Statistics
| Branch: | Revision:

root / block-qcow2.c @ 5efa9d5a

History | View | Annotate | Download (89.4 kB)

1
/*
2
 * Block driver for the QCOW version 2 format
3
 *
4
 * Copyright (c) 2004-2006 Fabrice Bellard
5
 *
6
 * Permission is hereby granted, free of charge, to any person obtaining a copy
7
 * of this software and associated documentation files (the "Software"), to deal
8
 * in the Software without restriction, including without limitation the rights
9
 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10
 * copies of the Software, and to permit persons to whom the Software is
11
 * furnished to do so, subject to the following conditions:
12
 *
13
 * The above copyright notice and this permission notice shall be included in
14
 * all copies or substantial portions of the Software.
15
 *
16
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21
 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22
 * THE SOFTWARE.
23
 */
24
#include "qemu-common.h"
25
#include "block_int.h"
26
#include "module.h"
27
#include <zlib.h>
28
#include "aes.h"
29

    
30
/*
31
  Differences with QCOW:
32

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

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

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

    
53
#define QCOW_CRYPT_NONE 0
54
#define QCOW_CRYPT_AES  1
55

    
56
#define QCOW_MAX_CRYPT_CLUSTERS 32
57

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

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

    
65
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 two concurrent writes happen to the same unallocated cluster
996
         * each write allocates separate cluster and writes data concurrently.
997
         * The first one to complete updates l2 table with pointer to its
998
         * cluster the second one has to do RMW (which is done above by
999
         * copy_sectors()), update l2 table with its cluster pointer and free
1000
         * old cluster. This is what this loop does */
1001
        if(l2_table[l2_index + i] != 0)
1002
            old_cluster[j++] = l2_table[l2_index + i];
1003

    
1004
        l2_table[l2_index + i] = cpu_to_be64((cluster_offset +
1005
                    (i << s->cluster_bits)) | QCOW_OFLAG_COPIED);
1006
     }
1007

    
1008
    if (bdrv_pwrite(s->hd, l2_offset + l2_index * sizeof(uint64_t),
1009
                l2_table + l2_index, m->nb_clusters * sizeof(uint64_t)) !=
1010
            m->nb_clusters * sizeof(uint64_t))
1011
        goto err;
1012

    
1013
    for (i = 0; i < j; i++)
1014
        free_any_clusters(bs, be64_to_cpu(old_cluster[i]) & ~QCOW_OFLAG_COPIED,
1015
                          1);
1016

    
1017
    ret = 0;
1018
err:
1019
    qemu_free(old_cluster);
1020
    return ret;
1021
 }
1022

    
1023
/*
1024
 * alloc_cluster_offset
1025
 *
1026
 * For a given offset of the disk image, return cluster offset in
1027
 * qcow2 file.
1028
 *
1029
 * If the offset is not found, allocate a new cluster.
1030
 *
1031
 * Return the cluster offset if successful,
1032
 * Return 0, otherwise.
1033
 *
1034
 */
1035

    
1036
static uint64_t alloc_cluster_offset(BlockDriverState *bs,
1037
                                     uint64_t offset,
1038
                                     int n_start, int n_end,
1039
                                     int *num, QCowL2Meta *m)
1040
{
1041
    BDRVQcowState *s = bs->opaque;
1042
    int l2_index, ret;
1043
    uint64_t l2_offset, *l2_table, cluster_offset;
1044
    int nb_clusters, i = 0;
1045

    
1046
    ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
1047
    if (ret == 0)
1048
        return 0;
1049

    
1050
    nb_clusters = size_to_clusters(s, n_end << 9);
1051

    
1052
    nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
1053

    
1054
    cluster_offset = be64_to_cpu(l2_table[l2_index]);
1055

    
1056
    /* We keep all QCOW_OFLAG_COPIED clusters */
1057

    
1058
    if (cluster_offset & QCOW_OFLAG_COPIED) {
1059
        nb_clusters = count_contiguous_clusters(nb_clusters, s->cluster_size,
1060
                &l2_table[l2_index], 0, 0);
1061

    
1062
        cluster_offset &= ~QCOW_OFLAG_COPIED;
1063
        m->nb_clusters = 0;
1064

    
1065
        goto out;
1066
    }
1067

    
1068
    /* for the moment, multiple compressed clusters are not managed */
1069

    
1070
    if (cluster_offset & QCOW_OFLAG_COMPRESSED)
1071
        nb_clusters = 1;
1072

    
1073
    /* how many available clusters ? */
1074

    
1075
    while (i < nb_clusters) {
1076
        i += count_contiguous_clusters(nb_clusters - i, s->cluster_size,
1077
                &l2_table[l2_index], i, 0);
1078

    
1079
        if(be64_to_cpu(l2_table[l2_index + i]))
1080
            break;
1081

    
1082
        i += count_contiguous_free_clusters(nb_clusters - i,
1083
                &l2_table[l2_index + i]);
1084

    
1085
        cluster_offset = be64_to_cpu(l2_table[l2_index + i]);
1086

    
1087
        if ((cluster_offset & QCOW_OFLAG_COPIED) ||
1088
                (cluster_offset & QCOW_OFLAG_COMPRESSED))
1089
            break;
1090
    }
1091
    nb_clusters = i;
1092

    
1093
    /* allocate a new cluster */
1094

    
1095
    cluster_offset = alloc_clusters(bs, nb_clusters * s->cluster_size);
1096

    
1097
    /* save info needed for meta data update */
1098
    m->offset = offset;
1099
    m->n_start = n_start;
1100
    m->nb_clusters = nb_clusters;
1101

    
1102
out:
1103
    m->nb_available = MIN(nb_clusters << (s->cluster_bits - 9), n_end);
1104

    
1105
    *num = m->nb_available - n_start;
1106

    
1107
    return cluster_offset;
1108
}
1109

    
1110
static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
1111
                             int nb_sectors, int *pnum)
1112
{
1113
    uint64_t cluster_offset;
1114

    
1115
    *pnum = nb_sectors;
1116
    cluster_offset = get_cluster_offset(bs, sector_num << 9, pnum);
1117

    
1118
    return (cluster_offset != 0);
1119
}
1120

    
1121
static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
1122
                             const uint8_t *buf, int buf_size)
1123
{
1124
    z_stream strm1, *strm = &strm1;
1125
    int ret, out_len;
1126

    
1127
    memset(strm, 0, sizeof(*strm));
1128

    
1129
    strm->next_in = (uint8_t *)buf;
1130
    strm->avail_in = buf_size;
1131
    strm->next_out = out_buf;
1132
    strm->avail_out = out_buf_size;
1133

    
1134
    ret = inflateInit2(strm, -12);
1135
    if (ret != Z_OK)
1136
        return -1;
1137
    ret = inflate(strm, Z_FINISH);
1138
    out_len = strm->next_out - out_buf;
1139
    if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
1140
        out_len != out_buf_size) {
1141
        inflateEnd(strm);
1142
        return -1;
1143
    }
1144
    inflateEnd(strm);
1145
    return 0;
1146
}
1147

    
1148
static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset)
1149
{
1150
    int ret, csize, nb_csectors, sector_offset;
1151
    uint64_t coffset;
1152

    
1153
    coffset = cluster_offset & s->cluster_offset_mask;
1154
    if (s->cluster_cache_offset != coffset) {
1155
        nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
1156
        sector_offset = coffset & 511;
1157
        csize = nb_csectors * 512 - sector_offset;
1158
        ret = bdrv_read(s->hd, coffset >> 9, s->cluster_data, nb_csectors);
1159
        if (ret < 0) {
1160
            return -1;
1161
        }
1162
        if (decompress_buffer(s->cluster_cache, s->cluster_size,
1163
                              s->cluster_data + sector_offset, csize) < 0) {
1164
            return -1;
1165
        }
1166
        s->cluster_cache_offset = coffset;
1167
    }
1168
    return 0;
1169
}
1170

    
1171
/* handle reading after the end of the backing file */
1172
static int backing_read1(BlockDriverState *bs,
1173
                         int64_t sector_num, uint8_t *buf, int nb_sectors)
1174
{
1175
    int n1;
1176
    if ((sector_num + nb_sectors) <= bs->total_sectors)
1177
        return nb_sectors;
1178
    if (sector_num >= bs->total_sectors)
1179
        n1 = 0;
1180
    else
1181
        n1 = bs->total_sectors - sector_num;
1182
    memset(buf + n1 * 512, 0, 512 * (nb_sectors - n1));
1183
    return n1;
1184
}
1185

    
1186
static int qcow_read(BlockDriverState *bs, int64_t sector_num,
1187
                     uint8_t *buf, int nb_sectors)
1188
{
1189
    BDRVQcowState *s = bs->opaque;
1190
    int ret, index_in_cluster, n, n1;
1191
    uint64_t cluster_offset;
1192

    
1193
    while (nb_sectors > 0) {
1194
        n = nb_sectors;
1195
        cluster_offset = get_cluster_offset(bs, sector_num << 9, &n);
1196
        index_in_cluster = sector_num & (s->cluster_sectors - 1);
1197
        if (!cluster_offset) {
1198
            if (bs->backing_hd) {
1199
                /* read from the base image */
1200
                n1 = backing_read1(bs->backing_hd, sector_num, buf, n);
1201
                if (n1 > 0) {
1202
                    ret = bdrv_read(bs->backing_hd, sector_num, buf, n1);
1203
                    if (ret < 0)
1204
                        return -1;
1205
                }
1206
            } else {
1207
                memset(buf, 0, 512 * n);
1208
            }
1209
        } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
1210
            if (decompress_cluster(s, cluster_offset) < 0)
1211
                return -1;
1212
            memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n);
1213
        } else {
1214
            ret = bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
1215
            if (ret != n * 512)
1216
                return -1;
1217
            if (s->crypt_method) {
1218
                encrypt_sectors(s, sector_num, buf, buf, n, 0,
1219
                                &s->aes_decrypt_key);
1220
            }
1221
        }
1222
        nb_sectors -= n;
1223
        sector_num += n;
1224
        buf += n * 512;
1225
    }
1226
    return 0;
1227
}
1228

