]> git.proxmox.com Git - qemu.git/blame - block-qcow2.c
Fix (at least one cause of) qcow2 corruption. (Nolan Leake)
[qemu.git] / block-qcow2.c
CommitLineData
585f8587
FB
1/*
2 * Block driver for the QCOW version 2 format
5fafdf24 3 *
585f8587 4 * Copyright (c) 2004-2006 Fabrice Bellard
5fafdf24 5 *
585f8587
FB
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 */
faf07963 24#include "qemu-common.h"
585f8587
FB
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.
5fafdf24 37 - Size of compressed clusters is stored in sectors to reduce bit usage
585f8587
FB
38 in the cluster offsets.
39 - Support for storing additional data (such as the VM state) in the
3b46e624 40 snapshots.
585f8587
FB
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
5fafdf24 48
585f8587
FB
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
095a9c58
AL
55#define QCOW_MAX_CRYPT_CLUSTERS 32
56
585f8587
FB
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
585f8587
FB
64typedef 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
80typedef 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
102typedef 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
113typedef 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
152static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset);
5fafdf24 153static int qcow_read(BlockDriverState *bs, int64_t sector_num,
585f8587
FB
154 uint8_t *buf, int nb_sectors);
155static int qcow_read_snapshots(BlockDriverState *bs);
156static void qcow_free_snapshots(BlockDriverState *bs);
157static int refcount_init(BlockDriverState *bs);
158static void refcount_close(BlockDriverState *bs);
159static int get_refcount(BlockDriverState *bs, int64_t cluster_index);
5fafdf24 160static int update_cluster_refcount(BlockDriverState *bs,
585f8587
FB
161 int64_t cluster_index,
162 int addend);
5fafdf24
TS
163static void update_refcount(BlockDriverState *bs,
164 int64_t offset, int64_t length,
585f8587
FB
165 int addend);
166static int64_t alloc_clusters(BlockDriverState *bs, int64_t size);
167static int64_t alloc_bytes(BlockDriverState *bs, int size);
5fafdf24 168static void free_clusters(BlockDriverState *bs,
585f8587
FB
169 int64_t offset, int64_t size);
170#ifdef DEBUG_ALLOC
171static void check_refcounts(BlockDriverState *bs);
172#endif
173
174static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
175{
176 const QCowHeader *cow_header = (const void *)buf;
3b46e624 177
585f8587
FB
178 if (buf_size >= sizeof(QCowHeader) &&
179 be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
5fafdf24 180 be32_to_cpu(cow_header->version) == QCOW_VERSION)
585f8587
FB
181 return 100;
182 else
183 return 0;
184}
185
186static 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
4dc822d7
AL
192 /* Performance is terrible right now with cache=writethrough due mainly
193 * to reference count updates. If the user does not explicitly specify
194 * a caching type, force to writeback caching.
195 */
196 if ((flags & BDRV_O_CACHE_DEF)) {
197 flags |= BDRV_O_CACHE_WB;
198 flags &= ~BDRV_O_CACHE_DEF;
199 }
b5eff355 200 ret = bdrv_file_open(&s->hd, filename, flags);
585f8587
FB
201 if (ret < 0)
202 return ret;
203 if (bdrv_pread(s->hd, 0, &header, sizeof(header)) != sizeof(header))
204 goto fail;
205 be32_to_cpus(&header.magic);
206 be32_to_cpus(&header.version);
207 be64_to_cpus(&header.backing_file_offset);
208 be32_to_cpus(&header.backing_file_size);
209 be64_to_cpus(&header.size);
210 be32_to_cpus(&header.cluster_bits);
211 be32_to_cpus(&header.crypt_method);
212 be64_to_cpus(&header.l1_table_offset);
213 be32_to_cpus(&header.l1_size);
214 be64_to_cpus(&header.refcount_table_offset);
215 be32_to_cpus(&header.refcount_table_clusters);
216 be64_to_cpus(&header.snapshots_offset);
217 be32_to_cpus(&header.nb_snapshots);
3b46e624 218
585f8587
FB
219 if (header.magic != QCOW_MAGIC || header.version != QCOW_VERSION)
220 goto fail;
5fafdf24
TS
221 if (header.size <= 1 ||
222 header.cluster_bits < 9 ||
585f8587
FB
223 header.cluster_bits > 16)
224 goto fail;
225 if (header.crypt_method > QCOW_CRYPT_AES)
226 goto fail;
227 s->crypt_method_header = header.crypt_method;
228 if (s->crypt_method_header)
229 bs->encrypted = 1;
230 s->cluster_bits = header.cluster_bits;
231 s->cluster_size = 1 << s->cluster_bits;
232 s->cluster_sectors = 1 << (s->cluster_bits - 9);
233 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
234 s->l2_size = 1 << s->l2_bits;
235 bs->total_sectors = header.size / 512;
236 s->csize_shift = (62 - (s->cluster_bits - 8));
237 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
238 s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
239 s->refcount_table_offset = header.refcount_table_offset;
5fafdf24 240 s->refcount_table_size =
585f8587
FB
241 header.refcount_table_clusters << (s->cluster_bits - 3);
242
243 s->snapshots_offset = header.snapshots_offset;
244 s->nb_snapshots = header.nb_snapshots;
245
246 /* read the level 1 table */
247 s->l1_size = header.l1_size;
248 shift = s->cluster_bits + s->l2_bits;
249 s->l1_vm_state_index = (header.size + (1LL << shift) - 1) >> shift;
250 /* the L1 table must contain at least enough entries to put
251 header.size bytes */
252 if (s->l1_size < s->l1_vm_state_index)
253 goto fail;
