]> git.proxmox.com Git - qemu.git/blob - block-qcow2.c
Cleanup {alloc|get}_cluster_offset() (Gleb Natapov)
[qemu.git] / block-qcow2.c
1 /*
2 * Block driver for the QCOW version 2 format
3 *
4 * Copyright (c) 2004-2006 Fabrice Bellard
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
24 #include "qemu-common.h"
25 #include "block_int.h"
26 #include <zlib.h>
27 #include "aes.h"
28 #include <assert.h>
29
30 /*
31 Differences with QCOW:
32
33 - Support for multiple incremental snapshots.
34 - Memory management by reference counts.
35 - Clusters which have a reference count of one have the bit
36 QCOW_OFLAG_COPIED to optimize write performance.
37 - Size of compressed clusters is stored in sectors to reduce bit usage
38 in the cluster offsets.
39 - Support for storing additional data (such as the VM state) in the
40 snapshots.
41 - If a backing store is used, the cluster size is not constrained
42 (could be backported to QCOW).
43 - L2 tables have always a size of one cluster.
44 */
45
46 //#define DEBUG_ALLOC
47 //#define DEBUG_ALLOC2
48
49 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
50 #define QCOW_VERSION 2
51
52 #define QCOW_CRYPT_NONE 0
53 #define QCOW_CRYPT_AES 1
54
55 #define QCOW_MAX_CRYPT_CLUSTERS 32
56
57 /* indicate that the refcount of the referenced cluster is exactly one. */
58 #define QCOW_OFLAG_COPIED (1LL << 63)
59 /* indicate that the cluster is compressed (they never have the copied flag) */
60 #define QCOW_OFLAG_COMPRESSED (1LL << 62)
61
62 #define REFCOUNT_SHIFT 1 /* refcount size is 2 bytes */
63
64 typedef struct QCowHeader {
65 uint32_t magic;
66 uint32_t version;
67 uint64_t backing_file_offset;
68 uint32_t backing_file_size;
69 uint32_t cluster_bits;
70 uint64_t size; /* in bytes */
71 uint32_t crypt_method;
72 uint32_t l1_size; /* XXX: save number of clusters instead ? */
73 uint64_t l1_table_offset;
74 uint64_t refcount_table_offset;
75 uint32_t refcount_table_clusters;
76 uint32_t nb_snapshots;
77 uint64_t snapshots_offset;
78 } QCowHeader;
79
80 typedef struct __attribute__((packed)) QCowSnapshotHeader {
81 /* header is 8 byte aligned */
82 uint64_t l1_table_offset;
83
84 uint32_t l1_size;
85 uint16_t id_str_size;
86 uint16_t name_size;
87
88 uint32_t date_sec;
89 uint32_t date_nsec;
90
91 uint64_t vm_clock_nsec;
92
93 uint32_t vm_state_size;
94 uint32_t extra_data_size; /* for extension */
95 /* extra data follows */
96 /* id_str follows */
97 /* name follows */
98 } QCowSnapshotHeader;
99
100 #define L2_CACHE_SIZE 16
101
102 typedef struct QCowSnapshot {
103 uint64_t l1_table_offset;
104 uint32_t l1_size;
105 char *id_str;
106 char *name;
107 uint32_t vm_state_size;
108 uint32_t date_sec;
109 uint32_t date_nsec;
110 uint64_t vm_clock_nsec;
111 } QCowSnapshot;
112
113 typedef struct BDRVQcowState {
114 BlockDriverState *hd;
115 int cluster_bits;
116 int cluster_size;
117 int cluster_sectors;
118 int l2_bits;
119 int l2_size;
120 int l1_size;
121 int l1_vm_state_index;
122 int csize_shift;
123 int csize_mask;
124 uint64_t cluster_offset_mask;
125 uint64_t l1_table_offset;
126 uint64_t *l1_table;
127 uint64_t *l2_cache;
128 uint64_t l2_cache_offsets[L2_CACHE_SIZE];
129 uint32_t l2_cache_counts[L2_CACHE_SIZE];
130 uint8_t *cluster_cache;
131 uint8_t *cluster_data;
132 uint64_t cluster_cache_offset;
133
134 uint64_t *refcount_table;
135 uint64_t refcount_table_offset;
136 uint32_t refcount_table_size;
137 uint64_t refcount_block_cache_offset;
138 uint16_t *refcount_block_cache;
139 int64_t free_cluster_index;
140 int64_t free_byte_offset;
141
142 uint32_t crypt_method; /* current crypt method, 0 if no key yet */
143 uint32_t crypt_method_header;
144 AES_KEY aes_encrypt_key;
145 AES_KEY aes_decrypt_key;
146 uint64_t snapshots_offset;
147 int snapshots_size;
148 int nb_snapshots;
149 QCowSnapshot *snapshots;
150 } BDRVQcowState;
151
152 static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset);
153 static int qcow_read(BlockDriverState *bs, int64_t sector_num,
154 uint8_t *buf, int nb_sectors);
155 static int qcow_read_snapshots(BlockDriverState *bs);
156 static void qcow_free_snapshots(BlockDriverState *bs);
157 static int refcount_init(BlockDriverState *bs);
158 static void refcount_close(BlockDriverState *bs);
159 static int get_refcount(BlockDriverState *bs, int64_t cluster_index);
160 static int update_cluster_refcount(BlockDriverState *bs,
161 int64_t cluster_index,
162 int addend);
163 static void update_refcount(BlockDriverState *bs,
164 int64_t offset, int64_t length,
165 int addend);
166 static int64_t alloc_clusters(BlockDriverState *bs, int64_t size);
167 static int64_t alloc_bytes(BlockDriverState *bs, int size);
168 static void free_clusters(BlockDriverState *bs,
169 int64_t offset, int64_t size);
170 #ifdef DEBUG_ALLOC
171 static void check_refcounts(BlockDriverState *bs);
172 #endif
173
174 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
175 {
176 const QCowHeader *cow_header = (const void *)buf;
177
178 if (buf_size >= sizeof(QCowHeader) &&
179 be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
180 be32_to_cpu(cow_header->version) == QCOW_VERSION)
181 return 100;
182 else
183 return 0;
184 }
185
186 static int qcow_open(BlockDriverState *bs, const char *filename, int flags)
187 {
188 BDRVQcowState *s = bs->opaque;
189 int len, i, shift, ret;
190 QCowHeader header;
191
192 ret = bdrv_file_open(&s->hd, filename, flags);
193 if (ret < 0)
194 return ret;
195 if (bdrv_pread(s->hd, 0, &header, sizeof(header)) != sizeof(header))
196 goto fail;
197 be32_to_cpus(&header.magic);
198 be32_to_cpus(&header.version);
199 be64_to_cpus(&header.backing_file_offset);
200 be32_to_cpus(&header.backing_file_size);
201 be64_to_cpus(&header.size);
202 be32_to_cpus(&header.cluster_bits);
203 be32_to_cpus(&header.crypt_method);
204 be64_to_cpus(&header.l1_table_offset);
205 be32_to_cpus(&header.l1_size);
206 be64_to_cpus(&header.refcount_table_offset);
207 be32_to_cpus(&header.refcount_table_clusters);
208 be64_to_cpus(&header.snapshots_offset);
209 be32_to_cpus(&header.nb_snapshots);
210
211 if (header.magic != QCOW_MAGIC || header.version != QCOW_VERSION)
212 goto fail;
213 if (header.size <= 1 ||
214 header.cluster_bits < 9 ||
215 header.cluster_bits > 16)
216 goto fail;
217 if (header.crypt_method > QCOW_CRYPT_AES)
218 goto fail;
219 s->crypt_method_header = header.crypt_method;
220 if (s->crypt_method_header)
221 bs->encrypted = 1;
222 s->cluster_bits = header.cluster_bits;
223 s->cluster_size = 1 << s->cluster_bits;
224 s->cluster_sectors = 1 << (s->cluster_bits - 9);
225 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
226 s->l2_size = 1 << s->l2_bits;
227 bs->total_sectors = header.size / 512;
228 s->csize_shift = (62 - (s->cluster_bits - 8));
229 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
230 s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
231 s->refcount_table_offset = header.refcount_table_offset;
232 s->refcount_table_size =
233 header.refcount_table_clusters << (s->cluster_bits - 3);
234
235 s->snapshots_offset = header.snapshots_offset;
236 s->nb_snapshots = header.nb_snapshots;
237
238 /* read the level 1 table */
239 s->l1_size = header.l1_size;
240 shift = s->cluster_bits + s->l2_bits;
241 s->l1_vm_state_index = (header.size + (1LL << shift) - 1) >> shift;
242 /* the L1 table must contain at least enough entries to put
243 header.size bytes */
244 if (s->l1_size < s->l1_vm_state_index)
245 goto fail;
246 s->l1_table_offset = header.l1_table_offset;
247 s->l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
248 if (!s->l1_table)
249 goto fail;
250 if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, s->l1_size * sizeof(uint64_t)) !=
251 s->l1_size * sizeof(uint64_t))
252 goto fail;
253 for(i = 0;i < s->l1_size; i++) {
254 be64_to_cpus(&s->l1_table[i]);
255 }
256 /* alloc L2 cache */
257 s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
258 if (!s->l2_cache)
259 goto fail;
260 s->cluster_cache = qemu_malloc(s->cluster_size);
261 if (!s->cluster_cache)
262 goto fail;
263 /* one more sector for decompressed data alignment */
264 s->cluster_data = qemu_malloc(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
265 + 512);
266 if (!s->cluster_data)
267 goto fail;
268 s->cluster_cache_offset = -1;
269
270 if (refcount_init(bs) < 0)
271 goto fail;
272
273 /* read the backing file name */
274 if (header.backing_file_offset != 0) {
275 len = header.backing_file_size;
276 if (len > 1023)
277 len = 1023;
278 if (bdrv_pread(s->hd, header.backing_file_offset, bs->backing_file, len) != len)
279 goto fail;
280 bs->backing_file[len] = '\0';
281 }
282 if (qcow_read_snapshots(bs) < 0)
283 goto fail;
284
285 #ifdef DEBUG_ALLOC
286 check_refcounts(bs);
287 #endif
288 return 0;
289
290 fail:
291 qcow_free_snapshots(bs);
292 refcount_close(bs);
293 qemu_free(s->l1_table);
294 qemu_free(s->l2_cache);
295 qemu_free(s->cluster_cache);
296 qemu_free(s->cluster_data);
297 bdrv_delete(s->hd);
298 return -1;
299 }
300
301 static int qcow_set_key(BlockDriverState *bs, const char *key)
302 {
303 BDRVQcowState *s = bs->opaque;
304 uint8_t keybuf[16];
305 int len, i;
306
307 memset(keybuf, 0, 16);
308 len = strlen(key);
309 if (len > 16)
310 len = 16;
311 /* XXX: we could compress the chars to 7 bits to increase
312 entropy */
313 for(i = 0;i < len;i++) {
314 keybuf[i] = key[i];
315 }
316 s->crypt_method = s->crypt_method_header;
317
318 if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
319 return -1;
320 if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
321 return -1;
322 #if 0
323 /* test */
324 {
325 uint8_t in[16];
326 uint8_t out[16];
327 uint8_t tmp[16];
328 for(i=0;i<16;i++)
329 in[i] = i;
330 AES_encrypt(in, tmp, &s->aes_encrypt_key);
331 AES_decrypt(tmp, out, &s->aes_decrypt_key);
332 for(i = 0; i < 16; i++)
333 printf(" %02x", tmp[i]);
334 printf("\n");
335 for(i = 0; i < 16; i++)
336 printf(" %02x", out[i]);
337 printf("\n");
338 }
339 #endif
340 return 0;
341 }
342
343 /* The crypt function is compatible with the linux cryptoloop
344 algorithm for < 4 GB images. NOTE: out_buf == in_buf is
345 supported */
346 static void encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
347 uint8_t *out_buf, const uint8_t *in_buf,
348 int nb_sectors, int enc,
349 const AES_KEY *key)
350 {
351 union {
352 uint64_t ll[2];
353 uint8_t b[16];
354 } ivec;
355 int i;
356
357 for(i = 0; i < nb_sectors; i++) {
358 ivec.ll[0] = cpu_to_le64(sector_num);
359 ivec.ll[1] = 0;
360 AES_cbc_encrypt(in_buf, out_buf, 512, key,
361 ivec.b, enc);
362 sector_num++;
363 in_buf += 512;
364 out_buf += 512;
365 }
366 }
367
368 static int copy_sectors(BlockDriverState *bs, uint64_t start_sect,
