block/parallels.c

   1 /*
   2  * Block driver for Parallels disk image format
   3  *
   4  * Copyright (c) 2007 Alex Beregszaszi
   5  * Copyright (c) 2015 Denis V. Lunev <den@openvz.org>
   6  *
   7  * This code was originally based on comparing different disk images created
   8  * by Parallels. Currently it is based on opened OpenVZ sources
   9  * available at
  10  *     http://git.openvz.org/?p=ploop;a=summary
  11  *
  12  * Permission is hereby granted, free of charge, to any person obtaining a copy
  13  * of this software and associated documentation files (the "Software"), to deal
  14  * in the Software without restriction, including without limitation the rights
  15  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  16  * copies of the Software, and to permit persons to whom the Software is
  17  * furnished to do so, subject to the following conditions:
  18  *
  19  * The above copyright notice and this permission notice shall be included in
  20  * all copies or substantial portions of the Software.
  21  *
  22  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  23  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  24  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  25  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  26  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  27  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  28  * THE SOFTWARE.
  29  */
  30 #include "qemu-common.h"
  31 #include "block/block_int.h"
  32 #include "qemu/module.h"
  33
  34 /**************************************************************/
  35
  36 #define HEADER_MAGIC "WithoutFreeSpace"
  37 #define HEADER_MAGIC2 "WithouFreSpacExt"
  38 #define HEADER_VERSION 2
  39
  40 #define DEFAULT_CLUSTER_SIZE 1048576        /* 1 MiB */
  41
  42
  43 // always little-endian
  44 typedef struct ParallelsHeader {
  45     char magic[16]; // "WithoutFreeSpace"
  46     uint32_t version;
  47     uint32_t heads;
  48     uint32_t cylinders;
  49     uint32_t tracks;
  50     uint32_t bat_entries;
  51     uint64_t nb_sectors;
  52     uint32_t inuse;
  53     uint32_t data_off;
  54     char padding[12];
  55 } QEMU_PACKED ParallelsHeader;
  56
  57 typedef struct BDRVParallelsState {
  58     /** Locking is conservative, the lock protects
  59      *   - image file extending (truncate, fallocate)
  60      *   - any access to block allocation table
  61      */
  62     CoMutex lock;
  63
  64     ParallelsHeader *header;
  65     uint32_t header_size;
  66     uint32_t *bat_bitmap;
  67     unsigned int bat_size;
  68
  69     unsigned int tracks;
  70
  71     unsigned int off_multiplier;
  72
  73     bool has_truncate;
  74 } BDRVParallelsState;
  75
  76
  77 static int64_t bat2sect(BDRVParallelsState *s, uint32_t idx)
  78 {
  79     return (uint64_t)le32_to_cpu(s->bat_bitmap[idx]) * s->off_multiplier;
  80 }
  81
  82 static int64_t seek_to_sector(BDRVParallelsState *s, int64_t sector_num)
  83 {
  84     uint32_t index, offset;
  85
  86     index = sector_num / s->tracks;
  87     offset = sector_num % s->tracks;
  88
  89     /* not allocated */
  90     if ((index >= s->bat_size) || (s->bat_bitmap[index] == 0)) {
  91         return -1;
  92     }
  93     return bat2sect(s, index) + offset;
  94 }
  95
  96 static int cluster_remainder(BDRVParallelsState *s, int64_t sector_num,
  97         int nb_sectors)
  98 {
  99     int ret = s->tracks - sector_num % s->tracks;
 100     return MIN(nb_sectors, ret);
 101 }
 102
 103 static int64_t allocate_cluster(BlockDriverState *bs, int64_t sector_num)
 104 {
