Statistics
| Branch: | Revision:

root / block-qcow2.c @ 94909d9f

History | View | Annotate | Download (88.9 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 <zlib.h>
27
#include "aes.h"
28
#include <assert.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
typedef struct QCowHeader {
66
    uint32_t magic;
67
    uint32_t version;
68
    uint64_t backing_file_offset;
69
    uint32_t backing_file_size;
70
    uint32_t cluster_bits;
71
    uint64_t size; /* in bytes */
72
    uint32_t crypt_method;
73
    uint32_t l1_size; /* XXX: save number of clusters instead ? */
74
    uint64_t l1_table_offset;
75
    uint64_t refcount_table_offset;
76
    uint32_t refcount_table_clusters;
77
    uint32_t nb_snapshots;
78
    uint64_t snapshots_offset;
79
} QCowHeader;
80

    
81

    
82
typedef struct {
83
    uint32_t magic;
84
    uint32_t len;
85
} QCowExtension;
86
#define  QCOW_EXT_MAGIC_END 0
87
#define  QCOW_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
88

    
89

    
90
typedef struct __attribute__((packed)) QCowSnapshotHeader {
91
    /* header is 8 byte aligned */
92
    uint64_t l1_table_offset;
93

    
94
    uint32_t l1_size;
95
    uint16_t id_str_size;
96
    uint16_t name_size;
97

    
98
    uint32_t date_sec;
99
    uint32_t date_nsec;
100

    
101
    uint64_t vm_clock_nsec;
102

    
103
    uint32_t vm_state_size;
104
    uint32_t extra_data_size; /* for extension */
105
    /* extra data follows */
106
    /* id_str follows */
107
    /* name follows  */
108
} QCowSnapshotHeader;
109

    
110
#define L2_CACHE_SIZE 16
111

    
112
typedef struct QCowSnapshot {
113
    uint64_t l1_table_offset;
114
    uint32_t l1_size;
115
    char *id_str;
116
    char *name;
117
    uint32_t vm_state_size;
118
    uint32_t date_sec;
119
    uint32_t date_nsec;
120
    uint64_t vm_clock_nsec;
121
} QCowSnapshot;
122

    
123
typedef struct BDRVQcowState {
124
    BlockDriverState *hd;
125
    int cluster_bits;
126
    int cluster_size;
127
    int cluster_sectors;
128
    int l2_bits;
129
    int l2_size;
130
    int l1_size;
131
    int l1_vm_state_index;
132
    int csize_shift;
133
    int csize_mask;
134
    uint64_t cluster_offset_mask;
135
    uint64_t l1_table_offset;
136
    uint64_t *l1_table;
137
    uint64_t *l2_cache;
138
    uint64_t l2_cache_offsets[L2_CACHE_SIZE];
139
    uint32_t l2_cache_counts[L2_CACHE_SIZE];
140
    uint8_t *cluster_cache;
141
    uint8_t *cluster_data;
142
    uint64_t cluster_cache_offset;
143

    
144
    uint64_t *refcount_table;
145
    uint64_t refcount_table_offset;
146
    uint32_t refcount_table_size;
147
    uint64_t refcount_block_cache_offset;
148
    uint16_t *refcount_block_cache;
149
    int64_t free_cluster_index;
150
    int64_t free_byte_offset;
151

    
152
    uint32_t crypt_method; /* current crypt method, 0 if no key yet */
153
    uint32_t crypt_method_header;
154
    AES_KEY aes_encrypt_key;
155
    AES_KEY aes_decrypt_key;
156
    uint64_t snapshots_offset;
157
    int snapshots_size;
158
    int nb_snapshots;
159
    QCowSnapshot *snapshots;
160
} BDRVQcowState;
161

    
162
static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset);
163
static int qcow_read(BlockDriverState *bs, int64_t sector_num,
164
                     uint8_t *buf, int nb_sectors);
165
static int qcow_read_snapshots(BlockDriverState *bs);
166
static void qcow_free_snapshots(BlockDriverState *bs);
167
static int refcount_init(BlockDriverState *bs);
168
static void refcount_close(BlockDriverState *bs);
169
static int get_refcount(BlockDriverState *bs, int64_t cluster_index);
170
static int update_cluster_refcount(BlockDriverState *bs,
171
                                   int64_t cluster_index,
172
                                   int addend);
173
static void update_refcount(BlockDriverState *bs,
174
                            int64_t offset, int64_t length,
175
                            int addend);
176
static int64_t alloc_clusters(BlockDriverState *bs, int64_t size);
177
static int64_t alloc_bytes(BlockDriverState *bs, int size);
178
static void free_clusters(BlockDriverState *bs,
179
                          int64_t offset, int64_t size);
180
static int check_refcounts(BlockDriverState *bs);
181

    
182
static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
183
{
184
    const QCowHeader *cow_header = (const void *)buf;
185

    
186
    if (buf_size >= sizeof(QCowHeader) &&
187
        be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
188
        be32_to_cpu(cow_header->version) == QCOW_VERSION)
189
        return 100;
190
    else
191
        return 0;
192
}
193

    
194

    
195
/* 
196
 * read qcow2 extension and fill bs
197
 * start reading from start_offset
198
 * finish reading upon magic of value 0 or when end_offset reached
199
 * unknown magic is skipped (future extension this version knows nothing about)
200
 * return 0 upon success, non-0 otherwise
201
 */
202
static int qcow_read_extensions(BlockDriverState *bs, uint64_t start_offset,
203
                                uint64_t end_offset)
204
{
205
    BDRVQcowState *s = bs->opaque;
206
    QCowExtension ext;
207
    uint64_t offset;
208

    
209
#ifdef DEBUG_EXT
210
    printf("qcow_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
211
#endif
212
    offset = start_offset;
213
    while (offset < end_offset) {
214

    
215
#ifdef DEBUG_EXT
216
        /* Sanity check */
217
        if (offset > s->cluster_size)
218
            printf("qcow_handle_extension: suspicious offset %lu\n", offset);
219

    
220
        printf("attemting to read extended header in offset %lu\n", offset);
221
#endif
222

    
223
        if (bdrv_pread(s->hd, offset, &ext, sizeof(ext)) != sizeof(ext)) {
224
            fprintf(stderr, "qcow_handle_extension: ERROR: pread fail from offset %llu\n",
225
                    (unsigned long long)offset);
226
            return 1;
227
        }
228
        be32_to_cpus(&ext.magic);
229
        be32_to_cpus(&ext.len);
230
        offset += sizeof(ext);
231
#ifdef DEBUG_EXT
232
        printf("ext.magic = 0x%x\n", ext.magic);
233
#endif
234
        switch (ext.magic) {
235
        case QCOW_EXT_MAGIC_END:
236
            return 0;
237

    
238
        case QCOW_EXT_MAGIC_BACKING_FORMAT:
239
            if (ext.len >= sizeof(bs->backing_format)) {
240
                fprintf(stderr, "ERROR: ext_backing_format: len=%u too large"
241
                        " (>=%zu)\n",
242
                        ext.len, sizeof(bs->backing_format));
243
                return 2;
244
            }
245
            if (bdrv_pread(s->hd, offset , bs->backing_format,
246
                           ext.len) != ext.len)
247
                return 3;
248
            bs->backing_format[ext.len] = '\0';
249
#ifdef DEBUG_EXT
250
            printf("Qcow2: Got format extension %s\n", bs->backing_format);
251
#endif
252
            offset += ((ext.len + 7) & ~7);
253
            break;
254

    
255
        default:
256
            /* unknown magic -- just skip it */
257
            offset += ((ext.len + 7) & ~7);
258
            break;
259
        }
260
    }
261

    
262
    return 0;
263
}
264

    
265

    
266
static int qcow_open(BlockDriverState *bs, const char *filename, int flags)
267
{
268
    BDRVQcowState *s = bs->opaque;
269
    int len, i, shift, ret;
270
    QCowHeader header;
271
    uint64_t ext_end;
272

    
273
    /* Performance is terrible right now with cache=writethrough due mainly
274
     * to reference count updates.  If the user does not explicitly specify
275
     * a caching type, force to writeback caching.
276
     */
277
    if ((flags & BDRV_O_CACHE_DEF)) {
278
        flags |= BDRV_O_CACHE_WB;
279
        flags &= ~BDRV_O_CACHE_DEF;
280
    }
281
    ret = bdrv_file_open(&s->hd, filename, flags);
282
    if (ret < 0)
283
        return ret;
284
    if (bdrv_pread(s->hd, 0, &header, sizeof(header)) != sizeof(header))
285
        goto fail;
286
    be32_to_cpus(&header.magic);
287
    be32_to_cpus(&header.version);
288
    be64_to_cpus(&header.backing_file_offset);
289
    be32_to_cpus(&header.backing_file_size);
290
    be64_to_cpus(&header.size);
291
    be32_to_cpus(&header.cluster_bits);
292
    be32_to_cpus(&header.crypt_method);
293
    be64_to_cpus(&header.l1_table_offset);
294
    be32_to_cpus(&header.l1_size);
295
    be64_to_cpus(&header.refcount_table_offset);
296
    be32_to_cpus(&header.refcount_table_clusters);
297
    be64_to_cpus(&header.snapshots_offset);
298
    be32_to_cpus(&header.nb_snapshots);
299

    
300
    if (header.magic != QCOW_MAGIC || header.version != QCOW_VERSION)
301
        goto fail;
302
    if (header.size <= 1 ||
303
        header.cluster_bits < 9 ||
304
        header.cluster_bits > 16)
305
        goto fail;
306
    if (header.crypt_method > QCOW_CRYPT_AES)
307
        goto fail;
308
    s->crypt_method_header = header.crypt_method;
309
    if (s->crypt_method_header)
310
        bs->encrypted = 1;
311
    s->cluster_bits = header.cluster_bits;
312
    s->cluster_size = 1 << s->cluster_bits;
313
    s->cluster_sectors = 1 << (s->cluster_bits - 9);
314
    s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
315
    s->l2_size = 1 << s->l2_bits;
316
    bs->total_sectors = header.size / 512;
317
    s->csize_shift = (62 - (s->cluster_bits - 8));
318
    s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
319
    s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
320
    s->refcount_table_offset = header.refcount_table_offset;
321
    s->refcount_table_size =
322
        header.refcount_table_clusters << (s->cluster_bits - 3);
323

    
324
    s->snapshots_offset = header.snapshots_offset;
325
    s->nb_snapshots = header.nb_snapshots;
326

    
327
    /* read the level 1 table */
328
    s->l1_size = header.l1_size;
329
    shift = s->cluster_bits + s->l2_bits;
330
    s->l1_vm_state_index = (header.size + (1LL << shift) - 1) >> shift;
331
    /* the L1 table must contain at least enough entries to put
332
       header.size bytes */
333
    if (s->l1_size < s->l1_vm_state_index)
334
        goto fail;
335
    s->l1_table_offset = header.l1_table_offset;
336
    s->l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
337
    if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, s->l1_size * sizeof(uint64_t)) !=
338
        s->l1_size * sizeof(uint64_t))
339
        goto fail;
340
    for(i = 0;i < s->l1_size; i++) {
341
        be64_to_cpus(&s->l1_table[i]);
342
    }
343
    /* alloc L2 cache */
344
    s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
345
    s->cluster_cache = qemu_malloc(s->cluster_size);
346
    /* one more sector for decompressed data alignment */
347
    s->cluster_data = qemu_malloc(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
348
                                  + 512);
349
    s->cluster_cache_offset = -1;
350

    
351
    if (refcount_init(bs) < 0)
352
        goto fail;
353

    
354
    /* read qcow2 extensions */
355
    if (header.backing_file_offset)
356
        ext_end = header.backing_file_offset;
357
    else
358
        ext_end = s->cluster_size;
359
    if (qcow_read_extensions(bs, sizeof(header), ext_end))
360
        goto fail;
361

    
362
    /* read the backing file name */
363
    if (header.backing_file_offset != 0) {
364
        len = header.backing_file_size;
365
        if (len > 1023)
366
            len = 1023;
367
        if (bdrv_pread(s->hd, header.backing_file_offset, bs->backing_file, len) != len)
368
            goto fail;
369
        bs->backing_file[len] = '\0';
370
    }
371
    if (qcow_read_snapshots(bs) < 0)
372
        goto fail;
373

    
374
#ifdef DEBUG_ALLOC
375
    check_refcounts(bs);
376
#endif
377
    return 0;
378

    
379
 fail:
380
    qcow_free_snapshots(bs);
381
    refcount_close(bs);
382
    qemu_free(s->l1_table);
383
    qemu_free(s->l2_cache);
384
    qemu_free(s->cluster_cache);
385
    qemu_free(s->cluster_data);
386
    bdrv_delete(s->hd);
387
    return -1;
388
}
389

    
390
static int qcow_set_key(BlockDriverState *bs, const char *key)
391
{
392
    BDRVQcowState *s = bs->opaque;
393
    uint8_t keybuf[16];
394
    int len, i;
395

    
396
    memset(keybuf, 0, 16);
397
    len = strlen(key);
398
    if (len > 16)
399
        len = 16;
400
    /* XXX: we could compress the chars to 7 bits to increase
401
       entropy */
402
    for(i = 0;i < len;i++) {
403
        keybuf[i] = key[i];
404
    }
405
    s->crypt_method = s->crypt_method_header;
406

    
407
    if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
408
        return -1;
409
    if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
410
        return -1;
411
#if 0
412
    /* test */
413
    {
414
        uint8_t in[16];
415
        uint8_t out[16];
416
        uint8_t tmp[16];
417
        for(i=0;i<16;i++)
418
            in[i] = i;
419
        AES_encrypt(in, tmp, &s->aes_encrypt_key);
420
        AES_decrypt(tmp, out, &s->aes_decrypt_key);
421
        for(i = 0; i < 16; i++)
422
            printf(" %02x", tmp[i]);
423
        printf("\n");
424
        for(i = 0; i < 16; i++)
425
            printf(" %02x", out[i]);
426
        printf("\n");
427
    }
428
#endif
429
    return 0;
430
}
431

