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     ret = bdrv_pwrite(s->hd, offsetof(QCowHeader, refcount_table_offset),
 172                     data, sizeof(data));
 173     if (ret != sizeof(data)) {
 174         goto fail;
 175     }
 176
 177     qemu_free(s->refcount_table);
 178     old_table_offset = s->refcount_table_offset;
 179     old_table_size = s->refcount_table_size;
 180     s->refcount_table = new_table;
 181     s->refcount_table_size = new_table_size;
 182     s->refcount_table_offset = table_offset;
 183
 184     update_refcount(bs, table_offset, new_table_size2, 1);
 185     qcow2_free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t));
 186     return 0;
 187  fail:
 188     qemu_free(new_table);
 189     return ret < 0 ? ret : -EIO;
 190 }
 191
 192
 193 static int64_t alloc_refcount_block(BlockDriverState *bs, int64_t cluster_index)
 194 {
 195     BDRVQcowState *s = bs->opaque;
 196     int64_t offset, refcount_block_offset;
 197     unsigned int refcount_table_index;
 198     int ret;
 199     uint64_t data64;
 200     int cache = cache_refcount_updates;
 201
 202     /* Find L1 index and grow refcount table if needed */
 203     refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
 204     if (refcount_table_index >= s->refcount_table_size) {
 205         ret = grow_refcount_table(bs, refcount_table_index + 1);
 206         if (ret < 0)
 207             return ret;
 208     }
 209
 210     /* Load or allocate the refcount block */
 211     refcount_block_offset = s->refcount_table[refcount_table_index];
 212     if (!refcount_block_offset) {
 213         if (cache_refcount_updates) {
 214             write_refcount_block(s);
 215             cache_refcount_updates = 0;
 216         }
 217         /* create a new refcount block */
 218         /* Note: we cannot update the refcount now to avoid recursion */
 219         offset = alloc_clusters_noref(bs, s->cluster_size);
 220         memset(s->refcount_block_cache, 0, s->cluster_size);
 221         ret = bdrv_pwrite(s->hd, offset, s->refcount_block_cache, s->cluster_size);
 222         if (ret != s->cluster_size)
 223             return -EINVAL;
 224         s->refcount_table[refcount_table_index] = offset;
 225         data64 = cpu_to_be64(offset);
 226         ret = bdrv_pwrite(s->hd, s->refcount_table_offset +
 227                           refcount_table_index * sizeof(uint64_t),
 228                           &data64, sizeof(data64));
 229         if (ret != sizeof(data64))
 230             return -EINVAL;
 231
 232         refcount_block_offset = offset;
 233         s->refcount_block_cache_offset = offset;
 234         update_refcount(bs, offset, s->cluster_size, 1);
 235         cache_refcount_updates = cache;
 236     } else {
 237         if (refcount_block_offset != s->refcount_block_cache_offset) {
 238             if (load_refcount_block(bs, refcount_block_offset) < 0)
 239                 return -EIO;
 240         }
 241     }
 242
 243     return refcount_block_offset;
 244 }
 245
 246 #define REFCOUNTS_PER_SECTOR (512 >> REFCOUNT_SHIFT)
 247 static int write_refcount_block_entries(BDRVQcowState *s,
 248     int64_t refcount_block_offset, int first_index, int last_index)
 249 {
 250     size_t size;
 251
 252     if (cache_refcount_updates) {
 253         return 0;
 254     }
 255
 256     first_index &= ~(REFCOUNTS_PER_SECTOR - 1);
 257     last_index = (last_index + REFCOUNTS_PER_SECTOR)
 258         & ~(REFCOUNTS_PER_SECTOR - 1);
 259
 260     size = (last_index - first_index) << REFCOUNT_SHIFT;
 261     if (bdrv_pwrite(s->hd,
 262         refcount_block_offset + (first_index << REFCOUNT_SHIFT),
 263         &s->refcount_block_cache[first_index], size) != size)
 264     {
 265         return -EIO;
 266     }
 267