369 uint64_t cluster_offset, int n_start, int n_end)
370 {
371 BDRVQcowState *s = bs->opaque;
372 int n, ret;
373
374 n = n_end - n_start;
375 if (n <= 0)
376 return 0;
377 ret = qcow_read(bs, start_sect + n_start, s->cluster_data, n);
378 if (ret < 0)
379 return ret;
380 if (s->crypt_method) {
381 encrypt_sectors(s, start_sect + n_start,
382 s->cluster_data,
383 s->cluster_data, n, 1,
384 &s->aes_encrypt_key);
385 }
386 ret = bdrv_write(s->hd, (cluster_offset >> 9) + n_start,
387 s->cluster_data, n);
388 if (ret < 0)
389 return ret;
390 return 0;
391 }
392
393 static void l2_cache_reset(BlockDriverState *bs)
394 {
395 BDRVQcowState *s = bs->opaque;
396
397 memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
398 memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
399 memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
400 }
401
402 static inline int l2_cache_new_entry(BlockDriverState *bs)
403 {
404 BDRVQcowState *s = bs->opaque;
405 uint32_t min_count;
406 int min_index, i;
407
408 /* find a new entry in the least used one */
409 min_index = 0;
410 min_count = 0xffffffff;
411 for(i = 0; i < L2_CACHE_SIZE; i++) {
412 if (s->l2_cache_counts[i] < min_count) {
413 min_count = s->l2_cache_counts[i];
414 min_index = i;
415 }
416 }
417 return min_index;
418 }
419
420 static int64_t align_offset(int64_t offset, int n)
421 {
422 offset = (offset + n - 1) & ~(n - 1);
423 return offset;
424 }
425
426 static int grow_l1_table(BlockDriverState *bs, int min_size)
427 {
428 BDRVQcowState *s = bs->opaque;
429 int new_l1_size, new_l1_size2, ret, i;
430 uint64_t *new_l1_table;
431 uint64_t new_l1_table_offset;
432 uint8_t data[12];
433
434 new_l1_size = s->l1_size;
435 if (min_size <= new_l1_size)
436 return 0;
437 while (min_size > new_l1_size) {
438 new_l1_size = (new_l1_size * 3 + 1) / 2;
439 }
440 #ifdef DEBUG_ALLOC2
441 printf("grow l1_table from %d to %d\n", s->l1_size, new_l1_size);
442 #endif
443
444 new_l1_size2 = sizeof(uint64_t) * new_l1_size;
445 new_l1_table = qemu_mallocz(new_l1_size2);
446 if (!new_l1_table)
447 return -ENOMEM;
448 memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t));
449
450 /* write new table (align to cluster) */
451 new_l1_table_offset = alloc_clusters(bs, new_l1_size2);
452
453 for(i = 0; i < s->l1_size; i++)
454 new_l1_table[i] = cpu_to_be64(new_l1_table[i]);
455 ret = bdrv_pwrite(s->hd, new_l1_table_offset, new_l1_table, new_l1_size2);
456 if (ret != new_l1_size2)
457 goto fail;
458 for(i = 0; i < s->l1_size; i++)
459 new_l1_table[i] = be64_to_cpu(new_l1_table[i]);
460
461 /* set new table */
462 cpu_to_be32w((uint32_t*)data, new_l1_size);
463 cpu_to_be64w((uint64_t*)(data + 4), new_l1_table_offset);
464 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, l1_size), data,
465 sizeof(data)) != sizeof(data))
466 goto fail;
467 qemu_free(s->l1_table);
468 free_clusters(bs, s->l1_table_offset, s->l1_size * sizeof(uint64_t));
469 s->l1_table_offset = new_l1_table_offset;
470 s->l1_table = new_l1_table;
471 s->l1_size = new_l1_size;
472 return 0;
473 fail:
474 qemu_free(s->l1_table);
475 return -EIO;
476 }
477
478 /*
479 * seek_l2_table
480 *
481 * seek l2_offset in the l2_cache table
482 * if not found, return NULL,
483 * if found,
484 * increments the l2 cache hit count of the entry,
485 * if counter overflow, divide by two all counters
486 * return the pointer to the l2 cache entry
487 *
488 */
489
490 static uint64_t *seek_l2_table(BDRVQcowState *s, uint64_t l2_offset)
491 {
492 int i, j;
493
494 for(i = 0; i < L2_CACHE_SIZE; i++) {
495 if (l2_offset == s->l2_cache_offsets[i]) {
496 /* increment the hit count */
497 if (++s->l2_cache_counts[i] == 0xffffffff) {
498 for(j = 0; j < L2_CACHE_SIZE; j++) {
499 s->l2_cache_counts[j] >>= 1;
500 }
501 }
502 return s->l2_cache + (i << s->l2_bits);
503 }
504 }
505 return NULL;
506 }
507
508 /*
509 * l2_load
510 *
511 * Loads a L2 table into memory. If the table is in the cache, the cache
512 * is used; otherwise the L2 table is loaded from the image file.
513 *
514 * Returns a pointer to the L2 table on success, or NULL if the read from
515 * the image file failed.
516 */
517
518 static uint64_t *l2_load(BlockDriverState *bs, uint64_t l2_offset)
519 {
520 BDRVQcowState *s = bs->opaque;
521 int min_index;
522 uint64_t *l2_table;
523
524 /* seek if the table for the given offset is in the cache */
525
526 l2_table = seek_l2_table(s, l2_offset);
527 if (l2_table != NULL)
528 return l2_table;
529
530 /* not found: load a new entry in the least used one */
531
532 min_index = l2_cache_new_entry(bs);
533 l2_table = s->l2_cache + (min_index << s->l2_bits);
534 if (bdrv_pread(s->hd, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
535 s->l2_size * sizeof(uint64_t))
536 return NULL;
537 s->l2_cache_offsets[min_index] = l2_offset;
538 s->l2_cache_counts[min_index] = 1;
539
540 return l2_table;
541 }
542
543 /*
544 * l2_allocate
545 *
546 * Allocate a new l2 entry in the file. If l1_index points to an already
547 * used entry in the L2 table (i.e. we are doing a copy on write for the L2
548 * table) copy the contents of the old L2 table into the newly allocated one.
549 * Otherwise the new table is initialized with zeros.
550 *
551 */
552
553 static uint64_t *l2_allocate(BlockDriverState *bs, int l1_index)
554 {
555 BDRVQcowState *s = bs->opaque;
556 int min_index;
557 uint64_t old_l2_offset, tmp;
558 uint64_t *l2_table, l2_offset;
559
560 old_l2_offset = s->l1_table[l1_index];
561
562 /* allocate a new l2 entry */
563
564 l2_offset = alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
565
566 /* update the L1 entry */
567
568 s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
569
570 tmp = cpu_to_be64(l2_offset | QCOW_OFLAG_COPIED);
571 if (bdrv_pwrite(s->hd, s->l1_table_offset + l1_index * sizeof(tmp),
572 &tmp, sizeof(tmp)) != sizeof(tmp))
573 return NULL;
574
575 /* allocate a new entry in the l2 cache */
576
577 min_index = l2_cache_new_entry(bs);
578 l2_table = s->l2_cache + (min_index << s->l2_bits);
579
580 if (old_l2_offset == 0) {
581 /* if there was no old l2 table, clear the new table */
582 memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
583 } else {
584 /* if there was an old l2 table, read it from the disk */
585 if (bdrv_pread(s->hd, old_l2_offset,
586 l2_table, s->l2_size * sizeof(uint64_t)) !=
587 s->l2_size * sizeof(uint64_t))
588 return NULL;
589 }
590 /* write the l2 table to the file */
591 if (bdrv_pwrite(s->hd, l2_offset,
592 l2_table, s->l2_size * sizeof(uint64_t)) !=
593 s->l2_size * sizeof(uint64_t))
594 return NULL;
595
596 /* update the l2 cache entry */
597
598 s->l2_cache_offsets[min_index] = l2_offset;
599 s->l2_cache_counts[min_index] = 1;
600
601 return l2_table;
602 }
603
604 static int size_to_clusters(BDRVQcowState *s, int64_t size)
605 {
606 return (size + (s->cluster_size - 1)) >> s->cluster_bits;
607 }
608
609 static int count_contiguous_clusters(uint64_t nb_clusters, int cluster_size,
610 uint64_t *l2_table, uint64_t mask)
611 {
612 int i;
613 uint64_t offset = be64_to_cpu(l2_table[0]) & ~mask;
614
615 for (i = 0; i < nb_clusters; i++)
616 if (offset + i * cluster_size != (be64_to_cpu(l2_table[i]) & ~mask))
617 break;
618
619 return i;
620 }
621
622 static int count_contiguous_free_clusters(uint64_t nb_clusters, uint64_t *l2_table)
623 {
624 int i = 0;
625
626 while(nb_clusters-- && l2_table[i] == 0)
627 i++;
628
629 return i;
630 }
631
632 /*
633 * get_cluster_offset
634 *
635 * For a given offset of the disk image, return cluster offset in
636 * qcow2 file.
637 *
638 * on entry, *num is the number of contiguous clusters we'd like to
639 * access following offset.
640 *
641 * on exit, *num is the number of contiguous clusters we can read.
642 *
643 * Return 1, if the offset is found
644 * Return 0, otherwise.
645 *
646 */
647
648 static uint64_t get_cluster_offset(BlockDriverState *bs,
649 uint64_t offset, int *num)
650 {
651 BDRVQcowState *s = bs->opaque;
652 int l1_index, l2_index;
653 uint64_t l2_offset, *l2_table, cluster_offset;
654 int l1_bits, c;
655 int index_in_cluster, nb_available, nb_needed, nb_clusters;
656
657 index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1);
658 nb_needed = *num + index_in_cluster;
659
660 l1_bits = s->l2_bits + s->cluster_bits;
661
662 /* compute how many bytes there are between the offset and
663 * the end of the l1 entry
664 */
665
666 nb_available = (1 << l1_bits) - (offset & ((1 << l1_bits) - 1));
667
668 /* compute the number of available sectors */
669
670 nb_available = (nb_available >> 9) + index_in_cluster;
671
672 cluster_offset = 0;
673
674 /* seek the the l2 offset in the l1 table */
675
676 l1_index = offset >> l1_bits;
677 if (l1_index >= s->l1_size)
678 goto out;
679
680 l2_offset = s->l1_table[l1_index];
681
682 /* seek the l2 table of the given l2 offset */
683
684 if (!l2_offset)
685 goto out;
686
687 /* load the l2 table in memory */
688
689 l2_offset &= ~QCOW_OFLAG_COPIED;
690 l2_table = l2_load(bs, l2_offset);
691 if (l2_table == NULL)
692 return 0;
693
694 /* find the cluster offset for the given disk offset */
695
696 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
697 cluster_offset = be64_to_cpu(l2_table[l2_index]);
698 nb_clusters = size_to_clusters(s, nb_needed << 9);
699
700 if (!cluster_offset) {
701 /* how many empty clusters ? */
702 c = count_contiguous_free_clusters(nb_clusters, &l2_table[l2_index]);
703 } else {
704 /* how many allocated clusters ? */
705 c = count_contiguous_clusters(nb_clusters, s->cluster_size,
706 &l2_table[l2_index], QCOW_OFLAG_COPIED);
707 }
708
709 nb_available = (c * s->cluster_sectors);
710 out:
711 if (nb_available > nb_needed)
712 nb_available = nb_needed;
713
714 *num = nb_available - index_in_cluster;
715
716 return cluster_offset & ~QCOW_OFLAG_COPIED;
717 }
718
719 /*
720 * free_any_clusters
721 *
722 * free clusters according to its type: compressed or not
723 *
724 */
725
726 static void free_any_clusters(BlockDriverState *bs,
727 uint64_t cluster_offset, int nb_clusters)
728 {
729 BDRVQcowState *s = bs->opaque;
730
731 /* free the cluster */
732
733 if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
734 int nb_csectors;
735 nb_csectors = ((cluster_offset >> s->csize_shift) &
736 s->csize_mask) + 1;
737 free_clusters(bs, (cluster_offset & s->cluster_offset_mask) & ~511,
738 nb_csectors * 512);
739 return;
740 }
741
742 free_clusters(bs, cluster_offset, nb_clusters << s->cluster_bits);
743
744 return;
745 }
746
747 /*
748 * get_cluster_table
749 *
750 * for a given disk offset, load (and allocate if needed)
751 * the l2 table.
