qcow2-refcount.c 41.5 KB
Newer Older
K
Kevin Wolf 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
/*
 * Block driver for the QCOW version 2 format
 *
 * Copyright (c) 2004-2006 Fabrice Bellard
 *
 * Permission is hereby granted, free of charge, to any person obtaining a copy
 * of this software and associated documentation files (the "Software"), to deal
 * in the Software without restriction, including without limitation the rights
 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
 * copies of the Software, and to permit persons to whom the Software is
 * furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice shall be included in
 * all copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
 * THE SOFTWARE.
 */

#include "qemu-common.h"
26
#include "block/block_int.h"
K
Kevin Wolf 已提交
27 28 29
#include "block/qcow2.h"

static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size);
30
static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
K
Kevin Wolf 已提交
31 32 33
                            int64_t offset, int64_t length,
                            int addend);

34

K
Kevin Wolf 已提交
35 36 37
/*********************************************************/
/* refcount handling */

K
Kevin Wolf 已提交
38
int qcow2_refcount_init(BlockDriverState *bs)
K
Kevin Wolf 已提交
39 40 41 42 43
{
    BDRVQcowState *s = bs->opaque;
    int ret, refcount_table_size2, i;

    refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t);
44
    s->refcount_table = g_malloc(refcount_table_size2);
K
Kevin Wolf 已提交
45
    if (s->refcount_table_size > 0) {
46 47
        BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_LOAD);
        ret = bdrv_pread(bs->file, s->refcount_table_offset,
K
Kevin Wolf 已提交
48 49 50 51 52 53 54 55 56 57 58
                         s->refcount_table, refcount_table_size2);
        if (ret != refcount_table_size2)
            goto fail;
        for(i = 0; i < s->refcount_table_size; i++)
            be64_to_cpus(&s->refcount_table[i]);
    }
    return 0;
 fail:
    return -ENOMEM;
}

K
Kevin Wolf 已提交
59
void qcow2_refcount_close(BlockDriverState *bs)
K
Kevin Wolf 已提交
60 61
{
    BDRVQcowState *s = bs->opaque;
62
    g_free(s->refcount_table);
K
Kevin Wolf 已提交
63 64 65 66
}


static int load_refcount_block(BlockDriverState *bs,
K
Kevin Wolf 已提交
67 68
                               int64_t refcount_block_offset,
                               void **refcount_block)
K
Kevin Wolf 已提交
69 70 71
{
    BDRVQcowState *s = bs->opaque;
    int ret;
72

73
    BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_LOAD);
K
Kevin Wolf 已提交
74 75
    ret = qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset,
        refcount_block);
76

K
Kevin Wolf 已提交
77
    return ret;
K
Kevin Wolf 已提交
78 79
}

80 81 82 83 84
/*
 * Returns the refcount of the cluster given by its index. Any non-negative
 * return value is the refcount of the cluster, negative values are -errno
 * and indicate an error.
 */
K
Kevin Wolf 已提交
85 86 87 88 89
static int get_refcount(BlockDriverState *bs, int64_t cluster_index)
{
    BDRVQcowState *s = bs->opaque;
    int refcount_table_index, block_index;
    int64_t refcount_block_offset;
90
    int ret;
K
Kevin Wolf 已提交
91 92
    uint16_t *refcount_block;
    uint16_t refcount;
K
Kevin Wolf 已提交
93 94 95 96 97 98 99

    refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
    if (refcount_table_index >= s->refcount_table_size)
        return 0;
    refcount_block_offset = s->refcount_table[refcount_table_index];
    if (!refcount_block_offset)
        return 0;
K
Kevin Wolf 已提交
100 101 102 103 104

    ret = qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset,
        (void**) &refcount_block);
    if (ret < 0) {
        return ret;
K
Kevin Wolf 已提交
105
    }
K
Kevin Wolf 已提交
106

K
Kevin Wolf 已提交
107 108
    block_index = cluster_index &
        ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
K
Kevin Wolf 已提交
109 110 111 112 113 114 115 116 117
    refcount = be16_to_cpu(refcount_block[block_index]);

    ret = qcow2_cache_put(bs, s->refcount_block_cache,
        (void**) &refcount_block);
    if (ret < 0) {
        return ret;
    }

    return refcount;
K
Kevin Wolf 已提交
118 119
}

120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137
/*
 * Rounds the refcount table size up to avoid growing the table for each single
 * refcount block that is allocated.
 */
static unsigned int next_refcount_table_size(BDRVQcowState *s,
    unsigned int min_size)
{
    unsigned int min_clusters = (min_size >> (s->cluster_bits - 3)) + 1;
    unsigned int refcount_table_clusters =
        MAX(1, s->refcount_table_size >> (s->cluster_bits - 3));

    while (min_clusters > refcount_table_clusters) {
        refcount_table_clusters = (refcount_table_clusters * 3 + 1) / 2;
    }

    return refcount_table_clusters << (s->cluster_bits - 3);
}

138 139 140 141 142 143 144 145 146 147 148 149 150 151 152

/* Checks if two offsets are described by the same refcount block */
static int in_same_refcount_block(BDRVQcowState *s, uint64_t offset_a,
    uint64_t offset_b)
{
    uint64_t block_a = offset_a >> (2 * s->cluster_bits - REFCOUNT_SHIFT);
    uint64_t block_b = offset_b >> (2 * s->cluster_bits - REFCOUNT_SHIFT);

    return (block_a == block_b);
}

/*
 * Loads a refcount block. If it doesn't exist yet, it is allocated first
 * (including growing the refcount table if needed).
 *
K
Kevin Wolf 已提交
153
 * Returns 0 on success or -errno in error case
154
 */
K
Kevin Wolf 已提交
155 156
static int alloc_refcount_block(BlockDriverState *bs,
    int64_t cluster_index, uint16_t **refcount_block)
K
Kevin Wolf 已提交
157 158
{
    BDRVQcowState *s = bs->opaque;
159 160 161
    unsigned int refcount_table_index;
    int ret;

162
    BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
K
Kevin Wolf 已提交
163

164 165 166 167 168 169
    /* Find the refcount block for the given cluster */
    refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);

    if (refcount_table_index < s->refcount_table_size) {

        uint64_t refcount_block_offset =
170
            s->refcount_table[refcount_table_index] & REFT_OFFSET_MASK;
171 172 173

        /* If it's already there, we're done */
        if (refcount_block_offset) {
K
Kevin Wolf 已提交
174 175
             return load_refcount_block(bs, refcount_block_offset,
                 (void**) refcount_block);
176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200
        }
    }