 268     return 0;
 269 }
 270
 271 /* XXX: cache several refcount block clusters ? */
 272 static int update_refcount(BlockDriverState *bs,
 273                             int64_t offset, int64_t length,
 274                             int addend)
 275 {
 276     BDRVQcowState *s = bs->opaque;
 277     int64_t start, last, cluster_offset;
 278     int64_t refcount_block_offset = 0;
 279     int64_t table_index = -1, old_table_index;
 280     int first_index = -1, last_index = -1;
 281     int ret;
 282
 283 #ifdef DEBUG_ALLOC2
 284     printf("update_refcount: offset=%" PRId64 " size=%" PRId64 " addend=%d\n",
 285            offset, length, addend);
 286 #endif
 287     if (length <= 0)
 288         return -EINVAL;
 289     start = offset & ~(s->cluster_size - 1);
 290     last = (offset + length - 1) & ~(s->cluster_size - 1);
 291     for(cluster_offset = start; cluster_offset <= last;
 292         cluster_offset += s->cluster_size)
 293     {
 294         int block_index, refcount;
 295         int64_t cluster_index = cluster_offset >> s->cluster_bits;
 296         int64_t new_block;
 297
 298         /* Only write refcount block to disk when we are done with it */
 299         old_table_index = table_index;
 300         table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
 301         if ((old_table_index >= 0) && (table_index != old_table_index)) {
 302
 303             if (write_refcount_block_entries(s, refcount_block_offset,
 304                 first_index, last_index) < 0)
 305             {
 306                 return -EIO;
 307             }
 308
 309             first_index = -1;
 310             last_index = -1;
 311         }
 312
 313         /* Load the refcount block and allocate it if needed */
 314         new_block = alloc_refcount_block(bs, cluster_index);
 315         if (new_block < 0) {
 316             ret = new_block;
 317             goto fail;
 318         }
 319         refcount_block_offset = new_block;
 320
 321         /* we can update the count and save it */
 322         block_index = cluster_index &
 323             ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
 324         if (first_index == -1 || block_index < first_index) {
 325             first_index = block_index;
 326         }
 327         if (block_index > last_index) {
 328             last_index = block_index;
 329         }
 330
 331         refcount = be16_to_cpu(s->refcount_block_cache[block_index]);
 332         refcount += addend;
 333         if (refcount < 0 || refcount > 0xffff) {
 334             ret = -EINVAL;
 335             goto fail;
 336         }
 337         if (refcount == 0 && cluster_index < s->free_cluster_index) {
 338             s->free_cluster_index = cluster_index;
 339         }
 340         s->refcount_block_cache[block_index] = cpu_to_be16(refcount);
 341     }
 342
 343     ret = 0;
 344 fail:
 345
 346     /* Write last changed block to disk */
 347     if (refcount_block_offset != 0) {
 348         if (write_refcount_block_entries(s, refcount_block_offset,
 349             first_index, last_index) < 0)
 350         {
 351             return ret < 0 ? ret : -EIO;
 352         }
 353     }
 354
 355     /*
 356      * Try do undo any updates if an error is returned (This may succeed in
 357      * some cases like ENOSPC for allocating a new refcount block)
 358      */
 359     if (ret < 0) {
 360         int dummy;
 361         dummy = update_refcount(bs, offset, cluster_offset - offset, -addend);
 362     }
 363
 364     return ret;
 365 }
 366
 367 /* addend must be 1 or -1 */
 368 static int update_cluster_refcount(BlockDriverState *bs,
 369                                    int64_t cluster_index,
 370                                    int addend)
 371 {
 372     BDRVQcowState *s = bs->opaque;