752 *
753 * the l2 table offset in the qcow2 file and the cluster index
754 * in the l2 table are given to the caller.
755 *
756 */
757
758 static int get_cluster_table(BlockDriverState *bs, uint64_t offset,
759 uint64_t **new_l2_table,
760 uint64_t *new_l2_offset,
761 int *new_l2_index)
762 {
763 BDRVQcowState *s = bs->opaque;
764 int l1_index, l2_index, ret;
765 uint64_t l2_offset, *l2_table;
766
767 /* seek the the l2 offset in the l1 table */
768
769 l1_index = offset >> (s->l2_bits + s->cluster_bits);
770 if (l1_index >= s->l1_size) {
771 ret = grow_l1_table(bs, l1_index + 1);
772 if (ret < 0)
773 return 0;
774 }
775 l2_offset = s->l1_table[l1_index];
776
777 /* seek the l2 table of the given l2 offset */
778
779 if (l2_offset & QCOW_OFLAG_COPIED) {
780 /* load the l2 table in memory */
781 l2_offset &= ~QCOW_OFLAG_COPIED;
782 l2_table = l2_load(bs, l2_offset);
783 if (l2_table == NULL)
784 return 0;
785 } else {
786 if (l2_offset)
787 free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t));
788 l2_table = l2_allocate(bs, l1_index);
789 if (l2_table == NULL)
790 return 0;
791 l2_offset = s->l1_table[l1_index] & ~QCOW_OFLAG_COPIED;
792 }
793
794 /* find the cluster offset for the given disk offset */
795
796 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
797
798 *new_l2_table = l2_table;
799 *new_l2_offset = l2_offset;
800 *new_l2_index = l2_index;
801
802 return 1;
803 }
804
805 /*
806 * alloc_compressed_cluster_offset
807 *
808 * For a given offset of the disk image, return cluster offset in
809 * qcow2 file.
810 *
811 * If the offset is not found, allocate a new compressed cluster.
812 *
813 * Return the cluster offset if successful,
814 * Return 0, otherwise.
815 *
816 */
817
818 static uint64_t alloc_compressed_cluster_offset(BlockDriverState *bs,
819 uint64_t offset,
820 int compressed_size)
821 {
822 BDRVQcowState *s = bs->opaque;
823 int l2_index, ret;
824 uint64_t l2_offset, *l2_table, cluster_offset;
825 int nb_csectors;
826
827 ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
828 if (ret == 0)
829 return 0;
830
831 cluster_offset = be64_to_cpu(l2_table[l2_index]);
832 if (cluster_offset & QCOW_OFLAG_COPIED)
833 return cluster_offset & ~QCOW_OFLAG_COPIED;
834
835 if (cluster_offset)
836 free_any_clusters(bs, cluster_offset, 1);
837
838 cluster_offset = alloc_bytes(bs, compressed_size);
839 nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) -
840 (cluster_offset >> 9);
841
842 cluster_offset |= QCOW_OFLAG_COMPRESSED |
843 ((uint64_t)nb_csectors << s->csize_shift);
844
845 /* update L2 table */
846
847 /* compressed clusters never have the copied flag */
848
849 l2_table[l2_index] = cpu_to_be64(cluster_offset);
850 if (bdrv_pwrite(s->hd,
851 l2_offset + l2_index * sizeof(uint64_t),
852 l2_table + l2_index,
853 sizeof(uint64_t)) != sizeof(uint64_t))
854 return 0;
855
856 return cluster_offset;
857 }
858
859 /*
860 * alloc_cluster_offset
861 *
862 * For a given offset of the disk image, return cluster offset in
863 * qcow2 file.
864 *
865 * If the offset is not found, allocate a new cluster.
866 *
867 * Return the cluster offset if successful,
868 * Return 0, otherwise.
869 *
870 */
871
872 static uint64_t alloc_cluster_offset(BlockDriverState *bs,
873 uint64_t offset,
874 int n_start, int n_end,
875 int *num)
876 {
877 BDRVQcowState *s = bs->opaque;
878 int l2_index, ret;
879 uint64_t l2_offset, *l2_table, cluster_offset;
880 int nb_available, nb_clusters, i = 0;
881 uint64_t start_sect;
882
883 ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
884 if (ret == 0)
885 return 0;
886
887 nb_clusters = size_to_clusters(s, n_end << 9);
888
889 if (nb_clusters > s->l2_size - l2_index)
890 nb_clusters = s->l2_size - l2_index;
891
892 cluster_offset = be64_to_cpu(l2_table[l2_index]);
893
894 /* We keep all QCOW_OFLAG_COPIED clusters */
895
896 if (cluster_offset & QCOW_OFLAG_COPIED) {
897 nb_clusters = count_contiguous_clusters(nb_clusters, s->cluster_size,
898 &l2_table[l2_index], 0);
899
900 nb_available = nb_clusters << (s->cluster_bits - 9);
901 if (nb_available > n_end)
902 nb_available = n_end;
903
904 cluster_offset &= ~QCOW_OFLAG_COPIED;
905
906 goto out;
907 }
908
909 /* for the moment, multiple compressed clusters are not managed */
910
911 if (cluster_offset & QCOW_OFLAG_COMPRESSED)
912 nb_clusters = 1;
913
914 /* how many available clusters ? */
915
916 while (i < nb_clusters) {
917 int j;
918 i += count_contiguous_free_clusters(nb_clusters - i,
919 &l2_table[l2_index + i]);
920
921 cluster_offset = be64_to_cpu(l2_table[l2_index + i]);
922
923 if ((cluster_offset & QCOW_OFLAG_COPIED) ||
924 (cluster_offset & QCOW_OFLAG_COMPRESSED))
925 break;
926
927 j = count_contiguous_clusters(nb_clusters - i, s->cluster_size,
928 &l2_table[l2_index + i], 0);
929
930 if (j)
931 free_any_clusters(bs, cluster_offset, j);
932
933 i += j;
934
935 if(be64_to_cpu(l2_table[l2_index + i]))
936 break;
937 }
938 nb_clusters = i;
939
940 /* allocate a new cluster */
941
942 cluster_offset = alloc_clusters(bs, nb_clusters * s->cluster_size);
943
944 /* we must initialize the cluster content which won't be
945 written */
946
947 nb_available = nb_clusters << (s->cluster_bits - 9);
948 if (nb_available > n_end)
949 nb_available = n_end;
950
951 /* copy content of unmodified sectors */
952
953 start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
954 if (n_start) {
955 ret = copy_sectors(bs, start_sect, cluster_offset, 0, n_start);
956 if (ret < 0)
957 return 0;
958 }
959
960 if (nb_available & (s->cluster_sectors - 1)) {
961 uint64_t end = nb_available & ~(uint64_t)(s->cluster_sectors - 1);
962 ret = copy_sectors(bs, start_sect + end,
963 cluster_offset + (end << 9),
964 nb_available - end,
965 s->cluster_sectors);
966 if (ret < 0)
967 return 0;
968 }
969
970 /* update L2 table */
971
972 for (i = 0; i < nb_clusters; i++)
973 l2_table[l2_index + i] = cpu_to_be64((cluster_offset +
974 (i << s->cluster_bits)) |
975 QCOW_OFLAG_COPIED);
976
977 if (bdrv_pwrite(s->hd,
978 l2_offset + l2_index * sizeof(uint64_t),
979 l2_table + l2_index,
980 nb_clusters * sizeof(uint64_t)) !=
981 nb_clusters * sizeof(uint64_t))
982 return 0;
983
984 out:
985 *num = nb_available - n_start;
986
987 return cluster_offset;
988 }
989
990 static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
991 int nb_sectors, int *pnum)
992 {
993 uint64_t cluster_offset;
994
995 *pnum = nb_sectors;
996 cluster_offset = get_cluster_offset(bs, sector_num << 9, pnum);
997
998 return (cluster_offset != 0);
999 }
1000
1001 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
1002 const uint8_t *buf, int buf_size)
1003 {
1004 z_stream strm1, *strm = &strm1;
1005 int ret, out_len;
1006
1007 memset(strm, 0, sizeof(*strm));
1008
1009 strm->next_in = (uint8_t *)buf;
1010 strm->avail_in = buf_size;
1011 strm->next_out = out_buf;
1012 strm->avail_out = out_buf_size;
1013
1014 ret = inflateInit2(strm, -12);
1015 if (ret != Z_OK)
1016 return -1;
1017 ret = inflate(strm, Z_FINISH);
1018 out_len = strm->next_out - out_buf;
1019 if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
1020 out_len != out_buf_size) {
1021 inflateEnd(strm);
1022 return -1;
1023 }
1024 inflateEnd(strm);
1025 return 0;
1026 }
1027
1028 static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset)
1029 {
1030 int ret, csize, nb_csectors, sector_offset;
1031 uint64_t coffset;
1032
1033 coffset = cluster_offset & s->cluster_offset_mask;
1034 if (s->cluster_cache_offset != coffset) {
1035 nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
1036 sector_offset = coffset & 511;
1037 csize = nb_csectors * 512 - sector_offset;
1038 ret = bdrv_read(s->hd, coffset >> 9, s->cluster_data, nb_csectors);
1039 if (ret < 0) {
1040 return -1;
1041 }
1042 if (decompress_buffer(s->cluster_cache, s->cluster_size,
1043 s->cluster_data + sector_offset, csize) < 0) {
1044 return -1;
1045 }
1046 s->cluster_cache_offset = coffset;
1047 }
1048 return 0;
1049 }
1050
1051 /* handle reading after the end of the backing file */
1052 static int backing_read1(BlockDriverState *bs,
1053 int64_t sector_num, uint8_t *buf, int nb_sectors)
1054 {
1055 int n1;
1056 if ((sector_num + nb_sectors) <= bs->total_sectors)
1057 return nb_sectors;
1058 if (sector_num >= bs->total_sectors)
1059 n1 = 0;
1060 else
1061 n1 = bs->total_sectors - sector_num;
1062 memset(buf + n1 * 512, 0, 512 * (nb_sectors - n1));
1063 return n1;
1064 }
1065
1066 static int qcow_read(BlockDriverState *bs, int64_t sector_num,
1067 uint8_t *buf, int nb_sectors)
1068 {
1069 BDRVQcowState *s = bs->opaque;
1070 int ret, index_in_cluster, n, n1;
1071 uint64_t cluster_offset;
1072
1073 while (nb_sectors > 0) {
1074 n = nb_sectors;
1075 cluster_offset = get_cluster_offset(bs, sector_num << 9, &n);
1076 index_in_cluster = sector_num & (s->cluster_sectors - 1);
1077 if (!cluster_offset) {
1078 if (bs->backing_hd) {
1079 /* read from the base image */
1080 n1 = backing_read1(bs->backing_hd, sector_num, buf, n);
1081 if (n1 > 0) {
1082 ret = bdrv_read(bs->backing_hd, sector_num, buf, n1);
1083 if (ret < 0)
1084 return -1;
1085 }
1086 } else {
1087 memset(buf, 0, 512 * n);
1088 }
1089 } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
1090 if (decompress_cluster(s, cluster_offset) < 0)
1091 return -1;
