qcow2-refcount.c 58.7 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
#include "block/qcow2.h"
M
Max Reitz 已提交
28 29
#include "qemu/range.h"
#include "qapi/qmp/types.h"
K
Kevin Wolf 已提交
30 31

static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size);
32
static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
K
Kevin Wolf 已提交
33
                            int64_t offset, int64_t length,
34
                            int addend, enum qcow2_discard_type type);
K
Kevin Wolf 已提交
35

36

K
Kevin Wolf 已提交
37 38 39
/*********************************************************/
/* refcount handling */

K
Kevin Wolf 已提交
40
int qcow2_refcount_init(BlockDriverState *bs)
K
Kevin Wolf 已提交
41 42 43 44 45
{
    BDRVQcowState *s = bs->opaque;
    int ret, refcount_table_size2, i;

    refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t);
46
    s->refcount_table = g_malloc(refcount_table_size2);
K
Kevin Wolf 已提交
47
    if (s->refcount_table_size > 0) {
48 49
        BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_LOAD);
        ret = bdrv_pread(bs->file, s->refcount_table_offset,
K
Kevin Wolf 已提交
50 51 52 53 54 55 56 57 58 59 60
                         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
61
void qcow2_refcount_close(BlockDriverState *bs)
K
Kevin Wolf 已提交
62 63
{
    BDRVQcowState *s = bs->opaque;
64
    g_free(s->refcount_table);
K
Kevin Wolf 已提交
65 66 67 68
}


static int load_refcount_block(BlockDriverState *bs,
K
Kevin Wolf 已提交
69 70
                               int64_t refcount_block_offset,
                               void **refcount_block)
K
Kevin Wolf 已提交
71 72 73
{
    BDRVQcowState *s = bs->opaque;
    int ret;
74

75
    BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_LOAD);
K
Kevin Wolf 已提交
76 77
    ret = qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset,
        refcount_block);
78

K
Kevin Wolf 已提交
79
    return ret;
K
Kevin Wolf 已提交
80 81
}

82 83 84 85 86
/*
 * 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
87 88 89 90 91
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;
92
    int ret;
K
Kevin Wolf 已提交
93 94
    uint16_t *refcount_block;
    uint16_t refcount;
K
Kevin Wolf 已提交
95 96 97 98 99 100 101

    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
102 103 104 105 106

    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
107
    }
K
Kevin Wolf 已提交
108

K
Kevin Wolf 已提交
109 110
    block_index = cluster_index &
        ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
K
Kevin Wolf 已提交
111 112 113 114 115 116 117 118 119
    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
120 121
}

122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139
/*
 * 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);
}

140 141 142 143 144 145 146 147 148 149 150 151 152 153 154

/* 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
155
 * Returns 0 on success or -errno in error case
156
 */
K
Kevin Wolf 已提交
157 158
static int alloc_refcount_block(BlockDriverState *bs,
    int64_t cluster_index, uint16_t **refcount_block)
K
Kevin Wolf 已提交
159 160
{
    BDRVQcowState *s = bs->opaque;
161 162 163
    unsigned int refcount_table_index;
    int ret;

164
    BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
K
Kevin Wolf 已提交
165

166 167 168 169 170 171
    /* 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 =
172
            s->refcount_table[refcount_table_index] & REFT_OFFSET_MASK;
173 174 175

        /* If it's already there, we're done */
        if (refcount_block_offset) {
K
Kevin Wolf 已提交
176 177
             return load_refcount_block(bs, refcount_block_offset,
                 (void**) refcount_block);
178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202
        }
    }

    /*
     * 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
203 204 205
    *refcount_block = NULL;

    /* We write to the refcount table, so we might depend on L2 tables */
206 207 208 209
    ret = qcow2_cache_flush(bs, s->l2_table_cache);
    if (ret < 0) {
        return ret;
    }
210 211

    /* Allocate the refcount block itself and mark it as used */
212 213 214 215
    int64_t new_block = alloc_clusters_noref(bs, s->cluster_size);
    if (new_block < 0) {
        return new_block;
    }
K
Kevin Wolf 已提交
216 217

#ifdef DEBUG_ALLOC2
218 219 220
    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
221
#endif
222 223

    if (in_same_refcount_block(s, new_block, cluster_index << s->cluster_bits)) {
224
        /* Zero the new refcount block before updating it */
K
Kevin Wolf 已提交
225 226 227 228 229 230 231
        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);
232

233 234 235
        /* 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
236
        (*refcount_block)[block_index] = cpu_to_be16(1);
237 238 239
    } else {
        /* Described somewhere else. This can recurse at most twice before we
         * arrive at a block that describes itself. */
240 241
        ret = update_refcount(bs, new_block, s->cluster_size, 1,
                              QCOW2_DISCARD_NEVER);
242 243 244
        if (ret < 0) {
            goto fail_block;
        }
245

246 247 248 249
        ret = qcow2_cache_flush(bs, s->refcount_block_cache);
        if (ret < 0) {
            goto fail_block;
        }
250

251 252
        /* Initialize the new refcount block only after updating its refcount,
         * update_refcount uses the refcount cache itself */
K
Kevin Wolf 已提交
253 254 255 256 257 258 259
        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);
260 261 262
    }

    /* Now the new refcount block needs to be written to disk */
263
    BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE);
K
Kevin Wolf 已提交
264 265
    qcow2_cache_entry_mark_dirty(s->refcount_block_cache, *refcount_block);
    ret = qcow2_cache_flush(bs, s->refcount_block_cache);
266 267 268 269 270 271 272
    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);
273
        BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_HOOKUP);
274
        ret = bdrv_pwrite_sync(bs->file,
275 276 277 278 279 280 281
            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
282 283 284 285 286 287
        return 0;
    }

    ret = qcow2_cache_put(bs, s->refcount_block_cache, (void**) refcount_block);
    if (ret < 0) {
        goto fail_block;
288 289 290 291 292 293 294 295 296 297 298 299 300
    }

    /*
     * 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.
     */
301
    BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_GROW);
K
Kevin Wolf 已提交
302

303 304 305 306 307 308 309 310 311 312
    /* 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 {
313 314
        uint64_t table_clusters =
            size_to_clusters(s, table_size * sizeof(uint64_t));
315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335
        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;
336 337
    uint16_t *new_blocks = g_malloc0(blocks_clusters * s->cluster_size);
    uint64_t *new_table = g_malloc0(table_size * sizeof(uint64_t));
338 339 340 341

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

    /* Fill the new refcount table */
K
Kevin Wolf 已提交
342
    memcpy(new_table, s->refcount_table,
343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358
        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 */
359
    BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_BLOCKS);
360
    ret = bdrv_pwrite_sync(bs->file, meta_offset, new_blocks,
361
        blocks_clusters * s->cluster_size);
362
    g_free(new_blocks);
363 364 365 366 367 368 369 370 371
    if (ret < 0) {
        goto fail_table;
    }

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

372
    BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_TABLE);
373
    ret = bdrv_pwrite_sync(bs->file, table_offset, new_table,
374 375 376 377 378 379
        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
380
        be64_to_cpus(&new_table[i]);
381
    }
K
Kevin Wolf 已提交
382