 373     int ret;
 374
 375     ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend);
 376     if (ret < 0) {
 377         return ret;
 378     }
 379
 380     return get_refcount(bs, cluster_index);
 381 }
 382
 383
 384
 385 /*********************************************************/
 386 /* cluster allocation functions */
 387
 388
 389
 390 /* return < 0 if error */
 391 static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size)
 392 {
 393     BDRVQcowState *s = bs->opaque;
 394     int i, nb_clusters;
 395
 396     nb_clusters = size_to_clusters(s, size);
 397 retry:
 398     for(i = 0; i < nb_clusters; i++) {
 399         int64_t i = s->free_cluster_index++;
 400         if (get_refcount(bs, i) != 0)
 401             goto retry;
 402     }
 403 #ifdef DEBUG_ALLOC2
 404     printf("alloc_clusters: size=%" PRId64 " -> %" PRId64 "\n",
 405             size,
 406             (s->free_cluster_index - nb_clusters) << s->cluster_bits);
 407 #endif
 408     return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
 409 }
 410
 411 int64_t qcow2_alloc_clusters(BlockDriverState *bs, int64_t size)
 412 {
 413     int64_t offset;
 414
 415     offset = alloc_clusters_noref(bs, size);
 416     update_refcount(bs, offset, size, 1);
 417     return offset;
 418 }
 419
 420 /* only used to allocate compressed sectors. We try to allocate
 421    contiguous sectors. size must be <= cluster_size */
 422 int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size)
 423 {
 424     BDRVQcowState *s = bs->opaque;
 425     int64_t offset, cluster_offset;
 426     int free_in_cluster;
 427
 428     assert(size > 0 && size <= s->cluster_size);
 429     if (s->free_byte_offset == 0) {
 430         s->free_byte_offset = qcow2_alloc_clusters(bs, s->cluster_size);
 431     }
 432  redo:
 433     free_in_cluster = s->cluster_size -
 434         (s->free_byte_offset & (s->cluster_size - 1));
 435     if (size <= free_in_cluster) {
 436         /* enough space in current cluster */
 437         offset = s->free_byte_offset;
 438         s->free_byte_offset += size;
 439         free_in_cluster -= size;
 440         if (free_in_cluster == 0)
 441             s->free_byte_offset = 0;
 442         if ((offset & (s->cluster_size - 1)) != 0)
 443             update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
 444     } else {
 445         offset = qcow2_alloc_clusters(bs, s->cluster_size);
 446         cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1);
 447         if ((cluster_offset + s->cluster_size) == offset) {
 448             /* we are lucky: contiguous data */
 449             offset = s->free_byte_offset;
 450             update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
 451             s->free_byte_offset += size;
 452         } else {
 453             s->free_byte_offset = offset;
 454             goto redo;
 455         }
 456     }
 457     return offset;
 458 }
 459
 460 void qcow2_free_clusters(BlockDriverState *bs,
 461                           int64_t offset, int64_t size)
 462 {
 463     update_refcount(bs, offset, size, -1);
 464 }
 465
 466 /*
 467  * free_any_clusters
 468  *
 469  * free clusters according to its type: compressed or not
 470  *
 471  */
 472
 473 void qcow2_free_any_clusters(BlockDriverState *bs,
 474     uint64_t cluster_offset, int nb_clusters)
 475 {
 476     BDRVQcowState *s = bs->opaque;
 477
 478     /* free the cluster */
 479
 480     if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
 481         int nb_csectors;
 482         nb_csectors = ((cluster_offset >> s->csize_shift) &
 483                        s->csize_mask) + 1;
 484         qcow2_free_clusters(bs,
 485             (cluster_offset & s->cluster_offset_mask) & ~511,
 486             nb_csectors * 512);