    
432
/* The crypt function is compatible with the linux cryptoloop
433
   algorithm for < 4 GB images. NOTE: out_buf == in_buf is
434
   supported */
435
static void encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
436
                            uint8_t *out_buf, const uint8_t *in_buf,
437
                            int nb_sectors, int enc,
438
                            const AES_KEY *key)
439
{
440
    union {
441
        uint64_t ll[2];
442
        uint8_t b[16];
443
    } ivec;
444
    int i;
445

    
446
    for(i = 0; i < nb_sectors; i++) {
447
        ivec.ll[0] = cpu_to_le64(sector_num);
448
        ivec.ll[1] = 0;
449
        AES_cbc_encrypt(in_buf, out_buf, 512, key,
450
                        ivec.b, enc);
451
        sector_num++;
452
        in_buf += 512;
453
        out_buf += 512;
454
    }
455
}
456

    
457
static int copy_sectors(BlockDriverState *bs, uint64_t start_sect,
458
                        uint64_t cluster_offset, int n_start, int n_end)
459
{
460
    BDRVQcowState *s = bs->opaque;
461
    int n, ret;
462

    
463
    n = n_end - n_start;
464
    if (n <= 0)
465
        return 0;
466
    ret = qcow_read(bs, start_sect + n_start, s->cluster_data, n);
467
    if (ret < 0)
468
        return ret;
469
    if (s->crypt_method) {
470
        encrypt_sectors(s, start_sect + n_start,
471
                        s->cluster_data,
472
                        s->cluster_data, n, 1,
473
                        &s->aes_encrypt_key);
474
    }
475
    ret = bdrv_write(s->hd, (cluster_offset >> 9) + n_start,
476
                     s->cluster_data, n);
477
    if (ret < 0)
478
        return ret;
479
    return 0;
480
}
481

    
482
static void l2_cache_reset(BlockDriverState *bs)
483
{
484
    BDRVQcowState *s = bs->opaque;
485

    
486
    memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
487
    memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
488
    memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
489
}
490

    
491
static inline int l2_cache_new_entry(BlockDriverState *bs)
492
{
493
    BDRVQcowState *s = bs->opaque;
494
    uint32_t min_count;
495
    int min_index, i;
496

    
497
    /* find a new entry in the least used one */
498
    min_index = 0;
499
    min_count = 0xffffffff;
500
    for(i = 0; i < L2_CACHE_SIZE; i++) {
501
        if (s->l2_cache_counts[i] < min_count) {
502
            min_count = s->l2_cache_counts[i];
503
            min_index = i;
504
        }
505
    }
506
    return min_index;
507
}
508

    
509
static int64_t align_offset(int64_t offset, int n)
510
{
511
    offset = (offset + n - 1) & ~(n - 1);
512
    return offset;
513
}
514

    
515
static int grow_l1_table(BlockDriverState *bs, int min_size)
516
{
517
    BDRVQcowState *s = bs->opaque;
518
    int new_l1_size, new_l1_size2, ret, i;
519
    uint64_t *new_l1_table;
520
    uint64_t new_l1_table_offset;
521
    uint8_t data[12];
522

    
523
    new_l1_size = s->l1_size;
524
    if (min_size <= new_l1_size)
525
        return 0;
526
    while (min_size > new_l1_size) {
527
        new_l1_size = (new_l1_size * 3 + 1) / 2;
528
    }
529
#ifdef DEBUG_ALLOC2
530
    printf("grow l1_table from %d to %d\n", s->l1_size, new_l1_size);
531
#endif
532

    
533
    new_l1_size2 = sizeof(uint64_t) * new_l1_size;
534
    new_l1_table = qemu_mallocz(new_l1_size2);
535
    memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t));
536

    
537
    /* write new table (align to cluster) */
538
    new_l1_table_offset = alloc_clusters(bs, new_l1_size2);
539

    
540
    for(i = 0; i < s->l1_size; i++)
541
        new_l1_table[i] = cpu_to_be64(new_l1_table[i]);
542
    ret = bdrv_pwrite(s->hd, new_l1_table_offset, new_l1_table, new_l1_size2);
543
    if (ret != new_l1_size2)
544
        goto fail;
545
    for(i = 0; i < s->l1_size; i++)
546
        new_l1_table[i] = be64_to_cpu(new_l1_table[i]);
547

    
548
    /* set new table */
549
    cpu_to_be32w((uint32_t*)data, new_l1_size);
550
    cpu_to_be64w((uint64_t*)(data + 4), new_l1_table_offset);
551
    if (bdrv_pwrite(s->hd, offsetof(QCowHeader, l1_size), data,
552
                sizeof(data)) != sizeof(data))
553
        goto fail;
554
    qemu_free(s->l1_table);
555
    free_clusters(bs, s->l1_table_offset, s->l1_size * sizeof(uint64_t));
556
    s->l1_table_offset = new_l1_table_offset;
557
    s->l1_table = new_l1_table;
558
    s->l1_size = new_l1_size;
559
    return 0;
560
 fail:
561
    qemu_free(s->l1_table);
562
    return -EIO;
563
}
564

    
565
/*
566
 * seek_l2_table
567
 *
568
 * seek l2_offset in the l2_cache table
569
 * if not found, return NULL,
570
 * if found,
571
 *   increments the l2 cache hit count of the entry,
572
 *   if counter overflow, divide by two all counters
573
 *   return the pointer to the l2 cache entry
574
 *
575
 */
576

    
577
static uint64_t *seek_l2_table(BDRVQcowState *s, uint64_t l2_offset)
578
{
579
    int i, j;
580

    
581
    for(i = 0; i < L2_CACHE_SIZE; i++) {
582
        if (l2_offset == s->l2_cache_offsets[i]) {
583
            /* increment the hit count */
584
            if (++s->l2_cache_counts[i] == 0xffffffff) {
585
                for(j = 0; j < L2_CACHE_SIZE; j++) {
586
                    s->l2_cache_counts[j] >>= 1;
587
                }
588
            }
589
            return s->l2_cache + (i << s->l2_bits);
590
        }
591
    }
592
    return NULL;
593
}
594

    
595
/*
596
 * l2_load
597
 *
598
 * Loads a L2 table into memory. If the table is in the cache, the cache
599
 * is used; otherwise the L2 table is loaded from the image file.
600
 *
601
 * Returns a pointer to the L2 table on success, or NULL if the read from
602
 * the image file failed.
603
 */
604

    
605
static uint64_t *l2_load(BlockDriverState *bs, uint64_t l2_offset)
606
{
607
    BDRVQcowState *s = bs->opaque;
608
    int min_index;
609
    uint64_t *l2_table;
610

    
611
    /* seek if the table for the given offset is in the cache */
612

    
613
    l2_table = seek_l2_table(s, l2_offset);
614
    if (l2_table != NULL)
615
        return l2_table;
616

    
617
    /* not found: load a new entry in the least used one */
618

    
619
    min_index = l2_cache_new_entry(bs);
620
    l2_table = s->l2_cache + (min_index << s->l2_bits);
621
    if (bdrv_pread(s->hd, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
622
        s->l2_size * sizeof(uint64_t))
623
        return NULL;
624
    s->l2_cache_offsets[min_index] = l2_offset;
625
    s->l2_cache_counts[min_index] = 1;
626

    
627
    return l2_table;
628
}
629

    
630
/*
631
 * l2_allocate
632
 *
633
 * Allocate a new l2 entry in the file. If l1_index points to an already
634
 * used entry in the L2 table (i.e. we are doing a copy on write for the L2
635
 * table) copy the contents of the old L2 table into the newly allocated one.
636
 * Otherwise the new table is initialized with zeros.
637
 *
638
 */
639

    
640
static uint64_t *l2_allocate(BlockDriverState *bs, int l1_index)
641
{
642
    BDRVQcowState *s = bs->opaque;
643
    int min_index;
644
    uint64_t old_l2_offset, tmp;
645
    uint64_t *l2_table, l2_offset;
646

    
647
    old_l2_offset = s->l1_table[l1_index];
648

    
649
    /* allocate a new l2 entry */
650

    
651
    l2_offset = alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
652

    
653
    /* update the L1 entry */
654

    
655
    s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
656

    
657
    tmp = cpu_to_be64(l2_offset | QCOW_OFLAG_COPIED);
658
    if (bdrv_pwrite(s->hd, s->l1_table_offset + l1_index * sizeof(tmp),
659
                    &tmp, sizeof(tmp)) != sizeof(tmp))
660
        return NULL;
661

    
662
    /* allocate a new entry in the l2 cache */
663

    
664
    min_index = l2_cache_new_entry(bs);
665
    l2_table = s->l2_cache + (min_index << s->l2_bits);
666

    
667
    if (old_l2_offset == 0) {
668
        /* if there was no old l2 table, clear the new table */
669
        memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
670
    } else {
671
        /* if there was an old l2 table, read it from the disk */
672
        if (bdrv_pread(s->hd, old_l2_offset,
673
                       l2_table, s->l2_size * sizeof(uint64_t)) !=
674
            s->l2_size * sizeof(uint64_t))
675
            return NULL;
676
    }
677
    /* write the l2 table to the file */
678
    if (bdrv_pwrite(s->hd, l2_offset,
679
                    l2_table, s->l2_size * sizeof(uint64_t)) !=
680
        s->l2_size * sizeof(uint64_t))
681
        return NULL;
682

    
683
    /* update the l2 cache entry */
684

    
685
    s->l2_cache_offsets[min_index] = l2_offset;
686
    s->l2_cache_counts[min_index] = 1;
687

    
688
    return l2_table;
689
}
690

    
691
static int size_to_clusters(BDRVQcowState *s, int64_t size)
692
{
693
    return (size + (s->cluster_size - 1)) >> s->cluster_bits;
694
}
695

    
696
static int count_contiguous_clusters(uint64_t nb_clusters, int cluster_size,
697
        uint64_t *l2_table, uint64_t start, uint64_t mask)
698
{
699
    int i;
700
    uint64_t offset = be64_to_cpu(l2_table[0]) & ~mask;
701

    
702
    if (!offset)
703
        return 0;
704

    
705
    for (i = start; i < start + nb_clusters; i++)
706
        if (offset + i * cluster_size != (be64_to_cpu(l2_table[i]) & ~mask))
707
            break;
708

    
709
        return (i - start);
710
}
711

    
712
static int count_contiguous_free_clusters(uint64_t nb_clusters, uint64_t *l2_table)
713
{
714
    int i = 0;
715

    
716
    while(nb_clusters-- && l2_table[i] == 0)
717
        i++;
718

    
719
    return i;
720
}
721

    
722
/*
723
 * get_cluster_offset
724
 *
725
 * For a given offset of the disk image, return cluster offset in
726
 * qcow2 file.
727
 *
728
 * on entry, *num is the number of contiguous clusters we'd like to
729
 * access following offset.
730
 *
731
 * on exit, *num is the number of contiguous clusters we can read.
732
 *
733
 * Return 1, if the offset is found
734
 * Return 0, otherwise.
735
 *
736
 */
737

    
738
static uint64_t get_cluster_offset(BlockDriverState *bs,
739
                                   uint64_t offset, int *num)
740
{
741
    BDRVQcowState *s = bs->opaque;
742
    int l1_index, l2_index;
743
    uint64_t l2_offset, *l2_table, cluster_offset;
744
    int l1_bits, c;
745
    int index_in_cluster, nb_available, nb_needed, nb_clusters;
746

    
747
    index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1);
748
    nb_needed = *num + index_in_cluster;
749

    
750
    l1_bits = s->l2_bits + s->cluster_bits;
751

    
752
    /* compute how many bytes there are between the offset and
753
     * the end of the l1 entry
754
     */
755

    
756
    nb_available = (1 << l1_bits) - (offset & ((1 << l1_bits) - 1));
757

    
758
    /* compute the number of available sectors */
759

    
760
    nb_available = (nb_available >> 9) + index_in_cluster;
761

    
762
    if (nb_needed > nb_available) {
763
        nb_needed = nb_available;
764
    }
765

    
766
    cluster_offset = 0;
767

    
768
    /* seek the the l2 offset in the l1 table */
769

    
770
    l1_index = offset >> l1_bits;
771
    if (l1_index >= s->l1_size)
772
        goto out;
773

    
774
    l2_offset = s->l1_table[l1_index];
775

    
776
    /* seek the l2 table of the given l2 offset */
777

    
778
    if (!l2_offset)
779
        goto out;
780

    
781
    /* load the l2 table in memory */
782

    
783
    l2_offset &= ~QCOW_OFLAG_COPIED;
784
    l2_table = l2_load(bs, l2_offset);
785
    if (l2_table == NULL)
786
        return 0;
787

    
788
    /* find the cluster offset for the given disk offset */
789

    
790
    l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
791
    cluster_offset = be64_to_cpu(l2_table[l2_index]);
792
    nb_clusters = size_to_clusters(s, nb_needed << 9);
793

    
794
    if (!cluster_offset) {
795
        /* how many empty clusters ? */
796
        c = count_contiguous_free_clusters(nb_clusters, &l2_table[l2_index]);
797
    } else {
798
        /* how many allocated clusters ? */
799
        c = count_contiguous_clusters(nb_clusters, s->cluster_size,
800
                &l2_table[l2_index], 0, QCOW_OFLAG_COPIED);
801
    }
802

    
803
   nb_available = (c * s->cluster_sectors);
804
out:
805
    if (nb_available > nb_needed)
806
        nb_available = nb_needed;
807

    
808
    *num = nb_available - index_in_cluster;
809

    
810
    return cluster_offset & ~QCOW_OFLAG_COPIED;
811
}
812

    
813
/*
814
 * free_any_clusters
815
 *
816
 * free clusters according to its type: compressed or not
817
 *
818
 */
819

