Statistics
| Branch: | Revision:

root / block / qcow2.c @ 73c632ed

History | View | Annotate | Download (91.3 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
#include "qemu-common.h"
25
#include "block_int.h"
26
#include "module.h"
27
#include <zlib.h>
28
#include "aes.h"
29

    
30
/*
31
  Differences with QCOW:
32

33
  - Support for multiple incremental snapshots.
34
  - Memory management by reference counts.
35
  - Clusters which have a reference count of one have the bit
36
    QCOW_OFLAG_COPIED to optimize write performance.
37
  - Size of compressed clusters is stored in sectors to reduce bit usage
38
    in the cluster offsets.
39
  - Support for storing additional data (such as the VM state) in the
40
    snapshots.
41
  - If a backing store is used, the cluster size is not constrained
42
    (could be backported to QCOW).
43
  - L2 tables have always a size of one cluster.
44
*/
45

    
46
//#define DEBUG_ALLOC
47
//#define DEBUG_ALLOC2
48
//#define DEBUG_EXT
49

    
50
#define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
51
#define QCOW_VERSION 2
52

    
53
#define QCOW_CRYPT_NONE 0
54
#define QCOW_CRYPT_AES  1
55

    
56
#define QCOW_MAX_CRYPT_CLUSTERS 32
57

    
58
/* indicate that the refcount of the referenced cluster is exactly one. */
59
#define QCOW_OFLAG_COPIED     (1LL << 63)
60
/* indicate that the cluster is compressed (they never have the copied flag) */
61
#define QCOW_OFLAG_COMPRESSED (1LL << 62)
62

    
63
#define REFCOUNT_SHIFT 1 /* refcount size is 2 bytes */
64

    
65
#define MIN_CLUSTER_BITS 9
66
#define MAX_CLUSTER_BITS 16
67

    
68
typedef struct QCowHeader {
69
    uint32_t magic;
70
    uint32_t version;
71
    uint64_t backing_file_offset;
72
    uint32_t backing_file_size;
73
    uint32_t cluster_bits;
74
    uint64_t size; /* in bytes */
75
    uint32_t crypt_method;
76
    uint32_t l1_size; /* XXX: save number of clusters instead ? */
77
    uint64_t l1_table_offset;
78
    uint64_t refcount_table_offset;
79
    uint32_t refcount_table_clusters;
80
    uint32_t nb_snapshots;
81
    uint64_t snapshots_offset;
82
} QCowHeader;
83

    
84

    
85
typedef struct {
86
    uint32_t magic;
87
    uint32_t len;
88
} QCowExtension;
89
#define  QCOW_EXT_MAGIC_END 0
90
#define  QCOW_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
91

    
92

    
93
typedef struct __attribute__((packed)) QCowSnapshotHeader {
94
    /* header is 8 byte aligned */
95
    uint64_t l1_table_offset;
96

    
97
    uint32_t l1_size;
98
    uint16_t id_str_size;
99
    uint16_t name_size;
100

    
101
    uint32_t date_sec;
102
    uint32_t date_nsec;
103

    
104
    uint64_t vm_clock_nsec;
105

    
106
    uint32_t vm_state_size;
107
    uint32_t extra_data_size; /* for extension */
108
    /* extra data follows */
109
    /* id_str follows */
110
    /* name follows  */
111
} QCowSnapshotHeader;
112

    
113
#define L2_CACHE_SIZE 16
114

    
115
typedef struct QCowSnapshot {
116
    uint64_t l1_table_offset;
117
    uint32_t l1_size;
118
    char *id_str;
119
    char *name;
120
    uint32_t vm_state_size;
121
    uint32_t date_sec;
122
    uint32_t date_nsec;
123
    uint64_t vm_clock_nsec;
124
} QCowSnapshot;
125

    
126
typedef struct BDRVQcowState {
127
    BlockDriverState *hd;
128
    int cluster_bits;
129
    int cluster_size;
130
    int cluster_sectors;
131
    int l2_bits;
132
    int l2_size;
133
    int l1_size;
134
    int l1_vm_state_index;
135
    int csize_shift;
136
    int csize_mask;
137
    uint64_t cluster_offset_mask;
138
    uint64_t l1_table_offset;
139
    uint64_t *l1_table;
140
    uint64_t *l2_cache;
141
    uint64_t l2_cache_offsets[L2_CACHE_SIZE];
142
    uint32_t l2_cache_counts[L2_CACHE_SIZE];
143
    uint8_t *cluster_cache;
144
    uint8_t *cluster_data;
145
    uint64_t cluster_cache_offset;
146

    
147
    uint64_t *refcount_table;
148
    uint64_t refcount_table_offset;
149
    uint32_t refcount_table_size;
150
    uint64_t refcount_block_cache_offset;
151
    uint16_t *refcount_block_cache;
152
    int64_t free_cluster_index;
153
    int64_t free_byte_offset;
154

    
155
    uint32_t crypt_method; /* current crypt method, 0 if no key yet */
156
    uint32_t crypt_method_header;
157
    AES_KEY aes_encrypt_key;
158
    AES_KEY aes_decrypt_key;
159
    uint64_t snapshots_offset;
160
    int snapshots_size;
161
    int nb_snapshots;
162
    QCowSnapshot *snapshots;
163
} BDRVQcowState;
164

    
165
static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset);
166
static int qcow_read(BlockDriverState *bs, int64_t sector_num,
167
                     uint8_t *buf, int nb_sectors);
168
static int qcow_read_snapshots(BlockDriverState *bs);
169
static void qcow_free_snapshots(BlockDriverState *bs);
170
static int refcount_init(BlockDriverState *bs);
171
static void refcount_close(BlockDriverState *bs);
172
static int get_refcount(BlockDriverState *bs, int64_t cluster_index);
173
static int update_cluster_refcount(BlockDriverState *bs,
174
                                   int64_t cluster_index,
175
                                   int addend);
176
static void update_refcount(BlockDriverState *bs,
177
                            int64_t offset, int64_t length,
178
                            int addend);
179
static int64_t alloc_clusters(BlockDriverState *bs, int64_t size);
180
static int64_t alloc_bytes(BlockDriverState *bs, int size);
181
static void free_clusters(BlockDriverState *bs,
182
                          int64_t offset, int64_t size);
183
static int check_refcounts(BlockDriverState *bs);
184

    
185
static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
186
{
187
    const QCowHeader *cow_header = (const void *)buf;
188

    
189
    if (buf_size >= sizeof(QCowHeader) &&
190
        be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
191
        be32_to_cpu(cow_header->version) == QCOW_VERSION)
192
        return 100;
193
    else
194
        return 0;
195
}
196

    
197

    
198
/* 
199
 * read qcow2 extension and fill bs
200
 * start reading from start_offset
201
 * finish reading upon magic of value 0 or when end_offset reached
202
 * unknown magic is skipped (future extension this version knows nothing about)
203
 * return 0 upon success, non-0 otherwise
204
 */
205
static int qcow_read_extensions(BlockDriverState *bs, uint64_t start_offset,
206
                                uint64_t end_offset)
207
{
208
    BDRVQcowState *s = bs->opaque;
209
    QCowExtension ext;
210
    uint64_t offset;
211

    
212
#ifdef DEBUG_EXT
213
    printf("qcow_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
214
#endif
215
    offset = start_offset;
216
    while (offset < end_offset) {
217

    
218
#ifdef DEBUG_EXT
219
        /* Sanity check */
220
        if (offset > s->cluster_size)
221
            printf("qcow_handle_extension: suspicious offset %lu\n", offset);
222

    
223
        printf("attemting to read extended header in offset %lu\n", offset);
224
#endif
225

    
226
        if (bdrv_pread(s->hd, offset, &ext, sizeof(ext)) != sizeof(ext)) {
227
            fprintf(stderr, "qcow_handle_extension: ERROR: pread fail from offset %llu\n",
228
                    (unsigned long long)offset);
229
            return 1;
230
        }
231
        be32_to_cpus(&ext.magic);
232
        be32_to_cpus(&ext.len);
233
        offset += sizeof(ext);
234
#ifdef DEBUG_EXT
235
        printf("ext.magic = 0x%x\n", ext.magic);
236
#endif
237
        switch (ext.magic) {
238
        case QCOW_EXT_MAGIC_END:
239
            return 0;
240

    
241
        case QCOW_EXT_MAGIC_BACKING_FORMAT:
242
            if (ext.len >= sizeof(bs->backing_format)) {
243
                fprintf(stderr, "ERROR: ext_backing_format: len=%u too large"
244
                        " (>=%zu)\n",
245
                        ext.len, sizeof(bs->backing_format));
246
                return 2;
247
            }
248
            if (bdrv_pread(s->hd, offset , bs->backing_format,
249
                           ext.len) != ext.len)
250
                return 3;
251
            bs->backing_format[ext.len] = '\0';
252
#ifdef DEBUG_EXT
253
            printf("Qcow2: Got format extension %s\n", bs->backing_format);
254
#endif
255
            offset += ((ext.len + 7) & ~7);
256
            break;
257

    
258
        default:
259
            /* unknown magic -- just skip it */
260
            offset += ((ext.len + 7) & ~7);
261
            break;
262
        }
263
    }
264

    
265
    return 0;
266
}
267

    
268

    
269
static int qcow_open(BlockDriverState *bs, const char *filename, int flags)
270
{
271
    BDRVQcowState *s = bs->opaque;
272
    int len, i, shift, ret;
273
    QCowHeader header;
274
    uint64_t ext_end;
275

    
276
    /* Performance is terrible right now with cache=writethrough due mainly
277
     * to reference count updates.  If the user does not explicitly specify
278
     * a caching type, force to writeback caching.
279
     */
280
    if ((flags & BDRV_O_CACHE_DEF)) {
281
        flags |= BDRV_O_CACHE_WB;
282
        flags &= ~BDRV_O_CACHE_DEF;
283
    }
284
    ret = bdrv_file_open(&s->hd, filename, flags);
285
    if (ret < 0)
286
        return ret;
287
    if (bdrv_pread(s->hd, 0, &header, sizeof(header)) != sizeof(header))
288
        goto fail;
289
    be32_to_cpus(&header.magic);
290
    be32_to_cpus(&header.version);
291
    be64_to_cpus(&header.backing_file_offset);
292
    be32_to_cpus(&header.backing_file_size);
293
    be64_to_cpus(&header.size);
294
    be32_to_cpus(&header.cluster_bits);
295
    be32_to_cpus(&header.crypt_method);
296
    be64_to_cpus(&header.l1_table_offset);
297
    be32_to_cpus(&header.l1_size);
298
    be64_to_cpus(&header.refcount_table_offset);
299
    be32_to_cpus(&header.refcount_table_clusters);
300
    be64_to_cpus(&header.snapshots_offset);
301
    be32_to_cpus(&header.nb_snapshots);
302

    
303
    if (header.magic != QCOW_MAGIC || header.version != QCOW_VERSION)
304
        goto fail;
305
    if (header.size <= 1 ||
306
        header.cluster_bits < MIN_CLUSTER_BITS ||
307
        header.cluster_bits > MAX_CLUSTER_BITS)
308
        goto fail;
309
    if (header.crypt_method > QCOW_CRYPT_AES)
310
        goto fail;
311
    s->crypt_method_header = header.crypt_method;
312
    if (s->crypt_method_header)
313
        bs->encrypted = 1;
314
    s->cluster_bits = header.cluster_bits;
315
    s->cluster_size = 1 << s->cluster_bits;
316
    s->cluster_sectors = 1 << (s->cluster_bits - 9);
317
    s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
318
    s->l2_size = 1 << s->l2_bits;
319
    bs->total_sectors = header.size / 512;
320
    s->csize_shift = (62 - (s->cluster_bits - 8));
321
    s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
322
    s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
323
    s->refcount_table_offset = header.refcount_table_offset;
324
    s->refcount_table_size =
325
        header.refcount_table_clusters << (s->cluster_bits - 3);
326

    
327
    s->snapshots_offset = header.snapshots_offset;
328
    s->nb_snapshots = header.nb_snapshots;
329

    
330
    /* read the level 1 table */
331
    s->l1_size = header.l1_size;
332
    shift = s->cluster_bits + s->l2_bits;
333
    s->l1_vm_state_index = (header.size + (1LL << shift) - 1) >> shift;
334
    /* the L1 table must contain at least enough entries to put
335
       header.size bytes */
336
    if (s->l1_size < s->l1_vm_state_index)
337
        goto fail;
338
    s->l1_table_offset = header.l1_table_offset;
339
    s->l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
340
    if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, s->l1_size * sizeof(uint64_t)) !=
341
        s->l1_size * sizeof(uint64_t))
342
        goto fail;
343
    for(i = 0;i < s->l1_size; i++) {
344
        be64_to_cpus(&s->l1_table[i]);
345
    }
346
    /* alloc L2 cache */
347
    s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
348
    s->cluster_cache = qemu_malloc(s->cluster_size);
349
    /* one more sector for decompressed data alignment */
350
    s->cluster_data = qemu_malloc(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
351
                                  + 512);
352
    s->cluster_cache_offset = -1;
353

    
354
    if (refcount_init(bs) < 0)
355
        goto fail;
356

    
357
    /* read qcow2 extensions */
358
    if (header.backing_file_offset)
359
        ext_end = header.backing_file_offset;
360
    else
361
        ext_end = s->cluster_size;
362
    if (qcow_read_extensions(bs, sizeof(header), ext_end))
363
        goto fail;
364

    
365
    /* read the backing file name */
366
    if (header.backing_file_offset != 0) {
367
        len = header.backing_file_size;
368
        if (len > 1023)
369
            len = 1023;
370
        if (bdrv_pread(s->hd, header.backing_file_offset, bs->backing_file, len) != len)
371
            goto fail;
372
        bs->backing_file[len] = '\0';
373
    }
374
    if (qcow_read_snapshots(bs) < 0)
375
        goto fail;
376

    
377
#ifdef DEBUG_ALLOC
378
    check_refcounts(bs);
379
#endif
380
    return 0;
381

    
382
 fail:
383
    qcow_free_snapshots(bs);
384
    refcount_close(bs);
385
    qemu_free(s->l1_table);
386
    qemu_free(s->l2_cache);
387
    qemu_free(s->cluster_cache);
388
    qemu_free(s->cluster_data);
389
    bdrv_delete(s->hd);
390
    return -1;
391
}
392

    
393
static int qcow_set_key(BlockDriverState *bs, const char *key)
394
{
395
    BDRVQcowState *s = bs->opaque;
396
    uint8_t keybuf[16];
397
    int len, i;
398