 487         return;
 488     }
 489
 490     qcow2_free_clusters(bs, cluster_offset, nb_clusters << s->cluster_bits);
 491
 492     return;
 493 }
 494
 495
 496
 497 /*********************************************************/
 498 /* snapshots and image creation */
 499
 500
 501
 502 void qcow2_create_refcount_update(QCowCreateState *s, int64_t offset,
 503     int64_t size)
 504 {
 505     int refcount;
 506     int64_t start, last, cluster_offset;
 507     uint16_t *p;
 508
 509     start = offset & ~(s->cluster_size - 1);
 510     last = (offset + size - 1)  & ~(s->cluster_size - 1);
 511     for(cluster_offset = start; cluster_offset <= last;
 512         cluster_offset += s->cluster_size) {
 513         p = &s->refcount_block[cluster_offset >> s->cluster_bits];
 514         refcount = be16_to_cpu(*p);
 515         refcount++;
 516         *p = cpu_to_be16(refcount);
 517     }
 518 }
 519
 520 /* update the refcounts of snapshots and the copied flag */
 521 int qcow2_update_snapshot_refcount(BlockDriverState *bs,
 522     int64_t l1_table_offset, int l1_size, int addend)
 523 {
 524     BDRVQcowState *s = bs->opaque;
 525     uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated;
 526     int64_t old_offset, old_l2_offset;
 527     int l2_size, i, j, l1_modified, l2_modified, nb_csectors, refcount;
 528
 529     qcow2_l2_cache_reset(bs);
 530     cache_refcount_updates = 1;
 531
 532     l2_table = NULL;
 533     l1_table = NULL;
 534     l1_size2 = l1_size * sizeof(uint64_t);
 535     if (l1_table_offset != s->l1_table_offset) {
 536         if (l1_size2 != 0) {
 537             l1_table = qemu_mallocz(align_offset(l1_size2, 512));
 538         } else {
 539             l1_table = NULL;
 540         }
 541         l1_allocated = 1;
 542         if (bdrv_pread(s->hd, l1_table_offset,
 543                        l1_table, l1_size2) != l1_size2)
 544             goto fail;
 545         for(i = 0;i < l1_size; i++)
 546             be64_to_cpus(&l1_table[i]);
 547     } else {
 548         assert(l1_size == s->l1_size);
 549         l1_table = s->l1_table;
 550         l1_allocated = 0;
 551     }
 552
 553     l2_size = s->l2_size * sizeof(uint64_t);
 554     l2_table = qemu_malloc(l2_size);
 555     l1_modified = 0;
 556     for(i = 0; i < l1_size; i++) {
 557         l2_offset = l1_table[i];
 558         if (l2_offset) {
 559             old_l2_offset = l2_offset;
 560             l2_offset &= ~QCOW_OFLAG_COPIED;
 561             l2_modified = 0;
 562             if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
 563                 goto fail;
 564             for(j = 0; j < s->l2_size; j++) {
 565                 offset = be64_to_cpu(l2_table[j]);
 566                 if (offset != 0) {
 567                     old_offset = offset;
 568                     offset &= ~QCOW_OFLAG_COPIED;
 569                     if (offset & QCOW_OFLAG_COMPRESSED) {
 570                         nb_csectors = ((offset >> s->csize_shift) &
 571                                        s->csize_mask) + 1;
 572                         if (addend != 0)
 573                             update_refcount(bs, (offset & s->cluster_offset_mask) & ~511,
 574                                             nb_csectors * 512, addend);
 575                         /* compressed clusters are never modified */
 576                         refcount = 2;
 577                     } else {
 578                         if (addend != 0) {
 579                             refcount = update_cluster_refcount(bs, offset >> s->cluster_bits, addend);
 580                         } else {
 581                             refcount = get_refcount(bs, offset >> s->cluster_bits);
 582                         }
 583                     }
 584
 585                     if (refcount == 1) {
