Statistics
| Branch: | Revision:

root / block / qcow2-refcount.c @ 83e3f76c

History | View | Annotate | Download (39.2 kB)

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

    
25
#include "qemu-common.h"
26
#include "block_int.h"
27
#include "block/qcow2.h"
28

    
29
static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size);
30
static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
31
                            int64_t offset, int64_t length,
32
                            int addend);
33

    
34

    
35
static int cache_refcount_updates = 0;
36

    
37
static int write_refcount_block(BlockDriverState *bs)
38
{
39
    BDRVQcowState *s = bs->opaque;
40
    size_t size = s->cluster_size;
41

    
42
    if (s->refcount_block_cache_offset == 0) {
43
        return 0;
44
    }
45

    
46
    BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_UPDATE);
47
    if (bdrv_pwrite_sync(bs->file, s->refcount_block_cache_offset,
48
            s->refcount_block_cache, size) < 0)
49
    {
50
        return -EIO;
51
    }
52

    
53
    return 0;
54
}
55

    
56
/*********************************************************/
57
/* refcount handling */
58

    
59
int qcow2_refcount_init(BlockDriverState *bs)
60
{
61
    BDRVQcowState *s = bs->opaque;
62
    int ret, refcount_table_size2, i;
63

    
64
    s->refcount_block_cache = qemu_malloc(s->cluster_size);
65
    refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t);
66
    s->refcount_table = qemu_malloc(refcount_table_size2);
67
    if (s->refcount_table_size > 0) {
68
        BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_LOAD);
69
        ret = bdrv_pread(bs->file, s->refcount_table_offset,
70
                         s->refcount_table, refcount_table_size2);
71
        if (ret != refcount_table_size2)
72
            goto fail;
73
        for(i = 0; i < s->refcount_table_size; i++)
74
            be64_to_cpus(&s->refcount_table[i]);
75
    }
76
    return 0;
77
 fail:
78
    return -ENOMEM;
79
}
80

    
81
void qcow2_refcount_close(BlockDriverState *bs)
82
{
83
    BDRVQcowState *s = bs->opaque;
84
    qemu_free(s->refcount_block_cache);
85
    qemu_free(s->refcount_table);
86
}
87

    
88

    
89
static int load_refcount_block(BlockDriverState *bs,
90
                               int64_t refcount_block_offset)
91
{
92
    BDRVQcowState *s = bs->opaque;
93
    int ret;
94

    
95
    if (cache_refcount_updates) {
96
        ret = write_refcount_block(bs);
97
        if (ret < 0) {
98
            return ret;
99
        }
100
    }
101

    
102
    BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_LOAD);
103
    ret = bdrv_pread(bs->file, refcount_block_offset, s->refcount_block_cache,
104
                     s->cluster_size);
105
    if (ret < 0) {
106
        return ret;
107
    }
108

    
109
    s->refcount_block_cache_offset = refcount_block_offset;
110
    return 0;
111
}
112

    
113
/*
114
 * Returns the refcount of the cluster given by its index. Any non-negative
115
 * return value is the refcount of the cluster, negative values are -errno
116
 * and indicate an error.
117
 */
118
static int get_refcount(BlockDriverState *bs, int64_t cluster_index)
119
{
120
    BDRVQcowState *s = bs->opaque;
121
    int refcount_table_index, block_index;
122
    int64_t refcount_block_offset;
123
    int ret;
124

    
125
    refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
126
    if (refcount_table_index >= s->refcount_table_size)
127
        return 0;
128
    refcount_block_offset = s->refcount_table[refcount_table_index];
129
    if (!refcount_block_offset)
130
        return 0;
131
    if (refcount_block_offset != s->refcount_block_cache_offset) {
132
        /* better than nothing: return allocated if read error */
133
        ret = load_refcount_block(bs, refcount_block_offset);
134
        if (ret < 0) {
135
            return ret;
136
        }
137
    }
138
    block_index = cluster_index &
139
        ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
140
    return be16_to_cpu(s->refcount_block_cache[block_index]);
141
}
142

    
143
/*
144
 * Rounds the refcount table size up to avoid growing the table for each single
145
 * refcount block that is allocated.
146
 */
147
static unsigned int next_refcount_table_size(BDRVQcowState *s,
148
    unsigned int min_size)
149
{
150
    unsigned int min_clusters = (min_size >> (s->cluster_bits - 3)) + 1;
151
    unsigned int refcount_table_clusters =
152
        MAX(1, s->refcount_table_size >> (s->cluster_bits - 3));
153

    
154
    while (min_clusters > refcount_table_clusters) {
155
        refcount_table_clusters = (refcount_table_clusters * 3 + 1) / 2;
156
    }
157

    
158
    return refcount_table_clusters << (s->cluster_bits - 3);
159
}
160

    
161

    
162
/* Checks if two offsets are described by the same refcount block */
163
static int in_same_refcount_block(BDRVQcowState *s, uint64_t offset_a,
164
    uint64_t offset_b)
165
{
166
    uint64_t block_a = offset_a >> (2 * s->cluster_bits - REFCOUNT_SHIFT);
167
    uint64_t block_b = offset_b >> (2 * s->cluster_bits - REFCOUNT_SHIFT);
168

    
169
    return (block_a == block_b);
170
}
171

    
172
/*
173
 * Loads a refcount block. If it doesn't exist yet, it is allocated first
174
 * (including growing the refcount table if needed).
175
 *
176
 * Returns the offset of the refcount block on success or -errno in error case
177
 */
178
static int64_t alloc_refcount_block(BlockDriverState *bs, int64_t cluster_index)
179
{
180
    BDRVQcowState *s = bs->opaque;
181
    unsigned int refcount_table_index;
182
    int ret;
183