    
399
    memset(keybuf, 0, 16);
400
    len = strlen(key);
401
    if (len > 16)
402
        len = 16;
403
    /* XXX: we could compress the chars to 7 bits to increase
404
       entropy */
405
    for(i = 0;i < len;i++) {
406
        keybuf[i] = key[i];
407
    }
408
    s->crypt_method = s->crypt_method_header;
409

    
410
    if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
411
        return -1;
412
    if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
413
        return -1;
414
#if 0
415
    /* test */
416
    {
417
        uint8_t in[16];
418
        uint8_t out[16];
419
        uint8_t tmp[16];
420
        for(i=0;i<16;i++)
421
            in[i] = i;
422
        AES_encrypt(in, tmp, &s->aes_encrypt_key);
423
        AES_decrypt(tmp, out, &s->aes_decrypt_key);
424
        for(i = 0; i < 16; i++)
425
            printf(" %02x", tmp[i]);
426
        printf("\n");
427
        for(i = 0; i < 16; i++)
428
            printf(" %02x", out[i]);
429
        printf("\n");
430
    }
431
#endif
432
    return 0;
433
}
434

    
435
/* The crypt function is compatible with the linux cryptoloop
436
   algorithm for < 4 GB images. NOTE: out_buf == in_buf is
437
   supported */
438
static void encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
439
                            uint8_t *out_buf, const uint8_t *in_buf,
440
                            int nb_sectors, int enc,
441
                            const AES_KEY *key)
442
{
443
    union {
444
        uint64_t ll[2];
445
        uint8_t b[16];
446
    } ivec;
447
    int i;
448

    
449
    for(i = 0; i < nb_sectors; i++) {
450
        ivec.ll[0] = cpu_to_le64(sector_num);
451
        ivec.ll[1] = 0;
452
        AES_cbc_encrypt(in_buf, out_buf, 512, key,
453
                        ivec.b, enc);
454
        sector_num++;
455
        in_buf += 512;
456
        out_buf += 512;
457
    }
458
}
459

    
460
static int copy_sectors(BlockDriverState *bs, uint64_t start_sect,
461
                        uint64_t cluster_offset, int n_start, int n_end)
462
{
463
    BDRVQcowState *s = bs->opaque;
464
    int n, ret;
465

    
466
    n = n_end - n_start;
467
    if (n <= 0)
468
        return 0;
469
    ret = qcow_read(bs, start_sect + n_start, s->cluster_data, n);
470
    if (ret < 0)
471
        return ret;
472
    if (s->crypt_method) {
473
        encrypt_sectors(s, start_sect + n_start,
474
                        s->cluster_data,
475
                        s->cluster_data, n, 1,
476
                        &s->aes_encrypt_key);
477
    }
478
    ret = bdrv_write(s->hd, (cluster_offset >> 9) + n_start,
479
                     s->cluster_data, n);
480
    if (ret < 0)
481
        return ret;
482
    return 0;
483
}
484

    
485
static void l2_cache_reset(BlockDriverState *bs)
486
{
487
    BDRVQcowState *s = bs->opaque;
488

    
489
    memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
490
    memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
491
    memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
492
}
493

    
494
static inline int l2_cache_new_entry(BlockDriverState *bs)
495
{
496
    BDRVQcowState *s = bs->opaque;
497
    uint32_t min_count;
498
    int min_index, i;
499

    
500
    /* find a new entry in the least used one */
501
    min_index = 0;
502
    min_count = 0xffffffff;
503
    for(i = 0; i < L2_CACHE_SIZE; i++) {
504
        if (s->l2_cache_counts[i] < min_count) {
505
            min_count = s->l2_cache_counts[i];
506
            min_index = i;
507
        }
508
    }
509
    return min_index;
510
}
511

    
512
static int64_t align_offset(int64_t offset, int n)
513
{
514
    offset = (offset + n - 1) & ~(n - 1);
515
    return offset;
516
}
517

    
518
static int grow_l1_table(BlockDriverState *bs, int min_size)
519
{
520
    BDRVQcowState *s = bs->opaque;
521
    int new_l1_size, new_l1_size2, ret, i;
522
    uint64_t *new_l1_table;
523
    uint64_t new_l1_table_offset;
524
    uint8_t data[12];
525

    
526
    new_l1_size = s->l1_size;
527
    if (min_size <= new_l1_size)
528
        return 0;
529
    while (min_size > new_l1_size) {
530
        new_l1_size = (new_l1_size * 3 + 1) / 2;
531
    }
532
#ifdef DEBUG_ALLOC2
533
    printf("grow l1_table from %d to %d\n", s->l1_size, new_l1_size);
534
#endif
535

    
536
    new_l1_size2 = sizeof(uint64_t) * new_l1_size;
537
    new_l1_table = qemu_mallocz(new_l1_size2);
538
    memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t));
539

    
540
    /* write new table (align to cluster) */
541
    new_l1_table_offset = alloc_clusters(bs, new_l1_size2);
542

    
543
    for(i = 0; i < s->l1_size; i++)
544
        new_l1_table[i] = cpu_to_be64(new_l1_table[i]);
545
    ret = bdrv_pwrite(s->hd, new_l1_table_offset, new_l1_table, new_l1_size2);
546
    if (ret != new_l1_size2)
547
        goto fail;
548
    for(i = 0; i < s->l1_size; i++)
549
        new_l1_table[i] = be64_to_cpu(new_l1_table[i]);
550

    
551
    /* set new table */
552
    cpu_to_be32w((uint32_t*)data, new_l1_size);
553
    cpu_to_be64w((uint64_t*)(data + 4), new_l1_table_offset);
554
    if (bdrv_pwrite(s->hd, offsetof(QCowHeader, l1_size), data,
555
                sizeof(data)) != sizeof(data))
556
        goto fail;
557
    qemu_free(s->l1_table);
558
    free_clusters(bs, s->l1_table_offset, s->l1_size * sizeof(uint64_t));
559
    s->l1_table_offset = new_l1_table_offset;
560
    s->l1_table = new_l1_table;
561
    s->l1_size = new_l1_size;
562
    return 0;
563
 fail:
564
    qemu_free(s->l1_table);
565
    return -EIO;
566
}
567

    
568
/*
569
 * seek_l2_table
570
 *
571
 * seek l2_offset in the l2_cache table
572
 * if not found, return NULL,
573
 * if found,
574
 *   increments the l2 cache hit count of the entry,
575
 *   if counter overflow, divide by two all counters
576
 *   return the pointer to the l2 cache entry
577
 *
578
 */
579

    
580
static uint64_t *seek_l2_table(BDRVQcowState *s, uint64_t l2_offset)
581
{
582
    int i, j;
583

    
584
    for(i = 0; i < L2_CACHE_SIZE; i++) {
585
        if (l2_offset == s->l2_cache_offsets[i]) {
586
            /* increment the hit count */
587
            if (++s->l2_cache_counts[i] == 0xffffffff) {
588
                for(j = 0; j < L2_CACHE_SIZE; j++) {
589
                    s->l2_cache_counts[j] >>= 1;
590
                }
591
            }
592
            return s->l2_cache + (i << s->l2_bits);
593
        }
594
    }
595
    return NULL;
596
}
597

    
598
/*
599
 * l2_load
600
 *
601
 * Loads a L2 table into memory. If the table is in the cache, the cache
602
 * is used; otherwise the L2 table is loaded from the image file.
603
 *
604
 * Returns a pointer to the L2 table on success, or NULL if the read from
605
 * the image file failed.
606
 */
607

    
608
static uint64_t *l2_load(BlockDriverState *bs, uint64_t l2_offset)
609
{
610
    BDRVQcowState *s = bs->opaque;
611
    int min_index;
612
    uint64_t *l2_table;
613

    
614
    /* seek if the table for the given offset is in the cache */
615

    
616
    l2_table = seek_l2_table(s, l2_offset);
617
    if (l2_table != NULL)
618
        return l2_table;
619

    
620
    /* not found: load a new entry in the least used one */
621

    
622
    min_index = l2_cache_new_entry(bs);
623
    l2_table = s->l2_cache + (min_index << s->l2_bits);
624
    if (bdrv_pread(s->hd, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
625
        s->l2_size * sizeof(uint64_t))
626
        return NULL;
627
    s->l2_cache_offsets[min_index] = l2_offset;
628
    s->l2_cache_counts[min_index] = 1;
629

    
630
    return l2_table;
631
}
632

    
633
/*
634
 * l2_allocate
635
 *
636
 * Allocate a new l2 entry in the file. If l1_index points to an already
637
 * used entry in the L2 table (i.e. we are doing a copy on write for the L2
638
 * table) copy the contents of the old L2 table into the newly allocated one.
639
 * Otherwise the new table is initialized with zeros.
640
 *
641
 */
642

    
643
static uint64_t *l2_allocate(BlockDriverState *bs, int l1_index)
644
{
645
    BDRVQcowState *s = bs->opaque;
646
    int min_index;
647
    uint64_t old_l2_offset, tmp;
648
    uint64_t *l2_table, l2_offset;
649

    
650
    old_l2_offset = s->l1_table[l1_index];
651

    
652
    /* allocate a new l2 entry */
653

    
654
    l2_offset = alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
655

    
656
    /* update the L1 entry */
657

    
658
    s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
659

    
660
    tmp = cpu_to_be64(l2_offset | QCOW_OFLAG_COPIED);
661
    if (bdrv_pwrite(s->hd, s->l1_table_offset + l1_index * sizeof(tmp),
662
                    &tmp, sizeof(tmp)) != sizeof(tmp))
663
        return NULL;
664

    
665
    /* allocate a new entry in the l2 cache */
666

    
667
    min_index = l2_cache_new_entry(bs);
668
    l2_table = s->l2_cache + (min_index << s->l2_bits);
669

    
670
    if (old_l2_offset == 0) {
671
        /* if there was no old l2 table, clear the new table */
672
        memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
673
    } else {
674
        /* if there was an old l2 table, read it from the disk */
675
        if (bdrv_pread(s->hd, old_l2_offset,
676
                       l2_table, s->l2_size * sizeof(uint64_t)) !=
677
            s->l2_size * sizeof(uint64_t))
678
            return NULL;
679
    }
680
    /* write the l2 table to the file */
681
    if (bdrv_pwrite(s->hd, l2_offset,
682
                    l2_table, s->l2_size * sizeof(uint64_t)) !=
683
        s->l2_size * sizeof(uint64_t))
684
        return NULL;
685

    
686
    /* update the l2 cache entry */
687

    
688
    s->l2_cache_offsets[min_index] = l2_offset;
689
    s->l2_cache_counts[min_index] = 1;
690

    
691
    return l2_table;
692
}
693

    
694
static int size_to_clusters(BDRVQcowState *s, int64_t size)
695
{
696
    return (size + (s->cluster_size - 1)) >> s->cluster_bits;
697
}
698

    
699
static int count_contiguous_clusters(uint64_t nb_clusters, int cluster_size,
700
        uint64_t *l2_table, uint64_t start, uint64_t mask)
701
{
702
    int i;
703
    uint64_t offset = be64_to_cpu(l2_table[0]) & ~mask;
704

    
705
    if (!offset)
706
        return 0;
707

    
708
    for (i = start; i < start + nb_clusters; i++)
709
        if (offset + i * cluster_size != (be64_to_cpu(l2_table[i]) & ~mask))
710
            break;
711

    
712
        return (i - start);
713
}
714

    
715
static int count_contiguous_free_clusters(uint64_t nb_clusters, uint64_t *l2_table)
716
{
717
    int i = 0;
718

    
719
    while(nb_clusters-- && l2_table[i] == 0)
720
        i++;
721

    
722
    return i;
723
}
724

    
725
/*
726
 * get_cluster_offset
727
 *
728
 * For a given offset of the disk image, return cluster offset in
729
 * qcow2 file.
730
 *
731
 * on entry, *num is the number of contiguous clusters we'd like to
732
 * access following offset.
733
 *
734
 * on exit, *num is the number of contiguous clusters we can read.
735
 *
736
 * Return 1, if the offset is found
737
 * Return 0, otherwise.
738
 *
739
 */
740

    
741
static uint64_t get_cluster_offset(BlockDriverState *bs,
742
                                   uint64_t offset, int *num)
743
{
744
    BDRVQcowState *s = bs->opaque;
745
    int l1_index, l2_index;
746
    uint64_t l2_offset, *l2_table, cluster_offset;
747
    int l1_bits, c;
748
    int index_in_cluster, nb_available, nb_needed, nb_clusters;
749

    
750
    index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1);
751
    nb_needed = *num + index_in_cluster;
752

    
753
    l1_bits = s->l2_bits + s->cluster_bits;
754

    
755
    /* compute how many bytes there are between the offset and
756
     * the end of the l1 entry
757
     */
758

    
759
    nb_available = (1 << l1_bits) - (offset & ((1 << l1_bits) - 1));
760

    
761
    /* compute the number of available sectors */
762

    
763
    nb_available = (nb_available >> 9) + index_in_cluster;
764

    
765
    if (nb_needed > nb_available) {
766
        nb_needed = nb_available;
767
    }
768

    
769
    cluster_offset = 0;
770

    
771
    /* seek the the l2 offset in the l1 table */
772

    
773
    l1_index = offset >> l1_bits;
774
    if (l1_index >= s->l1_size)
775
        goto out;
776

    
777
    l2_offset = s->l1_table[l1_index];
778

    
779
    /* seek the l2 table of the given l2 offset */
780

    
781
    if (!l2_offset)
782
        goto out;
783

    
784
    /* load the l2 table in memory */
785

    
786
    l2_offset &= ~QCOW_OFLAG_COPIED;
787
    l2_table = l2_load(bs, l2_offset);
788
    if (l2_table == NULL)
789
        return 0;
790

    
791
    /* find the cluster offset for the given disk offset */
792

    
793
    l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
794
    cluster_offset = be64_to_cpu(l2_table[l2_index]);
795
    nb_clusters = size_to_clusters(s, nb_needed << 9);
796

    
797
    if (!cluster_offset) {
798
        /* how many empty clusters ? */
799
        c = count_contiguous_free_clusters(nb_clusters, &l2_table[l2_index]);
800
    } else {
801
        /* how many allocated clusters ? */
802
        c = count_contiguous_clusters(nb_clusters, s->cluster_size,
803
                &l2_table[l2_index], 0, QCOW_OFLAG_COPIED);
804
    }
805

