Statistics
| Branch: | Revision:

root / block / qcow2-cluster.c @ 2dedf83e

History | View | Annotate | Download (25.5 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 <zlib.h>
26

    
27
#include "qemu-common.h"
28
#include "block_int.h"
29
#include "block/qcow2.h"
30

    
31
int qcow2_grow_l1_table(BlockDriverState *bs, int min_size)
32
{
33
    BDRVQcowState *s = bs->opaque;
34
    int new_l1_size, new_l1_size2, ret, i;
35
    uint64_t *new_l1_table;
36
    int64_t new_l1_table_offset;
37
    uint8_t data[12];
38

    
39
    new_l1_size = s->l1_size;
40
    if (min_size <= new_l1_size)
41
        return 0;
42
    if (new_l1_size == 0) {
43
        new_l1_size = 1;
44
    }
45
    while (min_size > new_l1_size) {
46
        new_l1_size = (new_l1_size * 3 + 1) / 2;
47
    }
48
#ifdef DEBUG_ALLOC2
49
    printf("grow l1_table from %d to %d\n", s->l1_size, new_l1_size);
50
#endif
51

    
52
    new_l1_size2 = sizeof(uint64_t) * new_l1_size;
53
    new_l1_table = qemu_mallocz(align_offset(new_l1_size2, 512));
54
    memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t));
55

    
56
    /* write new table (align to cluster) */
57
    new_l1_table_offset = qcow2_alloc_clusters(bs, new_l1_size2);
58
    if (new_l1_table_offset < 0) {
59
        qemu_free(new_l1_table);
60
        return new_l1_table_offset;
61
    }
62

    
63
    for(i = 0; i < s->l1_size; i++)
64
        new_l1_table[i] = cpu_to_be64(new_l1_table[i]);
65
    ret = bdrv_pwrite(s->hd, new_l1_table_offset, new_l1_table, new_l1_size2);
66
    if (ret != new_l1_size2)
67
        goto fail;
68
    for(i = 0; i < s->l1_size; i++)
69
        new_l1_table[i] = be64_to_cpu(new_l1_table[i]);
70

    
71
    /* set new table */
72
    cpu_to_be32w((uint32_t*)data, new_l1_size);
73
    cpu_to_be64w((uint64_t*)(data + 4), new_l1_table_offset);
74
    ret = bdrv_pwrite(s->hd, offsetof(QCowHeader, l1_size), data,sizeof(data));
75
    if (ret != sizeof(data)) {
76
        goto fail;
77
    }
78
    qemu_free(s->l1_table);
79
    qcow2_free_clusters(bs, s->l1_table_offset, s->l1_size * sizeof(uint64_t));
80
    s->l1_table_offset = new_l1_table_offset;
81
    s->l1_table = new_l1_table;
82
    s->l1_size = new_l1_size;
83
    return 0;
84
 fail:
85
    qemu_free(new_l1_table);
86
    qcow2_free_clusters(bs, new_l1_table_offset, new_l1_size2);
87
    return ret < 0 ? ret : -EIO;
88
}
89

    
90
void qcow2_l2_cache_reset(BlockDriverState *bs)
91
{
92
    BDRVQcowState *s = bs->opaque;
93

    
94
    memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
95
    memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
96
    memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
97
}
98

    
99
static inline int l2_cache_new_entry(BlockDriverState *bs)
100
{
101
    BDRVQcowState *s = bs->opaque;
102
    uint32_t min_count;
103
    int min_index, i;
104

    
105
    /* find a new entry in the least used one */
106
    min_index = 0;
107
    min_count = 0xffffffff;
108
    for(i = 0; i < L2_CACHE_SIZE; i++) {
109
        if (s->l2_cache_counts[i] < min_count) {
110
            min_count = s->l2_cache_counts[i];
111
            min_index = i;
112
        }
113
    }
114
    return min_index;
115
}
116

