block/qcow2-refcount.c

   1 /*
   2  * Block driver for the QCOW version 2 format
   3  *
   4  * Copyright (c) 2004-2006 Fabrice Bellard
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24
  25 #include "qemu-common.h"
  26 #include "block_int.h"
  27 #include "block/qcow2.h"
  28
  29 static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size);
  30 static int update_refcount(BlockDriverState *bs,
  31                             int64_t offset, int64_t length,
  32                             int addend);
  33
  34
  35 static int cache_refcount_updates = 0;
  36
  37 static int write_refcount_block(BDRVQcowState *s)
  38 {
  39     size_t size = s->cluster_size;
  40
  41     if (s->refcount_block_cache_offset == 0) {
  42         return 0;
  43     }
  44
  45     if (bdrv_pwrite(s->hd, s->refcount_block_cache_offset,
  46             s->refcount_block_cache, size) != size)
  47     {
  48         return -EIO;
  49     }
  50
  51     return 0;
  52 }
  53
  54 /*********************************************************/
  55 /* refcount handling */
  56
  57 int qcow2_refcount_init(BlockDriverState *bs)
  58 {
  59     BDRVQcowState *s = bs->opaque;
  60     int ret, refcount_table_size2, i;
  61
  62     s->refcount_block_cache = qemu_malloc(s->cluster_size);
  63     refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t);
  64     s->refcount_table = qemu_malloc(refcount_table_size2);
  65     if (s->refcount_table_size > 0) {
  66         ret = bdrv_pread(s->hd, s->refcount_table_offset,
  67                          s->refcount_table, refcount_table_size2);
  68         if (ret != refcount_table_size2)
  69             goto fail;
  70         for(i = 0; i < s->refcount_table_size; i++)
  71             be64_to_cpus(&s->refcount_table[i]);
  72     }
  73     return 0;
  74  fail:
  75     return -ENOMEM;
  76 }
  77
  78 void qcow2_refcount_close(BlockDriverState *bs)
  79 {
  80     BDRVQcowState *s = bs->opaque;
  81     qemu_free(s->refcount_block_cache);
  82     qemu_free(s->refcount_table);
  83 }
  84
  85
  86 static int load_refcount_block(BlockDriverState *bs,
  87                                int64_t refcount_block_offset)
  88 {
  89     BDRVQcowState *s = bs->opaque;
  90     int ret;
  91
  92     if (cache_refcount_updates) {
  93         write_refcount_block(s);
  94     }
  95
  96     ret = bdrv_pread(s->hd, refcount_block_offset, s->refcount_block_cache,
  97                      s->cluster_size);
  98     if (ret != s->cluster_size)
  99         return -EIO;
 100     s->refcount_block_cache_offset = refcount_block_offset;
 101     return 0;
 102 }
 103
 104 static int get_refcount(BlockDriverState *bs, int64_t cluster_index)
 105 {
 106     BDRVQcowState *s = bs->opaque;
 107     int refcount_table_index, block_index;
 108     int64_t refcount_block_offset;
 109
 110     refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
 111     if (refcount_table_index >= s->refcount_table_size)
 112         return 0;
 113     refcount_block_offset = s->refcount_table[refcount_table_index];
 114     if (!refcount_block_offset)
 115         return 0;
 116     if (refcount_block_offset != s->refcount_block_cache_offset) {
 117         /* better than nothing: return allocated if read error */
 118         if (load_refcount_block(bs, refcount_block_offset) < 0)
 119             return 1;
 120     }
 121     block_index = cluster_index &
 122         ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
 123     return be16_to_cpu(s->refcount_block_cache[block_index]);
 124 }
 125
 126 static int grow_refcount_table(BlockDriverState *bs, int min_size)
 127 {
 128     BDRVQcowState *s = bs->opaque;
 129     int new_table_size, new_table_size2, refcount_table_clusters, i, ret;
 130     uint64_t *new_table;
 131     int64_t table_offset;
 132     uint8_t data[12];
 133     int old_table_size;
 134     int64_t old_table_offset;
 135
 136     if (min_size <= s->refcount_table_size)