    
820
static void free_any_clusters(BlockDriverState *bs,
821
                              uint64_t cluster_offset, int nb_clusters)
822
{
823
    BDRVQcowState *s = bs->opaque;
824

    
825
    /* free the cluster */
826

    
827
    if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
828
        int nb_csectors;
829
        nb_csectors = ((cluster_offset >> s->csize_shift) &
830
                       s->csize_mask) + 1;
831
        free_clusters(bs, (cluster_offset & s->cluster_offset_mask) & ~511,
832
                      nb_csectors * 512);
833
        return;
834
    }
835

    
836
    free_clusters(bs, cluster_offset, nb_clusters << s->cluster_bits);
837

    
838
    return;
839
}
840

    
841
/*
842
 * get_cluster_table
843
 *
844
 * for a given disk offset, load (and allocate if needed)
845
 * the l2 table.
846
 *
847
 * the l2 table offset in the qcow2 file and the cluster index
848
 * in the l2 table are given to the caller.
849
 *
850
 */
851

    
852
static int get_cluster_table(BlockDriverState *bs, uint64_t offset,
853
                             uint64_t **new_l2_table,
854
                             uint64_t *new_l2_offset,
855
                             int *new_l2_index)
856
{
857
    BDRVQcowState *s = bs->opaque;
858
    int l1_index, l2_index, ret;
859
    uint64_t l2_offset, *l2_table;
860

    
861
    /* seek the the l2 offset in the l1 table */
862

    
863
    l1_index = offset >> (s->l2_bits + s->cluster_bits);
864
    if (l1_index >= s->l1_size) {
865
        ret = grow_l1_table(bs, l1_index + 1);
866
        if (ret < 0)
867
            return 0;
868
    }
869
    l2_offset = s->l1_table[l1_index];
870

    
871
    /* seek the l2 table of the given l2 offset */
872

    
873
    if (l2_offset & QCOW_OFLAG_COPIED) {
874
        /* load the l2 table in memory */
875
        l2_offset &= ~QCOW_OFLAG_COPIED;
876
        l2_table = l2_load(bs, l2_offset);
877
        if (l2_table == NULL)
878
            return 0;
879
    } else {
880
        if (l2_offset)
881
            free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t));
882
        l2_table = l2_allocate(bs, l1_index);
883
        if (l2_table == NULL)
884
            return 0;
885
        l2_offset = s->l1_table[l1_index] & ~QCOW_OFLAG_COPIED;
886
    }
887

    
888
    /* find the cluster offset for the given disk offset */
889

    
890
    l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
891

    
892
    *new_l2_table = l2_table;
893
    *new_l2_offset = l2_offset;
894
    *new_l2_index = l2_index;
895

    
896
    return 1;
897
}
898

    
899
/*
900
 * alloc_compressed_cluster_offset
901
 *
902
 * For a given offset of the disk image, return cluster offset in
903
 * qcow2 file.
904
 *
905
 * If the offset is not found, allocate a new compressed cluster.
906
 *
907
 * Return the cluster offset if successful,
908
 * Return 0, otherwise.
909
 *
910
 */
911

    
912
static uint64_t alloc_compressed_cluster_offset(BlockDriverState *bs,
913
                                                uint64_t offset,
914
                                                int compressed_size)
915
{
916
    BDRVQcowState *s = bs->opaque;
917
    int l2_index, ret;
918
    uint64_t l2_offset, *l2_table, cluster_offset;
919
    int nb_csectors;
920

    
921
    ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
922
    if (ret == 0)
923
        return 0;
924

    
925
    cluster_offset = be64_to_cpu(l2_table[l2_index]);
926
    if (cluster_offset & QCOW_OFLAG_COPIED)
927
        return cluster_offset & ~QCOW_OFLAG_COPIED;
928

    
929
    if (cluster_offset)
930
        free_any_clusters(bs, cluster_offset, 1);
931

    
932
    cluster_offset = alloc_bytes(bs, compressed_size);
933
    nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) -
934
                  (cluster_offset >> 9);
935

    
936
    cluster_offset |= QCOW_OFLAG_COMPRESSED |
937
                      ((uint64_t)nb_csectors << s->csize_shift);
938

    
939
    /* update L2 table */
940

    
941
    /* compressed clusters never have the copied flag */
942

    
943
    l2_table[l2_index] = cpu_to_be64(cluster_offset);
944
    if (bdrv_pwrite(s->hd,
945
                    l2_offset + l2_index * sizeof(uint64_t),
946
                    l2_table + l2_index,
947
                    sizeof(uint64_t)) != sizeof(uint64_t))
948
        return 0;
949

    
950
    return cluster_offset;
951
}
952

    
953
typedef struct QCowL2Meta
954
{
955
    uint64_t offset;
956
    int n_start;
957
    int nb_available;
958
    int nb_clusters;
959
} QCowL2Meta;
960

    
961
static int alloc_cluster_link_l2(BlockDriverState *bs, uint64_t cluster_offset,
962
        QCowL2Meta *m)
963
{
964
    BDRVQcowState *s = bs->opaque;
965
    int i, j = 0, l2_index, ret;
966
    uint64_t *old_cluster, start_sect, l2_offset, *l2_table;
967

    
968
    if (m->nb_clusters == 0)
969
        return 0;
970

    
971
    old_cluster = qemu_malloc(m->nb_clusters * sizeof(uint64_t));
972

    
973
    /* copy content of unmodified sectors */
974
    start_sect = (m->offset & ~(s->cluster_size - 1)) >> 9;
975
    if (m->n_start) {
976
        ret = copy_sectors(bs, start_sect, cluster_offset, 0, m->n_start);
977
        if (ret < 0)
978
            goto err;
979
    }
980

    
981
    if (m->nb_available & (s->cluster_sectors - 1)) {
982
        uint64_t end = m->nb_available & ~(uint64_t)(s->cluster_sectors - 1);
983
        ret = copy_sectors(bs, start_sect + end, cluster_offset + (end << 9),
984
                m->nb_available - end, s->cluster_sectors);
985
        if (ret < 0)
986
            goto err;
987
    }
988

    
989
    ret = -EIO;
990
    /* update L2 table */
991
    if (!get_cluster_table(bs, m->offset, &l2_table, &l2_offset, &l2_index))
992
        goto err;
993

    
994
    for (i = 0; i < m->nb_clusters; i++) {
995
        if(l2_table[l2_index + i] != 0)
996
            old_cluster[j++] = l2_table[l2_index + i];
997

    
998
        l2_table[l2_index + i] = cpu_to_be64((cluster_offset +
999
                    (i << s->cluster_bits)) | QCOW_OFLAG_COPIED);
1000
     }
1001

    
1002
    if (bdrv_pwrite(s->hd, l2_offset + l2_index * sizeof(uint64_t),
1003
                l2_table + l2_index, m->nb_clusters * sizeof(uint64_t)) !=
1004
            m->nb_clusters * sizeof(uint64_t))
1005
        goto err;
1006

    
1007
    for (i = 0; i < j; i++)
1008
        free_any_clusters(bs, be64_to_cpu(old_cluster[i]), 1);
1009

    
1010
    ret = 0;
1011
err:
1012
    qemu_free(old_cluster);
1013
    return ret;
1014
 }
1015

    
1016
/*
1017
 * alloc_cluster_offset
1018
 *
1019
 * For a given offset of the disk image, return cluster offset in
1020
 * qcow2 file.
1021
 *
1022
 * If the offset is not found, allocate a new cluster.
1023
 *
1024
 * Return the cluster offset if successful,
1025
 * Return 0, otherwise.
1026
 *
1027
 */
1028

    
1029
static uint64_t alloc_cluster_offset(BlockDriverState *bs,
1030
                                     uint64_t offset,
1031
                                     int n_start, int n_end,
1032
                                     int *num, QCowL2Meta *m)
1033
{
1034
    BDRVQcowState *s = bs->opaque;
1035
    int l2_index, ret;
1036
    uint64_t l2_offset, *l2_table, cluster_offset;
1037
    int nb_clusters, i = 0;
1038

    
1039
    ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
1040
    if (ret == 0)
1041
        return 0;
1042

    
1043
    nb_clusters = size_to_clusters(s, n_end << 9);
1044

    
1045
    nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
1046

    
1047
    cluster_offset = be64_to_cpu(l2_table[l2_index]);
1048

    
1049
    /* We keep all QCOW_OFLAG_COPIED clusters */
1050

    
1051
    if (cluster_offset & QCOW_OFLAG_COPIED) {
1052
        nb_clusters = count_contiguous_clusters(nb_clusters, s->cluster_size,
1053
                &l2_table[l2_index], 0, 0);
1054

    
1055
        cluster_offset &= ~QCOW_OFLAG_COPIED;
1056
        m->nb_clusters = 0;
1057

    
1058
        goto out;
1059
    }
1060

    
1061
    /* for the moment, multiple compressed clusters are not managed */
1062

    
1063
    if (cluster_offset & QCOW_OFLAG_COMPRESSED)
1064
        nb_clusters = 1;
1065

    
1066
    /* how many available clusters ? */
1067

    
1068
    while (i < nb_clusters) {
1069
        i += count_contiguous_clusters(nb_clusters - i, s->cluster_size,
1070
                &l2_table[l2_index], i, 0);
1071

    
1072
        if(be64_to_cpu(l2_table[l2_index + i]))
1073
            break;
1074

    
1075
        i += count_contiguous_free_clusters(nb_clusters - i,
1076
                &l2_table[l2_index + i]);
1077

    
1078
        cluster_offset = be64_to_cpu(l2_table[l2_index + i]);
1079

    
1080
        if ((cluster_offset & QCOW_OFLAG_COPIED) ||
1081
                (cluster_offset & QCOW_OFLAG_COMPRESSED))
1082
            break;
1083
    }
1084
    nb_clusters = i;
1085

    
1086
    /* allocate a new cluster */
1087

    
1088
    cluster_offset = alloc_clusters(bs, nb_clusters * s->cluster_size);
1089

    
1090
    /* save info needed for meta data update */
1091
    m->offset = offset;
1092
    m->n_start = n_start;
1093
    m->nb_clusters = nb_clusters;
1094

    
1095
out:
1096
    m->nb_available = MIN(nb_clusters << (s->cluster_bits - 9), n_end);
1097

    
1098
    *num = m->nb_available - n_start;
1099

    
1100
    return cluster_offset;
1101
}
1102

    
1103
static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
1104
                             int nb_sectors, int *pnum)
1105
{
1106
    uint64_t cluster_offset;
1107

    
1108
    *pnum = nb_sectors;
1109
    cluster_offset = get_cluster_offset(bs, sector_num << 9, pnum);
1110

    
1111
    return (cluster_offset != 0);
1112
}
1113

    
1114
static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
1115
                             const uint8_t *buf, int buf_size)
1116
{
1117
    z_stream strm1, *strm = &strm1;
1118
    int ret, out_len;
1119

    
1120
    memset(strm, 0, sizeof(*strm));
1121

    
1122
    strm->next_in = (uint8_t *)buf;
1123
    strm->avail_in = buf_size;
1124
    strm->next_out = out_buf;
1125
    strm->avail_out = out_buf_size;
1126

    
1127
    ret = inflateInit2(strm, -12);
1128
    if (ret != Z_OK)
1129
        return -1;
1130
    ret = inflate(strm, Z_FINISH);
1131
    out_len = strm->next_out - out_buf;
1132
    if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
1133
        out_len != out_buf_size) {
1134
        inflateEnd(strm);
1135
        return -1;
1136
    }
1137
    inflateEnd(strm);
1138
    return 0;
1139
}
1140

    
1141
static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset)
1142
{
1143
    int ret, csize, nb_csectors, sector_offset;
1144
    uint64_t coffset;
1145

    
1146
    coffset = cluster_offset & s->cluster_offset_mask;
1147
    if (s->cluster_cache_offset != coffset) {
1148
        nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
1149
        sector_offset = coffset & 511;
1150
        csize = nb_csectors * 512 - sector_offset;
1151
        ret = bdrv_read(s->hd, coffset >> 9, s->cluster_data, nb_csectors);
1152
        if (ret < 0) {
1153
            return -1;
1154
        }
1155
        if (decompress_buffer(s->cluster_cache, s->cluster_size,
1156
                              s->cluster_data + sector_offset, csize) < 0) {
1157
            return -1;
1158
        }
1159
        s->cluster_cache_offset = coffset;
1160
    }
1161
    return 0;
1162
}
1163

    
1164
/* handle reading after the end of the backing file */
1165
static int backing_read1(BlockDriverState *bs,
1166
                         int64_t sector_num, uint8_t *buf, int nb_sectors)
1167
{
1168
    int n1;
1169
    if ((sector_num + nb_sectors) <= bs->total_sectors)
1170
        return nb_sectors;
1171
    if (sector_num >= bs->total_sectors)
1172
        n1 = 0;
1173
    else
1174
        n1 = bs->total_sectors - sector_num;
1175
    memset(buf + n1 * 512, 0, 512 * (nb_sectors - n1));
1176
    return n1;
1177
}
1178

    
1179
static int qcow_read(BlockDriverState *bs, int64_t sector_num,
1180
                     uint8_t *buf, int nb_sectors)
1181
{
1182
    BDRVQcowState *s = bs->opaque;
1183
    int ret, index_in_cluster, n, n1;
1184
    uint64_t cluster_offset;
1185

    
1186
    while (nb_sectors > 0) {
1187
        n = nb_sectors;
1188
        cluster_offset = get_cluster_offset(bs, sector_num << 9, &n);
1189
        index_in_cluster = sector_num & (s->cluster_sectors - 1);
1190
        if (!cluster_offset) {
1191
            if (bs->backing_hd) {
1192
                /* read from the base image */
1193
                n1 = backing_read1(bs->backing_hd, sector_num, buf, n);
1194
                if (n1 > 0) {
1195
                    ret = bdrv_read(bs->backing_hd, sector_num, buf, n1);
1196
                    if (ret < 0)
1197
                        return -1;
1198
                }
1199
            } else {
1200
                memset(buf, 0, 512 * n);
1201
            }
1202
        } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
1203
            if (decompress_cluster(s, cluster_offset) < 0)
1204
                return -1;
1205
            memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n);
1206
        } else {
1207
            ret = bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
1208
            if (ret != n * 512)
1209
                return -1;
1210
            if (s->crypt_method) {
1211
                encrypt_sectors(s, sector_num, buf, buf, n, 0,
1212
                                &s->aes_decrypt_key);
1213
            }
1214
        }
1215
        nb_sectors -= n;
1216
        sector_num += n;
1217
        buf += n * 512;
1218
    }
1219
    return 0;
1220
}
1221

