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, write to the Free Software
  17  *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  18  */
  19
  20 #include <qemu-common.h>
  21 #include "block_int.h"
  22 #include "nbd.h"
  23
  24 #include <stdarg.h>
  25 #include <stdio.h>
  26 #include <getopt.h>
  27 #include <err.h>
  28 #include <sys/types.h>
  29 #include <sys/socket.h>
  30 #include <netinet/in.h>
  31 #include <netinet/tcp.h>
  32 #include <arpa/inet.h>
  33 #include <signal.h>
  34
  35 #define SOCKET_PATH    "/var/lock/qemu-nbd-%s"
  36
  37 #define NBD_BUFFER_SIZE (1024*1024)
  38
  39 int verbose;
  40
  41 static void usage(const char *name)
  42 {
  43     printf(
  44 "Usage: %s [OPTIONS] FILE\n"
  45 "QEMU Disk Network Block Device Server\n"
  46 "\n"
  47 "  -p, --port=PORT      port to listen on (default `1024')\n"
  48 "  -o, --offset=OFFSET  offset into the image\n"
  49 "  -b, --bind=IFACE     interface to bind to (default `0.0.0.0')\n"
  50 "  -k, --socket=PATH    path to the unix socket\n"
  51 "                       (default '"SOCKET_PATH"')\n"
  52 "  -r, --read-only      export read-only\n"
  53 "  -P, --partition=NUM  only expose partition NUM\n"
  54 "  -s, --snapshot       use snapshot file\n"
  55 "  -n, --nocache        disable host cache\n"
  56 "  -c, --connect=DEV    connect FILE to the local NBD device DEV\n"
  57 "  -d, --disconnect     disconnect the specified device\n"
  58 "  -v, --verbose        display extra debugging information\n"
  59 "  -h, --help           display this help and exit\n"
  60 "  -V, --version        output version information and exit\n"
  61 "\n"
  62 "Report bugs to <anthony@codemonkey.ws>\n"
  63     , name, "DEVICE");
  64 }
  65
  66 static void version(const char *name)
  67 {
  68     printf(
  69 "qemu-nbd version 0.0.1\n"
  70 "Written by Anthony Liguori.\n"
  71 "\n"
  72 "Copyright (C) 2006 Anthony Liguori <anthony@codemonkey.ws>.\n"
  73 "This is free software; see the source for copying conditions.  There is NO\n"
  74 "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
  75     );
  76 }
  77
  78 struct partition_record
  79 {
  80     uint8_t bootable;
  81     uint8_t start_head;
  82     uint32_t start_cylinder;
  83     uint8_t start_sector;
  84     uint8_t system;
  85     uint8_t end_head;
  86     uint8_t end_cylinder;
  87     uint8_t end_sector;
  88     uint32_t start_sector_abs;
  89     uint32_t nb_sectors_abs;
  90 };
  91
  92 static void read_partition(uint8_t *p, struct partition_record *r)
  93 {
  94     r->bootable = p[0];
  95     r->start_head = p[1];
  96     r->start_cylinder = p[3] | ((p[2] << 2) & 0x0300);
  97     r->start_sector = p[2] & 0x3f;
  98     r->system = p[4];
  99     r->end_head = p[5];
 100     r->end_cylinder = p[7] | ((p[6] << 2) & 0x300);
 101     r->end_sector = p[6] & 0x3f;
 102     r->start_sector_abs = p[8] | p[9] << 8 | p[10] << 16 | p[11] << 24;
 103     r->nb_sectors_abs = p[12] | p[13] << 8 | p[14] << 16 | p[15] << 24;
 104 }
 105
 106 static int find_partition(BlockDriverState *bs, int partition,
 107                           off_t *offset, off_t *size)
 108 {
 109     struct partition_record mbr[4];
 110     uint8_t data[512];
 111     int i;
 112     int ext_partnum = 4;
 113
 114     if (bdrv_read(bs, 0, data, 1))
 115         errx(EINVAL, "error while reading");
 116
 117     if (data[510] != 0x55 || data[511] != 0xaa) {
 118         errno = -EINVAL;
 119         return -1;
 120     }
 121
 122     for (i = 0; i < 4; i++) {
 123         read_partition(&data[446 + 16 * i], &mbr[i]);
 124
 125         if (!mbr[i].nb_sectors_abs)
 126             continue;
 127
 128         if (mbr[i].system == 0xF || mbr[i].system == 0x5) {
 129             struct partition_record ext[4];
 130             uint8_t data1[512];
 131             int j;
 132
 133             if (bdrv_read(bs, mbr[i].start_sector_abs, data1, 1))
 134                 errx(EINVAL, "error while reading");
 135
 136             for (j = 0; j < 4; j++) {
 137                 read_partition(&data1[446 + 16 * j], &ext[j]);
 138                 if (!ext[j].nb_sectors_abs)
 139                     continue;
 140