    
1229
static int qcow_write(BlockDriverState *bs, int64_t sector_num,
1230
                     const uint8_t *buf, int nb_sectors)
1231
{
1232
    BDRVQcowState *s = bs->opaque;
1233
    int ret, index_in_cluster, n;
1234
    uint64_t cluster_offset;
1235
    int n_end;
1236
    QCowL2Meta l2meta;
1237

    
1238
    while (nb_sectors > 0) {
1239
        index_in_cluster = sector_num & (s->cluster_sectors - 1);
1240
        n_end = index_in_cluster + nb_sectors;
1241
        if (s->crypt_method &&
1242
            n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
1243
            n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
1244
        cluster_offset = alloc_cluster_offset(bs, sector_num << 9,
1245
                                              index_in_cluster,
1246
                                              n_end, &n, &l2meta);
1247
        if (!cluster_offset)
1248
            return -1;
1249
        if (s->crypt_method) {
1250
            encrypt_sectors(s, sector_num, s->cluster_data, buf, n, 1,
1251
                            &s->aes_encrypt_key);
1252
            ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512,
1253
                              s->cluster_data, n * 512);
1254
        } else {
1255
            ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
1256
        }
1257
        if (ret != n * 512 || alloc_cluster_link_l2(bs, cluster_offset, &l2meta) < 0) {
1258
            free_any_clusters(bs, cluster_offset, l2meta.nb_clusters);
1259
            return -1;
1260
        }
1261
        nb_sectors -= n;
1262
        sector_num += n;
1263
        buf += n * 512;
1264
    }
1265
    s->cluster_cache_offset = -1; /* disable compressed cache */
1266
    return 0;
1267
}
1268

    
1269
typedef struct QCowAIOCB {
1270
    BlockDriverAIOCB common;
1271
    int64_t sector_num;
1272
    QEMUIOVector *qiov;
1273
    uint8_t *buf;
1274
    void *orig_buf;
1275
    int nb_sectors;
1276
    int n;
1277
    uint64_t cluster_offset;
1278
    uint8_t *cluster_data;
1279
    BlockDriverAIOCB *hd_aiocb;
1280
    struct iovec hd_iov;
1281
    QEMUIOVector hd_qiov;
1282
    QEMUBH *bh;
1283
    QCowL2Meta l2meta;
1284
} QCowAIOCB;
1285

    
1286
static void qcow_aio_read_cb(void *opaque, int ret);
1287
static void qcow_aio_read_bh(void *opaque)
1288
{
1289
    QCowAIOCB *acb = opaque;
1290
    qemu_bh_delete(acb->bh);
1291
    acb->bh = NULL;
1292
    qcow_aio_read_cb(opaque, 0);
1293
}
1294

    
1295
static int qcow_schedule_bh(QEMUBHFunc *cb, QCowAIOCB *acb)
1296
{
1297
    if (acb->bh)
1298
        return -EIO;
1299

    
1300
    acb->bh = qemu_bh_new(cb, acb);
1301
    if (!acb->bh)
1302
        return -EIO;
1303

    
1304
    qemu_bh_schedule(acb->bh);
1305

    
1306
    return 0;
1307
}
1308

    
1309
static void qcow_aio_read_cb(void *opaque, int ret)
1310
{
1311
    QCowAIOCB *acb = opaque;
1312
    BlockDriverState *bs = acb->common.bs;
1313
    BDRVQcowState *s = bs->opaque;
1314
    int index_in_cluster, n1;
1315

    
1316
    acb->hd_aiocb = NULL;
1317
    if (ret < 0)
1318
        goto done;
1319

    
1320
    /* post process the read buffer */
1321
    if (!acb->cluster_offset) {
1322
        /* nothing to do */
1323
    } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
1324
        /* nothing to do */
1325
    } else {
1326
        if (s->crypt_method) {
1327
            encrypt_sectors(s, acb->sector_num, acb->buf, acb->buf,
1328
                            acb->n, 0,
1329
                            &s->aes_decrypt_key);
1330
        }
1331
    }
1332

    
1333
    acb->nb_sectors -= acb->n;
1334
    acb->sector_num += acb->n;
1335
    acb->buf += acb->n * 512;
1336

    
1337
    if (acb->nb_sectors == 0) {
1338
        /* request completed */
1339
        ret = 0;
1340
        goto done;
1341
    }
1342

    
1343
    /* prepare next AIO request */
1344
    acb->n = acb->nb_sectors;
1345
    acb->cluster_offset = get_cluster_offset(bs, acb->sector_num << 9, &acb->n);
1346
    index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
1347

    
1348
    if (!acb->cluster_offset) {
1349
        if (bs->backing_hd) {
1350
            /* read from the base image */
1351
            n1 = backing_read1(bs->backing_hd, acb->sector_num,
1352
                               acb->buf, acb->n);
1353
            if (n1 > 0) {
1354
                acb->hd_iov.iov_base = (void *)acb->buf;
1355
                acb->hd_iov.iov_len = acb->n * 512;
1356
                qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
1357
                acb->hd_aiocb = bdrv_aio_readv(bs->backing_hd, acb->sector_num,
1358
                                    &acb->hd_qiov, acb->n,
1359
                                    qcow_aio_read_cb, acb);
1360
                if (acb->hd_aiocb == NULL)
1361
                    goto done;
1362
            } else {
1363
                ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
1364
                if (ret < 0)
1365
                    goto done;
1366
            }
1367
        } else {
1368
            /* Note: in this case, no need to wait */
1369
            memset(acb->buf, 0, 512 * acb->n);
1370
            ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
1371
            if (ret < 0)
1372
                goto done;
1373
        }
1374
    } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
1375
        /* add AIO support for compressed blocks ? */
1376
        if (decompress_cluster(s, acb->cluster_offset) < 0)
1377
            goto done;
1378
        memcpy(acb->buf,
1379
               s->cluster_cache + index_in_cluster * 512, 512 * acb->n);
1380
        ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
1381
        if (ret < 0)
1382
            goto done;
1383
    } else {
1384
        if ((acb->cluster_offset & 511) != 0) {
1385
            ret = -EIO;
1386
            goto done;
1387
        }
1388

    
1389
        acb->hd_iov.iov_base = (void *)acb->buf;
1390
        acb->hd_iov.iov_len = acb->n * 512;
1391
        qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
1392
        acb->hd_aiocb = bdrv_aio_readv(s->hd,
1393
                            (acb->cluster_offset >> 9) + index_in_cluster,
1394
                            &acb->hd_qiov, acb->n, qcow_aio_read_cb, acb);
1395
        if (acb->hd_aiocb == NULL)
1396
            goto done;
1397
    }
1398

    
1399
    return;
1400
done:
1401
    if (acb->qiov->niov > 1) {
1402
        qemu_iovec_from_buffer(acb->qiov, acb->orig_buf, acb->qiov->size);
1403
        qemu_vfree(acb->orig_buf);
1404
    }
1405
    acb->common.cb(acb->common.opaque, ret);
1406
    qemu_aio_release(acb);
1407
}
1408

    
1409
static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs,
1410
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1411
        BlockDriverCompletionFunc *cb, void *opaque, int is_write)
1412
{
1413
    QCowAIOCB *acb;
1414

    
1415
    acb = qemu_aio_get(bs, cb, opaque);
1416
    if (!acb)
1417
        return NULL;
1418
    acb->hd_aiocb = NULL;
1419
    acb->sector_num = sector_num;
1420
    acb->qiov = qiov;
1421
    if (qiov->niov > 1) {
1422
        acb->buf = acb->orig_buf = qemu_blockalign(bs, qiov->size);
1423
        if (is_write)
1424
            qemu_iovec_to_buffer(qiov, acb->buf);
1425
    } else {
1426
        acb->buf = (uint8_t *)qiov->iov->iov_base;
1427
    }
1428
    acb->nb_sectors = nb_sectors;
1429
    acb->n = 0;
1430
    acb->cluster_offset = 0;
1431
    acb->l2meta.nb_clusters = 0;
1432
    return acb;
1433
}
1434

    
1435
static BlockDriverAIOCB *qcow_aio_readv(BlockDriverState *bs,
1436
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1437
        BlockDriverCompletionFunc *cb, void *opaque)
1438
{
1439
    QCowAIOCB *acb;
1440

    
1441
    acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
1442
    if (!acb)
1443
        return NULL;
1444

    
1445
    qcow_aio_read_cb(acb, 0);
1446
    return &acb->common;
1447
}
1448

    
1449
static void qcow_aio_write_cb(void *opaque, int ret)
1450
{
1451
    QCowAIOCB *acb = opaque;
1452
    BlockDriverState *bs = acb->common.bs;
1453
    BDRVQcowState *s = bs->opaque;
1454
    int index_in_cluster;
1455
    const uint8_t *src_buf;
1456
    int n_end;
1457

    
1458
    acb->hd_aiocb = NULL;
1459

    
1460
    if (ret < 0)
1461
        goto done;
1462

    
1463
    if (alloc_cluster_link_l2(bs, acb->cluster_offset, &acb->l2meta) < 0) {
1464
        free_any_clusters(bs, acb->cluster_offset, acb->l2meta.nb_clusters);
1465
        goto done;
1466
    }
1467

    
1468
    acb->nb_sectors -= acb->n;
1469
    acb->sector_num += acb->n;
1470
    acb->buf += acb->n * 512;
1471

    
1472
    if (acb->nb_sectors == 0) {
1473
        /* request completed */
1474
        ret = 0;
1475
        goto done;
1476
    }
1477

    
1478
    index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
1479
    n_end = index_in_cluster + acb->nb_sectors;
1480
    if (s->crypt_method &&
1481
        n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
1482
        n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
1483

