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/block.h"
  21 #include "block/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 #define QEMU_NBD_OPT_DISCARD 3
  40
  41 static NBDExport *exp;
  42 static int verbose;
  43 static char *srcpath;
  44 static char *sockpath;
  45 static int persistent = 0;
  46 static enum { RUNNING, TERMINATE, TERMINATING, TERMINATED } state;
  47 static int shared = 1;
  48 static int nb_fds;
  49
  50 static void usage(const char *name)
  51 {
  52     (printf) (
  53 "Usage: %s [OPTIONS] FILE\n"
  54 "QEMU Disk Network Block Device Server\n"
  55 "\n"
  56 "  -h, --help           display this help and exit\n"
  57 "  -V, --version        output version information and exit\n"
  58 "\n"
  59 "Connection properties:\n"
  60 "  -p, --port=PORT      port to listen on (default `%d')\n"
  61 "  -b, --bind=IFACE     interface to bind to (default `0.0.0.0')\n"
  62 "  -k, --socket=PATH    path to the unix socket\n"
  63 "                       (default '"SOCKET_PATH"')\n"
  64 "  -e, --shared=NUM     device can be shared by NUM clients (default '1')\n"
  65 "  -t, --persistent     don't exit on the last connection\n"
  66 "  -v, --verbose        display extra debugging information\n"
  67 "\n"
  68 "Exposing part of the image:\n"
  69 "  -o, --offset=OFFSET  offset into the image\n"
  70 "  -P, --partition=NUM  only expose partition NUM\n"
  71 "\n"
  72 #ifdef __linux__
  73 "Kernel NBD client support:\n"
  74 "  -c, --connect=DEV    connect FILE to the local NBD device DEV\n"
  75 "  -d, --disconnect     disconnect the specified device\n"
  76 "\n"
  77 #endif
  78 "\n"
  79 "Block device options:\n"
  80 "  -r, --read-only      export read-only\n"
  81 "  -s, --snapshot       use snapshot file\n"
  82 "  -n, --nocache        disable host cache\n"
  83 "      --cache=MODE     set cache mode (none, writeback, ...)\n"
  84 #ifdef CONFIG_LINUX_AIO
  85 "      --aio=MODE       set AIO mode (native or threads)\n"
  86 #endif
  87 "\n"
  88 "Report bugs to <qemu-devel@nongnu.org>\n"
  89     , name, NBD_DEFAULT_PORT, "DEVICE");
  90 }
  91
  92 static void version(const char *name)
  93 {
  94     printf(
  95 "%s version 0.0.1\n"
  96 "Written by Anthony Liguori.\n"
  97 "\n"
  98 "Copyright (C) 2006 Anthony Liguori <anthony@codemonkey.ws>.\n"
  99 "This is free software; see the source for copying conditions.  There is NO\n"
 100 "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
 101     , name);
 102 }
 103
 104 struct partition_record
 105 {
 106     uint8_t bootable;
 107     uint8_t start_head;
 108     uint32_t start_cylinder;
 109     uint8_t start_sector;
 110     uint8_t system;
 111     uint8_t end_head;
 112     uint8_t end_cylinder;
 113     uint8_t end_sector;
 114     uint32_t start_sector_abs;
 115     uint32_t nb_sectors_abs;
 116 };
 117
 118 static void read_partition(uint8_t *p, struct partition_record *r)
 119 {
 120     r->bootable = p[0];
 121     r->start_head = p[1];
 122     r->start_cylinder = p[3] | ((p[2] << 2) & 0x0300);
 123     r->start_sector = p[2] & 0x3f;
 124     r->system = p[4];
 125     r->end_head = p[5];
 126     r->end_cylinder = p[7] | ((p[6] << 2) & 0x300);
 127     r->end_sector = p[6] & 0x3f;
 128     r->start_sector_abs = p[8] | p[9] << 8 | p[10] << 16 | p[11] << 24;
 129     r->nb_sectors_abs = p[12] | p[13] << 8 | p[14] << 16 | p[15] << 24;
 130 }
 131
 132 static int find_partition(BlockDriverState *bs, int partition,
 133                           off_t *offset, off_t *size)
 134 {
 135     struct partition_record mbr[4];
 136     uint8_t data[512];
 137     int i;
 138     int ext_partnum = 4;
 139     int ret;
 140
 141     if ((ret = bdrv_read(bs, 0, data, 1)) < 0) {
 142         errno = -ret;
 143         err(EXIT_FAILURE, "error while reading");
 144     }
 145
 146     if (data[510] != 0x55 || data[511] != 0xaa) {
 147         return -EINVAL;
 148     }
 149