    
1222
static int qcow_write(BlockDriverState *bs, int64_t sector_num,
1223
                     const uint8_t *buf, int nb_sectors)
1224
{
1225
    BDRVQcowState *s = bs->opaque;
1226
    int ret, index_in_cluster, n;
1227
    uint64_t cluster_offset;
1228
    int n_end;
1229
    QCowL2Meta l2meta;
1230

    
1231
    while (nb_sectors > 0) {
1232
        index_in_cluster = sector_num & (s->cluster_sectors - 1);
1233
        n_end = index_in_cluster + nb_sectors;
1234
        if (s->crypt_method &&
1235
            n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
1236
            n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
1237
        cluster_offset = alloc_cluster_offset(bs, sector_num << 9,
1238
                                              index_in_cluster,
1239
                                              n_end, &n, &l2meta);
1240
        if (!cluster_offset)
1241
            return -1;
1242
        if (s->crypt_method) {
1243
            encrypt_sectors(s, sector_num, s->cluster_data, buf, n, 1,
1244
                            &s->aes_encrypt_key);
1245
            ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512,
1246
                              s->cluster_data, n * 512);
1247
        } else {
1248
            ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
1249
        }
1250
        if (ret != n * 512 || alloc_cluster_link_l2(bs, cluster_offset, &l2meta) < 0) {
1251
            free_any_clusters(bs, cluster_offset, l2meta.nb_clusters);
1252
            return -1;
1253
        }
1254
        nb_sectors -= n;
1255
        sector_num += n;
1256
        buf += n * 512;
1257
    }
1258
    s->cluster_cache_offset = -1; /* disable compressed cache */
1259
    return 0;
1260
}
1261

    
1262
typedef struct QCowAIOCB {
1263
    BlockDriverAIOCB common;
1264
    int64_t sector_num;
1265
    QEMUIOVector *qiov;
1266
    uint8_t *buf;
1267
    void *orig_buf;
1268
    int nb_sectors;
1269
    int n;
1270
    uint64_t cluster_offset;
1271
    uint8_t *cluster_data;
1272
    BlockDriverAIOCB *hd_aiocb;
1273
    struct iovec hd_iov;
1274
    QEMUIOVector hd_qiov;
1275
    QEMUBH *bh;
1276
    QCowL2Meta l2meta;
1277
} QCowAIOCB;
1278

    
1279
static void qcow_aio_read_cb(void *opaque, int ret);
1280
static void qcow_aio_read_bh(void *opaque)
1281
{
1282
    QCowAIOCB *acb = opaque;
1283
    qemu_bh_delete(acb->bh);
1284
    acb->bh = NULL;
1285
    qcow_aio_read_cb(opaque, 0);
1286
}
1287

    
1288
static int qcow_schedule_bh(QEMUBHFunc *cb, QCowAIOCB *acb)
1289
{
1290
    if (acb->bh)
1291
        return -EIO;
1292

    
1293
    acb->bh = qemu_bh_new(cb, acb);
1294
    if (!acb->bh)
1295
        return -EIO;
1296

    
1297
    qemu_bh_schedule(acb->bh);
1298

    
1299
    return 0;
1300
}
1301

    
1302
static void qcow_aio_read_cb(void *opaque, int ret)
1303
{
1304
    QCowAIOCB *acb = opaque;
1305
    BlockDriverState *bs = acb->common.bs;
1306
    BDRVQcowState *s = bs->opaque;
1307
    int index_in_cluster, n1;
1308

    
1309
    acb->hd_aiocb = NULL;
1310
    if (ret < 0)
1311
        goto done;
1312

    
1313
    /* post process the read buffer */
1314
    if (!acb->cluster_offset) {
1315
        /* nothing to do */
1316
    } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
1317
        /* nothing to do */
1318
    } else {
1319
        if (s->crypt_method) {
1320
            encrypt_sectors(s, acb->sector_num, acb->buf, acb->buf,
1321
                            acb->n, 0,
1322
                            &s->aes_decrypt_key);
1323
        }
1324
    }
1325

    
1326
    acb->nb_sectors -= acb->n;
1327
    acb->sector_num += acb->n;
1328
    acb->buf += acb->n * 512;
1329

    
1330
    if (acb->nb_sectors == 0) {
1331
        /* request completed */
1332
        ret = 0;
1333
        goto done;
1334
    }
1335

    
1336
    /* prepare next AIO request */
1337
    acb->n = acb->nb_sectors;
1338
    acb->cluster_offset = get_cluster_offset(bs, acb->sector_num << 9, &acb->n);
1339
    index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
1340

    
1341
    if (!acb->cluster_offset) {
1342
        if (bs->backing_hd) {
1343
            /* read from the base image */
1344
            n1 = backing_read1(bs->backing_hd, acb->sector_num,
1345
                               acb->buf, acb->n);
1346
            if (n1 > 0) {
1347
                acb->hd_iov.iov_base = (void *)acb->buf;
1348
                acb->hd_iov.iov_len = acb->n * 512;
1349
                qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
1350
                acb->hd_aiocb = bdrv_aio_readv(bs->backing_hd, acb->sector_num,
1351
                                    &acb->hd_qiov, acb->n,
1352
                                    qcow_aio_read_cb, acb);
1353
                if (acb->hd_aiocb == NULL)
1354
                    goto done;
1355
            } else {
1356
                ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
1357
                if (ret < 0)
1358
                    goto done;
1359
            }
1360
        } else {
1361
            /* Note: in this case, no need to wait */
1362
            memset(acb->buf, 0, 512 * acb->n);
1363
            ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
1364
            if (ret < 0)
1365
                goto done;
1366
        }
1367
    } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
1368
        /* add AIO support for compressed blocks ? */
1369
        if (decompress_cluster(s, acb->cluster_offset) < 0)
1370
            goto done;
1371
        memcpy(acb->buf,
1372
               s->cluster_cache + index_in_cluster * 512, 512 * acb->n);
1373
        ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
1374
        if (ret < 0)
1375
            goto done;
1376
    } else {
1377
        if ((acb->cluster_offset & 511) != 0) {
1378
            ret = -EIO;
1379
            goto done;
1380
        }
1381

    
1382
        acb->hd_iov.iov_base = (void *)acb->buf;
1383
        acb->hd_iov.iov_len = acb->n * 512;
1384
        qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
1385
        acb->hd_aiocb = bdrv_aio_readv(s->hd,
1386
                            (acb->cluster_offset >> 9) + index_in_cluster,
1387
                            &acb->hd_qiov, acb->n, qcow_aio_read_cb, acb);
1388
        if (acb->hd_aiocb == NULL)
1389
            goto done;
1390
    }
1391

    
1392
    return;
1393
done:
1394
    if (acb->qiov->niov > 1) {
1395
        qemu_iovec_from_buffer(acb->qiov, acb->orig_buf, acb->qiov->size);
1396
        qemu_vfree(acb->orig_buf);
1397
    }
1398
    acb->common.cb(acb->common.opaque, ret);
1399
    qemu_aio_release(acb);
1400
}
1401

    
1402
static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs,
1403
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1404
        BlockDriverCompletionFunc *cb, void *opaque, int is_write)
1405
{
1406
    QCowAIOCB *acb;
1407

    
1408
    acb = qemu_aio_get(bs, cb, opaque);
1409
    if (!acb)
1410
        return NULL;
1411
    acb->hd_aiocb = NULL;
1412
    acb->sector_num = sector_num;
1413
    acb->qiov = qiov;
1414
    if (qiov->niov > 1) {
1415
        acb->buf = acb->orig_buf = qemu_memalign(512, qiov->size);
1416
        if (is_write)
1417
            qemu_iovec_to_buffer(qiov, acb->buf);
1418
    } else {
1419
        acb->buf = (uint8_t *)qiov->iov->iov_base;
1420
    }
1421
    acb->nb_sectors = nb_sectors;
1422
    acb->n = 0;
1423
    acb->cluster_offset = 0;
1424
    acb->l2meta.nb_clusters = 0;
1425
    return acb;
1426
}
1427

    
1428
static BlockDriverAIOCB *qcow_aio_readv(BlockDriverState *bs,
1429
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1430
        BlockDriverCompletionFunc *cb, void *opaque)
1431
{
1432
    QCowAIOCB *acb;
1433

    
1434
    acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
1435
    if (!acb)
1436
        return NULL;
1437

    
1438
    qcow_aio_read_cb(acb, 0);
1439
    return &acb->common;
1440
}
1441

    
1442
static void qcow_aio_write_cb(void *opaque, int ret)
1443
{
1444
    QCowAIOCB *acb = opaque;
1445
    BlockDriverState *bs = acb->common.bs;
1446
    BDRVQcowState *s = bs->opaque;
1447
    int index_in_cluster;
1448
    const uint8_t *src_buf;
1449
    int n_end;
1450

    
1451
    acb->hd_aiocb = NULL;
1452

    
1453
    if (ret < 0)
1454
        goto done;
1455

    
1456
    if (alloc_cluster_link_l2(bs, acb->cluster_offset, &acb->l2meta) < 0) {
1457
        free_any_clusters(bs, acb->cluster_offset, acb->l2meta.nb_clusters);
1458
        goto done;
1459
    }
1460

    
1461
    acb->nb_sectors -= acb->n;
1462
    acb->sector_num += acb->n;
1463
    acb->buf += acb->n * 512;
1464

    
1465
    if (acb->nb_sectors == 0) {
1466
        /* request completed */
1467
        ret = 0;
1468
        goto done;
1469
    }
1470

    
1471
    index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
1472
    n_end = index_in_cluster + acb->nb_sectors;
1473
    if (s->crypt_method &&
1474
        n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
1475
        n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
1476

    
1477
    acb->cluster_offset = alloc_cluster_offset(bs, acb->sector_num << 9,
1478
                                          index_in_cluster,
1479
                                          n_end, &acb->n, &acb->l2meta);
1480
    if (!acb->cluster_offset || (acb->cluster_offset & 511) != 0) {
1481
        ret = -EIO;
1482
        goto done;
1483
    }
1484
    if (s->crypt_method) {
1485
        if (!acb->cluster_data) {
1486
            acb->cluster_data = qemu_mallocz(QCOW_MAX_CRYPT_CLUSTERS *
1487
                                             s->cluster_size);
1488
        }
1489
        encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf,
1490
                        acb->n, 1, &s->aes_encrypt_key);
1491
        src_buf = acb->cluster_data;
1492
    } else {
1493
        src_buf = acb->buf;
1494
    }
1495
    acb->hd_iov.iov_base = (void *)src_buf;
1496
    acb->hd_iov.iov_len = acb->n * 512;
1497
    qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
1498
    acb->hd_aiocb = bdrv_aio_writev(s->hd,
1499
                                    (acb->cluster_offset >> 9) + index_in_cluster,
1500
                                    &acb->hd_qiov, acb->n,
1501
                                    qcow_aio_write_cb, acb);
1502
    if (acb->hd_aiocb == NULL)
1503
        goto done;
1504

    
1505
    return;
1506

    
1507
done:
1508
    if (acb->qiov->niov > 1)
1509
        qemu_vfree(acb->orig_buf);
1510
    acb->common.cb(acb->common.opaque, ret);
1511
    qemu_aio_release(acb);
1512
}
1513

    
1514
static BlockDriverAIOCB *qcow_aio_writev(BlockDriverState *bs,
1515
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1516
        BlockDriverCompletionFunc *cb, void *opaque)
1517
{
1518
    BDRVQcowState *s = bs->opaque;
1519
    QCowAIOCB *acb;
1520

    
1521
    s->cluster_cache_offset = -1; /* disable compressed cache */
1522

    
1523
    acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
1524
    if (!acb)
1525
        return NULL;
1526

    
1527
    qcow_aio_write_cb(acb, 0);
1528
    return &acb->common;
1529
}
1530

    
1531
static void qcow_aio_cancel(BlockDriverAIOCB *blockacb)
1532
{
1533
    QCowAIOCB *acb = (QCowAIOCB *)blockacb;
1534
    if (acb->hd_aiocb)
1535
        bdrv_aio_cancel(acb->hd_aiocb);
1536
    qemu_aio_release(acb);
1537
}
1538

    
1539
static void qcow_close(BlockDriverState *bs)
1540
{
1541
    BDRVQcowState *s = bs->opaque;
1542
    qemu_free(s->l1_table);
1543
    qemu_free(s->l2_cache);
1544
    qemu_free(s->cluster_cache);
1545
    qemu_free(s->cluster_data);
1546
    refcount_close(bs);
1547
    bdrv_delete(s->hd);
1548
}
1549

    
1550
/* XXX: use std qcow open function ? */
1551
typedef struct QCowCreateState {
1552
    int cluster_size;
1553
    int cluster_bits;
1554
    uint16_t *refcount_block;
1555
    uint64_t *refcount_table;
1556
    int64_t l1_table_offset;
1557
    int64_t refcount_table_offset;
1558
    int64_t refcount_block_offset;
1559
} QCowCreateState;
1560

    
1561
static void create_refcount_update(QCowCreateState *s,
1562
                                   int64_t offset, int64_t size)
1563
{
1564
    int refcount;
1565
    int64_t start, last, cluster_offset;
1566
    uint16_t *p;
1567