    
1484
    acb->cluster_offset = alloc_cluster_offset(bs, acb->sector_num << 9,
1485
                                          index_in_cluster,
1486
                                          n_end, &acb->n, &acb->l2meta);
1487
    if (!acb->cluster_offset || (acb->cluster_offset & 511) != 0) {
1488
        ret = -EIO;
1489
        goto done;
1490
    }
1491
    if (s->crypt_method) {
1492
        if (!acb->cluster_data) {
1493
            acb->cluster_data = qemu_mallocz(QCOW_MAX_CRYPT_CLUSTERS *
1494
                                             s->cluster_size);
1495
        }
1496
        encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf,
1497
                        acb->n, 1, &s->aes_encrypt_key);
1498
        src_buf = acb->cluster_data;
1499
    } else {
1500
        src_buf = acb->buf;
1501
    }
1502
    acb->hd_iov.iov_base = (void *)src_buf;
1503
    acb->hd_iov.iov_len = acb->n * 512;
1504
    qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
1505
    acb->hd_aiocb = bdrv_aio_writev(s->hd,
1506
                                    (acb->cluster_offset >> 9) + index_in_cluster,
1507
                                    &acb->hd_qiov, acb->n,
1508
                                    qcow_aio_write_cb, acb);
1509
    if (acb->hd_aiocb == NULL)
1510
        goto done;
1511

    
1512
    return;
1513

    
1514
done:
1515
    if (acb->qiov->niov > 1)
1516
        qemu_vfree(acb->orig_buf);
1517
    acb->common.cb(acb->common.opaque, ret);
1518
    qemu_aio_release(acb);
1519
}
1520

    
1521
static BlockDriverAIOCB *qcow_aio_writev(BlockDriverState *bs,
1522
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1523
        BlockDriverCompletionFunc *cb, void *opaque)
1524
{
1525
    BDRVQcowState *s = bs->opaque;
1526
    QCowAIOCB *acb;
1527

    
1528
    s->cluster_cache_offset = -1; /* disable compressed cache */
1529

    
1530
    acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
1531
    if (!acb)
1532
        return NULL;
1533

    
1534
    qcow_aio_write_cb(acb, 0);
1535
    return &acb->common;
1536
}
1537

    
1538
static void qcow_aio_cancel(BlockDriverAIOCB *blockacb)
1539
{
1540
    QCowAIOCB *acb = (QCowAIOCB *)blockacb;
1541
    if (acb->hd_aiocb)
1542
        bdrv_aio_cancel(acb->hd_aiocb);
1543
    qemu_aio_release(acb);
1544
}
1545

    
1546
static void qcow_close(BlockDriverState *bs)
1547
{
1548
    BDRVQcowState *s = bs->opaque;
1549
    qemu_free(s->l1_table);
1550
    qemu_free(s->l2_cache);
1551
    qemu_free(s->cluster_cache);
1552
    qemu_free(s->cluster_data);
1553
    refcount_close(bs);
1554
    bdrv_delete(s->hd);
1555
}
1556

    
1557
/* XXX: use std qcow open function ? */
1558
typedef struct QCowCreateState {
1559
    int cluster_size;
1560
    int cluster_bits;
1561
    uint16_t *refcount_block;
1562
    uint64_t *refcount_table;
1563
    int64_t l1_table_offset;
1564
    int64_t refcount_table_offset;
1565
    int64_t refcount_block_offset;
1566
} QCowCreateState;
1567

    
1568
static void create_refcount_update(QCowCreateState *s,
1569
                                   int64_t offset, int64_t size)
1570
{
1571
    int refcount;
1572
    int64_t start, last, cluster_offset;
1573
    uint16_t *p;
1574

    
1575
    start = offset & ~(s->cluster_size - 1);
1576
    last = (offset + size - 1)  & ~(s->cluster_size - 1);
1577
    for(cluster_offset = start; cluster_offset <= last;
1578
        cluster_offset += s->cluster_size) {
1579
        p = &s->refcount_block[cluster_offset >> s->cluster_bits];
1580
        refcount = be16_to_cpu(*p);
1581
        refcount++;
1582
        *p = cpu_to_be16(refcount);
1583
    }
1584
}
1585

    
1586
static int qcow_create2(const char *filename, int64_t total_size,
1587
                        const char *backing_file, const char *backing_format,
1588
                        int flags)
1589
{
1590

    
1591
    int fd, header_size, backing_filename_len, l1_size, i, shift, l2_bits;
1592
    int ref_clusters, backing_format_len = 0;
1593
    QCowHeader header;
1594
    uint64_t tmp, offset;
1595
    QCowCreateState s1, *s = &s1;
1596
    QCowExtension ext_bf = {0, 0};
1597

    
1598

    
1599
    memset(s, 0, sizeof(*s));
1600

    
1601
    fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
1602
    if (fd < 0)
1603
        return -1;
1604
    memset(&header, 0, sizeof(header));
1605
    header.magic = cpu_to_be32(QCOW_MAGIC);
1606
    header.version = cpu_to_be32(QCOW_VERSION);
1607
    header.size = cpu_to_be64(total_size * 512);
1608
    header_size = sizeof(header);
1609
    backing_filename_len = 0;
1610
    if (backing_file) {
1611
        if (backing_format) {
1612
            ext_bf.magic = QCOW_EXT_MAGIC_BACKING_FORMAT;
1613
            backing_format_len = strlen(backing_format);
1614
            ext_bf.len = (backing_format_len + 7) & ~7;
1615
            header_size += ((sizeof(ext_bf) + ext_bf.len + 7) & ~7);
1616
        }
1617
        header.backing_file_offset = cpu_to_be64(header_size);
1618
        backing_filename_len = strlen(backing_file);
1619
        header.backing_file_size = cpu_to_be32(backing_filename_len);
1620
        header_size += backing_filename_len;
1621
    }
1622
    s->cluster_bits = 12;  /* 4 KB clusters */
1623
    s->cluster_size = 1 << s->cluster_bits;
1624
    header.cluster_bits = cpu_to_be32(s->cluster_bits);
1625
    header_size = (header_size + 7) & ~7;
1626
    if (flags & BLOCK_FLAG_ENCRYPT) {
1627
        header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
1628
    } else {
1629
        header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
1630
    }
1631
    l2_bits = s->cluster_bits - 3;
1632
    shift = s->cluster_bits + l2_bits;
1633
    l1_size = (((total_size * 512) + (1LL << shift) - 1) >> shift);
1634
    offset = align_offset(header_size, s->cluster_size);
1635
    s->l1_table_offset = offset;
1636
    header.l1_table_offset = cpu_to_be64(s->l1_table_offset);
1637
    header.l1_size = cpu_to_be32(l1_size);
1638
    offset += align_offset(l1_size * sizeof(uint64_t), s->cluster_size);
1639

    
1640
    s->refcount_table = qemu_mallocz(s->cluster_size);
1641

    
1642
    s->refcount_table_offset = offset;
1643
    header.refcount_table_offset = cpu_to_be64(offset);
1644
    header.refcount_table_clusters = cpu_to_be32(1);
1645
    offset += s->cluster_size;
1646
    s->refcount_block_offset = offset;
1647

    
1648
    /* count how many refcount blocks needed */
1649
    tmp = offset >> s->cluster_bits;
1650
    ref_clusters = (tmp >> (s->cluster_bits - REFCOUNT_SHIFT)) + 1;
1651
    for (i=0; i < ref_clusters; i++) {
1652
        s->refcount_table[i] = cpu_to_be64(offset);
1653
        offset += s->cluster_size;
1654
    }
1655

    
1656
    s->refcount_block = qemu_mallocz(ref_clusters * s->cluster_size);
1657

    
1658
    /* update refcounts */
1659
    create_refcount_update(s, 0, header_size);
1660
    create_refcount_update(s, s->l1_table_offset, l1_size * sizeof(uint64_t));
1661
    create_refcount_update(s, s->refcount_table_offset, s->cluster_size);
1662
    create_refcount_update(s, s->refcount_block_offset, ref_clusters * s->cluster_size);
1663

    
1664
    /* write all the data */
1665
    write(fd, &header, sizeof(header));
1666
    if (backing_file) {
1667
        if (backing_format_len) {
1668
            char zero[16];
1669
            int d = ext_bf.len - backing_format_len;
1670

    
1671
            memset(zero, 0, sizeof(zero));
1672
            cpu_to_be32s(&ext_bf.magic);
1673
            cpu_to_be32s(&ext_bf.len);
1674
            write(fd, &ext_bf, sizeof(ext_bf));
1675
            write(fd, backing_format, backing_format_len);
1676
            if (d>0) {
1677
                write(fd, zero, d);
1678
            }
1679
        }
1680
        write(fd, backing_file, backing_filename_len);
1681
    }
1682
    lseek(fd, s->l1_table_offset, SEEK_SET);
1683
    tmp = 0;
1684
    for(i = 0;i < l1_size; i++) {
1685
        write(fd, &tmp, sizeof(tmp));
1686
    }
1687
    lseek(fd, s->refcount_table_offset, SEEK_SET);
1688
    write(fd, s->refcount_table, s->cluster_size);
1689

    
1690
    lseek(fd, s->refcount_block_offset, SEEK_SET);
1691
    write(fd, s->refcount_block, ref_clusters * s->cluster_size);
1692

    
1693
    qemu_free(s->refcount_table);
1694
    qemu_free(s->refcount_block);
1695
    close(fd);
1696
    return 0;
1697
}
1698

    
1699
static int qcow_create(const char *filename, int64_t total_size,
1700
                       const char *backing_file, int flags)
1701
{
1702
    return qcow_create2(filename, total_size, backing_file, NULL, flags);
1703
}
1704

    
1705
static int qcow_make_empty(BlockDriverState *bs)
1706
{
1707
#if 0
1708
    /* XXX: not correct */
1709
    BDRVQcowState *s = bs->opaque;
1710
    uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1711
    int ret;
1712

1713
    memset(s->l1_table, 0, l1_length);
1714
    if (bdrv_pwrite(s->hd, s->l1_table_offset, s->l1_table, l1_length) < 0)
1715
        return -1;
1716
    ret = bdrv_truncate(s->hd, s->l1_table_offset + l1_length);
1717
    if (ret < 0)
1718
        return ret;
1719

