
 283     }
 284     qemu_notify_event();
 285     nbd_client_put(client);
 286 }
 287
 288 static void nbd_accept(void *opaque)
 289 {
 290     int server_fd = (uintptr_t) opaque;
 291     struct sockaddr_in addr;
 292     socklen_t addr_len = sizeof(addr);
 293
 294     int fd = accept(server_fd, (struct sockaddr *)&addr, &addr_len);
 295     if (state >= TERMINATE) {
 296         close(fd);
 297         return;
 298     }
 299
 300     if (fd >= 0 && nbd_client_new(exp, fd, nbd_client_closed)) {
 301         nb_fds++;
 302     }
 303 }
 304
 305 int main(int argc, char **argv)
 306 {
 307     BlockDriverState *bs;
 308     off_t dev_offset = 0;
 309     uint32_t nbdflags = 0;
 310     bool disconnect = false;
 311     const char *bindto = "0.0.0.0";
 312     char *device = NULL;
 313     int port = NBD_DEFAULT_PORT;
 314     off_t fd_size;
 315     const char *sopt = "hVb:o:p:rsnP:c:dvk:e:t";
 316     struct option lopt[] = {
 317         { "help", 0, NULL, 'h' },
 318         { "version", 0, NULL, 'V' },
 319         { "bind", 1, NULL, 'b' },
 320         { "port", 1, NULL, 'p' },
 321         { "socket", 1, NULL, 'k' },
 322         { "offset", 1, NULL, 'o' },
 323         { "read-only", 0, NULL, 'r' },
 324         { "partition", 1, NULL, 'P' },
 325         { "connect", 1, NULL, 'c' },
 326         { "disconnect", 0, NULL, 'd' },
 327         { "snapshot", 0, NULL, 's' },
 328         { "nocache", 0, NULL, 'n' },
 329         { "cache", 1, NULL, QEMU_NBD_OPT_CACHE },
 330 #ifdef CONFIG_LINUX_AIO
 331         { "aio", 1, NULL, QEMU_NBD_OPT_AIO },
 332 #endif
 333         { "shared", 1, NULL, 'e' },
 334         { "persistent", 0, NULL, 't' },
 335         { "verbose", 0, NULL, 'v' },
 336         { NULL, 0, NULL, 0 }
 337     };
 338     int ch;
 339     int opt_ind = 0;
 340     int li;
 341     char *end;
 342     int flags = BDRV_O_RDWR;
 343     int partition = -1;
 344     int ret;
 345     int fd;
 346     bool seen_cache = false;
 347 #ifdef CONFIG_LINUX_AIO
 348     bool seen_aio = false;
 349 #endif
 350     pthread_t client_thread;
 351
 352     /* The client thread uses SIGTERM to interrupt the server.  A signal
 353      * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
 354      */
 355     struct sigaction sa_sigterm;
 356     memset(&sa_sigterm, 0, sizeof(sa_sigterm));
 357     sa_sigterm.sa_handler = termsig_handler;
 358     sigaction(SIGTERM, &sa_sigterm, NULL);
 359
 360     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 361         switch (ch) {
 362         case 's':
 363             flags |= BDRV_O_SNAPSHOT;
 364             break;
 365         case 'n':
 366             optarg = (char *) "none";
 367             /* fallthrough */
 368         case QEMU_NBD_OPT_CACHE:
 369             if (seen_cache) {
 370                 errx(EXIT_FAILURE, "-n and --cache can only be specified once");
 371             }
 372             seen_cache = true;
 373             if (bdrv_parse_cache_flags(optarg, &flags) == -1) {
 374                 errx(EXIT_FAILURE, "Invalid cache mode `%s'", optarg);
 375             }
 376             break;
 377 #ifdef CONFIG_LINUX_AIO
 378         case QEMU_NBD_OPT_AIO:
 379             if (seen_aio) {
 380                 errx(EXIT_FAILURE, "--aio can only be specified once");
 381             }
 382             seen_aio = true;
 383             if (!strcmp(optarg, "native")) {
 384                 flags |= BDRV_O_NATIVE_AIO;
 385             } else if (!strcmp(optarg, "threads")) {
 386                 /* this is the default */
 387             } else {
 388                errx(EXIT_FAILURE, "invalid aio mode `%s'", optarg);