    
1568
    start = offset & ~(s->cluster_size - 1);
1569
    last = (offset + size - 1)  & ~(s->cluster_size - 1);
1570
    for(cluster_offset = start; cluster_offset <= last;
1571
        cluster_offset += s->cluster_size) {
1572
        p = &s->refcount_block[cluster_offset >> s->cluster_bits];
1573
        refcount = be16_to_cpu(*p);
1574
        refcount++;
1575
        *p = cpu_to_be16(refcount);
1576
    }
1577
}
1578

    
1579
static int qcow_create2(const char *filename, int64_t total_size,
1580
                        const char *backing_file, const char *backing_format,
1581
                        int flags)
1582
{
1583

    
1584
    int fd, header_size, backing_filename_len, l1_size, i, shift, l2_bits;
1585
    int ref_clusters, backing_format_len = 0;
1586
    QCowHeader header;
1587
    uint64_t tmp, offset;
1588
    QCowCreateState s1, *s = &s1;
1589
    QCowExtension ext_bf = {0, 0};
1590

    
1591

    
1592
    memset(s, 0, sizeof(*s));
1593

    
1594
    fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
1595
    if (fd < 0)
1596
        return -1;
1597
    memset(&header, 0, sizeof(header));
1598
    header.magic = cpu_to_be32(QCOW_MAGIC);
1599
    header.version = cpu_to_be32(QCOW_VERSION);
1600
    header.size = cpu_to_be64(total_size * 512);
1601
    header_size = sizeof(header);
1602
    backing_filename_len = 0;
1603
    if (backing_file) {
1604
        if (backing_format) {
1605
            ext_bf.magic = QCOW_EXT_MAGIC_BACKING_FORMAT;
1606
            backing_format_len = strlen(backing_format);
1607
            ext_bf.len = (backing_format_len + 7) & ~7;
1608
            header_size += ((sizeof(ext_bf) + ext_bf.len + 7) & ~7);
1609
        }
1610
        header.backing_file_offset = cpu_to_be64(header_size);
1611
        backing_filename_len = strlen(backing_file);
1612
        header.backing_file_size = cpu_to_be32(backing_filename_len);
1613
        header_size += backing_filename_len;
1614
    }
1615
    s->cluster_bits = 12;  /* 4 KB clusters */
1616
    s->cluster_size = 1 << s->cluster_bits;
1617
    header.cluster_bits = cpu_to_be32(s->cluster_bits);
1618
    header_size = (header_size + 7) & ~7;
1619
    if (flags & BLOCK_FLAG_ENCRYPT) {
1620
        header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
1621
    } else {
1622
        header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
1623
    }
1624
    l2_bits = s->cluster_bits - 3;
1625
    shift = s->cluster_bits + l2_bits;
1626
    l1_size = (((total_size * 512) + (1LL << shift) - 1) >> shift);
1627
    offset = align_offset(header_size, s->cluster_size);
1628
    s->l1_table_offset = offset;
1629
    header.l1_table_offset = cpu_to_be64(s->l1_table_offset);
1630
    header.l1_size = cpu_to_be32(l1_size);
1631
    offset += align_offset(l1_size * sizeof(uint64_t), s->cluster_size);
1632

    
1633
    s->refcount_table = qemu_mallocz(s->cluster_size);
1634

    
1635
    s->refcount_table_offset = offset;
1636
    header.refcount_table_offset = cpu_to_be64(offset);
1637
    header.refcount_table_clusters = cpu_to_be32(1);
1638
    offset += s->cluster_size;
1639
    s->refcount_block_offset = offset;
1640

    
1641
    /* count how many refcount blocks needed */
1642
    tmp = offset >> s->cluster_bits;
1643
    ref_clusters = (tmp >> (s->cluster_bits - REFCOUNT_SHIFT)) + 1;
1644
    for (i=0; i < ref_clusters; i++) {
1645
        s->refcount_table[i] = cpu_to_be64(offset);
1646
        offset += s->cluster_size;
1647
    }
1648

    
1649
    s->refcount_block = qemu_mallocz(ref_clusters * s->cluster_size);
1650

    
1651
    /* update refcounts */
1652
    create_refcount_update(s, 0, header_size);
1653
    create_refcount_update(s, s->l1_table_offset, l1_size * sizeof(uint64_t));
1654
    create_refcount_update(s, s->refcount_table_offset, s->cluster_size);
1655
    create_refcount_update(s, s->refcount_block_offset, ref_clusters * s->cluster_size);
1656

    
1657
    /* write all the data */
1658
    write(fd, &header, sizeof(header));
1659
    if (backing_file) {
1660
        if (backing_format_len) {
1661
            char zero[16];
1662
            int d = ext_bf.len - backing_format_len;
1663

    
1664
            memset(zero, 0, sizeof(zero));
1665
            cpu_to_be32s(&ext_bf.magic);
1666
            cpu_to_be32s(&ext_bf.len);
1667
            write(fd, &ext_bf, sizeof(ext_bf));
1668
            write(fd, backing_format, backing_format_len);
1669
            if (d>0) {
1670
                write(fd, zero, d);
1671
            }
1672
        }
1673
        write(fd, backing_file, backing_filename_len);
1674
    }
1675
    lseek(fd, s->l1_table_offset, SEEK_SET);
1676
    tmp = 0;
1677
    for(i = 0;i < l1_size; i++) {
1678
        write(fd, &tmp, sizeof(tmp));
1679
    }
1680
    lseek(fd, s->refcount_table_offset, SEEK_SET);
1681
    write(fd, s->refcount_table, s->cluster_size);
1682

    
1683
    lseek(fd, s->refcount_block_offset, SEEK_SET);
1684
    write(fd, s->refcount_block, ref_clusters * s->cluster_size);
1685

    
1686
    qemu_free(s->refcount_table);
1687
    qemu_free(s->refcount_block);
1688
    close(fd);
1689
    return 0;
1690
}
1691

    
1692
static int qcow_create(const char *filename, int64_t total_size,
1693
                       const char *backing_file, int flags)
1694
{
1695
    return qcow_create2(filename, total_size, backing_file, NULL, flags);
1696
}
1697

    
1698
static int qcow_make_empty(BlockDriverState *bs)
1699
{
1700
#if 0
1701
    /* XXX: not correct */
1702
    BDRVQcowState *s = bs->opaque;
1703
    uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1704
    int ret;
1705

1706
    memset(s->l1_table, 0, l1_length);
1707
    if (bdrv_pwrite(s->hd, s->l1_table_offset, s->l1_table, l1_length) < 0)
1708
        return -1;
1709
    ret = bdrv_truncate(s->hd, s->l1_table_offset + l1_length);
1710
    if (ret < 0)
1711
        return ret;
1712

1713
    l2_cache_reset(bs);
1714
#endif
1715
    return 0;
1716
}
1717

    
1718
/* XXX: put compressed sectors first, then all the cluster aligned
1719
   tables to avoid losing bytes in alignment */
1720
static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
1721
                                 const uint8_t *buf, int nb_sectors)
1722
{
1723
    BDRVQcowState *s = bs->opaque;
1724
    z_stream strm;
1725
    int ret, out_len;
1726
    uint8_t *out_buf;
1727
    uint64_t cluster_offset;
1728

    
1729
    if (nb_sectors == 0) {
1730
        /* align end of file to a sector boundary to ease reading with
1731
           sector based I/Os */
1732
        cluster_offset = bdrv_getlength(s->hd);
1733
        cluster_offset = (cluster_offset + 511) & ~511;
1734
        bdrv_truncate(s->hd, cluster_offset);
1735
        return 0;
1736
    }
1737

    
1738
    if (nb_sectors != s->cluster_sectors)
1739
        return -EINVAL;
1740

    
1741
    out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
1742

    
1743
    /* best compression, small window, no zlib header */
1744
    memset(&strm, 0, sizeof(strm));
1745
    ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1746
                       Z_DEFLATED, -12,
1747
                       9, Z_DEFAULT_STRATEGY);
1748
    if (ret != 0) {
1749
        qemu_free(out_buf);
1750
        return -1;
1751
    }
1752

    
1753
    strm.avail_in = s->cluster_size;
1754
    strm.next_in = (uint8_t *)buf;
1755
    strm.avail_out = s->cluster_size;
1756
    strm.next_out = out_buf;
1757

    
1758
    ret = deflate(&strm, Z_FINISH);
1759
    if (ret != Z_STREAM_END && ret != Z_OK) {
1760
        qemu_free(out_buf);
1761
        deflateEnd(&strm);
1762
        return -1;
1763
    }
1764
    out_len = strm.next_out - out_buf;
1765

    
1766
    deflateEnd(&strm);
1767

    
1768
    if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1769
        /* could not compress: write normal cluster */
1770
        qcow_write(bs, sector_num, buf, s->cluster_sectors);
1771
    } else {
1772
        cluster_offset = alloc_compressed_cluster_offset(bs, sector_num << 9,
1773
                                              out_len);
1774
        if (!cluster_offset)
1775
            return -1;
1776
        cluster_offset &= s->cluster_offset_mask;
1777
        if (bdrv_pwrite(s->hd, cluster_offset, out_buf, out_len) != out_len) {
1778
            qemu_free(out_buf);
1779
            return -1;
1780
        }
1781
    }
1782

    
1783
    qemu_free(out_buf);
1784
    return 0;
1785
}
1786

    
1787
static void qcow_flush(BlockDriverState *bs)
1788
{
1789
    BDRVQcowState *s = bs->opaque;
1790
    bdrv_flush(s->hd);
1791
}
1792

    
1793
static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1794
{
1795
    BDRVQcowState *s = bs->opaque;
1796
    bdi->cluster_size = s->cluster_size;
1797
    bdi->vm_state_offset = (int64_t)s->l1_vm_state_index <<
1798
        (s->cluster_bits + s->l2_bits);
1799
    return 0;
1800
}
1801

    
1802
/*********************************************************/
1803
/* snapshot support */
1804

    
1805
/* update the refcounts of snapshots and the copied flag */
1806
static int update_snapshot_refcount(BlockDriverState *bs,
1807
                                    int64_t l1_table_offset,
1808
                                    int l1_size,
1809
                                    int addend)
1810
{
1811
    BDRVQcowState *s = bs->opaque;
1812
    uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated;
1813
    int64_t old_offset, old_l2_offset;
1814
    int l2_size, i, j, l1_modified, l2_modified, nb_csectors, refcount;
1815

    
1816
    l2_cache_reset(bs);
1817

    
1818
    l2_table = NULL;
1819
    l1_table = NULL;
1820
    l1_size2 = l1_size * sizeof(uint64_t);
1821
    l1_allocated = 0;
1822
    if (l1_table_offset != s->l1_table_offset) {
1823
        l1_table = qemu_malloc(l1_size2);
1824
        l1_allocated = 1;
1825
        if (bdrv_pread(s->hd, l1_table_offset,
1826
                       l1_table, l1_size2) != l1_size2)
1827
            goto fail;
1828
        for(i = 0;i < l1_size; i++)
1829
            be64_to_cpus(&l1_table[i]);
1830
    } else {
1831
        assert(l1_size == s->l1_size);
1832
        l1_table = s->l1_table;
1833
        l1_allocated = 0;
1834
    }
1835

    
1836
    l2_size = s->l2_size * sizeof(uint64_t);
1837
    l2_table = qemu_malloc(l2_size);
1838
    l1_modified = 0;
1839
    for(i = 0; i < l1_size; i++) {
1840
        l2_offset = l1_table[i];
1841
        if (l2_offset) {
1842
            old_l2_offset = l2_offset;
1843
            l2_offset &= ~QCOW_OFLAG_COPIED;
1844
            l2_modified = 0;
1845
            if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
1846
                goto fail;
1847
            for(j = 0; j < s->l2_size; j++) {
1848
                offset = be64_to_cpu(l2_table[j]);
1849
                if (offset != 0) {
1850
                    old_offset = offset;
1851
                    offset &= ~QCOW_OFLAG_COPIED;
1852
                    if (offset & QCOW_OFLAG_COMPRESSED) {
1853
                        nb_csectors = ((offset >> s->csize_shift) &
1854
                                       s->csize_mask) + 1;
1855
                        if (addend != 0)
1856
                            update_refcount(bs, (offset & s->cluster_offset_mask) & ~511,
1857
                                            nb_csectors * 512, addend);
1858
                        /* compressed clusters are never modified */
1859
                        refcount = 2;
1860
                    } else {
1861
                        if (addend != 0) {
1862
                            refcount = update_cluster_refcount(bs, offset >> s->cluster_bits, addend);
1863
                        } else {
1864
                            refcount = get_refcount(bs, offset >> s->cluster_bits);
1865
                        }
1866
                    }
1867

    
1868
                    if (refcount == 1) {
1869
                        offset |= QCOW_OFLAG_COPIED;
1870
                    }
1871
                    if (offset != old_offset) {
1872
                        l2_table[j] = cpu_to_be64(offset);
1873
                        l2_modified = 1;
1874
                    }
1875
                }
1876
            }
1877
            if (l2_modified) {
1878
                if (bdrv_pwrite(s->hd,
1879
                                l2_offset, l2_table, l2_size) != l2_size)
1880
                    goto fail;
1881
            }
1882

    
1883
            if (addend != 0) {
1884
                refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend);
1885
            } else {
1886
                refcount = get_refcount(bs, l2_offset >> s->cluster_bits);
1887
            }
1888
            if (refcount == 1) {
1889
                l2_offset |= QCOW_OFLAG_COPIED;
1890
            }
1891
            if (l2_offset != old_l2_offset) {
1892
                l1_table[i] = l2_offset;
1893
                l1_modified = 1;
1894
            }
1895
        }
1896
    }
1897
    if (l1_modified) {
1898
        for(i = 0; i < l1_size; i++)
1899
            cpu_to_be64s(&l1_table[i]);
1900
        if (bdrv_pwrite(s->hd, l1_table_offset, l1_table,
1901
                        l1_size2) != l1_size2)
1902
            goto fail;
1903
        for(i = 0; i < l1_size; i++)
1904
            be64_to_cpus(&l1_table[i]);
1905
    }
1906
    if (l1_allocated)
1907
        qemu_free(l1_table);
1908
    qemu_free(l2_table);
1909
    return 0;
1910
 fail:
1911
    if (l1_allocated)
1912
        qemu_free(l1_table);
1913
    qemu_free(l2_table);
1914
    return -EIO;
1915
}
1916