    /*
     * If we came here, we need to allocate something. Something is at least
     * a cluster for the new refcount block. It may also include a new refcount
     * table if the old refcount table is too small.
     *
     * Note that allocating clusters here needs some special care:
     *
     * - We can't use the normal qcow2_alloc_clusters(), it would try to
     *   increase the refcount and very likely we would end up with an endless
     *   recursion. Instead we must place the refcount blocks in a way that
     *   they can describe them themselves.
     *
     * - We need to consider that at this point we are inside update_refcounts
     *   and doing the initial refcount increase. This means that some clusters
     *   have already been allocated by the caller, but their refcount isn't
     *   accurate yet. free_cluster_index tells us where this allocation ends
     *   as long as we don't overwrite it by freeing clusters.
     *
     * - alloc_clusters_noref and qcow2_free_clusters may load a different
     *   refcount block into the cache
     */

K
Kevin Wolf 已提交
201 202 203
    *refcount_block = NULL;

    /* We write to the refcount table, so we might depend on L2 tables */
204 205 206 207
    ret = qcow2_cache_flush(bs, s->l2_table_cache);
    if (ret < 0) {
        return ret;
    }
208 209

    /* Allocate the refcount block itself and mark it as used */
210 211 212 213
    int64_t new_block = alloc_clusters_noref(bs, s->cluster_size);
    if (new_block < 0) {
        return new_block;
    }
K
Kevin Wolf 已提交
214 215

#ifdef DEBUG_ALLOC2
216 217 218
    fprintf(stderr, "qcow2: Allocate refcount block %d for %" PRIx64
        " at %" PRIx64 "\n",
        refcount_table_index, cluster_index << s->cluster_bits, new_block);
K
Kevin Wolf 已提交
219
#endif
220 221

    if (in_same_refcount_block(s, new_block, cluster_index << s->cluster_bits)) {
222
        /* Zero the new refcount block before updating it */
K
Kevin Wolf 已提交
223 224 225 226 227 228 229
        ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block,
            (void**) refcount_block);
        if (ret < 0) {
            goto fail_block;
        }

        memset(*refcount_block, 0, s->cluster_size);
230

231 232 233
        /* The block describes itself, need to update the cache */
        int block_index = (new_block >> s->cluster_bits) &
            ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
K
Kevin Wolf 已提交
234
        (*refcount_block)[block_index] = cpu_to_be16(1);
235 236 237 238 239 240 241
    } else {
        /* Described somewhere else. This can recurse at most twice before we
         * arrive at a block that describes itself. */
        ret = update_refcount(bs, new_block, s->cluster_size, 1);
        if (ret < 0) {
            goto fail_block;
        }
242

243 244 245 246
        ret = qcow2_cache_flush(bs, s->refcount_block_cache);
        if (ret < 0) {
            goto fail_block;
        }
247

248 249
        /* Initialize the new refcount block only after updating its refcount,
         * update_refcount uses the refcount cache itself */
K
Kevin Wolf 已提交
250 251 252 253 254 255 256
        ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block,
            (void**) refcount_block);
        if (ret < 0) {
            goto fail_block;
        }

        memset(*refcount_block, 0, s->cluster_size);
257 258 259
    }

    /* Now the new refcount block needs to be written to disk */
260
    BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE);
K
Kevin Wolf 已提交
261 262
    qcow2_cache_entry_mark_dirty(s->refcount_block_cache, *refcount_block);
    ret = qcow2_cache_flush(bs, s->refcount_block_cache);
263 264 265 266 267 268 269
    if (ret < 0) {
        goto fail_block;
    }

    /* If the refcount table is big enough, just hook the block up there */
    if (refcount_table_index < s->refcount_table_size) {
        uint64_t data64 = cpu_to_be64(new_block);
270
        BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_HOOKUP);
271
        ret = bdrv_pwrite_sync(bs->file,
272 273 274 275 276 277 278
            s->refcount_table_offset + refcount_table_index * sizeof(uint64_t),
            &data64, sizeof(data64));
        if (ret < 0) {
            goto fail_block;
        }

        s->refcount_table[refcount_table_index] = new_block;
K
Kevin Wolf 已提交
279 280 281 282 283 284
        return 0;
    }

    ret = qcow2_cache_put(bs, s->refcount_block_cache, (void**) refcount_block);
    if (ret < 0) {
        goto fail_block;
285 286 287 288 289 290 291 292 293 294 295 296 297
    }

    /*
     * If we come here, we need to grow the refcount table. Again, a new
     * refcount table needs some space and we can't simply allocate to avoid
     * endless recursion.
     *
     * Therefore let's grab new refcount blocks at the end of the image, which
     * will describe themselves and the new refcount table. This way we can
     * reference them only in the new table and do the switch to the new
     * refcount table at once without producing an inconsistent state in
     * between.
     */
298
    BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_GROW);
K
Kevin Wolf 已提交
299

300 301 302 303 304 305 306 307 308 309
    /* Calculate the number of refcount blocks needed so far */
    uint64_t refcount_block_clusters = 1 << (s->cluster_bits - REFCOUNT_SHIFT);
    uint64_t blocks_used = (s->free_cluster_index +
        refcount_block_clusters - 1) / refcount_block_clusters;

    /* And now we need at least one block more for the new metadata */
    uint64_t table_size = next_refcount_table_size(s, blocks_used + 1);
    uint64_t last_table_size;
    uint64_t blocks_clusters;
    do {
310 311
        uint64_t table_clusters =
            size_to_clusters(s, table_size * sizeof(uint64_t));
312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332
        blocks_clusters = 1 +
            ((table_clusters + refcount_block_clusters - 1)
            / refcount_block_clusters);
        uint64_t meta_clusters = table_clusters + blocks_clusters;

        last_table_size = table_size;
        table_size = next_refcount_table_size(s, blocks_used +
            ((meta_clusters + refcount_block_clusters - 1)
            / refcount_block_clusters));

    } while (last_table_size != table_size);

#ifdef DEBUG_ALLOC2
    fprintf(stderr, "qcow2: Grow refcount table %" PRId32 " => %" PRId64 "\n",
        s->refcount_table_size, table_size);
#endif

    /* Create the new refcount table and blocks */
    uint64_t meta_offset = (blocks_used * refcount_block_clusters) *
        s->cluster_size;
    uint64_t table_offset = meta_offset + blocks_clusters * s->cluster_size;
333 334
    uint16_t *new_blocks = g_malloc0(blocks_clusters * s->cluster_size);
    uint64_t *new_table = g_malloc0(table_size * sizeof(uint64_t));
335 336 337 338

    assert(meta_offset >= (s->free_cluster_index * s->cluster_size));