    
117
/*
118
 * seek_l2_table
119
 *
120
 * seek l2_offset in the l2_cache table
121
 * if not found, return NULL,
122
 * if found,
123
 *   increments the l2 cache hit count of the entry,
124
 *   if counter overflow, divide by two all counters
125
 *   return the pointer to the l2 cache entry
126
 *
127
 */
128

    
129
static uint64_t *seek_l2_table(BDRVQcowState *s, uint64_t l2_offset)
130
{
131
    int i, j;
132

    
133
    for(i = 0; i < L2_CACHE_SIZE; i++) {
134
        if (l2_offset == s->l2_cache_offsets[i]) {
135
            /* increment the hit count */
136
            if (++s->l2_cache_counts[i] == 0xffffffff) {
137
                for(j = 0; j < L2_CACHE_SIZE; j++) {
138
                    s->l2_cache_counts[j] >>= 1;
139
                }
140
            }
141
            return s->l2_cache + (i << s->l2_bits);
142
        }
143
    }
144
    return NULL;
145
}
146

    
147
/*
148
 * l2_load
149
 *
150
 * Loads a L2 table into memory. If the table is in the cache, the cache
151
 * is used; otherwise the L2 table is loaded from the image file.
152
 *
153
 * Returns a pointer to the L2 table on success, or NULL if the read from
154
 * the image file failed.
155
 */
156

    
157
static uint64_t *l2_load(BlockDriverState *bs, uint64_t l2_offset)
158
{
159
    BDRVQcowState *s = bs->opaque;
160
    int min_index;
161
    uint64_t *l2_table;
162

    
163
    /* seek if the table for the given offset is in the cache */
164

    
165
    l2_table = seek_l2_table(s, l2_offset);
166
    if (l2_table != NULL)
167
        return l2_table;
168

    
169
    /* not found: load a new entry in the least used one */
170

    
171
    min_index = l2_cache_new_entry(bs);
172
    l2_table = s->l2_cache + (min_index << s->l2_bits);
173
    if (bdrv_pread(s->hd, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
174
        s->l2_size * sizeof(uint64_t))
175
        return NULL;
176
    s->l2_cache_offsets[min_index] = l2_offset;
177
    s->l2_cache_counts[min_index] = 1;
178

    
179
    return l2_table;
180
}
181

    
182
/*
183
 * Writes one sector of the L1 table to the disk (can't update single entries
184
 * and we really don't want bdrv_pread to perform a read-modify-write)
185
 */
186
#define L1_ENTRIES_PER_SECTOR (512 / 8)
187
static int write_l1_entry(BDRVQcowState *s, int l1_index)
188
{
189
    uint64_t buf[L1_ENTRIES_PER_SECTOR];
190
    int l1_start_index;
191
    int i;
192

    
193
    l1_start_index = l1_index & ~(L1_ENTRIES_PER_SECTOR - 1);
194
    for (i = 0; i < L1_ENTRIES_PER_SECTOR; i++) {
195
        buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]);
196
    }
197

    
198
    if (bdrv_pwrite(s->hd, s->l1_table_offset + 8 * l1_start_index,
199
        buf, sizeof(buf)) != sizeof(buf))
200
    {
201
        return -1;
202
    }
203

    
204
    return 0;
205
}
206

    
207
/*
208
 * l2_allocate
209
 *
210
 * Allocate a new l2 entry in the file. If l1_index points to an already
211
 * used entry in the L2 table (i.e. we are doing a copy on write for the L2
212
 * table) copy the contents of the old L2 table into the newly allocated one.
213
 * Otherwise the new table is initialized with zeros.
214
 *
215
 */
216

    
217
static uint64_t *l2_allocate(BlockDriverState *bs, int l1_index)
218
{
219
    BDRVQcowState *s = bs->opaque;
220
    int min_index;
221
    uint64_t old_l2_offset;
222
    uint64_t *l2_table, l2_offset;
223