    
1917
static void qcow_free_snapshots(BlockDriverState *bs)
1918
{
1919
    BDRVQcowState *s = bs->opaque;
1920
    int i;
1921

    
1922
    for(i = 0; i < s->nb_snapshots; i++) {
1923
        qemu_free(s->snapshots[i].name);
1924
        qemu_free(s->snapshots[i].id_str);
1925
    }
1926
    qemu_free(s->snapshots);
1927
    s->snapshots = NULL;
1928
    s->nb_snapshots = 0;
1929
}
1930

    
1931
static int qcow_read_snapshots(BlockDriverState *bs)
1932
{
1933
    BDRVQcowState *s = bs->opaque;
1934
    QCowSnapshotHeader h;
1935
    QCowSnapshot *sn;
1936
    int i, id_str_size, name_size;
1937
    int64_t offset;
1938
    uint32_t extra_data_size;
1939

    
1940
    if (!s->nb_snapshots) {
1941
        s->snapshots = NULL;
1942
        s->snapshots_size = 0;
1943
        return 0;
1944
    }
1945

    
1946
    offset = s->snapshots_offset;
1947
    s->snapshots = qemu_mallocz(s->nb_snapshots * sizeof(QCowSnapshot));
1948
    for(i = 0; i < s->nb_snapshots; i++) {
1949
        offset = align_offset(offset, 8);
1950
        if (bdrv_pread(s->hd, offset, &h, sizeof(h)) != sizeof(h))
1951
            goto fail;
1952
        offset += sizeof(h);
1953
        sn = s->snapshots + i;
1954
        sn->l1_table_offset = be64_to_cpu(h.l1_table_offset);
1955
        sn->l1_size = be32_to_cpu(h.l1_size);
1956
        sn->vm_state_size = be32_to_cpu(h.vm_state_size);
1957
        sn->date_sec = be32_to_cpu(h.date_sec);
1958
        sn->date_nsec = be32_to_cpu(h.date_nsec);
1959
        sn->vm_clock_nsec = be64_to_cpu(h.vm_clock_nsec);
1960
        extra_data_size = be32_to_cpu(h.extra_data_size);
1961

    
1962
        id_str_size = be16_to_cpu(h.id_str_size);
1963
        name_size = be16_to_cpu(h.name_size);
1964

    
1965
        offset += extra_data_size;
1966

    
1967
        sn->id_str = qemu_malloc(id_str_size + 1);
1968
        if (bdrv_pread(s->hd, offset, sn->id_str, id_str_size) != id_str_size)
1969
            goto fail;
1970
        offset += id_str_size;
1971
        sn->id_str[id_str_size] = '\0';
1972

    
1973
        sn->name = qemu_malloc(name_size + 1);
1974
        if (bdrv_pread(s->hd, offset, sn->name, name_size) != name_size)
1975
            goto fail;
1976
        offset += name_size;
1977
        sn->name[name_size] = '\0';
1978
    }
1979
    s->snapshots_size = offset - s->snapshots_offset;
1980
    return 0;
1981
 fail:
1982
    qcow_free_snapshots(bs);
1983
    return -1;
1984
}
1985

    
1986
/* add at the end of the file a new list of snapshots */
1987
static int qcow_write_snapshots(BlockDriverState *bs)
1988
{
1989
    BDRVQcowState *s = bs->opaque;
1990
    QCowSnapshot *sn;
1991
    QCowSnapshotHeader h;
1992
    int i, name_size, id_str_size, snapshots_size;
1993
    uint64_t data64;
1994
    uint32_t data32;
1995
    int64_t offset, snapshots_offset;
1996

    
1997
    /* compute the size of the snapshots */
1998
    offset = 0;
1999
    for(i = 0; i < s->nb_snapshots; i++) {
2000
        sn = s->snapshots + i;
2001
        offset = align_offset(offset, 8);
2002
        offset += sizeof(h);
2003
        offset += strlen(sn->id_str);
2004
        offset += strlen(sn->name);
2005
    }
2006
    snapshots_size = offset;
2007

    
2008
    snapshots_offset = alloc_clusters(bs, snapshots_size);
2009
    offset = snapshots_offset;
2010

    
2011
    for(i = 0; i < s->nb_snapshots; i++) {
2012
        sn = s->snapshots + i;
2013
        memset(&h, 0, sizeof(h));
2014
        h.l1_table_offset = cpu_to_be64(sn->l1_table_offset);
2015
        h.l1_size = cpu_to_be32(sn->l1_size);
2016
        h.vm_state_size = cpu_to_be32(sn->vm_state_size);
2017
        h.date_sec = cpu_to_be32(sn->date_sec);
2018
        h.date_nsec = cpu_to_be32(sn->date_nsec);
2019
        h.vm_clock_nsec = cpu_to_be64(sn->vm_clock_nsec);
2020

    
2021
        id_str_size = strlen(sn->id_str);
2022
        name_size = strlen(sn->name);
2023
        h.id_str_size = cpu_to_be16(id_str_size);
2024
        h.name_size = cpu_to_be16(name_size);
2025
        offset = align_offset(offset, 8);
2026
        if (bdrv_pwrite(s->hd, offset, &h, sizeof(h)) != sizeof(h))
2027
            goto fail;
2028
        offset += sizeof(h);
2029
        if (bdrv_pwrite(s->hd, offset, sn->id_str, id_str_size) != id_str_size)
2030
            goto fail;
2031
        offset += id_str_size;
2032
        if (bdrv_pwrite(s->hd, offset, sn->name, name_size) != name_size)
2033
            goto fail;
2034
        offset += name_size;
2035
    }
2036

    
2037
    /* update the various header fields */
2038
    data64 = cpu_to_be64(snapshots_offset);
2039
    if (bdrv_pwrite(s->hd, offsetof(QCowHeader, snapshots_offset),
2040
                    &data64, sizeof(data64)) != sizeof(data64))
2041
        goto fail;
2042
    data32 = cpu_to_be32(s->nb_snapshots);
2043
    if (bdrv_pwrite(s->hd, offsetof(QCowHeader, nb_snapshots),
2044
                    &data32, sizeof(data32)) != sizeof(data32))
2045
        goto fail;
2046

    
2047
    /* free the old snapshot table */
2048
    free_clusters(bs, s->snapshots_offset, s->snapshots_size);
2049
    s->snapshots_offset = snapshots_offset;
2050
    s->snapshots_size = snapshots_size;
2051
    return 0;
2052
 fail:
2053
    return -1;
2054
}
2055

    
2056
static void find_new_snapshot_id(BlockDriverState *bs,
2057
                                 char *id_str, int id_str_size)
2058
{
2059
    BDRVQcowState *s = bs->opaque;
2060
    QCowSnapshot *sn;
2061
    int i, id, id_max = 0;
2062

    
2063
    for(i = 0; i < s->nb_snapshots; i++) {
2064
        sn = s->snapshots + i;
2065
        id = strtoul(sn->id_str, NULL, 10);
2066
        if (id > id_max)
2067
            id_max = id;
2068
    }
2069
    snprintf(id_str, id_str_size, "%d", id_max + 1);
2070
}
2071

    
2072
static int find_snapshot_by_id(BlockDriverState *bs, const char *id_str)
2073
{
2074
    BDRVQcowState *s = bs->opaque;
2075
    int i;
2076

    
2077
    for(i = 0; i < s->nb_snapshots; i++) {
2078
        if (!strcmp(s->snapshots[i].id_str, id_str))
2079
            return i;
2080
    }
2081
    return -1;
2082
}
2083

    
2084
static int find_snapshot_by_id_or_name(BlockDriverState *bs, const char *name)
2085
{
2086
    BDRVQcowState *s = bs->opaque;
2087
    int i, ret;
2088

    
2089
    ret = find_snapshot_by_id(bs, name);
2090
    if (ret >= 0)
2091
        return ret;
2092
    for(i = 0; i < s->nb_snapshots; i++) {
2093
        if (!strcmp(s->snapshots[i].name, name))
2094
            return i;
2095
    }
2096
    return -1;
2097
}
2098

    
2099
/* if no id is provided, a new one is constructed */
2100
static int qcow_snapshot_create(BlockDriverState *bs,
2101
                                QEMUSnapshotInfo *sn_info)
2102
{
2103
    BDRVQcowState *s = bs->opaque;
2104
    QCowSnapshot *snapshots1, sn1, *sn = &sn1;
2105
    int i, ret;
2106
    uint64_t *l1_table = NULL;
2107

    
2108
    memset(sn, 0, sizeof(*sn));
2109

    
2110
    if (sn_info->id_str[0] == '\0') {
2111
        /* compute a new id */
2112
        find_new_snapshot_id(bs, sn_info->id_str, sizeof(sn_info->id_str));
2113
    }
2114

    
2115
    /* check that the ID is unique */
2116
    if (find_snapshot_by_id(bs, sn_info->id_str) >= 0)
2117
        return -ENOENT;
2118

    
2119
    sn->id_str = qemu_strdup(sn_info->id_str);
2120
    if (!sn->id_str)
2121
        goto fail;
2122
    sn->name = qemu_strdup(sn_info->name);
2123
    if (!sn->name)
2124
        goto fail;
2125
    sn->vm_state_size = sn_info->vm_state_size;
2126
    sn->date_sec = sn_info->date_sec;
2127
    sn->date_nsec = sn_info->date_nsec;
2128
    sn->vm_clock_nsec = sn_info->vm_clock_nsec;
2129

    
2130
    ret = update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1);
2131
    if (ret < 0)
2132
        goto fail;
2133

    
2134
    /* create the L1 table of the snapshot */
2135
    sn->l1_table_offset = alloc_clusters(bs, s->l1_size * sizeof(uint64_t));
2136
    sn->l1_size = s->l1_size;
2137

    
2138
    l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
2139
    for(i = 0; i < s->l1_size; i++) {
2140
        l1_table[i] = cpu_to_be64(s->l1_table[i]);
2141
    }
2142
    if (bdrv_pwrite(s->hd, sn->l1_table_offset,
2143
                    l1_table, s->l1_size * sizeof(uint64_t)) !=
2144
        (s->l1_size * sizeof(uint64_t)))
2145
        goto fail;
2146
    qemu_free(l1_table);
2147
    l1_table = NULL;
2148

    
2149
    snapshots1 = qemu_malloc((s->nb_snapshots + 1) * sizeof(QCowSnapshot));
2150
    if (s->snapshots) {
2151
        memcpy(snapshots1, s->snapshots, s->nb_snapshots * sizeof(QCowSnapshot));
2152
        qemu_free(s->snapshots);
2153
    }
2154
    s->snapshots = snapshots1;
2155
    s->snapshots[s->nb_snapshots++] = *sn;
2156

    
2157
    if (qcow_write_snapshots(bs) < 0)
2158
        goto fail;
2159
#ifdef DEBUG_ALLOC
2160
    check_refcounts(bs);
2161
#endif
2162
    return 0;
2163
 fail:
2164
    qemu_free(sn->name);
2165
    qemu_free(l1_table);
2166
    return -1;
2167
}
2168

    
2169
/* copy the snapshot 'snapshot_name' into the current disk image */
2170
static int qcow_snapshot_goto(BlockDriverState *bs,
2171
                              const char *snapshot_id)
2172
{
2173
    BDRVQcowState *s = bs->opaque;
2174
    QCowSnapshot *sn;
2175
    int i, snapshot_index, l1_size2;
2176

    
2177
    snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
2178
    if (snapshot_index < 0)
2179
        return -ENOENT;
2180
    sn = &s->snapshots[snapshot_index];
2181

    
2182
    if (update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, -1) < 0)
2183
        goto fail;
2184

    
2185
    if (grow_l1_table(bs, sn->l1_size) < 0)
2186
        goto fail;
2187

    
2188
    s->l1_size = sn->l1_size;
2189
    l1_size2 = s->l1_size * sizeof(uint64_t);
2190
    /* copy the snapshot l1 table to the current l1 table */
2191
    if (bdrv_pread(s->hd, sn->l1_table_offset,
2192
                   s->l1_table, l1_size2) != l1_size2)
2193
        goto fail;
2194
    if (bdrv_pwrite(s->hd, s->l1_table_offset,
2195
                    s->l1_table, l1_size2) != l1_size2)
2196
        goto fail;
2197
    for(i = 0;i < s->l1_size; i++) {
2198
        be64_to_cpus(&s->l1_table[i]);
2199
    }
2200

    
2201
    if (update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1) < 0)
2202
        goto fail;
2203

    
2204
#ifdef DEBUG_ALLOC
2205
    check_refcounts(bs);
2206
#endif
2207
    return 0;
2208
 fail:
2209
    return -EIO;
2210
}
2211

    
2212
static int qcow_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
2213
{
2214
    BDRVQcowState *s = bs->opaque;
2215
    QCowSnapshot *sn;
2216
    int snapshot_index, ret;
2217

    
2218
    snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
2219
    if (snapshot_index < 0)
2220
        return -ENOENT;
2221
    sn = &s->snapshots[snapshot_index];
2222

    
2223
    ret = update_snapshot_refcount(bs, sn->l1_table_offset, sn->l1_size, -1);
2224
    if (ret < 0)
2225
        return ret;
2226
    /* must update the copied flag on the current cluster offsets */
2227
    ret = update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 0);
2228
    if (ret < 0)
2229
        return ret;
2230
    free_clusters(bs, sn->l1_table_offset, sn->l1_size * sizeof(uint64_t));
2231