1092 memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n);
1093 } else {
1094 ret = bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
1095 if (ret != n * 512)
1096 return -1;
1097 if (s->crypt_method) {
1098 encrypt_sectors(s, sector_num, buf, buf, n, 0,
1099 &s->aes_decrypt_key);
1100 }
1101 }
1102 nb_sectors -= n;
1103 sector_num += n;
1104 buf += n * 512;
1105 }
1106 return 0;
1107 }
1108
1109 static int qcow_write(BlockDriverState *bs, int64_t sector_num,
1110 const uint8_t *buf, int nb_sectors)
1111 {
1112 BDRVQcowState *s = bs->opaque;
1113 int ret, index_in_cluster, n;
1114 uint64_t cluster_offset;
1115 int n_end;
1116
1117 while (nb_sectors > 0) {
1118 index_in_cluster = sector_num & (s->cluster_sectors - 1);
1119 n_end = index_in_cluster + nb_sectors;
1120 if (s->crypt_method &&
1121 n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
1122 n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
1123 cluster_offset = alloc_cluster_offset(bs, sector_num << 9,
1124 index_in_cluster,
1125 n_end, &n);
1126 if (!cluster_offset)
1127 return -1;
1128 if (s->crypt_method) {
1129 encrypt_sectors(s, sector_num, s->cluster_data, buf, n, 1,
1130 &s->aes_encrypt_key);
1131 ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512,
1132 s->cluster_data, n * 512);
1133 } else {
1134 ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
1135 }
1136 if (ret != n * 512)
1137 return -1;
1138 nb_sectors -= n;
1139 sector_num += n;
1140 buf += n * 512;
1141 }
1142 s->cluster_cache_offset = -1; /* disable compressed cache */
1143 return 0;
1144 }
1145
1146 typedef struct QCowAIOCB {
1147 BlockDriverAIOCB common;
1148 int64_t sector_num;
1149 uint8_t *buf;
1150 int nb_sectors;
1151 int n;
1152 uint64_t cluster_offset;
1153 uint8_t *cluster_data;
1154 BlockDriverAIOCB *hd_aiocb;
1155 QEMUBH *bh;
1156 } QCowAIOCB;
1157
1158 static void qcow_aio_read_cb(void *opaque, int ret);
1159 static void qcow_aio_read_bh(void *opaque)
1160 {
1161 QCowAIOCB *acb = opaque;
1162 qemu_bh_delete(acb->bh);
1163 acb->bh = NULL;
1164 qcow_aio_read_cb(opaque, 0);
1165 }
1166
1167 static int qcow_schedule_bh(QEMUBHFunc *cb, QCowAIOCB *acb)
1168 {
1169 if (acb->bh)
1170 return -EIO;
1171
1172 acb->bh = qemu_bh_new(cb, acb);
1173 if (!acb->bh)
1174 return -EIO;
1175
1176 qemu_bh_schedule(acb->bh);
1177
1178 return 0;
1179 }
1180
1181 static void qcow_aio_read_cb(void *opaque, int ret)
1182 {
1183 QCowAIOCB *acb = opaque;
1184 BlockDriverState *bs = acb->common.bs;
1185 BDRVQcowState *s = bs->opaque;
1186 int index_in_cluster, n1;
1187
1188 acb->hd_aiocb = NULL;
1189 if (ret < 0) {
1190 fail:
1191 acb->common.cb(acb->common.opaque, ret);
1192 qemu_aio_release(acb);
1193 return;
1194 }
1195
1196 /* post process the read buffer */
1197 if (!acb->cluster_offset) {
1198 /* nothing to do */
1199 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
1200 /* nothing to do */
1201 } else {
1202 if (s->crypt_method) {
1203 encrypt_sectors(s, acb->sector_num, acb->buf, acb->buf,
1204 acb->n, 0,
1205 &s->aes_decrypt_key);
1206 }
1207 }
1208
1209 acb->nb_sectors -= acb->n;
1210 acb->sector_num += acb->n;
1211 acb->buf += acb->n * 512;
1212
1213 if (acb->nb_sectors == 0) {
1214 /* request completed */
1215 acb->common.cb(acb->common.opaque, 0);
1216 qemu_aio_release(acb);
1217 return;
1218 }
1219
1220 /* prepare next AIO request */
1221 acb->n = acb->nb_sectors;
1222 acb->cluster_offset = get_cluster_offset(bs, acb->sector_num << 9, &acb->n);
1223 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
1224
1225 if (!acb->cluster_offset) {
1226 if (bs->backing_hd) {
1227 /* read from the base image */
1228 n1 = backing_read1(bs->backing_hd, acb->sector_num,
1229 acb->buf, acb->n);
1230 if (n1 > 0) {
1231 acb->hd_aiocb = bdrv_aio_read(bs->backing_hd, acb->sector_num,
1232 acb->buf, acb->n, qcow_aio_read_cb, acb);
1233 if (acb->hd_aiocb == NULL)
1234 goto fail;
1235 } else {
1236 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
1237 if (ret < 0)
1238 goto fail;
1239 }
1240 } else {
1241 /* Note: in this case, no need to wait */
1242 memset(acb->buf, 0, 512 * acb->n);
1243 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
1244 if (ret < 0)
1245 goto fail;
1246 }
1247 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
1248 /* add AIO support for compressed blocks ? */
1249 if (decompress_cluster(s, acb->cluster_offset) < 0)
1250 goto fail;
1251 memcpy(acb->buf,
1252 s->cluster_cache + index_in_cluster * 512, 512 * acb->n);
1253 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
1254 if (ret < 0)
1255 goto fail;
1256 } else {
1257 if ((acb->cluster_offset & 511) != 0) {
1258 ret = -EIO;
1259 goto fail;
1260 }
1261 acb->hd_aiocb = bdrv_aio_read(s->hd,
1262 (acb->cluster_offset >> 9) + index_in_cluster,
1263 acb->buf, acb->n, qcow_aio_read_cb, acb);
1264 if (acb->hd_aiocb == NULL)
1265 goto fail;
1266 }
1267 }
1268
1269 static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs,
1270 int64_t sector_num, uint8_t *buf, int nb_sectors,
1271 BlockDriverCompletionFunc *cb, void *opaque)
1272 {
1273 QCowAIOCB *acb;
1274
1275 acb = qemu_aio_get(bs, cb, opaque);
1276 if (!acb)
1277 return NULL;
1278 acb->hd_aiocb = NULL;
1279 acb->sector_num = sector_num;
1280 acb->buf = buf;
1281 acb->nb_sectors = nb_sectors;
1282 acb->n = 0;
1283 acb->cluster_offset = 0;
1284 return acb;
1285 }
1286
1287 static BlockDriverAIOCB *qcow_aio_read(BlockDriverState *bs,
1288 int64_t sector_num, uint8_t *buf, int nb_sectors,
1289 BlockDriverCompletionFunc *cb, void *opaque)
1290 {
1291 QCowAIOCB *acb;
1292
1293 acb = qcow_aio_setup(bs, sector_num, buf, nb_sectors, cb, opaque);
1294 if (!acb)
1295 return NULL;
1296
1297 qcow_aio_read_cb(acb, 0);
1298 return &acb->common;
1299 }
1300
1301 static void qcow_aio_write_cb(void *opaque, int ret)
1302 {
1303 QCowAIOCB *acb = opaque;
1304 BlockDriverState *bs = acb->common.bs;
1305 BDRVQcowState *s = bs->opaque;
1306 int index_in_cluster;
1307 uint64_t cluster_offset;
1308 const uint8_t *src_buf;
1309 int n_end;
1310
1311 acb->hd_aiocb = NULL;
1312
1313 if (ret < 0) {
1314 fail:
1315 acb->common.cb(acb->common.opaque, ret);
1316 qemu_aio_release(acb);
1317 return;
1318 }
1319
1320 acb->nb_sectors -= acb->n;
1321 acb->sector_num += acb->n;
1322 acb->buf += acb->n * 512;
1323
1324 if (acb->nb_sectors == 0) {
1325 /* request completed */
1326 acb->common.cb(acb->common.opaque, 0);
1327 qemu_aio_release(acb);
1328 return;
1329 }
1330
1331 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
1332 n_end = index_in_cluster + acb->nb_sectors;
1333 if (s->crypt_method &&
1334 n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
1335 n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
1336
1337 cluster_offset = alloc_cluster_offset(bs, acb->sector_num << 9,
1338 index_in_cluster,
1339 n_end, &acb->n);
1340 if (!cluster_offset || (cluster_offset & 511) != 0) {
1341 ret = -EIO;
1342 goto fail;
1343 }
1344 if (s->crypt_method) {
1345 if (!acb->cluster_data) {
1346 acb->cluster_data = qemu_mallocz(QCOW_MAX_CRYPT_CLUSTERS *
1347 s->cluster_size);
1348 if (!acb->cluster_data) {
1349 ret = -ENOMEM;
1350 goto fail;
1351 }
1352 }
1353 encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf,
1354 acb->n, 1, &s->aes_encrypt_key);
1355 src_buf = acb->cluster_data;
1356 } else {
1357 src_buf = acb->buf;
1358 }
1359 acb->hd_aiocb = bdrv_aio_write(s->hd,
1360 (cluster_offset >> 9) + index_in_cluster,
1361 src_buf, acb->n,
1362 qcow_aio_write_cb, acb);
1363 if (acb->hd_aiocb == NULL)
1364 goto fail;
1365 }
1366
1367 static BlockDriverAIOCB *qcow_aio_write(BlockDriverState *bs,
1368 int64_t sector_num, const uint8_t *buf, int nb_sectors,
1369 BlockDriverCompletionFunc *cb, void *opaque)
1370 {
1371 BDRVQcowState *s = bs->opaque;
1372 QCowAIOCB *acb;
1373
1374 s->cluster_cache_offset = -1; /* disable compressed cache */
1375
1376 acb = qcow_aio_setup(bs, sector_num, (uint8_t*)buf, nb_sectors, cb, opaque);
1377 if (!acb)
1378 return NULL;
1379
1380 qcow_aio_write_cb(acb, 0);
1381 return &acb->common;
1382 }
1383
1384 static void qcow_aio_cancel(BlockDriverAIOCB *blockacb)
1385 {
1386 QCowAIOCB *acb = (QCowAIOCB *)blockacb;
1387 if (acb->hd_aiocb)
1388 bdrv_aio_cancel(acb->hd_aiocb);
1389 qemu_aio_release(acb);
1390 }
1391
1392 static void qcow_close(BlockDriverState *bs)
1393 {
1394 BDRVQcowState *s = bs->opaque;
1395 qemu_free(s->l1_table);
1396 qemu_free(s->l2_cache);
1397 qemu_free(s->cluster_cache);
1398 qemu_free(s->cluster_data);
1399 refcount_close(bs);
1400 bdrv_delete(s->hd);
1401 }
1402
1403 /* XXX: use std qcow open function ? */
1404 typedef struct QCowCreateState {
1405 int cluster_size;
1406 int cluster_bits;
1407 uint16_t *refcount_block;
1408 uint64_t *refcount_table;
1409 int64_t l1_table_offset;
1410 int64_t refcount_table_offset;
1411 int64_t refcount_block_offset;
1412 } QCowCreateState;
1413
1414 static void create_refcount_update(QCowCreateState *s,
1415 int64_t offset, int64_t size)
1416 {
1417 int refcount;
1418 int64_t start, last, cluster_offset;
1419 uint16_t *p;
1420
1421 start = offset & ~(s->cluster_size - 1);
1422 last = (offset + size - 1) & ~(s->cluster_size - 1);
1423 for(cluster_offset = start; cluster_offset <= last;
1424 cluster_offset += s->cluster_size) {
1425 p = &s->refcount_block[cluster_offset >> s->cluster_bits];