    
184
    BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
185

    
186
    /* Find the refcount block for the given cluster */
187
    refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
188

    
189
    if (refcount_table_index < s->refcount_table_size) {
190

    
191
        uint64_t refcount_block_offset =
192
            s->refcount_table[refcount_table_index];
193

    
194
        /* If it's already there, we're done */
195
        if (refcount_block_offset) {
196
            if (refcount_block_offset != s->refcount_block_cache_offset) {
197
                ret = load_refcount_block(bs, refcount_block_offset);
198
                if (ret < 0) {
199
                    return ret;
200
                }
201
            }
202
            return refcount_block_offset;
203
        }
204
    }
205

    
206
    /*
207
     * If we came here, we need to allocate something. Something is at least
208
     * a cluster for the new refcount block. It may also include a new refcount
209
     * table if the old refcount table is too small.
210
     *
211
     * Note that allocating clusters here needs some special care:
212
     *
213
     * - We can't use the normal qcow2_alloc_clusters(), it would try to
214
     *   increase the refcount and very likely we would end up with an endless
215
     *   recursion. Instead we must place the refcount blocks in a way that
216
     *   they can describe them themselves.
217
     *
218
     * - We need to consider that at this point we are inside update_refcounts
219
     *   and doing the initial refcount increase. This means that some clusters
220
     *   have already been allocated by the caller, but their refcount isn't
221
     *   accurate yet. free_cluster_index tells us where this allocation ends
222
     *   as long as we don't overwrite it by freeing clusters.
223
     *
224
     * - alloc_clusters_noref and qcow2_free_clusters may load a different
225
     *   refcount block into the cache
226
     */
227

    
228
    if (cache_refcount_updates) {
229
        ret = write_refcount_block(bs);
230
        if (ret < 0) {
231
            return ret;
232
        }
233
    }
234

    
235
    /* Allocate the refcount block itself and mark it as used */
236
    int64_t new_block = alloc_clusters_noref(bs, s->cluster_size);
237
    if (new_block < 0) {
238
        return new_block;
239
    }
240

    
241
#ifdef DEBUG_ALLOC2
242
    fprintf(stderr, "qcow2: Allocate refcount block %d for %" PRIx64
243
        " at %" PRIx64 "\n",
244
        refcount_table_index, cluster_index << s->cluster_bits, new_block);
245
#endif
246

    
247
    if (in_same_refcount_block(s, new_block, cluster_index << s->cluster_bits)) {
248
        /* Zero the new refcount block before updating it */
249
        memset(s->refcount_block_cache, 0, s->cluster_size);
250
        s->refcount_block_cache_offset = new_block;
251

    
252
        /* The block describes itself, need to update the cache */
253
        int block_index = (new_block >> s->cluster_bits) &
254
            ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
255
        s->refcount_block_cache[block_index] = cpu_to_be16(1);
256
    } else {
257
        /* Described somewhere else. This can recurse at most twice before we
258
         * arrive at a block that describes itself. */
259
        ret = update_refcount(bs, new_block, s->cluster_size, 1);
260
        if (ret < 0) {
261
            goto fail_block;
262
        }
263

    
264
        bdrv_flush(bs->file);
265

    
266
        /* Initialize the new refcount block only after updating its refcount,
267
         * update_refcount uses the refcount cache itself */
268
        memset(s->refcount_block_cache, 0, s->cluster_size);
269
        s->refcount_block_cache_offset = new_block;
270
    }
271

    
272
    /* Now the new refcount block needs to be written to disk */
273
    BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE);
274
    ret = bdrv_pwrite_sync(bs->file, new_block, s->refcount_block_cache,
275
        s->cluster_size);
276
    if (ret < 0) {
277
        goto fail_block;
278
    }
279

    
280
    /* If the refcount table is big enough, just hook the block up there */
281
    if (refcount_table_index < s->refcount_table_size) {
282
        uint64_t data64 = cpu_to_be64(new_block);
283
        BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_HOOKUP);
284
        ret = bdrv_pwrite_sync(bs->file,
285
            s->refcount_table_offset + refcount_table_index * sizeof(uint64_t),
286
            &data64, sizeof(data64));
287
        if (ret < 0) {
288
            goto fail_block;
289
        }
290

    
291
        s->refcount_table[refcount_table_index] = new_block;
292
        return new_block;
293
    }
294

    
295
    /*
296
     * If we come here, we need to grow the refcount table. Again, a new
297
     * refcount table needs some space and we can't simply allocate to avoid
298
     * endless recursion.
299
     *
300
     * Therefore let's grab new refcount blocks at the end of the image, which
301
     * will describe themselves and the new refcount table. This way we can
302
     * reference them only in the new table and do the switch to the new
303
     * refcount table at once without producing an inconsistent state in
304
     * between.
305
     */
306
    BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_GROW);
307

    
308
    /* Calculate the number of refcount blocks needed so far */
309
    uint64_t refcount_block_clusters = 1 << (s->cluster_bits - REFCOUNT_SHIFT);
310
    uint64_t blocks_used = (s->free_cluster_index +
311
        refcount_block_clusters - 1) / refcount_block_clusters;
312