    
224
    old_l2_offset = s->l1_table[l1_index];
225

    
226
    /* allocate a new l2 entry */
227

    
228
    l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
229
    if (l2_offset < 0) {
230
        return NULL;
231
    }
232

    
233
    /* update the L1 entry */
234

    
235
    s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
236
    if (write_l1_entry(s, l1_index) < 0) {
237
        return NULL;
238
    }
239

    
240
    /* allocate a new entry in the l2 cache */
241

    
242
    min_index = l2_cache_new_entry(bs);
243
    l2_table = s->l2_cache + (min_index << s->l2_bits);
244

    
245
    if (old_l2_offset == 0) {
246
        /* if there was no old l2 table, clear the new table */
247
        memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
248
    } else {
249
        /* if there was an old l2 table, read it from the disk */
250
        if (bdrv_pread(s->hd, old_l2_offset,
251
                       l2_table, s->l2_size * sizeof(uint64_t)) !=
252
            s->l2_size * sizeof(uint64_t))
253
            return NULL;
254
    }
255
    /* write the l2 table to the file */
256
    if (bdrv_pwrite(s->hd, l2_offset,
257
                    l2_table, s->l2_size * sizeof(uint64_t)) !=
258
        s->l2_size * sizeof(uint64_t))
259
        return NULL;
260

    
261
    /* update the l2 cache entry */
262

    
263
    s->l2_cache_offsets[min_index] = l2_offset;
264
    s->l2_cache_counts[min_index] = 1;
265

    
266
    return l2_table;
267
}
268

    
269
static int count_contiguous_clusters(uint64_t nb_clusters, int cluster_size,
270
        uint64_t *l2_table, uint64_t start, uint64_t mask)
271
{
272
    int i;
273
    uint64_t offset = be64_to_cpu(l2_table[0]) & ~mask;
274

    
275
    if (!offset)
276
        return 0;
277

    
278
    for (i = start; i < start + nb_clusters; i++)
279
        if (offset + (uint64_t) i * cluster_size != (be64_to_cpu(l2_table[i]) & ~mask))
280
            break;
281

    
282
        return (i - start);
283
}
284

    
285
static int count_contiguous_free_clusters(uint64_t nb_clusters, uint64_t *l2_table)
286
{
287
    int i = 0;
288

    
289
    while(nb_clusters-- && l2_table[i] == 0)
290
        i++;
291

    
292
    return i;
293
}
294

    
295
/* The crypt function is compatible with the linux cryptoloop
296
   algorithm for < 4 GB images. NOTE: out_buf == in_buf is
297
   supported */
298
void qcow2_encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
299
                           uint8_t *out_buf, const uint8_t *in_buf,
300
                           int nb_sectors, int enc,
301
                           const AES_KEY *key)
302
{
303
    union {
304
        uint64_t ll[2];
305
        uint8_t b[16];
306
    } ivec;
307
    int i;
308

    
309
    for(i = 0; i < nb_sectors; i++) {
310
        ivec.ll[0] = cpu_to_le64(sector_num);
311
        ivec.ll[1] = 0;
312
        AES_cbc_encrypt(in_buf, out_buf, 512, key,
313
                        ivec.b, enc);
314
        sector_num++;
315
        in_buf += 512;
316
        out_buf += 512;
317
    }
318
}
319

    
320

    
321
static int qcow_read(BlockDriverState *bs, int64_t sector_num,
322
                     uint8_t *buf, int nb_sectors)
323
{
324
    BDRVQcowState *s = bs->opaque;
325
    int ret, index_in_cluster, n, n1;
326
    uint64_t cluster_offset;
327