383 384
    /* Hook up the new refcount table in the qcow2 header */
    uint8_t data[12];
K
Kevin Wolf 已提交
385
    cpu_to_be64w((uint64_t*)data, table_offset);
386
    cpu_to_be32w((uint32_t*)(data + 8), table_clusters);
387
    BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_SWITCH_TABLE);
388
    ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, refcount_table_offset),
389 390 391
        data, sizeof(data));
    if (ret < 0) {
        goto fail_table;
392 393
    }

394 395 396 397
    /* And switch it in memory */
    uint64_t old_table_offset = s->refcount_table_offset;
    uint64_t old_table_size = s->refcount_table_size;

398
    g_free(s->refcount_table);
K
Kevin Wolf 已提交
399
    s->refcount_table = new_table;
400
    s->refcount_table_size = table_size;
K
Kevin Wolf 已提交
401 402
    s->refcount_table_offset = table_offset;

403 404
    /* Free old table. Remember, we must not change free_cluster_index */
    uint64_t old_free_cluster_index = s->free_cluster_index;
405 406
    qcow2_free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t),
                        QCOW2_DISCARD_OTHER);
407
    s->free_cluster_index = old_free_cluster_index;
K
Kevin Wolf 已提交
408

K
Kevin Wolf 已提交
409
    ret = load_refcount_block(bs, new_block, (void**) refcount_block);
410
    if (ret < 0) {
K
Kevin Wolf 已提交
411
        return ret;
K
Kevin Wolf 已提交
412 413
    }

414
    return 0;
K
Kevin Wolf 已提交
415

416
fail_table:
417
    g_free(new_table);
418
fail_block:
K
Kevin Wolf 已提交
419 420
    if (*refcount_block != NULL) {
        qcow2_cache_put(bs, s->refcount_block_cache, (void**) refcount_block);
421
    }
K
Kevin Wolf 已提交
422
    return ret;
423 424
}

K
Kevin Wolf 已提交
425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492
void qcow2_process_discards(BlockDriverState *bs, int ret)
{
    BDRVQcowState *s = bs->opaque;
    Qcow2DiscardRegion *d, *next;

    QTAILQ_FOREACH_SAFE(d, &s->discards, next, next) {
        QTAILQ_REMOVE(&s->discards, d, next);

        /* Discard is optional, ignore the return value */
        if (ret >= 0) {
            bdrv_discard(bs->file,
                         d->offset >> BDRV_SECTOR_BITS,
                         d->bytes >> BDRV_SECTOR_BITS);
        }

        g_free(d);
    }
}

static void update_refcount_discard(BlockDriverState *bs,
                                    uint64_t offset, uint64_t length)
{
    BDRVQcowState *s = bs->opaque;
    Qcow2DiscardRegion *d, *p, *next;

    QTAILQ_FOREACH(d, &s->discards, next) {
        uint64_t new_start = MIN(offset, d->offset);
        uint64_t new_end = MAX(offset + length, d->offset + d->bytes);

        if (new_end - new_start <= length + d->bytes) {
            /* There can't be any overlap, areas ending up here have no
             * references any more and therefore shouldn't get freed another
             * time. */
            assert(d->bytes + length == new_end - new_start);
            d->offset = new_start;
            d->bytes = new_end - new_start;
            goto found;
        }
    }

    d = g_malloc(sizeof(*d));
    *d = (Qcow2DiscardRegion) {
        .bs     = bs,
        .offset = offset,
        .bytes  = length,
    };
    QTAILQ_INSERT_TAIL(&s->discards, d, next);

found:
    /* Merge discard requests if they are adjacent now */
    QTAILQ_FOREACH_SAFE(p, &s->discards, next, next) {
        if (p == d
            || p->offset > d->offset + d->bytes
            || d->offset > p->offset + p->bytes)
        {
            continue;
        }

        /* Still no overlap possible */
        assert(p->offset == d->offset + d->bytes
            || d->offset == p->offset + p->bytes);

        QTAILQ_REMOVE(&s->discards, p, next);
        d->offset = MIN(d->offset, p->offset);
        d->bytes += p->bytes;
    }
}

K
Kevin Wolf 已提交
493
/* XXX: cache several refcount block clusters ? */
494
static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
495
    int64_t offset, int64_t length, int addend, enum qcow2_discard_type type)
K
Kevin Wolf 已提交
496 497 498
{
    BDRVQcowState *s = bs->opaque;
    int64_t start, last, cluster_offset;
K
Kevin Wolf 已提交
499 500
    uint16_t *refcount_block = NULL;
    int64_t old_table_index = -1;
501
    int ret;
K
Kevin Wolf 已提交
502 503

#ifdef DEBUG_ALLOC2
504
    fprintf(stderr, "update_refcount: offset=%" PRId64 " size=%" PRId64 " addend=%d\n",
K
Kevin Wolf 已提交
505 506
           offset, length, addend);
#endif
K
Kevin Wolf 已提交
507
    if (length < 0) {
K
Kevin Wolf 已提交
508
        return -EINVAL;
K
Kevin Wolf 已提交
509 510 511 512
    } else if (length == 0) {
        return 0;
    }

K
Kevin Wolf 已提交
513 514 515 516 517
    if (addend < 0) {
        qcow2_cache_set_dependency(bs, s->refcount_block_cache,
            s->l2_table_cache);
    }

K
Kevin Wolf 已提交
518 519 520 521 522 523 524
    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
525 526
        int64_t table_index =
            cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
K
Kevin Wolf 已提交
527

K
Kevin Wolf 已提交
528 529 530 531 532 533 534 535 536
        /* 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;
                }
            }
537

K
Kevin Wolf 已提交
538
            ret = alloc_refcount_block(bs, cluster_index, &refcount_block);
539
            if (ret < 0) {
K
Kevin Wolf 已提交
540
                goto fail;
K
Kevin Wolf 已提交
541 542
            }
        }
K
Kevin Wolf 已提交
543
        old_table_index = table_index;
K
Kevin Wolf 已提交
544

K
Kevin Wolf 已提交
545
        qcow2_cache_entry_mark_dirty(s->refcount_block_cache, refcount_block);
K
Kevin Wolf 已提交
546 547 548 549 550

        /* 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
551
        refcount = be16_to_cpu(refcount_block[block_index]);
K
Kevin Wolf 已提交
552
        refcount += addend;
553 554 555 556
        if (refcount < 0 || refcount > 0xffff) {
            ret = -EINVAL;
            goto fail;
        }
K
Kevin Wolf 已提交
557 558 559
        if (refcount == 0 && cluster_index < s->free_cluster_index) {
            s->free_cluster_index = cluster_index;
        }
K
Kevin Wolf 已提交
560
        refcount_block[block_index] = cpu_to_be16(refcount);
K
Kevin Wolf 已提交
561

562
        if (refcount == 0 && s->discard_passthrough[type]) {
K
Kevin Wolf 已提交
563
            update_refcount_discard(bs, cluster_offset, s->cluster_size);
564
        }
K
Kevin Wolf 已提交
565 566
    }