    
313
    /* And now we need at least one block more for the new metadata */
314
    uint64_t table_size = next_refcount_table_size(s, blocks_used + 1);
315
    uint64_t last_table_size;
316
    uint64_t blocks_clusters;
317
    do {
318
        uint64_t table_clusters = size_to_clusters(s, table_size);
319
        blocks_clusters = 1 +
320
            ((table_clusters + refcount_block_clusters - 1)
321
            / refcount_block_clusters);
322
        uint64_t meta_clusters = table_clusters + blocks_clusters;
323

    
324
        last_table_size = table_size;
325
        table_size = next_refcount_table_size(s, blocks_used +
326
            ((meta_clusters + refcount_block_clusters - 1)
327
            / refcount_block_clusters));
328

    
329
    } while (last_table_size != table_size);
330

    
331
#ifdef DEBUG_ALLOC2
332
    fprintf(stderr, "qcow2: Grow refcount table %" PRId32 " => %" PRId64 "\n",
333
        s->refcount_table_size, table_size);
334
#endif
335

    
336
    /* Create the new refcount table and blocks */
337
    uint64_t meta_offset = (blocks_used * refcount_block_clusters) *
338
        s->cluster_size;
339
    uint64_t table_offset = meta_offset + blocks_clusters * s->cluster_size;
340
    uint16_t *new_blocks = qemu_mallocz(blocks_clusters * s->cluster_size);
341
    uint64_t *new_table = qemu_mallocz(table_size * sizeof(uint64_t));
342

    
343
    assert(meta_offset >= (s->free_cluster_index * s->cluster_size));
344

    
345
    /* Fill the new refcount table */
346
    memcpy(new_table, s->refcount_table,
347
        s->refcount_table_size * sizeof(uint64_t));
348
    new_table[refcount_table_index] = new_block;
349

    
350
    int i;
351
    for (i = 0; i < blocks_clusters; i++) {
352
        new_table[blocks_used + i] = meta_offset + (i * s->cluster_size);
353
    }
354

    
355
    /* Fill the refcount blocks */
356
    uint64_t table_clusters = size_to_clusters(s, table_size * sizeof(uint64_t));
357
    int block = 0;
358
    for (i = 0; i < table_clusters + blocks_clusters; i++) {
359
        new_blocks[block++] = cpu_to_be16(1);
360
    }
361

    
362
    /* Write refcount blocks to disk */
363
    BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_BLOCKS);
364
    ret = bdrv_pwrite_sync(bs->file, meta_offset, new_blocks,
365
        blocks_clusters * s->cluster_size);
366
    qemu_free(new_blocks);
367
    if (ret < 0) {
368
        goto fail_table;
369
    }
370

    
371
    /* Write refcount table to disk */
372
    for(i = 0; i < table_size; i++) {
373
        cpu_to_be64s(&new_table[i]);
374
    }
375

    
376
    BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_TABLE);
377
    ret = bdrv_pwrite_sync(bs->file, table_offset, new_table,
378
        table_size * sizeof(uint64_t));
379
    if (ret < 0) {
380
        goto fail_table;
381
    }
382

    
383
    for(i = 0; i < table_size; i++) {
384
        cpu_to_be64s(&new_table[i]);
385
    }
386

    
387
    /* Hook up the new refcount table in the qcow2 header */
388
    uint8_t data[12];
389
    cpu_to_be64w((uint64_t*)data, table_offset);
390
    cpu_to_be32w((uint32_t*)(data + 8), table_clusters);
391
    BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_SWITCH_TABLE);
392
    ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, refcount_table_offset),
393
        data, sizeof(data));
394
    if (ret < 0) {
395
        goto fail_table;
396
    }
397

    
398
    /* And switch it in memory */
399
    uint64_t old_table_offset = s->refcount_table_offset;
400
    uint64_t old_table_size = s->refcount_table_size;
401

    
402
    qemu_free(s->refcount_table);
403
    s->refcount_table = new_table;
404
    s->refcount_table_size = table_size;
405
    s->refcount_table_offset = table_offset;
406

    
407
    /* Free old table. Remember, we must not change free_cluster_index */
408
    uint64_t old_free_cluster_index = s->free_cluster_index;
409
    qcow2_free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t));
410
    s->free_cluster_index = old_free_cluster_index;
411

    
412
    ret = load_refcount_block(bs, new_block);
413
    if (ret < 0) {
414
        goto fail_block;
415
    }
416

    
417
    return new_block;
418

    
419
fail_table:
420
    qemu_free(new_table);
421
fail_block:
422
    s->refcount_block_cache_offset = 0;
423
    return ret;
424
}
425

    
426
#define REFCOUNTS_PER_SECTOR (512 >> REFCOUNT_SHIFT)
427
static int write_refcount_block_entries(BlockDriverState *bs,
428
    int64_t refcount_block_offset, int first_index, int last_index)
429
{
430
    BDRVQcowState *s = bs->opaque;
431
    size_t size;
432
    int ret;
433

    
434
    if (cache_refcount_updates) {
435
        return 0;
436
    }
437

    
438
    if (first_index < 0) {
439
        return 0;
440
    }
441

    
442
    first_index &= ~(REFCOUNTS_PER_SECTOR - 1);
443
    last_index = (last_index + REFCOUNTS_PER_SECTOR)
444
        & ~(REFCOUNTS_PER_SECTOR - 1);
445

    
446
    size = (last_index - first_index) << REFCOUNT_SHIFT;
447

    
448
    BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_UPDATE_PART);
449
    ret = bdrv_pwrite(bs->file,
450
        refcount_block_offset + (first_index << REFCOUNT_SHIFT),
451
        &s->refcount_block_cache[first_index], size);
452
    if (ret < 0) {
453
        return ret;
454
    }
455