 137         return 0;
 138     /* compute new table size */
 139     refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
 140     for(;;) {
 141         if (refcount_table_clusters == 0) {
 142             refcount_table_clusters = 1;
 143         } else {
 144             refcount_table_clusters = (refcount_table_clusters * 3 + 1) / 2;
 145         }
 146         new_table_size = refcount_table_clusters << (s->cluster_bits - 3);
 147         if (min_size <= new_table_size)
 148             break;
 149     }
 150 #ifdef DEBUG_ALLOC2
 151     printf("grow_refcount_table from %d to %d\n",
 152            s->refcount_table_size,
 153            new_table_size);
 154 #endif
 155     new_table_size2 = new_table_size * sizeof(uint64_t);
 156     new_table = qemu_mallocz(new_table_size2);
 157     memcpy(new_table, s->refcount_table,
 158            s->refcount_table_size * sizeof(uint64_t));
 159     for(i = 0; i < s->refcount_table_size; i++)
 160         cpu_to_be64s(&new_table[i]);
 161     /* Note: we cannot update the refcount now to avoid recursion */
 162     table_offset = alloc_clusters_noref(bs, new_table_size2);
 163     ret = bdrv_pwrite(s->hd, table_offset, new_table, new_table_size2);
 164     if (ret != new_table_size2)
 165         goto fail;
 166     for(i = 0; i < s->refcount_table_size; i++)
 167         be64_to_cpus(&new_table[i]);
 168
 169     cpu_to_be64w((uint64_t*)data, table_offset);
 170     cpu_to_be32w((uint32_t*)(data + 8), refcount_table_clusters);
 171     if (bdrv_pwrite(s->hd, offsetof(QCowHeader, refcount_table_offset),
 172                     data, sizeof(data)) != sizeof(data))
 173         goto fail;
 174     qemu_free(s->refcount_table);
 175     old_table_offset = s->refcount_table_offset;
 176     old_table_size = s->refcount_table_size;
 177     s->refcount_table = new_table;
 178     s->refcount_table_size = new_table_size;
 179     s->refcount_table_offset = table_offset;
 180
 181     update_refcount(bs, table_offset, new_table_size2, 1);
 182     qcow2_free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t));
 183     return 0;
 184  fail:
 185     qcow2_free_clusters(bs, table_offset, new_table_size2);
 186     qemu_free(new_table);
 187     return -EIO;
 188 }
 189
 190
 191 static int64_t alloc_refcount_block(BlockDriverState *bs, int64_t cluster_index)
 192 {
 193     BDRVQcowState *s = bs->opaque;
 194     int64_t offset, refcount_block_offset;
 195     unsigned int refcount_table_index;
 196     int ret;
 197     uint64_t data64;
 198     int cache = cache_refcount_updates;
 199
 200     /* Find L1 index and grow refcount table if needed */
 201     refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
 202     if (refcount_table_index >= s->refcount_table_size) {
 203         ret = grow_refcount_table(bs, refcount_table_index + 1);
 204         if (ret < 0)
 205             return ret;
 206     }
 207
 208     /* Load or allocate the refcount block */
 209     refcount_block_offset = s->refcount_table[refcount_table_index];
 210     if (!refcount_block_offset) {
 211         if (cache_refcount_updates) {
 212             write_refcount_block(s);
 213             cache_refcount_updates = 0;
 214         }
 215         /* create a new refcount block */
 216         /* Note: we cannot update the refcount now to avoid recursion */
 217         offset = alloc_clusters_noref(bs, s->cluster_size);
 218         memset(s->refcount_block_cache, 0, s->cluster_size);
 219         ret = bdrv_pwrite(s->hd, offset, s->refcount_block_cache, s->cluster_size);
 220         if (ret != s->cluster_size)
 221             return -EINVAL;
 222         s->refcount_table[refcount_table_index] = offset;
 223         data64 = cpu_to_be64(offset);
 224         ret = bdrv_pwrite(s->hd, s->refcount_table_offset +
 225                           refcount_table_index * sizeof(uint64_t),
 226                           &data64, sizeof(data64));
 227         if (ret != sizeof(data64))