 389             }
 390             break;
 391 #endif
 392         case 'b':
 393             bindto = optarg;
 394             break;
 395         case 'p':
 396             li = strtol(optarg, &end, 0);
 397             if (*end) {
 398                 errx(EXIT_FAILURE, "Invalid port `%s'", optarg);
 399             }
 400             if (li < 1 || li > 65535) {
 401                 errx(EXIT_FAILURE, "Port out of range `%s'", optarg);
 402             }
 403             port = (uint16_t)li;
 404             break;
 405         case 'o':
 406                 dev_offset = strtoll (optarg, &end, 0);
 407             if (*end) {
 408                 errx(EXIT_FAILURE, "Invalid offset `%s'", optarg);
 409             }
 410             if (dev_offset < 0) {
 411                 errx(EXIT_FAILURE, "Offset must be positive `%s'", optarg);
 412             }
 413             break;
 414         case 'r':
 415             nbdflags |= NBD_FLAG_READ_ONLY;
 416             flags &= ~BDRV_O_RDWR;
 417             break;
 418         case 'P':
 419             partition = strtol(optarg, &end, 0);
 420             if (*end)
 421                 errx(EXIT_FAILURE, "Invalid partition `%s'", optarg);
 422             if (partition < 1 || partition > 8)
 423                 errx(EXIT_FAILURE, "Invalid partition %d", partition);
 424             break;
 425         case 'k':
 426             sockpath = optarg;
 427             if (sockpath[0] != '/')
 428                 errx(EXIT_FAILURE, "socket path must be absolute\n");
 429             break;
 430         case 'd':
 431             disconnect = true;
 432             break;
 433         case 'c':
 434             device = optarg;
 435             break;
 436         case 'e':
 437             shared = strtol(optarg, &end, 0);
 438             if (*end) {
 439                 errx(EXIT_FAILURE, "Invalid shared device number '%s'", optarg);
 440             }
 441             if (shared < 1) {
 442                 errx(EXIT_FAILURE, "Shared device number must be greater than 0\n");
 443             }
 444             break;
 445         case 't':
 446             persistent = 1;
 447             break;
 448         case 'v':
 449             verbose = 1;
 450             break;
 451         case 'V':
 452             version(argv[0]);
 453             exit(0);
 454             break;
 455         case 'h':
 456             usage(argv[0]);
 457             exit(0);
 458             break;
 459         case '?':
 460             errx(EXIT_FAILURE, "Try `%s --help' for more information.",
 461                  argv[0]);
 462         }
 463     }
 464
 465     if ((argc - optind) != 1) {
 466         errx(EXIT_FAILURE, "Invalid number of argument.\n"
 467              "Try `%s --help' for more information.",
 468              argv[0]);
 469     }
 470
 471     if (disconnect) {
 472         fd = open(argv[optind], O_RDWR);
 473         if (fd < 0) {
 474             err(EXIT_FAILURE, "Cannot open %s", argv[optind]);
 475         }
 476         nbd_disconnect(fd);
 477
 478         close(fd);
 479
 480         printf("%s disconnected\n", argv[optind]);
 481
 482         return 0;
 483     }
 484
 485     if (device && !verbose) {
 486         int stderr_fd[2];
 487         pid_t pid;
 488         int ret;
 489
 490         if (qemu_pipe(stderr_fd) < 0) {
 491             err(EXIT_FAILURE, "Error setting up communication pipe");
 492         }
 493
 494         /* Now daemonize, but keep a communication channel open to
 495          * print errors and exit with the proper status code.