    
456
    return 0;
457
}
458

    
459
/* XXX: cache several refcount block clusters ? */
460
static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
461
    int64_t offset, int64_t length, int addend)
462
{
463
    BDRVQcowState *s = bs->opaque;
464
    int64_t start, last, cluster_offset;
465
    int64_t refcount_block_offset = 0;
466
    int64_t table_index = -1, old_table_index;
467
    int first_index = -1, last_index = -1;
468
    int ret;
469

    
470
#ifdef DEBUG_ALLOC2
471
    printf("update_refcount: offset=%" PRId64 " size=%" PRId64 " addend=%d\n",
472
           offset, length, addend);
473
#endif
474
    if (length < 0) {
475
        return -EINVAL;
476
    } else if (length == 0) {
477
        return 0;
478
    }
479

    
480
    start = offset & ~(s->cluster_size - 1);
481
    last = (offset + length - 1) & ~(s->cluster_size - 1);
482
    for(cluster_offset = start; cluster_offset <= last;
483
        cluster_offset += s->cluster_size)
484
    {
485
        int block_index, refcount;
486
        int64_t cluster_index = cluster_offset >> s->cluster_bits;
487
        int64_t new_block;
488

    
489
        /* Only write refcount block to disk when we are done with it */
490
        old_table_index = table_index;
491
        table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
492
        if ((old_table_index >= 0) && (table_index != old_table_index)) {
493

    
494
            ret = write_refcount_block_entries(bs, refcount_block_offset,
495
                first_index, last_index);
496
            if (ret < 0) {
497
                return ret;
498
            }
499

    
500
            first_index = -1;
501
            last_index = -1;
502
        }
503

    
504
        /* Load the refcount block and allocate it if needed */
505
        new_block = alloc_refcount_block(bs, cluster_index);
506
        if (new_block < 0) {
507
            ret = new_block;
508
            goto fail;
509
        }
510
        refcount_block_offset = new_block;
511

    
512
        /* we can update the count and save it */
513
        block_index = cluster_index &
514
            ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
515
        if (first_index == -1 || block_index < first_index) {
516
            first_index = block_index;
517
        }
518
        if (block_index > last_index) {
519
            last_index = block_index;
520
        }
521

    
522
        refcount = be16_to_cpu(s->refcount_block_cache[block_index]);
523
        refcount += addend;
524
        if (refcount < 0 || refcount > 0xffff) {
525
            ret = -EINVAL;
526
            goto fail;
527
        }
528
        if (refcount == 0 && cluster_index < s->free_cluster_index) {
529
            s->free_cluster_index = cluster_index;
530
        }
531
        s->refcount_block_cache[block_index] = cpu_to_be16(refcount);
532
    }
533

    
534
    ret = 0;
535
fail:
536

    
537
    /* Write last changed block to disk */
538
    if (refcount_block_offset != 0) {
539
        int wret;
540
        wret = write_refcount_block_entries(bs, refcount_block_offset,
541
            first_index, last_index);
542
        if (wret < 0) {
543
            return ret < 0 ? ret : wret;
544
        }
545
    }
546

    
547
    /*
548
     * Try do undo any updates if an error is returned (This may succeed in
549
     * some cases like ENOSPC for allocating a new refcount block)
550
     */
551
    if (ret < 0) {
552
        int dummy;
553
        dummy = update_refcount(bs, offset, cluster_offset - offset, -addend);
554
        (void)dummy;
555
    }
556

    
557
    return ret;
558
}
559

    
560
/*
561
 * Increases or decreases the refcount of a given cluster by one.
562
 * addend must be 1 or -1.
563
 *
564
 * If the return value is non-negative, it is the new refcount of the cluster.
565
 * If it is negative, it is -errno and indicates an error.
566
 */
567
static int update_cluster_refcount(BlockDriverState *bs,
568
                                   int64_t cluster_index,
569
                                   int addend)
570
{
571
    BDRVQcowState *s = bs->opaque;
572
    int ret;
573

    
574
    ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend);
575
    if (ret < 0) {
576
        return ret;
577
    }
578

    
579
    bdrv_flush(bs->file);
580

    
581
    return get_refcount(bs, cluster_index);
582
}
583

    
584

    
585

    
586
/*********************************************************/
587
/* cluster allocation functions */
588

    
589

    
590

    
591
/* return < 0 if error */
592
static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size)
593
{
594
    BDRVQcowState *s = bs->opaque;
595
    int i, nb_clusters, refcount;
596

    
597
    nb_clusters = size_to_clusters(s, size);
598
retry:
599
    for(i = 0; i < nb_clusters; i++) {
600
        int64_t next_cluster_index = s->free_cluster_index++;
601
        refcount = get_refcount(bs, next_cluster_index);
602

    
603
        if (refcount < 0) {
604
            return refcount;
605
        } else if (refcount != 0) {
606
            goto retry;
607
        }
608
    }
609
#ifdef DEBUG_ALLOC2
610
    printf("alloc_clusters: size=%" PRId64 " -> %" PRId64 "\n",
611
            size,
612
            (s->free_cluster_index - nb_clusters) << s->cluster_bits);
613
#endif
614
    return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
615
}
616

    
617
int64_t qcow2_alloc_clusters(BlockDriverState *bs, int64_t size)
618
{
619
    int64_t offset;
620
    int ret;
621

    
622
    BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
623
    offset = alloc_clusters_noref(bs, size);
624
    if (offset < 0) {
625
        return offset;
626
    }
627

    
628
    ret = update_refcount(bs, offset, size, 1);
629
    if (ret < 0) {
630
        return ret;
631
    }
632