    
328
    while (nb_sectors > 0) {
329
        n = nb_sectors;
330
        cluster_offset = qcow2_get_cluster_offset(bs, sector_num << 9, &n);
331
        index_in_cluster = sector_num & (s->cluster_sectors - 1);
332
        if (!cluster_offset) {
333
            if (bs->backing_hd) {
334
                /* read from the base image */
335
                n1 = qcow2_backing_read1(bs->backing_hd, sector_num, buf, n);
336
                if (n1 > 0) {
337
                    ret = bdrv_read(bs->backing_hd, sector_num, buf, n1);
338
                    if (ret < 0)
339
                        return -1;
340
                }
341
            } else {
342
                memset(buf, 0, 512 * n);
343
            }
344
        } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
345
            if (qcow2_decompress_cluster(s, cluster_offset) < 0)
346
                return -1;
347
            memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n);
348
        } else {
349
            ret = bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
350
            if (ret != n * 512)
351
                return -1;
352
            if (s->crypt_method) {
353
                qcow2_encrypt_sectors(s, sector_num, buf, buf, n, 0,
354
                                &s->aes_decrypt_key);
355
            }
356
        }
357
        nb_sectors -= n;
358
        sector_num += n;
359
        buf += n * 512;
360
    }
361
    return 0;
362
}
363

    
364
static int copy_sectors(BlockDriverState *bs, uint64_t start_sect,
365
                        uint64_t cluster_offset, int n_start, int n_end)
366
{
367
    BDRVQcowState *s = bs->opaque;
368
    int n, ret;
369

    
370
    n = n_end - n_start;
371
    if (n <= 0)
372
        return 0;
373
    ret = qcow_read(bs, start_sect + n_start, s->cluster_data, n);
374
    if (ret < 0)
375
        return ret;
376
    if (s->crypt_method) {
377
        qcow2_encrypt_sectors(s, start_sect + n_start,
378
                        s->cluster_data,
379
                        s->cluster_data, n, 1,
380
                        &s->aes_encrypt_key);
381
    }
382
    ret = bdrv_write(s->hd, (cluster_offset >> 9) + n_start,
383
                     s->cluster_data, n);
384
    if (ret < 0)
385
        return ret;
386
    return 0;
387
}
388

    
389

    
390
/*
391
 * get_cluster_offset
392
 *
393
 * For a given offset of the disk image, return cluster offset in
394
 * qcow2 file.
395
 *
396
 * on entry, *num is the number of contiguous clusters we'd like to
397
 * access following offset.
398
 *
399
 * on exit, *num is the number of contiguous clusters we can read.
400
 *
401
 * Return 1, if the offset is found
402
 * Return 0, otherwise.
403
 *
404
 */
405

    
406
uint64_t qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
407
    int *num)
408
{
409
    BDRVQcowState *s = bs->opaque;
410
    unsigned int l1_index, l2_index;
411
    uint64_t l2_offset, *l2_table, cluster_offset;
412
    int l1_bits, c;
413
    unsigned int index_in_cluster, nb_clusters;
414
    uint64_t nb_available, nb_needed;
415

    
416
    index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1);
417
    nb_needed = *num + index_in_cluster;
418

    
419
    l1_bits = s->l2_bits + s->cluster_bits;
420

    
421
    /* compute how many bytes there are between the offset and
422
     * the end of the l1 entry
423
     */
424

    
425
    nb_available = (1ULL << l1_bits) - (offset & ((1ULL << l1_bits) - 1));
426

    
427
    /* compute the number of available sectors */
428

    
429
    nb_available = (nb_available >> 9) + index_in_cluster;
430

    
431
    if (nb_needed > nb_available) {
432
        nb_needed = nb_available;
433
    }
434

    
435
    cluster_offset = 0;
436

    
437
    /* seek the the l2 offset in the l1 table */
438

    
439
    l1_index = offset >> l1_bits;
440
    if (l1_index >= s->l1_size)
441
        goto out;
442

    
443
    l2_offset = s->l1_table[l1_index];
444

    
445
    /* seek the l2 table of the given l2 offset */
446

    
447
    if (!l2_offset)
448
        goto out;
449

    
450
    /* load the l2 table in memory */
451

    
452
    l2_offset &= ~QCOW_OFLAG_COPIED;
453
    l2_table = l2_load(bs, l2_offset);
454
    if (l2_table == NULL)
455
        return 0;
456