    /* Fill the new refcount table */
K
Kevin Wolf 已提交
339
    memcpy(new_table, s->refcount_table,
340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355
        s->refcount_table_size * sizeof(uint64_t));
    new_table[refcount_table_index] = new_block;

    int i;
    for (i = 0; i < blocks_clusters; i++) {
        new_table[blocks_used + i] = meta_offset + (i * s->cluster_size);
    }

    /* Fill the refcount blocks */
    uint64_t table_clusters = size_to_clusters(s, table_size * sizeof(uint64_t));
    int block = 0;
    for (i = 0; i < table_clusters + blocks_clusters; i++) {
        new_blocks[block++] = cpu_to_be16(1);
    }

    /* Write refcount blocks to disk */
356
    BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_BLOCKS);
357
    ret = bdrv_pwrite_sync(bs->file, meta_offset, new_blocks,
358
        blocks_clusters * s->cluster_size);
359
    g_free(new_blocks);
360 361 362 363 364 365 366 367 368
    if (ret < 0) {
        goto fail_table;
    }

    /* Write refcount table to disk */
    for(i = 0; i < table_size; i++) {
        cpu_to_be64s(&new_table[i]);
    }

369
    BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_TABLE);
370
    ret = bdrv_pwrite_sync(bs->file, table_offset, new_table,
371 372 373 374 375 376
        table_size * sizeof(uint64_t));
    if (ret < 0) {
        goto fail_table;
    }

    for(i = 0; i < table_size; i++) {
Z
Zhi Yong Wu 已提交
377
        be64_to_cpus(&new_table[i]);
378
    }
K
Kevin Wolf 已提交
379

380 381
    /* Hook up the new refcount table in the qcow2 header */
    uint8_t data[12];
K
Kevin Wolf 已提交
382
    cpu_to_be64w((uint64_t*)data, table_offset);
383
    cpu_to_be32w((uint32_t*)(data + 8), table_clusters);
384
    BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_SWITCH_TABLE);
385
    ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, refcount_table_offset),
386 387 388
        data, sizeof(data));
    if (ret < 0) {
        goto fail_table;
389 390
    }

391 392 393 394
    /* And switch it in memory */
    uint64_t old_table_offset = s->refcount_table_offset;
    uint64_t old_table_size = s->refcount_table_size;

395
    g_free(s->refcount_table);
K
Kevin Wolf 已提交
396
    s->refcount_table = new_table;
397
    s->refcount_table_size = table_size;
K
Kevin Wolf 已提交
398 399
    s->refcount_table_offset = table_offset;

400 401
    /* Free old table. Remember, we must not change free_cluster_index */
    uint64_t old_free_cluster_index = s->free_cluster_index;
K
Kevin Wolf 已提交
402
    qcow2_free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t));
403
    s->free_cluster_index = old_free_cluster_index;
K
Kevin Wolf 已提交
404

K
Kevin Wolf 已提交
405
    ret = load_refcount_block(bs, new_block, (void**) refcount_block);
406
    if (ret < 0) {
K
Kevin Wolf 已提交
407
        return ret;
K
Kevin Wolf 已提交
408 409
    }

410
    return 0;
K
Kevin Wolf 已提交
411

412
fail_table:
413
    g_free(new_table);
414
fail_block:
K
Kevin Wolf 已提交
415 416
    if (*refcount_block != NULL) {
        qcow2_cache_put(bs, s->refcount_block_cache, (void**) refcount_block);
417
    }
K
Kevin Wolf 已提交
418
    return ret;
419 420
}

K
Kevin Wolf 已提交
421
/* XXX: cache several refcount block clusters ? */
422 423
static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
    int64_t offset, int64_t length, int addend)
K
Kevin Wolf 已提交
424 425 426
{
    BDRVQcowState *s = bs->opaque;
    int64_t start, last, cluster_offset;
K
Kevin Wolf 已提交
427 428
    uint16_t *refcount_block = NULL;
    int64_t old_table_index = -1;
429
    int ret;
K
Kevin Wolf 已提交
430 431

#ifdef DEBUG_ALLOC2
432
    fprintf(stderr, "update_refcount: offset=%" PRId64 " size=%" PRId64 " addend=%d\n",
K
Kevin Wolf 已提交
433 434
           offset, length, addend);
#endif
K
Kevin Wolf 已提交
435
    if (length < 0) {
K
Kevin Wolf 已提交
436
        return -EINVAL;
K
Kevin Wolf 已提交
437 438 439 440
    } else if (length == 0) {
        return 0;
    }

K
Kevin Wolf 已提交
441 442 443 444 445
    if (addend < 0) {
        qcow2_cache_set_dependency(bs, s->refcount_block_cache,
            s->l2_table_cache);
    }

K
Kevin Wolf 已提交
446 447 448 449 450 451 452
    start = offset & ~(s->cluster_size - 1);
    last = (offset + length - 1) & ~(s->cluster_size - 1);
    for(cluster_offset = start; cluster_offset <= last;
        cluster_offset += s->cluster_size)
    {
        int block_index, refcount;
        int64_t cluster_index = cluster_offset >> s->cluster_bits;
K
Kevin Wolf 已提交
453 454
        int64_t table_index =
            cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
K
Kevin Wolf 已提交
455

K
Kevin Wolf 已提交
456 457 458 459 460 461 462 463 464
        /* Load the refcount block and allocate it if needed */
        if (table_index != old_table_index) {
            if (refcount_block) {
                ret = qcow2_cache_put(bs, s->refcount_block_cache,
                    (void**) &refcount_block);
                if (ret < 0) {
                    goto fail;
                }
            }
465

K
Kevin Wolf 已提交
466
            ret = alloc_refcount_block(bs, cluster_index, &refcount_block);
467
            if (ret < 0) {
K
Kevin Wolf 已提交
468
                goto fail;
K
Kevin Wolf 已提交
469 470
            }
        }
K
Kevin Wolf 已提交
471
        old_table_index = table_index;
K
Kevin Wolf 已提交
472

K
Kevin Wolf 已提交
473
        qcow2_cache_entry_mark_dirty(s->refcount_block_cache, refcount_block);
K
Kevin Wolf 已提交
474 475 476 477 478

        /* we can update the count and save it */
        block_index = cluster_index &
            ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);