 586                         offset |= QCOW_OFLAG_COPIED;
 587                     }
 588                     if (offset != old_offset) {
 589                         l2_table[j] = cpu_to_be64(offset);
 590                         l2_modified = 1;
 591                     }
 592                 }
 593             }
 594             if (l2_modified) {
 595                 if (bdrv_pwrite(s->hd,
 596                                 l2_offset, l2_table, l2_size) != l2_size)
 597                     goto fail;
 598             }
 599
 600             if (addend != 0) {
 601                 refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend);
 602             } else {
 603                 refcount = get_refcount(bs, l2_offset >> s->cluster_bits);
 604             }
 605             if (refcount == 1) {
 606                 l2_offset |= QCOW_OFLAG_COPIED;
 607             }
 608             if (l2_offset != old_l2_offset) {
 609                 l1_table[i] = l2_offset;
 610                 l1_modified = 1;
 611             }
 612         }
 613     }
 614     if (l1_modified) {
 615         for(i = 0; i < l1_size; i++)
 616             cpu_to_be64s(&l1_table[i]);
 617         if (bdrv_pwrite(s->hd, l1_table_offset, l1_table,
 618                         l1_size2) != l1_size2)
 619             goto fail;
 620         for(i = 0; i < l1_size; i++)
 621             be64_to_cpus(&l1_table[i]);
 622     }
 623     if (l1_allocated)
 624         qemu_free(l1_table);
 625     qemu_free(l2_table);
 626     cache_refcount_updates = 0;
 627     write_refcount_block(s);
 628     return 0;
 629  fail:
 630     if (l1_allocated)
 631         qemu_free(l1_table);
 632     qemu_free(l2_table);
 633     cache_refcount_updates = 0;
 634     write_refcount_block(s);
 635     return -EIO;
 636 }
 637
 638
 639
 640
 641 /*********************************************************/
 642 /* refcount checking functions */
 643
 644
 645
 646 /*
 647  * Increases the refcount for a range of clusters in a given refcount table.
 648  * This is used to construct a temporary refcount table out of L1 and L2 tables
 649  * which can be compared the the refcount table saved in the image.
 650  *
 651  * Returns the number of errors in the image that were found
 652  */
 653 static int inc_refcounts(BlockDriverState *bs,
 654                           uint16_t *refcount_table,
 655                           int refcount_table_size,
 656                           int64_t offset, int64_t size)
 657 {
 658     BDRVQcowState *s = bs->opaque;
 659     int64_t start, last, cluster_offset;
 660     int k;
 661     int errors = 0;
 662
 663     if (size <= 0)
 664         return 0;
 665
 666     start = offset & ~(s->cluster_size - 1);
 667     last = (offset + size - 1) & ~(s->cluster_size - 1);
 668     for(cluster_offset = start; cluster_offset <= last;
 669         cluster_offset += s->cluster_size) {
 670         k = cluster_offset >> s->cluster_bits;
 671         if (k < 0 || k >= refcount_table_size) {
 672             fprintf(stderr, "ERROR: invalid cluster offset=0x%" PRIx64 "\n",
 673                 cluster_offset);
 674             errors++;
 675         } else {
 676             if (++refcount_table[k] == 0) {
 677                 fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64
 678                     "\n", cluster_offset);
 679                 errors++;
 680             }
 681         }
 682     }
 683
 684     return errors;
 685 }
 686
 687 /*
 688  * Increases the refcount in the given refcount table for the all clusters
 689  * referenced in the L2 table. While doing so, performs some checks on L2
 690  * entries.
 691  *
 692  * Returns the number of errors found by the checks or -errno if an internal
 693  * error occurred.