1720
    l2_cache_reset(bs);
1721
#endif
1722
    return 0;
1723
}
1724

    
1725
/* XXX: put compressed sectors first, then all the cluster aligned
1726
   tables to avoid losing bytes in alignment */
1727
static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
1728
                                 const uint8_t *buf, int nb_sectors)
1729
{
1730
    BDRVQcowState *s = bs->opaque;
1731
    z_stream strm;
1732
    int ret, out_len;
1733
    uint8_t *out_buf;
1734
    uint64_t cluster_offset;
1735

    
1736
    if (nb_sectors == 0) {
1737
        /* align end of file to a sector boundary to ease reading with
1738
           sector based I/Os */
1739
        cluster_offset = bdrv_getlength(s->hd);
1740
        cluster_offset = (cluster_offset + 511) & ~511;
1741
        bdrv_truncate(s->hd, cluster_offset);
1742
        return 0;
1743
    }
1744

    
1745
    if (nb_sectors != s->cluster_sectors)
1746
        return -EINVAL;
1747

    
1748
    out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
1749

    
1750
    /* best compression, small window, no zlib header */
1751
    memset(&strm, 0, sizeof(strm));
1752
    ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1753
                       Z_DEFLATED, -12,
1754
                       9, Z_DEFAULT_STRATEGY);
1755
    if (ret != 0) {
1756
        qemu_free(out_buf);
1757
        return -1;
1758
    }
1759

    
1760
    strm.avail_in = s->cluster_size;
1761
    strm.next_in = (uint8_t *)buf;
1762
    strm.avail_out = s->cluster_size;
1763
    strm.next_out = out_buf;
1764

    
1765
    ret = deflate(&strm, Z_FINISH);
1766
    if (ret != Z_STREAM_END && ret != Z_OK) {
1767
        qemu_free(out_buf);
1768
        deflateEnd(&strm);
1769
        return -1;
1770
    }
1771
    out_len = strm.next_out - out_buf;
1772

    
1773
    deflateEnd(&strm);
1774

    
1775
    if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1776
        /* could not compress: write normal cluster */
1777
        qcow_write(bs, sector_num, buf, s->cluster_sectors);
1778
    } else {
1779
        cluster_offset = alloc_compressed_cluster_offset(bs, sector_num << 9,
1780
                                              out_len);
1781
        if (!cluster_offset)
1782
            return -1;
1783
        cluster_offset &= s->cluster_offset_mask;
1784
        if (bdrv_pwrite(s->hd, cluster_offset, out_buf, out_len) != out_len) {
1785
            qemu_free(out_buf);
1786
            return -1;
1787
        }
1788
    }
1789

    
1790
    qemu_free(out_buf);
1791
    return 0;
1792
}
1793

    
1794
static void qcow_flush(BlockDriverState *bs)
1795
{
1796
    BDRVQcowState *s = bs->opaque;
1797
    bdrv_flush(s->hd);
1798
}
1799

    
1800
static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1801
{
1802
    BDRVQcowState *s = bs->opaque;
1803
    bdi->cluster_size = s->cluster_size;
1804
    bdi->vm_state_offset = (int64_t)s->l1_vm_state_index <<
1805
        (s->cluster_bits + s->l2_bits);
1806
    return 0;
1807
}
1808

    
1809
/*********************************************************/
1810
/* snapshot support */
1811

    
1812
/* update the refcounts of snapshots and the copied flag */
1813
static int update_snapshot_refcount(BlockDriverState *bs,
1814
                                    int64_t l1_table_offset,
1815
                                    int l1_size,
1816
                                    int addend)
1817
{
1818
    BDRVQcowState *s = bs->opaque;
1819
    uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated;
1820
    int64_t old_offset, old_l2_offset;
1821
    int l2_size, i, j, l1_modified, l2_modified, nb_csectors, refcount;
1822

    
1823
    l2_cache_reset(bs);
1824

    
1825
    l2_table = NULL;
1826
    l1_table = NULL;
1827
    l1_size2 = l1_size * sizeof(uint64_t);
1828
    l1_allocated = 0;
1829
    if (l1_table_offset != s->l1_table_offset) {
1830
        l1_table = qemu_malloc(l1_size2);
1831
        l1_allocated = 1;
1832
        if (bdrv_pread(s->hd, l1_table_offset,
1833
                       l1_table, l1_size2) != l1_size2)
1834
            goto fail;
1835
        for(i = 0;i < l1_size; i++)
1836
            be64_to_cpus(&l1_table[i]);
1837
    } else {
1838
        assert(l1_size == s->l1_size);
1839
        l1_table = s->l1_table;
1840
        l1_allocated = 0;
1841
    }
1842

    
1843
    l2_size = s->l2_size * sizeof(uint64_t);
1844
    l2_table = qemu_malloc(l2_size);
1845
    l1_modified = 0;
1846
    for(i = 0; i < l1_size; i++) {
1847
        l2_offset = l1_table[i];
1848
        if (l2_offset) {
1849
            old_l2_offset = l2_offset;
1850
            l2_offset &= ~QCOW_OFLAG_COPIED;
1851
            l2_modified = 0;
1852
            if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
1853
                goto fail;
1854
            for(j = 0; j < s->l2_size; j++) {
1855
                offset = be64_to_cpu(l2_table[j]);
1856
                if (offset != 0) {
1857
                    old_offset = offset;
1858
                    offset &= ~QCOW_OFLAG_COPIED;
1859
                    if (offset & QCOW_OFLAG_COMPRESSED) {
1860
                        nb_csectors = ((offset >> s->csize_shift) &
1861
                                       s->csize_mask) + 1;
1862
                        if (addend != 0)
1863
                            update_refcount(bs, (offset & s->cluster_offset_mask) & ~511,
1864
                                            nb_csectors * 512, addend);
1865
                        /* compressed clusters are never modified */
1866
                        refcount = 2;
1867
                    } else {
1868
                        if (addend != 0) {
1869
                            refcount = update_cluster_refcount(bs, offset >> s->cluster_bits, addend);
1870
                        } else {
1871
                            refcount = get_refcount(bs, offset >> s->cluster_bits);
1872
                        }
1873
                    }
1874

    
1875
                    if (refcount == 1) {
1876
                        offset |= QCOW_OFLAG_COPIED;
1877
                    }
1878
                    if (offset != old_offset) {
1879
                        l2_table[j] = cpu_to_be64(offset);
1880
                        l2_modified = 1;
1881
                    }
1882
                }
1883
            }
1884
            if (l2_modified) {
1885
                if (bdrv_pwrite(s->hd,
1886
                                l2_offset, l2_table, l2_size) != l2_size)
1887
                    goto fail;
1888
            }
1889

    
1890
            if (addend != 0) {
1891
                refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend);
1892
            } else {
1893
                refcount = get_refcount(bs, l2_offset >> s->cluster_bits);
1894
            }
1895
            if (refcount == 1) {
1896
                l2_offset |= QCOW_OFLAG_COPIED;
1897
            }
1898
            if (l2_offset != old_l2_offset) {
1899
                l1_table[i] = l2_offset;
1900
                l1_modified = 1;
1901
            }
1902
        }
1903
    }
1904
    if (l1_modified) {
1905
        for(i = 0; i < l1_size; i++)
1906
            cpu_to_be64s(&l1_table[i]);
1907
        if (bdrv_pwrite(s->hd, l1_table_offset, l1_table,
1908
                        l1_size2) != l1_size2)
1909
            goto fail;
1910
        for(i = 0; i < l1_size; i++)
1911
            be64_to_cpus(&l1_table[i]);
1912
    }
1913
    if (l1_allocated)
1914
        qemu_free(l1_table);
1915
    qemu_free(l2_table);
1916
    return 0;
1917
 fail:
1918
    if (l1_allocated)
1919
        qemu_free(l1_table);
1920
    qemu_free(l2_table);
1921
    return -EIO;
1922
}
1923

    
1924
static void qcow_free_snapshots(BlockDriverState *bs)
1925
{
1926
    BDRVQcowState *s = bs->opaque;
1927
    int i;
1928

    
1929
    for(i = 0; i < s->nb_snapshots; i++) {
1930
        qemu_free(s->snapshots[i].name);
1931
        qemu_free(s->snapshots[i].id_str);
1932
    }
1933
    qemu_free(s->snapshots);
1934
    s->snapshots = NULL;
1935
    s->nb_snapshots = 0;
1936
}
1937

    
1938
static int qcow_read_snapshots(BlockDriverState *bs)
1939
{
1940
    BDRVQcowState *s = bs->opaque;
1941
    QCowSnapshotHeader h;
1942
    QCowSnapshot *sn;
1943
    int i, id_str_size, name_size;
1944
    int64_t offset;
1945
    uint32_t extra_data_size;
1946

    
1947
    if (!s->nb_snapshots) {
1948
        s->snapshots = NULL;
1949
        s->snapshots_size = 0;
1950
        return 0;
1951
    }
1952

    
1953
    offset = s->snapshots_offset;
1954
    s->snapshots = qemu_mallocz(s->nb_snapshots * sizeof(QCowSnapshot));
1955
    for(i = 0; i < s->nb_snapshots; i++) {
1956
        offset = align_offset(offset, 8);
1957
        if (bdrv_pread(s->hd, offset, &h, sizeof(h)) != sizeof(h))
1958
            goto fail;
1959
        offset += sizeof(h);
1960
        sn = s->snapshots + i;
1961
        sn->l1_table_offset = be64_to_cpu(h.l1_table_offset);
1962
        sn->l1_size = be32_to_cpu(h.l1_size);
1963
        sn->vm_state_size = be32_to_cpu(h.vm_state_size);
1964
        sn->date_sec = be32_to_cpu(h.date_sec);
1965
        sn->date_nsec = be32_to_cpu(h.date_nsec);
1966
        sn->vm_clock_nsec = be64_to_cpu(h.vm_clock_nsec);
1967
        extra_data_size = be32_to_cpu(h.extra_data_size);
1968