K
Kevin Wolf 已提交
479
        refcount = be16_to_cpu(refcount_block[block_index]);
K
Kevin Wolf 已提交
480
        refcount += addend;
481 482 483 484
        if (refcount < 0 || refcount > 0xffff) {
            ret = -EINVAL;
            goto fail;
        }
K
Kevin Wolf 已提交
485 486 487
        if (refcount == 0 && cluster_index < s->free_cluster_index) {
            s->free_cluster_index = cluster_index;
        }
K
Kevin Wolf 已提交
488
        refcount_block[block_index] = cpu_to_be16(refcount);
K
Kevin Wolf 已提交
489 490
    }

491 492
    ret = 0;
fail:
K
Kevin Wolf 已提交
493
    /* Write last changed block to disk */
K
Kevin Wolf 已提交
494
    if (refcount_block) {
495
        int wret;
K
Kevin Wolf 已提交
496 497
        wret = qcow2_cache_put(bs, s->refcount_block_cache,
            (void**) &refcount_block);
498 499
        if (wret < 0) {
            return ret < 0 ? ret : wret;
K
Kevin Wolf 已提交
500 501 502
        }
    }

503 504 505 506 507 508 509
    /*
     * Try do undo any updates if an error is returned (This may succeed in
     * some cases like ENOSPC for allocating a new refcount block)
     */
    if (ret < 0) {
        int dummy;
        dummy = update_refcount(bs, offset, cluster_offset - offset, -addend);
B
Blue Swirl 已提交
510
        (void)dummy;
511 512 513
    }

    return ret;
K
Kevin Wolf 已提交
514 515
}

516 517 518 519 520 521 522
/*
 * Increases or decreases the refcount of a given cluster by one.
 * addend must be 1 or -1.
 *
 * If the return value is non-negative, it is the new refcount of the cluster.
 * If it is negative, it is -errno and indicates an error.
 */
K
Kevin Wolf 已提交
523 524 525 526 527 528 529 530 531 532 533 534
static int update_cluster_refcount(BlockDriverState *bs,
                                   int64_t cluster_index,
                                   int addend)
{
    BDRVQcowState *s = bs->opaque;
    int ret;

    ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend);
    if (ret < 0) {
        return ret;
    }

535 536
    bdrv_flush(bs->file);

K
Kevin Wolf 已提交
537 538 539 540 541 542 543 544 545 546 547 548 549 550
    return get_refcount(bs, cluster_index);
}



/*********************************************************/
/* cluster allocation functions */



/* return < 0 if error */
static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size)
{
    BDRVQcowState *s = bs->opaque;
551
    int i, nb_clusters, refcount;
K
Kevin Wolf 已提交
552 553 554 555

    nb_clusters = size_to_clusters(s, size);
retry:
    for(i = 0; i < nb_clusters; i++) {
556
        int64_t next_cluster_index = s->free_cluster_index++;
557 558 559 560 561
        refcount = get_refcount(bs, next_cluster_index);

        if (refcount < 0) {
            return refcount;
        } else if (refcount != 0) {
K
Kevin Wolf 已提交
562
            goto retry;
563
        }
K
Kevin Wolf 已提交
564 565
    }
#ifdef DEBUG_ALLOC2
566
    fprintf(stderr, "alloc_clusters: size=%" PRId64 " -> %" PRId64 "\n",
K
Kevin Wolf 已提交
567 568 569 570 571 572
            size,
            (s->free_cluster_index - nb_clusters) << s->cluster_bits);
#endif
    return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
}

K
Kevin Wolf 已提交
573
int64_t qcow2_alloc_clusters(BlockDriverState *bs, int64_t size)
K
Kevin Wolf 已提交
574 575
{
    int64_t offset;
576
    int ret;
K
Kevin Wolf 已提交
577

578
    BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
K
Kevin Wolf 已提交
579
    offset = alloc_clusters_noref(bs, size);
580 581 582 583
    if (offset < 0) {
        return offset;
    }

584 585 586 587
    ret = update_refcount(bs, offset, size, 1);
    if (ret < 0) {
        return ret;
    }
588

K
Kevin Wolf 已提交
589 590 591
    return offset;
}

592 593 594 595 596
int qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset,
    int nb_clusters)
{
    BDRVQcowState *s = bs->opaque;
    uint64_t cluster_index;
597
    uint64_t old_free_cluster_index;
598 599 600 601 602 603 604 605 606 607 608 609 610 611 612
    int i, refcount, ret;

    /* Check how many clusters there are free */
    cluster_index = offset >> s->cluster_bits;
    for(i = 0; i < nb_clusters; i++) {
        refcount = get_refcount(bs, cluster_index++);

        if (refcount < 0) {
            return refcount;
        } else if (refcount != 0) {
            break;
        }
    }

    /* And then allocate them */
613 614 615
    old_free_cluster_index = s->free_cluster_index;
    s->free_cluster_index = cluster_index + i;

616 617 618 619 620
    ret = update_refcount(bs, offset, i << s->cluster_bits, 1);
    if (ret < 0) {
        return ret;
    }

621 622
    s->free_cluster_index = old_free_cluster_index;

623 624 625
    return i;
}

K
Kevin Wolf 已提交
626 627
/* only used to allocate compressed sectors. We try to allocate
   contiguous sectors. size must be <= cluster_size */
K
Kevin Wolf 已提交
628
int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size)
K
Kevin Wolf 已提交
629 630 631 632 633
{
    BDRVQcowState *s = bs->opaque;
    int64_t offset, cluster_offset;
    int free_in_cluster;

634
    BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_BYTES);
K
Kevin Wolf 已提交
635 636
    assert(size > 0 && size <= s->cluster_size);
    if (s->free_byte_offset == 0) {
637 638 639
        offset = qcow2_alloc_clusters(bs, s->cluster_size);
        if (offset < 0) {
            return offset;
640
        }
641
        s->free_byte_offset = offset;
K
Kevin Wolf 已提交
642 643 644 645 646 647 648 649 650 651 652 653 654 655
    }
 redo:
    free_in_cluster = s->cluster_size -
        (s->free_byte_offset & (s->cluster_size - 1));
    if (size <= free_in_cluster) {
        /* enough space in current cluster */
        offset = s->free_byte_offset;
        s->free_byte_offset += size;
        free_in_cluster -= size;
        if (free_in_cluster == 0)
            s->free_byte_offset = 0;
        if ((offset & (s->cluster_size - 1)) != 0)
            update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
    } else {
K
Kevin Wolf 已提交
656
        offset = qcow2_alloc_clusters(bs, s->cluster_size);
657 658 659
        if (offset < 0) {
            return offset;
        }
K
Kevin Wolf 已提交
660 661 662 663 664 665 666 667 668 669 670
        cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1);
        if ((cluster_offset + s->cluster_size) == offset) {
            /* we are lucky: contiguous data */
            offset = s->free_byte_offset;
            update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
            s->free_byte_offset += size;
        } else {
            s->free_byte_offset = offset;
            goto redo;
        }
    }
671