    
633
    return offset;
634
}
635

    
636
/* only used to allocate compressed sectors. We try to allocate
637
   contiguous sectors. size must be <= cluster_size */
638
int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size)
639
{
640
    BDRVQcowState *s = bs->opaque;
641
    int64_t offset, cluster_offset;
642
    int free_in_cluster;
643

    
644
    BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_BYTES);
645
    assert(size > 0 && size <= s->cluster_size);
646
    if (s->free_byte_offset == 0) {
647
        s->free_byte_offset = qcow2_alloc_clusters(bs, s->cluster_size);
648
        if (s->free_byte_offset < 0) {
649
            return s->free_byte_offset;
650
        }
651
    }
652
 redo:
653
    free_in_cluster = s->cluster_size -
654
        (s->free_byte_offset & (s->cluster_size - 1));
655
    if (size <= free_in_cluster) {
656
        /* enough space in current cluster */
657
        offset = s->free_byte_offset;
658
        s->free_byte_offset += size;
659
        free_in_cluster -= size;
660
        if (free_in_cluster == 0)
661
            s->free_byte_offset = 0;
662
        if ((offset & (s->cluster_size - 1)) != 0)
663
            update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
664
    } else {
665
        offset = qcow2_alloc_clusters(bs, s->cluster_size);
666
        if (offset < 0) {
667
            return offset;
668
        }
669
        cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1);
670
        if ((cluster_offset + s->cluster_size) == offset) {
671
            /* we are lucky: contiguous data */
672
            offset = s->free_byte_offset;
673
            update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
674
            s->free_byte_offset += size;
675
        } else {
676
            s->free_byte_offset = offset;
677
            goto redo;
678
        }
679
    }
680

    
681
    bdrv_flush(bs->file);
682
    return offset;
683
}
684

    
685
void qcow2_free_clusters(BlockDriverState *bs,
686
                          int64_t offset, int64_t size)
687
{
688
    int ret;
689

    
690
    BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_FREE);
691
    ret = update_refcount(bs, offset, size, -1);
692
    if (ret < 0) {
693
        fprintf(stderr, "qcow2_free_clusters failed: %s\n", strerror(-ret));
694
        /* TODO Remember the clusters to free them later and avoid leaking */
695
    }
696
}
697

    
698
/*
699
 * free_any_clusters
700
 *
701
 * free clusters according to its type: compressed or not
702
 *
703
 */
704

    
705
void qcow2_free_any_clusters(BlockDriverState *bs,
706
    uint64_t cluster_offset, int nb_clusters)
707
{
708
    BDRVQcowState *s = bs->opaque;
709

    
710
    /* free the cluster */
711

    
712
    if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
713
        int nb_csectors;
714
        nb_csectors = ((cluster_offset >> s->csize_shift) &
715
                       s->csize_mask) + 1;
716
        qcow2_free_clusters(bs,
717
            (cluster_offset & s->cluster_offset_mask) & ~511,
718
            nb_csectors * 512);
719
        return;
720
    }
721

    
722
    qcow2_free_clusters(bs, cluster_offset, nb_clusters << s->cluster_bits);
723

    
724
    return;
725
}
726

    
727

    
728

    
729
/*********************************************************/
730
/* snapshots and image creation */
731

    
732

    
733

    
734
void qcow2_create_refcount_update(QCowCreateState *s, int64_t offset,
735
    int64_t size)
736
{
737
    int refcount;
738
    int64_t start, last, cluster_offset;
739
    uint16_t *p;
740

    
741
    start = offset & ~(s->cluster_size - 1);
742
    last = (offset + size - 1)  & ~(s->cluster_size - 1);
743
    for(cluster_offset = start; cluster_offset <= last;
744
        cluster_offset += s->cluster_size) {
745
        p = &s->refcount_block[cluster_offset >> s->cluster_bits];
746
        refcount = be16_to_cpu(*p);
747
        refcount++;
748
        *p = cpu_to_be16(refcount);
749
    }
750
}
751

    
752
/* update the refcounts of snapshots and the copied flag */
753
int qcow2_update_snapshot_refcount(BlockDriverState *bs,
754
    int64_t l1_table_offset, int l1_size, int addend)
755
{
756
    BDRVQcowState *s = bs->opaque;
757
    uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated;
758
    int64_t old_offset, old_l2_offset;
759
    int l2_size, i, j, l1_modified, l2_modified, nb_csectors, refcount;
760

    
761
    qcow2_l2_cache_reset(bs);
762
    cache_refcount_updates = 1;
763

    
764
    l2_table = NULL;
765
    l1_table = NULL;
766
    l1_size2 = l1_size * sizeof(uint64_t);
767
    if (l1_table_offset != s->l1_table_offset) {
768
        if (l1_size2 != 0) {
769
            l1_table = qemu_mallocz(align_offset(l1_size2, 512));
770
        } else {
771
            l1_table = NULL;
772
        }
773
        l1_allocated = 1;
774
        if (bdrv_pread(bs->file, l1_table_offset,
775
                       l1_table, l1_size2) != l1_size2)
776
            goto fail;
777
        for(i = 0;i < l1_size; i++)
778
            be64_to_cpus(&l1_table[i]);
779
    } else {
780
        assert(l1_size == s->l1_size);
781
        l1_table = s->l1_table;
782
        l1_allocated = 0;
783
    }
784