 141                 if ((ext_partnum + j + 1) == partition) {
 142                     *offset = (uint64_t)ext[j].start_sector_abs << 9;
 143                     *size = (uint64_t)ext[j].nb_sectors_abs << 9;
 144                     return 0;
 145                 }
 146             }
 147             ext_partnum += 4;
 148         } else if ((i + 1) == partition) {
 149             *offset = (uint64_t)mbr[i].start_sector_abs << 9;
 150             *size = (uint64_t)mbr[i].nb_sectors_abs << 9;
 151             return 0;
 152         }
 153     }
 154
 155     errno = -ENOENT;
 156     return -1;
 157 }
 158
 159 static void show_parts(const char *device)
 160 {
 161     if (fork() == 0) {
 162         int nbd;
 163
 164         /* linux just needs an open() to trigger
 165          * the partition table update
 166          * but remember to load the module with max_part != 0 :
 167          *     modprobe nbd max_part=63
 168          */
 169         nbd = open(device, O_RDWR);
 170         if (nbd != -1)
 171               close(nbd);
 172         exit(0);
 173     }
 174 }
 175
 176 int main(int argc, char **argv)
 177 {
 178     BlockDriverState *bs;
 179     off_t dev_offset = 0;
 180     off_t offset = 0;
 181     bool readonly = false;
 182     bool disconnect = false;
 183     const char *bindto = "0.0.0.0";
 184     int port = 1024;
 185     int sock, csock;
 186     struct sockaddr_in addr;
 187     socklen_t addr_len = sizeof(addr);
 188     off_t fd_size;
 189     char *device = NULL;
 190     char *socket = NULL;
 191     char sockpath[128];
 192     const char *sopt = "hVbo:p:rsnP:c:dvk:";
 193     struct option lopt[] = {
 194         { "help", 0, 0, 'h' },
 195         { "version", 0, 0, 'V' },
 196         { "bind", 1, 0, 'b' },
 197         { "port", 1, 0, 'p' },
 198         { "socket", 1, 0, 'k' },
 199         { "offset", 1, 0, 'o' },
 200         { "read-only", 0, 0, 'r' },
 201         { "partition", 1, 0, 'P' },
 202         { "connect", 1, 0, 'c' },
 203         { "disconnect", 0, 0, 'd' },
 204         { "snapshot", 0, 0, 's' },
 205         { "nocache", 0, 0, 'n' },
 206         { "verbose", 0, 0, 'v' },
 207         { NULL, 0, 0, 0 }
 208     };
 209     int ch;
 210     int opt_ind = 0;
 211     int li;
 212     char *end;
 213     int flags = 0;
 214     int partition = -1;
 215     int fd;
 216     int ret;
 217     uint8_t *data;
 218
 219     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
 220         switch (ch) {
 221         case 's':
 222             flags |= BDRV_O_SNAPSHOT;
 223             break;
 224         case 'n':
 225             flags |= BDRV_O_DIRECT;
 226             break;
 227         case 'b':
 228             bindto = optarg;
 229             break;
 230         case 'p':
 231             li = strtol(optarg, &end, 0);
 232             if (*end) {
 233                 errx(EINVAL, "Invalid port `%s'", optarg);
 234             }
 235             if (li < 1 || li > 65535) {
 236                 errx(EINVAL, "Port out of range `%s'", optarg);
 237             }
 238             port = (uint16_t)li;
 239             break;
 240         case 'o':
 241                 dev_offset = strtoll (optarg, &end, 0);
 242             if (*end) {
 243                 errx(EINVAL, "Invalid offset `%s'", optarg);
 244             }
 245             if (dev_offset < 0) {
 246                 errx(EINVAL, "Offset must be positive `%s'", optarg);
 247             }
 248             break;
 249         case 'r':
 250             readonly = true;
 251             break;
 252         case 'P':
 253             partition = strtol(optarg, &end, 0);
 254             if (*end)
 255                 errx(EINVAL, "Invalid partition `%s'", optarg);
 256             if (partition < 1 || partition > 8)
 257                 errx(EINVAL, "Invalid partition %d", partition);
 258             break;
 259         case 'k':
 260             socket = optarg;
 261             if (socket[0] != '/')
 262                 errx(EINVAL, "socket path must be absolute\n");
 263             break;
 264         case 'd':
 265             disconnect = true;
 266             break;
 267         case 'c':
 268             device = optarg;