 105     BDRVParallelsState *s = bs->opaque;
 106     uint32_t idx, offset;
 107     int64_t pos;
 108     int ret;
 109
 110     idx = sector_num / s->tracks;
 111     offset = sector_num % s->tracks;
 112
 113     if (idx >= s->bat_size) {
 114         return -EINVAL;
 115     }
 116     if (s->bat_bitmap[idx] != 0) {
 117         return bat2sect(s, idx) + offset;
 118     }
 119
 120     pos = bdrv_getlength(bs->file) >> BDRV_SECTOR_BITS;
 121     if (s->has_truncate) {
 122         ret = bdrv_truncate(bs->file, (pos + s->tracks) << BDRV_SECTOR_BITS);
 123     } else {
 124         ret = bdrv_write_zeroes(bs->file, pos, s->tracks, 0);
 125     }
 126     if (ret < 0) {
 127         return ret;
 128     }
 129
 130     s->bat_bitmap[idx] = cpu_to_le32(pos / s->off_multiplier);
 131     ret = bdrv_pwrite(bs->file,
 132             sizeof(ParallelsHeader) + idx * sizeof(s->bat_bitmap[idx]),
 133             s->bat_bitmap + idx, sizeof(s->bat_bitmap[idx]));
 134     if (ret < 0) {
 135         s->bat_bitmap[idx] = 0;
 136         return ret;
 137     }
 138     return bat2sect(s, idx) + offset;
 139 }
 140
 141 static int64_t coroutine_fn parallels_co_get_block_status(BlockDriverState *bs,
 142         int64_t sector_num, int nb_sectors, int *pnum)
 143 {
 144     BDRVParallelsState *s = bs->opaque;
 145     int64_t offset;
 146
 147     qemu_co_mutex_lock(&s->lock);
 148     offset = seek_to_sector(s, sector_num);
 149     qemu_co_mutex_unlock(&s->lock);
 150
 151     *pnum = cluster_remainder(s, sector_num, nb_sectors);
 152
 153     if (offset < 0) {
 154         return 0;
 155     }
 156
 157     return (offset << BDRV_SECTOR_BITS) |
 158         BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
 159 }
 160
 161 static coroutine_fn int parallels_co_writev(BlockDriverState *bs,
 162         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
 163 {
 164     BDRVParallelsState *s = bs->opaque;
 165     uint64_t bytes_done = 0;
 166     QEMUIOVector hd_qiov;
 167     int ret = 0;
 168
 169     qemu_iovec_init(&hd_qiov, qiov->niov);
 170
 171     while (nb_sectors > 0) {
 172         int64_t position;
 173         int n, nbytes;
 174
 175         qemu_co_mutex_lock(&s->lock);
 176         position = allocate_cluster(bs, sector_num);
 177         qemu_co_mutex_unlock(&s->lock);
 178         if (position < 0) {
 179             ret = (int)position;
 180             break;
 181         }
 182
 183         n = cluster_remainder(s, sector_num, nb_sectors);
 184         nbytes = n << BDRV_SECTOR_BITS;
 185
 186         qemu_iovec_reset(&hd_qiov);
 187         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
 188
 189         ret = bdrv_co_writev(bs->file, position, n, &hd_qiov);
 190         if (ret < 0) {
 191             break;
 192         }
 193
 194         nb_sectors -= n;
 195         sector_num += n;
 196         bytes_done += nbytes;
 197     }
 198
 199     qemu_iovec_destroy(&hd_qiov);
 200     return ret;
 201 }
 202
 203 static coroutine_fn int parallels_co_readv(BlockDriverState *bs,
 204         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
 205 {
 206     BDRVParallelsState *s = bs->opaque;
 207     uint64_t bytes_done = 0;
 208     QEMUIOVector hd_qiov;
 209     int ret = 0;
 210
 211     qemu_iovec_init(&hd_qiov, qiov->niov);
 212
 213     while (nb_sectors > 0) {
 214         int64_t position;
 215         int n, nbytes;
 216
 217         qemu_co_mutex_lock(&s->lock);
 218         position = seek_to_sector(s, sector_num);
 219         qemu_co_mutex_unlock(&s->lock);
 220
 221         n = cluster_remainder(s, sector_num, nb_sectors);