 228             return -EINVAL;
 229
 230         refcount_block_offset = offset;
 231         s->refcount_block_cache_offset = offset;
 232         update_refcount(bs, offset, s->cluster_size, 1);
 233         cache_refcount_updates = cache;
 234     } else {
 235         if (refcount_block_offset != s->refcount_block_cache_offset) {
 236             if (load_refcount_block(bs, refcount_block_offset) < 0)
 237                 return -EIO;
 238         }
 239     }
 240
 241     return refcount_block_offset;
 242 }
 243
 244 #define REFCOUNTS_PER_SECTOR (512 >> REFCOUNT_SHIFT)
 245 static int write_refcount_block_entries(BDRVQcowState *s,
 246     int64_t refcount_block_offset, int first_index, int last_index)
 247 {
 248     size_t size;
 249
 250     if (cache_refcount_updates) {
 251         return 0;
 252     }
 253
 254     first_index &= ~(REFCOUNTS_PER_SECTOR - 1);
 255     last_index = (last_index + REFCOUNTS_PER_SECTOR)
 256         & ~(REFCOUNTS_PER_SECTOR - 1);
 257
 258     size = (last_index - first_index) << REFCOUNT_SHIFT;
 259     if (bdrv_pwrite(s->hd,
 260         refcount_block_offset + (first_index << REFCOUNT_SHIFT),
 261         &s->refcount_block_cache[first_index], size) != size)
 262     {
 263         return -EIO;
 264     }
 265
 266     return 0;
 267 }
 268
 269 /* XXX: cache several refcount block clusters ? */
 270 static int update_refcount(BlockDriverState *bs,
 271                             int64_t offset, int64_t length,
 272                             int addend)
 273 {
 274     BDRVQcowState *s = bs->opaque;
 275     int64_t start, last, cluster_offset;
 276     int64_t refcount_block_offset = 0;
 277     int64_t table_index = -1, old_table_index;
 278     int first_index = -1, last_index = -1;
 279
 280 #ifdef DEBUG_ALLOC2
 281     printf("update_refcount: offset=%" PRId64 " size=%" PRId64 " addend=%d\n",
 282            offset, length, addend);
 283 #endif
 284     if (length <= 0)
 285         return -EINVAL;
 286     start = offset & ~(s->cluster_size - 1);
 287     last = (offset + length - 1) & ~(s->cluster_size - 1);
 288     for(cluster_offset = start; cluster_offset <= last;
 289         cluster_offset += s->cluster_size)
 290     {
 291         int block_index, refcount;
 292         int64_t cluster_index = cluster_offset >> s->cluster_bits;
 293
 294         /* Only write refcount block to disk when we are done with it */
 295         old_table_index = table_index;
 296         table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
 297         if ((old_table_index >= 0) && (table_index != old_table_index)) {
 298
 299             if (write_refcount_block_entries(s, refcount_block_offset,
 300                 first_index, last_index) < 0)
 301             {
 302                 return -EIO;
 303             }
 304
 305             first_index = -1;
 306             last_index = -1;
 307         }
 308
 309         /* Load the refcount block and allocate it if needed */
 310         refcount_block_offset = alloc_refcount_block(bs, cluster_index);
 311         if (refcount_block_offset < 0) {
 312             return refcount_block_offset;
 313         }
 314
 315         /* we can update the count and save it */
 316         block_index = cluster_index &
 317             ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
 318         if (first_index == -1 || block_index < first_index) {
 319             first_index = block_index;
 320         }
 321         if (block_index > last_index) {
 322             last_index = block_index;
 323         }
 324
 325         refcount = be16_to_cpu(s->refcount_block_cache[block_index]);
 326         refcount += addend;
 327         if (refcount < 0 || refcount > 0xffff)
 328             return -EINVAL;
 329         if (refcount == 0 && cluster_index < s->free_cluster_index) {
 330             s->free_cluster_index = cluster_index;
 331         }
 332         s->refcount_block_cache[block_index] = cpu_to_be16(refcount);
