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