    
806
   nb_available = (c * s->cluster_sectors);
807
out:
808
    if (nb_available > nb_needed)
809
        nb_available = nb_needed;
810

    
811
    *num = nb_available - index_in_cluster;
812

    
813
    return cluster_offset & ~QCOW_OFLAG_COPIED;
814
}
815

    
816
/*
817
 * free_any_clusters
818
 *
819
 * free clusters according to its type: compressed or not
820
 *
821
 */
822

    
823
static void free_any_clusters(BlockDriverState *bs,
824
                              uint64_t cluster_offset, int nb_clusters)
825
{
826
    BDRVQcowState *s = bs->opaque;
827

    
828
    /* free the cluster */
829

    
830
    if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
831
        int nb_csectors;
832
        nb_csectors = ((cluster_offset >> s->csize_shift) &
833
                       s->csize_mask) + 1;
834
        free_clusters(bs, (cluster_offset & s->cluster_offset_mask) & ~511,
835
                      nb_csectors * 512);
836
        return;
837
    }
838

    
839
    free_clusters(bs, cluster_offset, nb_clusters << s->cluster_bits);
840

    
841
    return;
842
}
843

    
844
/*
845
 * get_cluster_table
846
 *
847
 * for a given disk offset, load (and allocate if needed)
848
 * the l2 table.
849
 *
850
 * the l2 table offset in the qcow2 file and the cluster index
851
 * in the l2 table are given to the caller.
852
 *
853
 */
854

    
855
static int get_cluster_table(BlockDriverState *bs, uint64_t offset,
856
                             uint64_t **new_l2_table,
857
                             uint64_t *new_l2_offset,
858
                             int *new_l2_index)
859
{
860
    BDRVQcowState *s = bs->opaque;
861
    int l1_index, l2_index, ret;
862
    uint64_t l2_offset, *l2_table;
863

    
864
    /* seek the the l2 offset in the l1 table */
865

    
866
    l1_index = offset >> (s->l2_bits + s->cluster_bits);
867
    if (l1_index >= s->l1_size) {
868
        ret = grow_l1_table(bs, l1_index + 1);
869
        if (ret < 0)
870
            return 0;
871
    }
872
    l2_offset = s->l1_table[l1_index];
873

    
874
    /* seek the l2 table of the given l2 offset */
875

    
876
    if (l2_offset & QCOW_OFLAG_COPIED) {
877
        /* load the l2 table in memory */
878
        l2_offset &= ~QCOW_OFLAG_COPIED;
879
        l2_table = l2_load(bs, l2_offset);
880
        if (l2_table == NULL)
881
            return 0;
882
    } else {
883
        if (l2_offset)
884
            free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t));
885
        l2_table = l2_allocate(bs, l1_index);
886
        if (l2_table == NULL)
887
            return 0;
888
        l2_offset = s->l1_table[l1_index] & ~QCOW_OFLAG_COPIED;
889
    }
890

    
891
    /* find the cluster offset for the given disk offset */
892

    
893
    l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
894

    
895
    *new_l2_table = l2_table;
896
    *new_l2_offset = l2_offset;
897
    *new_l2_index = l2_index;
898

    
899
    return 1;
900
}
901

    
902
/*
903
 * alloc_compressed_cluster_offset
904
 *
905
 * For a given offset of the disk image, return cluster offset in
906
 * qcow2 file.
907
 *
908
 * If the offset is not found, allocate a new compressed cluster.
909
 *
910
 * Return the cluster offset if successful,
911
 * Return 0, otherwise.
912
 *
913
 */
914

    
915
static uint64_t alloc_compressed_cluster_offset(BlockDriverState *bs,
916
                                                uint64_t offset,
917
                                                int compressed_size)
918
{
919
    BDRVQcowState *s = bs->opaque;
920
    int l2_index, ret;
921
    uint64_t l2_offset, *l2_table, cluster_offset;
922
    int nb_csectors;
923

    
924
    ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
925
    if (ret == 0)
926
        return 0;
927

    
928
    cluster_offset = be64_to_cpu(l2_table[l2_index]);
929
    if (cluster_offset & QCOW_OFLAG_COPIED)
930
        return cluster_offset & ~QCOW_OFLAG_COPIED;
931

    
932
    if (cluster_offset)
933
        free_any_clusters(bs, cluster_offset, 1);
934

    
935
    cluster_offset = alloc_bytes(bs, compressed_size);
936
    nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) -
937
                  (cluster_offset >> 9);
938

    
939
    cluster_offset |= QCOW_OFLAG_COMPRESSED |
940
                      ((uint64_t)nb_csectors << s->csize_shift);
941

    
942
    /* update L2 table */
943

    
944
    /* compressed clusters never have the copied flag */
945

    
946
    l2_table[l2_index] = cpu_to_be64(cluster_offset);
947
    if (bdrv_pwrite(s->hd,
948
                    l2_offset + l2_index * sizeof(uint64_t),
949
                    l2_table + l2_index,
950
                    sizeof(uint64_t)) != sizeof(uint64_t))
951
        return 0;
952

    
953
    return cluster_offset;
954
}
955

    
956
typedef struct QCowL2Meta
957
{
958
    uint64_t offset;
959
    int n_start;
960
    int nb_available;
961
    int nb_clusters;
962
} QCowL2Meta;
963

    
964
static int alloc_cluster_link_l2(BlockDriverState *bs, uint64_t cluster_offset,
965
        QCowL2Meta *m)
966
{
967
    BDRVQcowState *s = bs->opaque;
968
    int i, j = 0, l2_index, ret;
969
    uint64_t *old_cluster, start_sect, l2_offset, *l2_table;
970

    
971
    if (m->nb_clusters == 0)
972
        return 0;
973

    
974
    old_cluster = qemu_malloc(m->nb_clusters * sizeof(uint64_t));
975

    
976
    /* copy content of unmodified sectors */
977
    start_sect = (m->offset & ~(s->cluster_size - 1)) >> 9;
978
    if (m->n_start) {
979
        ret = copy_sectors(bs, start_sect, cluster_offset, 0, m->n_start);
980
        if (ret < 0)
981
            goto err;
982
    }
983

    
984
    if (m->nb_available & (s->cluster_sectors - 1)) {
985
        uint64_t end = m->nb_available & ~(uint64_t)(s->cluster_sectors - 1);
986
        ret = copy_sectors(bs, start_sect + end, cluster_offset + (end << 9),
987
                m->nb_available - end, s->cluster_sectors);
988
        if (ret < 0)
989
            goto err;
990
    }
991

    
992
    ret = -EIO;
993
    /* update L2 table */
994
    if (!get_cluster_table(bs, m->offset, &l2_table, &l2_offset, &l2_index))
995
        goto err;
996

    
997
    for (i = 0; i < m->nb_clusters; i++) {
998
        /* if two concurrent writes happen to the same unallocated cluster
999
         * each write allocates separate cluster and writes data concurrently.
1000
         * The first one to complete updates l2 table with pointer to its
1001
         * cluster the second one has to do RMW (which is done above by
1002
         * copy_sectors()), update l2 table with its cluster pointer and free
1003
         * old cluster. This is what this loop does */
1004
        if(l2_table[l2_index + i] != 0)
1005
            old_cluster[j++] = l2_table[l2_index + i];
1006

    
1007
        l2_table[l2_index + i] = cpu_to_be64((cluster_offset +
1008
                    (i << s->cluster_bits)) | QCOW_OFLAG_COPIED);
1009
     }
1010

    
1011
    if (bdrv_pwrite(s->hd, l2_offset + l2_index * sizeof(uint64_t),
1012
                l2_table + l2_index, m->nb_clusters * sizeof(uint64_t)) !=
1013
            m->nb_clusters * sizeof(uint64_t))
1014
        goto err;
1015

    
1016
    for (i = 0; i < j; i++)
1017
        free_any_clusters(bs, be64_to_cpu(old_cluster[i]) & ~QCOW_OFLAG_COPIED,
1018
                          1);
1019

    
1020
    ret = 0;
1021
err:
1022
    qemu_free(old_cluster);
1023
    return ret;
1024
 }
1025

    
1026
/*
1027
 * alloc_cluster_offset
1028
 *
1029
 * For a given offset of the disk image, return cluster offset in
1030
 * qcow2 file.
1031
 *
1032
 * If the offset is not found, allocate a new cluster.
1033
 *
1034
 * Return the cluster offset if successful,
1035
 * Return 0, otherwise.
1036
 *
1037
 */
1038

    
1039
static uint64_t alloc_cluster_offset(BlockDriverState *bs,
1040
                                     uint64_t offset,
1041
                                     int n_start, int n_end,
1042
                                     int *num, QCowL2Meta *m)
1043
{
1044
    BDRVQcowState *s = bs->opaque;
1045
    int l2_index, ret;
1046
    uint64_t l2_offset, *l2_table, cluster_offset;
1047
    int nb_clusters, i = 0;
1048

    
1049
    ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
1050
    if (ret == 0)
1051
        return 0;
1052

    
1053
    nb_clusters = size_to_clusters(s, n_end << 9);
1054

    
1055
    nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
1056

    
1057
    cluster_offset = be64_to_cpu(l2_table[l2_index]);
1058

    
1059
    /* We keep all QCOW_OFLAG_COPIED clusters */
1060

    
1061
    if (cluster_offset & QCOW_OFLAG_COPIED) {
1062
        nb_clusters = count_contiguous_clusters(nb_clusters, s->cluster_size,
1063
                &l2_table[l2_index], 0, 0);
1064

    
1065
        cluster_offset &= ~QCOW_OFLAG_COPIED;
1066
        m->nb_clusters = 0;
1067

    
1068
        goto out;
1069
    }
1070

    
1071
    /* for the moment, multiple compressed clusters are not managed */
1072

    
1073
    if (cluster_offset & QCOW_OFLAG_COMPRESSED)
1074
        nb_clusters = 1;
1075

    
1076
    /* how many available clusters ? */
1077

    
1078
    while (i < nb_clusters) {
1079
        i += count_contiguous_clusters(nb_clusters - i, s->cluster_size,
1080
                &l2_table[l2_index], i, 0);
1081

    
1082
        if(be64_to_cpu(l2_table[l2_index + i]))
1083
            break;
1084

    
1085
        i += count_contiguous_free_clusters(nb_clusters - i,
1086
                &l2_table[l2_index + i]);
1087

    
1088
        cluster_offset = be64_to_cpu(l2_table[l2_index + i]);
1089

    
1090
        if ((cluster_offset & QCOW_OFLAG_COPIED) ||
1091
                (cluster_offset & QCOW_OFLAG_COMPRESSED))
1092
            break;
1093
    }
1094
    nb_clusters = i;
1095

    
1096
    /* allocate a new cluster */
1097

    
1098
    cluster_offset = alloc_clusters(bs, nb_clusters * s->cluster_size);
1099

    
1100
    /* save info needed for meta data update */
1101
    m->offset = offset;
1102
    m->n_start = n_start;
1103
    m->nb_clusters = nb_clusters;
1104

    
1105
out:
1106
    m->nb_available = MIN(nb_clusters << (s->cluster_bits - 9), n_end);
1107

    
1108
    *num = m->nb_available - n_start;
1109

    
1110
    return cluster_offset;
1111
}
1112

    
1113
static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
1114
                             int nb_sectors, int *pnum)
1115
{
1116
    uint64_t cluster_offset;
1117

    
1118
    *pnum = nb_sectors;
1119
    cluster_offset = get_cluster_offset(bs, sector_num << 9, pnum);
1120

    
1121
    return (cluster_offset != 0);
1122
}
1123

    
1124
static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
1125
                             const uint8_t *buf, int buf_size)
1126
{
1127
    z_stream strm1, *strm = &strm1;
1128
    int ret, out_len;
1129

    
1130
    memset(strm, 0, sizeof(*strm));
1131

    
1132
    strm->next_in = (uint8_t *)buf;
1133
    strm->avail_in = buf_size;
1134
    strm->next_out = out_buf;
1135
    strm->avail_out = out_buf_size;
1136

    
1137
    ret = inflateInit2(strm, -12);
1138
    if (ret != Z_OK)
1139
        return -1;
1140
    ret = inflate(strm, Z_FINISH);
1141
    out_len = strm->next_out - out_buf;
1142
    if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
1143
        out_len != out_buf_size) {
1144
        inflateEnd(strm);
1145
        return -1;
1146
    }
1147
    inflateEnd(strm);
1148
    return 0;
1149
}
1150

    
1151
static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset)
1152
{
1153
    int ret, csize, nb_csectors, sector_offset;
1154
    uint64_t coffset;
1155

    
1156
    coffset = cluster_offset & s->cluster_offset_mask;
1157
    if (s->cluster_cache_offset != coffset) {
1158
        nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
1159
        sector_offset = coffset & 511;
1160
        csize = nb_csectors * 512 - sector_offset;
1161
        ret = bdrv_read(s->hd, coffset >> 9, s->cluster_data, nb_csectors);
1162
        if (ret < 0) {
1163
            return -1;
1164
        }
1165
        if (decompress_buffer(s->cluster_cache, s->cluster_size,
1166
                              s->cluster_data + sector_offset, csize) < 0) {
1167
            return -1;
1168
        }
1169
        s->cluster_cache_offset = coffset;
1170
    }
1171
    return 0;
1172
}
1173

    
1174
/* handle reading after the end of the backing file */
1175
static int backing_read1(BlockDriverState *bs,
1176
                         int64_t sector_num, uint8_t *buf, int nb_sectors)
1177
{
1178
    int n1;
1179
    if ((sector_num + nb_sectors) <= bs->total_sectors)
1180
        return nb_sectors;
1181
    if (sector_num >= bs->total_sectors)
1182
        n1 = 0;
1183
    else
1184
        n1 = bs->total_sectors - sector_num;
1185
    memset(buf + n1 * 512, 0, 512 * (nb_sectors - n1));
1186
    return n1;
1187
}
1188

    
1189
static int qcow_read(BlockDriverState *bs, int64_t sector_num,
1190
                     uint8_t *buf, int nb_sectors)
1191
{
1192
    BDRVQcowState *s = bs->opaque;
1193
    int ret, index_in_cluster, n, n1;
1194
    uint64_t cluster_offset;
1195