    
457
    /* find the cluster offset for the given disk offset */
458

    
459
    l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
460
    cluster_offset = be64_to_cpu(l2_table[l2_index]);
461
    nb_clusters = size_to_clusters(s, nb_needed << 9);
462

    
463
    if (!cluster_offset) {
464
        /* how many empty clusters ? */
465
        c = count_contiguous_free_clusters(nb_clusters, &l2_table[l2_index]);
466
    } else {
467
        /* how many allocated clusters ? */
468
        c = count_contiguous_clusters(nb_clusters, s->cluster_size,
469
                &l2_table[l2_index], 0, QCOW_OFLAG_COPIED);
470
    }
471

    
472
   nb_available = (c * s->cluster_sectors);
473
out:
474
    if (nb_available > nb_needed)
475
        nb_available = nb_needed;
476

    
477
    *num = nb_available - index_in_cluster;
478

    
479
    return cluster_offset & ~QCOW_OFLAG_COPIED;
480
}
481

    
482
/*
483
 * get_cluster_table
484
 *
485
 * for a given disk offset, load (and allocate if needed)
486
 * the l2 table.
487
 *
488
 * the l2 table offset in the qcow2 file and the cluster index
489
 * in the l2 table are given to the caller.
490
 *
491
 * Returns 0 on success, -errno in failure case
492
 */
493
static int get_cluster_table(BlockDriverState *bs, uint64_t offset,
494
                             uint64_t **new_l2_table,
495
                             uint64_t *new_l2_offset,
496
                             int *new_l2_index)
497
{
498
    BDRVQcowState *s = bs->opaque;
499
    unsigned int l1_index, l2_index;
500
    uint64_t l2_offset, *l2_table;
501
    int ret;
502

    
503
    /* seek the the l2 offset in the l1 table */
504

    
505
    l1_index = offset >> (s->l2_bits + s->cluster_bits);
506
    if (l1_index >= s->l1_size) {
507
        ret = qcow2_grow_l1_table(bs, l1_index + 1);
508
        if (ret < 0) {
509
            return ret;
510
        }
511
    }
512
    l2_offset = s->l1_table[l1_index];
513

    
514
    /* seek the l2 table of the given l2 offset */
515

    
516
    if (l2_offset & QCOW_OFLAG_COPIED) {
517
        /* load the l2 table in memory */
518
        l2_offset &= ~QCOW_OFLAG_COPIED;
519
        l2_table = l2_load(bs, l2_offset);
520
        if (l2_table == NULL) {
521
            return -EIO;
522
        }
523
    } else {
524
        if (l2_offset)
525
            qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t));
526
        l2_table = l2_allocate(bs, l1_index);
527
        if (l2_table == NULL) {
528
            return -EIO;
529
        }
530
        l2_offset = s->l1_table[l1_index] & ~QCOW_OFLAG_COPIED;
531
    }
532

    
533
    /* find the cluster offset for the given disk offset */
534

    
535
    l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
536

    
537
    *new_l2_table = l2_table;
538
    *new_l2_offset = l2_offset;
539
    *new_l2_index = l2_index;
540

    
541
    return 0;
542
}
543

    
544
/*
545
 * alloc_compressed_cluster_offset
546
 *
547
 * For a given offset of the disk image, return cluster offset in
548
 * qcow2 file.
549
 *
550
 * If the offset is not found, allocate a new compressed cluster.
551
 *
552
 * Return the cluster offset if successful,
553
 * Return 0, otherwise.
554
 *
555
 */
556

    
557
uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs,
558
                                               uint64_t offset,
559
                                               int compressed_size)
560
{
561
    BDRVQcowState *s = bs->opaque;
562
    int l2_index, ret;
563
    uint64_t l2_offset, *l2_table, cluster_offset;
564
    int nb_csectors;
565

