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