1426 refcount = be16_to_cpu(*p);
1427 refcount++;
1428 *p = cpu_to_be16(refcount);
1429 }
1430 }
1431
1432 static int qcow_create(const char *filename, int64_t total_size,
1433 const char *backing_file, int flags)
1434 {
1435 int fd, header_size, backing_filename_len, l1_size, i, shift, l2_bits;
1436 QCowHeader header;
1437 uint64_t tmp, offset;
1438 QCowCreateState s1, *s = &s1;
1439
1440 memset(s, 0, sizeof(*s));
1441
1442 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
1443 if (fd < 0)
1444 return -1;
1445 memset(&header, 0, sizeof(header));
1446 header.magic = cpu_to_be32(QCOW_MAGIC);
1447 header.version = cpu_to_be32(QCOW_VERSION);
1448 header.size = cpu_to_be64(total_size * 512);
1449 header_size = sizeof(header);
1450 backing_filename_len = 0;
1451 if (backing_file) {
1452 header.backing_file_offset = cpu_to_be64(header_size);
1453 backing_filename_len = strlen(backing_file);
1454 header.backing_file_size = cpu_to_be32(backing_filename_len);
1455 header_size += backing_filename_len;
1456 }
1457 s->cluster_bits = 12; /* 4 KB clusters */
1458 s->cluster_size = 1 << s->cluster_bits;
1459 header.cluster_bits = cpu_to_be32(s->cluster_bits);
1460 header_size = (header_size + 7) & ~7;
1461 if (flags & BLOCK_FLAG_ENCRYPT) {
1462 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
1463 } else {
1464 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
1465 }
1466 l2_bits = s->cluster_bits - 3;
1467 shift = s->cluster_bits + l2_bits;
1468 l1_size = (((total_size * 512) + (1LL << shift) - 1) >> shift);
1469 offset = align_offset(header_size, s->cluster_size);
1470 s->l1_table_offset = offset;
1471 header.l1_table_offset = cpu_to_be64(s->l1_table_offset);
1472 header.l1_size = cpu_to_be32(l1_size);
1473 offset += align_offset(l1_size * sizeof(uint64_t), s->cluster_size);
1474
1475 s->refcount_table = qemu_mallocz(s->cluster_size);
1476 if (!s->refcount_table)
1477 goto fail;
1478 s->refcount_block = qemu_mallocz(s->cluster_size);
1479 if (!s->refcount_block)
1480 goto fail;
1481
1482 s->refcount_table_offset = offset;
1483 header.refcount_table_offset = cpu_to_be64(offset);
1484 header.refcount_table_clusters = cpu_to_be32(1);
1485 offset += s->cluster_size;
1486
1487 s->refcount_table[0] = cpu_to_be64(offset);
1488 s->refcount_block_offset = offset;
1489 offset += s->cluster_size;
1490
1491 /* update refcounts */
1492 create_refcount_update(s, 0, header_size);
1493 create_refcount_update(s, s->l1_table_offset, l1_size * sizeof(uint64_t));
1494 create_refcount_update(s, s->refcount_table_offset, s->cluster_size);
1495 create_refcount_update(s, s->refcount_block_offset, s->cluster_size);
1496
1497 /* write all the data */
1498 write(fd, &header, sizeof(header));
1499 if (backing_file) {
1500 write(fd, backing_file, backing_filename_len);
1501 }
1502 lseek(fd, s->l1_table_offset, SEEK_SET);
1503 tmp = 0;
1504 for(i = 0;i < l1_size; i++) {
1505 write(fd, &tmp, sizeof(tmp));
1506 }
1507 lseek(fd, s->refcount_table_offset, SEEK_SET);
1508 write(fd, s->refcount_table, s->cluster_size);
1509
1510 lseek(fd, s->refcount_block_offset, SEEK_SET);
1511 write(fd, s->refcount_block, s->cluster_size);
1512
1513 qemu_free(s->refcount_table);
1514 qemu_free(s->refcount_block);
1515 close(fd);
1516 return 0;
1517 fail:
1518 qemu_free(s->refcount_table);
1519 qemu_free(s->refcount_block);
1520 close(fd);
1521 return -ENOMEM;
1522 }
1523
1524 static int qcow_make_empty(BlockDriverState *bs)
1525 {
1526 #if 0
1527 /* XXX: not correct */
1528 BDRVQcowState *s = bs->opaque;
1529 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1530 int ret;
1531
1532 memset(s->l1_table, 0, l1_length);
1533 if (bdrv_pwrite(s->hd, s->l1_table_offset, s->l1_table, l1_length) < 0)
1534 return -1;
1535 ret = bdrv_truncate(s->hd, s->l1_table_offset + l1_length);
1536 if (ret < 0)
1537 return ret;
1538
1539 l2_cache_reset(bs);
1540 #endif
1541 return 0;
1542 }
1543
1544 /* XXX: put compressed sectors first, then all the cluster aligned
1545 tables to avoid losing bytes in alignment */
1546 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
1547 const uint8_t *buf, int nb_sectors)
1548 {
1549 BDRVQcowState *s = bs->opaque;
1550 z_stream strm;
1551 int ret, out_len;
1552 uint8_t *out_buf;
1553 uint64_t cluster_offset;
1554
1555 if (nb_sectors == 0) {
1556 /* align end of file to a sector boundary to ease reading with
1557 sector based I/Os */
1558 cluster_offset = bdrv_getlength(s->hd);
1559 cluster_offset = (cluster_offset + 511) & ~511;
1560 bdrv_truncate(s->hd, cluster_offset);
1561 return 0;
1562 }
1563
1564 if (nb_sectors != s->cluster_sectors)
1565 return -EINVAL;
1566
1567 out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
1568 if (!out_buf)
1569 return -ENOMEM;
1570
1571 /* best compression, small window, no zlib header */
1572 memset(&strm, 0, sizeof(strm));
1573 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1574 Z_DEFLATED, -12,
1575 9, Z_DEFAULT_STRATEGY);
1576 if (ret != 0) {
1577 qemu_free(out_buf);
1578 return -1;
1579 }
1580
1581 strm.avail_in = s->cluster_size;
1582 strm.next_in = (uint8_t *)buf;
1583 strm.avail_out = s->cluster_size;
1584 strm.next_out = out_buf;
1585
1586 ret = deflate(&strm, Z_FINISH);
1587 if (ret != Z_STREAM_END && ret != Z_OK) {
1588 qemu_free(out_buf);
1589 deflateEnd(&strm);
1590 return -1;
1591 }
1592 out_len = strm.next_out - out_buf;
1593
1594 deflateEnd(&strm);
1595
1596 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1597 /* could not compress: write normal cluster */
1598 qcow_write(bs, sector_num, buf, s->cluster_sectors);
1599 } else {
1600 cluster_offset = alloc_compressed_cluster_offset(bs, sector_num << 9,
1601 out_len);
1602 if (!cluster_offset)
1603 return -1;
1604 cluster_offset &= s->cluster_offset_mask;
1605 if (bdrv_pwrite(s->hd, cluster_offset, out_buf, out_len) != out_len) {
1606 qemu_free(out_buf);
1607 return -1;
1608 }
1609 }
1610
1611 qemu_free(out_buf);
1612 return 0;
1613 }
1614
1615 static void qcow_flush(BlockDriverState *bs)
1616 {
1617 BDRVQcowState *s = bs->opaque;
1618 bdrv_flush(s->hd);
1619 }
1620
1621 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1622 {
1623 BDRVQcowState *s = bs->opaque;
1624 bdi->cluster_size = s->cluster_size;
1625 bdi->vm_state_offset = (int64_t)s->l1_vm_state_index <<
1626 (s->cluster_bits + s->l2_bits);
1627 return 0;
1628 }
1629
1630 /*********************************************************/
1631 /* snapshot support */
1632
1633 /* update the refcounts of snapshots and the copied flag */
1634 static int update_snapshot_refcount(BlockDriverState *bs,
1635 int64_t l1_table_offset,
1636 int l1_size,
1637 int addend)
1638 {
1639 BDRVQcowState *s = bs->opaque;
1640 uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated;
1641 int64_t old_offset, old_l2_offset;
1642 int l2_size, i, j, l1_modified, l2_modified, nb_csectors, refcount;
1643
1644 l2_cache_reset(bs);
1645
1646 l2_table = NULL;
1647 l1_table = NULL;
1648 l1_size2 = l1_size * sizeof(uint64_t);
1649 l1_allocated = 0;
1650 if (l1_table_offset != s->l1_table_offset) {
1651 l1_table = qemu_malloc(l1_size2);
1652 if (!l1_table)
1653 goto fail;
1654 l1_allocated = 1;
1655 if (bdrv_pread(s->hd, l1_table_offset,
1656 l1_table, l1_size2) != l1_size2)
1657 goto fail;
1658 for(i = 0;i < l1_size; i++)
1659 be64_to_cpus(&l1_table[i]);
1660 } else {
1661 assert(l1_size == s->l1_size);
1662 l1_table = s->l1_table;
1663 l1_allocated = 0;
1664 }
1665
1666 l2_size = s->l2_size * sizeof(uint64_t);
1667 l2_table = qemu_malloc(l2_size);
1668 if (!l2_table)
1669 goto fail;
1670 l1_modified = 0;
1671 for(i = 0; i < l1_size; i++) {
1672 l2_offset = l1_table[i];
1673 if (l2_offset) {
1674 old_l2_offset = l2_offset;
1675 l2_offset &= ~QCOW_OFLAG_COPIED;
1676 l2_modified = 0;
1677 if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
1678 goto fail;
1679 for(j = 0; j < s->l2_size; j++) {
1680 offset = be64_to_cpu(l2_table[j]);
1681 if (offset != 0) {
1682 old_offset = offset;
1683 offset &= ~QCOW_OFLAG_COPIED;
1684 if (offset & QCOW_OFLAG_COMPRESSED) {
1685 nb_csectors = ((offset >> s->csize_shift) &
1686 s->csize_mask) + 1;
1687 if (addend != 0)
1688 update_refcount(bs, (offset & s->cluster_offset_mask) & ~511,
1689 nb_csectors * 512, addend);
1690 /* compressed clusters are never modified */
1691 refcount = 2;
1692 } else {
1693 if (addend != 0) {
1694 refcount = update_cluster_refcount(bs, offset >> s->cluster_bits, addend);
1695 } else {
1696 refcount = get_refcount(bs, offset >> s->cluster_bits);
1697 }
1698 }
1699
1700 if (refcount == 1) {
1701 offset |= QCOW_OFLAG_COPIED;
1702 }
1703 if (offset != old_offset) {
1704 l2_table[j] = cpu_to_be64(offset);
1705 l2_modified = 1;
1706 }
1707 }
1708 }
1709 if (l2_modified) {
1710 if (bdrv_pwrite(s->hd,
1711 l2_offset, l2_table, l2_size) != l2_size)
1712 goto fail;
1713 }
1714
1715 if (addend != 0) {
1716 refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend);
1717 } else {
1718 refcount = get_refcount(bs, l2_offset >> s->cluster_bits);
1719 }
1720 if (refcount == 1) {
1721 l2_offset |= QCOW_OFLAG_COPIED;
1722 }
1723 if (l2_offset != old_l2_offset) {
1724 l1_table[i] = l2_offset;
1725 l1_modified = 1;
1726 }
1727 }
1728 }
1729 if (l1_modified) {
1730 for(i = 0; i < l1_size; i++)