 150     for (i = 0; i < 4; i++) {
 151         read_partition(&data[446 + 16 * i], &mbr[i]);
 152
 153         if (!mbr[i].nb_sectors_abs)
 154             continue;
 155
 156         if (mbr[i].system == 0xF || mbr[i].system == 0x5) {
 157             struct partition_record ext[4];
 158             uint8_t data1[512];
 159             int j;
 160
 161             if ((ret = bdrv_read(bs, mbr[i].start_sector_abs, data1, 1)) < 0) {
 162                 errno = -ret;
 163                 err(EXIT_FAILURE, "error while reading");
 164             }
 165
 166             for (j = 0; j < 4; j++) {
 167                 read_partition(&data1[446 + 16 * j], &ext[j]);
 168                 if (!ext[j].nb_sectors_abs)
 169                     continue;
 170
 171                 if ((ext_partnum + j + 1) == partition) {
 172                     *offset = (uint64_t)ext[j].start_sector_abs << 9;
 173                     *size = (uint64_t)ext[j].nb_sectors_abs << 9;
 174                     return 0;
 175                 }
 176             }
 177             ext_partnum += 4;
 178         } else if ((i + 1) == partition) {
 179             *offset = (uint64_t)mbr[i].start_sector_abs << 9;
 180             *size = (uint64_t)mbr[i].nb_sectors_abs << 9;
 181             return 0;
 182         }
 183     }
 184
 185     return -ENOENT;
 186 }
 187
 188 static void termsig_handler(int signum)
 189 {
 190     state = TERMINATE;
 191     qemu_notify_event();
 192 }
 193
 194 static void *show_parts(void *arg)
 195 {
 196     char *device = arg;
 197     int nbd;
 198
 199     /* linux just needs an open() to trigger
 200      * the partition table update
 201      * but remember to load the module with max_part != 0 :
 202      *     modprobe nbd max_part=63
 203      */
 204     nbd = open(device, O_RDWR);
 205     if (nbd >= 0) {
 206         close(nbd);
 207     }
 208     return NULL;
 209 }
 210
 211 static void *nbd_client_thread(void *arg)
 212 {
 213     char *device = arg;
 214     off_t size;
 215     size_t blocksize;
 216     uint32_t nbdflags;
 217     int fd, sock;
 218     int ret;
 219     pthread_t show_parts_thread;
 220
 221     sock = unix_socket_outgoing(sockpath);
 222     if (sock < 0) {
 223         goto out;
 224     }
 225
 226     ret = nbd_receive_negotiate(sock, NULL, &nbdflags,
 227                                 &size, &blocksize);
 228     if (ret < 0) {
 229         goto out;
 230     }
 231
 232     fd = open(device, O_RDWR);
 233     if (fd < 0) {
 234         /* Linux-only, we can use %m in printf.  */
 235         fprintf(stderr, "Failed to open %s: %m", device);
 236         goto out;
 237     }
 238
 239     ret = nbd_init(fd, sock, nbdflags, size, blocksize);
 240     if (ret < 0) {
 241         goto out;
 242     }
 243
 244     /* update partition table */
 245     pthread_create(&show_parts_thread, NULL, show_parts, device);
 246
 247     if (verbose) {
 248         fprintf(stderr, "NBD device %s is now connected to %s\n",
 249                 device, srcpath);
 250     } else {
 251         /* Close stderr so that the qemu-nbd process exits.  */
 252         dup2(STDOUT_FILENO, STDERR_FILENO);
 253     }
 254
 255     ret = nbd_client(fd);
 256     if (ret) {
 257         goto out;
 258     }
 259     close(fd);
 260     kill(getpid(), SIGTERM);
 261     return (void *) EXIT_SUCCESS;
 262
 263 out:
 264     kill(getpid(), SIGTERM);
 265     return (void *) EXIT_FAILURE;
 266 }
 267
 268 static int nbd_can_accept(void *opaque)
 269 {
 270     return nb_fds < shared;
 271 }
 272
 273 static void nbd_export_closed(NBDExport *exp)
 274 {
 275     assert(state == TERMINATING);
 276     state = TERMINATED;
 277 }
 278
 279 static void nbd_client_closed(NBDClient *client)
 280 {
 281     nb_fds--;
 282     if (nb_fds == 0 && !persistent && state == RUNNING) {
 283         state = TERMINATE;
 284     }
 285     qemu_notify_event();
 286     nbd_client_put(client);