254 s->l1_table_offset = header.l1_table_offset;
255 s->l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
5fafdf24 256 if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, s->l1_size * sizeof(uint64_t)) !=
585f8587
FB
257 s->l1_size * sizeof(uint64_t))
258 goto fail;
259 for(i = 0;i < s->l1_size; i++) {
260 be64_to_cpus(&s->l1_table[i]);
261 }
262 /* alloc L2 cache */
263 s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
585f8587 264 s->cluster_cache = qemu_malloc(s->cluster_size);
585f8587 265 /* one more sector for decompressed data alignment */
095a9c58
AL
266 s->cluster_data = qemu_malloc(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
267 + 512);
585f8587 268 s->cluster_cache_offset = -1;
3b46e624 269
585f8587
FB
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
301static int qcow_set_key(BlockDriverState *bs, const char *key)
302{
303 BDRVQcowState *s = bs->opaque;
304 uint8_t keybuf[16];
305 int len, i;
3b46e624 306
585f8587
FB
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 */
346static 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;
5fafdf24 360 AES_cbc_encrypt(in_buf, out_buf, 512, key,
585f8587
FB
361 ivec.b, enc);
362 sector_num++;
363 in_buf += 512;
364 out_buf += 512;
365 }
366}
367
368static 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) {
5fafdf24
TS
381 encrypt_sectors(s, start_sect + n_start,
382 s->cluster_data,
585f8587
FB
383 s->cluster_data, n, 1,
384 &s->aes_encrypt_key);
385 }
5fafdf24 386 ret = bdrv_write(s->hd, (cluster_offset >> 9) + n_start,
585f8587
FB
387 s->cluster_data, n);
388 if (ret < 0)
389 return ret;
390 return 0;
391}
392
393static 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
402static 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
420static int64_t align_offset(int64_t offset, int n)
421{
422 offset = (offset + n - 1) & ~(n - 1);
423 return offset;
424}
425
426static 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;
643e5399 432 uint8_t data[12];
585f8587
FB
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);
585f8587
FB
446 memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t));
447
448 /* write new table (align to cluster) */
449 new_l1_table_offset = alloc_clusters(bs, new_l1_size2);
3b46e624 450
585f8587
FB
451 for(i = 0; i < s->l1_size; i++)
452 new_l1_table[i] = cpu_to_be64(new_l1_table[i]);
453 ret = bdrv_pwrite(s->hd, new_l1_table_offset, new_l1_table, new_l1_size2);
454 if (ret != new_l1_size2)
455 goto fail;
456 for(i = 0; i < s->l1_size; i++)
457 new_l1_table[i] = be64_to_cpu(new_l1_table[i]);
3b46e624 458
585f8587 459 /* set new table */
643e5399
AL
460 cpu_to_be32w((uint32_t*)data, new_l1_size);
461 cpu_to_be64w((uint64_t*)(data + 4), new_l1_table_offset);
462 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, l1_size), data,
463 sizeof(data)) != sizeof(data))
585f8587
FB
464 goto fail;
465 qemu_free(s->l1_table);
466 free_clusters(bs, s->l1_table_offset, s->l1_size * sizeof(uint64_t));
467 s->l1_table_offset = new_l1_table_offset;
468 s->l1_table = new_l1_table;
469 s->l1_size = new_l1_size;
470 return 0;
471 fail:
472 qemu_free(s->l1_table);
473 return -EIO;
474}
475
108534b9
AL
476/*
477 * seek_l2_table
585f8587 478 *
108534b9
AL
479 * seek l2_offset in the l2_cache table
480 * if not found, return NULL,
481 * if found,
482 * increments the l2 cache hit count of the entry,
483 * if counter overflow, divide by two all counters
484 * return the pointer to the l2 cache entry
585f8587 485 *
108534b9
AL
486 */
487
488static uint64_t *seek_l2_table(BDRVQcowState *s, uint64_t l2_offset)
489{
490 int i, j;
491
492 for(i = 0; i < L2_CACHE_SIZE; i++) {
493 if (l2_offset == s->l2_cache_offsets[i]) {
494 /* increment the hit count */
495 if (++s->l2_cache_counts[i] == 0xffffffff) {
496 for(j = 0; j < L2_CACHE_SIZE; j++) {
497 s->l2_cache_counts[j] >>= 1;
498 }
499 }
500 return s->l2_cache + (i << s->l2_bits);
501 }
502 }
503 return NULL;
504}
505
506/*
507 * l2_load
508 *
509 * Loads a L2 table into memory. If the table is in the cache, the cache
510 * is used; otherwise the L2 table is loaded from the image file.
511 *
512 * Returns a pointer to the L2 table on success, or NULL if the read from
513 * the image file failed.
514 */
515
516static uint64_t *l2_load(BlockDriverState *bs, uint64_t l2_offset)
517{
518 BDRVQcowState *s = bs->opaque;
519 int min_index;
520 uint64_t *l2_table;
521
522 /* seek if the table for the given offset is in the cache */
523
524 l2_table = seek_l2_table(s, l2_offset);
525 if (l2_table != NULL)
526 return l2_table;
527
528 /* not found: load a new entry in the least used one */
529
530 min_index = l2_cache_new_entry(bs);
531 l2_table = s->l2_cache + (min_index << s->l2_bits);
532 if (bdrv_pread(s->hd, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
533 s->l2_size * sizeof(uint64_t))
534 return NULL;
535 s->l2_cache_offsets[min_index] = l2_offset;
536 s->l2_cache_counts[min_index] = 1;
537
538 return l2_table;
539}
540
541/*
542 * l2_allocate
585f8587 543 *
108534b9
AL
544 * Allocate a new l2 entry in the file. If l1_index points to an already
545 * used entry in the L2 table (i.e. we are doing a copy on write for the L2
546 * table) copy the contents of the old L2 table into the newly allocated one.