    
1196
    while (nb_sectors > 0) {
1197
        n = nb_sectors;
1198
        cluster_offset = get_cluster_offset(bs, sector_num << 9, &n);
1199
        index_in_cluster = sector_num & (s->cluster_sectors - 1);
1200
        if (!cluster_offset) {
1201
            if (bs->backing_hd) {
1202
                /* read from the base image */
1203
                n1 = backing_read1(bs->backing_hd, sector_num, buf, n);
1204
                if (n1 > 0) {
1205
                    ret = bdrv_read(bs->backing_hd, sector_num, buf, n1);
1206
                    if (ret < 0)
1207
                        return -1;
1208
                }
1209
            } else {
1210
                memset(buf, 0, 512 * n);
1211
            }
1212
        } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
1213
            if (decompress_cluster(s, cluster_offset) < 0)
1214
                return -1;
1215
            memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n);
1216
        } else {
1217
            ret = bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
1218
            if (ret != n * 512)
1219
                return -1;
1220
            if (s->crypt_method) {
1221
                encrypt_sectors(s, sector_num, buf, buf, n, 0,
1222
                                &s->aes_decrypt_key);
1223
            }
1224
        }
1225
        nb_sectors -= n;
1226
        sector_num += n;
1227
        buf += n * 512;
1228
    }
1229
    return 0;
1230
}
1231

    
1232
static int qcow_write(BlockDriverState *bs, int64_t sector_num,
1233
                     const uint8_t *buf, int nb_sectors)
1234
{
1235
    BDRVQcowState *s = bs->opaque;
1236
    int ret, index_in_cluster, n;
1237
    uint64_t cluster_offset;
1238
    int n_end;
1239
    QCowL2Meta l2meta;
1240

    
1241
    while (nb_sectors > 0) {
1242
        index_in_cluster = sector_num & (s->cluster_sectors - 1);
1243
        n_end = index_in_cluster + nb_sectors;
1244
        if (s->crypt_method &&
1245
            n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
1246
            n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
1247
        cluster_offset = alloc_cluster_offset(bs, sector_num << 9,
1248
                                              index_in_cluster,
1249
                                              n_end, &n, &l2meta);
1250
        if (!cluster_offset)
1251
            return -1;
1252
        if (s->crypt_method) {
1253
            encrypt_sectors(s, sector_num, s->cluster_data, buf, n, 1,
1254
                            &s->aes_encrypt_key);
1255
            ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512,
1256
                              s->cluster_data, n * 512);
1257
        } else {
1258
            ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
1259
        }
1260
        if (ret != n * 512 || alloc_cluster_link_l2(bs, cluster_offset, &l2meta) < 0) {
1261
            free_any_clusters(bs, cluster_offset, l2meta.nb_clusters);
1262
            return -1;
1263
        }
1264
        nb_sectors -= n;
1265
        sector_num += n;
1266
        buf += n * 512;
1267
    }
1268
    s->cluster_cache_offset = -1; /* disable compressed cache */
1269
    return 0;
1270
}
1271

    
1272
typedef struct QCowAIOCB {
1273
    BlockDriverAIOCB common;
1274
    int64_t sector_num;
1275
    QEMUIOVector *qiov;
1276
    uint8_t *buf;
1277
    void *orig_buf;
1278
    int nb_sectors;
1279
    int n;
1280
    uint64_t cluster_offset;
1281
    uint8_t *cluster_data;
1282
    BlockDriverAIOCB *hd_aiocb;
1283
    struct iovec hd_iov;
1284
    QEMUIOVector hd_qiov;
1285
    QEMUBH *bh;
1286
    QCowL2Meta l2meta;
1287
} QCowAIOCB;
1288

    
1289
static void qcow_aio_read_cb(void *opaque, int ret);
1290
static void qcow_aio_read_bh(void *opaque)
1291
{
1292
    QCowAIOCB *acb = opaque;
1293
    qemu_bh_delete(acb->bh);
1294
    acb->bh = NULL;
1295
    qcow_aio_read_cb(opaque, 0);
1296
}
1297

    
1298
static int qcow_schedule_bh(QEMUBHFunc *cb, QCowAIOCB *acb)
1299
{
1300
    if (acb->bh)
1301
        return -EIO;
1302

    
1303
    acb->bh = qemu_bh_new(cb, acb);
1304
    if (!acb->bh)
1305
        return -EIO;
1306

    
1307
    qemu_bh_schedule(acb->bh);
1308

    
1309
    return 0;
1310
}
1311

    
1312
static void qcow_aio_read_cb(void *opaque, int ret)
1313
{
1314
    QCowAIOCB *acb = opaque;
1315
    BlockDriverState *bs = acb->common.bs;
1316
    BDRVQcowState *s = bs->opaque;
1317
    int index_in_cluster, n1;
1318

    
1319
    acb->hd_aiocb = NULL;
1320
    if (ret < 0)
1321
        goto done;
1322

    
1323
    /* post process the read buffer */
1324
    if (!acb->cluster_offset) {
1325
        /* nothing to do */
1326
    } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
1327
        /* nothing to do */
1328
    } else {
1329
        if (s->crypt_method) {
1330
            encrypt_sectors(s, acb->sector_num, acb->buf, acb->buf,
1331
                            acb->n, 0,
1332
                            &s->aes_decrypt_key);
1333
        }
1334
    }
1335

    
1336
    acb->nb_sectors -= acb->n;
1337
    acb->sector_num += acb->n;
1338
    acb->buf += acb->n * 512;
1339

    
1340
    if (acb->nb_sectors == 0) {
1341
        /* request completed */
1342
        ret = 0;
1343
        goto done;
1344
    }
1345

    
1346
    /* prepare next AIO request */
1347
    acb->n = acb->nb_sectors;
1348
    acb->cluster_offset = get_cluster_offset(bs, acb->sector_num << 9, &acb->n);
1349
    index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
1350

    
1351
    if (!acb->cluster_offset) {
1352
        if (bs->backing_hd) {
1353
            /* read from the base image */
1354
            n1 = backing_read1(bs->backing_hd, acb->sector_num,
1355
                               acb->buf, acb->n);
1356
            if (n1 > 0) {
1357
                acb->hd_iov.iov_base = (void *)acb->buf;
1358
                acb->hd_iov.iov_len = acb->n * 512;
1359
                qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
1360
                acb->hd_aiocb = bdrv_aio_readv(bs->backing_hd, acb->sector_num,
1361
                                    &acb->hd_qiov, acb->n,
1362
                                    qcow_aio_read_cb, acb);
1363
                if (acb->hd_aiocb == NULL)
1364
                    goto done;
1365
            } else {
1366
                ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
1367
                if (ret < 0)
1368
                    goto done;
1369
            }
1370
        } else {
1371
            /* Note: in this case, no need to wait */
1372
            memset(acb->buf, 0, 512 * acb->n);
1373
            ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
1374
            if (ret < 0)
1375
                goto done;
1376
        }
1377
    } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
1378
        /* add AIO support for compressed blocks ? */
1379
        if (decompress_cluster(s, acb->cluster_offset) < 0)
1380
            goto done;
1381
        memcpy(acb->buf,
1382
               s->cluster_cache + index_in_cluster * 512, 512 * acb->n);
1383
        ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
1384
        if (ret < 0)
1385
            goto done;
1386
    } else {
1387
        if ((acb->cluster_offset & 511) != 0) {
1388
            ret = -EIO;
1389
            goto done;
1390
        }
1391

    
1392
        acb->hd_iov.iov_base = (void *)acb->buf;
1393
        acb->hd_iov.iov_len = acb->n * 512;
1394
        qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
1395
        acb->hd_aiocb = bdrv_aio_readv(s->hd,
1396
                            (acb->cluster_offset >> 9) + index_in_cluster,
1397
                            &acb->hd_qiov, acb->n, qcow_aio_read_cb, acb);
1398
        if (acb->hd_aiocb == NULL)
1399
            goto done;
1400
    }
1401

    
1402
    return;
1403
done:
1404
    if (acb->qiov->niov > 1) {
1405
        qemu_iovec_from_buffer(acb->qiov, acb->orig_buf, acb->qiov->size);
1406
        qemu_vfree(acb->orig_buf);
1407
    }
1408
    acb->common.cb(acb->common.opaque, ret);
1409
    qemu_aio_release(acb);
1410
}
1411

    
1412
static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs,
1413
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1414
        BlockDriverCompletionFunc *cb, void *opaque, int is_write)
1415
{
1416
    QCowAIOCB *acb;
1417

    
1418
    acb = qemu_aio_get(bs, cb, opaque);
1419
    if (!acb)
1420
        return NULL;
1421
    acb->hd_aiocb = NULL;
1422
    acb->sector_num = sector_num;
1423
    acb->qiov = qiov;
1424
    if (qiov->niov > 1) {
1425
        acb->buf = acb->orig_buf = qemu_blockalign(bs, qiov->size);
1426
        if (is_write)
1427
            qemu_iovec_to_buffer(qiov, acb->buf);
1428
    } else {
1429
        acb->buf = (uint8_t *)qiov->iov->iov_base;
1430
    }
1431
    acb->nb_sectors = nb_sectors;
1432
    acb->n = 0;
1433
    acb->cluster_offset = 0;
1434
    acb->l2meta.nb_clusters = 0;
1435
    return acb;
1436
}
1437

    
1438
static BlockDriverAIOCB *qcow_aio_readv(BlockDriverState *bs,
1439
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1440
        BlockDriverCompletionFunc *cb, void *opaque)
1441
{
1442
    QCowAIOCB *acb;
1443

    
1444
    acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
1445
    if (!acb)
1446
        return NULL;
1447

    
1448
    qcow_aio_read_cb(acb, 0);
1449
    return &acb->common;
1450
}
1451

    
1452
static void qcow_aio_write_cb(void *opaque, int ret)
1453
{
1454
    QCowAIOCB *acb = opaque;
1455
    BlockDriverState *bs = acb->common.bs;
1456
    BDRVQcowState *s = bs->opaque;
1457
    int index_in_cluster;
1458
    const uint8_t *src_buf;
1459
    int n_end;
1460

    
1461
    acb->hd_aiocb = NULL;
1462

    
1463
    if (ret < 0)
1464
        goto done;
1465

    
1466
    if (alloc_cluster_link_l2(bs, acb->cluster_offset, &acb->l2meta) < 0) {
1467
        free_any_clusters(bs, acb->cluster_offset, acb->l2meta.nb_clusters);
1468
        goto done;
1469
    }
1470

    
1471
    acb->nb_sectors -= acb->n;
1472
    acb->sector_num += acb->n;
1473
    acb->buf += acb->n * 512;
1474

    
1475
    if (acb->nb_sectors == 0) {
1476
        /* request completed */
1477
        ret = 0;
1478
        goto done;
1479
    }
1480

    
1481
    index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
1482
    n_end = index_in_cluster + acb->nb_sectors;
1483
    if (s->crypt_method &&
1484
        n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
1485
        n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
1486

    
1487
    acb->cluster_offset = alloc_cluster_offset(bs, acb->sector_num << 9,
1488
                                          index_in_cluster,
1489
                                          n_end, &acb->n, &acb->l2meta);
1490
    if (!acb->cluster_offset || (acb->cluster_offset & 511) != 0) {
1491
        ret = -EIO;
1492
        goto done;
1493
    }
1494
    if (s->crypt_method) {
1495
        if (!acb->cluster_data) {
1496
            acb->cluster_data = qemu_mallocz(QCOW_MAX_CRYPT_CLUSTERS *
1497
                                             s->cluster_size);
1498
        }
1499
        encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf,
1500
                        acb->n, 1, &s->aes_encrypt_key);
1501
        src_buf = acb->cluster_data;
1502
    } else {
1503
        src_buf = acb->buf;
1504
    }
1505
    acb->hd_iov.iov_base = (void *)src_buf;
1506
    acb->hd_iov.iov_len = acb->n * 512;
1507
    qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
1508
    acb->hd_aiocb = bdrv_aio_writev(s->hd,
1509
                                    (acb->cluster_offset >> 9) + index_in_cluster,
1510
                                    &acb->hd_qiov, acb->n,
1511
                                    qcow_aio_write_cb, acb);
1512
    if (acb->hd_aiocb == NULL)
1513
        goto done;
1514

    
1515
    return;
1516

    
1517
done:
1518
    if (acb->qiov->niov > 1)
1519
        qemu_vfree(acb->orig_buf);
1520
    acb->common.cb(acb->common.opaque, ret);
1521
    qemu_aio_release(acb);
1522
}
1523

    
1524
static BlockDriverAIOCB *qcow_aio_writev(BlockDriverState *bs,
1525
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1526
        BlockDriverCompletionFunc *cb, void *opaque)
1527
{
1528
    BDRVQcowState *s = bs->opaque;
1529
    QCowAIOCB *acb;
1530

    
1531
    s->cluster_cache_offset = -1; /* disable compressed cache */
1532

    
1533
    acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
1534
    if (!acb)
1535
        return NULL;
1536

    
1537
    qcow_aio_write_cb(acb, 0);
1538
    return &acb->common;
1539
}
1540

    
1541
static void qcow_aio_cancel(BlockDriverAIOCB *blockacb)
1542
{
1543
    QCowAIOCB *acb = (QCowAIOCB *)blockacb;
1544
    if (acb->hd_aiocb)
1545
        bdrv_aio_cancel(acb->hd_aiocb);
1546
    qemu_aio_release(acb);
1547
}
1548

    
1549
static void qcow_close(BlockDriverState *bs)
1550
{
1551
    BDRVQcowState *s = bs->opaque;
1552
    qemu_free(s->l1_table);
1553
    qemu_free(s->l2_cache);
1554
    qemu_free(s->cluster_cache);
1555
    qemu_free(s->cluster_data);
1556
    refcount_close(bs);
1557
    bdrv_delete(s->hd);
1558
}
1559

    
1560
/* XXX: use std qcow open function ? */
1561
typedef struct QCowCreateState {
1562
    int cluster_size;
1563
    int cluster_bits;
1564
    uint16_t *refcount_block;
1565
    uint64_t *refcount_table;
1566
    int64_t l1_table_offset;
1567
    int64_t refcount_table_offset;
1568
    int64_t refcount_block_offset;
1569
} QCowCreateState;
1570

    
1571
static void create_refcount_update(QCowCreateState *s,
1572
                                   int64_t offset, int64_t size)
1573
{
1574
    int refcount;
1575
    int64_t start, last, cluster_offset;
1576
    uint16_t *p;
1577

    
1578
    start = offset & ~(s->cluster_size - 1);
1579
    last = (offset + size - 1)  & ~(s->cluster_size - 1);
1580
    for(cluster_offset = start; cluster_offset <= last;
1581
        cluster_offset += s->cluster_size) {
1582
        p = &s->refcount_block[cluster_offset >> s->cluster_bits];
1583
        refcount = be16_to_cpu(*p);
1584
        refcount++;
1585
        *p = cpu_to_be16(refcount);
1586
    }
1587
}
1588