1731 cpu_to_be64s(&l1_table[i]);
1732 if (bdrv_pwrite(s->hd, l1_table_offset, l1_table,
1733 l1_size2) != l1_size2)
1734 goto fail;
1735 for(i = 0; i < l1_size; i++)
1736 be64_to_cpus(&l1_table[i]);
1737 }
1738 if (l1_allocated)
1739 qemu_free(l1_table);
1740 qemu_free(l2_table);
1741 return 0;
1742 fail:
1743 if (l1_allocated)
1744 qemu_free(l1_table);
1745 qemu_free(l2_table);
1746 return -EIO;
1747 }
1748
1749 static void qcow_free_snapshots(BlockDriverState *bs)
1750 {
1751 BDRVQcowState *s = bs->opaque;
1752 int i;
1753
1754 for(i = 0; i < s->nb_snapshots; i++) {
1755 qemu_free(s->snapshots[i].name);
1756 qemu_free(s->snapshots[i].id_str);
1757 }
1758 qemu_free(s->snapshots);
1759 s->snapshots = NULL;
1760 s->nb_snapshots = 0;
1761 }
1762
1763 static int qcow_read_snapshots(BlockDriverState *bs)
1764 {
1765 BDRVQcowState *s = bs->opaque;
1766 QCowSnapshotHeader h;
1767 QCowSnapshot *sn;
1768 int i, id_str_size, name_size;
1769 int64_t offset;
1770 uint32_t extra_data_size;
1771
1772 offset = s->snapshots_offset;
1773 s->snapshots = qemu_mallocz(s->nb_snapshots * sizeof(QCowSnapshot));
1774 if (!s->snapshots)
1775 goto fail;
1776 for(i = 0; i < s->nb_snapshots; i++) {
1777 offset = align_offset(offset, 8);
1778 if (bdrv_pread(s->hd, offset, &h, sizeof(h)) != sizeof(h))
1779 goto fail;
1780 offset += sizeof(h);
1781 sn = s->snapshots + i;
1782 sn->l1_table_offset = be64_to_cpu(h.l1_table_offset);
1783 sn->l1_size = be32_to_cpu(h.l1_size);
1784 sn->vm_state_size = be32_to_cpu(h.vm_state_size);
1785 sn->date_sec = be32_to_cpu(h.date_sec);
1786 sn->date_nsec = be32_to_cpu(h.date_nsec);
1787 sn->vm_clock_nsec = be64_to_cpu(h.vm_clock_nsec);
1788 extra_data_size = be32_to_cpu(h.extra_data_size);
1789
1790 id_str_size = be16_to_cpu(h.id_str_size);
1791 name_size = be16_to_cpu(h.name_size);
1792
1793 offset += extra_data_size;
1794
1795 sn->id_str = qemu_malloc(id_str_size + 1);
1796 if (!sn->id_str)
1797 goto fail;
1798 if (bdrv_pread(s->hd, offset, sn->id_str, id_str_size) != id_str_size)
1799 goto fail;
1800 offset += id_str_size;
1801 sn->id_str[id_str_size] = '\0';
1802
1803 sn->name = qemu_malloc(name_size + 1);
1804 if (!sn->name)
1805 goto fail;
1806 if (bdrv_pread(s->hd, offset, sn->name, name_size) != name_size)
1807 goto fail;
1808 offset += name_size;
1809 sn->name[name_size] = '\0';
1810 }
1811 s->snapshots_size = offset - s->snapshots_offset;
1812 return 0;
1813 fail:
1814 qcow_free_snapshots(bs);
1815 return -1;
1816 }
1817
1818 /* add at the end of the file a new list of snapshots */
1819 static int qcow_write_snapshots(BlockDriverState *bs)
1820 {
1821 BDRVQcowState *s = bs->opaque;
1822 QCowSnapshot *sn;
1823 QCowSnapshotHeader h;
1824 int i, name_size, id_str_size, snapshots_size;
1825 uint64_t data64;
1826 uint32_t data32;
1827 int64_t offset, snapshots_offset;
1828
1829 /* compute the size of the snapshots */
1830 offset = 0;
1831 for(i = 0; i < s->nb_snapshots; i++) {
1832 sn = s->snapshots + i;
1833 offset = align_offset(offset, 8);
1834 offset += sizeof(h);
1835 offset += strlen(sn->id_str);
1836 offset += strlen(sn->name);
1837 }
1838 snapshots_size = offset;
1839
1840 snapshots_offset = alloc_clusters(bs, snapshots_size);
1841 offset = snapshots_offset;
1842
1843 for(i = 0; i < s->nb_snapshots; i++) {
1844 sn = s->snapshots + i;
1845 memset(&h, 0, sizeof(h));
1846 h.l1_table_offset = cpu_to_be64(sn->l1_table_offset);
1847 h.l1_size = cpu_to_be32(sn->l1_size);
1848 h.vm_state_size = cpu_to_be32(sn->vm_state_size);
1849 h.date_sec = cpu_to_be32(sn->date_sec);
1850 h.date_nsec = cpu_to_be32(sn->date_nsec);
1851 h.vm_clock_nsec = cpu_to_be64(sn->vm_clock_nsec);
1852
1853 id_str_size = strlen(sn->id_str);
1854 name_size = strlen(sn->name);
1855 h.id_str_size = cpu_to_be16(id_str_size);
1856 h.name_size = cpu_to_be16(name_size);
1857 offset = align_offset(offset, 8);
1858 if (bdrv_pwrite(s->hd, offset, &h, sizeof(h)) != sizeof(h))
1859 goto fail;
1860 offset += sizeof(h);
1861 if (bdrv_pwrite(s->hd, offset, sn->id_str, id_str_size) != id_str_size)
1862 goto fail;
1863 offset += id_str_size;
1864 if (bdrv_pwrite(s->hd, offset, sn->name, name_size) != name_size)
1865 goto fail;
1866 offset += name_size;
1867 }
1868
1869 /* update the various header fields */
1870 data64 = cpu_to_be64(snapshots_offset);
1871 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, snapshots_offset),
1872 &data64, sizeof(data64)) != sizeof(data64))
1873 goto fail;
1874 data32 = cpu_to_be32(s->nb_snapshots);
1875 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, nb_snapshots),
1876 &data32, sizeof(data32)) != sizeof(data32))
1877 goto fail;
1878
1879 /* free the old snapshot table */
1880 free_clusters(bs, s->snapshots_offset, s->snapshots_size);
1881 s->snapshots_offset = snapshots_offset;
1882 s->snapshots_size = snapshots_size;
1883 return 0;
1884 fail:
1885 return -1;
1886 }
1887
1888 static void find_new_snapshot_id(BlockDriverState *bs,
1889 char *id_str, int id_str_size)
1890 {
1891 BDRVQcowState *s = bs->opaque;
1892 QCowSnapshot *sn;
1893 int i, id, id_max = 0;
1894
1895 for(i = 0; i < s->nb_snapshots; i++) {
1896 sn = s->snapshots + i;
1897 id = strtoul(sn->id_str, NULL, 10);
1898 if (id > id_max)
1899 id_max = id;
1900 }
1901 snprintf(id_str, id_str_size, "%d", id_max + 1);
1902 }
1903
1904 static int find_snapshot_by_id(BlockDriverState *bs, const char *id_str)
1905 {
1906 BDRVQcowState *s = bs->opaque;
1907 int i;
1908
1909 for(i = 0; i < s->nb_snapshots; i++) {
1910 if (!strcmp(s->snapshots[i].id_str, id_str))
1911 return i;
1912 }
1913 return -1;
1914 }
1915
1916 static int find_snapshot_by_id_or_name(BlockDriverState *bs, const char *name)
1917 {
1918 BDRVQcowState *s = bs->opaque;
1919 int i, ret;
1920
1921 ret = find_snapshot_by_id(bs, name);
1922 if (ret >= 0)
1923 return ret;
1924 for(i = 0; i < s->nb_snapshots; i++) {
1925 if (!strcmp(s->snapshots[i].name, name))
1926 return i;
1927 }
1928 return -1;
1929 }
1930
1931 /* if no id is provided, a new one is constructed */
1932 static int qcow_snapshot_create(BlockDriverState *bs,
1933 QEMUSnapshotInfo *sn_info)
1934 {
1935 BDRVQcowState *s = bs->opaque;
1936 QCowSnapshot *snapshots1, sn1, *sn = &sn1;
1937 int i, ret;
1938 uint64_t *l1_table = NULL;
1939
1940 memset(sn, 0, sizeof(*sn));
1941
1942 if (sn_info->id_str[0] == '\0') {
1943 /* compute a new id */
1944 find_new_snapshot_id(bs, sn_info->id_str, sizeof(sn_info->id_str));
1945 }
1946
1947 /* check that the ID is unique */
1948 if (find_snapshot_by_id(bs, sn_info->id_str) >= 0)
1949 return -ENOENT;
1950
1951 sn->id_str = qemu_strdup(sn_info->id_str);
1952 if (!sn->id_str)
1953 goto fail;
1954 sn->name = qemu_strdup(sn_info->name);
1955 if (!sn->name)
1956 goto fail;
1957 sn->vm_state_size = sn_info->vm_state_size;
1958 sn->date_sec = sn_info->date_sec;
1959 sn->date_nsec = sn_info->date_nsec;
1960 sn->vm_clock_nsec = sn_info->vm_clock_nsec;
1961
1962 ret = update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1);
1963 if (ret < 0)
1964 goto fail;
1965
1966 /* create the L1 table of the snapshot */
1967 sn->l1_table_offset = alloc_clusters(bs, s->l1_size * sizeof(uint64_t));
1968 sn->l1_size = s->l1_size;
1969
1970 l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
1971 if (!l1_table)
1972 goto fail;
1973 for(i = 0; i < s->l1_size; i++) {
1974 l1_table[i] = cpu_to_be64(s->l1_table[i]);
1975 }
1976 if (bdrv_pwrite(s->hd, sn->l1_table_offset,
1977 l1_table, s->l1_size * sizeof(uint64_t)) !=
1978 (s->l1_size * sizeof(uint64_t)))
1979 goto fail;
1980 qemu_free(l1_table);
1981 l1_table = NULL;
1982
1983 snapshots1 = qemu_malloc((s->nb_snapshots + 1) * sizeof(QCowSnapshot));
1984 if (!snapshots1)
1985 goto fail;
1986 memcpy(snapshots1, s->snapshots, s->nb_snapshots * sizeof(QCowSnapshot));
1987 s->snapshots = snapshots1;
1988 s->snapshots[s->nb_snapshots++] = *sn;
1989
1990 if (qcow_write_snapshots(bs) < 0)
1991 goto fail;
1992 #ifdef DEBUG_ALLOC
1993 check_refcounts(bs);
1994 #endif
1995 return 0;
1996 fail:
1997 qemu_free(sn->name);
1998 qemu_free(l1_table);
1999 return -1;
2000 }
2001
2002 /* copy the snapshot 'snapshot_name' into the current disk image */
2003 static int qcow_snapshot_goto(BlockDriverState *bs,
2004 const char *snapshot_id)
2005 {
2006 BDRVQcowState *s = bs->opaque;
2007 QCowSnapshot *sn;
2008 int i, snapshot_index, l1_size2;
2009
2010 snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
2011 if (snapshot_index < 0)
2012 return -ENOENT;
2013 sn = &s->snapshots[snapshot_index];
2014
2015 if (update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, -1) < 0)
2016 goto fail;
2017
2018 if (grow_l1_table(bs, sn->l1_size) < 0)
2019 goto fail;
2020
2021 s->l1_size = sn->l1_size;
2022 l1_size2 = s->l1_size * sizeof(uint64_t);
2023 /* copy the snapshot l1 table to the current l1 table */
2024 if (bdrv_pread(s->hd, sn->l1_table_offset,
2025 s->l1_table, l1_size2) != l1_size2)
2026 goto fail;
2027 if (bdrv_pwrite(s->hd, s->l1_table_offset,
2028 s->l1_table, l1_size2) != l1_size2)
2029 goto fail;
2030 for(i = 0;i < s->l1_size; i++) {
2031 be64_to_cpus(&s->l1_table[i]);
2032 }
2033
2034 if (update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1) < 0)
2035 goto fail;
2036
2037 #ifdef DEBUG_ALLOC
2038 check_refcounts(bs);
2039 #endif
2040 return 0;