 269             break;
 270         case 'v':
 271             verbose = 1;
 272             break;
 273         case 'V':
 274             version(argv[0]);
 275             exit(0);
 276             break;
 277         case 'h':
 278             usage(argv[0]);
 279             exit(0);
 280             break;
 281         case '?':
 282             errx(EINVAL, "Try `%s --help' for more information.",
 283                  argv[0]);
 284         }
 285     }
 286
 287     if ((argc - optind) != 1) {
 288         errx(EINVAL, "Invalid number of argument.\n"
 289              "Try `%s --help' for more information.",
 290              argv[0]);
 291     }
 292
 293     if (disconnect) {
 294         fd = open(argv[optind], O_RDWR);
 295         if (fd == -1)
 296             errx(errno, "Cannot open %s", argv[optind]);
 297
 298         nbd_disconnect(fd);
 299
 300         close(fd);
 301
 302         printf("%s disconnected\n", argv[optind]);
 303
 304         return 0;
 305     }
 306
 307     bdrv_init();
 308
 309     bs = bdrv_new("hda");
 310     if (bs == NULL)
 311         return 1;
 312
 313     if (bdrv_open(bs, argv[optind], flags) == -1)
 314         return 1;
 315
 316     fd_size = bs->total_sectors * 512;
 317
 318     if (partition != -1 &&
 319         find_partition(bs, partition, &dev_offset, &fd_size))
 320         errx(errno, "Could not find partition %d", partition);
 321
 322     if (device) {
 323         pid_t pid;
 324         if (!verbose)
 325             daemon(0, 0);       /* detach client and server */
 326
 327         if (socket == NULL) {
 328             sprintf(sockpath, SOCKET_PATH, basename(device));
 329             socket = sockpath;
 330         }
 331
 332         pid = fork();
 333         if (pid < 0)
 334             return 1;
 335         if (pid != 0) {
 336             off_t size;
 337             size_t blocksize;
 338
 339             ret = 0;
 340             bdrv_close(bs);
 341
 342             do {
 343                 sock = unix_socket_outgoing(socket);
 344                 if (sock == -1) {
 345                     if (errno != ENOENT && errno != ECONNREFUSED)
 346                         goto out;
 347                     sleep(1);   /* wait children */
 348                 }
 349             } while (sock == -1);
 350
 351             fd = open(device, O_RDWR);
 352             if (fd == -1) {
 353                 ret = 1;
 354                 goto out;
 355             }
 356
 357             ret = nbd_receive_negotiate(sock, &size, &blocksize);
 358             if (ret == -1) {
 359                 ret = 1;
 360                 goto out;
 361             }
 362
 363             ret = nbd_init(fd, sock, size, blocksize);
 364             if (ret == -1) {
 365                 ret = 1;
 366                 goto out;
 367             }
 368
 369             printf("NBD device %s is now connected to file %s\n",
 370                     device, argv[optind]);
 371
 372             /* update partition table */
 373
 374             show_parts(device);
 375
 376             nbd_client(fd, sock);
 377             close(fd);
 378  out:
 379             kill(pid, SIGTERM);
 380             unlink(socket);
 381
 382             return ret;
 383         }
 384         /* children */
 385     }
 386
 387     if (socket) {
 388         sock = unix_socket_incoming(socket);
 389     } else {
 390         sock = tcp_socket_incoming(bindto, port);
 391     }
 392
 393     if (sock == -1)
 394         return 1;
 395
 396     csock = accept(sock,
 397                (struct sockaddr *)&addr,
 398                &addr_len);
 399     if (csock == -1)
 400         return 1;
 401
 402     /* new fd_size is calculated by find_partition */
 403     if (nbd_negotiate(bs, csock, fd_size) == -1)
 404         return 1;
 405
 406     data = qemu_memalign(512, NBD_BUFFER_SIZE);
 407     while (nbd_trip(bs, csock, fd_size, dev_offset, &offset, readonly,
 408                     data, NBD_BUFFER_SIZE) == 0);
 409     qemu_free(data);
 410
 411     close(csock);
 412     close(sock);
 413     bdrv_close(bs);
 414     if (socket)
 415         unlink(socket);
 416
 417     return 0;
 418 }