    
566
    ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
567
    if (ret < 0) {
568
        return 0;
569
    }
570

    
571
    cluster_offset = be64_to_cpu(l2_table[l2_index]);
572
    if (cluster_offset & QCOW_OFLAG_COPIED)
573
        return cluster_offset & ~QCOW_OFLAG_COPIED;
574

    
575
    if (cluster_offset)
576
        qcow2_free_any_clusters(bs, cluster_offset, 1);
577

    
578
    cluster_offset = qcow2_alloc_bytes(bs, compressed_size);
579
    if (cluster_offset < 0) {
580
        return 0;
581
    }
582

    
583
    nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) -
584
                  (cluster_offset >> 9);
585

    
586
    cluster_offset |= QCOW_OFLAG_COMPRESSED |
587
                      ((uint64_t)nb_csectors << s->csize_shift);
588

    
589
    /* update L2 table */
590

    
591
    /* compressed clusters never have the copied flag */
592

    
593
    l2_table[l2_index] = cpu_to_be64(cluster_offset);
594
    if (bdrv_pwrite(s->hd,
595
                    l2_offset + l2_index * sizeof(uint64_t),
596
                    l2_table + l2_index,
597
                    sizeof(uint64_t)) != sizeof(uint64_t))
598
        return 0;
599

    
600
    return cluster_offset;
601
}
602

    
603
/*
604
 * Write L2 table updates to disk, writing whole sectors to avoid a
605
 * read-modify-write in bdrv_pwrite
606
 */
607
#define L2_ENTRIES_PER_SECTOR (512 / 8)
608
static int write_l2_entries(BDRVQcowState *s, uint64_t *l2_table,
609
    uint64_t l2_offset, int l2_index, int num)
610
{
611
    int l2_start_index = l2_index & ~(L1_ENTRIES_PER_SECTOR - 1);
612
    int start_offset = (8 * l2_index) & ~511;
613
    int end_offset = (8 * (l2_index + num) + 511) & ~511;
614
    size_t len = end_offset - start_offset;
615

    
616
    if (bdrv_pwrite(s->hd, l2_offset + start_offset, &l2_table[l2_start_index],
617
        len) != len)
618
    {
619
        return -1;
620
    }
621

    
622
    return 0;
623
}
624

    
625
int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m)
626
{
627
    BDRVQcowState *s = bs->opaque;
628
    int i, j = 0, l2_index, ret;
629
    uint64_t *old_cluster, start_sect, l2_offset, *l2_table;
630
    uint64_t cluster_offset = m->cluster_offset;
631

    
632
    if (m->nb_clusters == 0)
633
        return 0;
634

    
635
    old_cluster = qemu_malloc(m->nb_clusters * sizeof(uint64_t));
636

    
637
    /* copy content of unmodified sectors */
638
    start_sect = (m->offset & ~(s->cluster_size - 1)) >> 9;
639
    if (m->n_start) {
640
        ret = copy_sectors(bs, start_sect, cluster_offset, 0, m->n_start);
641
        if (ret < 0)
642
            goto err;
643
    }
644

    
645
    if (m->nb_available & (s->cluster_sectors - 1)) {
646
        uint64_t end = m->nb_available & ~(uint64_t)(s->cluster_sectors - 1);
647
        ret = copy_sectors(bs, start_sect + end, cluster_offset + (end << 9),
648
                m->nb_available - end, s->cluster_sectors);
649
        if (ret < 0)
650
            goto err;
651
    }
652

    
653
    /* update L2 table */
654
    ret = get_cluster_table(bs, m->offset, &l2_table, &l2_offset, &l2_index);
655
    if (ret < 0) {
656
        goto err;
657
    }
658