 222         nbytes = n << BDRV_SECTOR_BITS;
 223
 224         if (position < 0) {
 225             qemu_iovec_memset(qiov, bytes_done, 0, nbytes);
 226         } else {
 227             qemu_iovec_reset(&hd_qiov);
 228             qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
 229
 230             ret = bdrv_co_readv(bs->file, position, n, &hd_qiov);
 231             if (ret < 0) {
 232                 break;
 233             }
 234         }
 235
 236         nb_sectors -= n;
 237         sector_num += n;
 238         bytes_done += nbytes;
 239     }
 240
 241     qemu_iovec_destroy(&hd_qiov);
 242     return ret;
 243 }
 244
 245
 246 static int parallels_check(BlockDriverState *bs, BdrvCheckResult *res,
 247                            BdrvCheckMode fix)
 248 {
 249     BDRVParallelsState *s = bs->opaque;
 250     int64_t size, prev_off, high_off;
 251     int ret;
 252     uint32_t i;
 253     bool flush_bat = false;
 254     int cluster_size = s->tracks << BDRV_SECTOR_BITS;
 255
 256     size = bdrv_getlength(bs->file);
 257     if (size < 0) {
 258         res->check_errors++;
 259         return size;
 260     }
 261
 262     res->bfi.total_clusters = s->bat_size;
 263     res->bfi.compressed_clusters = 0; /* compression is not supported */
 264
 265     high_off = 0;
 266     prev_off = 0;
 267     for (i = 0; i < s->bat_size; i++) {
 268         int64_t off = bat2sect(s, i) << BDRV_SECTOR_BITS;
 269         if (off == 0) {
 270             prev_off = 0;
 271             continue;
 272         }
 273
 274         /* cluster outside the image */
 275         if (off > size) {
 276             fprintf(stderr, "%s cluster %u is outside image\n",
 277                     fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i);
 278             res->corruptions++;
 279             if (fix & BDRV_FIX_ERRORS) {
 280                 prev_off = 0;
 281                 s->bat_bitmap[i] = 0;
 282                 res->corruptions_fixed++;
 283                 flush_bat = true;
 284                 continue;
 285             }
 286         }
 287
 288         res->bfi.allocated_clusters++;
 289         if (off > high_off) {
 290             high_off = off;
 291         }
 292
 293         if (prev_off != 0 && (prev_off + cluster_size) != off) {
 294             res->bfi.fragmented_clusters++;
 295         }
 296         prev_off = off;
 297     }
 298
 299     if (flush_bat) {
 300         ret = bdrv_pwrite_sync(bs->file, 0, s->header, s->header_size);
 301         if (ret < 0) {
 302             res->check_errors++;
 303             return ret;
 304         }
 305     }
 306
 307     res->image_end_offset = high_off + cluster_size;
 308     if (size > res->image_end_offset) {
 309         int64_t count;
 310         count = DIV_ROUND_UP(size - res->image_end_offset, cluster_size);
 311         fprintf(stderr, "%s space leaked at the end of the image %" PRId64 "\n",
 312                 fix & BDRV_FIX_LEAKS ? "Repairing" : "ERROR",
 313                 size - res->image_end_offset);
 314         res->leaks += count;
 315         if (fix & BDRV_FIX_LEAKS) {
 316             ret = bdrv_truncate(bs->file, res->image_end_offset);
 317             if (ret < 0) {
 318                 res->check_errors++;
 319                 return ret;
 320             }
 321             res->leaks_fixed += count;
 322         }
 323     }
 324
 325     return 0;
 326 }
 327
 328
 329 static int parallels_create(const char *filename, QemuOpts *opts, Error **errp)
 330 {
 331     int64_t total_size, cl_size;
 332     uint8_t tmp[BDRV_SECTOR_SIZE];