2041 fail:
2042 return -EIO;
2043 }
2044
2045 static int qcow_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
2046 {
2047 BDRVQcowState *s = bs->opaque;
2048 QCowSnapshot *sn;
2049 int snapshot_index, ret;
2050
2051 snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
2052 if (snapshot_index < 0)
2053 return -ENOENT;
2054 sn = &s->snapshots[snapshot_index];
2055
2056 ret = update_snapshot_refcount(bs, sn->l1_table_offset, sn->l1_size, -1);
2057 if (ret < 0)
2058 return ret;
2059 /* must update the copied flag on the current cluster offsets */
2060 ret = update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 0);
2061 if (ret < 0)
2062 return ret;
2063 free_clusters(bs, sn->l1_table_offset, sn->l1_size * sizeof(uint64_t));
2064
2065 qemu_free(sn->id_str);
2066 qemu_free(sn->name);
2067 memmove(sn, sn + 1, (s->nb_snapshots - snapshot_index - 1) * sizeof(*sn));
2068 s->nb_snapshots--;
2069 ret = qcow_write_snapshots(bs);
2070 if (ret < 0) {
2071 /* XXX: restore snapshot if error ? */
2072 return ret;
2073 }
2074 #ifdef DEBUG_ALLOC
2075 check_refcounts(bs);
2076 #endif
2077 return 0;
2078 }
2079
2080 static int qcow_snapshot_list(BlockDriverState *bs,
2081 QEMUSnapshotInfo **psn_tab)
2082 {
2083 BDRVQcowState *s = bs->opaque;
2084 QEMUSnapshotInfo *sn_tab, *sn_info;
2085 QCowSnapshot *sn;
2086 int i;
2087
2088 sn_tab = qemu_mallocz(s->nb_snapshots * sizeof(QEMUSnapshotInfo));
2089 if (!sn_tab)
2090 goto fail;
2091 for(i = 0; i < s->nb_snapshots; i++) {
2092 sn_info = sn_tab + i;
2093 sn = s->snapshots + i;
2094 pstrcpy(sn_info->id_str, sizeof(sn_info->id_str),
2095 sn->id_str);
2096 pstrcpy(sn_info->name, sizeof(sn_info->name),
2097 sn->name);
2098 sn_info->vm_state_size = sn->vm_state_size;
2099 sn_info->date_sec = sn->date_sec;
2100 sn_info->date_nsec = sn->date_nsec;
2101 sn_info->vm_clock_nsec = sn->vm_clock_nsec;
2102 }
2103 *psn_tab = sn_tab;
2104 return s->nb_snapshots;
2105 fail:
2106 qemu_free(sn_tab);
2107 *psn_tab = NULL;
2108 return -ENOMEM;
2109 }
2110
2111 /*********************************************************/
2112 /* refcount handling */
2113
2114 static int refcount_init(BlockDriverState *bs)
2115 {
2116 BDRVQcowState *s = bs->opaque;
2117 int ret, refcount_table_size2, i;
2118
2119 s->refcount_block_cache = qemu_malloc(s->cluster_size);
2120 if (!s->refcount_block_cache)
2121 goto fail;
2122 refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t);
2123 s->refcount_table = qemu_malloc(refcount_table_size2);
2124 if (!s->refcount_table)
2125 goto fail;
2126 if (s->refcount_table_size > 0) {
2127 ret = bdrv_pread(s->hd, s->refcount_table_offset,
2128 s->refcount_table, refcount_table_size2);
2129 if (ret != refcount_table_size2)
2130 goto fail;
2131 for(i = 0; i < s->refcount_table_size; i++)
2132 be64_to_cpus(&s->refcount_table[i]);
2133 }
2134 return 0;
2135 fail:
2136 return -ENOMEM;
2137 }
2138
2139 static void refcount_close(BlockDriverState *bs)
2140 {
2141 BDRVQcowState *s = bs->opaque;
2142 qemu_free(s->refcount_block_cache);
2143 qemu_free(s->refcount_table);
2144 }
2145
2146
2147 static int load_refcount_block(BlockDriverState *bs,
2148 int64_t refcount_block_offset)
2149 {
2150 BDRVQcowState *s = bs->opaque;
2151 int ret;
2152 ret = bdrv_pread(s->hd, refcount_block_offset, s->refcount_block_cache,
2153 s->cluster_size);
2154 if (ret != s->cluster_size)
2155 return -EIO;
2156 s->refcount_block_cache_offset = refcount_block_offset;
2157 return 0;
2158 }
2159
2160 static int get_refcount(BlockDriverState *bs, int64_t cluster_index)
2161 {
2162 BDRVQcowState *s = bs->opaque;
2163 int refcount_table_index, block_index;
2164 int64_t refcount_block_offset;
2165
2166 refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
2167 if (refcount_table_index >= s->refcount_table_size)
2168 return 0;
2169 refcount_block_offset = s->refcount_table[refcount_table_index];
2170 if (!refcount_block_offset)
2171 return 0;
2172 if (refcount_block_offset != s->refcount_block_cache_offset) {
2173 /* better than nothing: return allocated if read error */
2174 if (load_refcount_block(bs, refcount_block_offset) < 0)
2175 return 1;
2176 }
2177 block_index = cluster_index &
2178 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
2179 return be16_to_cpu(s->refcount_block_cache[block_index]);
2180 }
2181
2182 /* return < 0 if error */
2183 static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size)
2184 {
2185 BDRVQcowState *s = bs->opaque;
2186 int i, nb_clusters;
2187
2188 nb_clusters = size_to_clusters(s, size);
2189 retry:
2190 for(i = 0; i < nb_clusters; i++) {
2191 int64_t i = s->free_cluster_index++;
2192 if (get_refcount(bs, i) != 0)
2193 goto retry;
2194 }
2195 #ifdef DEBUG_ALLOC2
2196 printf("alloc_clusters: size=%lld -> %lld\n",
2197 size,
2198 (s->free_cluster_index - nb_clusters) << s->cluster_bits);
2199 #endif
2200 return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
2201 }
2202
2203 static int64_t alloc_clusters(BlockDriverState *bs, int64_t size)
2204 {
2205 int64_t offset;
2206
2207 offset = alloc_clusters_noref(bs, size);
2208 update_refcount(bs, offset, size, 1);
2209 return offset;
2210 }
2211
2212 /* only used to allocate compressed sectors. We try to allocate
2213 contiguous sectors. size must be <= cluster_size */
2214 static int64_t alloc_bytes(BlockDriverState *bs, int size)
2215 {
2216 BDRVQcowState *s = bs->opaque;
2217 int64_t offset, cluster_offset;
2218 int free_in_cluster;
2219
2220 assert(size > 0 && size <= s->cluster_size);
2221 if (s->free_byte_offset == 0) {
2222 s->free_byte_offset = alloc_clusters(bs, s->cluster_size);
2223 }
2224 redo:
2225 free_in_cluster = s->cluster_size -
2226 (s->free_byte_offset & (s->cluster_size - 1));
2227 if (size <= free_in_cluster) {
2228 /* enough space in current cluster */
2229 offset = s->free_byte_offset;
2230 s->free_byte_offset += size;
2231 free_in_cluster -= size;
2232 if (free_in_cluster == 0)
2233 s->free_byte_offset = 0;
2234 if ((offset & (s->cluster_size - 1)) != 0)
2235 update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
2236 } else {
2237 offset = alloc_clusters(bs, s->cluster_size);
2238 cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1);
2239 if ((cluster_offset + s->cluster_size) == offset) {
2240 /* we are lucky: contiguous data */
2241 offset = s->free_byte_offset;
2242 update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
2243 s->free_byte_offset += size;
2244 } else {
2245 s->free_byte_offset = offset;
2246 goto redo;
2247 }
2248 }
2249 return offset;
2250 }
2251
2252 static void free_clusters(BlockDriverState *bs,
2253 int64_t offset, int64_t size)
2254 {
2255 update_refcount(bs, offset, size, -1);
2256 }
2257
2258 static int grow_refcount_table(BlockDriverState *bs, int min_size)
2259 {
2260 BDRVQcowState *s = bs->opaque;
2261 int new_table_size, new_table_size2, refcount_table_clusters, i, ret;
2262 uint64_t *new_table;
2263 int64_t table_offset;
2264 uint8_t data[12];
2265 int old_table_size;
2266 int64_t old_table_offset;
2267
2268 if (min_size <= s->refcount_table_size)
2269 return 0;
2270 /* compute new table size */
2271 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
2272 for(;;) {
2273 if (refcount_table_clusters == 0) {
2274 refcount_table_clusters = 1;
2275 } else {
2276 refcount_table_clusters = (refcount_table_clusters * 3 + 1) / 2;
2277 }
2278 new_table_size = refcount_table_clusters << (s->cluster_bits - 3);
2279 if (min_size <= new_table_size)
2280 break;
2281 }
2282 #ifdef DEBUG_ALLOC2
2283 printf("grow_refcount_table from %d to %d\n",
2284 s->refcount_table_size,
2285 new_table_size);
2286 #endif
2287 new_table_size2 = new_table_size * sizeof(uint64_t);
2288 new_table = qemu_mallocz(new_table_size2);
2289 if (!new_table)
2290 return -ENOMEM;
2291 memcpy(new_table, s->refcount_table,
2292 s->refcount_table_size * sizeof(uint64_t));
2293 for(i = 0; i < s->refcount_table_size; i++)
2294 cpu_to_be64s(&new_table[i]);
2295 /* Note: we cannot update the refcount now to avoid recursion */
2296 table_offset = alloc_clusters_noref(bs, new_table_size2);
2297 ret = bdrv_pwrite(s->hd, table_offset, new_table, new_table_size2);
2298 if (ret != new_table_size2)
2299 goto fail;
2300 for(i = 0; i < s->refcount_table_size; i++)
2301 be64_to_cpus(&new_table[i]);
2302
2303 cpu_to_be64w((uint64_t*)data, table_offset);
2304 cpu_to_be32w((uint32_t*)(data + 8), refcount_table_clusters);
2305 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, refcount_table_offset),
2306 data, sizeof(data)) != sizeof(data))
2307 goto fail;
2308 qemu_free(s->refcount_table);
2309 old_table_offset = s->refcount_table_offset;
2310 old_table_size = s->refcount_table_size;
2311 s->refcount_table = new_table;
2312 s->refcount_table_size = new_table_size;
2313 s->refcount_table_offset = table_offset;
2314
2315 update_refcount(bs, table_offset, new_table_size2, 1);
2316 free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t));
2317 return 0;
2318 fail:
2319 free_clusters(bs, table_offset, new_table_size2);
2320 qemu_free(new_table);
2321 return -EIO;
2322 }
2323
2324 /* addend must be 1 or -1 */
2325 /* XXX: cache several refcount block clusters ? */
2326 static int update_cluster_refcount(BlockDriverState *bs,
2327 int64_t cluster_index,
2328 int addend)