 287 }
 288
 289 static void nbd_accept(void *opaque)
 290 {
 291     int server_fd = (uintptr_t) opaque;
 292     struct sockaddr_in addr;
 293     socklen_t addr_len = sizeof(addr);
 294
 295     int fd = accept(server_fd, (struct sockaddr *)&addr, &addr_len);
 296     if (state >= TERMINATE) {
 297         close(fd);
 298         return;
 299     }
 300
 301     if (fd >= 0 && nbd_client_new(exp, fd, nbd_client_closed)) {
 302         nb_fds++;
 303     }
 304 }
 305
 306 int main(int argc, char **argv)
 307 {
 308     BlockDriverState *bs;
 309     BlockDriver *drv;
 310     off_t dev_offset = 0;
 311     uint32_t nbdflags = 0;
 312     bool disconnect = false;
 313     const char *bindto = "0.0.0.0";
 314     char *device = NULL;
 315     int port = NBD_DEFAULT_PORT;
 316     off_t fd_size;
 317     const char *sopt = "hVb:o:p:rsnP:c:dvk:e:f:t";
 318     struct option lopt[] = {
 319         { "help", 0, NULL, 'h' },
 320         { "version", 0, NULL, 'V' },
 321         { "bind", 1, NULL, 'b' },
 322         { "port", 1, NULL, 'p' },
 323         { "socket", 1, NULL, 'k' },
 324         { "offset", 1, NULL, 'o' },
 325         { "read-only", 0, NULL, 'r' },
 326         { "partition", 1, NULL, 'P' },
 327         { "connect", 1, NULL, 'c' },
 328         { "disconnect", 0, NULL, 'd' },
 329         { "snapshot", 0, NULL, 's' },
 330         { "nocache", 0, NULL, 'n' },
 331         { "cache", 1, NULL, QEMU_NBD_OPT_CACHE },
 332 #ifdef CONFIG_LINUX_AIO
 333         { "aio", 1, NULL, QEMU_NBD_OPT_AIO },
 334 #endif
 335         { "discard", 1, NULL, QEMU_NBD_OPT_DISCARD },
 336         { "shared", 1, NULL, 'e' },
 337         { "format", 1, NULL, 'f' },
 338         { "persistent", 0, NULL, 't' },
 339         { "verbose", 0, NULL, 'v' },
 340         { NULL, 0, NULL, 0 }
 341     };
 342     int ch;
 343     int opt_ind = 0;
 344     int li;
 345     char *end;
 346     int flags = BDRV_O_RDWR;
 347     int partition = -1;
 348     int ret;
 349     int fd;
 350     bool seen_cache = false;
 351     bool seen_discard = false;
 352 #ifdef CONFIG_LINUX_AIO
 353     bool seen_aio = false;
 354 #endif
 355     pthread_t client_thread;
 356     const char *fmt = NULL;
 357
 358     /* The client thread uses SIGTERM to interrupt the server.  A signal
 359      * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
 360      */
 361     struct sigaction sa_sigterm;
 362     memset(&sa_sigterm, 0, sizeof(sa_sigterm));
 363     sa_sigterm.sa_handler = termsig_handler;
 364     sigaction(SIGTERM, &sa_sigterm, NULL);
 365
 366     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 367         switch (ch) {
 368         case 's':
 369             flags |= BDRV_O_SNAPSHOT;
 370             break;
 371         case 'n':
 372             optarg = (char *) "none";
 373             /* fallthrough */
 374         case QEMU_NBD_OPT_CACHE:
 375             if (seen_cache) {
 376                 errx(EXIT_FAILURE, "-n and --cache can only be specified once");
 377             }
 378             seen_cache = true;
 379             if (bdrv_parse_cache_flags(optarg, &flags) == -1) {
 380                 errx(EXIT_FAILURE, "Invalid cache mode `%s'", optarg);
 381             }
 382             break;
 383 #ifdef CONFIG_LINUX_AIO
 384         case QEMU_NBD_OPT_AIO:
 385             if (seen_aio) {
 386                 errx(EXIT_FAILURE, "--aio can only be specified once");
 387             }
 388             seen_aio = true;
 389             if (!strcmp(optarg, "native")) {
 390                 flags |= BDRV_O_NATIVE_AIO;
 391             } else if (!strcmp(optarg, "threads")) {
 392                 /* this is the default */
 393             } else {
 394                errx(EXIT_FAILURE, "invalid aio mode `%s'", optarg);
 395             }
 396             break;
 397 #endif
 398         case QEMU_NBD_OPT_DISCARD:
 399             if (seen_discard) {