    
1589
static int get_bits_from_size(size_t size)
1590
{
1591
    int res = 0;
1592

    
1593
    if (size == 0) {
1594
        return -1;
1595
    }
1596

    
1597
    while (size != 1) {
1598
        /* Not a power of two */
1599
        if (size & 1) {
1600
            return -1;
1601
        }
1602

    
1603
        size >>= 1;
1604
        res++;
1605
    }
1606

    
1607
    return res;
1608
}
1609

    
1610
static int qcow_create2(const char *filename, int64_t total_size,
1611
                        const char *backing_file, const char *backing_format,
1612
                        int flags, size_t cluster_size)
1613
{
1614

    
1615
    int fd, header_size, backing_filename_len, l1_size, i, shift, l2_bits;
1616
    int ref_clusters, backing_format_len = 0;
1617
    QCowHeader header;
1618
    uint64_t tmp, offset;
1619
    QCowCreateState s1, *s = &s1;
1620
    QCowExtension ext_bf = {0, 0};
1621

    
1622

    
1623
    memset(s, 0, sizeof(*s));
1624

    
1625
    fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
1626
    if (fd < 0)
1627
        return -1;
1628
    memset(&header, 0, sizeof(header));
1629
    header.magic = cpu_to_be32(QCOW_MAGIC);
1630
    header.version = cpu_to_be32(QCOW_VERSION);
1631
    header.size = cpu_to_be64(total_size * 512);
1632
    header_size = sizeof(header);
1633
    backing_filename_len = 0;
1634
    if (backing_file) {
1635
        if (backing_format) {
1636
            ext_bf.magic = QCOW_EXT_MAGIC_BACKING_FORMAT;
1637
            backing_format_len = strlen(backing_format);
1638
            ext_bf.len = (backing_format_len + 7) & ~7;
1639
            header_size += ((sizeof(ext_bf) + ext_bf.len + 7) & ~7);
1640
        }
1641
        header.backing_file_offset = cpu_to_be64(header_size);
1642
        backing_filename_len = strlen(backing_file);
1643
        header.backing_file_size = cpu_to_be32(backing_filename_len);
1644
        header_size += backing_filename_len;
1645
    }
1646

    
1647
    /* Cluster size */
1648
    s->cluster_bits = get_bits_from_size(cluster_size);
1649
    if (s->cluster_bits < MIN_CLUSTER_BITS ||
1650
        s->cluster_bits > MAX_CLUSTER_BITS)
1651
    {
1652
        fprintf(stderr, "Cluster size must be a power of two between "
1653
            "%d and %dk\n",
1654
            1 << MIN_CLUSTER_BITS,
1655
            1 << (MAX_CLUSTER_BITS - 10));
1656
        return -EINVAL;
1657
    }
1658
    s->cluster_size = 1 << s->cluster_bits;
1659

    
1660
    header.cluster_bits = cpu_to_be32(s->cluster_bits);
1661
    header_size = (header_size + 7) & ~7;
1662
    if (flags & BLOCK_FLAG_ENCRYPT) {
1663
        header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
1664
    } else {
1665
        header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
1666
    }
1667
    l2_bits = s->cluster_bits - 3;
1668
    shift = s->cluster_bits + l2_bits;
1669
    l1_size = (((total_size * 512) + (1LL << shift) - 1) >> shift);
1670
    offset = align_offset(header_size, s->cluster_size);
1671
    s->l1_table_offset = offset;
1672
    header.l1_table_offset = cpu_to_be64(s->l1_table_offset);
1673
    header.l1_size = cpu_to_be32(l1_size);
1674
    offset += align_offset(l1_size * sizeof(uint64_t), s->cluster_size);
1675

    
1676
    s->refcount_table = qemu_mallocz(s->cluster_size);
1677

    
1678
    s->refcount_table_offset = offset;
1679
    header.refcount_table_offset = cpu_to_be64(offset);
1680
    header.refcount_table_clusters = cpu_to_be32(1);
1681
    offset += s->cluster_size;
1682
    s->refcount_block_offset = offset;
1683

    
1684
    /* count how many refcount blocks needed */
1685
    tmp = offset >> s->cluster_bits;
1686
    ref_clusters = (tmp >> (s->cluster_bits - REFCOUNT_SHIFT)) + 1;
1687
    for (i=0; i < ref_clusters; i++) {
1688
        s->refcount_table[i] = cpu_to_be64(offset);
1689
        offset += s->cluster_size;
1690
    }
1691

    
1692
    s->refcount_block = qemu_mallocz(ref_clusters * s->cluster_size);
1693

    
1694
    /* update refcounts */
1695
    create_refcount_update(s, 0, header_size);
1696
    create_refcount_update(s, s->l1_table_offset, l1_size * sizeof(uint64_t));
1697
    create_refcount_update(s, s->refcount_table_offset, s->cluster_size);
1698
    create_refcount_update(s, s->refcount_block_offset, ref_clusters * s->cluster_size);
1699

    
1700
    /* write all the data */
1701
    write(fd, &header, sizeof(header));
1702
    if (backing_file) {
1703
        if (backing_format_len) {
1704
            char zero[16];
1705
            int d = ext_bf.len - backing_format_len;
1706

    
1707
            memset(zero, 0, sizeof(zero));
1708
            cpu_to_be32s(&ext_bf.magic);
1709
            cpu_to_be32s(&ext_bf.len);
1710
            write(fd, &ext_bf, sizeof(ext_bf));
1711
            write(fd, backing_format, backing_format_len);
1712
            if (d>0) {
1713
                write(fd, zero, d);
1714
            }
1715
        }
1716
        write(fd, backing_file, backing_filename_len);
1717
    }
1718
    lseek(fd, s->l1_table_offset, SEEK_SET);
1719
    tmp = 0;
1720
    for(i = 0;i < l1_size; i++) {
1721
        write(fd, &tmp, sizeof(tmp));
1722
    }
1723
    lseek(fd, s->refcount_table_offset, SEEK_SET);
1724
    write(fd, s->refcount_table, s->cluster_size);
1725

    
1726
    lseek(fd, s->refcount_block_offset, SEEK_SET);
1727
    write(fd, s->refcount_block, ref_clusters * s->cluster_size);
1728

    
1729
    qemu_free(s->refcount_table);
1730
    qemu_free(s->refcount_block);
1731
    close(fd);
1732
    return 0;
1733
}
1734

    
1735
static int qcow_create(const char *filename, QEMUOptionParameter *options)
1736
{
1737
    const char *backing_file = NULL;
1738
    const char *backing_fmt = NULL;
1739
    uint64_t sectors = 0;
1740
    int flags = 0;
1741
    size_t cluster_size = 4096;
1742

    
1743
    /* Read out options */
1744
    while (options && options->name) {
1745
        if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
1746
            sectors = options->value.n / 512;
1747
        } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
1748
            backing_file = options->value.s;
1749
        } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FMT)) {
1750
            backing_fmt = options->value.s;
1751
        } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) {
1752
            flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
1753
        } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
1754
            if (options->value.n) {
1755
                cluster_size = options->value.n;
1756
            }
1757
        }
1758
        options++;
1759
    }
1760

    
1761
    return qcow_create2(filename, sectors, backing_file, backing_fmt, flags,
1762
        cluster_size);
1763
}
1764

    
1765
static int qcow_make_empty(BlockDriverState *bs)
1766
{
1767
#if 0
1768
    /* XXX: not correct */
1769
    BDRVQcowState *s = bs->opaque;
1770
    uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1771
    int ret;
1772

1773
    memset(s->l1_table, 0, l1_length);
1774
    if (bdrv_pwrite(s->hd, s->l1_table_offset, s->l1_table, l1_length) < 0)
1775
        return -1;
1776
    ret = bdrv_truncate(s->hd, s->l1_table_offset + l1_length);
1777
    if (ret < 0)
1778
        return ret;
1779

1780
    l2_cache_reset(bs);
1781
#endif
1782
    return 0;
1783
}
1784

    
1785
/* XXX: put compressed sectors first, then all the cluster aligned
1786
   tables to avoid losing bytes in alignment */
1787
static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
1788
                                 const uint8_t *buf, int nb_sectors)
1789
{
1790
    BDRVQcowState *s = bs->opaque;
1791
    z_stream strm;
1792
    int ret, out_len;
1793
    uint8_t *out_buf;
1794
    uint64_t cluster_offset;
1795

    
1796
    if (nb_sectors == 0) {
1797
        /* align end of file to a sector boundary to ease reading with
1798
           sector based I/Os */
1799
        cluster_offset = bdrv_getlength(s->hd);
1800
        cluster_offset = (cluster_offset + 511) & ~511;
1801
        bdrv_truncate(s->hd, cluster_offset);
1802
        return 0;
1803
    }
1804

    
1805
    if (nb_sectors != s->cluster_sectors)
1806
        return -EINVAL;
1807

    
1808
    out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
1809

    
1810
    /* best compression, small window, no zlib header */
1811
    memset(&strm, 0, sizeof(strm));
1812
    ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1813
                       Z_DEFLATED, -12,
1814
                       9, Z_DEFAULT_STRATEGY);
1815
    if (ret != 0) {
1816
        qemu_free(out_buf);
1817
        return -1;
1818
    }
1819

    
1820
    strm.avail_in = s->cluster_size;
1821
    strm.next_in = (uint8_t *)buf;
1822
    strm.avail_out = s->cluster_size;
1823
    strm.next_out = out_buf;
1824

    
1825
    ret = deflate(&strm, Z_FINISH);
1826
    if (ret != Z_STREAM_END && ret != Z_OK) {
1827
        qemu_free(out_buf);
1828
        deflateEnd(&strm);
1829
        return -1;
1830
    }
1831
    out_len = strm.next_out - out_buf;
1832

    
1833
    deflateEnd(&strm);
1834

    
1835
    if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1836
        /* could not compress: write normal cluster */
1837
        qcow_write(bs, sector_num, buf, s->cluster_sectors);
1838
    } else {
1839
        cluster_offset = alloc_compressed_cluster_offset(bs, sector_num << 9,
1840
                                              out_len);
1841
        if (!cluster_offset)
1842
            return -1;
1843
        cluster_offset &= s->cluster_offset_mask;
1844
        if (bdrv_pwrite(s->hd, cluster_offset, out_buf, out_len) != out_len) {
1845
            qemu_free(out_buf);
1846
            return -1;
1847
        }
1848
    }
1849

    
1850
    qemu_free(out_buf);
1851
    return 0;
1852
}
1853

    
1854
static void qcow_flush(BlockDriverState *bs)
1855
{
1856
    BDRVQcowState *s = bs->opaque;
1857
    bdrv_flush(s->hd);
1858
}
1859

    
1860
static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1861
{
1862
    BDRVQcowState *s = bs->opaque;
1863
    bdi->cluster_size = s->cluster_size;
1864
    bdi->vm_state_offset = (int64_t)s->l1_vm_state_index <<
1865
        (s->cluster_bits + s->l2_bits);
1866
    return 0;
1867
}
1868

    
1869
/*********************************************************/
1870
/* snapshot support */
1871

    
1872
/* update the refcounts of snapshots and the copied flag */
1873
static int update_snapshot_refcount(BlockDriverState *bs,
1874
                                    int64_t l1_table_offset,
1875
                                    int l1_size,
1876
                                    int addend)
1877
{
1878
    BDRVQcowState *s = bs->opaque;
1879
    uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated;
1880
    int64_t old_offset, old_l2_offset;
1881
    int l2_size, i, j, l1_modified, l2_modified, nb_csectors, refcount;
1882

    
1883
    l2_cache_reset(bs);
1884

    
1885
    l2_table = NULL;
1886
    l1_table = NULL;
1887
    l1_size2 = l1_size * sizeof(uint64_t);
1888
    l1_allocated = 0;
1889
    if (l1_table_offset != s->l1_table_offset) {
1890
        l1_table = qemu_malloc(l1_size2);
1891
        l1_allocated = 1;
1892
        if (bdrv_pread(s->hd, l1_table_offset,
1893
                       l1_table, l1_size2) != l1_size2)
1894
            goto fail;
1895
        for(i = 0;i < l1_size; i++)
1896
            be64_to_cpus(&l1_table[i]);
1897
    } else {
1898
        assert(l1_size == s->l1_size);
1899
        l1_table = s->l1_table;
1900
        l1_allocated = 0;
1901
    }
1902

    
1903
    l2_size = s->l2_size * sizeof(uint64_t);
1904
    l2_table = qemu_malloc(l2_size);
1905
    l1_modified = 0;
1906
    for(i = 0; i < l1_size; i++) {
1907
        l2_offset = l1_table[i];
1908
        if (l2_offset) {
1909
            old_l2_offset = l2_offset;
1910
            l2_offset &= ~QCOW_OFLAG_COPIED;
1911
            l2_modified = 0;
1912
            if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
1913
                goto fail;
1914
            for(j = 0; j < s->l2_size; j++) {
1915
                offset = be64_to_cpu(l2_table[j]);
1916
                if (offset != 0) {
1917
                    old_offset = offset;
1918
                    offset &= ~QCOW_OFLAG_COPIED;
1919
                    if (offset & QCOW_OFLAG_COMPRESSED) {
1920
                        nb_csectors = ((offset >> s->csize_shift) &
1921
                                       s->csize_mask) + 1;
1922
                        if (addend != 0)
1923
                            update_refcount(bs, (offset & s->cluster_offset_mask) & ~511,
1924
                                            nb_csectors * 512, addend);
1925
                        /* compressed clusters are never modified */
1926
                        refcount = 2;
1927
                    } else {
1928
                        if (addend != 0) {
1929
                            refcount = update_cluster_refcount(bs, offset >> s->cluster_bits, addend);
1930
                        } else {
1931
                            refcount = get_refcount(bs, offset >> s->cluster_bits);
1932
                        }
1933
                    }
1934

    
1935
                    if (refcount == 1) {
1936
                        offset |= QCOW_OFLAG_COPIED;
1937
                    }
1938
                    if (offset != old_offset) {
1939
                        l2_table[j] = cpu_to_be64(offset);
1940
                        l2_modified = 1;
1941
                    }
1942
                }
1943
            }
1944
            if (l2_modified) {
1945
                if (bdrv_pwrite(s->hd,
1946
                                l2_offset, l2_table, l2_size) != l2_size)
1947
                    goto fail;
1948
            }
1949