672 673 674 675 676
    /* The cluster refcount was incremented, either by qcow2_alloc_clusters()
     * or explicitly by update_cluster_refcount().  Refcount blocks must be
     * flushed before the caller's L2 table updates.
     */
    qcow2_cache_set_dependency(bs, s->l2_table_cache, s->refcount_block_cache);
K
Kevin Wolf 已提交
677 678 679
    return offset;
}

K
Kevin Wolf 已提交
680
void qcow2_free_clusters(BlockDriverState *bs,
K
Kevin Wolf 已提交
681 682
                          int64_t offset, int64_t size)
{
683 684
    int ret;

685
    BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_FREE);
686 687 688
    ret = update_refcount(bs, offset, size, -1);
    if (ret < 0) {
        fprintf(stderr, "qcow2_free_clusters failed: %s\n", strerror(-ret));
689
        /* TODO Remember the clusters to free them later and avoid leaking */
690
    }
K
Kevin Wolf 已提交
691 692
}

693
/*
694 695
 * Free a cluster using its L2 entry (handles clusters of all types, e.g.
 * normal cluster, compressed cluster, etc.)
696
 */
K
Kevin Wolf 已提交
697
void qcow2_free_any_clusters(BlockDriverState *bs,
698
    uint64_t l2_entry, int nb_clusters)
699 700 701
{
    BDRVQcowState *s = bs->opaque;

702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717
    switch (qcow2_get_cluster_type(l2_entry)) {
    case QCOW2_CLUSTER_COMPRESSED:
        {
            int nb_csectors;
            nb_csectors = ((l2_entry >> s->csize_shift) &
                           s->csize_mask) + 1;
            qcow2_free_clusters(bs,
                (l2_entry & s->cluster_offset_mask) & ~511,
                nb_csectors * 512);
        }
        break;
    case QCOW2_CLUSTER_NORMAL:
        qcow2_free_clusters(bs, l2_entry & L2E_OFFSET_MASK,
                            nb_clusters << s->cluster_bits);
        break;
    case QCOW2_CLUSTER_UNALLOCATED:
718
    case QCOW2_CLUSTER_ZERO:
719 720 721
        break;
    default:
        abort();
722 723 724
    }
}

K
Kevin Wolf 已提交
725 726 727 728 729 730 731 732


/*********************************************************/
/* snapshots and image creation */



/* update the refcounts of snapshots and the copied flag */
K
Kevin Wolf 已提交
733 734
int qcow2_update_snapshot_refcount(BlockDriverState *bs,
    int64_t l1_table_offset, int l1_size, int addend)
K
Kevin Wolf 已提交
735 736 737 738
{
    BDRVQcowState *s = bs->opaque;
    uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated;
    int64_t old_offset, old_l2_offset;
739
    int i, j, l1_modified = 0, nb_csectors, refcount;
K
Kevin Wolf 已提交
740
    int ret;
K
Kevin Wolf 已提交
741 742 743 744

    l2_table = NULL;
    l1_table = NULL;
    l1_size2 = l1_size * sizeof(uint64_t);
745 746 747 748

    /* WARNING: qcow2_snapshot_goto relies on this function not using the
     * l1_table_offset when it is the current s->l1_table_offset! Be careful
     * when changing this! */
K
Kevin Wolf 已提交
749
    if (l1_table_offset != s->l1_table_offset) {
750
        l1_table = g_malloc0(align_offset(l1_size2, 512));
K
Kevin Wolf 已提交
751
        l1_allocated = 1;
752
        if (bdrv_pread(bs->file, l1_table_offset,
K
Kevin Wolf 已提交
753
                       l1_table, l1_size2) != l1_size2)
754 755
        {
            ret = -EIO;
K
Kevin Wolf 已提交
756
            goto fail;
757 758
        }

K
Kevin Wolf 已提交
759 760 761 762 763 764 765 766 767 768 769 770
        for(i = 0;i < l1_size; i++)
            be64_to_cpus(&l1_table[i]);
    } else {
        assert(l1_size == s->l1_size);
        l1_table = s->l1_table;
        l1_allocated = 0;
    }

    for(i = 0; i < l1_size; i++) {
        l2_offset = l1_table[i];
        if (l2_offset) {
            old_l2_offset = l2_offset;
771
            l2_offset &= L1E_OFFSET_MASK;
K
Kevin Wolf 已提交
772 773 774 775

            ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset,
                (void**) &l2_table);
            if (ret < 0) {
K
Kevin Wolf 已提交
776
                goto fail;
K
Kevin Wolf 已提交
777 778
            }

K
Kevin Wolf 已提交
779 780 781 782 783 784 785 786
            for(j = 0; j < s->l2_size; j++) {
                offset = be64_to_cpu(l2_table[j]);
                if (offset != 0) {
                    old_offset = offset;
                    offset &= ~QCOW_OFLAG_COPIED;
                    if (offset & QCOW_OFLAG_COMPRESSED) {
                        nb_csectors = ((offset >> s->csize_shift) &
                                       s->csize_mask) + 1;
787 788 789 790 791 792 793 794
                        if (addend != 0) {
                            int ret;
                            ret = update_refcount(bs,
                                (offset & s->cluster_offset_mask) & ~511,
                                nb_csectors * 512, addend);
                            if (ret < 0) {
                                goto fail;
                            }
795 796 797 798

                            /* TODO Flushing once for the whole function should
                             * be enough */
                            bdrv_flush(bs->file);
799
                        }
K
Kevin Wolf 已提交
800 801 802
                        /* compressed clusters are never modified */
                        refcount = 2;
                    } else {
803
                        uint64_t cluster_index = (offset & L2E_OFFSET_MASK) >> s->cluster_bits;
K
Kevin Wolf 已提交
804
                        if (addend != 0) {
805
                            refcount = update_cluster_refcount(bs, cluster_index, addend);
K
Kevin Wolf 已提交
806
                        } else {
807
                            refcount = get_refcount(bs, cluster_index);
K
Kevin Wolf 已提交
808
                        }
809 810