    
1969
        id_str_size = be16_to_cpu(h.id_str_size);
1970
        name_size = be16_to_cpu(h.name_size);
1971

    
1972
        offset += extra_data_size;
1973

    
1974
        sn->id_str = qemu_malloc(id_str_size + 1);
1975
        if (bdrv_pread(s->hd, offset, sn->id_str, id_str_size) != id_str_size)
1976
            goto fail;
1977
        offset += id_str_size;
1978
        sn->id_str[id_str_size] = '\0';
1979

    
1980
        sn->name = qemu_malloc(name_size + 1);
1981
        if (bdrv_pread(s->hd, offset, sn->name, name_size) != name_size)
1982
            goto fail;
1983
        offset += name_size;
1984
        sn->name[name_size] = '\0';
1985
    }
1986
    s->snapshots_size = offset - s->snapshots_offset;
1987
    return 0;
1988
 fail:
1989
    qcow_free_snapshots(bs);
1990
    return -1;
1991
}
1992

    
1993
/* add at the end of the file a new list of snapshots */
1994
static int qcow_write_snapshots(BlockDriverState *bs)
1995
{
1996
    BDRVQcowState *s = bs->opaque;
1997
    QCowSnapshot *sn;
1998
    QCowSnapshotHeader h;
1999
    int i, name_size, id_str_size, snapshots_size;
2000
    uint64_t data64;
2001
    uint32_t data32;
2002
    int64_t offset, snapshots_offset;
2003

    
2004
    /* compute the size of the snapshots */
2005
    offset = 0;
2006
    for(i = 0; i < s->nb_snapshots; i++) {
2007
        sn = s->snapshots + i;
2008
        offset = align_offset(offset, 8);
2009
        offset += sizeof(h);
2010
        offset += strlen(sn->id_str);
2011
        offset += strlen(sn->name);
2012
    }
2013
    snapshots_size = offset;
2014

    
2015
    snapshots_offset = alloc_clusters(bs, snapshots_size);
2016
    offset = snapshots_offset;
2017

    
2018
    for(i = 0; i < s->nb_snapshots; i++) {
2019
        sn = s->snapshots + i;
2020
        memset(&h, 0, sizeof(h));
2021
        h.l1_table_offset = cpu_to_be64(sn->l1_table_offset);
2022
        h.l1_size = cpu_to_be32(sn->l1_size);
2023
        h.vm_state_size = cpu_to_be32(sn->vm_state_size);
2024
        h.date_sec = cpu_to_be32(sn->date_sec);
2025
        h.date_nsec = cpu_to_be32(sn->date_nsec);
2026
        h.vm_clock_nsec = cpu_to_be64(sn->vm_clock_nsec);
2027

    
2028
        id_str_size = strlen(sn->id_str);
2029
        name_size = strlen(sn->name);
2030
        h.id_str_size = cpu_to_be16(id_str_size);
2031
        h.name_size = cpu_to_be16(name_size);
2032
        offset = align_offset(offset, 8);
2033
        if (bdrv_pwrite(s->hd, offset, &h, sizeof(h)) != sizeof(h))
2034
            goto fail;
2035
        offset += sizeof(h);
2036
        if (bdrv_pwrite(s->hd, offset, sn->id_str, id_str_size) != id_str_size)
2037
            goto fail;
2038
        offset += id_str_size;
2039
        if (bdrv_pwrite(s->hd, offset, sn->name, name_size) != name_size)
2040
            goto fail;
2041
        offset += name_size;
2042
    }
2043

    
2044
    /* update the various header fields */
2045
    data64 = cpu_to_be64(snapshots_offset);
2046
    if (bdrv_pwrite(s->hd, offsetof(QCowHeader, snapshots_offset),
2047
                    &data64, sizeof(data64)) != sizeof(data64))
2048
        goto fail;
2049
    data32 = cpu_to_be32(s->nb_snapshots);
2050
    if (bdrv_pwrite(s->hd, offsetof(QCowHeader, nb_snapshots),
2051
                    &data32, sizeof(data32)) != sizeof(data32))
2052
        goto fail;
2053

    
2054
    /* free the old snapshot table */
2055
    free_clusters(bs, s->snapshots_offset, s->snapshots_size);
2056
    s->snapshots_offset = snapshots_offset;
2057
    s->snapshots_size = snapshots_size;
2058
    return 0;
2059
 fail:
2060
    return -1;
2061
}
2062

    
2063
static void find_new_snapshot_id(BlockDriverState *bs,
2064
                                 char *id_str, int id_str_size)
2065
{
2066
    BDRVQcowState *s = bs->opaque;
2067
    QCowSnapshot *sn;
2068
    int i, id, id_max = 0;
2069

    
2070
    for(i = 0; i < s->nb_snapshots; i++) {
2071
        sn = s->snapshots + i;
2072
        id = strtoul(sn->id_str, NULL, 10);
2073
        if (id > id_max)
2074
            id_max = id;
2075
    }
2076
    snprintf(id_str, id_str_size, "%d", id_max + 1);
2077
}
2078

    
2079
static int find_snapshot_by_id(BlockDriverState *bs, const char *id_str)
2080
{
2081
    BDRVQcowState *s = bs->opaque;
2082
    int i;
2083

    
2084
    for(i = 0; i < s->nb_snapshots; i++) {
2085
        if (!strcmp(s->snapshots[i].id_str, id_str))
2086
            return i;
2087
    }
2088
    return -1;
2089
}
2090

    
2091
static int find_snapshot_by_id_or_name(BlockDriverState *bs, const char *name)
2092
{
2093
    BDRVQcowState *s = bs->opaque;
2094
    int i, ret;
2095

    
2096
    ret = find_snapshot_by_id(bs, name);
2097
    if (ret >= 0)
2098
        return ret;
2099
    for(i = 0; i < s->nb_snapshots; i++) {
2100
        if (!strcmp(s->snapshots[i].name, name))
2101
            return i;
2102
    }
2103
    return -1;
2104
}
2105

    
2106
/* if no id is provided, a new one is constructed */
2107
static int qcow_snapshot_create(BlockDriverState *bs,
2108
                                QEMUSnapshotInfo *sn_info)
2109
{
2110
    BDRVQcowState *s = bs->opaque;
2111
    QCowSnapshot *snapshots1, sn1, *sn = &sn1;
2112
    int i, ret;
2113
    uint64_t *l1_table = NULL;
2114

    
2115
    memset(sn, 0, sizeof(*sn));
2116

    
2117
    if (sn_info->id_str[0] == '\0') {
2118
        /* compute a new id */
2119
        find_new_snapshot_id(bs, sn_info->id_str, sizeof(sn_info->id_str));
2120
    }
2121

    
2122
    /* check that the ID is unique */
2123
    if (find_snapshot_by_id(bs, sn_info->id_str) >= 0)
2124
        return -ENOENT;
2125

    
2126
    sn->id_str = qemu_strdup(sn_info->id_str);
2127
    if (!sn->id_str)
2128
        goto fail;
2129
    sn->name = qemu_strdup(sn_info->name);
2130
    if (!sn->name)
2131
        goto fail;
2132
    sn->vm_state_size = sn_info->vm_state_size;
2133
    sn->date_sec = sn_info->date_sec;
2134
    sn->date_nsec = sn_info->date_nsec;
2135
    sn->vm_clock_nsec = sn_info->vm_clock_nsec;
2136

    
2137
    ret = update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1);
2138
    if (ret < 0)
2139
        goto fail;
2140

    
2141
    /* create the L1 table of the snapshot */
2142
    sn->l1_table_offset = alloc_clusters(bs, s->l1_size * sizeof(uint64_t));
2143
    sn->l1_size = s->l1_size;
2144

    
2145
    l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
2146
    for(i = 0; i < s->l1_size; i++) {
2147
        l1_table[i] = cpu_to_be64(s->l1_table[i]);
2148
    }
2149
    if (bdrv_pwrite(s->hd, sn->l1_table_offset,
2150
                    l1_table, s->l1_size * sizeof(uint64_t)) !=
2151
        (s->l1_size * sizeof(uint64_t)))
2152
        goto fail;
2153
    qemu_free(l1_table);
2154
    l1_table = NULL;
2155

    
2156
    snapshots1 = qemu_malloc((s->nb_snapshots + 1) * sizeof(QCowSnapshot));
2157
    if (s->snapshots) {
2158
        memcpy(snapshots1, s->snapshots, s->nb_snapshots * sizeof(QCowSnapshot));
2159
        qemu_free(s->snapshots);
2160
    }
2161
    s->snapshots = snapshots1;
2162
    s->snapshots[s->nb_snapshots++] = *sn;
2163

    
2164
    if (qcow_write_snapshots(bs) < 0)
2165
        goto fail;
2166
#ifdef DEBUG_ALLOC
2167
    check_refcounts(bs);
2168
#endif
2169
    return 0;
2170
 fail:
2171
    qemu_free(sn->name);
2172
    qemu_free(l1_table);
2173
    return -1;
2174
}
2175

    
2176
/* copy the snapshot 'snapshot_name' into the current disk image */
2177
static int qcow_snapshot_goto(BlockDriverState *bs,
2178
                              const char *snapshot_id)
2179
{
2180
    BDRVQcowState *s = bs->opaque;
2181
    QCowSnapshot *sn;
2182
    int i, snapshot_index, l1_size2;
2183

    
2184
    snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
2185
    if (snapshot_index < 0)
2186
        return -ENOENT;
2187
    sn = &s->snapshots[snapshot_index];
2188

    
2189
    if (update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, -1) < 0)
2190
        goto fail;
2191