547 * Otherwise the new table is initialized with zeros.
585f8587 548 *
585f8587 549 */
108534b9
AL
550
551static uint64_t *l2_allocate(BlockDriverState *bs, int l1_index)
552{
553 BDRVQcowState *s = bs->opaque;
554 int min_index;
555 uint64_t old_l2_offset, tmp;
556 uint64_t *l2_table, l2_offset;
557
558 old_l2_offset = s->l1_table[l1_index];
559
560 /* allocate a new l2 entry */
561
562 l2_offset = alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
563
564 /* update the L1 entry */
565
566 s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
567
568 tmp = cpu_to_be64(l2_offset | QCOW_OFLAG_COPIED);
569 if (bdrv_pwrite(s->hd, s->l1_table_offset + l1_index * sizeof(tmp),
570 &tmp, sizeof(tmp)) != sizeof(tmp))
571 return NULL;
572
573 /* allocate a new entry in the l2 cache */
574
575 min_index = l2_cache_new_entry(bs);
576 l2_table = s->l2_cache + (min_index << s->l2_bits);
577
578 if (old_l2_offset == 0) {
579 /* if there was no old l2 table, clear the new table */
580 memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
581 } else {
582 /* if there was an old l2 table, read it from the disk */
583 if (bdrv_pread(s->hd, old_l2_offset,
584 l2_table, s->l2_size * sizeof(uint64_t)) !=
585 s->l2_size * sizeof(uint64_t))
586 return NULL;
587 }
588 /* write the l2 table to the file */
589 if (bdrv_pwrite(s->hd, l2_offset,
590 l2_table, s->l2_size * sizeof(uint64_t)) !=
591 s->l2_size * sizeof(uint64_t))
592 return NULL;
593
594 /* update the l2 cache entry */
595
596 s->l2_cache_offsets[min_index] = l2_offset;
597 s->l2_cache_counts[min_index] = 1;
598
599 return l2_table;
600}
601
6db6c638
AL
602static int size_to_clusters(BDRVQcowState *s, int64_t size)
603{
604 return (size + (s->cluster_size - 1)) >> s->cluster_bits;
605}
606
607static int count_contiguous_clusters(uint64_t nb_clusters, int cluster_size,
ff4b91c2 608 uint64_t *l2_table, uint64_t start, uint64_t mask)
6db6c638
AL
609{
610 int i;
611 uint64_t offset = be64_to_cpu(l2_table[0]) & ~mask;
612
ab5ccbd6
AL
613 if (!offset)
614 return 0;
615
ff4b91c2 616 for (i = start; i < start + nb_clusters; i++)
6db6c638
AL
617 if (offset + i * cluster_size != (be64_to_cpu(l2_table[i]) & ~mask))
618 break;
619
ff4b91c2 620 return (i - start);
6db6c638
AL
621}
622
623static int count_contiguous_free_clusters(uint64_t nb_clusters, uint64_t *l2_table)
624{
625 int i = 0;
626
627 while(nb_clusters-- && l2_table[i] == 0)
628 i++;
629
630 return i;
631}
632
05203524
AL
633/*
634 * get_cluster_offset
635 *
636 * For a given offset of the disk image, return cluster offset in
637 * qcow2 file.
638 *
095a9c58
AL
639 * on entry, *num is the number of contiguous clusters we'd like to
640 * access following offset.
641 *
642 * on exit, *num is the number of contiguous clusters we can read.
643 *
05203524
AL
644 * Return 1, if the offset is found
645 * Return 0, otherwise.
646 *
647 */
648
095a9c58
AL
649static uint64_t get_cluster_offset(BlockDriverState *bs,
650 uint64_t offset, int *num)
05203524
AL
651{
652 BDRVQcowState *s = bs->opaque;
653 int l1_index, l2_index;
6db6c638
AL
654 uint64_t l2_offset, *l2_table, cluster_offset;
655 int l1_bits, c;
656 int index_in_cluster, nb_available, nb_needed, nb_clusters;
095a9c58
AL
657
658 index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1);
659 nb_needed = *num + index_in_cluster;
660
661 l1_bits = s->l2_bits + s->cluster_bits;
662
663 /* compute how many bytes there are between the offset and
6db6c638 664 * the end of the l1 entry
095a9c58
AL
665 */
666
667 nb_available = (1 << l1_bits) - (offset & ((1 << l1_bits) - 1));
668
669 /* compute the number of available sectors */
670
671 nb_available = (nb_available >> 9) + index_in_cluster;
672
44988194
AL
673 if (nb_needed > nb_available) {
674 nb_needed = nb_available;
675 }
676
095a9c58 677 cluster_offset = 0;
05203524
AL
678
679 /* seek the the l2 offset in the l1 table */
680
095a9c58 681 l1_index = offset >> l1_bits;
05203524 682 if (l1_index >= s->l1_size)
095a9c58 683 goto out;
05203524
AL
684
685 l2_offset = s->l1_table[l1_index];