    
2232
    qemu_free(sn->id_str);
2233
    qemu_free(sn->name);
2234
    memmove(sn, sn + 1, (s->nb_snapshots - snapshot_index - 1) * sizeof(*sn));
2235
    s->nb_snapshots--;
2236
    ret = qcow_write_snapshots(bs);
2237
    if (ret < 0) {
2238
        /* XXX: restore snapshot if error ? */
2239
        return ret;
2240
    }
2241
#ifdef DEBUG_ALLOC
2242
    check_refcounts(bs);
2243
#endif
2244
    return 0;
2245
}
2246

    
2247
static int qcow_snapshot_list(BlockDriverState *bs,
2248
                              QEMUSnapshotInfo **psn_tab)
2249
{
2250
    BDRVQcowState *s = bs->opaque;
2251
    QEMUSnapshotInfo *sn_tab, *sn_info;
2252
    QCowSnapshot *sn;
2253
    int i;
2254

    
2255
    sn_tab = qemu_mallocz(s->nb_snapshots * sizeof(QEMUSnapshotInfo));
2256
    for(i = 0; i < s->nb_snapshots; i++) {
2257
        sn_info = sn_tab + i;
2258
        sn = s->snapshots + i;
2259
        pstrcpy(sn_info->id_str, sizeof(sn_info->id_str),
2260
                sn->id_str);
2261
        pstrcpy(sn_info->name, sizeof(sn_info->name),
2262
                sn->name);
2263
        sn_info->vm_state_size = sn->vm_state_size;
2264
        sn_info->date_sec = sn->date_sec;
2265
        sn_info->date_nsec = sn->date_nsec;
2266
        sn_info->vm_clock_nsec = sn->vm_clock_nsec;
2267
    }
2268
    *psn_tab = sn_tab;
2269
    return s->nb_snapshots;
2270
}
2271

    
2272
/*********************************************************/
2273
/* refcount handling */
2274

    
2275
static int refcount_init(BlockDriverState *bs)
2276
{
2277
    BDRVQcowState *s = bs->opaque;
2278
    int ret, refcount_table_size2, i;
2279

    
2280
    s->refcount_block_cache = qemu_malloc(s->cluster_size);
2281
    refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t);
2282
    s->refcount_table = qemu_malloc(refcount_table_size2);
2283
    if (s->refcount_table_size > 0) {
2284
        ret = bdrv_pread(s->hd, s->refcount_table_offset,
2285
                         s->refcount_table, refcount_table_size2);
2286
        if (ret != refcount_table_size2)
2287
            goto fail;
2288
        for(i = 0; i < s->refcount_table_size; i++)
2289
            be64_to_cpus(&s->refcount_table[i]);
2290
    }
2291
    return 0;
2292
 fail:
2293
    return -ENOMEM;
2294
}
2295

    
2296
static void refcount_close(BlockDriverState *bs)
2297
{
2298
    BDRVQcowState *s = bs->opaque;
2299
    qemu_free(s->refcount_block_cache);
2300
    qemu_free(s->refcount_table);
2301
}
2302

    
2303

    
2304
static int load_refcount_block(BlockDriverState *bs,
2305
                               int64_t refcount_block_offset)
2306
{
2307
    BDRVQcowState *s = bs->opaque;
2308
    int ret;
2309
    ret = bdrv_pread(s->hd, refcount_block_offset, s->refcount_block_cache,
2310
                     s->cluster_size);
2311
    if (ret != s->cluster_size)
2312
        return -EIO;
2313
    s->refcount_block_cache_offset = refcount_block_offset;
2314
    return 0;
2315
}
2316

    
2317
static int get_refcount(BlockDriverState *bs, int64_t cluster_index)
2318
{
2319
    BDRVQcowState *s = bs->opaque;
2320
    int refcount_table_index, block_index;
2321
    int64_t refcount_block_offset;
2322

    
2323
    refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
2324
    if (refcount_table_index >= s->refcount_table_size)
2325
        return 0;
2326
    refcount_block_offset = s->refcount_table[refcount_table_index];
2327
    if (!refcount_block_offset)
2328
        return 0;
2329
    if (refcount_block_offset != s->refcount_block_cache_offset) {
2330
        /* better than nothing: return allocated if read error */
2331
        if (load_refcount_block(bs, refcount_block_offset) < 0)
2332
            return 1;
2333
    }
2334
    block_index = cluster_index &
2335
        ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
2336
    return be16_to_cpu(s->refcount_block_cache[block_index]);
2337
}
2338

    
2339
/* return < 0 if error */
2340
static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size)
2341
{
2342
    BDRVQcowState *s = bs->opaque;
2343
    int i, nb_clusters;
2344

    
2345
    nb_clusters = size_to_clusters(s, size);
2346
retry:
2347
    for(i = 0; i < nb_clusters; i++) {
2348
        int64_t i = s->free_cluster_index++;
2349
        if (get_refcount(bs, i) != 0)
2350
            goto retry;
2351
    }
2352
#ifdef DEBUG_ALLOC2
2353
    printf("alloc_clusters: size=%lld -> %lld\n",
2354
            size,
2355
            (s->free_cluster_index - nb_clusters) << s->cluster_bits);
2356
#endif
2357
    return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
2358
}
2359

    
2360
static int64_t alloc_clusters(BlockDriverState *bs, int64_t size)
2361
{
2362
    int64_t offset;
2363

    
2364
    offset = alloc_clusters_noref(bs, size);
2365
    update_refcount(bs, offset, size, 1);
2366
    return offset;
2367
}
2368

    
2369
/* only used to allocate compressed sectors. We try to allocate
2370
   contiguous sectors. size must be <= cluster_size */
2371
static int64_t alloc_bytes(BlockDriverState *bs, int size)
2372
{
2373
    BDRVQcowState *s = bs->opaque;
2374
    int64_t offset, cluster_offset;
2375
    int free_in_cluster;
2376

    
2377
    assert(size > 0 && size <= s->cluster_size);
2378
    if (s->free_byte_offset == 0) {
2379
        s->free_byte_offset = alloc_clusters(bs, s->cluster_size);
2380
    }
2381
 redo:
2382
    free_in_cluster = s->cluster_size -
2383
        (s->free_byte_offset & (s->cluster_size - 1));
2384
    if (size <= free_in_cluster) {
2385
        /* enough space in current cluster */
2386
        offset = s->free_byte_offset;
2387
        s->free_byte_offset += size;
2388
        free_in_cluster -= size;
2389
        if (free_in_cluster == 0)
2390
            s->free_byte_offset = 0;
2391
        if ((offset & (s->cluster_size - 1)) != 0)
2392
            update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
2393
    } else {
2394
        offset = alloc_clusters(bs, s->cluster_size);
2395
        cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1);
2396
        if ((cluster_offset + s->cluster_size) == offset) {
2397
            /* we are lucky: contiguous data */
2398
            offset = s->free_byte_offset;
2399
            update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
2400
            s->free_byte_offset += size;
2401
        } else {
2402
            s->free_byte_offset = offset;
2403
            goto redo;
2404
        }
2405
    }
2406
    return offset;
2407
}
2408

    
2409
static void free_clusters(BlockDriverState *bs,
2410
                          int64_t offset, int64_t size)
2411
{
2412
    update_refcount(bs, offset, size, -1);
2413
}
2414

    
2415
static int grow_refcount_table(BlockDriverState *bs, int min_size)
2416
{
2417
    BDRVQcowState *s = bs->opaque;
2418
    int new_table_size, new_table_size2, refcount_table_clusters, i, ret;
2419
    uint64_t *new_table;
2420
    int64_t table_offset;
2421
    uint8_t data[12];
2422
    int old_table_size;
2423
    int64_t old_table_offset;
2424

    
2425
    if (min_size <= s->refcount_table_size)
2426
        return 0;
2427
    /* compute new table size */
2428
    refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
2429
    for(;;) {
2430
        if (refcount_table_clusters == 0) {
2431
            refcount_table_clusters = 1;
2432
        } else {
2433
            refcount_table_clusters = (refcount_table_clusters * 3 + 1) / 2;
2434
        }
2435
        new_table_size = refcount_table_clusters << (s->cluster_bits - 3);
2436
        if (min_size <= new_table_size)
2437
            break;
2438
    }
2439
#ifdef DEBUG_ALLOC2
2440
    printf("grow_refcount_table from %d to %d\n",
2441
           s->refcount_table_size,
2442
           new_table_size);
2443
#endif
2444
    new_table_size2 = new_table_size * sizeof(uint64_t);
2445
    new_table = qemu_mallocz(new_table_size2);
2446
    memcpy(new_table, s->refcount_table,
2447
           s->refcount_table_size * sizeof(uint64_t));
2448
    for(i = 0; i < s->refcount_table_size; i++)
2449
        cpu_to_be64s(&new_table[i]);
2450
    /* Note: we cannot update the refcount now to avoid recursion */
2451
    table_offset = alloc_clusters_noref(bs, new_table_size2);
2452
    ret = bdrv_pwrite(s->hd, table_offset, new_table, new_table_size2);
2453
    if (ret != new_table_size2)
2454
        goto fail;
2455
    for(i = 0; i < s->refcount_table_size; i++)
2456
        be64_to_cpus(&new_table[i]);
2457

    
2458
    cpu_to_be64w((uint64_t*)data, table_offset);
2459
    cpu_to_be32w((uint32_t*)(data + 8), refcount_table_clusters);
2460
    if (bdrv_pwrite(s->hd, offsetof(QCowHeader, refcount_table_offset),
2461
                    data, sizeof(data)) != sizeof(data))
2462
        goto fail;
2463
    qemu_free(s->refcount_table);
2464
    old_table_offset = s->refcount_table_offset;
2465
    old_table_size = s->refcount_table_size;
2466
    s->refcount_table = new_table;
2467
    s->refcount_table_size = new_table_size;
2468
    s->refcount_table_offset = table_offset;
2469

    
2470
    update_refcount(bs, table_offset, new_table_size2, 1);
2471
    free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t));
2472
    return 0;
2473
 fail:
2474
    free_clusters(bs, table_offset, new_table_size2);
2475
    qemu_free(new_table);
2476
    return -EIO;
2477
}
2478

    
2479
/* addend must be 1 or -1 */
2480
/* XXX: cache several refcount block clusters ? */
2481
static int update_cluster_refcount(BlockDriverState *bs,
2482
                                   int64_t cluster_index,
2483
                                   int addend)
2484
{
2485
    BDRVQcowState *s = bs->opaque;
2486
    int64_t offset, refcount_block_offset;
2487
    int ret, refcount_table_index, block_index, refcount;
2488
    uint64_t data64;
2489

    
2490
    refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
2491
    if (refcount_table_index >= s->refcount_table_size) {
2492
        if (addend < 0)
2493
            return -EINVAL;
2494
        ret = grow_refcount_table(bs, refcount_table_index + 1);
2495
        if (ret < 0)
2496
            return ret;
2497
    }
2498
    refcount_block_offset = s->refcount_table[refcount_table_index];
2499
    if (!refcount_block_offset) {
2500
        if (addend < 0)
2501
            return -EINVAL;
2502
        /* create a new refcount block */
2503
        /* Note: we cannot update the refcount now to avoid recursion */
2504
        offset = alloc_clusters_noref(bs, s->cluster_size);
2505
        memset(s->refcount_block_cache, 0, s->cluster_size);
2506
        ret = bdrv_pwrite(s->hd, offset, s->refcount_block_cache, s->cluster_size);
2507
        if (ret != s->cluster_size)
2508
            return -EINVAL;
2509
        s->refcount_table[refcount_table_index] = offset;
2510
        data64 = cpu_to_be64(offset);
2511
        ret = bdrv_pwrite(s->hd, s->refcount_table_offset +
2512
                          refcount_table_index * sizeof(uint64_t),
2513
                          &data64, sizeof(data64));
2514
        if (ret != sizeof(data64))
2515
            return -EINVAL;
2516

    
2517
        refcount_block_offset = offset;
2518
        s->refcount_block_cache_offset = offset;
2519
        update_refcount(bs, offset, s->cluster_size, 1);
2520
    } else {
2521
        if (refcount_block_offset != s->refcount_block_cache_offset) {
2522
            if (load_refcount_block(bs, refcount_block_offset) < 0)
2523
                return -EIO;
2524
        }
2525
    }
2526
    /* we can update the count and save it */
2527
    block_index = cluster_index &
2528
        ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
2529
    refcount = be16_to_cpu(s->refcount_block_cache[block_index]);
2530
    refcount += addend;
2531
    if (refcount < 0 || refcount > 0xffff)
2532
        return -EINVAL;
2533
    if (refcount == 0 && cluster_index < s->free_cluster_index) {
2534
        s->free_cluster_index = cluster_index;
2535
    }
2536
    s->refcount_block_cache[block_index] = cpu_to_be16(refcount);
2537
    if (bdrv_pwrite(s->hd,
2538
                    refcount_block_offset + (block_index << REFCOUNT_SHIFT),
2539
                    &s->refcount_block_cache[block_index], 2) != 2)
2540
        return -EIO;
2541
    return refcount;
2542
}
2543

    
2544
static void update_refcount(BlockDriverState *bs,
2545
                            int64_t offset, int64_t length,
2546
                            int addend)
2547
{
2548
    BDRVQcowState *s = bs->opaque;
2549
    int64_t start, last, cluster_offset;
2550

    
2551
#ifdef DEBUG_ALLOC2
2552
    printf("update_refcount: offset=%lld size=%lld addend=%d\n",
2553
           offset, length, addend);
2554
#endif
2555
    if (length <= 0)
2556
        return;
2557
    start = offset & ~(s->cluster_size - 1);
2558
    last = (offset + length - 1) & ~(s->cluster_size - 1);
2559
    for(cluster_offset = start; cluster_offset <= last;
2560
        cluster_offset += s->cluster_size) {
2561
        update_cluster_refcount(bs, cluster_offset >> s->cluster_bits, addend);
2562
    }
2563
}
2564