 333     }
 334
 335     /* Write last changed block to disk */
 336     if (refcount_block_offset != 0) {
 337         if (write_refcount_block_entries(s, refcount_block_offset,
 338             first_index, last_index) < 0)
 339         {
 340             return -EIO;
 341         }
 342     }
 343
 344     return 0;
 345 }
 346
 347 /* addend must be 1 or -1 */
 348 static int update_cluster_refcount(BlockDriverState *bs,
 349                                    int64_t cluster_index,
 350                                    int addend)
 351 {
 352     BDRVQcowState *s = bs->opaque;
 353     int ret;
 354
 355     ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend);
 356     if (ret < 0) {
 357         return ret;
 358     }
 359
 360     return get_refcount(bs, cluster_index);
 361 }
 362
 363
 364
 365 /*********************************************************/
 366 /* cluster allocation functions */
 367
 368
 369
 370 /* return < 0 if error */
 371 static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size)
 372 {
 373     BDRVQcowState *s = bs->opaque;
 374     int i, nb_clusters;
 375
 376     nb_clusters = size_to_clusters(s, size);
 377 retry:
 378     for(i = 0; i < nb_clusters; i++) {
 379         int64_t i = s->free_cluster_index++;
 380         if (get_refcount(bs, i) != 0)
 381             goto retry;
 382     }
 383 #ifdef DEBUG_ALLOC2
 384     printf("alloc_clusters: size=%" PRId64 " -> %" PRId64 "\n",
 385             size,
 386             (s->free_cluster_index - nb_clusters) << s->cluster_bits);
 387 #endif
 388     return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
 389 }
 390
 391 int64_t qcow2_alloc_clusters(BlockDriverState *bs, int64_t size)
 392 {
 393     int64_t offset;
 394
 395     offset = alloc_clusters_noref(bs, size);
 396     update_refcount(bs, offset, size, 1);
 397     return offset;
 398 }
 399
 400 /* only used to allocate compressed sectors. We try to allocate
 401    contiguous sectors. size must be <= cluster_size */
 402 int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size)
 403 {
 404     BDRVQcowState *s = bs->opaque;
 405     int64_t offset, cluster_offset;
 406     int free_in_cluster;
 407
 408     assert(size > 0 && size <= s->cluster_size);
 409     if (s->free_byte_offset == 0) {
 410         s->free_byte_offset = qcow2_alloc_clusters(bs, s->cluster_size);
 411     }
 412  redo:
 413     free_in_cluster = s->cluster_size -
 414         (s->free_byte_offset & (s->cluster_size - 1));
 415     if (size <= free_in_cluster) {
 416         /* enough space in current cluster */
 417         offset = s->free_byte_offset;
 418         s->free_byte_offset += size;
 419         free_in_cluster -= size;
 420         if (free_in_cluster == 0)
 421             s->free_byte_offset = 0;
 422         if ((offset & (s->cluster_size - 1)) != 0)
 423             update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
 424     } else {
 425         offset = qcow2_alloc_clusters(bs, s->cluster_size);
 426         cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1);
 427         if ((cluster_offset + s->cluster_size) == offset) {
 428             /* we are lucky: contiguous data */
 429             offset = s->free_byte_offset;
 430             update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
 431             s->free_byte_offset += size;
 432         } else {
 433             s->free_byte_offset = offset;
 434             goto redo;
 435         }
 436     }
 437     return offset;
 438 }
 439
 440 void qcow2_free_clusters(BlockDriverState *bs,
 441                           int64_t offset, int64_t size)
 442 {
 443     update_refcount(bs, offset, size, -1);
 444 }
 445
 446 /*
 447  * free_any_clusters
 448  *
 449  * free clusters according to its type: compressed or not
 450  *
 451  */
 452
 453 void qcow2_free_any_clusters(BlockDriverState *bs,
 454     uint64_t cluster_offset, int nb_clusters)
 455 {