686
687 /* seek the l2 table of the given l2 offset */
688
689 if (!l2_offset)
095a9c58 690 goto out;
05203524
AL
691
692 /* load the l2 table in memory */
693
694 l2_offset &= ~QCOW_OFLAG_COPIED;
695 l2_table = l2_load(bs, l2_offset);
696 if (l2_table == NULL)
768706a5 697 return 0;
05203524
AL
698
699 /* find the cluster offset for the given disk offset */
700
701 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
702 cluster_offset = be64_to_cpu(l2_table[l2_index]);
6db6c638 703 nb_clusters = size_to_clusters(s, nb_needed << 9);
095a9c58
AL
704
705 if (!cluster_offset) {
6db6c638
AL
706 /* how many empty clusters ? */
707 c = count_contiguous_free_clusters(nb_clusters, &l2_table[l2_index]);
095a9c58 708 } else {
6db6c638
AL
709 /* how many allocated clusters ? */
710 c = count_contiguous_clusters(nb_clusters, s->cluster_size,
ff4b91c2 711 &l2_table[l2_index], 0, QCOW_OFLAG_COPIED);
6db6c638 712 }
095a9c58 713
6db6c638 714 nb_available = (c * s->cluster_sectors);
095a9c58
AL
715out:
716 if (nb_available > nb_needed)
717 nb_available = nb_needed;
718
719 *num = nb_available - index_in_cluster;
720
6db6c638 721 return cluster_offset & ~QCOW_OFLAG_COPIED;
05203524
AL
722}
723
724/*
52d893ec 725 * free_any_clusters
05203524 726 *
52d893ec 727 * free clusters according to its type: compressed or not
05203524 728 *
52d893ec
AL
729 */
730
731static void free_any_clusters(BlockDriverState *bs,
095a9c58 732 uint64_t cluster_offset, int nb_clusters)
52d893ec
AL
733{
734 BDRVQcowState *s = bs->opaque;
735
52d893ec
AL
736 /* free the cluster */
737
738 if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
739 int nb_csectors;
740 nb_csectors = ((cluster_offset >> s->csize_shift) &
741 s->csize_mask) + 1;
742 free_clusters(bs, (cluster_offset & s->cluster_offset_mask) & ~511,
743 nb_csectors * 512);
744 return;
745 }
746
095a9c58
AL
747 free_clusters(bs, cluster_offset, nb_clusters << s->cluster_bits);
748
749 return;
52d893ec
AL
750}
751
752/*
753 * get_cluster_table
05203524 754 *
52d893ec
AL
755 * for a given disk offset, load (and allocate if needed)
756 * the l2 table.
757 *
758 * the l2 table offset in the qcow2 file and the cluster index
759 * in the l2 table are given to the caller.
05203524
AL
760 *
761 */
762
52d893ec
AL
763static int get_cluster_table(BlockDriverState *bs, uint64_t offset,
764 uint64_t **new_l2_table,
765 uint64_t *new_l2_offset,
766 int *new_l2_index)
585f8587
FB
767{
768 BDRVQcowState *s = bs->opaque;
108534b9 769 int l1_index, l2_index, ret;
52d893ec 770 uint64_t l2_offset, *l2_table;
108534b9
AL
771
772 /* seek the the l2 offset in the l1 table */
3b46e624 773
585f8587
FB
774 l1_index = offset >> (s->l2_bits + s->cluster_bits);
775 if (l1_index >= s->l1_size) {
108534b9
AL
776 ret = grow_l1_table(bs, l1_index + 1);
777 if (ret < 0)
585f8587
FB
778 return 0;
779 }
780 l2_offset = s->l1_table[l1_index];
108534b9
AL
781
782 /* seek the l2 table of the given l2 offset */
783
05203524
AL
784 if (l2_offset & QCOW_OFLAG_COPIED) {
785 /* load the l2 table in memory */
786 l2_offset &= ~QCOW_OFLAG_COPIED;
787 l2_table = l2_load(bs, l2_offset);
788 if (l2_table == NULL)
585f8587 789 return 0;
05203524
AL
790 } else {
791 if (l2_offset)
792 free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t));
108534b9
AL
793 l2_table = l2_allocate(bs, l1_index);
794 if (l2_table == NULL)
585f8587 795 return 0;
108534b9 796 l2_offset = s->l1_table[l1_index] & ~QCOW_OFLAG_COPIED;
585f8587 797 }
108534b9
AL
798
799 /* find the cluster offset for the given disk offset */
800
585f8587 801 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
05203524 802
52d893ec
AL
803 *new_l2_table = l2_table;
804 *new_l2_offset = l2_offset;
805 *new_l2_index = l2_index;
806
807 return 1;
808}
809
810/*
811 * alloc_compressed_cluster_offset
812 *
813 * For a given offset of the disk image, return cluster offset in
814 * qcow2 file.
815 *
816 * If the offset is not found, allocate a new compressed cluster.
817 *
818 * Return the cluster offset if successful,
819 * Return 0, otherwise.