    
2192
    if (grow_l1_table(bs, sn->l1_size) < 0)
2193
        goto fail;
2194

    
2195
    s->l1_size = sn->l1_size;
2196
    l1_size2 = s->l1_size * sizeof(uint64_t);
2197
    /* copy the snapshot l1 table to the current l1 table */
2198
    if (bdrv_pread(s->hd, sn->l1_table_offset,
2199
                   s->l1_table, l1_size2) != l1_size2)
2200
        goto fail;
2201
    if (bdrv_pwrite(s->hd, s->l1_table_offset,
2202
                    s->l1_table, l1_size2) != l1_size2)
2203
        goto fail;
2204
    for(i = 0;i < s->l1_size; i++) {
2205
        be64_to_cpus(&s->l1_table[i]);
2206
    }
2207

    
2208
    if (update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1) < 0)
2209
        goto fail;
2210

    
2211
#ifdef DEBUG_ALLOC
2212
    check_refcounts(bs);
2213
#endif
2214
    return 0;
2215
 fail:
2216
    return -EIO;
2217
}
2218

    
2219
static int qcow_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
2220
{
2221
    BDRVQcowState *s = bs->opaque;
2222
    QCowSnapshot *sn;
2223
    int snapshot_index, ret;
2224

    
2225
    snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
2226
    if (snapshot_index < 0)
2227
        return -ENOENT;
2228
    sn = &s->snapshots[snapshot_index];
2229

    
2230
    ret = update_snapshot_refcount(bs, sn->l1_table_offset, sn->l1_size, -1);
2231
    if (ret < 0)
2232
        return ret;
2233
    /* must update the copied flag on the current cluster offsets */
2234
    ret = update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 0);
2235
    if (ret < 0)
2236
        return ret;
2237
    free_clusters(bs, sn->l1_table_offset, sn->l1_size * sizeof(uint64_t));
2238

    
2239
    qemu_free(sn->id_str);
2240
    qemu_free(sn->name);
2241
    memmove(sn, sn + 1, (s->nb_snapshots - snapshot_index - 1) * sizeof(*sn));
2242
    s->nb_snapshots--;
2243
    ret = qcow_write_snapshots(bs);
2244
    if (ret < 0) {
2245
        /* XXX: restore snapshot if error ? */
2246
        return ret;
2247
    }
2248
#ifdef DEBUG_ALLOC
2249
    check_refcounts(bs);
2250
#endif
2251
    return 0;
2252
}
2253

    
2254
static int qcow_snapshot_list(BlockDriverState *bs,
2255
                              QEMUSnapshotInfo **psn_tab)
2256
{
2257
    BDRVQcowState *s = bs->opaque;
2258
    QEMUSnapshotInfo *sn_tab, *sn_info;
2259
    QCowSnapshot *sn;
2260
    int i;
2261

    
2262
    sn_tab = qemu_mallocz(s->nb_snapshots * sizeof(QEMUSnapshotInfo));
2263
    for(i = 0; i < s->nb_snapshots; i++) {
2264
        sn_info = sn_tab + i;
2265
        sn = s->snapshots + i;
2266
        pstrcpy(sn_info->id_str, sizeof(sn_info->id_str),
2267
                sn->id_str);
2268
        pstrcpy(sn_info->name, sizeof(sn_info->name),
2269
                sn->name);
2270
        sn_info->vm_state_size = sn->vm_state_size;
2271
        sn_info->date_sec = sn->date_sec;
2272
        sn_info->date_nsec = sn->date_nsec;
2273
        sn_info->vm_clock_nsec = sn->vm_clock_nsec;
2274
    }
2275
    *psn_tab = sn_tab;
2276
    return s->nb_snapshots;
2277
}
2278

    
2279
/*********************************************************/
2280
/* refcount handling */
2281

    
2282
static int refcount_init(BlockDriverState *bs)
2283
{
2284
    BDRVQcowState *s = bs->opaque;
2285
    int ret, refcount_table_size2, i;
2286

    
2287
    s->refcount_block_cache = qemu_malloc(s->cluster_size);
2288
    refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t);
2289
    s->refcount_table = qemu_malloc(refcount_table_size2);
2290
    if (s->refcount_table_size > 0) {
2291
        ret = bdrv_pread(s->hd, s->refcount_table_offset,
2292
                         s->refcount_table, refcount_table_size2);
2293
        if (ret != refcount_table_size2)
2294
            goto fail;
2295
        for(i = 0; i < s->refcount_table_size; i++)
2296
            be64_to_cpus(&s->refcount_table[i]);
2297
    }
2298
    return 0;
2299
 fail:
2300
    return -ENOMEM;
2301
}
2302

    
2303
static void refcount_close(BlockDriverState *bs)
2304
{
2305
    BDRVQcowState *s = bs->opaque;
2306
    qemu_free(s->refcount_block_cache);
2307
    qemu_free(s->refcount_table);
2308
}
2309

    
2310

    
2311
static int load_refcount_block(BlockDriverState *bs,
2312
                               int64_t refcount_block_offset)
2313
{
2314
    BDRVQcowState *s = bs->opaque;
2315
    int ret;
2316
    ret = bdrv_pread(s->hd, refcount_block_offset, s->refcount_block_cache,
2317
                     s->cluster_size);
2318
    if (ret != s->cluster_size)
2319
        return -EIO;
2320
    s->refcount_block_cache_offset = refcount_block_offset;
2321
    return 0;
2322
}
2323

    
2324
static int get_refcount(BlockDriverState *bs, int64_t cluster_index)
2325
{
2326
    BDRVQcowState *s = bs->opaque;
2327
    int refcount_table_index, block_index;
2328
    int64_t refcount_block_offset;
2329

    
2330
    refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
2331
    if (refcount_table_index >= s->refcount_table_size)
2332
        return 0;
2333
    refcount_block_offset = s->refcount_table[refcount_table_index];
2334
    if (!refcount_block_offset)
2335
        return 0;
2336
    if (refcount_block_offset != s->refcount_block_cache_offset) {
2337
        /* better than nothing: return allocated if read error */
2338
        if (load_refcount_block(bs, refcount_block_offset) < 0)
2339
            return 1;
2340
    }
2341
    block_index = cluster_index &
2342
        ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
2343
    return be16_to_cpu(s->refcount_block_cache[block_index]);
2344
}
2345

    
2346
/* return < 0 if error */
2347
static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size)
2348
{
2349
    BDRVQcowState *s = bs->opaque;
2350
    int i, nb_clusters;
2351

    
2352
    nb_clusters = size_to_clusters(s, size);
2353
retry:
2354
    for(i = 0; i < nb_clusters; i++) {
2355
        int64_t i = s->free_cluster_index++;
2356
        if (get_refcount(bs, i) != 0)
2357
            goto retry;
2358
    }
2359
#ifdef DEBUG_ALLOC2
2360
    printf("alloc_clusters: size=%lld -> %lld\n",
2361
            size,
2362
            (s->free_cluster_index - nb_clusters) << s->cluster_bits);
2363
#endif
2364
    return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
2365
}
2366

    
2367
static int64_t alloc_clusters(BlockDriverState *bs, int64_t size)
2368
{
2369
    int64_t offset;
2370

    
2371
    offset = alloc_clusters_noref(bs, size);
2372
    update_refcount(bs, offset, size, 1);
2373
    return offset;
2374
}
2375

    
2376
/* only used to allocate compressed sectors. We try to allocate
2377
   contiguous sectors. size must be <= cluster_size */
2378
static int64_t alloc_bytes(BlockDriverState *bs, int size)
2379
{
2380
    BDRVQcowState *s = bs->opaque;
2381
    int64_t offset, cluster_offset;
2382
    int free_in_cluster;
2383

    
2384
    assert(size > 0 && size <= s->cluster_size);
2385
    if (s->free_byte_offset == 0) {
2386
        s->free_byte_offset = alloc_clusters(bs, s->cluster_size);
2387
    }
2388
 redo:
2389
    free_in_cluster = s->cluster_size -
2390
        (s->free_byte_offset & (s->cluster_size - 1));
2391
    if (size <= free_in_cluster) {
2392
        /* enough space in current cluster */
2393
        offset = s->free_byte_offset;
2394
        s->free_byte_offset += size;
2395
        free_in_cluster -= size;
2396
        if (free_in_cluster == 0)
2397
            s->free_byte_offset = 0;
2398
        if ((offset & (s->cluster_size - 1)) != 0)
2399
            update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
2400
    } else {
2401
        offset = alloc_clusters(bs, s->cluster_size);
2402
        cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1);
2403
        if ((cluster_offset + s->cluster_size) == offset) {
2404
            /* we are lucky: contiguous data */
2405
            offset = s->free_byte_offset;
2406
            update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
2407
            s->free_byte_offset += size;
2408
        } else {
2409
            s->free_byte_offset = offset;
2410
            goto redo;
2411
        }
2412
    }
2413
    return offset;
2414
}
2415

    
2416
static void free_clusters(BlockDriverState *bs,
2417
                          int64_t offset, int64_t size)
2418
{
2419
    update_refcount(bs, offset, size, -1);
2420
}
2421

    
2422
static int grow_refcount_table(BlockDriverState *bs, int min_size)
2423
{
2424
    BDRVQcowState *s = bs->opaque;
2425
    int new_table_size, new_table_size2, refcount_table_clusters, i, ret;
2426
    uint64_t *new_table;
2427
    int64_t table_offset;
2428
    uint8_t data[12];
2429
    int old_table_size;
2430
    int64_t old_table_offset;
2431