    
1950
            if (addend != 0) {
1951
                refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend);
1952
            } else {
1953
                refcount = get_refcount(bs, l2_offset >> s->cluster_bits);
1954
            }
1955
            if (refcount == 1) {
1956
                l2_offset |= QCOW_OFLAG_COPIED;
1957
            }
1958
            if (l2_offset != old_l2_offset) {
1959
                l1_table[i] = l2_offset;
1960
                l1_modified = 1;
1961
            }
1962
        }
1963
    }
1964
    if (l1_modified) {
1965
        for(i = 0; i < l1_size; i++)
1966
            cpu_to_be64s(&l1_table[i]);
1967
        if (bdrv_pwrite(s->hd, l1_table_offset, l1_table,
1968
                        l1_size2) != l1_size2)
1969
            goto fail;
1970
        for(i = 0; i < l1_size; i++)
1971
            be64_to_cpus(&l1_table[i]);
1972
    }
1973
    if (l1_allocated)
1974
        qemu_free(l1_table);
1975
    qemu_free(l2_table);
1976
    return 0;
1977
 fail:
1978
    if (l1_allocated)
1979
        qemu_free(l1_table);
1980
    qemu_free(l2_table);
1981
    return -EIO;
1982
}
1983

    
1984
static void qcow_free_snapshots(BlockDriverState *bs)
1985
{
1986
    BDRVQcowState *s = bs->opaque;
1987
    int i;
1988

    
1989
    for(i = 0; i < s->nb_snapshots; i++) {
1990
        qemu_free(s->snapshots[i].name);
1991
        qemu_free(s->snapshots[i].id_str);
1992
    }
1993
    qemu_free(s->snapshots);
1994
    s->snapshots = NULL;
1995
    s->nb_snapshots = 0;
1996
}
1997

    
1998
static int qcow_read_snapshots(BlockDriverState *bs)
1999
{
2000
    BDRVQcowState *s = bs->opaque;
2001
    QCowSnapshotHeader h;
2002
    QCowSnapshot *sn;
2003
    int i, id_str_size, name_size;
2004
    int64_t offset;
2005
    uint32_t extra_data_size;
2006

    
2007
    if (!s->nb_snapshots) {
2008
        s->snapshots = NULL;
2009
        s->snapshots_size = 0;
2010
        return 0;
2011
    }
2012

    
2013
    offset = s->snapshots_offset;
2014
    s->snapshots = qemu_mallocz(s->nb_snapshots * sizeof(QCowSnapshot));
2015
    for(i = 0; i < s->nb_snapshots; i++) {
2016
        offset = align_offset(offset, 8);
2017
        if (bdrv_pread(s->hd, offset, &h, sizeof(h)) != sizeof(h))
2018
            goto fail;
2019
        offset += sizeof(h);
2020
        sn = s->snapshots + i;
2021
        sn->l1_table_offset = be64_to_cpu(h.l1_table_offset);
2022
        sn->l1_size = be32_to_cpu(h.l1_size);
2023
        sn->vm_state_size = be32_to_cpu(h.vm_state_size);
2024
        sn->date_sec = be32_to_cpu(h.date_sec);
2025
        sn->date_nsec = be32_to_cpu(h.date_nsec);
2026
        sn->vm_clock_nsec = be64_to_cpu(h.vm_clock_nsec);
2027
        extra_data_size = be32_to_cpu(h.extra_data_size);
2028

    
2029
        id_str_size = be16_to_cpu(h.id_str_size);
2030
        name_size = be16_to_cpu(h.name_size);
2031

    
2032
        offset += extra_data_size;
2033

    
2034
        sn->id_str = qemu_malloc(id_str_size + 1);
2035
        if (bdrv_pread(s->hd, offset, sn->id_str, id_str_size) != id_str_size)
2036
            goto fail;
2037
        offset += id_str_size;
2038
        sn->id_str[id_str_size] = '\0';
2039

    
2040
        sn->name = qemu_malloc(name_size + 1);
2041
        if (bdrv_pread(s->hd, offset, sn->name, name_size) != name_size)
2042
            goto fail;
2043
        offset += name_size;
2044
        sn->name[name_size] = '\0';
2045
    }
2046
    s->snapshots_size = offset - s->snapshots_offset;
2047
    return 0;
2048
 fail:
2049
    qcow_free_snapshots(bs);
2050
    return -1;
2051
}
2052

    
2053
/* add at the end of the file a new list of snapshots */
2054
static int qcow_write_snapshots(BlockDriverState *bs)
2055
{
2056
    BDRVQcowState *s = bs->opaque;
2057
    QCowSnapshot *sn;
2058
    QCowSnapshotHeader h;
2059
    int i, name_size, id_str_size, snapshots_size;
2060
    uint64_t data64;
2061
    uint32_t data32;
2062
    int64_t offset, snapshots_offset;
2063

    
2064
    /* compute the size of the snapshots */
2065
    offset = 0;
2066
    for(i = 0; i < s->nb_snapshots; i++) {
2067
        sn = s->snapshots + i;
2068
        offset = align_offset(offset, 8);
2069
        offset += sizeof(h);
2070
        offset += strlen(sn->id_str);
2071
        offset += strlen(sn->name);
2072
    }
2073
    snapshots_size = offset;
2074

    
2075
    snapshots_offset = alloc_clusters(bs, snapshots_size);
2076
    offset = snapshots_offset;
2077

    
2078
    for(i = 0; i < s->nb_snapshots; i++) {
2079
        sn = s->snapshots + i;
2080
        memset(&h, 0, sizeof(h));
2081
        h.l1_table_offset = cpu_to_be64(sn->l1_table_offset);
2082
        h.l1_size = cpu_to_be32(sn->l1_size);
2083
        h.vm_state_size = cpu_to_be32(sn->vm_state_size);
2084
        h.date_sec = cpu_to_be32(sn->date_sec);
2085
        h.date_nsec = cpu_to_be32(sn->date_nsec);
2086
        h.vm_clock_nsec = cpu_to_be64(sn->vm_clock_nsec);
2087

    
2088
        id_str_size = strlen(sn->id_str);
2089
        name_size = strlen(sn->name);
2090
        h.id_str_size = cpu_to_be16(id_str_size);
2091
        h.name_size = cpu_to_be16(name_size);
2092
        offset = align_offset(offset, 8);
2093
        if (bdrv_pwrite(s->hd, offset, &h, sizeof(h)) != sizeof(h))
2094
            goto fail;
2095
        offset += sizeof(h);
2096
        if (bdrv_pwrite(s->hd, offset, sn->id_str, id_str_size) != id_str_size)
2097
            goto fail;
2098
        offset += id_str_size;
2099
        if (bdrv_pwrite(s->hd, offset, sn->name, name_size) != name_size)
2100
            goto fail;
2101
        offset += name_size;
2102
    }
2103

    
2104
    /* update the various header fields */
2105
    data64 = cpu_to_be64(snapshots_offset);
2106
    if (bdrv_pwrite(s->hd, offsetof(QCowHeader, snapshots_offset),
2107
                    &data64, sizeof(data64)) != sizeof(data64))
2108
        goto fail;
2109
    data32 = cpu_to_be32(s->nb_snapshots);
2110
    if (bdrv_pwrite(s->hd, offsetof(QCowHeader, nb_snapshots),
2111
                    &data32, sizeof(data32)) != sizeof(data32))
2112
        goto fail;
2113

    
2114
    /* free the old snapshot table */
2115
    free_clusters(bs, s->snapshots_offset, s->snapshots_size);
2116
    s->snapshots_offset = snapshots_offset;
2117
    s->snapshots_size = snapshots_size;
2118
    return 0;
2119
 fail:
2120
    return -1;
2121
}
2122

    
2123
static void find_new_snapshot_id(BlockDriverState *bs,
2124
                                 char *id_str, int id_str_size)
2125
{
2126
    BDRVQcowState *s = bs->opaque;
2127
    QCowSnapshot *sn;
2128
    int i, id, id_max = 0;
2129

    
2130
    for(i = 0; i < s->nb_snapshots; i++) {
2131
        sn = s->snapshots + i;
2132
        id = strtoul(sn->id_str, NULL, 10);
2133
        if (id > id_max)
2134
            id_max = id;
2135
    }
2136
    snprintf(id_str, id_str_size, "%d", id_max + 1);
2137
}
2138

    
2139
static int find_snapshot_by_id(BlockDriverState *bs, const char *id_str)
2140
{
2141
    BDRVQcowState *s = bs->opaque;
2142
    int i;
2143

    
2144
    for(i = 0; i < s->nb_snapshots; i++) {
2145
        if (!strcmp(s->snapshots[i].id_str, id_str))
2146
            return i;
2147
    }
2148
    return -1;
2149
}
2150

    
2151
static int find_snapshot_by_id_or_name(BlockDriverState *bs, const char *name)
2152
{
2153
    BDRVQcowState *s = bs->opaque;
2154
    int i, ret;
2155

    
2156
    ret = find_snapshot_by_id(bs, name);
2157
    if (ret >= 0)
2158
        return ret;
2159
    for(i = 0; i < s->nb_snapshots; i++) {
2160
        if (!strcmp(s->snapshots[i].name, name))
2161
            return i;
2162
    }
2163
    return -1;
2164
}
2165

    
2166
/* if no id is provided, a new one is constructed */
2167
static int qcow_snapshot_create(BlockDriverState *bs,
2168
                                QEMUSnapshotInfo *sn_info)
2169
{
2170
    BDRVQcowState *s = bs->opaque;
2171
    QCowSnapshot *snapshots1, sn1, *sn = &sn1;
2172
    int i, ret;
2173
    uint64_t *l1_table = NULL;
2174

    
2175
    memset(sn, 0, sizeof(*sn));
2176

    
2177
    if (sn_info->id_str[0] == '\0') {
2178
        /* compute a new id */
2179
        find_new_snapshot_id(bs, sn_info->id_str, sizeof(sn_info->id_str));
2180
    }
2181

    
2182
    /* check that the ID is unique */
2183
    if (find_snapshot_by_id(bs, sn_info->id_str) >= 0)
2184
        return -ENOENT;
2185

    
2186
    sn->id_str = qemu_strdup(sn_info->id_str);
2187
    if (!sn->id_str)
2188
        goto fail;
2189
    sn->name = qemu_strdup(sn_info->name);
2190
    if (!sn->name)
2191
        goto fail;
2192
    sn->vm_state_size = sn_info->vm_state_size;
2193
    sn->date_sec = sn_info->date_sec;
2194
    sn->date_nsec = sn_info->date_nsec;
2195
    sn->vm_clock_nsec = sn_info->vm_clock_nsec;
2196

    
2197
    ret = update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1);
2198
    if (ret < 0)
2199
        goto fail;
2200

    
2201
    /* create the L1 table of the snapshot */
2202
    sn->l1_table_offset = alloc_clusters(bs, s->l1_size * sizeof(uint64_t));
2203
    sn->l1_size = s->l1_size;
2204

    
2205
    l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
2206
    for(i = 0; i < s->l1_size; i++) {
2207
        l1_table[i] = cpu_to_be64(s->l1_table[i]);
2208
    }
2209
    if (bdrv_pwrite(s->hd, sn->l1_table_offset,
2210
                    l1_table, s->l1_size * sizeof(uint64_t)) !=
2211
        (s->l1_size * sizeof(uint64_t)))
2212
        goto fail;
2213
    qemu_free(l1_table);
2214
    l1_table = NULL;
2215

    
2216
    snapshots1 = qemu_malloc((s->nb_snapshots + 1) * sizeof(QCowSnapshot));
2217
    if (s->snapshots) {
2218
        memcpy(snapshots1, s->snapshots, s->nb_snapshots * sizeof(QCowSnapshot));
2219
        qemu_free(s->snapshots);
2220
    }
2221
    s->snapshots = snapshots1;
2222
    s->snapshots[s->nb_snapshots++] = *sn;
2223

    
2224
    if (qcow_write_snapshots(bs) < 0)
2225
        goto fail;
2226
#ifdef DEBUG_ALLOC
2227
    check_refcounts(bs);
2228
#endif
2229
    return 0;
2230
 fail:
2231
    qemu_free(sn->name);
2232
    qemu_free(l1_table);
2233
    return -1;
2234
}
2235

    
2236
/* copy the snapshot 'snapshot_name' into the current disk image */
2237
static int qcow_snapshot_goto(BlockDriverState *bs,
2238
                              const char *snapshot_id)
2239
{
2240
    BDRVQcowState *s = bs->opaque;
2241
    QCowSnapshot *sn;
2242
    int i, snapshot_index, l1_size2;
2243

    
2244
    snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
2245
    if (snapshot_index < 0)
2246
        return -ENOENT;
2247
    sn = &s->snapshots[snapshot_index];
2248

    
2249
    if (update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, -1) < 0)
2250
        goto fail;
2251

    
2252
    if (grow_l1_table(bs, sn->l1_size) < 0)
2253
        goto fail;
2254

    
2255
    s->l1_size = sn->l1_size;
2256
    l1_size2 = s->l1_size * sizeof(uint64_t);
2257
    /* copy the snapshot l1 table to the current l1 table */
2258
    if (bdrv_pread(s->hd, sn->l1_table_offset,
2259
                   s->l1_table, l1_size2) != l1_size2)
2260
        goto fail;
2261
    if (bdrv_pwrite(s->hd, s->l1_table_offset,
2262
                    s->l1_table, l1_size2) != l1_size2)
2263
        goto fail;
2264
    for(i = 0;i < s->l1_size; i++) {
2265
        be64_to_cpus(&s->l1_table[i]);
2266
    }
2267

    
2268
    if (update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1) < 0)
2269
        goto fail;
2270

    
2271
#ifdef DEBUG_ALLOC
2272
    check_refcounts(bs);
2273
#endif
2274
    return 0;
2275
 fail:
2276
    return -EIO;
2277
}
2278

    
2279
static int qcow_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
2280
{
2281
    BDRVQcowState *s = bs->opaque;
2282
    QCowSnapshot *sn;
2283
    int snapshot_index, ret;
2284

    
2285
    snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
2286
    if (snapshot_index < 0)
2287
        return -ENOENT;
2288
    sn = &s->snapshots[snapshot_index];
2289

    
2290
    ret = update_snapshot_refcount(bs, sn->l1_table_offset, sn->l1_size, -1);
2291
    if (ret < 0)
2292
        return ret;
2293
    /* must update the copied flag on the current cluster offsets */
2294
    ret = update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 0);
2295
    if (ret < 0)
2296
        return ret;
2297
    free_clusters(bs, sn->l1_table_offset, sn->l1_size * sizeof(uint64_t));
2298