    
785
    l2_size = s->l2_size * sizeof(uint64_t);
786
    l2_table = qemu_malloc(l2_size);
787
    l1_modified = 0;
788
    for(i = 0; i < l1_size; i++) {
789
        l2_offset = l1_table[i];
790
        if (l2_offset) {
791
            old_l2_offset = l2_offset;
792
            l2_offset &= ~QCOW_OFLAG_COPIED;
793
            l2_modified = 0;
794
            if (bdrv_pread(bs->file, l2_offset, l2_table, l2_size) != l2_size)
795
                goto fail;
796
            for(j = 0; j < s->l2_size; j++) {
797
                offset = be64_to_cpu(l2_table[j]);
798
                if (offset != 0) {
799
                    old_offset = offset;
800
                    offset &= ~QCOW_OFLAG_COPIED;
801
                    if (offset & QCOW_OFLAG_COMPRESSED) {
802
                        nb_csectors = ((offset >> s->csize_shift) &
803
                                       s->csize_mask) + 1;
804
                        if (addend != 0) {
805
                            int ret;
806
                            ret = update_refcount(bs,
807
                                (offset & s->cluster_offset_mask) & ~511,
808
                                nb_csectors * 512, addend);
809
                            if (ret < 0) {
810
                                goto fail;
811
                            }
812

    
813
                            /* TODO Flushing once for the whole function should
814
                             * be enough */
815
                            bdrv_flush(bs->file);
816
                        }
817
                        /* compressed clusters are never modified */
818
                        refcount = 2;
819
                    } else {
820
                        if (addend != 0) {
821
                            refcount = update_cluster_refcount(bs, offset >> s->cluster_bits, addend);
822
                        } else {
823
                            refcount = get_refcount(bs, offset >> s->cluster_bits);
824
                        }
825

    
826
                        if (refcount < 0) {
827
                            goto fail;
828
                        }
829
                    }
830

    
831
                    if (refcount == 1) {
832
                        offset |= QCOW_OFLAG_COPIED;
833
                    }
834
                    if (offset != old_offset) {
835
                        l2_table[j] = cpu_to_be64(offset);
836
                        l2_modified = 1;
837
                    }
838
                }
839
            }
840
            if (l2_modified) {
841
                if (bdrv_pwrite_sync(bs->file,
842
                                l2_offset, l2_table, l2_size) < 0)
843
                    goto fail;
844
            }
845

    
846
            if (addend != 0) {
847
                refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend);
848
            } else {
849
                refcount = get_refcount(bs, l2_offset >> s->cluster_bits);
850
            }
851
            if (refcount < 0) {
852
                goto fail;
853
            } else if (refcount == 1) {
854
                l2_offset |= QCOW_OFLAG_COPIED;
855
            }
856
            if (l2_offset != old_l2_offset) {
857
                l1_table[i] = l2_offset;
858
                l1_modified = 1;
859
            }
860
        }
861
    }
862
    if (l1_modified) {
863
        for(i = 0; i < l1_size; i++)
864
            cpu_to_be64s(&l1_table[i]);
865
        if (bdrv_pwrite_sync(bs->file, l1_table_offset, l1_table,
866
                        l1_size2) < 0)
867
            goto fail;
868
        for(i = 0; i < l1_size; i++)
869
            be64_to_cpus(&l1_table[i]);
870
    }
871
    if (l1_allocated)
872
        qemu_free(l1_table);
873
    qemu_free(l2_table);
874
    cache_refcount_updates = 0;
875
    write_refcount_block(bs);
876
    return 0;
877
 fail:
878
    if (l1_allocated)
879
        qemu_free(l1_table);
880
    qemu_free(l2_table);
881
    cache_refcount_updates = 0;
882
    write_refcount_block(bs);
883
    return -EIO;
884
}
885

    
886

    
887

    
888

    
889
/*********************************************************/
890
/* refcount checking functions */
891

    
892

    
893

    
894
/*
895
 * Increases the refcount for a range of clusters in a given refcount table.
896
 * This is used to construct a temporary refcount table out of L1 and L2 tables
897
 * which can be compared the the refcount table saved in the image.
898
 *
899
 * Modifies the number of errors in res.
900
 */
901
static void inc_refcounts(BlockDriverState *bs,
902
                          BdrvCheckResult *res,
903
                          uint16_t *refcount_table,
904
                          int refcount_table_size,
905
                          int64_t offset, int64_t size)
906
{
907
    BDRVQcowState *s = bs->opaque;
908
    int64_t start, last, cluster_offset;
909
    int k;
910

    
911
    if (size <= 0)
912
        return;
913

    
914
    start = offset & ~(s->cluster_size - 1);
915
    last = (offset + size - 1) & ~(s->cluster_size - 1);
916
    for(cluster_offset = start; cluster_offset <= last;
917
        cluster_offset += s->cluster_size) {
918
        k = cluster_offset >> s->cluster_bits;
919
        if (k < 0) {
920
            fprintf(stderr, "ERROR: invalid cluster offset=0x%" PRIx64 "\n",
921
                cluster_offset);
922
            res->corruptions++;
923
        } else if (k >= refcount_table_size) {
924
            fprintf(stderr, "Warning: cluster offset=0x%" PRIx64 " is after "
925
                "the end of the image file, can't properly check refcounts.\n",
926
                cluster_offset);
927
            res->check_errors++;
928
        } else {
929
            if (++refcount_table[k] == 0) {
930
                fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64
931
                    "\n", cluster_offset);
932
                res->corruptions++;
933
            }
934
        }
935
    }
936
}
937