567 568
    ret = 0;
fail:
K
Kevin Wolf 已提交
569 570 571 572
    if (!s->cache_discards) {
        qcow2_process_discards(bs, ret);
    }

K
Kevin Wolf 已提交
573
    /* Write last changed block to disk */
K
Kevin Wolf 已提交
574
    if (refcount_block) {
575
        int wret;
K
Kevin Wolf 已提交
576 577
        wret = qcow2_cache_put(bs, s->refcount_block_cache,
            (void**) &refcount_block);
578 579
        if (wret < 0) {
            return ret < 0 ? ret : wret;
K
Kevin Wolf 已提交
580 581 582
        }
    }

583 584 585 586 587 588
    /*
     * 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;
589 590
        dummy = update_refcount(bs, offset, cluster_offset - offset, -addend,
                                QCOW2_DISCARD_NEVER);
B
Blue Swirl 已提交
591
        (void)dummy;
592 593 594
    }

    return ret;
K
Kevin Wolf 已提交
595 596
}

597 598 599 600 601 602 603
/*
 * 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.
 */
M
Max Reitz 已提交
604 605 606 607
int qcow2_update_cluster_refcount(BlockDriverState *bs,
                                  int64_t cluster_index,
                                  int addend,
                                  enum qcow2_discard_type type)
K
Kevin Wolf 已提交
608 609 610 611
{
    BDRVQcowState *s = bs->opaque;
    int ret;

612 613
    ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend,
                          type);
K
Kevin Wolf 已提交
614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631
    if (ret < 0) {
        return ret;
    }

    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;
632
    int i, nb_clusters, refcount;
K
Kevin Wolf 已提交
633 634 635 636

    nb_clusters = size_to_clusters(s, size);
retry:
    for(i = 0; i < nb_clusters; i++) {
637
        int64_t next_cluster_index = s->free_cluster_index++;
638 639 640 641 642
        refcount = get_refcount(bs, next_cluster_index);

        if (refcount < 0) {
            return refcount;
        } else if (refcount != 0) {
K
Kevin Wolf 已提交
643
            goto retry;
644
        }
K
Kevin Wolf 已提交
645 646
    }
#ifdef DEBUG_ALLOC2
647
    fprintf(stderr, "alloc_clusters: size=%" PRId64 " -> %" PRId64 "\n",
K
Kevin Wolf 已提交
648 649 650 651 652 653
            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
654
int64_t qcow2_alloc_clusters(BlockDriverState *bs, int64_t size)
K
Kevin Wolf 已提交
655 656
{
    int64_t offset;
657
    int ret;
K
Kevin Wolf 已提交
658

659
    BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
K
Kevin Wolf 已提交
660
    offset = alloc_clusters_noref(bs, size);
661 662 663 664
    if (offset < 0) {
        return offset;
    }

665
    ret = update_refcount(bs, offset, size, 1, QCOW2_DISCARD_NEVER);
666 667 668
    if (ret < 0) {
        return ret;
    }
669

K
Kevin Wolf 已提交
670 671 672
    return offset;
}

673 674 675 676 677
int qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset,
    int nb_clusters)
{
    BDRVQcowState *s = bs->opaque;
    uint64_t cluster_index;
678
    uint64_t old_free_cluster_index;
679 680 681 682 683 684 685 686 687 688 689 690 691 692 693
    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 */
694 695 696
    old_free_cluster_index = s->free_cluster_index;
    s->free_cluster_index = cluster_index + i;

697 698
    ret = update_refcount(bs, offset, i << s->cluster_bits, 1,
                          QCOW2_DISCARD_NEVER);
699 700 701 702
    if (ret < 0) {
        return ret;
    }

703 704
    s->free_cluster_index = old_free_cluster_index;

705 706 707
    return i;
}

K
Kevin Wolf 已提交
708 709
/* only used to allocate compressed sectors. We try to allocate
   contiguous sectors. size must be <= cluster_size */
K
Kevin Wolf 已提交
710
int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size)
K
Kevin Wolf 已提交
711 712 713 714 715
{
    BDRVQcowState *s = bs->opaque;
    int64_t offset, cluster_offset;
    int free_in_cluster;

716
    BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_BYTES);
K
Kevin Wolf 已提交
717 718
    assert(size > 0 && size <= s->cluster_size);
    if (s->free_byte_offset == 0) {
719 720 721
        offset = qcow2_alloc_clusters(bs, s->cluster_size);
        if (offset < 0) {
            return offset;
722
        }
723
        s->free_byte_offset = offset;
K
Kevin Wolf 已提交
724 725 726 727 728 729 730 731 732 733 734 735
    }
 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)
M
Max Reitz 已提交
736 737
            qcow2_update_cluster_refcount(bs, offset >> s->cluster_bits, 1,
                                          QCOW2_DISCARD_NEVER);
K
Kevin Wolf 已提交
738
    } else {
K
Kevin Wolf 已提交
739
        offset = qcow2_alloc_clusters(bs, s->cluster_size);
740 741 742
        if (offset < 0) {
            return offset;
        }
K
Kevin Wolf 已提交
743 744 745 746
        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;
M
Max Reitz 已提交
747 748
            qcow2_update_cluster_refcount(bs, offset >> s->cluster_bits, 1,
                                          QCOW2_DISCARD_NEVER);
K
Kevin Wolf 已提交
749 750 751 752 753 754
            s->free_byte_offset += size;
        } else {
            s->free_byte_offset = offset;
            goto redo;
        }
    }
755

756
    /* The cluster refcount was incremented, either by qcow2_alloc_clusters()
M
Max Reitz 已提交
757 758
     * or explicitly by qcow2_update_cluster_refcount().  Refcount blocks must
     * be flushed before the caller's L2 table updates.
759 760
     */
    qcow2_cache_set_dependency(bs, s->l2_table_cache, s->refcount_block_cache);
K
Kevin Wolf 已提交
761 762 763
    return offset;
}

K
Kevin Wolf 已提交
764
void qcow2_free_clusters(BlockDriverState *bs,
765 766
                          int64_t offset, int64_t size,
                          enum qcow2_discard_type type)
K
Kevin Wolf 已提交
767
{
768 769
    int ret;

770
    BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_FREE);
771
    ret = update_refcount(bs, offset, size, -1, type);
772 773
    if (ret < 0) {
        fprintf(stderr, "qcow2_free_clusters failed: %s\n", strerror(-ret));
774
        /* TODO Remember the clusters to free them later and avoid leaking */
775
    }
K
Kevin Wolf 已提交
776 777
}

778
/*
779 780
 * Free a cluster using its L2 entry (handles clusters of all types, e.g.
 * normal cluster, compressed cluster, etc.)
781
 */
782 783
void qcow2_free_any_clusters(BlockDriverState *bs, uint64_t l2_entry,
                             int nb_clusters, enum qcow2_discard_type type)
784 785 786
{
    BDRVQcowState *s = bs->opaque;