    
2432
    if (min_size <= s->refcount_table_size)
2433
        return 0;
2434
    /* compute new table size */
2435
    refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
2436
    for(;;) {
2437
        if (refcount_table_clusters == 0) {
2438
            refcount_table_clusters = 1;
2439
        } else {
2440
            refcount_table_clusters = (refcount_table_clusters * 3 + 1) / 2;
2441
        }
2442
        new_table_size = refcount_table_clusters << (s->cluster_bits - 3);
2443
        if (min_size <= new_table_size)
2444
            break;
2445
    }
2446
#ifdef DEBUG_ALLOC2
2447
    printf("grow_refcount_table from %d to %d\n",
2448
           s->refcount_table_size,
2449
           new_table_size);
2450
#endif
2451
    new_table_size2 = new_table_size * sizeof(uint64_t);
2452
    new_table = qemu_mallocz(new_table_size2);
2453
    memcpy(new_table, s->refcount_table,
2454
           s->refcount_table_size * sizeof(uint64_t));
2455
    for(i = 0; i < s->refcount_table_size; i++)
2456
        cpu_to_be64s(&new_table[i]);
2457
    /* Note: we cannot update the refcount now to avoid recursion */
2458
    table_offset = alloc_clusters_noref(bs, new_table_size2);
2459
    ret = bdrv_pwrite(s->hd, table_offset, new_table, new_table_size2);
2460
    if (ret != new_table_size2)
2461
        goto fail;
2462
    for(i = 0; i < s->refcount_table_size; i++)
2463
        be64_to_cpus(&new_table[i]);
2464

    
2465
    cpu_to_be64w((uint64_t*)data, table_offset);
2466
    cpu_to_be32w((uint32_t*)(data + 8), refcount_table_clusters);
2467
    if (bdrv_pwrite(s->hd, offsetof(QCowHeader, refcount_table_offset),
2468
                    data, sizeof(data)) != sizeof(data))
2469
        goto fail;
2470
    qemu_free(s->refcount_table);
2471
    old_table_offset = s->refcount_table_offset;
2472
    old_table_size = s->refcount_table_size;
2473
    s->refcount_table = new_table;
2474
    s->refcount_table_size = new_table_size;
2475
    s->refcount_table_offset = table_offset;
2476

    
2477
    update_refcount(bs, table_offset, new_table_size2, 1);
2478
    free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t));
2479
    return 0;
2480
 fail:
2481
    free_clusters(bs, table_offset, new_table_size2);
2482
    qemu_free(new_table);
2483
    return -EIO;
2484
}
2485

    
2486
/* addend must be 1 or -1 */
2487
/* XXX: cache several refcount block clusters ? */
2488
static int update_cluster_refcount(BlockDriverState *bs,
2489
                                   int64_t cluster_index,
2490
                                   int addend)
2491
{
2492
    BDRVQcowState *s = bs->opaque;
2493
    int64_t offset, refcount_block_offset;
2494
    int ret, refcount_table_index, block_index, refcount;
2495
    uint64_t data64;
2496

    
2497
    refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
2498
    if (refcount_table_index >= s->refcount_table_size) {
2499
        if (addend < 0)
2500
            return -EINVAL;
2501
        ret = grow_refcount_table(bs, refcount_table_index + 1);
2502
        if (ret < 0)
2503
            return ret;
2504
    }
2505
    refcount_block_offset = s->refcount_table[refcount_table_index];
2506
    if (!refcount_block_offset) {
2507
        if (addend < 0)
2508
            return -EINVAL;
2509
        /* create a new refcount block */
2510
        /* Note: we cannot update the refcount now to avoid recursion */
2511
        offset = alloc_clusters_noref(bs, s->cluster_size);
2512
        memset(s->refcount_block_cache, 0, s->cluster_size);
2513
        ret = bdrv_pwrite(s->hd, offset, s->refcount_block_cache, s->cluster_size);
2514
        if (ret != s->cluster_size)
2515
            return -EINVAL;
2516
        s->refcount_table[refcount_table_index] = offset;
2517
        data64 = cpu_to_be64(offset);
2518
        ret = bdrv_pwrite(s->hd, s->refcount_table_offset +
2519
                          refcount_table_index * sizeof(uint64_t),
2520
                          &data64, sizeof(data64));
2521
        if (ret != sizeof(data64))
2522
            return -EINVAL;
2523

    
2524
        refcount_block_offset = offset;
2525
        s->refcount_block_cache_offset = offset;
2526
        update_refcount(bs, offset, s->cluster_size, 1);
2527
    } else {
2528
        if (refcount_block_offset != s->refcount_block_cache_offset) {
2529
            if (load_refcount_block(bs, refcount_block_offset) < 0)
2530
                return -EIO;
2531
        }
2532
    }
2533
    /* we can update the count and save it */
2534
    block_index = cluster_index &
2535
        ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
2536
    refcount = be16_to_cpu(s->refcount_block_cache[block_index]);
2537
    refcount += addend;
2538
    if (refcount < 0 || refcount > 0xffff)
2539
        return -EINVAL;
2540
    if (refcount == 0 && cluster_index < s->free_cluster_index) {
2541
        s->free_cluster_index = cluster_index;
2542
    }
2543
    s->refcount_block_cache[block_index] = cpu_to_be16(refcount);
2544
    if (bdrv_pwrite(s->hd,
2545
                    refcount_block_offset + (block_index << REFCOUNT_SHIFT),
2546
                    &s->refcount_block_cache[block_index], 2) != 2)
2547
        return -EIO;
2548
    return refcount;
2549
}
2550

    
2551
static void update_refcount(BlockDriverState *bs,
2552
                            int64_t offset, int64_t length,
2553
                            int addend)
2554
{
2555
    BDRVQcowState *s = bs->opaque;
2556
    int64_t start, last, cluster_offset;
2557

    
2558
#ifdef DEBUG_ALLOC2
2559
    printf("update_refcount: offset=%lld size=%lld addend=%d\n",
2560
           offset, length, addend);
2561
#endif
2562
    if (length <= 0)
2563
        return;
2564
    start = offset & ~(s->cluster_size - 1);
2565
    last = (offset + length - 1) & ~(s->cluster_size - 1);
2566
    for(cluster_offset = start; cluster_offset <= last;
2567
        cluster_offset += s->cluster_size) {
2568
        update_cluster_refcount(bs, cluster_offset >> s->cluster_bits, addend);
2569
    }
2570
}
2571

    
2572
/*
2573
 * Increases the refcount for a range of clusters in a given refcount table.
2574
 * This is used to construct a temporary refcount table out of L1 and L2 tables
2575
 * which can be compared the the refcount table saved in the image.
2576
 *
2577
 * Returns the number of errors in the image that were found
2578
 */
2579
static int inc_refcounts(BlockDriverState *bs,
2580
                          uint16_t *refcount_table,
2581
                          int refcount_table_size,
2582
                          int64_t offset, int64_t size)
2583
{
2584
    BDRVQcowState *s = bs->opaque;
2585
    int64_t start, last, cluster_offset;
2586
    int k;
2587
    int errors = 0;
2588

    
2589
    if (size <= 0)
2590
        return 0;
2591

    
2592
    start = offset & ~(s->cluster_size - 1);
2593
    last = (offset + size - 1) & ~(s->cluster_size - 1);
2594
    for(cluster_offset = start; cluster_offset <= last;
2595
        cluster_offset += s->cluster_size) {
2596
        k = cluster_offset >> s->cluster_bits;
2597
        if (k < 0 || k >= refcount_table_size) {
2598
            fprintf(stderr, "ERROR: invalid cluster offset=0x%" PRIx64 "\n",
2599
                cluster_offset);
2600
            errors++;
2601
        } else {
2602
            if (++refcount_table[k] == 0) {
2603
                fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64
2604
                    "\n", cluster_offset);
2605
                errors++;
2606
            }
2607
        }
2608
    }
2609

    
2610
    return errors;
2611
}
2612

    
2613
/*
2614
 * Increases the refcount in the given refcount table for the all clusters
2615
 * referenced in the L2 table. While doing so, performs some checks on L2
2616
 * entries.
2617
 *
2618
 * Returns the number of errors found by the checks or -errno if an internal
2619
 * error occurred.
2620
 */
2621
static int check_refcounts_l2(BlockDriverState *bs,
2622
    uint16_t *refcount_table, int refcount_table_size, int64_t l2_offset,
2623
    int check_copied)
2624
{
2625
    BDRVQcowState *s = bs->opaque;
2626
    uint64_t *l2_table, offset;
2627
    int i, l2_size, nb_csectors, refcount;
2628
    int errors = 0;
2629

    
2630
    /* Read L2 table from disk */
2631
    l2_size = s->l2_size * sizeof(uint64_t);
2632
    l2_table = qemu_malloc(l2_size);
2633

    
2634
    if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
2635
        goto fail;
2636

    
2637
    /* Do the actual checks */
2638
    for(i = 0; i < s->l2_size; i++) {
2639
        offset = be64_to_cpu(l2_table[i]);
2640
        if (offset != 0) {
2641
            if (offset & QCOW_OFLAG_COMPRESSED) {
2642
                /* Compressed clusters don't have QCOW_OFLAG_COPIED */
2643
                if (offset & QCOW_OFLAG_COPIED) {
2644
                    fprintf(stderr, "ERROR: cluster %" PRId64 ": "
2645
                        "copied flag must never be set for compressed "
2646
                        "clusters\n", offset >> s->cluster_bits);
2647
                    offset &= ~QCOW_OFLAG_COPIED;
2648
                    errors++;
2649
                }
2650

    
2651
                /* Mark cluster as used */
2652
                nb_csectors = ((offset >> s->csize_shift) &
2653
                               s->csize_mask) + 1;
2654
                offset &= s->cluster_offset_mask;
2655
                errors += inc_refcounts(bs, refcount_table,
2656
                              refcount_table_size,
2657
                              offset & ~511, nb_csectors * 512);
2658
            } else {
2659
                /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
2660
                if (check_copied) {
2661
                    uint64_t entry = offset;
2662
                    offset &= ~QCOW_OFLAG_COPIED;
2663
                    refcount = get_refcount(bs, offset >> s->cluster_bits);
2664
                    if ((refcount == 1) != ((entry & QCOW_OFLAG_COPIED) != 0)) {
2665
                        fprintf(stderr, "ERROR OFLAG_COPIED: offset=%"
2666
                            PRIx64 " refcount=%d\n", entry, refcount);
2667
                        errors++;
2668
                    }
2669
                }
2670

    
2671
                /* Mark cluster as used */
2672
                offset &= ~QCOW_OFLAG_COPIED;
2673
                errors += inc_refcounts(bs, refcount_table,
2674
                              refcount_table_size,
2675
                              offset, s->cluster_size);
2676