 400                 errx(EXIT_FAILURE, "--discard can only be specified once");
 401             }
 402             seen_discard = true;
 403             if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
 404                 errx(EXIT_FAILURE, "Invalid discard mode `%s'", optarg);
 405             }
 406             break;
 407         case 'b':
 408             bindto = optarg;
 409             break;
 410         case 'p':
 411             li = strtol(optarg, &end, 0);
 412             if (*end) {
 413                 errx(EXIT_FAILURE, "Invalid port `%s'", optarg);
 414             }
 415             if (li < 1 || li > 65535) {
 416                 errx(EXIT_FAILURE, "Port out of range `%s'", optarg);
 417             }
 418             port = (uint16_t)li;
 419             break;
 420         case 'o':
 421                 dev_offset = strtoll (optarg, &end, 0);
 422             if (*end) {
 423                 errx(EXIT_FAILURE, "Invalid offset `%s'", optarg);
 424             }
 425             if (dev_offset < 0) {
 426                 errx(EXIT_FAILURE, "Offset must be positive `%s'", optarg);
 427             }
 428             break;
 429         case 'r':
 430             nbdflags |= NBD_FLAG_READ_ONLY;
 431             flags &= ~BDRV_O_RDWR;
 432             break;
 433         case 'P':
 434             partition = strtol(optarg, &end, 0);
 435             if (*end)
 436                 errx(EXIT_FAILURE, "Invalid partition `%s'", optarg);
 437             if (partition < 1 || partition > 8)
 438                 errx(EXIT_FAILURE, "Invalid partition %d", partition);
 439             break;
 440         case 'k':
 441             sockpath = optarg;
 442             if (sockpath[0] != '/')
 443                 errx(EXIT_FAILURE, "socket path must be absolute\n");
 444             break;
 445         case 'd':
 446             disconnect = true;
 447             break;
 448         case 'c':
 449             device = optarg;
 450             break;
 451         case 'e':
 452             shared = strtol(optarg, &end, 0);
 453             if (*end) {
 454                 errx(EXIT_FAILURE, "Invalid shared device number '%s'", optarg);
 455             }
 456             if (shared < 1) {
 457                 errx(EXIT_FAILURE, "Shared device number must be greater than 0\n");
 458             }
 459             break;
 460         case 'f':
 461             fmt = optarg;
 462             break;
 463         case 't':
 464             persistent = 1;
 465             break;
 466         case 'v':
 467             verbose = 1;
 468             break;
 469         case 'V':
 470             version(argv[0]);
 471             exit(0);
 472             break;
 473         case 'h':
 474             usage(argv[0]);
 475             exit(0);
 476             break;
 477         case '?':
 478             errx(EXIT_FAILURE, "Try `%s --help' for more information.",
 479                  argv[0]);
 480         }
 481     }
 482
 483     if ((argc - optind) != 1) {
 484         errx(EXIT_FAILURE, "Invalid number of argument.\n"
 485              "Try `%s --help' for more information.",
 486              argv[0]);
 487     }
 488
 489     if (disconnect) {
 490         fd = open(argv[optind], O_RDWR);
 491         if (fd < 0) {
 492             err(EXIT_FAILURE, "Cannot open %s", argv[optind]);
 493         }
 494         nbd_disconnect(fd);
 495
 496         close(fd);
 497
 498         printf("%s disconnected\n", argv[optind]);
 499
 500         return 0;
 501     }
 502
 503     if (device && !verbose) {
 504         int stderr_fd[2];
 505         pid_t pid;
 506         int ret;
 507
 508         if (qemu_pipe(stderr_fd) < 0) {
 509             err(EXIT_FAILURE, "Error setting up communication pipe");
 510         }
 511
 512         /* Now daemonize, but keep a communication channel open to
 513          * print errors and exit with the proper status code.