787 788 789 790 791 792 793 794
    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,
795
                nb_csectors * 512, type);
796 797 798 799
        }
        break;
    case QCOW2_CLUSTER_NORMAL:
        qcow2_free_clusters(bs, l2_entry & L2E_OFFSET_MASK,
800
                            nb_clusters << s->cluster_bits, type);
801 802
        break;
    case QCOW2_CLUSTER_UNALLOCATED:
803
    case QCOW2_CLUSTER_ZERO:
804 805 806
        break;
    default:
        abort();
807 808 809
    }
}

K
Kevin Wolf 已提交
810 811 812 813 814 815 816 817


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



/* update the refcounts of snapshots and the copied flag */
K
Kevin Wolf 已提交
818 819
int qcow2_update_snapshot_refcount(BlockDriverState *bs,
    int64_t l1_table_offset, int l1_size, int addend)
K
Kevin Wolf 已提交
820 821 822 823
{
    BDRVQcowState *s = bs->opaque;
    uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated;
    int64_t old_offset, old_l2_offset;
824
    int i, j, l1_modified = 0, nb_csectors, refcount;
K
Kevin Wolf 已提交
825
    int ret;
K
Kevin Wolf 已提交
826 827 828 829

    l2_table = NULL;
    l1_table = NULL;
    l1_size2 = l1_size * sizeof(uint64_t);
830

K
Kevin Wolf 已提交
831 832
    s->cache_discards = true;

833 834 835
    /* 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
836
    if (l1_table_offset != s->l1_table_offset) {
837
        l1_table = g_malloc0(align_offset(l1_size2, 512));
K
Kevin Wolf 已提交
838
        l1_allocated = 1;
839 840 841

        ret = bdrv_pread(bs->file, l1_table_offset, l1_table, l1_size2);
        if (ret < 0) {
K
Kevin Wolf 已提交
842
            goto fail;
843 844
        }

K
Kevin Wolf 已提交
845 846 847 848 849 850 851 852 853 854 855 856
        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;
857
            l2_offset &= L1E_OFFSET_MASK;
K
Kevin Wolf 已提交
858 859 860 861

            ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset,
                (void**) &l2_table);
            if (ret < 0) {
K
Kevin Wolf 已提交
862
                goto fail;
K
Kevin Wolf 已提交
863 864
            }

K
Kevin Wolf 已提交
865
            for(j = 0; j < s->l2_size; j++) {
866 867
                uint64_t cluster_index;

K
Kevin Wolf 已提交
868
                offset = be64_to_cpu(l2_table[j]);
869 870 871 872 873
                old_offset = offset;
                offset &= ~QCOW_OFLAG_COPIED;

                switch (qcow2_get_cluster_type(offset)) {
                    case QCOW2_CLUSTER_COMPRESSED:
K
Kevin Wolf 已提交
874 875
                        nb_csectors = ((offset >> s->csize_shift) &
                                       s->csize_mask) + 1;
876 877 878
                        if (addend != 0) {
                            ret = update_refcount(bs,
                                (offset & s->cluster_offset_mask) & ~511,
879 880
                                nb_csectors * 512, addend,
                                QCOW2_DISCARD_SNAPSHOT);
881 882 883 884
                            if (ret < 0) {
                                goto fail;
                            }
                        }
K
Kevin Wolf 已提交
885 886
                        /* compressed clusters are never modified */
                        refcount = 2;
887 888 889 890 891 892 893 894 895 896
                        break;

                    case QCOW2_CLUSTER_NORMAL:
                    case QCOW2_CLUSTER_ZERO:
                        cluster_index = (offset & L2E_OFFSET_MASK) >> s->cluster_bits;
                        if (!cluster_index) {
                            /* unallocated */
                            refcount = 0;
                            break;
                        }
K
Kevin Wolf 已提交
897
                        if (addend != 0) {
M
Max Reitz 已提交
898 899 900
                            refcount = qcow2_update_cluster_refcount(bs,
                                    cluster_index, addend,
                                    QCOW2_DISCARD_SNAPSHOT);
K
Kevin Wolf 已提交
901
                        } else {
902
                            refcount = get_refcount(bs, cluster_index);
K
Kevin Wolf 已提交
903
                        }
904 905

                        if (refcount < 0) {
906
                            ret = refcount;
907 908
                            goto fail;
                        }
909
                        break;
K
Kevin Wolf 已提交
910

911 912 913 914 915 916 917 918 919 920 921 922 923 924 925
                    case QCOW2_CLUSTER_UNALLOCATED:
                        refcount = 0;
                        break;

                    default:
                        abort();
                }

                if (refcount == 1) {
                    offset |= QCOW_OFLAG_COPIED;
                }
                if (offset != old_offset) {
                    if (addend > 0) {
                        qcow2_cache_set_dependency(bs, s->l2_table_cache,
                            s->refcount_block_cache);
K
Kevin Wolf 已提交
926
                    }
927 928
                    l2_table[j] = cpu_to_be64(offset);
                    qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
K
Kevin Wolf 已提交
929 930
                }
            }
K
Kevin Wolf 已提交
931 932 933 934

            ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
            if (ret < 0) {
                goto fail;
K
Kevin Wolf 已提交
935 936
            }

K
Kevin Wolf 已提交
937

K
Kevin Wolf 已提交
938
            if (addend != 0) {
M
Max Reitz 已提交
939 940
                refcount = qcow2_update_cluster_refcount(bs, l2_offset >>
                        s->cluster_bits, addend, QCOW2_DISCARD_SNAPSHOT);
K
Kevin Wolf 已提交
941 942 943
            } else {
                refcount = get_refcount(bs, l2_offset >> s->cluster_bits);
            }
944
            if (refcount < 0) {
945
                ret = refcount;
946 947
                goto fail;
            } else if (refcount == 1) {
K
Kevin Wolf 已提交
948 949 950 951 952 953 954 955
                l2_offset |= QCOW_OFLAG_COPIED;
            }
            if (l2_offset != old_l2_offset) {
                l1_table[i] = l2_offset;
                l1_modified = 1;
            }
        }
    }
956

957
    ret = bdrv_flush(bs);
958 959 960 961 962
fail:
    if (l2_table) {
        qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
    }

K
Kevin Wolf 已提交
963 964 965
    s->cache_discards = false;
    qcow2_process_discards(bs, ret);

966
    /* Update L1 only if it isn't deleted anyway (addend = -1) */
967 968
    if (ret == 0 && addend >= 0 && l1_modified) {
        for (i = 0; i < l1_size; i++) {
K
Kevin Wolf 已提交
969
            cpu_to_be64s(&l1_table[i]);
970 971 972 973 974
        }

        ret = bdrv_pwrite_sync(bs->file, l1_table_offset, l1_table, l1_size2);

        for (i = 0; i < l1_size; i++) {
K
Kevin Wolf 已提交
975
            be64_to_cpus(&l1_table[i]);
976
        }
K
Kevin Wolf 已提交
977 978
    }
    if (l1_allocated)
979
        g_free(l1_table);
980
    return ret;
K
Kevin Wolf 已提交
981 982 983 984 985 986 987 988 989 990 991 992 993 994 995
}