820 *
821 */
822
823static uint64_t alloc_compressed_cluster_offset(BlockDriverState *bs,
824 uint64_t offset,
825 int compressed_size)
826{
827 BDRVQcowState *s = bs->opaque;
828 int l2_index, ret;
829 uint64_t l2_offset, *l2_table, cluster_offset;
830 int nb_csectors;
831
832 ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
833 if (ret == 0)
834 return 0;
835
836 cluster_offset = be64_to_cpu(l2_table[l2_index]);
05203524
AL
837 if (cluster_offset & QCOW_OFLAG_COPIED)
838 return cluster_offset & ~QCOW_OFLAG_COPIED;
839
095a9c58
AL
840 if (cluster_offset)
841 free_any_clusters(bs, cluster_offset, 1);
108534b9 842
52d893ec
AL
843 cluster_offset = alloc_bytes(bs, compressed_size);
844 nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) -
845 (cluster_offset >> 9);
05203524 846
52d893ec
AL
847 cluster_offset |= QCOW_OFLAG_COMPRESSED |
848 ((uint64_t)nb_csectors << s->csize_shift);
05203524 849
52d893ec 850 /* update L2 table */
05203524 851
52d893ec 852 /* compressed clusters never have the copied flag */
05203524 853
52d893ec
AL
854 l2_table[l2_index] = cpu_to_be64(cluster_offset);
855 if (bdrv_pwrite(s->hd,
856 l2_offset + l2_index * sizeof(uint64_t),
857 l2_table + l2_index,
858 sizeof(uint64_t)) != sizeof(uint64_t))
859 return 0;
05203524 860
52d893ec
AL
861 return cluster_offset;
862}
05203524 863
e976c6a1
AL
864typedef struct QCowL2Meta
865{
866 uint64_t offset;
867 int n_start;
868 int nb_available;
869 int nb_clusters;
870} QCowL2Meta;
871
872static int alloc_cluster_link_l2(BlockDriverState *bs, uint64_t cluster_offset,
873 QCowL2Meta *m)
874{
875 BDRVQcowState *s = bs->opaque;
876 int i, j = 0, l2_index, ret;
877 uint64_t *old_cluster, start_sect, l2_offset, *l2_table;
878
879 if (m->nb_clusters == 0)
880 return 0;
881
3ec88e80 882 old_cluster = qemu_malloc(m->nb_clusters * sizeof(uint64_t));
e976c6a1
AL
883
884 /* copy content of unmodified sectors */
885 start_sect = (m->offset & ~(s->cluster_size - 1)) >> 9;
886 if (m->n_start) {
887 ret = copy_sectors(bs, start_sect, cluster_offset, 0, m->n_start);
888 if (ret < 0)
889 goto err;
890 }
891
892 if (m->nb_available & (s->cluster_sectors - 1)) {
893 uint64_t end = m->nb_available & ~(uint64_t)(s->cluster_sectors - 1);
894 ret = copy_sectors(bs, start_sect + end, cluster_offset + (end << 9),
895 m->nb_available - end, s->cluster_sectors);
896 if (ret < 0)
897 goto err;
898 }
899
900 ret = -EIO;
901 /* update L2 table */
902 if (!get_cluster_table(bs, m->offset, &l2_table, &l2_offset, &l2_index))
903 goto err;
904
905 for (i = 0; i < m->nb_clusters; i++) {
906 if(l2_table[l2_index + i] != 0)
907 old_cluster[j++] = l2_table[l2_index + i];
908
909 l2_table[l2_index + i] = cpu_to_be64((cluster_offset +
910 (i << s->cluster_bits)) | QCOW_OFLAG_COPIED);
911 }
912
913 if (bdrv_pwrite(s->hd, l2_offset + l2_index * sizeof(uint64_t),
914 l2_table + l2_index, m->nb_clusters * sizeof(uint64_t)) !=
915 m->nb_clusters * sizeof(uint64_t))
916 goto err;
917
918 for (i = 0; i < j; i++)
919 free_any_clusters(bs, old_cluster[i], 1);
920
921 ret = 0;
922err:
923 qemu_free(old_cluster);
924 return ret;
925 }
926
52d893ec
AL
927/*
928 * alloc_cluster_offset
929 *
930 * For a given offset of the disk image, return cluster offset in
931 * qcow2 file.
932 *
933 * If the offset is not found, allocate a new cluster.
934 *
935 * Return the cluster offset if successful,
936 * Return 0, otherwise.
937 *
938 */
939
940static uint64_t alloc_cluster_offset(BlockDriverState *bs,
941 uint64_t offset,
095a9c58 942 int n_start, int n_end,
e976c6a1 943 int *num, QCowL2Meta *m)
52d893ec
AL
944{
945 BDRVQcowState *s = bs->opaque;
946 int l2_index, ret;
947 uint64_t l2_offset, *l2_table, cluster_offset;
e976c6a1 948 int nb_clusters, i = 0;
52d893ec
AL
949
950 ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
951 if (ret == 0)
952 return 0;
953
6db6c638
AL
954 nb_clusters = size_to_clusters(s, n_end << 9);
955
e976c6a1 956 nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
095a9c58 957
52d893ec 958 cluster_offset = be64_to_cpu(l2_table[l2_index]);
52d893ec 959
095a9c58
AL
960 /* We keep all QCOW_OFLAG_COPIED clusters */
961
962 if (cluster_offset & QCOW_OFLAG_COPIED) {
6db6c638 963 nb_clusters = count_contiguous_clusters(nb_clusters, s->cluster_size,
ff4b91c2 964 &l2_table[l2_index], 0, 0);
095a9c58 965
095a9c58 966 cluster_offset &= ~QCOW_OFLAG_COPIED;
e976c6a1 967 m->nb_clusters = 0;
095a9c58
AL