    
659
    for (i = 0; i < m->nb_clusters; i++) {
660
        /* if two concurrent writes happen to the same unallocated cluster
661
         * each write allocates separate cluster and writes data concurrently.
662
         * The first one to complete updates l2 table with pointer to its
663
         * cluster the second one has to do RMW (which is done above by
664
         * copy_sectors()), update l2 table with its cluster pointer and free
665
         * old cluster. This is what this loop does */
666
        if(l2_table[l2_index + i] != 0)
667
            old_cluster[j++] = l2_table[l2_index + i];
668

    
669
        l2_table[l2_index + i] = cpu_to_be64((cluster_offset +
670
                    (i << s->cluster_bits)) | QCOW_OFLAG_COPIED);
671
     }
672

    
673
    if (write_l2_entries(s, l2_table, l2_offset, l2_index, m->nb_clusters) < 0) {
674
        ret = -1;
675
        goto err;
676
    }
677

    
678
    for (i = 0; i < j; i++)
679
        qcow2_free_any_clusters(bs,
680
            be64_to_cpu(old_cluster[i]) & ~QCOW_OFLAG_COPIED, 1);
681

    
682
    ret = 0;
683
err:
684
    qemu_free(old_cluster);
685
    return ret;
686
 }
687

    
688
/*
689
 * alloc_cluster_offset
690
 *
691
 * For a given offset of the disk image, return cluster offset in qcow2 file.
692
 * If the offset is not found, allocate a new cluster.
693
 *
694
 * If the cluster was already allocated, m->nb_clusters is set to 0,
695
 * m->depends_on is set to NULL and the other fields in m are meaningless.
696
 *
697
 * If the cluster is newly allocated, m->nb_clusters is set to the number of
698
 * contiguous clusters that have been allocated. This may be 0 if the request
699
 * conflict with another write request in flight; in this case, m->depends_on
700
 * is set and the remaining fields of m are meaningless.
701
 *
702
 * If m->nb_clusters is non-zero, the other fields of m are valid and contain
703
 * information about the first allocated cluster.
704
 *
705
 * Return 0 on success and -errno in error cases
706
 */
707
uint64_t qcow2_alloc_cluster_offset(BlockDriverState *bs,
708
                                    uint64_t offset,
709
                                    int n_start, int n_end,
710
                                    int *num, QCowL2Meta *m)
711
{
712
    BDRVQcowState *s = bs->opaque;
713
    int l2_index, ret;
714
    uint64_t l2_offset, *l2_table;
715
    int64_t cluster_offset;
716
    unsigned int nb_clusters, i = 0;
717
    QCowL2Meta *old_alloc;
718

    
719
    ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
720
    if (ret < 0) {
721
        return ret;
722
    }
723

    
724
    nb_clusters = size_to_clusters(s, n_end << 9);
725

    
726
    nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
727

    
728
    cluster_offset = be64_to_cpu(l2_table[l2_index]);
729

    
730
    /* We keep all QCOW_OFLAG_COPIED clusters */
731

    
732
    if (cluster_offset & QCOW_OFLAG_COPIED) {
733
        nb_clusters = count_contiguous_clusters(nb_clusters, s->cluster_size,
734
                &l2_table[l2_index], 0, 0);
735

    
736
        cluster_offset &= ~QCOW_OFLAG_COPIED;
737
        m->nb_clusters = 0;
738
        m->depends_on = NULL;
739

    
740
        goto out;
741
    }
742

    
743
    /* for the moment, multiple compressed clusters are not managed */
744

    
745
    if (cluster_offset & QCOW_OFLAG_COMPRESSED)
746
        nb_clusters = 1;
747

    
748
    /* how many available clusters ? */
749

    
750
    while (i < nb_clusters) {
751
        i += count_contiguous_clusters(nb_clusters - i, s->cluster_size,
752
                &l2_table[l2_index], i, 0);
753

    
754
        if(be64_to_cpu(l2_table[l2_index + i]))
755
            break;
756

    
757
        i += count_contiguous_free_clusters(nb_clusters - i,
758
                &l2_table[l2_index + i]);
759