/*********************************************************/
/* 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.
 *
996
 * Modifies the number of errors in res.
K
Kevin Wolf 已提交
997
 */
998 999
static void inc_refcounts(BlockDriverState *bs,
                          BdrvCheckResult *res,
K
Kevin Wolf 已提交
1000 1001 1002 1003 1004 1005 1006 1007 1008
                          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)
1009
        return;
K
Kevin Wolf 已提交
1010 1011 1012 1013 1014 1015

    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;
1016
        if (k < 0) {
K
Kevin Wolf 已提交
1017 1018
            fprintf(stderr, "ERROR: invalid cluster offset=0x%" PRIx64 "\n",
                cluster_offset);
1019 1020 1021 1022 1023 1024
            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
1025 1026 1027 1028
        } else {
            if (++refcount_table[k] == 0) {
                fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64
                    "\n", cluster_offset);
1029
                res->corruptions++;
K
Kevin Wolf 已提交
1030 1031 1032 1033 1034
            }
        }
    }
}

1035 1036
/* Flags for check_refcounts_l1() and check_refcounts_l2() */
enum {
1037
    CHECK_FRAG_INFO = 0x2,      /* update BlockFragInfo counters */
1038 1039
};

K
Kevin Wolf 已提交
1040 1041 1042 1043 1044 1045 1046 1047
/*
 * 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.
 */
1048
static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res,
K
Kevin Wolf 已提交
1049
    uint16_t *refcount_table, int refcount_table_size, int64_t l2_offset,
1050
    int flags)
K
Kevin Wolf 已提交
1051 1052
{
    BDRVQcowState *s = bs->opaque;
1053
    uint64_t *l2_table, l2_entry;
1054
    uint64_t next_contiguous_offset = 0;
1055
    int i, l2_size, nb_csectors;
K
Kevin Wolf 已提交
1056 1057 1058

    /* Read L2 table from disk */
    l2_size = s->l2_size * sizeof(uint64_t);
1059
    l2_table = g_malloc(l2_size);
K
Kevin Wolf 已提交
1060

1061
    if (bdrv_pread(bs->file, l2_offset, l2_table, l2_size) != l2_size)
K
Kevin Wolf 已提交
1062 1063 1064 1065
        goto fail;

    /* Do the actual checks */
    for(i = 0; i < s->l2_size; i++) {
1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077
        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
1078

1079 1080 1081 1082 1083 1084
            /* 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);
1085 1086 1087

            if (flags & CHECK_FRAG_INFO) {
                res->bfi.allocated_clusters++;
1088
                res->bfi.compressed_clusters++;
1089 1090 1091 1092 1093 1094 1095 1096

                /* 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++;
            }
1097
            break;
K
Kevin Wolf 已提交
1098

1099 1100 1101 1102 1103 1104
        case QCOW2_CLUSTER_ZERO:
            if ((l2_entry & L2E_OFFSET_MASK) == 0) {
                break;
            }
            /* fall through */

1105 1106 1107
        case QCOW2_CLUSTER_NORMAL:
        {
            uint64_t offset = l2_entry & L2E_OFFSET_MASK;
K
Kevin Wolf 已提交
1108

1109 1110 1111 1112 1113 1114 1115 1116 1117
            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;
            }

1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135
            /* 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
1136 1137 1138
        }
    }

1139
    g_free(l2_table);
1140
    return 0;
K
Kevin Wolf 已提交
1141 1142

fail:
1143
    fprintf(stderr, "ERROR: I/O error in check_refcounts_l2\n");
1144
    g_free(l2_table);
K
Kevin Wolf 已提交
1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156
    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,
1157
                              BdrvCheckResult *res,
K
Kevin Wolf 已提交
1158 1159 1160
                              uint16_t *refcount_table,
                              int refcount_table_size,
                              int64_t l1_table_offset, int l1_size,
1161
                              int flags)
K
Kevin Wolf 已提交
1162 1163 1164
{
    BDRVQcowState *s = bs->opaque;
    uint64_t *l1_table, l2_offset, l1_size2;
1165
    int i, ret;
K
Kevin Wolf 已提交
1166 1167 1168 1169

    l1_size2 = l1_size * sizeof(uint64_t);

    /* Mark L1 table as used */
1170 1171
    inc_refcounts(bs, res, refcount_table, refcount_table_size,
        l1_table_offset, l1_size2);
K
Kevin Wolf 已提交
1172 1173

    /* Read L1 table entries from disk */
1174 1175 1176
    if (l1_size2 == 0) {
        l1_table = NULL;
    } else {
1177
        l1_table = g_malloc(l1_size2);
1178
        if (bdrv_pread(bs->file, l1_table_offset,
1179 1180 1181 1182 1183
                       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
1184 1185 1186 1187 1188 1189

    /* Do the actual checks */
    for(i = 0; i < l1_size; i++) {
        l2_offset = l1_table[i];
        if (l2_offset) {
            /* Mark L2 table as used */
1190
            l2_offset &= L1E_OFFSET_MASK;
1191 1192
            inc_refcounts(bs, res, refcount_table, refcount_table_size,
                l2_offset, s->cluster_size);
K
Kevin Wolf 已提交
1193 1194 1195 1196 1197

            /* 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);
1198
                res->corruptions++;
K
Kevin Wolf 已提交
1199 1200 1201
            }

            /* Process and check L2 entries */
1202
            ret = check_refcounts_l2(bs, res, refcount_table,
1203
                                     refcount_table_size, l2_offset, flags);
K
Kevin Wolf 已提交
1204 1205 1206 1207 1208
            if (ret < 0) {
                goto fail;
            }
        }
    }
1209
    g_free(l1_table);
1210
    return 0;
K
Kevin Wolf 已提交
1211 1212 1213

fail:
    fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
1214
    res->check_errors++;
1215
    g_free(l1_table);
K
Kevin Wolf 已提交
1216 1217 1218
    return -EIO;
}

1219 1220 1221 1222 1223 1224 1225 1226
/*
 * Checks the OFLAG_COPIED flag for all L1 and L2 entries.
 *
 * This function does not print an error message nor does it increment
 * check_errors if get_refcount fails (this is because such an error will have
 * been already detected and sufficiently signaled by the calling function
 * (qcow2_check_refcounts) by the time this function is called).
 */
1227 1228
static int check_oflag_copied(BlockDriverState *bs, BdrvCheckResult *res,
                              BdrvCheckMode fix)
1229 1230 1231 1232 1233 1234 1235 1236 1237 1238
{
    BDRVQcowState *s = bs->opaque;
    uint64_t *l2_table = qemu_blockalign(bs, s->cluster_size);
    int ret;
    int refcount;
    int i, j;

    for (i = 0; i < s->l1_size; i++) {
        uint64_t l1_entry = s->l1_table[i];
        uint64_t l2_offset = l1_entry & L1E_OFFSET_MASK;
1239
        bool l2_dirty = false;
1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250