 456     BDRVQcowState *s = bs->opaque;
 457
 458     /* free the cluster */
 459
 460     if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
 461         int nb_csectors;
 462         nb_csectors = ((cluster_offset >> s->csize_shift) &
 463                        s->csize_mask) + 1;
 464         qcow2_free_clusters(bs,
 465             (cluster_offset & s->cluster_offset_mask) & ~511,
 466             nb_csectors * 512);
 467         return;
 468     }
 469
 470     qcow2_free_clusters(bs, cluster_offset, nb_clusters << s->cluster_bits);
 471
 472     return;
 473 }
 474
 475
 476
 477 /*********************************************************/
 478 /* snapshots and image creation */
 479
 480
 481
 482 void qcow2_create_refcount_update(QCowCreateState *s, int64_t offset,
 483     int64_t size)
 484 {
 485     int refcount;
 486     int64_t start, last, cluster_offset;
 487     uint16_t *p;
 488
 489     start = offset & ~(s->cluster_size - 1);
 490     last = (offset + size - 1)  & ~(s->cluster_size - 1);
 491     for(cluster_offset = start; cluster_offset <= last;
 492         cluster_offset += s->cluster_size) {
 493         p = &s->refcount_block[cluster_offset >> s->cluster_bits];
 494         refcount = be16_to_cpu(*p);
 495         refcount++;
 496         *p = cpu_to_be16(refcount);
 497     }
 498 }
 499
 500 /* update the refcounts of snapshots and the copied flag */
 501 int qcow2_update_snapshot_refcount(BlockDriverState *bs,
 502     int64_t l1_table_offset, int l1_size, int addend)
 503 {
 504     BDRVQcowState *s = bs->opaque;
 505     uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated;
 506     int64_t old_offset, old_l2_offset;
 507     int l2_size, i, j, l1_modified, l2_modified, nb_csectors, refcount;
 508
 509     qcow2_l2_cache_reset(bs);
 510     cache_refcount_updates = 1;
 511
 512     l2_table = NULL;
 513     l1_table = NULL;
 514     l1_size2 = l1_size * sizeof(uint64_t);
 515     l1_allocated = 0;
 516     if (l1_table_offset != s->l1_table_offset) {
 517         l1_table = qemu_mallocz(align_offset(l1_size2, 512));
 518         l1_allocated = 1;
 519         if (bdrv_pread(s->hd, l1_table_offset,
 520                        l1_table, l1_size2) != l1_size2)
 521             goto fail;
 522         for(i = 0;i < l1_size; i++)
 523             be64_to_cpus(&l1_table[i]);
 524     } else {
 525         assert(l1_size == s->l1_size);
 526         l1_table = s->l1_table;
 527         l1_allocated = 0;
 528     }
 529
 530     l2_size = s->l2_size * sizeof(uint64_t);
 531     l2_table = qemu_malloc(l2_size);
 532     l1_modified = 0;
 533     for(i = 0; i < l1_size; i++) {
 534         l2_offset = l1_table[i];
 535         if (l2_offset) {
 536             old_l2_offset = l2_offset;
 537             l2_offset &= ~QCOW_OFLAG_COPIED;
 538             l2_modified = 0;
 539             if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
 540                 goto fail;
 541             for(j = 0; j < s->l2_size; j++) {
 542                 offset = be64_to_cpu(l2_table[j]);
 543                 if (offset != 0) {
 544                     old_offset = offset;
 545                     offset &= ~QCOW_OFLAG_COPIED;
 546                     if (offset & QCOW_OFLAG_COMPRESSED) {
 547                         nb_csectors = ((offset >> s->csize_shift) &
 548                                        s->csize_mask) + 1;
 549                         if (addend != 0)
 550                             update_refcount(bs, (offset & s->cluster_offset_mask) & ~511,
 551                                             nb_csectors * 512, addend);
 552                         /* compressed clusters are never modified */
 553                         refcount = 2;
 554                     } else {
 555                         if (addend != 0) {
 556                             refcount = update_cluster_refcount(bs, offset >> s->cluster_bits, addend);
 557                         } else {