                        if (refcount < 0) {
811
                            ret = -EIO;
812 813
                            goto fail;
                        }
K
Kevin Wolf 已提交
814 815 816 817 818 819
                    }

                    if (refcount == 1) {
                        offset |= QCOW_OFLAG_COPIED;
                    }
                    if (offset != old_offset) {
K
Kevin Wolf 已提交
820 821 822 823
                        if (addend > 0) {
                            qcow2_cache_set_dependency(bs, s->l2_table_cache,
                                s->refcount_block_cache);
                        }
K
Kevin Wolf 已提交
824
                        l2_table[j] = cpu_to_be64(offset);
K
Kevin Wolf 已提交
825
                        qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
K
Kevin Wolf 已提交
826 827 828
                    }
                }
            }
K
Kevin Wolf 已提交
829 830 831 832

            ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
            if (ret < 0) {
                goto fail;
K
Kevin Wolf 已提交
833 834
            }

K
Kevin Wolf 已提交
835

K
Kevin Wolf 已提交
836 837 838 839 840
            if (addend != 0) {
                refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend);
            } else {
                refcount = get_refcount(bs, l2_offset >> s->cluster_bits);
            }
841
            if (refcount < 0) {
842
                ret = -EIO;
843 844
                goto fail;
            } else if (refcount == 1) {
K
Kevin Wolf 已提交
845 846 847 848 849 850 851 852
                l2_offset |= QCOW_OFLAG_COPIED;
            }
            if (l2_offset != old_l2_offset) {
                l1_table[i] = l2_offset;
                l1_modified = 1;
            }
        }
    }
853 854 855 856 857 858 859

    ret = 0;
fail:
    if (l2_table) {
        qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
    }

860 861
    /* Update L1 only if it isn't deleted anyway (addend = -1) */
    if (addend >= 0 && l1_modified) {
K
Kevin Wolf 已提交
862 863
        for(i = 0; i < l1_size; i++)
            cpu_to_be64s(&l1_table[i]);
864 865
        if (bdrv_pwrite_sync(bs->file, l1_table_offset, l1_table,
                        l1_size2) < 0)
K
Kevin Wolf 已提交
866 867 868 869 870
            goto fail;
        for(i = 0; i < l1_size; i++)
            be64_to_cpus(&l1_table[i]);
    }
    if (l1_allocated)
871
        g_free(l1_table);
872
    return ret;
K
Kevin Wolf 已提交
873 874 875 876 877 878 879 880 881 882 883 884 885 886 887
}




/*********************************************************/
/* refcount checking functions */



/*
 * Increases the refcount for a range of clusters in a given refcount table.
 * This is used to construct a temporary refcount table out of L1 and L2 tables
 * which can be compared the the refcount table saved in the image.
 *
888
 * Modifies the number of errors in res.
K
Kevin Wolf 已提交
889
 */
890 891
static void inc_refcounts(BlockDriverState *bs,
                          BdrvCheckResult *res,
K
Kevin Wolf 已提交
892 893 894 895 896 897 898 899 900
                          uint16_t *refcount_table,
                          int refcount_table_size,
                          int64_t offset, int64_t size)
{
    BDRVQcowState *s = bs->opaque;
    int64_t start, last, cluster_offset;
    int k;

    if (size <= 0)
901
        return;
K
Kevin Wolf 已提交
902 903 904 905 906 907

    start = offset & ~(s->cluster_size - 1);
    last = (offset + size - 1) & ~(s->cluster_size - 1);
    for(cluster_offset = start; cluster_offset <= last;
        cluster_offset += s->cluster_size) {
        k = cluster_offset >> s->cluster_bits;
908
        if (k < 0) {
K
Kevin Wolf 已提交
909 910
            fprintf(stderr, "ERROR: invalid cluster offset=0x%" PRIx64 "\n",
                cluster_offset);
911 912 913 914 915 916
            res->corruptions++;
        } else if (k >= refcount_table_size) {
            fprintf(stderr, "Warning: cluster offset=0x%" PRIx64 " is after "
                "the end of the image file, can't properly check refcounts.\n",
                cluster_offset);
            res->check_errors++;
K
Kevin Wolf 已提交
917 918 919 920
        } else {
            if (++refcount_table[k] == 0) {
                fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64
                    "\n", cluster_offset);
921
                res->corruptions++;
K
Kevin Wolf 已提交
922 923 924 925 926
            }
        }
    }
}

927 928 929
/* Flags for check_refcounts_l1() and check_refcounts_l2() */
enum {
    CHECK_OFLAG_COPIED = 0x1,   /* check QCOW_OFLAG_COPIED matches refcount */
930
    CHECK_FRAG_INFO = 0x2,      /* update BlockFragInfo counters */
931 932
};

K
Kevin Wolf 已提交
933 934 935 936 937 938 939 940
/*
 * Increases the refcount in the given refcount table for the all clusters
 * referenced in the L2 table. While doing so, performs some checks on L2
 * entries.
 *
 * Returns the number of errors found by the checks or -errno if an internal
 * error occurred.
 */
941
static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res,
K
Kevin Wolf 已提交
942
    uint16_t *refcount_table, int refcount_table_size, int64_t l2_offset,
943
    int flags)
K
Kevin Wolf 已提交
944 945
{
    BDRVQcowState *s = bs->opaque;
946
    uint64_t *l2_table, l2_entry;
947
    uint64_t next_contiguous_offset = 0;
K
Kevin Wolf 已提交
948 949 950 951
    int i, l2_size, nb_csectors, refcount;

    /* Read L2 table from disk */
    l2_size = s->l2_size * sizeof(uint64_t);
952
    l2_table = g_malloc(l2_size);
K
Kevin Wolf 已提交
953

954
    if (bdrv_pread(bs->file, l2_offset, l2_table, l2_size) != l2_size)
K
Kevin Wolf 已提交
955 956 957 958
        goto fail;

    /* Do the actual checks */
    for(i = 0; i < s->l2_size; i++) {
959 960 961 962 963 964 965 966 967 968 969 970
        l2_entry = be64_to_cpu(l2_table[i]);

        switch (qcow2_get_cluster_type(l2_entry)) {
        case QCOW2_CLUSTER_COMPRESSED:
            /* Compressed clusters don't have QCOW_OFLAG_COPIED */
            if (l2_entry & QCOW_OFLAG_COPIED) {
                fprintf(stderr, "ERROR: cluster %" PRId64 ": "
                    "copied flag must never be set for compressed "
                    "clusters\n", l2_entry >> s->cluster_bits);
                l2_entry &= ~QCOW_OFLAG_COPIED;
                res->corruptions++;
            }
K
Kevin Wolf 已提交
971