 694  */
 695 static int check_refcounts_l2(BlockDriverState *bs,
 696     uint16_t *refcount_table, int refcount_table_size, int64_t l2_offset,
 697     int check_copied)
 698 {
 699     BDRVQcowState *s = bs->opaque;
 700     uint64_t *l2_table, offset;
 701     int i, l2_size, nb_csectors, refcount;
 702     int errors = 0;
 703
 704     /* Read L2 table from disk */
 705     l2_size = s->l2_size * sizeof(uint64_t);
 706     l2_table = qemu_malloc(l2_size);
 707
 708     if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
 709         goto fail;
 710
 711     /* Do the actual checks */
 712     for(i = 0; i < s->l2_size; i++) {
 713         offset = be64_to_cpu(l2_table[i]);
 714         if (offset != 0) {
 715             if (offset & QCOW_OFLAG_COMPRESSED) {
 716                 /* Compressed clusters don't have QCOW_OFLAG_COPIED */
 717                 if (offset & QCOW_OFLAG_COPIED) {
 718                     fprintf(stderr, "ERROR: cluster %" PRId64 ": "
 719                         "copied flag must never be set for compressed "
 720                         "clusters\n", offset >> s->cluster_bits);
 721                     offset &= ~QCOW_OFLAG_COPIED;
 722                     errors++;
 723                 }
 724
 725                 /* Mark cluster as used */
 726                 nb_csectors = ((offset >> s->csize_shift) &
 727                                s->csize_mask) + 1;
 728                 offset &= s->cluster_offset_mask;
 729                 errors += inc_refcounts(bs, refcount_table,
 730                               refcount_table_size,
 731                               offset & ~511, nb_csectors * 512);
 732             } else {
 733                 /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
 734                 if (check_copied) {
 735                     uint64_t entry = offset;
 736                     offset &= ~QCOW_OFLAG_COPIED;
 737                     refcount = get_refcount(bs, offset >> s->cluster_bits);
 738                     if ((refcount == 1) != ((entry & QCOW_OFLAG_COPIED) != 0)) {
 739                         fprintf(stderr, "ERROR OFLAG_COPIED: offset=%"
 740                             PRIx64 " refcount=%d\n", entry, refcount);
 741                         errors++;
 742                     }
 743                 }
 744
 745                 /* Mark cluster as used */
 746                 offset &= ~QCOW_OFLAG_COPIED;
 747                 errors += inc_refcounts(bs, refcount_table,
 748                               refcount_table_size,
 749                               offset, s->cluster_size);
 750
 751                 /* Correct offsets are cluster aligned */
 752                 if (offset & (s->cluster_size - 1)) {
 753                     fprintf(stderr, "ERROR offset=%" PRIx64 ": Cluster is not "
 754                         "properly aligned; L2 entry corrupted.\n", offset);
 755                     errors++;
 756                 }
 757             }
 758         }
 759     }
 760
 761     qemu_free(l2_table);
 762     return errors;
 763
 764 fail:
 765     fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
 766     qemu_free(l2_table);
 767     return -EIO;
 768 }
 769
 770 /*
 771  * Increases the refcount for the L1 table, its L2 tables and all referenced
 772  * clusters in the given refcount table. While doing so, performs some checks
 773  * on L1 and L2 entries.
 774  *
 775  * Returns the number of errors found by the checks or -errno if an internal
 776  * error occurred.
 777  */
 778 static int check_refcounts_l1(BlockDriverState *bs,
 779                               uint16_t *refcount_table,
 780                               int refcount_table_size,
 781                               int64_t l1_table_offset, int l1_size,
 782                               int check_copied)
 783 {
 784     BDRVQcowState *s = bs->opaque;
 785     uint64_t *l1_table, l2_offset, l1_size2;
 786     int i, refcount, ret;
 787     int errors = 0;
 788
 789     l1_size2 = l1_size * sizeof(uint64_t);
 790
 791     /* Mark L1 table as used */
 792     errors += inc_refcounts(bs, refcount_table, refcount_table_size,
 793                   l1_table_offset, l1_size2);
 794
 795     /* Read L1 table entries from disk */
 796     if (l1_size2 == 0) {
 797         l1_table = NULL;
 798     } else {
 799         l1_table = qemu_malloc(l1_size2);
 800         if (bdrv_pread(s->hd, l1_table_offset,