968
969 goto out;
970 }
971
972 /* for the moment, multiple compressed clusters are not managed */
973
974 if (cluster_offset & QCOW_OFLAG_COMPRESSED)
975 nb_clusters = 1;
976
bc352085 977 /* how many available clusters ? */
095a9c58 978
bc352085 979 while (i < nb_clusters) {
ab5ccbd6 980 i += count_contiguous_clusters(nb_clusters - i, s->cluster_size,
ff4b91c2 981 &l2_table[l2_index], i, 0);
ab5ccbd6
AL
982
983 if(be64_to_cpu(l2_table[l2_index + i]))
984 break;
985
6db6c638
AL
986 i += count_contiguous_free_clusters(nb_clusters - i,
987 &l2_table[l2_index + i]);
095a9c58 988
6db6c638 989 cluster_offset = be64_to_cpu(l2_table[l2_index + i]);
095a9c58 990
6db6c638 991 if ((cluster_offset & QCOW_OFLAG_COPIED) ||
bc352085 992 (cluster_offset & QCOW_OFLAG_COMPRESSED))
6db6c638 993 break;
095a9c58 994 }
bc352085 995 nb_clusters = i;
05203524
AL
996
997 /* allocate a new cluster */
998
095a9c58 999 cluster_offset = alloc_clusters(bs, nb_clusters * s->cluster_size);
05203524 1000
e976c6a1
AL
1001 /* save info needed for meta data update */
1002 m->offset = offset;
1003 m->n_start = n_start;
1004 m->nb_clusters = nb_clusters;
05203524 1005
095a9c58 1006out:
e976c6a1
AL
1007 m->nb_available = MIN(nb_clusters << (s->cluster_bits - 9), n_end);
1008
1009 *num = m->nb_available - n_start;
095a9c58 1010
585f8587
FB
1011 return cluster_offset;
1012}
1013
5fafdf24 1014static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
585f8587
FB
1015 int nb_sectors, int *pnum)
1016{
585f8587
FB
1017 uint64_t cluster_offset;
1018
095a9c58
AL
1019 *pnum = nb_sectors;
1020 cluster_offset = get_cluster_offset(bs, sector_num << 9, pnum);
1021
585f8587
FB
1022 return (cluster_offset != 0);
1023}
1024
1025static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
1026 const uint8_t *buf, int buf_size)
1027{
1028 z_stream strm1, *strm = &strm1;
1029 int ret, out_len;
1030
1031 memset(strm, 0, sizeof(*strm));
1032
1033 strm->next_in = (uint8_t *)buf;
1034 strm->avail_in = buf_size;
1035 strm->next_out = out_buf;
1036 strm->avail_out = out_buf_size;
1037
1038 ret = inflateInit2(strm, -12);
1039 if (ret != Z_OK)
1040 return -1;
1041 ret = inflate(strm, Z_FINISH);
1042 out_len = strm->next_out - out_buf;
1043 if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
1044 out_len != out_buf_size) {
1045 inflateEnd(strm);
1046 return -1;
1047 }
1048 inflateEnd(strm);
1049 return 0;
1050}
3b46e624 1051
585f8587
FB
1052static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset)
1053{
1054 int ret, csize, nb_csectors, sector_offset;
1055 uint64_t coffset;
1056
1057 coffset = cluster_offset & s->cluster_offset_mask;
1058 if (s->cluster_cache_offset != coffset) {
1059 nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
1060 sector_offset = coffset & 511;
1061 csize = nb_csectors * 512 - sector_offset;
1062 ret = bdrv_read(s->hd, coffset >> 9, s->cluster_data, nb_csectors);
1063 if (ret < 0) {
1064 return -1;
1065 }
1066 if (decompress_buffer(s->cluster_cache, s->cluster_size,
1067 s->cluster_data + sector_offset, csize) < 0) {
1068 return -1;
1069 }
1070 s->cluster_cache_offset = coffset;
1071 }
1072 return 0;
1073}
1074
a9465922 1075/* handle reading after the end of the backing file */
5fafdf24 1076static int backing_read1(BlockDriverState *bs,
a9465922
FB
1077 int64_t sector_num, uint8_t *buf, int nb_sectors)
1078{
1079 int n1;
1080 if ((sector_num + nb_sectors) <= bs->total_sectors)
1081 return nb_sectors;
1082 if (sector_num >= bs->total_sectors)
1083 n1 = 0;
1084 else
1085 n1 = bs->total_sectors - sector_num;
1086 memset(buf + n1 * 512, 0, 512 * (nb_sectors - n1));
1087 return n1;
1088}
1089
5fafdf24 1090static int qcow_read(BlockDriverState *bs, int64_t sector_num,
585f8587
FB
1091 uint8_t *buf, int nb_sectors)
1092{
1093 BDRVQcowState *s = bs->opaque;
a9465922 1094 int ret, index_in_cluster, n, n1;
585f8587 1095 uint64_t cluster_offset;
3b46e624 1096
585f8587 1097 while (nb_sectors > 0) {
095a9c58
AL
1098 n = nb_sectors;
1099 cluster_offset = get_cluster_offset(bs, sector_num << 9, &n);
585f8587 1100 index_in_cluster = sector_num & (s->cluster_sectors - 1);
585f8587
FB
1101 if (!cluster_offset) {
1102 if (bs->backing_hd) {
1103 /* read from the base image */
a9465922
FB
1104 n1 = backing_read1(bs->backing_hd, sector_num, buf, n);
1105 if (n1 > 0) {
1106 ret = bdrv_read(bs->backing_hd, sector_num, buf, n1);
1107 if (ret < 0)
1108 return -1;
1109 }
585f8587