 496          */
 497         pid = fork();
 498         if (pid == 0) {
 499             close(stderr_fd[0]);
 500             ret = qemu_daemon(1, 0);
 501
 502             /* Temporarily redirect stderr to the parent's pipe...  */
 503             dup2(stderr_fd[1], STDERR_FILENO);
 504             if (ret < 0) {
 505                 err(EXIT_FAILURE, "Failed to daemonize");
 506             }
 507
 508             /* ... close the descriptor we inherited and go on.  */
 509             close(stderr_fd[1]);
 510         } else {
 511             bool errors = false;
 512             char *buf;
 513
 514             /* In the parent.  Print error messages from the child until
 515              * it closes the pipe.
 516              */
 517             close(stderr_fd[1]);
 518             buf = g_malloc(1024);
 519             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
 520                 errors = true;
 521                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
 522                 if (ret < 0) {
 523                     exit(EXIT_FAILURE);
 524                 }
 525             }
 526             if (ret < 0) {
 527                 err(EXIT_FAILURE, "Cannot read from daemon");
 528             }
 529
 530             /* Usually the daemon should not print any message.
 531              * Exit with zero status in that case.
 532              */
 533             exit(errors);
 534         }
 535     }
 536
 537     if (device != NULL && sockpath == NULL) {
 538         sockpath = g_malloc(128);
 539         snprintf(sockpath, 128, SOCKET_PATH, basename(device));
 540     }
 541
 542     qemu_init_main_loop();
 543     bdrv_init();
 544     atexit(bdrv_close_all);
 545
 546     bs = bdrv_new("hda");
 547     srcpath = argv[optind];
 548     if ((ret = bdrv_open(bs, srcpath, flags, NULL)) < 0) {
 549         errno = -ret;
 550         err(EXIT_FAILURE, "Failed to bdrv_open '%s'", argv[optind]);
 551     }
 552
 553     fd_size = bdrv_getlength(bs);
 554
 555     if (partition != -1) {
 556         ret = find_partition(bs, partition, &dev_offset, &fd_size);
 557         if (ret < 0) {
 558             errno = -ret;
 559             err(EXIT_FAILURE, "Could not find partition %d", partition);
 560         }
 561     }
 562
 563     exp = nbd_export_new(bs, dev_offset, fd_size, nbdflags, nbd_export_closed);
 564
 565     if (sockpath) {
 566         fd = unix_socket_incoming(sockpath);
 567     } else {
 568         fd = tcp_socket_incoming(bindto, port);
 569     }
 570
 571     if (fd < 0) {
 572         return 1;
 573     }
 574
 575     if (device) {
 576         int ret;
 577
 578         ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
 579         if (ret != 0) {
 580             errx(EXIT_FAILURE, "Failed to create client thread: %s",
 581                  strerror(ret));
 582         }
 583     } else {
 584         /* Shut up GCC warnings.  */
 585         memset(&client_thread, 0, sizeof(client_thread));
 586     }
 587
 588     qemu_set_fd_handler2(fd, nbd_can_accept, nbd_accept, NULL,
 589                          (void *)(uintptr_t)fd);
 590
 591     /* now when the initialization is (almost) complete, chdir("/")
 592      * to free any busy filesystems */
 593     if (chdir("/") < 0) {
 594         err(EXIT_FAILURE, "Could not chdir to root directory");
 595     }
 596
 597     state = RUNNING;
 598     do {
 599         main_loop_wait(false);
 600         if (state == TERMINATE) {
 601             state = TERMINATING;
 602             nbd_export_close(exp);
 603             nbd_export_put(exp);
 604             exp = NULL;
 605         }
 606     } while (state != TERMINATED);
 607
 608     bdrv_close(bs);
 609     if (sockpath) {
 610         unlink(sockpath);
 611     }
 612
 613     if (device) {
 614         void *ret;
 615         pthread_join(client_thread, &ret);
 616         exit(ret != NULL);
 617     } else {
 618         exit(EXIT_SUCCESS);
 619     }
 620 }