    
2565
/*
2566
 * Increases the refcount for a range of clusters in a given refcount table.
2567
 * This is used to construct a temporary refcount table out of L1 and L2 tables
2568
 * which can be compared the the refcount table saved in the image.
2569
 *
2570
 * Returns the number of errors in the image that were found
2571
 */
2572
static int inc_refcounts(BlockDriverState *bs,
2573
                          uint16_t *refcount_table,
2574
                          int refcount_table_size,
2575
                          int64_t offset, int64_t size)
2576
{
2577
    BDRVQcowState *s = bs->opaque;
2578
    int64_t start, last, cluster_offset;
2579
    int k;
2580
    int errors = 0;
2581

    
2582
    if (size <= 0)
2583
        return 0;
2584

    
2585
    start = offset & ~(s->cluster_size - 1);
2586
    last = (offset + size - 1) & ~(s->cluster_size - 1);
2587
    for(cluster_offset = start; cluster_offset <= last;
2588
        cluster_offset += s->cluster_size) {
2589
        k = cluster_offset >> s->cluster_bits;
2590
        if (k < 0 || k >= refcount_table_size) {
2591
            fprintf(stderr, "ERROR: invalid cluster offset=0x%" PRIx64 "\n",
2592
                cluster_offset);
2593
            errors++;
2594
        } else {
2595
            if (++refcount_table[k] == 0) {
2596
                fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64
2597
                    "\n", cluster_offset);
2598
                errors++;
2599
            }
2600
        }
2601
    }
2602

    
2603
    return errors;
2604
}
2605

    
2606
/*
2607
 * Increases the refcount in the given refcount table for the all clusters
2608
 * referenced in the L2 table. While doing so, performs some checks on L2
2609
 * entries.
2610
 *
2611
 * Returns the number of errors found by the checks or -errno if an internal
2612
 * error occurred.
2613
 */
2614
static int check_refcounts_l2(BlockDriverState *bs,
2615
    uint16_t *refcount_table, int refcount_table_size, int64_t l2_offset,
2616
    int check_copied)
2617
{
2618
    BDRVQcowState *s = bs->opaque;
2619
    uint64_t *l2_table, offset;
2620
    int i, l2_size, nb_csectors, refcount;
2621
    int errors = 0;
2622

    
2623
    /* Read L2 table from disk */
2624
    l2_size = s->l2_size * sizeof(uint64_t);
2625
    l2_table = qemu_malloc(l2_size);
2626

    
2627
    if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
2628
        goto fail;
2629

    
2630
    /* Do the actual checks */
2631
    for(i = 0; i < s->l2_size; i++) {
2632
        offset = be64_to_cpu(l2_table[i]);
2633
        if (offset != 0) {
2634
            if (offset & QCOW_OFLAG_COMPRESSED) {
2635
                /* Compressed clusters don't have QCOW_OFLAG_COPIED */
2636
                if (offset & QCOW_OFLAG_COPIED) {
2637
                    fprintf(stderr, "ERROR: cluster %" PRId64 ": "
2638
                        "copied flag must never be set for compressed "
2639
                        "clusters\n", offset >> s->cluster_bits);
2640
                    offset &= ~QCOW_OFLAG_COPIED;
2641
                    errors++;
2642
                }
2643

    
2644
                /* Mark cluster as used */
2645
                nb_csectors = ((offset >> s->csize_shift) &
2646
                               s->csize_mask) + 1;
2647
                offset &= s->cluster_offset_mask;
2648
                errors += inc_refcounts(bs, refcount_table,
2649
                              refcount_table_size,
2650
                              offset & ~511, nb_csectors * 512);
2651
            } else {
2652
                /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
2653
                if (check_copied) {
2654
                    uint64_t entry = offset;
2655
                    offset &= ~QCOW_OFLAG_COPIED;
2656
                    refcount = get_refcount(bs, offset >> s->cluster_bits);
2657
                    if ((refcount == 1) != ((entry & QCOW_OFLAG_COPIED) != 0)) {
2658
                        fprintf(stderr, "ERROR OFLAG_COPIED: offset=%"
2659
                            PRIx64 " refcount=%d\n", entry, refcount);
2660
                        errors++;
2661
                    }
2662
                }
2663

    
2664
                /* Mark cluster as used */
2665
                offset &= ~QCOW_OFLAG_COPIED;
2666
                errors += inc_refcounts(bs, refcount_table,
2667
                              refcount_table_size,
2668
                              offset, s->cluster_size);
2669

    
2670
                /* Correct offsets are cluster aligned */
2671
                if (offset & (s->cluster_size - 1)) {
2672
                    fprintf(stderr, "ERROR offset=%" PRIx64 ": Cluster is not "
2673
                        "properly aligned; L2 entry corrupted.\n", offset);
2674
                    errors++;
2675
                }
2676
            }
2677
        }
2678
    }
2679

    
2680
    qemu_free(l2_table);
2681
    return errors;
2682

    
2683
fail:
2684
    fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
2685
    qemu_free(l2_table);
2686
    return -EIO;
2687
}
2688

    
2689
/*
2690
 * Increases the refcount for the L1 table, its L2 tables and all referenced
2691
 * clusters in the given refcount table. While doing so, performs some checks
2692
 * on L1 and L2 entries.
2693
 *
2694
 * Returns the number of errors found by the checks or -errno if an internal
2695
 * error occurred.
2696
 */
2697
static int check_refcounts_l1(BlockDriverState *bs,
2698
                              uint16_t *refcount_table,
2699
                              int refcount_table_size,
2700
                              int64_t l1_table_offset, int l1_size,
2701
                              int check_copied)
2702
{
2703
    BDRVQcowState *s = bs->opaque;
2704
    uint64_t *l1_table, l2_offset, l1_size2;
2705
    int i, refcount, ret;
2706
    int errors = 0;
2707

    
2708
    l1_size2 = l1_size * sizeof(uint64_t);
2709

    
2710
    /* Mark L1 table as used */
2711
    errors += inc_refcounts(bs, refcount_table, refcount_table_size,
2712
                  l1_table_offset, l1_size2);
2713

    
2714
    /* Read L1 table entries from disk */
2715
    l1_table = qemu_malloc(l1_size2);
2716
    if (bdrv_pread(s->hd, l1_table_offset,
2717
                   l1_table, l1_size2) != l1_size2)
2718
        goto fail;
2719
    for(i = 0;i < l1_size; i++)
2720
        be64_to_cpus(&l1_table[i]);
2721

    
2722
    /* Do the actual checks */
2723
    for(i = 0; i < l1_size; i++) {
2724
        l2_offset = l1_table[i];
2725
        if (l2_offset) {
2726
            /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
2727
            if (check_copied) {
2728
                refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED)
2729
                    >> s->cluster_bits);
2730
                if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) {
2731
                    fprintf(stderr, "ERROR OFLAG_COPIED: l2_offset=%" PRIx64
2732
                        " refcount=%d\n", l2_offset, refcount);
2733
                    errors++;
2734
                }
2735
            }
2736

    
2737
            /* Mark L2 table as used */
2738
            l2_offset &= ~QCOW_OFLAG_COPIED;
2739
            errors += inc_refcounts(bs, refcount_table,
2740
                          refcount_table_size,
2741
                          l2_offset,
2742
                          s->cluster_size);
2743

    
2744
            /* L2 tables are cluster aligned */
2745
            if (l2_offset & (s->cluster_size - 1)) {
2746
                fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not "
2747
                    "cluster aligned; L1 entry corrupted\n", l2_offset);
2748
                errors++;
2749
            }
2750

    
2751
            /* Process and check L2 entries */
2752
            ret = check_refcounts_l2(bs, refcount_table, refcount_table_size,
2753
                l2_offset, check_copied);
2754
            if (ret < 0) {
2755
                goto fail;
2756
            }
2757
            errors += ret;
2758
        }
2759
    }
2760
    qemu_free(l1_table);
2761
    return errors;
2762

    
2763
fail:
2764
    fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
2765
    qemu_free(l1_table);
2766
    return -EIO;
2767
}
2768

    
2769
/*
2770
 * Checks an image for refcount consistency.
2771
 *
2772
 * Returns 0 if no errors are found, the number of errors in case the image is
2773
 * detected as corrupted, and -errno when an internal error occured.
2774
 */
2775
static int check_refcounts(BlockDriverState *bs)
2776
{
2777
    BDRVQcowState *s = bs->opaque;
2778
    int64_t size;
2779
    int nb_clusters, refcount1, refcount2, i;
2780
    QCowSnapshot *sn;
2781
    uint16_t *refcount_table;
2782
    int ret, errors = 0;
2783

    
2784
    size = bdrv_getlength(s->hd);
2785
    nb_clusters = size_to_clusters(s, size);
2786
    refcount_table = qemu_mallocz(nb_clusters * sizeof(uint16_t));
2787

    
2788
    /* header */
2789
    errors += inc_refcounts(bs, refcount_table, nb_clusters,
2790
                  0, s->cluster_size);
2791

    
2792
    /* current L1 table */
2793
    ret = check_refcounts_l1(bs, refcount_table, nb_clusters,
2794
                       s->l1_table_offset, s->l1_size, 1);
2795
    if (ret < 0) {
2796
        return ret;
2797
    }
2798
    errors += ret;
2799

    
2800
    /* snapshots */
2801
    for(i = 0; i < s->nb_snapshots; i++) {
2802
        sn = s->snapshots + i;
2803
        check_refcounts_l1(bs, refcount_table, nb_clusters,
2804
                           sn->l1_table_offset, sn->l1_size, 0);
2805
    }
2806
    errors += inc_refcounts(bs, refcount_table, nb_clusters,
2807
                  s->snapshots_offset, s->snapshots_size);
2808

    
2809
    /* refcount data */
2810
    errors += inc_refcounts(bs, refcount_table, nb_clusters,
2811
                  s->refcount_table_offset,
2812
                  s->refcount_table_size * sizeof(uint64_t));
2813
    for(i = 0; i < s->refcount_table_size; i++) {
2814
        int64_t offset;
2815
        offset = s->refcount_table[i];
2816
        if (offset != 0) {
2817
            errors += inc_refcounts(bs, refcount_table, nb_clusters,
2818
                          offset, s->cluster_size);
2819
        }
2820
    }
2821

    
2822
    /* compare ref counts */
2823
    for(i = 0; i < nb_clusters; i++) {
2824
        refcount1 = get_refcount(bs, i);
2825
        refcount2 = refcount_table[i];
2826
        if (refcount1 != refcount2) {
2827
            fprintf(stderr, "ERROR cluster %d refcount=%d reference=%d\n",
2828
                   i, refcount1, refcount2);
2829
            errors++;
2830
        }
2831
    }
2832

    
2833
    qemu_free(refcount_table);
2834

    
2835
    return errors;
2836
}
2837

    
2838
static int qcow_check(BlockDriverState *bs)
2839
{
2840
    return check_refcounts(bs);
2841
}
2842

    
2843
#if 0
2844
static void dump_refcounts(BlockDriverState *bs)
2845
{
2846
    BDRVQcowState *s = bs->opaque;
2847
    int64_t nb_clusters, k, k1, size;
2848
    int refcount;
2849

2850
    size = bdrv_getlength(s->hd);
2851
    nb_clusters = size_to_clusters(s, size);
2852
    for(k = 0; k < nb_clusters;) {
2853
        k1 = k;
2854
        refcount = get_refcount(bs, k);
2855
        k++;
2856
        while (k < nb_clusters && get_refcount(bs, k) == refcount)
2857
            k++;
2858
        printf("%lld: refcount=%d nb=%lld\n", k, refcount, k - k1);
2859
    }
2860
}
2861
#endif
2862

    
2863
static int qcow_put_buffer(BlockDriverState *bs, const uint8_t *buf,
2864
                           int64_t pos, int size)
2865
{
2866
    int growable = bs->growable;
2867

    
2868
    bs->growable = 1;
2869
    bdrv_pwrite(bs, pos, buf, size);
2870
    bs->growable = growable;
2871

    
2872
    return size;
2873
}
2874

    
2875
static int qcow_get_buffer(BlockDriverState *bs, uint8_t *buf,
2876
                           int64_t pos, int size)
2877
{
2878
    int growable = bs->growable;
2879
    int ret;
2880

    
2881
    bs->growable = 1;
2882
    ret = bdrv_pread(bs, pos, buf, size);
2883
    bs->growable = growable;
2884

    
2885
    return ret;
2886
}
2887

    
2888
BlockDriver bdrv_qcow2 = {
2889
    .format_name        = "qcow2",
2890
    .instance_size        = sizeof(BDRVQcowState),
2891
    .bdrv_probe                = qcow_probe,
2892
    .bdrv_open                = qcow_open,
2893
    .bdrv_close                = qcow_close,
2894
    .bdrv_create        = qcow_create,
2895
    .bdrv_flush                = qcow_flush,
2896
    .bdrv_is_allocated        = qcow_is_allocated,
2897
    .bdrv_set_key        = qcow_set_key,
2898
    .bdrv_make_empty        = qcow_make_empty,
2899

    
2900
    .bdrv_aio_readv        = qcow_aio_readv,
2901
    .bdrv_aio_writev        = qcow_aio_writev,
2902
    .bdrv_aio_cancel        = qcow_aio_cancel,
2903
    .aiocb_size                = sizeof(QCowAIOCB),
2904
    .bdrv_write_compressed = qcow_write_compressed,
2905

    
2906
    .bdrv_snapshot_create = qcow_snapshot_create,
2907
    .bdrv_snapshot_goto        = qcow_snapshot_goto,
2908
    .bdrv_snapshot_delete = qcow_snapshot_delete,
2909
    .bdrv_snapshot_list        = qcow_snapshot_list,
2910
    .bdrv_get_info        = qcow_get_info,
2911

    
2912
    .bdrv_put_buffer    = qcow_put_buffer,
2913
    .bdrv_get_buffer    = qcow_get_buffer,
2914

    
2915
    .bdrv_create2 = qcow_create2,
2916
    .bdrv_check = qcow_check,
2917
};