 514          */
 515         pid = fork();
 516         if (pid == 0) {
 517             close(stderr_fd[0]);
 518             ret = qemu_daemon(1, 0);
 519
 520             /* Temporarily redirect stderr to the parent's pipe...  */
 521             dup2(stderr_fd[1], STDERR_FILENO);
 522             if (ret < 0) {
 523                 err(EXIT_FAILURE, "Failed to daemonize");
 524             }
 525
 526             /* ... close the descriptor we inherited and go on.  */
 527             close(stderr_fd[1]);
 528         } else {
 529             bool errors = false;
 530             char *buf;
 531
 532             /* In the parent.  Print error messages from the child until
 533              * it closes the pipe.
 534              */
 535             close(stderr_fd[1]);
 536             buf = g_malloc(1024);
 537             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
 538                 errors = true;
 539                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
 540                 if (ret < 0) {
 541                     exit(EXIT_FAILURE);
 542                 }
 543             }
 544             if (ret < 0) {
 545                 err(EXIT_FAILURE, "Cannot read from daemon");
 546             }
 547
 548             /* Usually the daemon should not print any message.
 549              * Exit with zero status in that case.
 550              */
 551             exit(errors);
 552         }
 553     }
 554
 555     if (device != NULL && sockpath == NULL) {
 556         sockpath = g_malloc(128);
 557         snprintf(sockpath, 128, SOCKET_PATH, basename(device));
 558     }
 559
 560     qemu_init_main_loop();
 561     bdrv_init();
 562     atexit(bdrv_close_all);
 563
 564     if (fmt) {
 565         drv = bdrv_find_format(fmt);
 566         if (!drv) {
 567             errx(EXIT_FAILURE, "Unknown file format '%s'", fmt);
 568         }
 569     } else {
 570         drv = NULL;
 571     }
 572
 573     bs = bdrv_new("hda");
 574     srcpath = argv[optind];
 575     ret = bdrv_open(bs, srcpath, NULL, flags, drv);
 576     if (ret < 0) {
 577         errno = -ret;
 578         err(EXIT_FAILURE, "Failed to bdrv_open '%s'", argv[optind]);
 579     }
 580
 581     fd_size = bdrv_getlength(bs);
 582
 583     if (partition != -1) {
 584         ret = find_partition(bs, partition, &dev_offset, &fd_size);
 585         if (ret < 0) {
 586             errno = -ret;
 587             err(EXIT_FAILURE, "Could not find partition %d", partition);
 588         }
 589     }
 590
 591     exp = nbd_export_new(bs, dev_offset, fd_size, nbdflags, nbd_export_closed);
 592
 593     if (sockpath) {
 594         fd = unix_socket_incoming(sockpath);
 595     } else {
 596         fd = tcp_socket_incoming(bindto, port);
 597     }
 598
 599     if (fd < 0) {
 600         return 1;
 601     }
 602
 603     if (device) {
 604         int ret;
 605
 606         ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
 607         if (ret != 0) {
 608             errx(EXIT_FAILURE, "Failed to create client thread: %s",
 609                  strerror(ret));
 610         }
 611     } else {
 612         /* Shut up GCC warnings.  */
 613         memset(&client_thread, 0, sizeof(client_thread));
 614     }
 615
 616     qemu_set_fd_handler2(fd, nbd_can_accept, nbd_accept, NULL,
 617                          (void *)(uintptr_t)fd);
 618
 619     /* now when the initialization is (almost) complete, chdir("/")
 620      * to free any busy filesystems */
 621     if (chdir("/") < 0) {
 622         err(EXIT_FAILURE, "Could not chdir to root directory");
 623     }
 624
 625     state = RUNNING;
 626     do {
 627         main_loop_wait(false);
 628         if (state == TERMINATE) {
 629             state = TERMINATING;
 630             nbd_export_close(exp);
 631             nbd_export_put(exp);
 632             exp = NULL;
 633         }
 634     } while (state != TERMINATED);
 635
 636     bdrv_close(bs);
 637     if (sockpath) {
 638         unlink(sockpath);
 639     }
 640
 641     if (device) {
 642         void *ret;
 643         pthread_join(client_thread, &ret);
 644         exit(ret != NULL);
 645     } else {
 646         exit(EXIT_SUCCESS);
 647     }
 648 }