FB
1110 } else {
1111 memset(buf, 0, 512 * n);
1112 }
1113 } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
1114 if (decompress_cluster(s, cluster_offset) < 0)
1115 return -1;
1116 memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n);
1117 } else {
1118 ret = bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
5fafdf24 1119 if (ret != n * 512)
585f8587
FB
1120 return -1;
1121 if (s->crypt_method) {
5fafdf24 1122 encrypt_sectors(s, sector_num, buf, buf, n, 0,
585f8587
FB
1123 &s->aes_decrypt_key);
1124 }
1125 }
1126 nb_sectors -= n;
1127 sector_num += n;
1128 buf += n * 512;
1129 }
1130 return 0;
1131}
1132
5fafdf24 1133static int qcow_write(BlockDriverState *bs, int64_t sector_num,
585f8587
FB
1134 const uint8_t *buf, int nb_sectors)
1135{
1136 BDRVQcowState *s = bs->opaque;
1137 int ret, index_in_cluster, n;
1138 uint64_t cluster_offset;
095a9c58 1139 int n_end;
e976c6a1 1140 QCowL2Meta l2meta;
3b46e624 1141
585f8587
FB
1142 while (nb_sectors > 0) {
1143 index_in_cluster = sector_num & (s->cluster_sectors - 1);
095a9c58
AL
1144 n_end = index_in_cluster + nb_sectors;
1145 if (s->crypt_method &&
1146 n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
1147 n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
52d893ec 1148 cluster_offset = alloc_cluster_offset(bs, sector_num << 9,
05203524 1149 index_in_cluster,
e976c6a1 1150 n_end, &n, &l2meta);
585f8587
FB
1151 if (!cluster_offset)
1152 return -1;
1153 if (s->crypt_method) {
1154 encrypt_sectors(s, sector_num, s->cluster_data, buf, n, 1,
1155 &s->aes_encrypt_key);
5fafdf24 1156 ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512,
585f8587
FB
1157 s->cluster_data, n * 512);
1158 } else {
1159 ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
1160 }
e976c6a1
AL
1161 if (ret != n * 512 || alloc_cluster_link_l2(bs, cluster_offset, &l2meta) < 0) {
1162 free_any_clusters(bs, cluster_offset, l2meta.nb_clusters);
585f8587 1163 return -1;
e976c6a1 1164 }
585f8587
FB
1165 nb_sectors -= n;
1166 sector_num += n;
1167 buf += n * 512;
1168 }
1169 s->cluster_cache_offset = -1; /* disable compressed cache */
1170 return 0;
1171}
1172
ce1a14dc
PB
1173typedef struct QCowAIOCB {
1174 BlockDriverAIOCB common;
585f8587
FB
1175 int64_t sector_num;
1176 uint8_t *buf;
1177 int nb_sectors;
1178 int n;
1179 uint64_t cluster_offset;
5fafdf24 1180 uint8_t *cluster_data;
585f8587 1181 BlockDriverAIOCB *hd_aiocb;
1490791f 1182 QEMUBH *bh;
e976c6a1 1183 QCowL2Meta l2meta;
585f8587
FB
1184} QCowAIOCB;
1185
1490791f
AL
1186static void qcow_aio_read_cb(void *opaque, int ret);
1187static void qcow_aio_read_bh(void *opaque)
1188{
1189 QCowAIOCB *acb = opaque;
1190 qemu_bh_delete(acb->bh);
1191 acb->bh = NULL;
1192 qcow_aio_read_cb(opaque, 0);
1193}
1194
a32ef786
AL
1195static int qcow_schedule_bh(QEMUBHFunc *cb, QCowAIOCB *acb)
1196{
1197 if (acb->bh)
1198 return -EIO;
1199
1200 acb->bh = qemu_bh_new(cb, acb);
1201 if (!acb->bh)
1202 return -EIO;
1203
1204 qemu_bh_schedule(acb->bh);
1205
1206 return 0;
1207}
1208
585f8587
FB
1209static void qcow_aio_read_cb(void *opaque, int ret)
1210{
ce1a14dc
PB
1211 QCowAIOCB *acb = opaque;
1212 BlockDriverState *bs = acb->common.bs;
585f8587 1213 BDRVQcowState *s = bs->opaque;
a9465922 1214 int index_in_cluster, n1;
585f8587 1215
ce1a14dc 1216 acb->hd_aiocb = NULL;
585f8587 1217 if (ret < 0) {
ac674887 1218fail:
ce1a14dc
PB
1219 acb->common.cb(acb->common.opaque, ret);
1220 qemu_aio_release(acb);
585f8587
FB
1221 return;
1222 }
1223
585f8587 1224 /* post process the read buffer */
ce1a14dc 1225 if (!acb->cluster_offset) {
585f8587 1226 /* nothing to do */
ce1a14dc 1227 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
585f8587
FB
1228 /* nothing to do */
1229 } else {
1230 if (s->crypt_method) {
5fafdf24
TS
1231 encrypt_sectors(s, acb->sector_num, acb->buf, acb->buf,
1232 acb->n, 0,
585f8587
FB
1233 &s->aes_decrypt_key);
1234 }
1235 }
1236
ce1a14dc
PB
1237 acb->nb_sectors -= acb->n;
1238 acb->sector_num += acb->n;
1239 acb->buf += acb->n * 512;
585f8587 1240
ce1a14dc 1241 if (acb->nb_sectors == 0) {
585f8587 1242 /* request completed */
ce1a14dc
PB
1243 acb->common.cb(acb->common.opaque, 0);
1244 qemu_aio_release(acb);
585f8587
FB
1245 return;
1246 }
3b46e624 1247
585f8587 1248 /* prepare next AIO request */
095a9c58
AL
1249 acb->n = acb->nb_sectors;
1250 acb->cluster_offset = get_cluster_offset(bs, acb->sector_num << 9, &acb->n);