        if (!l2_offset) {
            continue;
        }

        refcount = get_refcount(bs, l2_offset >> s->cluster_bits);
        if (refcount < 0) {
            /* don't print message nor increment check_errors */
            continue;
        }
        if ((refcount == 1) != ((l1_entry & QCOW_OFLAG_COPIED) != 0)) {
1251
            fprintf(stderr, "%s OFLAG_COPIED L2 cluster: l1_index=%d "
1252
                    "l1_entry=%" PRIx64 " refcount=%d\n",
1253 1254
                    fix & BDRV_FIX_ERRORS ? "Repairing" :
                                            "ERROR",
1255
                    i, l1_entry, refcount);
1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268
            if (fix & BDRV_FIX_ERRORS) {
                s->l1_table[i] = refcount == 1
                               ? l1_entry |  QCOW_OFLAG_COPIED
                               : l1_entry & ~QCOW_OFLAG_COPIED;
                ret = qcow2_write_l1_entry(bs, i);
                if (ret < 0) {
                    res->check_errors++;
                    goto fail;
                }
                res->corruptions_fixed++;
            } else {
                res->corruptions++;
            }
1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292
        }

        ret = bdrv_pread(bs->file, l2_offset, l2_table,
                         s->l2_size * sizeof(uint64_t));
        if (ret < 0) {
            fprintf(stderr, "ERROR: Could not read L2 table: %s\n",
                    strerror(-ret));
            res->check_errors++;
            goto fail;
        }

        for (j = 0; j < s->l2_size; j++) {
            uint64_t l2_entry = be64_to_cpu(l2_table[j]);
            uint64_t data_offset = l2_entry & L2E_OFFSET_MASK;
            int cluster_type = qcow2_get_cluster_type(l2_entry);

            if ((cluster_type == QCOW2_CLUSTER_NORMAL) ||
                ((cluster_type == QCOW2_CLUSTER_ZERO) && (data_offset != 0))) {
                refcount = get_refcount(bs, data_offset >> s->cluster_bits);
                if (refcount < 0) {
                    /* don't print message nor increment check_errors */
                    continue;
                }
                if ((refcount == 1) != ((l2_entry & QCOW_OFLAG_COPIED) != 0)) {
1293
                    fprintf(stderr, "%s OFLAG_COPIED data cluster: "
1294
                            "l2_entry=%" PRIx64 " refcount=%d\n",
1295 1296
                            fix & BDRV_FIX_ERRORS ? "Repairing" :
                                                    "ERROR",
1297
                            l2_entry, refcount);
1298 1299 1300 1301 1302 1303 1304 1305 1306
                    if (fix & BDRV_FIX_ERRORS) {
                        l2_table[j] = cpu_to_be64(refcount == 1
                                    ? l2_entry |  QCOW_OFLAG_COPIED
                                    : l2_entry & ~QCOW_OFLAG_COPIED);
                        l2_dirty = true;
                        res->corruptions_fixed++;
                    } else {
                        res->corruptions++;
                    }
1307 1308 1309
                }
            }
        }
1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329

        if (l2_dirty) {
            ret = qcow2_pre_write_overlap_check(bs,
                    QCOW2_OL_DEFAULT & ~QCOW2_OL_ACTIVE_L2, l2_offset,
                    s->cluster_size);
            if (ret < 0) {
                fprintf(stderr, "ERROR: Could not write L2 table; metadata "
                        "overlap check failed: %s\n", strerror(-ret));
                res->check_errors++;
                goto fail;
            }

            ret = bdrv_pwrite(bs->file, l2_offset, l2_table, s->cluster_size);
            if (ret < 0) {
                fprintf(stderr, "ERROR: Could not write L2 table: %s\n",
                        strerror(-ret));
                res->check_errors++;
                goto fail;
            }
        }
1330 1331 1332 1333 1334 1335 1336 1337 1338
    }

    ret = 0;

fail:
    qemu_vfree(l2_table);
    return ret;
}

1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453
/*
 * Writes one sector of the refcount table to the disk
 */
#define RT_ENTRIES_PER_SECTOR (512 / sizeof(uint64_t))
static int write_reftable_entry(BlockDriverState *bs, int rt_index)
{
    BDRVQcowState *s = bs->opaque;
    uint64_t buf[RT_ENTRIES_PER_SECTOR];
    int rt_start_index;
    int i, ret;

    rt_start_index = rt_index & ~(RT_ENTRIES_PER_SECTOR - 1);
    for (i = 0; i < RT_ENTRIES_PER_SECTOR; i++) {
        buf[i] = cpu_to_be64(s->refcount_table[rt_start_index + i]);
    }

    ret = qcow2_pre_write_overlap_check(bs,
            QCOW2_OL_DEFAULT & ~QCOW2_OL_REFCOUNT_TABLE,
            s->refcount_table_offset + rt_start_index * sizeof(uint64_t),
            sizeof(buf));
    if (ret < 0) {
        return ret;
    }

    BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE);
    ret = bdrv_pwrite_sync(bs->file, s->refcount_table_offset +
            rt_start_index * sizeof(uint64_t), buf, sizeof(buf));
    if (ret < 0) {
        return ret;
    }

    return 0;
}

/*
 * Allocates a new cluster for the given refcount block (represented by its
 * offset in the image file) and copies the current content there. This function
 * does _not_ decrement the reference count for the currently occupied cluster.
 *
 * This function prints an informative message to stderr on error (and returns
 * -errno); on success, 0 is returned.
 */
static int64_t realloc_refcount_block(BlockDriverState *bs, int reftable_index,
                                      uint64_t offset)
{
    BDRVQcowState *s = bs->opaque;
    int64_t new_offset = 0;
    void *refcount_block = NULL;
    int ret;

    /* allocate new refcount block */
    new_offset = qcow2_alloc_clusters(bs, s->cluster_size);
    if (new_offset < 0) {
        fprintf(stderr, "Could not allocate new cluster: %s\n",
                strerror(-new_offset));
        ret = new_offset;
        goto fail;
    }

    /* fetch current refcount block content */
    ret = qcow2_cache_get(bs, s->refcount_block_cache, offset, &refcount_block);
    if (ret < 0) {
        fprintf(stderr, "Could not fetch refcount block: %s\n", strerror(-ret));
        goto fail;
    }

    /* new block has not yet been entered into refcount table, therefore it is
     * no refcount block yet (regarding this check) */
    ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_DEFAULT, new_offset,
            s->cluster_size);
    if (ret < 0) {
        fprintf(stderr, "Could not write refcount block; metadata overlap "
                "check failed: %s\n", strerror(-ret));
        /* the image will be marked corrupt, so don't even attempt on freeing
         * the cluster */
        new_offset = 0;
        goto fail;
    }

    /* write to new block */
    ret = bdrv_write(bs->file, new_offset / BDRV_SECTOR_SIZE, refcount_block,
            s->cluster_sectors);
    if (ret < 0) {
        fprintf(stderr, "Could not write refcount block: %s\n", strerror(-ret));
        goto fail;
    }