972 973 974 975 976 977
            /* Mark cluster as used */
            nb_csectors = ((l2_entry >> s->csize_shift) &
                           s->csize_mask) + 1;
            l2_entry &= s->cluster_offset_mask;
            inc_refcounts(bs, res, refcount_table, refcount_table_size,
                l2_entry & ~511, nb_csectors * 512);
978 979 980

            if (flags & CHECK_FRAG_INFO) {
                res->bfi.allocated_clusters++;
981
                res->bfi.compressed_clusters++;
982 983 984 985 986 987 988 989

                /* Compressed clusters are fragmented by nature.  Since they
                 * take up sub-sector space but we only have sector granularity
                 * I/O we need to re-read the same sectors even for adjacent
                 * compressed clusters.
                 */
                res->bfi.fragmented_clusters++;
            }
990
            break;
K
Kevin Wolf 已提交
991

992 993 994 995 996 997
        case QCOW2_CLUSTER_ZERO:
            if ((l2_entry & L2E_OFFSET_MASK) == 0) {
                break;
            }
            /* fall through */

998 999 1000 1001
        case QCOW2_CLUSTER_NORMAL:
        {
            /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
            uint64_t offset = l2_entry & L2E_OFFSET_MASK;
K
Kevin Wolf 已提交
1002

1003
            if (flags & CHECK_OFLAG_COPIED) {
1004 1005 1006 1007 1008 1009 1010 1011 1012
                refcount = get_refcount(bs, offset >> s->cluster_bits);
                if (refcount < 0) {
                    fprintf(stderr, "Can't get refcount for offset %"
                        PRIx64 ": %s\n", l2_entry, strerror(-refcount));
                    goto fail;
                }
                if ((refcount == 1) != ((l2_entry & QCOW_OFLAG_COPIED) != 0)) {
                    fprintf(stderr, "ERROR OFLAG_COPIED: offset=%"
                        PRIx64 " refcount=%d\n", l2_entry, refcount);
1013
                    res->corruptions++;
K
Kevin Wolf 已提交
1014 1015
                }
            }
1016

1017 1018 1019 1020 1021 1022 1023 1024 1025
            if (flags & CHECK_FRAG_INFO) {
                res->bfi.allocated_clusters++;
                if (next_contiguous_offset &&
                    offset != next_contiguous_offset) {
                    res->bfi.fragmented_clusters++;
                }
                next_contiguous_offset = offset + s->cluster_size;
            }

1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043
            /* Mark cluster as used */
            inc_refcounts(bs, res, refcount_table,refcount_table_size,
                offset, s->cluster_size);

            /* Correct offsets are cluster aligned */
            if (offset & (s->cluster_size - 1)) {
                fprintf(stderr, "ERROR offset=%" PRIx64 ": Cluster is not "
                    "properly aligned; L2 entry corrupted.\n", offset);
                res->corruptions++;
            }
            break;
        }

        case QCOW2_CLUSTER_UNALLOCATED:
            break;

        default:
            abort();
K
Kevin Wolf 已提交
1044 1045 1046
        }
    }

1047
    g_free(l2_table);
1048
    return 0;
K
Kevin Wolf 已提交
1049 1050

fail:
1051
    fprintf(stderr, "ERROR: I/O error in check_refcounts_l2\n");
1052
    g_free(l2_table);
K
Kevin Wolf 已提交
1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064
    return -EIO;
}

/*
 * Increases the refcount for the L1 table, its L2 tables and all referenced
 * clusters in the given refcount table. While doing so, performs some checks
 * on L1 and L2 entries.
 *
 * Returns the number of errors found by the checks or -errno if an internal
 * error occurred.
 */
static int check_refcounts_l1(BlockDriverState *bs,
1065
                              BdrvCheckResult *res,
K
Kevin Wolf 已提交
1066 1067 1068
                              uint16_t *refcount_table,
                              int refcount_table_size,
                              int64_t l1_table_offset, int l1_size,
1069
                              int flags)
K
Kevin Wolf 已提交
1070 1071 1072 1073 1074 1075 1076 1077
{
    BDRVQcowState *s = bs->opaque;
    uint64_t *l1_table, l2_offset, l1_size2;
    int i, refcount, ret;

    l1_size2 = l1_size * sizeof(uint64_t);

    /* Mark L1 table as used */
1078 1079
    inc_refcounts(bs, res, refcount_table, refcount_table_size,
        l1_table_offset, l1_size2);
K
Kevin Wolf 已提交
1080 1081

    /* Read L1 table entries from disk */
1082 1083 1084
    if (l1_size2 == 0) {
        l1_table = NULL;
    } else {
1085
        l1_table = g_malloc(l1_size2);
1086
        if (bdrv_pread(bs->file, l1_table_offset,
1087 1088 1089 1090 1091
                       l1_table, l1_size2) != l1_size2)
            goto fail;
        for(i = 0;i < l1_size; i++)
            be64_to_cpus(&l1_table[i]);
    }
K
Kevin Wolf 已提交
1092 1093 1094 1095 1096 1097

    /* Do the actual checks */
    for(i = 0; i < l1_size; i++) {
        l2_offset = l1_table[i];
        if (l2_offset) {
            /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
1098
            if (flags & CHECK_OFLAG_COPIED) {
K
Kevin Wolf 已提交
1099 1100
                refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED)
                    >> s->cluster_bits);
1101 1102 1103
                if (refcount < 0) {
                    fprintf(stderr, "Can't get refcount for l2_offset %"
                        PRIx64 ": %s\n", l2_offset, strerror(-refcount));
1104
                    goto fail;
1105
                }
K
Kevin Wolf 已提交
1106 1107 1108
                if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) {
                    fprintf(stderr, "ERROR OFLAG_COPIED: l2_offset=%" PRIx64
                        " refcount=%d\n", l2_offset, refcount);
1109
                    res->corruptions++;
K
Kevin Wolf 已提交
1110 1111 1112 1113
                }
            }

            /* Mark L2 table as used */
1114
            l2_offset &= L1E_OFFSET_MASK;
1115 1116
            inc_refcounts(bs, res, refcount_table, refcount_table_size,
                l2_offset, s->cluster_size);
K
Kevin Wolf 已提交
1117 1118 1119 1120 1121

            /* L2 tables are cluster aligned */
            if (l2_offset & (s->cluster_size - 1)) {
                fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not "
                    "cluster aligned; L1 entry corrupted\n", l2_offset);
1122
                res->corruptions++;
K
Kevin Wolf 已提交
1123 1124 1125
            }