    
2299
    qemu_free(sn->id_str);
2300
    qemu_free(sn->name);
2301
    memmove(sn, sn + 1, (s->nb_snapshots - snapshot_index - 1) * sizeof(*sn));
2302
    s->nb_snapshots--;
2303
    ret = qcow_write_snapshots(bs);
2304
    if (ret < 0) {
2305
        /* XXX: restore snapshot if error ? */
2306
        return ret;
2307
    }
2308
#ifdef DEBUG_ALLOC
2309
    check_refcounts(bs);
2310
#endif
2311
    return 0;
2312
}
2313

    
2314
static int qcow_snapshot_list(BlockDriverState *bs,
2315
                              QEMUSnapshotInfo **psn_tab)
2316
{
2317
    BDRVQcowState *s = bs->opaque;
2318
    QEMUSnapshotInfo *sn_tab, *sn_info;
2319
    QCowSnapshot *sn;
2320
    int i;
2321

    
2322
    if (!s->nb_snapshots) {
2323
        *psn_tab = NULL;
2324
        return s->nb_snapshots;
2325
    }
2326

    
2327
    sn_tab = qemu_mallocz(s->nb_snapshots * sizeof(QEMUSnapshotInfo));
2328
    for(i = 0; i < s->nb_snapshots; i++) {
2329
        sn_info = sn_tab + i;
2330
        sn = s->snapshots + i;
2331
        pstrcpy(sn_info->id_str, sizeof(sn_info->id_str),
2332
                sn->id_str);
2333
        pstrcpy(sn_info->name, sizeof(sn_info->name),
2334
                sn->name);
2335
        sn_info->vm_state_size = sn->vm_state_size;
2336
        sn_info->date_sec = sn->date_sec;
2337
        sn_info->date_nsec = sn->date_nsec;
2338
        sn_info->vm_clock_nsec = sn->vm_clock_nsec;
2339
    }
2340
    *psn_tab = sn_tab;
2341
    return s->nb_snapshots;
2342
}
2343

    
2344
/*********************************************************/
2345
/* refcount handling */
2346

    
2347
static int refcount_init(BlockDriverState *bs)
2348
{
2349
    BDRVQcowState *s = bs->opaque;
2350
    int ret, refcount_table_size2, i;
2351

    
2352
    s->refcount_block_cache = qemu_malloc(s->cluster_size);
2353
    refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t);
2354
    s->refcount_table = qemu_malloc(refcount_table_size2);
2355
    if (s->refcount_table_size > 0) {
2356
        ret = bdrv_pread(s->hd, s->refcount_table_offset,
2357
                         s->refcount_table, refcount_table_size2);
2358
        if (ret != refcount_table_size2)
2359
            goto fail;
2360
        for(i = 0; i < s->refcount_table_size; i++)
2361
            be64_to_cpus(&s->refcount_table[i]);
2362
    }
2363
    return 0;
2364
 fail:
2365
    return -ENOMEM;
2366
}
2367

    
2368
static void refcount_close(BlockDriverState *bs)
2369
{
2370
    BDRVQcowState *s = bs->opaque;
2371
    qemu_free(s->refcount_block_cache);
2372
    qemu_free(s->refcount_table);
2373
}
2374

    
2375

    
2376
static int load_refcount_block(BlockDriverState *bs,
2377
                               int64_t refcount_block_offset)
2378
{
2379
    BDRVQcowState *s = bs->opaque;
2380
    int ret;
2381
    ret = bdrv_pread(s->hd, refcount_block_offset, s->refcount_block_cache,
2382
                     s->cluster_size);
2383
    if (ret != s->cluster_size)
2384
        return -EIO;
2385
    s->refcount_block_cache_offset = refcount_block_offset;
2386
    return 0;
2387
}
2388

    
2389
static int get_refcount(BlockDriverState *bs, int64_t cluster_index)
2390
{
2391
    BDRVQcowState *s = bs->opaque;
2392
    int refcount_table_index, block_index;
2393
    int64_t refcount_block_offset;
2394

    
2395
    refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
2396
    if (refcount_table_index >= s->refcount_table_size)
2397
        return 0;
2398
    refcount_block_offset = s->refcount_table[refcount_table_index];
2399
    if (!refcount_block_offset)
2400
        return 0;
2401
    if (refcount_block_offset != s->refcount_block_cache_offset) {
2402
        /* better than nothing: return allocated if read error */
2403
        if (load_refcount_block(bs, refcount_block_offset) < 0)
2404
            return 1;
2405
    }
2406
    block_index = cluster_index &
2407
        ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
2408
    return be16_to_cpu(s->refcount_block_cache[block_index]);
2409
}
2410

    
2411
/* return < 0 if error */
2412
static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size)
2413
{
2414
    BDRVQcowState *s = bs->opaque;
2415
    int i, nb_clusters;
2416

    
2417
    nb_clusters = size_to_clusters(s, size);
2418
retry:
2419
    for(i = 0; i < nb_clusters; i++) {
2420
        int64_t i = s->free_cluster_index++;
2421
        if (get_refcount(bs, i) != 0)
2422
            goto retry;
2423
    }
2424
#ifdef DEBUG_ALLOC2
2425
    printf("alloc_clusters: size=%lld -> %lld\n",
2426
            size,
2427
            (s->free_cluster_index - nb_clusters) << s->cluster_bits);
2428
#endif
2429
    return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
2430
}
2431

    
2432
static int64_t alloc_clusters(BlockDriverState *bs, int64_t size)
2433
{
2434
    int64_t offset;
2435

    
2436
    offset = alloc_clusters_noref(bs, size);
2437
    update_refcount(bs, offset, size, 1);
2438
    return offset;
2439
}
2440

    
2441
/* only used to allocate compressed sectors. We try to allocate
2442
   contiguous sectors. size must be <= cluster_size */
2443
static int64_t alloc_bytes(BlockDriverState *bs, int size)
2444
{
2445
    BDRVQcowState *s = bs->opaque;
2446
    int64_t offset, cluster_offset;
2447
    int free_in_cluster;
2448

    
2449
    assert(size > 0 && size <= s->cluster_size);
2450
    if (s->free_byte_offset == 0) {
2451
        s->free_byte_offset = alloc_clusters(bs, s->cluster_size);
2452
    }
2453
 redo:
2454
    free_in_cluster = s->cluster_size -
2455
        (s->free_byte_offset & (s->cluster_size - 1));
2456
    if (size <= free_in_cluster) {
2457
        /* enough space in current cluster */
2458
        offset = s->free_byte_offset;
2459
        s->free_byte_offset += size;
2460
        free_in_cluster -= size;
2461
        if (free_in_cluster == 0)
2462
            s->free_byte_offset = 0;
2463
        if ((offset & (s->cluster_size - 1)) != 0)
2464
            update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
2465
    } else {
2466
        offset = alloc_clusters(bs, s->cluster_size);
2467
        cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1);
2468
        if ((cluster_offset + s->cluster_size) == offset) {
2469
            /* we are lucky: contiguous data */
2470
            offset = s->free_byte_offset;
2471
            update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
2472
            s->free_byte_offset += size;
2473
        } else {
2474
            s->free_byte_offset = offset;
2475
            goto redo;
2476
        }
2477
    }
2478
    return offset;
2479
}
2480

    
2481
static void free_clusters(BlockDriverState *bs,
2482
                          int64_t offset, int64_t size)
2483
{
2484
    update_refcount(bs, offset, size, -1);
2485
}
2486

    
2487
static int grow_refcount_table(BlockDriverState *bs, int min_size)
2488
{
2489
    BDRVQcowState *s = bs->opaque;
2490
    int new_table_size, new_table_size2, refcount_table_clusters, i, ret;
2491
    uint64_t *new_table;
2492
    int64_t table_offset;
2493
    uint8_t data[12];
2494
    int old_table_size;
2495
    int64_t old_table_offset;
2496

    
2497
    if (min_size <= s->refcount_table_size)
2498
        return 0;
2499
    /* compute new table size */
2500
    refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
2501
    for(;;) {
2502
        if (refcount_table_clusters == 0) {
2503
            refcount_table_clusters = 1;
2504
        } else {
2505
            refcount_table_clusters = (refcount_table_clusters * 3 + 1) / 2;
2506
        }
2507
        new_table_size = refcount_table_clusters << (s->cluster_bits - 3);
2508
        if (min_size <= new_table_size)
2509
            break;
2510
    }
2511
#ifdef DEBUG_ALLOC2
2512
    printf("grow_refcount_table from %d to %d\n",
2513
           s->refcount_table_size,
2514
           new_table_size);
2515
#endif
2516
    new_table_size2 = new_table_size * sizeof(uint64_t);
2517
    new_table = qemu_mallocz(new_table_size2);
2518
    memcpy(new_table, s->refcount_table,
2519
           s->refcount_table_size * sizeof(uint64_t));
2520
    for(i = 0; i < s->refcount_table_size; i++)
2521
        cpu_to_be64s(&new_table[i]);
2522
    /* Note: we cannot update the refcount now to avoid recursion */
2523
    table_offset = alloc_clusters_noref(bs, new_table_size2);
2524
    ret = bdrv_pwrite(s->hd, table_offset, new_table, new_table_size2);
2525
    if (ret != new_table_size2)
2526
        goto fail;
2527
    for(i = 0; i < s->refcount_table_size; i++)
2528
        be64_to_cpus(&new_table[i]);
2529

    
2530
    cpu_to_be64w((uint64_t*)data, table_offset);
2531
    cpu_to_be32w((uint32_t*)(data + 8), refcount_table_clusters);
2532
    if (bdrv_pwrite(s->hd, offsetof(QCowHeader, refcount_table_offset),
2533
                    data, sizeof(data)) != sizeof(data))
2534
        goto fail;
2535
    qemu_free(s->refcount_table);
2536
    old_table_offset = s->refcount_table_offset;
2537
    old_table_size = s->refcount_table_size;
2538
    s->refcount_table = new_table;
2539
    s->refcount_table_size = new_table_size;
2540
    s->refcount_table_offset = table_offset;
2541

    
2542
    update_refcount(bs, table_offset, new_table_size2, 1);
2543
    free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t));
2544
    return 0;
2545
 fail:
2546
    free_clusters(bs, table_offset, new_table_size2);
2547
    qemu_free(new_table);
2548
    return -EIO;
2549
}
2550

    
2551
/* addend must be 1 or -1 */
2552
/* XXX: cache several refcount block clusters ? */
2553
static int update_cluster_refcount(BlockDriverState *bs,
2554
                                   int64_t cluster_index,
2555
                                   int addend)
2556
{
2557
    BDRVQcowState *s = bs->opaque;
2558
    int64_t offset, refcount_block_offset;
2559
    int ret, refcount_table_index, block_index, refcount;
2560
    uint64_t data64;
2561

    
2562
    refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
2563
    if (refcount_table_index >= s->refcount_table_size) {
2564
        if (addend < 0)
2565
            return -EINVAL;
2566
        ret = grow_refcount_table(bs, refcount_table_index + 1);
2567
        if (ret < 0)
2568
            return ret;
2569
    }
2570
    refcount_block_offset = s->refcount_table[refcount_table_index];
2571
    if (!refcount_block_offset) {
2572
        if (addend < 0)
2573
            return -EINVAL;
2574
        /* create a new refcount block */
2575
        /* Note: we cannot update the refcount now to avoid recursion */
2576
        offset = alloc_clusters_noref(bs, s->cluster_size);
2577
        memset(s->refcount_block_cache, 0, s->cluster_size);
2578
        ret = bdrv_pwrite(s->hd, offset, s->refcount_block_cache, s->cluster_size);
2579
        if (ret != s->cluster_size)
2580
            return -EINVAL;
2581
        s->refcount_table[refcount_table_index] = offset;
2582
        data64 = cpu_to_be64(offset);
2583
        ret = bdrv_pwrite(s->hd, s->refcount_table_offset +
2584
                          refcount_table_index * sizeof(uint64_t),
2585
                          &data64, sizeof(data64));
2586
        if (ret != sizeof(data64))
2587
            return -EINVAL;
2588

    
2589
        refcount_block_offset = offset;
2590
        s->refcount_block_cache_offset = offset;
2591
        update_refcount(bs, offset, s->cluster_size, 1);
2592
    } else {
2593
        if (refcount_block_offset != s->refcount_block_cache_offset) {
2594
            if (load_refcount_block(bs, refcount_block_offset) < 0)
2595
                return -EIO;
2596
        }
2597
    }
2598
    /* we can update the count and save it */
2599
    block_index = cluster_index &
2600
        ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
2601
    refcount = be16_to_cpu(s->refcount_block_cache[block_index]);
2602
    refcount += addend;
2603
    if (refcount < 0 || refcount > 0xffff)
2604
        return -EINVAL;
2605
    if (refcount == 0 && cluster_index < s->free_cluster_index) {
2606
        s->free_cluster_index = cluster_index;
2607
    }
2608
    s->refcount_block_cache[block_index] = cpu_to_be16(refcount);
2609
    if (bdrv_pwrite(s->hd,
2610
                    refcount_block_offset + (block_index << REFCOUNT_SHIFT),
2611
                    &s->refcount_block_cache[block_index], 2) != 2)
2612
        return -EIO;
2613
    return refcount;
2614
}
2615

    
2616
static void update_refcount(BlockDriverState *bs,
2617
                            int64_t offset, int64_t length,
2618
                            int addend)
2619
{
2620
    BDRVQcowState *s = bs->opaque;
2621
    int64_t start, last, cluster_offset;
2622

    
2623
#ifdef DEBUG_ALLOC2
2624
    printf("update_refcount: offset=%lld size=%lld addend=%d\n",
2625
           offset, length, addend);
2626
#endif
2627
    if (length <= 0)
2628
        return;
2629
    start = offset & ~(s->cluster_size - 1);
2630
    last = (offset + length - 1) & ~(s->cluster_size - 1);
2631
    for(cluster_offset = start; cluster_offset <= last;
2632
        cluster_offset += s->cluster_size) {
2633
        update_cluster_refcount(bs, cluster_offset >> s->cluster_bits, addend);
2634
    }
2635
}
2636