    /* update refcount table */
    assert(!(new_offset & (s->cluster_size - 1)));
    s->refcount_table[reftable_index] = new_offset;
    ret = write_reftable_entry(bs, reftable_index);
    if (ret < 0) {
        fprintf(stderr, "Could not update refcount table: %s\n",
                strerror(-ret));
        goto fail;
    }

fail:
    if (new_offset && (ret < 0)) {
        qcow2_free_clusters(bs, new_offset, s->cluster_size,
                QCOW2_DISCARD_ALWAYS);
    }
    if (refcount_block) {
        if (ret < 0) {
            qcow2_cache_put(bs, s->refcount_block_cache, &refcount_block);
        } else {
            ret = qcow2_cache_put(bs, s->refcount_block_cache, &refcount_block);
        }
    }
    if (ret < 0) {
        return ret;
    }
    return new_offset;
}

K
Kevin Wolf 已提交
1454 1455 1456 1457
/*
 * Checks an image for refcount consistency.
 *
 * Returns 0 if no errors are found, the number of errors in case the image is
1458
 * detected as corrupted, and -errno when an internal error occurred.
K
Kevin Wolf 已提交
1459
 */
1460 1461
int qcow2_check_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
                          BdrvCheckMode fix)
K
Kevin Wolf 已提交
1462 1463
{
    BDRVQcowState *s = bs->opaque;
1464
    int64_t size, i, highest_cluster;
1465
    int nb_clusters, refcount1, refcount2;
K
Kevin Wolf 已提交
1466 1467
    QCowSnapshot *sn;
    uint16_t *refcount_table;
1468
    int ret;
K
Kevin Wolf 已提交
1469

1470
    size = bdrv_getlength(bs->file);
K
Kevin Wolf 已提交
1471
    nb_clusters = size_to_clusters(s, size);
1472
    refcount_table = g_malloc0(nb_clusters * sizeof(uint16_t));
K
Kevin Wolf 已提交
1473

1474 1475 1476
    res->bfi.total_clusters =
        size_to_clusters(s, bs->total_sectors * BDRV_SECTOR_SIZE);

K
Kevin Wolf 已提交
1477
    /* header */
1478 1479
    inc_refcounts(bs, res, refcount_table, nb_clusters,
        0, s->cluster_size);
K
Kevin Wolf 已提交
1480 1481

    /* current L1 table */
1482
    ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
M
Max Reitz 已提交
1483
                             s->l1_table_offset, s->l1_size, CHECK_FRAG_INFO);
K
Kevin Wolf 已提交
1484
    if (ret < 0) {
1485
        goto fail;
K
Kevin Wolf 已提交
1486 1487 1488 1489 1490
    }

    /* snapshots */
    for(i = 0; i < s->nb_snapshots; i++) {
        sn = s->snapshots + i;
1491 1492 1493
        ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
            sn->l1_table_offset, sn->l1_size, 0);
        if (ret < 0) {
1494
            goto fail;
1495
        }
K
Kevin Wolf 已提交
1496
    }
1497 1498
    inc_refcounts(bs, res, refcount_table, nb_clusters,
        s->snapshots_offset, s->snapshots_size);
K
Kevin Wolf 已提交
1499 1500

    /* refcount data */
1501 1502 1503 1504
    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
1505
    for(i = 0; i < s->refcount_table_size; i++) {
1506
        uint64_t offset, cluster;
K
Kevin Wolf 已提交
1507
        offset = s->refcount_table[i];
1508
        cluster = offset >> s->cluster_bits;
1509 1510 1511

        /* Refcount blocks are cluster aligned */
        if (offset & (s->cluster_size - 1)) {
1512
            fprintf(stderr, "ERROR refcount block %" PRId64 " is not "
1513
                "cluster aligned; refcount table entry corrupted\n", i);
1514
            res->corruptions++;
1515 1516 1517 1518
            continue;
        }

        if (cluster >= nb_clusters) {
1519 1520
            fprintf(stderr, "ERROR refcount block %" PRId64
                    " is outside image\n", i);
1521
            res->corruptions++;
1522
            continue;
1523 1524
        }

K
Kevin Wolf 已提交
1525
        if (offset != 0) {
1526 1527
            inc_refcounts(bs, res, refcount_table, nb_clusters,
                offset, s->cluster_size);
1528
            if (refcount_table[cluster] != 1) {
1529
                fprintf(stderr, "%s refcount block %" PRId64
1530
                    " refcount=%d\n",
1531 1532
                    fix & BDRV_FIX_ERRORS ? "Repairing" :
                                            "ERROR",
1533
                    i, refcount_table[cluster]);
1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561

                if (fix & BDRV_FIX_ERRORS) {
                    int64_t new_offset;

                    new_offset = realloc_refcount_block(bs, i, offset);
                    if (new_offset < 0) {
                        res->corruptions++;
                        continue;
                    }

                    /* update refcounts */
                    if ((new_offset >> s->cluster_bits) >= nb_clusters) {
                        /* increase refcount_table size if necessary */
                        int old_nb_clusters = nb_clusters;
                        nb_clusters = (new_offset >> s->cluster_bits) + 1;
                        refcount_table = g_realloc(refcount_table,
                                nb_clusters * sizeof(uint16_t));
                        memset(&refcount_table[old_nb_clusters], 0, (nb_clusters
                                - old_nb_clusters) * sizeof(uint16_t));
                    }
                    refcount_table[cluster]--;
                    inc_refcounts(bs, res, refcount_table, nb_clusters,
                            new_offset, s->cluster_size);

                    res->corruptions_fixed++;
                } else {
                    res->corruptions++;
                }
1562
            }
K
Kevin Wolf 已提交
1563 1564 1565 1566
        }
    }

    /* compare ref counts */
1567
    for (i = 0, highest_cluster = 0; i < nb_clusters; i++) {
K
Kevin Wolf 已提交
1568
        refcount1 = get_refcount(bs, i);
1569
        if (refcount1 < 0) {
1570
            fprintf(stderr, "Can't get refcount for cluster %" PRId64 ": %s\n",
1571
                i, strerror(-refcount1));
1572
            res->check_errors++;
1573
            continue;
1574 1575
        }

K
Kevin Wolf 已提交
1576
        refcount2 = refcount_table[i];
1577 1578 1579 1580 1581

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

K
Kevin Wolf 已提交
1582
        if (refcount1 != refcount2) {
1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595

            /* 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
1596
                   i, refcount1, refcount2);
1597 1598 1599

            if (num_fixed) {
                ret = update_refcount(bs, i << s->cluster_bits, 1,
1600 1601
                                      refcount2 - refcount1,
                                      QCOW2_DISCARD_ALWAYS);
1602 1603 1604 1605 1606 1607 1608
                if (ret >= 0) {
                    (*num_fixed)++;
                    continue;
                }
            }

            /* And if we couldn't, print an error */
1609 1610 1611 1612 1613
            if (refcount1 < refcount2) {
                res->corruptions++;
            } else {
                res->leaks++;
            }
K
Kevin Wolf 已提交
1614 1615 1616
        }
    }

1617
    /* check OFLAG_COPIED */
1618
    ret = check_oflag_copied(bs, res, fix);
1619 1620 1621 1622
    if (ret < 0) {
        goto fail;
    }