 801                        l1_table, l1_size2) != l1_size2)
 802             goto fail;
 803         for(i = 0;i < l1_size; i++)
 804             be64_to_cpus(&l1_table[i]);
 805     }
 806
 807     /* Do the actual checks */
 808     for(i = 0; i < l1_size; i++) {
 809         l2_offset = l1_table[i];
 810         if (l2_offset) {
 811             /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
 812             if (check_copied) {
 813                 refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED)
 814                     >> s->cluster_bits);
 815                 if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) {
 816                     fprintf(stderr, "ERROR OFLAG_COPIED: l2_offset=%" PRIx64
 817                         " refcount=%d\n", l2_offset, refcount);
 818                     errors++;
 819                 }
 820             }
 821
 822             /* Mark L2 table as used */
 823             l2_offset &= ~QCOW_OFLAG_COPIED;
 824             errors += inc_refcounts(bs, refcount_table,
 825                           refcount_table_size,
 826                           l2_offset,
 827                           s->cluster_size);
 828
 829             /* L2 tables are cluster aligned */
 830             if (l2_offset & (s->cluster_size - 1)) {
 831                 fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not "
 832                     "cluster aligned; L1 entry corrupted\n", l2_offset);
 833                 errors++;
 834             }
 835
 836             /* Process and check L2 entries */
 837             ret = check_refcounts_l2(bs, refcount_table, refcount_table_size,
 838                 l2_offset, check_copied);
 839             if (ret < 0) {
 840                 goto fail;
 841             }
 842             errors += ret;
 843         }
 844     }
 845     qemu_free(l1_table);
 846     return errors;
 847
 848 fail:
 849     fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
 850     qemu_free(l1_table);
 851     return -EIO;
 852 }
 853
 854 /*
 855  * Checks an image for refcount consistency.
 856  *
 857  * Returns 0 if no errors are found, the number of errors in case the image is
 858  * detected as corrupted, and -errno when an internal error occured.
 859  */
 860 int qcow2_check_refcounts(BlockDriverState *bs)
 861 {
 862     BDRVQcowState *s = bs->opaque;
 863     int64_t size;
 864     int nb_clusters, refcount1, refcount2, i;
 865     QCowSnapshot *sn;
 866     uint16_t *refcount_table;
 867     int ret, errors = 0;
 868
 869     size = bdrv_getlength(s->hd);
 870     nb_clusters = size_to_clusters(s, size);
 871     refcount_table = qemu_mallocz(nb_clusters * sizeof(uint16_t));
 872
 873     /* header */
 874     errors += inc_refcounts(bs, refcount_table, nb_clusters,
 875                   0, s->cluster_size);
 876
 877     /* current L1 table */
 878     ret = check_refcounts_l1(bs, refcount_table, nb_clusters,
 879                        s->l1_table_offset, s->l1_size, 1);
 880     if (ret < 0) {
 881         return ret;
 882     }
 883     errors += ret;
 884
 885     /* snapshots */
 886     for(i = 0; i < s->nb_snapshots; i++) {
 887         sn = s->snapshots + i;
 888         check_refcounts_l1(bs, refcount_table, nb_clusters,
 889                            sn->l1_table_offset, sn->l1_size, 0);
 890     }
 891     errors += inc_refcounts(bs, refcount_table, nb_clusters,
 892                   s->snapshots_offset, s->snapshots_size);
 893
 894     /* refcount data */
 895     errors += inc_refcounts(bs, refcount_table, nb_clusters,
 896                   s->refcount_table_offset,
 897                   s->refcount_table_size * sizeof(uint64_t));
 898     for(i = 0; i < s->refcount_table_size; i++) {
 899         int64_t offset;
 900         offset = s->refcount_table[i];
 901         if (offset != 0) {
 902             errors += inc_refcounts(bs, refcount_table, nb_clusters,
 903                           offset, s->cluster_size);
 904         }
 905     }
 906
 907     /* compare ref counts */
 908     for(i = 0; i < nb_clusters; i++) {
 909         refcount1 = get_refcount(bs, i);
 910         refcount2 = refcount_table[i];
 911         if (refcount1 != refcount2) {
 912             fprintf(stderr, "ERROR cluster %d refcount=%d reference=%d\n",
 913                    i, refcount1, refcount2);
 914             errors++;
 915         }
 916     }
 917
 918     qemu_free(refcount_table);
 919
 920     return errors;
 921 }
 922