 558                             refcount = get_refcount(bs, offset >> s->cluster_bits);
 559                         }
 560                     }
 561
 562                     if (refcount == 1) {
 563                         offset |= QCOW_OFLAG_COPIED;
 564                     }
 565                     if (offset != old_offset) {
 566                         l2_table[j] = cpu_to_be64(offset);
 567                         l2_modified = 1;
 568                     }
 569                 }
 570             }
 571             if (l2_modified) {
 572                 if (bdrv_pwrite(s->hd,
 573                                 l2_offset, l2_table, l2_size) != l2_size)
 574                     goto fail;
 575             }
 576
 577             if (addend != 0) {
 578                 refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend);
 579             } else {
 580                 refcount = get_refcount(bs, l2_offset >> s->cluster_bits);
 581             }
 582             if (refcount == 1) {
 583                 l2_offset |= QCOW_OFLAG_COPIED;
 584             }
 585             if (l2_offset != old_l2_offset) {
 586                 l1_table[i] = l2_offset;
 587                 l1_modified = 1;
 588             }
 589         }
 590     }
 591     if (l1_modified) {
 592         for(i = 0; i < l1_size; i++)
 593             cpu_to_be64s(&l1_table[i]);
 594         if (bdrv_pwrite(s->hd, l1_table_offset, l1_table,
 595                         l1_size2) != l1_size2)
 596             goto fail;
 597         for(i = 0; i < l1_size; i++)
 598             be64_to_cpus(&l1_table[i]);
 599     }
 600     if (l1_allocated)
 601         qemu_free(l1_table);
 602     qemu_free(l2_table);
 603     cache_refcount_updates = 0;
 604     write_refcount_block(s);
 605     return 0;
 606  fail:
 607     if (l1_allocated)
 608         qemu_free(l1_table);
 609     qemu_free(l2_table);
 610     cache_refcount_updates = 0;
 611     write_refcount_block(s);
 612     return -EIO;
 613 }
 614
 615
 616
 617
 618 /*********************************************************/
 619 /* refcount checking functions */
 620
 621
 622
 623 /*
 624  * Increases the refcount for a range of clusters in a given refcount table.
 625  * This is used to construct a temporary refcount table out of L1 and L2 tables
 626  * which can be compared the the refcount table saved in the image.
 627  *
 628  * Returns the number of errors in the image that were found
 629  */
 630 static int inc_refcounts(BlockDriverState *bs,
 631                           uint16_t *refcount_table,
 632                           int refcount_table_size,
 633                           int64_t offset, int64_t size)
 634 {
 635     BDRVQcowState *s = bs->opaque;
 636     int64_t start, last, cluster_offset;
 637     int k;
 638     int errors = 0;
 639
 640     if (size <= 0)
 641         return 0;
 642
 643     start = offset & ~(s->cluster_size - 1);
 644     last = (offset + size - 1) & ~(s->cluster_size - 1);
 645     for(cluster_offset = start; cluster_offset <= last;
 646         cluster_offset += s->cluster_size) {
 647         k = cluster_offset >> s->cluster_bits;
 648         if (k < 0 || k >= refcount_table_size) {
 649             fprintf(stderr, "ERROR: invalid cluster offset=0x%" PRIx64 "\n",
 650                 cluster_offset);
 651             errors++;
 652         } else {
 653             if (++refcount_table[k] == 0) {
 654                 fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64
 655                     "\n", cluster_offset);
 656                 errors++;
 657             }
 658         }
 659     }
 660
 661     return errors;
 662 }
 663
 664 /*
 665  * Increases the refcount in the given refcount table for the all clusters
 666  * referenced in the L2 table. While doing so, performs some checks on L2
 667  * entries.
 668  *
 669  * Returns the number of errors found by the checks or -errno if an internal
 670  * error occurred.