    
2637
/*
2638
 * Increases the refcount for a range of clusters in a given refcount table.
2639
 * This is used to construct a temporary refcount table out of L1 and L2 tables
2640
 * which can be compared the the refcount table saved in the image.
2641
 *
2642
 * Returns the number of errors in the image that were found
2643
 */
2644
static int inc_refcounts(BlockDriverState *bs,
2645
                          uint16_t *refcount_table,
2646
                          int refcount_table_size,
2647
                          int64_t offset, int64_t size)
2648
{
2649
    BDRVQcowState *s = bs->opaque;
2650
    int64_t start, last, cluster_offset;
2651
    int k;
2652
    int errors = 0;
2653

    
2654
    if (size <= 0)
2655
        return 0;
2656

    
2657
    start = offset & ~(s->cluster_size - 1);
2658
    last = (offset + size - 1) & ~(s->cluster_size - 1);
2659
    for(cluster_offset = start; cluster_offset <= last;
2660
        cluster_offset += s->cluster_size) {
2661
        k = cluster_offset >> s->cluster_bits;
2662
        if (k < 0 || k >= refcount_table_size) {
2663
            fprintf(stderr, "ERROR: invalid cluster offset=0x%" PRIx64 "\n",
2664
                cluster_offset);
2665
            errors++;
2666
        } else {
2667
            if (++refcount_table[k] == 0) {
2668
                fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64
2669
                    "\n", cluster_offset);
2670
                errors++;
2671
            }
2672
        }
2673
    }
2674

    
2675
    return errors;
2676
}
2677

    
2678
/*
2679
 * Increases the refcount in the given refcount table for the all clusters
2680
 * referenced in the L2 table. While doing so, performs some checks on L2
2681
 * entries.
2682
 *
2683
 * Returns the number of errors found by the checks or -errno if an internal
2684
 * error occurred.
2685
 */
2686
static int check_refcounts_l2(BlockDriverState *bs,
2687
    uint16_t *refcount_table, int refcount_table_size, int64_t l2_offset,
2688
    int check_copied)
2689
{
2690
    BDRVQcowState *s = bs->opaque;
2691
    uint64_t *l2_table, offset;
2692
    int i, l2_size, nb_csectors, refcount;
2693
    int errors = 0;
2694

    
2695
    /* Read L2 table from disk */
2696
    l2_size = s->l2_size * sizeof(uint64_t);
2697
    l2_table = qemu_malloc(l2_size);
2698

    
2699
    if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
2700
        goto fail;
2701

    
2702
    /* Do the actual checks */
2703
    for(i = 0; i < s->l2_size; i++) {
2704
        offset = be64_to_cpu(l2_table[i]);
2705
        if (offset != 0) {
2706
            if (offset & QCOW_OFLAG_COMPRESSED) {
2707
                /* Compressed clusters don't have QCOW_OFLAG_COPIED */
2708
                if (offset & QCOW_OFLAG_COPIED) {
2709
                    fprintf(stderr, "ERROR: cluster %" PRId64 ": "
2710
                        "copied flag must never be set for compressed "
2711
                        "clusters\n", offset >> s->cluster_bits);
2712
                    offset &= ~QCOW_OFLAG_COPIED;
2713
                    errors++;
2714
                }
2715

    
2716
                /* Mark cluster as used */
2717
                nb_csectors = ((offset >> s->csize_shift) &
2718
                               s->csize_mask) + 1;
2719
                offset &= s->cluster_offset_mask;
2720
                errors += inc_refcounts(bs, refcount_table,
2721
                              refcount_table_size,
2722
                              offset & ~511, nb_csectors * 512);
2723
            } else {
2724
                /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
2725
                if (check_copied) {
2726
                    uint64_t entry = offset;
2727
                    offset &= ~QCOW_OFLAG_COPIED;
2728
                    refcount = get_refcount(bs, offset >> s->cluster_bits);
2729
                    if ((refcount == 1) != ((entry & QCOW_OFLAG_COPIED) != 0)) {
2730
                        fprintf(stderr, "ERROR OFLAG_COPIED: offset=%"
2731
                            PRIx64 " refcount=%d\n", entry, refcount);
2732
                        errors++;
2733
                    }
2734
                }
2735

    
2736
                /* Mark cluster as used */
2737
                offset &= ~QCOW_OFLAG_COPIED;
2738
                errors += inc_refcounts(bs, refcount_table,
2739
                              refcount_table_size,
2740
                              offset, s->cluster_size);
2741

    
2742
                /* Correct offsets are cluster aligned */
2743
                if (offset & (s->cluster_size - 1)) {
2744
                    fprintf(stderr, "ERROR offset=%" PRIx64 ": Cluster is not "
2745
                        "properly aligned; L2 entry corrupted.\n", offset);
2746
                    errors++;
2747
                }
2748
            }
2749
        }
2750
    }
2751

    
2752
    qemu_free(l2_table);
2753
    return errors;
2754

    
2755
fail:
2756
    fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
2757
    qemu_free(l2_table);
2758
    return -EIO;
2759
}
2760

    
2761
/*
2762
 * Increases the refcount for the L1 table, its L2 tables and all referenced
2763
 * clusters in the given refcount table. While doing so, performs some checks
2764
 * on L1 and L2 entries.
2765
 *
2766
 * Returns the number of errors found by the checks or -errno if an internal
2767
 * error occurred.
2768
 */
2769
static int check_refcounts_l1(BlockDriverState *bs,
2770
                              uint16_t *refcount_table,
2771
                              int refcount_table_size,
2772
                              int64_t l1_table_offset, int l1_size,
2773
                              int check_copied)
2774
{
2775
    BDRVQcowState *s = bs->opaque;
2776
    uint64_t *l1_table, l2_offset, l1_size2;
2777
    int i, refcount, ret;
2778
    int errors = 0;
2779

    
2780
    l1_size2 = l1_size * sizeof(uint64_t);
2781

    
2782
    /* Mark L1 table as used */
2783
    errors += inc_refcounts(bs, refcount_table, refcount_table_size,
2784
                  l1_table_offset, l1_size2);
2785

    
2786
    /* Read L1 table entries from disk */
2787
    l1_table = qemu_malloc(l1_size2);
2788
    if (bdrv_pread(s->hd, l1_table_offset,
2789
                   l1_table, l1_size2) != l1_size2)
2790
        goto fail;
2791
    for(i = 0;i < l1_size; i++)
2792
        be64_to_cpus(&l1_table[i]);
2793

    
2794
    /* Do the actual checks */
2795
    for(i = 0; i < l1_size; i++) {
2796
        l2_offset = l1_table[i];
2797
        if (l2_offset) {
2798
            /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
2799
            if (check_copied) {
2800
                refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED)
2801
                    >> s->cluster_bits);
2802
                if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) {
2803
                    fprintf(stderr, "ERROR OFLAG_COPIED: l2_offset=%" PRIx64
2804
                        " refcount=%d\n", l2_offset, refcount);
2805
                    errors++;
2806
                }
2807
            }
2808

    
2809
            /* Mark L2 table as used */
2810
            l2_offset &= ~QCOW_OFLAG_COPIED;
2811
            errors += inc_refcounts(bs, refcount_table,
2812
                          refcount_table_size,
2813
                          l2_offset,
2814
                          s->cluster_size);
2815

    
2816
            /* L2 tables are cluster aligned */
2817
            if (l2_offset & (s->cluster_size - 1)) {
2818
                fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not "
2819
                    "cluster aligned; L1 entry corrupted\n", l2_offset);
2820
                errors++;
2821
            }
2822

    
2823
            /* Process and check L2 entries */
2824
            ret = check_refcounts_l2(bs, refcount_table, refcount_table_size,
2825
                l2_offset, check_copied);
2826
            if (ret < 0) {
2827
                goto fail;
2828
            }
2829
            errors += ret;
2830
        }
2831
    }
2832
    qemu_free(l1_table);
2833
    return errors;
2834

    
2835
fail:
2836
    fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
2837
    qemu_free(l1_table);
2838
    return -EIO;
2839
}
2840

    
2841
/*
2842
 * Checks an image for refcount consistency.
2843
 *
2844
 * Returns 0 if no errors are found, the number of errors in case the image is
2845
 * detected as corrupted, and -errno when an internal error occured.
2846
 */
2847
static int check_refcounts(BlockDriverState *bs)
2848
{
2849
    BDRVQcowState *s = bs->opaque;
2850
    int64_t size;
2851
    int nb_clusters, refcount1, refcount2, i;
2852
    QCowSnapshot *sn;
2853
    uint16_t *refcount_table;
2854
    int ret, errors = 0;
2855

    
2856
    size = bdrv_getlength(s->hd);
2857
    nb_clusters = size_to_clusters(s, size);
2858
    refcount_table = qemu_mallocz(nb_clusters * sizeof(uint16_t));
2859

    
2860
    /* header */
2861
    errors += inc_refcounts(bs, refcount_table, nb_clusters,
2862
                  0, s->cluster_size);
2863

    
2864
    /* current L1 table */
2865
    ret = check_refcounts_l1(bs, refcount_table, nb_clusters,
2866
                       s->l1_table_offset, s->l1_size, 1);
2867
    if (ret < 0) {
2868
        return ret;
2869
    }
2870
    errors += ret;
2871

    
2872
    /* snapshots */
2873
    for(i = 0; i < s->nb_snapshots; i++) {
2874
        sn = s->snapshots + i;
2875
        check_refcounts_l1(bs, refcount_table, nb_clusters,
2876
                           sn->l1_table_offset, sn->l1_size, 0);
2877
    }
2878
    errors += inc_refcounts(bs, refcount_table, nb_clusters,
2879
                  s->snapshots_offset, s->snapshots_size);
2880

    
2881
    /* refcount data */
2882
    errors += inc_refcounts(bs, refcount_table, nb_clusters,
2883
                  s->refcount_table_offset,
2884
                  s->refcount_table_size * sizeof(uint64_t));
2885
    for(i = 0; i < s->refcount_table_size; i++) {
2886
        int64_t offset;
2887
        offset = s->refcount_table[i];
2888
        if (offset != 0) {
2889
            errors += inc_refcounts(bs, refcount_table, nb_clusters,
2890
                          offset, s->cluster_size);
2891
        }
2892
    }
2893

    
2894
    /* compare ref counts */
2895
    for(i = 0; i < nb_clusters; i++) {
2896
        refcount1 = get_refcount(bs, i);
2897
        refcount2 = refcount_table[i];
2898
        if (refcount1 != refcount2) {
2899
            fprintf(stderr, "ERROR cluster %d refcount=%d reference=%d\n",
2900
                   i, refcount1, refcount2);
2901
            errors++;
2902
        }
2903
    }
2904

    
2905
    qemu_free(refcount_table);
2906

    
2907
    return errors;
2908
}
2909

    
2910
static int qcow_check(BlockDriverState *bs)
2911
{
2912
    return check_refcounts(bs);
2913
}
2914

    
2915
#if 0
2916
static void dump_refcounts(BlockDriverState *bs)
2917
{
2918
    BDRVQcowState *s = bs->opaque;
2919
    int64_t nb_clusters, k, k1, size;
2920
    int refcount;
2921

2922
    size = bdrv_getlength(s->hd);
2923
    nb_clusters = size_to_clusters(s, size);
2924
    for(k = 0; k < nb_clusters;) {
2925
        k1 = k;
2926
        refcount = get_refcount(bs, k);
2927
        k++;
2928
        while (k < nb_clusters && get_refcount(bs, k) == refcount)
2929
            k++;
2930
        printf("%lld: refcount=%d nb=%lld\n", k, refcount, k - k1);
2931
    }
2932
}
2933
#endif
2934

    
2935
static int qcow_put_buffer(BlockDriverState *bs, const uint8_t *buf,
2936
                           int64_t pos, int size)
2937
{
2938
    int growable = bs->growable;
2939

    
2940
    bs->growable = 1;
2941
    bdrv_pwrite(bs, pos, buf, size);
2942
    bs->growable = growable;
2943

    
2944
    return size;
2945
}
2946

    
2947
static int qcow_get_buffer(BlockDriverState *bs, uint8_t *buf,
2948
                           int64_t pos, int size)
2949
{
2950
    int growable = bs->growable;
2951
    int ret;
2952

    
2953
    bs->growable = 1;
2954
    ret = bdrv_pread(bs, pos, buf, size);
2955
    bs->growable = growable;
2956

    
2957
    return ret;
2958
}
2959

    
2960
static QEMUOptionParameter qcow_create_options[] = {
2961
    { BLOCK_OPT_SIZE,           OPT_SIZE },
2962
    { BLOCK_OPT_BACKING_FILE,   OPT_STRING },
2963
    { BLOCK_OPT_BACKING_FMT,    OPT_STRING },
2964
    { BLOCK_OPT_ENCRYPT,        OPT_FLAG },
2965
    { BLOCK_OPT_CLUSTER_SIZE,   OPT_SIZE },
2966
    { NULL }
2967
};
2968

    
2969
static BlockDriver bdrv_qcow2 = {
2970
    .format_name        = "qcow2",
2971
    .instance_size        = sizeof(BDRVQcowState),
2972
    .bdrv_probe                = qcow_probe,
2973
    .bdrv_open                = qcow_open,
2974
    .bdrv_close                = qcow_close,
2975
    .bdrv_create        = qcow_create,
2976
    .bdrv_flush                = qcow_flush,
2977
    .bdrv_is_allocated        = qcow_is_allocated,
2978
    .bdrv_set_key        = qcow_set_key,
2979
    .bdrv_make_empty        = qcow_make_empty,
2980

    
2981
    .bdrv_aio_readv        = qcow_aio_readv,
2982
    .bdrv_aio_writev        = qcow_aio_writev,
2983
    .bdrv_aio_cancel        = qcow_aio_cancel,
2984
    .aiocb_size                = sizeof(QCowAIOCB),
2985
    .bdrv_write_compressed = qcow_write_compressed,
2986

    
2987
    .bdrv_snapshot_create = qcow_snapshot_create,
2988
    .bdrv_snapshot_goto        = qcow_snapshot_goto,
2989
    .bdrv_snapshot_delete = qcow_snapshot_delete,
2990
    .bdrv_snapshot_list        = qcow_snapshot_list,
2991
    .bdrv_get_info        = qcow_get_info,
2992

    
2993
    .bdrv_put_buffer    = qcow_put_buffer,
2994
    .bdrv_get_buffer    = qcow_get_buffer,
2995

    
2996
    .create_options = qcow_create_options,
2997
    .bdrv_check = qcow_check,
2998
};
2999

    
3000
static void bdrv_qcow2_init(void)
3001
{
3002
    bdrv_register(&bdrv_qcow2);
3003
}
3004

    
3005
block_init(bdrv_qcow2_init);