1623
    res->image_end_offset = (highest_cluster + 1) * s->cluster_size;
1624 1625 1626
    ret = 0;

fail:
1627
    g_free(refcount_table);
K
Kevin Wolf 已提交
1628

1629
    return ret;
K
Kevin Wolf 已提交
1630 1631
}

M
Max Reitz 已提交
1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733
#define overlaps_with(ofs, sz) \
    ranges_overlap(offset, size, ofs, sz)

/*
 * Checks if the given offset into the image file is actually free to use by
 * looking for overlaps with important metadata sections (L1/L2 tables etc.),
 * i.e. a sanity check without relying on the refcount tables.
 *
 * The chk parameter specifies exactly what checks to perform (being a bitmask
 * of QCow2MetadataOverlap values).
 *
 * Returns:
 * - 0 if writing to this offset will not affect the mentioned metadata
 * - a positive QCow2MetadataOverlap value indicating one overlapping section
 * - a negative value (-errno) indicating an error while performing a check,
 *   e.g. when bdrv_read failed on QCOW2_OL_INACTIVE_L2
 */
int qcow2_check_metadata_overlap(BlockDriverState *bs, int chk, int64_t offset,
                                 int64_t size)
{
    BDRVQcowState *s = bs->opaque;
    int i, j;

    if (!size) {
        return 0;
    }

    if (chk & QCOW2_OL_MAIN_HEADER) {
        if (offset < s->cluster_size) {
            return QCOW2_OL_MAIN_HEADER;
        }
    }

    /* align range to test to cluster boundaries */
    size = align_offset(offset_into_cluster(s, offset) + size, s->cluster_size);
    offset = start_of_cluster(s, offset);

    if ((chk & QCOW2_OL_ACTIVE_L1) && s->l1_size) {
        if (overlaps_with(s->l1_table_offset, s->l1_size * sizeof(uint64_t))) {
            return QCOW2_OL_ACTIVE_L1;
        }
    }

    if ((chk & QCOW2_OL_REFCOUNT_TABLE) && s->refcount_table_size) {
        if (overlaps_with(s->refcount_table_offset,
            s->refcount_table_size * sizeof(uint64_t))) {
            return QCOW2_OL_REFCOUNT_TABLE;
        }
    }

    if ((chk & QCOW2_OL_SNAPSHOT_TABLE) && s->snapshots_size) {
        if (overlaps_with(s->snapshots_offset, s->snapshots_size)) {
            return QCOW2_OL_SNAPSHOT_TABLE;
        }
    }

    if ((chk & QCOW2_OL_INACTIVE_L1) && s->snapshots) {
        for (i = 0; i < s->nb_snapshots; i++) {
            if (s->snapshots[i].l1_size &&
                overlaps_with(s->snapshots[i].l1_table_offset,
                s->snapshots[i].l1_size * sizeof(uint64_t))) {
                return QCOW2_OL_INACTIVE_L1;
            }
        }
    }

    if ((chk & QCOW2_OL_ACTIVE_L2) && s->l1_table) {
        for (i = 0; i < s->l1_size; i++) {
            if ((s->l1_table[i] & L1E_OFFSET_MASK) &&
                overlaps_with(s->l1_table[i] & L1E_OFFSET_MASK,
                s->cluster_size)) {
                return QCOW2_OL_ACTIVE_L2;
            }
        }
    }

    if ((chk & QCOW2_OL_REFCOUNT_BLOCK) && s->refcount_table) {
        for (i = 0; i < s->refcount_table_size; i++) {
            if ((s->refcount_table[i] & REFT_OFFSET_MASK) &&
                overlaps_with(s->refcount_table[i] & REFT_OFFSET_MASK,
                s->cluster_size)) {
                return QCOW2_OL_REFCOUNT_BLOCK;
            }
        }
    }

    if ((chk & QCOW2_OL_INACTIVE_L2) && s->snapshots) {
        for (i = 0; i < s->nb_snapshots; i++) {
            uint64_t l1_ofs = s->snapshots[i].l1_table_offset;
            uint32_t l1_sz  = s->snapshots[i].l1_size;
            uint64_t *l1 = g_malloc(l1_sz * sizeof(uint64_t));
            int ret;

            ret = bdrv_read(bs->file, l1_ofs / BDRV_SECTOR_SIZE, (uint8_t *)l1,
                            l1_sz * sizeof(uint64_t) / BDRV_SECTOR_SIZE);

            if (ret < 0) {
                g_free(l1);
                return ret;
            }

            for (j = 0; j < l1_sz; j++) {
1734 1735
                uint64_t l2_ofs = be64_to_cpu(l1[j]) & L1E_OFFSET_MASK;
                if (l2_ofs && overlaps_with(l2_ofs, s->cluster_size)) {
M
Max Reitz 已提交
1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801
                    g_free(l1);
                    return QCOW2_OL_INACTIVE_L2;
                }
            }

            g_free(l1);
        }
    }

    return 0;
}

static const char *metadata_ol_names[] = {
    [QCOW2_OL_MAIN_HEADER_BITNR]    = "qcow2_header",
    [QCOW2_OL_ACTIVE_L1_BITNR]      = "active L1 table",
    [QCOW2_OL_ACTIVE_L2_BITNR]      = "active L2 table",
    [QCOW2_OL_REFCOUNT_TABLE_BITNR] = "refcount table",
    [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = "refcount block",
    [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = "snapshot table",
    [QCOW2_OL_INACTIVE_L1_BITNR]    = "inactive L1 table",
    [QCOW2_OL_INACTIVE_L2_BITNR]    = "inactive L2 table",
};

/*
 * First performs a check for metadata overlaps (through
 * qcow2_check_metadata_overlap); if that fails with a negative value (error
 * while performing a check), that value is returned. If an impending overlap
 * is detected, the BDS will be made unusable, the qcow2 file marked corrupt
 * and -EIO returned.
 *
 * Returns 0 if there were neither overlaps nor errors while checking for
 * overlaps; or a negative value (-errno) on error.
 */
int qcow2_pre_write_overlap_check(BlockDriverState *bs, int chk, int64_t offset,
                                  int64_t size)
{
    int ret = qcow2_check_metadata_overlap(bs, chk, offset, size);

    if (ret < 0) {
        return ret;
    } else if (ret > 0) {
        int metadata_ol_bitnr = ffs(ret) - 1;
        char *message;
        QObject *data;

        assert(metadata_ol_bitnr < QCOW2_OL_MAX_BITNR);

        fprintf(stderr, "qcow2: Preventing invalid write on metadata (overlaps "
                "with %s); image marked as corrupt.\n",
                metadata_ol_names[metadata_ol_bitnr]);
        message = g_strdup_printf("Prevented %s overwrite",
                metadata_ol_names[metadata_ol_bitnr]);
        data = qobject_from_jsonf("{ 'device': %s, 'msg': %s, 'offset': %"
                PRId64 ", 'size': %" PRId64 " }", bs->device_name, message,
                offset, size);
        monitor_protocol_event(QEVENT_BLOCK_IMAGE_CORRUPTED, data);
        g_free(message);
        qobject_decref(data);

        qcow2_mark_corrupt(bs);
        bs->drv = NULL; /* make BDS unusable */
        return -EIO;
    }

    return 0;
}