 671  */
 672 static int check_refcounts_l2(BlockDriverState *bs,
 673     uint16_t *refcount_table, int refcount_table_size, int64_t l2_offset,
 674     int check_copied)
 675 {
 676     BDRVQcowState *s = bs->opaque;
 677     uint64_t *l2_table, offset;
 678     int i, l2_size, nb_csectors, refcount;
 679     int errors = 0;
 680
 681     /* Read L2 table from disk */
 682     l2_size = s->l2_size * sizeof(uint64_t);
 683     l2_table = qemu_malloc(l2_size);
 684
 685     if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
 686         goto fail;
 687
 688     /* Do the actual checks */
 689     for(i = 0; i < s->l2_size; i++) {
 690         offset = be64_to_cpu(l2_table[i]);
 691         if (offset != 0) {
 692             if (offset & QCOW_OFLAG_COMPRESSED) {
 693                 /* Compressed clusters don't have QCOW_OFLAG_COPIED */
 694                 if (offset & QCOW_OFLAG_COPIED) {
 695                     fprintf(stderr, "ERROR: cluster %" PRId64 ": "
 696                         "copied flag must never be set for compressed "
 697                         "clusters\n", offset >> s->cluster_bits);
 698                     offset &= ~QCOW_OFLAG_COPIED;
 699                     errors++;
 700                 }
 701
 702                 /* Mark cluster as used */
 703                 nb_csectors = ((offset >> s->csize_shift) &
 704                                s->csize_mask) + 1;
 705                 offset &= s->cluster_offset_mask;
 706                 errors += inc_refcounts(bs, refcount_table,
 707                               refcount_table_size,
 708                               offset & ~511, nb_csectors * 512);
 709             } else {
 710                 /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
 711                 if (check_copied) {
 712                     uint64_t entry = offset;
 713                     offset &= ~QCOW_OFLAG_COPIED;
 714                     refcount = get_refcount(bs, offset >> s->cluster_bits);
 715                     if ((refcount == 1) != ((entry & QCOW_OFLAG_COPIED) != 0)) {
 716                         fprintf(stderr, "ERROR OFLAG_COPIED: offset=%"
 717                             PRIx64 " refcount=%d\n", entry, refcount);
 718                         errors++;
 719                     }
 720                 }
 721
 722                 /* Mark cluster as used */
 723                 offset &= ~QCOW_OFLAG_COPIED;
 724                 errors += inc_refcounts(bs, refcount_table,
 725                               refcount_table_size,
 726                               offset, s->cluster_size);
 727
 728                 /* Correct offsets are cluster aligned */
 729                 if (offset & (s->cluster_size - 1)) {
 730                     fprintf(stderr, "ERROR offset=%" PRIx64 ": Cluster is not "
 731                         "properly aligned; L2 entry corrupted.\n", offset);
 732                     errors++;
 733                 }
 734             }
 735         }
 736     }
 737
 738     qemu_free(l2_table);
 739     return errors;
 740
 741 fail:
 742     fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
 743     qemu_free(l2_table);
 744     return -EIO;
 745 }
 746
 747 /*
 748  * Increases the refcount for the L1 table, its L2 tables and all referenced
 749  * clusters in the given refcount table. While doing so, performs some checks
 750  * on L1 and L2 entries.
 751  *
 752  * Returns the number of errors found by the checks or -errno if an internal
 753  * error occurred.