    
938
/*
939
 * Increases the refcount in the given refcount table for the all clusters
940
 * referenced in the L2 table. While doing so, performs some checks on L2
941
 * entries.
942
 *
943
 * Returns the number of errors found by the checks or -errno if an internal
944
 * error occurred.
945
 */
946
static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res,
947
    uint16_t *refcount_table, int refcount_table_size, int64_t l2_offset,
948
    int check_copied)
949
{
950
    BDRVQcowState *s = bs->opaque;
951
    uint64_t *l2_table, offset;
952
    int i, l2_size, nb_csectors, refcount;
953

    
954
    /* Read L2 table from disk */
955
    l2_size = s->l2_size * sizeof(uint64_t);
956
    l2_table = qemu_malloc(l2_size);
957

    
958
    if (bdrv_pread(bs->file, l2_offset, l2_table, l2_size) != l2_size)
959
        goto fail;
960

    
961
    /* Do the actual checks */
962
    for(i = 0; i < s->l2_size; i++) {
963
        offset = be64_to_cpu(l2_table[i]);
964
        if (offset != 0) {
965
            if (offset & QCOW_OFLAG_COMPRESSED) {
966
                /* Compressed clusters don't have QCOW_OFLAG_COPIED */
967
                if (offset & QCOW_OFLAG_COPIED) {
968
                    fprintf(stderr, "ERROR: cluster %" PRId64 ": "
969
                        "copied flag must never be set for compressed "
970
                        "clusters\n", offset >> s->cluster_bits);
971
                    offset &= ~QCOW_OFLAG_COPIED;
972
                    res->corruptions++;
973
                }
974

    
975
                /* Mark cluster as used */
976
                nb_csectors = ((offset >> s->csize_shift) &
977
                               s->csize_mask) + 1;
978
                offset &= s->cluster_offset_mask;
979
                inc_refcounts(bs, res, refcount_table, refcount_table_size,
980
                    offset & ~511, nb_csectors * 512);
981
            } else {
982
                /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
983
                if (check_copied) {
984
                    uint64_t entry = offset;
985
                    offset &= ~QCOW_OFLAG_COPIED;
986
                    refcount = get_refcount(bs, offset >> s->cluster_bits);
987
                    if (refcount < 0) {
988
                        fprintf(stderr, "Can't get refcount for offset %"
989
                            PRIx64 ": %s\n", entry, strerror(-refcount));
990
                        goto fail;
991
                    }
992
                    if ((refcount == 1) != ((entry & QCOW_OFLAG_COPIED) != 0)) {
993
                        fprintf(stderr, "ERROR OFLAG_COPIED: offset=%"
994
                            PRIx64 " refcount=%d\n", entry, refcount);
995
                        res->corruptions++;
996
                    }
997
                }
998

    
999
                /* Mark cluster as used */
1000
                offset &= ~QCOW_OFLAG_COPIED;
1001
                inc_refcounts(bs, res, refcount_table,refcount_table_size,
1002
                    offset, s->cluster_size);
1003

    
1004
                /* Correct offsets are cluster aligned */
1005
                if (offset & (s->cluster_size - 1)) {
1006
                    fprintf(stderr, "ERROR offset=%" PRIx64 ": Cluster is not "
1007
                        "properly aligned; L2 entry corrupted.\n", offset);
1008
                    res->corruptions++;
1009
                }
1010
            }
1011
        }
1012
    }
1013

    
1014
    qemu_free(l2_table);
1015
    return 0;
1016

    
1017
fail:
1018
    fprintf(stderr, "ERROR: I/O error in check_refcounts_l2\n");
1019
    qemu_free(l2_table);
1020
    return -EIO;
1021
}
1022

    
1023
/*
1024
 * Increases the refcount for the L1 table, its L2 tables and all referenced
1025
 * clusters in the given refcount table. While doing so, performs some checks
1026
 * on L1 and L2 entries.
1027
 *
1028
 * Returns the number of errors found by the checks or -errno if an internal
1029
 * error occurred.
1030
 */
1031
static int check_refcounts_l1(BlockDriverState *bs,
1032
                              BdrvCheckResult *res,
1033
                              uint16_t *refcount_table,
1034
                              int refcount_table_size,
1035
                              int64_t l1_table_offset, int l1_size,
1036
                              int check_copied)
1037
{
1038
    BDRVQcowState *s = bs->opaque;
1039
    uint64_t *l1_table, l2_offset, l1_size2;
1040
    int i, refcount, ret;
1041

    
1042
    l1_size2 = l1_size * sizeof(uint64_t);
1043

    
1044
    /* Mark L1 table as used */
1045
    inc_refcounts(bs, res, refcount_table, refcount_table_size,
1046
        l1_table_offset, l1_size2);
1047

    
1048
    /* Read L1 table entries from disk */
1049
    if (l1_size2 == 0) {
1050
        l1_table = NULL;
1051
    } else {
1052
        l1_table = qemu_malloc(l1_size2);
1053
        if (bdrv_pread(bs->file, l1_table_offset,
1054
                       l1_table, l1_size2) != l1_size2)
1055
            goto fail;
1056
        for(i = 0;i < l1_size; i++)
1057
            be64_to_cpus(&l1_table[i]);
1058
    }
1059

    
1060
    /* Do the actual checks */
1061
    for(i = 0; i < l1_size; i++) {
1062
        l2_offset = l1_table[i];
1063
        if (l2_offset) {
1064
            /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
1065
            if (check_copied) {
1066
                refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED)
1067
                    >> s->cluster_bits);
1068
                if (refcount < 0) {
1069
                    fprintf(stderr, "Can't get refcount for l2_offset %"
1070
                        PRIx64 ": %s\n", l2_offset, strerror(-refcount));
1071
                    goto fail;
1072
                }
1073
                if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) {
1074
                    fprintf(stderr, "ERROR OFLAG_COPIED: l2_offset=%" PRIx64
1075
                        " refcount=%d\n", l2_offset, refcount);
1076
                    res->corruptions++;
1077
                }
1078
            }
1079