            /* Process and check L2 entries */
1126
            ret = check_refcounts_l2(bs, res, refcount_table,
1127
                                     refcount_table_size, l2_offset, flags);
K
Kevin Wolf 已提交
1128 1129 1130 1131 1132
            if (ret < 0) {
                goto fail;
            }
        }
    }
1133
    g_free(l1_table);
1134
    return 0;
K
Kevin Wolf 已提交
1135 1136 1137

fail:
    fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
1138
    res->check_errors++;
1139
    g_free(l1_table);
K
Kevin Wolf 已提交
1140 1141 1142 1143 1144 1145 1146
    return -EIO;
}

/*
 * Checks an image for refcount consistency.
 *
 * Returns 0 if no errors are found, the number of errors in case the image is
1147
 * detected as corrupted, and -errno when an internal error occurred.
K
Kevin Wolf 已提交
1148
 */
1149 1150
int qcow2_check_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
                          BdrvCheckMode fix)
K
Kevin Wolf 已提交
1151 1152
{
    BDRVQcowState *s = bs->opaque;
1153
    int64_t size, i, highest_cluster;
1154
    int nb_clusters, refcount1, refcount2;
K
Kevin Wolf 已提交
1155 1156
    QCowSnapshot *sn;
    uint16_t *refcount_table;
1157
    int ret;
K
Kevin Wolf 已提交
1158

1159
    size = bdrv_getlength(bs->file);
K
Kevin Wolf 已提交
1160
    nb_clusters = size_to_clusters(s, size);
1161
    res->bfi.total_clusters = nb_clusters;
1162
    refcount_table = g_malloc0(nb_clusters * sizeof(uint16_t));
K
Kevin Wolf 已提交
1163 1164

    /* header */
1165 1166
    inc_refcounts(bs, res, refcount_table, nb_clusters,
        0, s->cluster_size);
K
Kevin Wolf 已提交
1167 1168

    /* current L1 table */
1169
    ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
1170
                             s->l1_table_offset, s->l1_size,
1171
                             CHECK_OFLAG_COPIED | CHECK_FRAG_INFO);
K
Kevin Wolf 已提交
1172
    if (ret < 0) {
1173
        goto fail;
K
Kevin Wolf 已提交
1174 1175 1176 1177 1178
    }

    /* snapshots */
    for(i = 0; i < s->nb_snapshots; i++) {
        sn = s->snapshots + i;
1179 1180 1181
        ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
            sn->l1_table_offset, sn->l1_size, 0);
        if (ret < 0) {
1182
            goto fail;
1183
        }
K
Kevin Wolf 已提交
1184
    }
1185 1186
    inc_refcounts(bs, res, refcount_table, nb_clusters,
        s->snapshots_offset, s->snapshots_size);
K
Kevin Wolf 已提交
1187 1188

    /* refcount data */
1189 1190 1191 1192
    inc_refcounts(bs, res, refcount_table, nb_clusters,
        s->refcount_table_offset,
        s->refcount_table_size * sizeof(uint64_t));

K
Kevin Wolf 已提交
1193
    for(i = 0; i < s->refcount_table_size; i++) {
1194
        uint64_t offset, cluster;
K
Kevin Wolf 已提交
1195
        offset = s->refcount_table[i];
1196
        cluster = offset >> s->cluster_bits;
1197 1198 1199

        /* Refcount blocks are cluster aligned */
        if (offset & (s->cluster_size - 1)) {
1200
            fprintf(stderr, "ERROR refcount block %" PRId64 " is not "
1201
                "cluster aligned; refcount table entry corrupted\n", i);
1202
            res->corruptions++;
1203 1204 1205 1206
            continue;
        }

        if (cluster >= nb_clusters) {
1207 1208
            fprintf(stderr, "ERROR refcount block %" PRId64
                    " is outside image\n", i);
1209
            res->corruptions++;
1210
            continue;
1211 1212
        }

K
Kevin Wolf 已提交
1213
        if (offset != 0) {
1214 1215
            inc_refcounts(bs, res, refcount_table, nb_clusters,
                offset, s->cluster_size);
1216
            if (refcount_table[cluster] != 1) {
1217 1218
                fprintf(stderr, "ERROR refcount block %" PRId64
                    " refcount=%d\n",
1219
                    i, refcount_table[cluster]);
1220
                res->corruptions++;
1221
            }
K
Kevin Wolf 已提交
1222 1223 1224 1225
        }
    }

    /* compare ref counts */
1226
    for (i = 0, highest_cluster = 0; i < nb_clusters; i++) {
K
Kevin Wolf 已提交
1227
        refcount1 = get_refcount(bs, i);
1228
        if (refcount1 < 0) {
1229
            fprintf(stderr, "Can't get refcount for cluster %" PRId64 ": %s\n",
1230
                i, strerror(-refcount1));
1231
            res->check_errors++;
1232
            continue;
1233 1234
        }

K
Kevin Wolf 已提交
1235
        refcount2 = refcount_table[i];
1236 1237 1238 1239 1240

        if (refcount1 > 0 || refcount2 > 0) {
            highest_cluster = i;
        }

K
Kevin Wolf 已提交
1241
        if (refcount1 != refcount2) {
1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254

            /* Check if we're allowed to fix the mismatch */
            int *num_fixed = NULL;
            if (refcount1 > refcount2 && (fix & BDRV_FIX_LEAKS)) {
                num_fixed = &res->leaks_fixed;
            } else if (refcount1 < refcount2 && (fix & BDRV_FIX_ERRORS)) {
                num_fixed = &res->corruptions_fixed;
            }

            fprintf(stderr, "%s cluster %" PRId64 " refcount=%d reference=%d\n",
                   num_fixed != NULL     ? "Repairing" :
                   refcount1 < refcount2 ? "ERROR" :
                                           "Leaked",
K
Kevin Wolf 已提交
1255
                   i, refcount1, refcount2);
1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266

            if (num_fixed) {
                ret = update_refcount(bs, i << s->cluster_bits, 1,
                                      refcount2 - refcount1);
                if (ret >= 0) {
                    (*num_fixed)++;
                    continue;
                }
            }

            /* And if we couldn't, print an error */
1267 1268 1269 1270 1271
            if (refcount1 < refcount2) {
                res->corruptions++;
            } else {
                res->leaks++;
            }
K
Kevin Wolf 已提交
1272 1273 1274
        }
    }

1275
    res->image_end_offset = (highest_cluster + 1) * s->cluster_size;
1276 1277 1278
    ret = 0;

fail:
1279
    g_free(refcount_table);
K
Kevin Wolf 已提交
1280

1281
    return ret;
K
Kevin Wolf 已提交
1282 1283
}