 754  */
 755 static int check_refcounts_l1(BlockDriverState *bs,
 756                               uint16_t *refcount_table,
 757                               int refcount_table_size,
 758                               int64_t l1_table_offset, int l1_size,
 759                               int check_copied)
 760 {
 761     BDRVQcowState *s = bs->opaque;
 762     uint64_t *l1_table, l2_offset, l1_size2;
 763     int i, refcount, ret;
 764     int errors = 0;
 765
 766     l1_size2 = l1_size * sizeof(uint64_t);
 767
 768     /* Mark L1 table as used */
 769     errors += inc_refcounts(bs, refcount_table, refcount_table_size,
 770                   l1_table_offset, l1_size2);
 771
 772     /* Read L1 table entries from disk */
 773     l1_table = qemu_malloc(l1_size2);
 774     if (bdrv_pread(s->hd, l1_table_offset,
 775                    l1_table, l1_size2) != l1_size2)
 776         goto fail;
 777     for(i = 0;i < l1_size; i++)
 778         be64_to_cpus(&l1_table[i]);
 779
 780     /* Do the actual checks */
 781     for(i = 0; i < l1_size; i++) {
 782         l2_offset = l1_table[i];
 783         if (l2_offset) {
 784             /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
 785             if (check_copied) {
 786                 refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED)
 787                     >> s->cluster_bits);
 788                 if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) {
 789                     fprintf(stderr, "ERROR OFLAG_COPIED: l2_offset=%" PRIx64
 790                         " refcount=%d\n", l2_offset, refcount);
 791                     errors++;
 792                 }
 793             }
 794
 795             /* Mark L2 table as used */
 796             l2_offset &= ~QCOW_OFLAG_COPIED;
 797             errors += inc_refcounts(bs, refcount_table,
 798                           refcount_table_size,
 799                           l2_offset,
 800                           s->cluster_size);
 801
 802             /* L2 tables are cluster aligned */
 803             if (l2_offset & (s->cluster_size - 1)) {
 804                 fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not "
 805                     "cluster aligned; L1 entry corrupted\n", l2_offset);
 806                 errors++;
 807             }
 808
 809             /* Process and check L2 entries */
 810             ret = check_refcounts_l2(bs, refcount_table, refcount_table_size,
 811                 l2_offset, check_copied);
 812             if (ret < 0) {
 813                 goto fail;
 814             }
 815             errors += ret;
 816         }
 817     }
 818     qemu_free(l1_table);
 819     return errors;
 820
 821 fail:
 822     fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
 823     qemu_free(l1_table);
 824     return -EIO;
 825 }
 826
 827 /*
 828  * Checks an image for refcount consistency.
 829  *
 830  * Returns 0 if no errors are found, the number of errors in case the image is
 831  * detected as corrupted, and -errno when an internal error occured.
 832  */
 833 int qcow2_check_refcounts(BlockDriverState *bs)
 834 {
 835     BDRVQcowState *s = bs->opaque;
 836     int64_t size;
 837     int nb_clusters, refcount1, refcount2, i;
 838     QCowSnapshot *sn;
 839     uint16_t *refcount_table;
 840     int ret, errors = 0;
 841
 842     size = bdrv_getlength(s->hd);
 843     nb_clusters = size_to_clusters(s, size);
 844     refcount_table = qemu_mallocz(nb_clusters * sizeof(uint16_t));
 845
 846     /* header */
 847     errors += inc_refcounts(bs, refcount_table, nb_clusters,
 848                   0, s->cluster_size);
 849
 850     /* current L1 table */
 851     ret = check_refcounts_l1(bs, refcount_table, nb_clusters,
 852                        s->l1_table_offset, s->l1_size, 1);
 853     if (ret < 0) {
 854         return ret;
 855     }
 856     errors += ret;
 857
 858     /* snapshots */
 859     for(i = 0; i < s->nb_snapshots; i++) {
 860         sn = s->snapshots + i;
 861         check_refcounts_l1(bs, refcount_table, nb_clusters,
 862                            sn->l1_table_offset, sn->l1_size, 0);
 863     }
 864     errors += inc_refcounts(bs, refcount_table, nb_clusters,
 865                   s->snapshots_offset, s->snapshots_size);
 866
 867     /* refcount data */
 868     errors += inc_refcounts(bs, refcount_table, nb_clusters,
 869                   s->refcount_table_offset,
 870                   s->refcount_table_size * sizeof(uint64_t));
 871     for(i = 0; i < s->refcount_table_size; i++) {
 872         int64_t offset;
 873         offset = s->refcount_table[i];
 874         if (offset != 0) {
 875             errors += inc_refcounts(bs, refcount_table, nb_clusters,
 876                           offset, s->cluster_size);
 877         }
 878     }
 879
 880     /* compare ref counts */
 881     for(i = 0; i < nb_clusters; i++) {
 882         refcount1 = get_refcount(bs, i);
 883         refcount2 = refcount_table[i];
 884         if (refcount1 != refcount2) {
 885             fprintf(stderr, "ERROR cluster %d refcount=%d reference=%d\n",
 886                    i, refcount1, refcount2);
 887             errors++;
 888         }
 889     }
 890
 891     qemu_free(refcount_table);
 892
 893     return errors;
 894 }
 895