 333     Error *local_err = NULL;
 334     BlockDriverState *file;
 335     uint32_t bat_entries, bat_sectors;
 336     ParallelsHeader header;
 337     int ret;
 338
 339     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
 340                           BDRV_SECTOR_SIZE);
 341     cl_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
 342                           DEFAULT_CLUSTER_SIZE), BDRV_SECTOR_SIZE);
 343
 344     ret = bdrv_create_file(filename, opts, &local_err);
 345     if (ret < 0) {
 346         error_propagate(errp, local_err);
 347         return ret;
 348     }
 349
 350     file = NULL;
 351     ret = bdrv_open(&file, filename, NULL, NULL,
 352                     BDRV_O_RDWR | BDRV_O_PROTOCOL, NULL, &local_err);
 353     if (ret < 0) {
 354         error_propagate(errp, local_err);
 355         return ret;
 356     }
 357     ret = bdrv_truncate(file, 0);
 358     if (ret < 0) {
 359         goto exit;
 360     }
 361
 362     bat_entries = DIV_ROUND_UP(total_size, cl_size);
 363     bat_sectors = DIV_ROUND_UP(bat_entries * sizeof(uint32_t) +
 364                                sizeof(ParallelsHeader), cl_size);
 365     bat_sectors = (bat_sectors *  cl_size) >> BDRV_SECTOR_BITS;
 366
 367     memset(&header, 0, sizeof(header));
 368     memcpy(header.magic, HEADER_MAGIC2, sizeof(header.magic));
 369     header.version = cpu_to_le32(HEADER_VERSION);
 370     /* don't care much about geometry, it is not used on image level */
 371     header.heads = cpu_to_le32(16);
 372     header.cylinders = cpu_to_le32(total_size / BDRV_SECTOR_SIZE / 16 / 32);
 373     header.tracks = cpu_to_le32(cl_size >> BDRV_SECTOR_BITS);
 374     header.bat_entries = cpu_to_le32(bat_entries);
 375     header.nb_sectors = cpu_to_le64(DIV_ROUND_UP(total_size, BDRV_SECTOR_SIZE));
 376     header.data_off = cpu_to_le32(bat_sectors);
 377
 378     /* write all the data */
 379     memset(tmp, 0, sizeof(tmp));
 380     memcpy(tmp, &header, sizeof(header));
 381
 382     ret = bdrv_pwrite(file, 0, tmp, BDRV_SECTOR_SIZE);
 383     if (ret < 0) {
 384         goto exit;
 385     }
 386     ret = bdrv_write_zeroes(file, 1, bat_sectors - 1, 0);
 387     if (ret < 0) {
 388         goto exit;
 389     }
 390     ret = 0;
 391
 392 done:
 393     bdrv_unref(file);
 394     return ret;
 395
 396 exit:
 397     error_setg_errno(errp, -ret, "Failed to create Parallels image");
 398     goto done;
 399 }
 400
 401
 402 static int parallels_probe(const uint8_t *buf, int buf_size,
 403                            const char *filename)
 404 {
 405     const ParallelsHeader *ph = (const void *)buf;
 406
 407     if (buf_size < sizeof(ParallelsHeader)) {
 408         return 0;
 409     }
 410
 411     if ((!memcmp(ph->magic, HEADER_MAGIC, 16) ||
 412            !memcmp(ph->magic, HEADER_MAGIC2, 16)) &&
 413            (le32_to_cpu(ph->version) == HEADER_VERSION)) {
 414         return 100;
 415     }
 416
 417     return 0;
 418 }
 419
 420 static int parallels_open(BlockDriverState *bs, QDict *options, int flags,
 421                           Error **errp)
 422 {
 423     BDRVParallelsState *s = bs->opaque;
 424     ParallelsHeader ph;
 425     int ret, size;
 426
 427     ret = bdrv_pread(bs->file, 0, &ph, sizeof(ph));
 428     if (ret < 0) {
 429         goto fail;
 430     }
 431
 432     bs->total_sectors = le64_to_cpu(ph.nb_sectors);
 433
 434     if (le32_to_cpu(ph.version) != HEADER_VERSION) {
 435         goto fail_format;
 436     }
 437     if (!memcmp(ph.magic, HEADER_MAGIC, 16)) {
 438         s->off_multiplier = 1;