    
760
        cluster_offset = be64_to_cpu(l2_table[l2_index + i]);
761

    
762
        if ((cluster_offset & QCOW_OFLAG_COPIED) ||
763
                (cluster_offset & QCOW_OFLAG_COMPRESSED))
764
            break;
765
    }
766
    nb_clusters = i;
767

    
768
    /*
769
     * Check if there already is an AIO write request in flight which allocates
770
     * the same cluster. In this case we need to wait until the previous
771
     * request has completed and updated the L2 table accordingly.
772
     */
773
    QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) {
774

    
775
        uint64_t end_offset = offset + nb_clusters * s->cluster_size;
776
        uint64_t old_offset = old_alloc->offset;
777
        uint64_t old_end_offset = old_alloc->offset +
778
            old_alloc->nb_clusters * s->cluster_size;
779

    
780
        if (end_offset < old_offset || offset > old_end_offset) {
781
            /* No intersection */
782
        } else {
783
            if (offset < old_offset) {
784
                /* Stop at the start of a running allocation */
785
                nb_clusters = (old_offset - offset) >> s->cluster_bits;
786
            } else {
787
                nb_clusters = 0;
788
            }
789

    
790
            if (nb_clusters == 0) {
791
                /* Set dependency and wait for a callback */
792
                m->depends_on = old_alloc;
793
                m->nb_clusters = 0;
794
                *num = 0;
795
                return 0;
796
            }
797
        }
798
    }
799

    
800
    if (!nb_clusters) {
801
        abort();
802
    }
803

    
804
    QLIST_INSERT_HEAD(&s->cluster_allocs, m, next_in_flight);
805

    
806
    /* allocate a new cluster */
807

    
808
    cluster_offset = qcow2_alloc_clusters(bs, nb_clusters * s->cluster_size);
809
    if (cluster_offset < 0) {
810
        return cluster_offset;
811
    }
812

    
813
    /* save info needed for meta data update */
814
    m->offset = offset;
815
    m->n_start = n_start;
816
    m->nb_clusters = nb_clusters;
817

    
818
out:
819
    m->nb_available = MIN(nb_clusters << (s->cluster_bits - 9), n_end);
820
    m->cluster_offset = cluster_offset;
821

    
822
    *num = m->nb_available - n_start;
823

    
824
    return 0;
825
}
826

    
827
static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
828
                             const uint8_t *buf, int buf_size)
829
{
830
    z_stream strm1, *strm = &strm1;
831
    int ret, out_len;
832

    
833
    memset(strm, 0, sizeof(*strm));
834

    
835
    strm->next_in = (uint8_t *)buf;
836
    strm->avail_in = buf_size;
837
    strm->next_out = out_buf;
838
    strm->avail_out = out_buf_size;
839

    
840
    ret = inflateInit2(strm, -12);
841
    if (ret != Z_OK)
842
        return -1;
843
    ret = inflate(strm, Z_FINISH);
844
    out_len = strm->next_out - out_buf;
845
    if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
846
        out_len != out_buf_size) {
847
        inflateEnd(strm);
848
        return -1;
849
    }
850
    inflateEnd(strm);
851
    return 0;
852
}
853

    
854
int qcow2_decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset)
855
{
856
    int ret, csize, nb_csectors, sector_offset;
857
    uint64_t coffset;
858

    
859
    coffset = cluster_offset & s->cluster_offset_mask;
860
    if (s->cluster_cache_offset != coffset) {
861
        nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
862
        sector_offset = coffset & 511;
863
        csize = nb_csectors * 512 - sector_offset;
864
        ret = bdrv_read(s->hd, coffset >> 9, s->cluster_data, nb_csectors);
865
        if (ret < 0) {
866
            return -1;
867
        }
868
        if (decompress_buffer(s->cluster_cache, s->cluster_size,
869
                              s->cluster_data + sector_offset, csize) < 0) {
870
            return -1;
871
        }
872
        s->cluster_cache_offset = coffset;
873
    }
874
    return 0;
875
}