2329 {
2330 BDRVQcowState *s = bs->opaque;
2331 int64_t offset, refcount_block_offset;
2332 int ret, refcount_table_index, block_index, refcount;
2333 uint64_t data64;
2334
2335 refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
2336 if (refcount_table_index >= s->refcount_table_size) {
2337 if (addend < 0)
2338 return -EINVAL;
2339 ret = grow_refcount_table(bs, refcount_table_index + 1);
2340 if (ret < 0)
2341 return ret;
2342 }
2343 refcount_block_offset = s->refcount_table[refcount_table_index];
2344 if (!refcount_block_offset) {
2345 if (addend < 0)
2346 return -EINVAL;
2347 /* create a new refcount block */
2348 /* Note: we cannot update the refcount now to avoid recursion */
2349 offset = alloc_clusters_noref(bs, s->cluster_size);
2350 memset(s->refcount_block_cache, 0, s->cluster_size);
2351 ret = bdrv_pwrite(s->hd, offset, s->refcount_block_cache, s->cluster_size);
2352 if (ret != s->cluster_size)
2353 return -EINVAL;
2354 s->refcount_table[refcount_table_index] = offset;
2355 data64 = cpu_to_be64(offset);
2356 ret = bdrv_pwrite(s->hd, s->refcount_table_offset +
2357 refcount_table_index * sizeof(uint64_t),
2358 &data64, sizeof(data64));
2359 if (ret != sizeof(data64))
2360 return -EINVAL;
2361
2362 refcount_block_offset = offset;
2363 s->refcount_block_cache_offset = offset;
2364 update_refcount(bs, offset, s->cluster_size, 1);
2365 } else {
2366 if (refcount_block_offset != s->refcount_block_cache_offset) {
2367 if (load_refcount_block(bs, refcount_block_offset) < 0)
2368 return -EIO;
2369 }
2370 }
2371 /* we can update the count and save it */
2372 block_index = cluster_index &
2373 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
2374 refcount = be16_to_cpu(s->refcount_block_cache[block_index]);
2375 refcount += addend;
2376 if (refcount < 0 || refcount > 0xffff)
2377 return -EINVAL;
2378 if (refcount == 0 && cluster_index < s->free_cluster_index) {
2379 s->free_cluster_index = cluster_index;
2380 }
2381 s->refcount_block_cache[block_index] = cpu_to_be16(refcount);
2382 if (bdrv_pwrite(s->hd,
2383 refcount_block_offset + (block_index << REFCOUNT_SHIFT),
2384 &s->refcount_block_cache[block_index], 2) != 2)
2385 return -EIO;
2386 return refcount;
2387 }
2388
2389 static void update_refcount(BlockDriverState *bs,
2390 int64_t offset, int64_t length,
2391 int addend)
2392 {
2393 BDRVQcowState *s = bs->opaque;
2394 int64_t start, last, cluster_offset;
2395
2396 #ifdef DEBUG_ALLOC2
2397 printf("update_refcount: offset=%lld size=%lld addend=%d\n",
2398 offset, length, addend);
2399 #endif
2400 if (length <= 0)
2401 return;
2402 start = offset & ~(s->cluster_size - 1);
2403 last = (offset + length - 1) & ~(s->cluster_size - 1);
2404 for(cluster_offset = start; cluster_offset <= last;
2405 cluster_offset += s->cluster_size) {
2406 update_cluster_refcount(bs, cluster_offset >> s->cluster_bits, addend);
2407 }
2408 }
2409
2410 #ifdef DEBUG_ALLOC
2411 static void inc_refcounts(BlockDriverState *bs,
2412 uint16_t *refcount_table,
2413 int refcount_table_size,
2414 int64_t offset, int64_t size)
2415 {
2416 BDRVQcowState *s = bs->opaque;
2417 int64_t start, last, cluster_offset;
2418 int k;
2419
2420 if (size <= 0)
2421 return;
2422
2423 start = offset & ~(s->cluster_size - 1);
2424 last = (offset + size - 1) & ~(s->cluster_size - 1);
2425 for(cluster_offset = start; cluster_offset <= last;
2426 cluster_offset += s->cluster_size) {
2427 k = cluster_offset >> s->cluster_bits;
2428 if (k < 0 || k >= refcount_table_size) {
2429 printf("ERROR: invalid cluster offset=0x%llx\n", cluster_offset);
2430 } else {
2431 if (++refcount_table[k] == 0) {
2432 printf("ERROR: overflow cluster offset=0x%llx\n", cluster_offset);
2433 }
2434 }
2435 }
2436 }
2437
2438 static int check_refcounts_l1(BlockDriverState *bs,
2439 uint16_t *refcount_table,
2440 int refcount_table_size,
2441 int64_t l1_table_offset, int l1_size,
2442 int check_copied)
2443 {
2444 BDRVQcowState *s = bs->opaque;
2445 uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2;
2446 int l2_size, i, j, nb_csectors, refcount;
2447
2448 l2_table = NULL;
2449 l1_size2 = l1_size * sizeof(uint64_t);
2450
2451 inc_refcounts(bs, refcount_table, refcount_table_size,
2452 l1_table_offset, l1_size2);
2453
2454 l1_table = qemu_malloc(l1_size2);
2455 if (!l1_table)
2456 goto fail;
2457 if (bdrv_pread(s->hd, l1_table_offset,
2458 l1_table, l1_size2) != l1_size2)
2459 goto fail;
2460 for(i = 0;i < l1_size; i++)
2461 be64_to_cpus(&l1_table[i]);
2462
2463 l2_size = s->l2_size * sizeof(uint64_t);
2464 l2_table = qemu_malloc(l2_size);
2465 if (!l2_table)
2466 goto fail;
2467 for(i = 0; i < l1_size; i++) {
2468 l2_offset = l1_table[i];
2469 if (l2_offset) {
2470 if (check_copied) {
2471 refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED) >> s->cluster_bits);
2472 if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) {
2473 printf("ERROR OFLAG_COPIED: l2_offset=%llx refcount=%d\n",
2474 l2_offset, refcount);
2475 }
2476 }
2477 l2_offset &= ~QCOW_OFLAG_COPIED;
2478 if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
2479 goto fail;
2480 for(j = 0; j < s->l2_size; j++) {
2481 offset = be64_to_cpu(l2_table[j]);
2482 if (offset != 0) {
2483 if (offset & QCOW_OFLAG_COMPRESSED) {
2484 if (offset & QCOW_OFLAG_COPIED) {
2485 printf("ERROR: cluster %lld: copied flag must never be set for compressed clusters\n",
2486 offset >> s->cluster_bits);
2487 offset &= ~QCOW_OFLAG_COPIED;
2488 }
2489 nb_csectors = ((offset >> s->csize_shift) &
2490 s->csize_mask) + 1;
2491 offset &= s->cluster_offset_mask;
2492 inc_refcounts(bs, refcount_table,
2493 refcount_table_size,
2494 offset & ~511, nb_csectors * 512);
2495 } else {
2496 if (check_copied) {
2497 refcount = get_refcount(bs, (offset & ~QCOW_OFLAG_COPIED) >> s->cluster_bits);
2498 if ((refcount == 1) != ((offset & QCOW_OFLAG_COPIED) != 0)) {
2499 printf("ERROR OFLAG_COPIED: offset=%llx refcount=%d\n",
2500 offset, refcount);
2501 }
2502 }
2503 offset &= ~QCOW_OFLAG_COPIED;
2504 inc_refcounts(bs, refcount_table,
2505 refcount_table_size,
2506 offset, s->cluster_size);
2507 }
2508 }
2509 }
2510 inc_refcounts(bs, refcount_table,
2511 refcount_table_size,
2512 l2_offset,
2513 s->cluster_size);
2514 }
2515 }
2516 qemu_free(l1_table);
2517 qemu_free(l2_table);
2518 return 0;
2519 fail:
2520 printf("ERROR: I/O error in check_refcounts_l1\n");
2521 qemu_free(l1_table);
2522 qemu_free(l2_table);
2523 return -EIO;
2524 }
2525
2526 static void check_refcounts(BlockDriverState *bs)
2527 {
2528 BDRVQcowState *s = bs->opaque;
2529 int64_t size;
2530 int nb_clusters, refcount1, refcount2, i;
2531 QCowSnapshot *sn;
2532 uint16_t *refcount_table;
2533
2534 size = bdrv_getlength(s->hd);
2535 nb_clusters = size_to_clusters(s, size);
2536 refcount_table = qemu_mallocz(nb_clusters * sizeof(uint16_t));
2537
2538 /* header */
2539 inc_refcounts(bs, refcount_table, nb_clusters,
2540 0, s->cluster_size);
2541
2542 check_refcounts_l1(bs, refcount_table, nb_clusters,
2543 s->l1_table_offset, s->l1_size, 1);
2544
2545 /* snapshots */
2546 for(i = 0; i < s->nb_snapshots; i++) {
2547 sn = s->snapshots + i;
2548 check_refcounts_l1(bs, refcount_table, nb_clusters,
2549 sn->l1_table_offset, sn->l1_size, 0);
2550 }
2551 inc_refcounts(bs, refcount_table, nb_clusters,
2552 s->snapshots_offset, s->snapshots_size);
2553
2554 /* refcount data */
2555 inc_refcounts(bs, refcount_table, nb_clusters,
2556 s->refcount_table_offset,
2557 s->refcount_table_size * sizeof(uint64_t));
2558 for(i = 0; i < s->refcount_table_size; i++) {
2559 int64_t offset;
2560 offset = s->refcount_table[i];
2561 if (offset != 0) {
2562 inc_refcounts(bs, refcount_table, nb_clusters,
2563 offset, s->cluster_size);
2564 }
2565 }
2566
2567 /* compare ref counts */
2568 for(i = 0; i < nb_clusters; i++) {
2569 refcount1 = get_refcount(bs, i);
2570 refcount2 = refcount_table[i];
2571 if (refcount1 != refcount2)
2572 printf("ERROR cluster %d refcount=%d reference=%d\n",
2573 i, refcount1, refcount2);
2574 }
2575
2576 qemu_free(refcount_table);
2577 }
2578
2579 #if 0
2580 static void dump_refcounts(BlockDriverState *bs)
2581 {
2582 BDRVQcowState *s = bs->opaque;
2583 int64_t nb_clusters, k, k1, size;
2584 int refcount;
2585
2586 size = bdrv_getlength(s->hd);
2587 nb_clusters = size_to_clusters(s, size);
2588 for(k = 0; k < nb_clusters;) {
2589 k1 = k;
2590 refcount = get_refcount(bs, k);
2591 k++;
2592 while (k < nb_clusters && get_refcount(bs, k) == refcount)
2593 k++;
2594 printf("%lld: refcount=%d nb=%lld\n", k, refcount, k - k1);
2595 }
2596 }
2597 #endif
2598 #endif
2599
2600 BlockDriver bdrv_qcow2 = {
2601 "qcow2",
2602 sizeof(BDRVQcowState),
2603 qcow_probe,
2604 qcow_open,
2605 NULL,
2606 NULL,
2607 qcow_close,
2608 qcow_create,
2609 qcow_flush,
2610 qcow_is_allocated,
2611 qcow_set_key,
2612 qcow_make_empty,
2613
2614 .bdrv_aio_read = qcow_aio_read,
2615 .bdrv_aio_write = qcow_aio_write,
2616 .bdrv_aio_cancel = qcow_aio_cancel,
2617 .aiocb_size = sizeof(QCowAIOCB),
2618 .bdrv_write_compressed = qcow_write_compressed,
2619
2620 .bdrv_snapshot_create = qcow_snapshot_create,
2621 .bdrv_snapshot_goto = qcow_snapshot_goto,
2622 .bdrv_snapshot_delete = qcow_snapshot_delete,
2623 .bdrv_snapshot_list = qcow_snapshot_list,
2624 .bdrv_get_info = qcow_get_info,
2625 };