    
2677
                /* Correct offsets are cluster aligned */
2678
                if (offset & (s->cluster_size - 1)) {
2679
                    fprintf(stderr, "ERROR offset=%" PRIx64 ": Cluster is not "
2680
                        "properly aligned; L2 entry corrupted.\n", offset);
2681
                    errors++;
2682
                }
2683
            }
2684
        }
2685
    }
2686

    
2687
    qemu_free(l2_table);
2688
    return errors;
2689

    
2690
fail:
2691
    fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
2692
    qemu_free(l2_table);
2693
    return -EIO;
2694
}
2695

    
2696
/*
2697
 * Increases the refcount for the L1 table, its L2 tables and all referenced
2698
 * clusters in the given refcount table. While doing so, performs some checks
2699
 * on L1 and L2 entries.
2700
 *
2701
 * Returns the number of errors found by the checks or -errno if an internal
2702
 * error occurred.
2703
 */
2704
static int check_refcounts_l1(BlockDriverState *bs,
2705
                              uint16_t *refcount_table,
2706
                              int refcount_table_size,
2707
                              int64_t l1_table_offset, int l1_size,
2708
                              int check_copied)
2709
{
2710
    BDRVQcowState *s = bs->opaque;
2711
    uint64_t *l1_table, l2_offset, l1_size2;
2712
    int i, refcount, ret;
2713
    int errors = 0;
2714

    
2715
    l1_size2 = l1_size * sizeof(uint64_t);
2716

    
2717
    /* Mark L1 table as used */
2718
    errors += inc_refcounts(bs, refcount_table, refcount_table_size,
2719
                  l1_table_offset, l1_size2);
2720

    
2721
    /* Read L1 table entries from disk */
2722
    l1_table = qemu_malloc(l1_size2);
2723
    if (bdrv_pread(s->hd, l1_table_offset,
2724
                   l1_table, l1_size2) != l1_size2)
2725
        goto fail;
2726
    for(i = 0;i < l1_size; i++)
2727
        be64_to_cpus(&l1_table[i]);
2728

    
2729
    /* Do the actual checks */
2730
    for(i = 0; i < l1_size; i++) {
2731
        l2_offset = l1_table[i];
2732
        if (l2_offset) {
2733
            /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
2734
            if (check_copied) {
2735
                refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED)
2736
                    >> s->cluster_bits);
2737
                if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) {
2738
                    fprintf(stderr, "ERROR OFLAG_COPIED: l2_offset=%" PRIx64
2739
                        " refcount=%d\n", l2_offset, refcount);
2740
                    errors++;
2741
                }
2742
            }
2743

    
2744
            /* Mark L2 table as used */
2745
            l2_offset &= ~QCOW_OFLAG_COPIED;
2746
            errors += inc_refcounts(bs, refcount_table,
2747
                          refcount_table_size,
2748
                          l2_offset,
2749
                          s->cluster_size);
2750

    
2751
            /* L2 tables are cluster aligned */
2752
            if (l2_offset & (s->cluster_size - 1)) {
2753
                fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not "
2754
                    "cluster aligned; L1 entry corrupted\n", l2_offset);
2755
                errors++;
2756
            }
2757

    
2758
            /* Process and check L2 entries */
2759
            ret = check_refcounts_l2(bs, refcount_table, refcount_table_size,
2760
                l2_offset, check_copied);
2761
            if (ret < 0) {
2762
                goto fail;
2763
            }
2764
            errors += ret;
2765
        }
2766
    }
2767
    qemu_free(l1_table);
2768
    return errors;
2769

    
2770
fail:
2771
    fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
2772
    qemu_free(l1_table);
2773
    return -EIO;
2774
}
2775

    
2776
/*
2777
 * Checks an image for refcount consistency.
2778
 *
2779
 * Returns 0 if no errors are found, the number of errors in case the image is
2780
 * detected as corrupted, and -errno when an internal error occured.
2781
 */
2782
static int check_refcounts(BlockDriverState *bs)
2783
{
2784
    BDRVQcowState *s = bs->opaque;
2785
    int64_t size;
2786
    int nb_clusters, refcount1, refcount2, i;
2787
    QCowSnapshot *sn;
2788
    uint16_t *refcount_table;
2789
    int ret, errors = 0;
2790

    
2791
    size = bdrv_getlength(s->hd);
2792
    nb_clusters = size_to_clusters(s, size);
2793
    refcount_table = qemu_mallocz(nb_clusters * sizeof(uint16_t));
2794

    
2795
    /* header */
2796
    errors += inc_refcounts(bs, refcount_table, nb_clusters,
2797
                  0, s->cluster_size);
2798

    
2799
    /* current L1 table */
2800
    ret = check_refcounts_l1(bs, refcount_table, nb_clusters,
2801
                       s->l1_table_offset, s->l1_size, 1);
2802
    if (ret < 0) {
2803
        return ret;
2804
    }
2805
    errors += ret;
2806

    
2807
    /* snapshots */
2808
    for(i = 0; i < s->nb_snapshots; i++) {
2809
        sn = s->snapshots + i;
2810
        check_refcounts_l1(bs, refcount_table, nb_clusters,
2811
                           sn->l1_table_offset, sn->l1_size, 0);
2812
    }
2813
    errors += inc_refcounts(bs, refcount_table, nb_clusters,
2814
                  s->snapshots_offset, s->snapshots_size);
2815

    
2816
    /* refcount data */
2817
    errors += inc_refcounts(bs, refcount_table, nb_clusters,
2818
                  s->refcount_table_offset,
2819
                  s->refcount_table_size * sizeof(uint64_t));
2820
    for(i = 0; i < s->refcount_table_size; i++) {
2821
        int64_t offset;
2822
        offset = s->refcount_table[i];
2823
        if (offset != 0) {
2824
            errors += inc_refcounts(bs, refcount_table, nb_clusters,
2825
                          offset, s->cluster_size);
2826
        }
2827
    }
2828

    
2829
    /* compare ref counts */
2830
    for(i = 0; i < nb_clusters; i++) {
2831
        refcount1 = get_refcount(bs, i);
2832
        refcount2 = refcount_table[i];
2833
        if (refcount1 != refcount2) {
2834
            fprintf(stderr, "ERROR cluster %d refcount=%d reference=%d\n",
2835
                   i, refcount1, refcount2);
2836
            errors++;
2837
        }
2838
    }
2839

    
2840
    qemu_free(refcount_table);
2841

    
2842
    return errors;
2843
}
2844

    
2845
static int qcow_check(BlockDriverState *bs)
2846
{
2847
    return check_refcounts(bs);
2848
}
2849

    
2850
#if 0
2851
static void dump_refcounts(BlockDriverState *bs)
2852
{
2853
    BDRVQcowState *s = bs->opaque;
2854
    int64_t nb_clusters, k, k1, size;
2855
    int refcount;
2856

2857
    size = bdrv_getlength(s->hd);
2858
    nb_clusters = size_to_clusters(s, size);
2859
    for(k = 0; k < nb_clusters;) {
2860
        k1 = k;
2861
        refcount = get_refcount(bs, k);
2862
        k++;
2863
        while (k < nb_clusters && get_refcount(bs, k) == refcount)
2864
            k++;
2865
        printf("%lld: refcount=%d nb=%lld\n", k, refcount, k - k1);
2866
    }
2867
}
2868
#endif
2869

    
2870
static int qcow_put_buffer(BlockDriverState *bs, const uint8_t *buf,
2871
                           int64_t pos, int size)
2872
{
2873
    int growable = bs->growable;
2874

    
2875
    bs->growable = 1;
2876
    bdrv_pwrite(bs, pos, buf, size);
2877
    bs->growable = growable;
2878

    
2879
    return size;
2880
}
2881

    
2882
static int qcow_get_buffer(BlockDriverState *bs, uint8_t *buf,
2883
                           int64_t pos, int size)
2884
{
2885
    int growable = bs->growable;
2886
    int ret;
2887

    
2888
    bs->growable = 1;
2889
    ret = bdrv_pread(bs, pos, buf, size);
2890
    bs->growable = growable;
2891

    
2892
    return ret;
2893
}
2894

    
2895
static BlockDriver bdrv_qcow2 = {
2896
    .format_name        = "qcow2",
2897
    .instance_size        = sizeof(BDRVQcowState),
2898
    .bdrv_probe                = qcow_probe,
2899
    .bdrv_open                = qcow_open,
2900
    .bdrv_close                = qcow_close,
2901
    .bdrv_create        = qcow_create,
2902
    .bdrv_flush                = qcow_flush,
2903
    .bdrv_is_allocated        = qcow_is_allocated,
2904
    .bdrv_set_key        = qcow_set_key,
2905
    .bdrv_make_empty        = qcow_make_empty,
2906

    
2907
    .bdrv_aio_readv        = qcow_aio_readv,
2908
    .bdrv_aio_writev        = qcow_aio_writev,
2909
    .bdrv_aio_cancel        = qcow_aio_cancel,
2910
    .aiocb_size                = sizeof(QCowAIOCB),
2911
    .bdrv_write_compressed = qcow_write_compressed,
2912

    
2913
    .bdrv_snapshot_create = qcow_snapshot_create,
2914
    .bdrv_snapshot_goto        = qcow_snapshot_goto,
2915
    .bdrv_snapshot_delete = qcow_snapshot_delete,
2916
    .bdrv_snapshot_list        = qcow_snapshot_list,
2917
    .bdrv_get_info        = qcow_get_info,
2918

    
2919
    .bdrv_put_buffer    = qcow_put_buffer,
2920
    .bdrv_get_buffer    = qcow_get_buffer,
2921

    
2922
    .bdrv_create2 = qcow_create2,
2923
    .bdrv_check = qcow_check,
2924
};
2925

    
2926
static void bdrv_qcow2_init(void)
2927
{
2928
    bdrv_register(&bdrv_qcow2);
2929
}
2930

    
2931
block_init(bdrv_qcow2_init);