    
1080
            /* Mark L2 table as used */
1081
            l2_offset &= ~QCOW_OFLAG_COPIED;
1082
            inc_refcounts(bs, res, refcount_table, refcount_table_size,
1083
                l2_offset, s->cluster_size);
1084

    
1085
            /* L2 tables are cluster aligned */
1086
            if (l2_offset & (s->cluster_size - 1)) {
1087
                fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not "
1088
                    "cluster aligned; L1 entry corrupted\n", l2_offset);
1089
                res->corruptions++;
1090
            }
1091

    
1092
            /* Process and check L2 entries */
1093
            ret = check_refcounts_l2(bs, res, refcount_table,
1094
                refcount_table_size, l2_offset, check_copied);
1095
            if (ret < 0) {
1096
                goto fail;
1097
            }
1098
        }
1099
    }
1100
    qemu_free(l1_table);
1101
    return 0;
1102

    
1103
fail:
1104
    fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
1105
    res->check_errors++;
1106
    qemu_free(l1_table);
1107
    return -EIO;
1108
}
1109

    
1110
/*
1111
 * Checks an image for refcount consistency.
1112
 *
1113
 * Returns 0 if no errors are found, the number of errors in case the image is
1114
 * detected as corrupted, and -errno when an internal error occured.
1115
 */
1116
int qcow2_check_refcounts(BlockDriverState *bs, BdrvCheckResult *res)
1117
{
1118
    BDRVQcowState *s = bs->opaque;
1119
    int64_t size;
1120
    int nb_clusters, refcount1, refcount2, i;
1121
    QCowSnapshot *sn;
1122
    uint16_t *refcount_table;
1123
    int ret;
1124

    
1125
    size = bdrv_getlength(bs->file);
1126
    nb_clusters = size_to_clusters(s, size);
1127
    refcount_table = qemu_mallocz(nb_clusters * sizeof(uint16_t));
1128

    
1129
    /* header */
1130
    inc_refcounts(bs, res, refcount_table, nb_clusters,
1131
        0, s->cluster_size);
1132

    
1133
    /* current L1 table */
1134
    ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
1135
                       s->l1_table_offset, s->l1_size, 1);
1136
    if (ret < 0) {
1137
        return ret;
1138
    }
1139

    
1140
    /* snapshots */
1141
    for(i = 0; i < s->nb_snapshots; i++) {
1142
        sn = s->snapshots + i;
1143
        ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
1144
            sn->l1_table_offset, sn->l1_size, 0);
1145
        if (ret < 0) {
1146
            return ret;
1147
        }
1148
    }
1149
    inc_refcounts(bs, res, refcount_table, nb_clusters,
1150
        s->snapshots_offset, s->snapshots_size);
1151

    
1152
    /* refcount data */
1153
    inc_refcounts(bs, res, refcount_table, nb_clusters,
1154
        s->refcount_table_offset,
1155
        s->refcount_table_size * sizeof(uint64_t));
1156

    
1157
    for(i = 0; i < s->refcount_table_size; i++) {
1158
        uint64_t offset, cluster;
1159
        offset = s->refcount_table[i];
1160
        cluster = offset >> s->cluster_bits;
1161

    
1162
        /* Refcount blocks are cluster aligned */
1163
        if (offset & (s->cluster_size - 1)) {
1164
            fprintf(stderr, "ERROR refcount block %d is not "
1165
                "cluster aligned; refcount table entry corrupted\n", i);
1166
            res->corruptions++;
1167
            continue;
1168
        }
1169

    
1170
        if (cluster >= nb_clusters) {
1171
            fprintf(stderr, "ERROR refcount block %d is outside image\n", i);
1172
            res->corruptions++;
1173
            continue;
1174
        }
1175

    
1176
        if (offset != 0) {
1177
            inc_refcounts(bs, res, refcount_table, nb_clusters,
1178
                offset, s->cluster_size);
1179
            if (refcount_table[cluster] != 1) {
1180
                fprintf(stderr, "ERROR refcount block %d refcount=%d\n",
1181
                    i, refcount_table[cluster]);
1182
                res->corruptions++;
1183
            }
1184
        }
1185
    }
1186

    
1187
    /* compare ref counts */
1188
    for(i = 0; i < nb_clusters; i++) {
1189
        refcount1 = get_refcount(bs, i);
1190
        if (refcount1 < 0) {
1191
            fprintf(stderr, "Can't get refcount for cluster %d: %s\n",
1192
                i, strerror(-refcount1));
1193
            res->check_errors++;
1194
            continue;
1195
        }
1196

    
1197
        refcount2 = refcount_table[i];
1198
        if (refcount1 != refcount2) {
1199
            fprintf(stderr, "%s cluster %d refcount=%d reference=%d\n",
1200
                   refcount1 < refcount2 ? "ERROR" : "Leaked",
1201
                   i, refcount1, refcount2);
1202
            if (refcount1 < refcount2) {
1203
                res->corruptions++;
1204
            } else {
1205
                res->leaks++;
1206
            }
1207
        }
1208
    }
1209

    
1210
    qemu_free(refcount_table);
1211

    
1212
    return 0;
1213
}
1214