 439         bs->total_sectors = 0xffffffff & bs->total_sectors;
 440     } else if (!memcmp(ph.magic, HEADER_MAGIC2, 16)) {
 441         s->off_multiplier = le32_to_cpu(ph.tracks);
 442     } else {
 443         goto fail_format;
 444     }
 445
 446     s->tracks = le32_to_cpu(ph.tracks);
 447     if (s->tracks == 0) {
 448         error_setg(errp, "Invalid image: Zero sectors per track");
 449         ret = -EINVAL;
 450         goto fail;
 451     }
 452     if (s->tracks > INT32_MAX/513) {
 453         error_setg(errp, "Invalid image: Too big cluster");
 454         ret = -EFBIG;
 455         goto fail;
 456     }
 457
 458     s->bat_size = le32_to_cpu(ph.bat_entries);
 459     if (s->bat_size > INT_MAX / sizeof(uint32_t)) {
 460         error_setg(errp, "Catalog too large");
 461         ret = -EFBIG;
 462         goto fail;
 463     }
 464
 465     size = sizeof(ParallelsHeader) + sizeof(uint32_t) * s->bat_size;
 466     s->header_size = ROUND_UP(size, bdrv_opt_mem_align(bs->file));
 467     s->header = qemu_try_blockalign(bs->file, s->header_size);
 468     if (s->header == NULL) {
 469         ret = -ENOMEM;
 470         goto fail;
 471     }
 472     if (le32_to_cpu(ph.data_off) < s->header_size) {
 473         /* there is not enough unused space to fit to block align between BAT
 474            and actual data. We can't avoid read-modify-write... */
 475         s->header_size = size;
 476     }
 477
 478     ret = bdrv_pread(bs->file, 0, s->header, s->header_size);
 479     if (ret < 0) {
 480         goto fail;
 481     }
 482     s->bat_bitmap = (uint32_t *)(s->header + 1);
 483
 484     s->has_truncate = bdrv_has_zero_init(bs->file) &&
 485                       bdrv_truncate(bs->file, bdrv_getlength(bs->file)) == 0;
 486
 487     qemu_co_mutex_init(&s->lock);
 488     return 0;
 489
 490 fail_format:
 491     error_setg(errp, "Image not in Parallels format");
 492     ret = -EINVAL;
 493 fail:
 494     qemu_vfree(s->header);
 495     return ret;
 496 }
 497
 498
 499 static void parallels_close(BlockDriverState *bs)
 500 {
 501     BDRVParallelsState *s = bs->opaque;
 502     qemu_vfree(s->header);
 503 }
 504
 505 static QemuOptsList parallels_create_opts = {
 506     .name = "parallels-create-opts",
 507     .head = QTAILQ_HEAD_INITIALIZER(parallels_create_opts.head),
 508     .desc = {
 509         {
 510             .name = BLOCK_OPT_SIZE,
 511             .type = QEMU_OPT_SIZE,
 512             .help = "Virtual disk size",
 513         },
 514         {
 515             .name = BLOCK_OPT_CLUSTER_SIZE,
 516             .type = QEMU_OPT_SIZE,
 517             .help = "Parallels image cluster size",
 518             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE),
 519         },
 520         { /* end of list */ }
 521     }
 522 };
 523
 524 static BlockDriver bdrv_parallels = {
 525     .format_name        = "parallels",
 526     .instance_size      = sizeof(BDRVParallelsState),
 527     .bdrv_probe         = parallels_probe,
 528     .bdrv_open          = parallels_open,
 529     .bdrv_close         = parallels_close,
 530     .bdrv_co_get_block_status = parallels_co_get_block_status,
 531     .bdrv_has_zero_init       = bdrv_has_zero_init_1,
 532     .bdrv_co_readv  = parallels_co_readv,
 533     .bdrv_co_writev = parallels_co_writev,
 534
 535     .bdrv_create    = parallels_create,
 536     .bdrv_check     = parallels_check,
 537     .create_opts    = &parallels_create_opts,
 538 };
 539
 540 static void bdrv_parallels_init(void)
 541 {
 542     bdrv_register(&bdrv_parallels);
 543 }
 544
 545 block_init(bdrv_parallels_init);