ce1a14dc 1251 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
ce1a14dc
PB
1252
1253 if (!acb->cluster_offset) {
585f8587
FB
1254 if (bs->backing_hd) {
1255 /* read from the base image */
5fafdf24 1256 n1 = backing_read1(bs->backing_hd, acb->sector_num,
ce1a14dc 1257 acb->buf, acb->n);
a9465922 1258 if (n1 > 0) {
5fafdf24 1259 acb->hd_aiocb = bdrv_aio_read(bs->backing_hd, acb->sector_num,
ce1a14dc
PB
1260 acb->buf, acb->n, qcow_aio_read_cb, acb);
1261 if (acb->hd_aiocb == NULL)
a9465922
FB
1262 goto fail;
1263 } else {
a32ef786
AL
1264 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
1265 if (ret < 0)
ac674887 1266 goto fail;
a9465922 1267 }
585f8587
FB
1268 } else {
1269 /* Note: in this case, no need to wait */
ce1a14dc 1270 memset(acb->buf, 0, 512 * acb->n);
a32ef786
AL
1271 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
1272 if (ret < 0)
ac674887 1273 goto fail;
585f8587 1274 }
ce1a14dc 1275 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
585f8587 1276 /* add AIO support for compressed blocks ? */
ce1a14dc 1277 if (decompress_cluster(s, acb->cluster_offset) < 0)
585f8587 1278 goto fail;
5fafdf24 1279 memcpy(acb->buf,
ce1a14dc 1280 s->cluster_cache + index_in_cluster * 512, 512 * acb->n);
a32ef786
AL
1281 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
1282 if (ret < 0)
ac674887 1283 goto fail;
585f8587 1284 } else {
ce1a14dc 1285 if ((acb->cluster_offset & 511) != 0) {
585f8587
FB
1286 ret = -EIO;
1287 goto fail;
1288 }
ce1a14dc 1289 acb->hd_aiocb = bdrv_aio_read(s->hd,
5fafdf24 1290 (acb->cluster_offset >> 9) + index_in_cluster,
ce1a14dc
PB
1291 acb->buf, acb->n, qcow_aio_read_cb, acb);
1292 if (acb->hd_aiocb == NULL)
585f8587
FB
1293 goto fail;
1294 }
1295}
1296
ce1a14dc
PB
1297static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs,
1298 int64_t sector_num, uint8_t *buf, int nb_sectors,
1299 BlockDriverCompletionFunc *cb, void *opaque)
585f8587 1300{
ce1a14dc
PB
1301 QCowAIOCB *acb;
1302
1303 acb = qemu_aio_get(bs, cb, opaque);
1304 if (!acb)
1305 return NULL;
1306 acb->hd_aiocb = NULL;
1307 acb->sector_num = sector_num;
1308 acb->buf = buf;
1309 acb->nb_sectors = nb_sectors;
1310 acb->n = 0;
1311 acb->cluster_offset = 0;
e976c6a1 1312 acb->l2meta.nb_clusters = 0;
ce1a14dc
PB
1313 return acb;
1314}
1315
1316static BlockDriverAIOCB *qcow_aio_read(BlockDriverState *bs,
1317 int64_t sector_num, uint8_t *buf, int nb_sectors,
1318 BlockDriverCompletionFunc *cb, void *opaque)
1319{
1320 QCowAIOCB *acb;
1321
1322 acb = qcow_aio_setup(bs, sector_num, buf, nb_sectors, cb, opaque);
1323 if (!acb)
1324 return NULL;
585f8587
FB
1325
1326 qcow_aio_read_cb(acb, 0);
ce1a14dc 1327 return &acb->common;
585f8587
FB
1328}
1329
1330static void qcow_aio_write_cb(void *opaque, int ret)
1331{
ce1a14dc
PB
1332 QCowAIOCB *acb = opaque;
1333 BlockDriverState *bs = acb->common.bs;
585f8587 1334 BDRVQcowState *s = bs->opaque;
585f8587 1335 int index_in_cluster;
585f8587 1336 const uint8_t *src_buf;
095a9c58 1337 int n_end;
ce1a14dc
PB
1338
1339 acb->hd_aiocb = NULL;
1340
585f8587
FB
1341 if (ret < 0) {
1342 fail:
ce1a14dc
PB
1343 acb->common.cb(acb->common.opaque, ret);
1344 qemu_aio_release(acb);
585f8587
FB
1345 return;
1346 }
1347
e976c6a1
AL
1348 if (alloc_cluster_link_l2(bs, acb->cluster_offset, &acb->l2meta) < 0) {
1349 free_any_clusters(bs, acb->cluster_offset, acb->l2meta.nb_clusters);
1350 goto fail;
1351 }
1352
ce1a14dc
PB
1353 acb->nb_sectors -= acb->n;
1354 acb->sector_num += acb->n;
1355 acb->buf += acb->n * 512;
585f8587 1356
ce1a14dc 1357 if (acb->nb_sectors == 0) {
585f8587 1358 /* request completed */
ce1a14dc
PB
1359 acb->common.cb(acb->common.opaque, 0);
1360 qemu_aio_release(acb);
585f8587
FB
1361 return;
1362 }
3b46e624 1363
ce1a14dc 1364 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
095a9c58
AL
1365 n_end = index_in_cluster + acb->nb_sectors;
1366 if (s->crypt_method &&
1367 n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
1368 n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
1369
e976c6a1 1370 acb->cluster_offset = alloc_cluster_offset(bs, acb->sector_num << 9,
05203524 1371 index_in_cluster,
e976c6a1
AL
1372 n_end, &acb->n, &acb->l2meta);
1373 if (!acb->cluster_offset || (acb->cluster_offset & 511) != 0) {
585f8587
FB
1374 ret = -EIO;
1375 goto fail;
1376 }
1377 if (s->crypt_method) {
ce1a14dc 1378 if (!acb->cluster_data) {
095a9c58
AL
1379 acb->cluster_data = qemu_mallocz(QCOW_MAX_CRYPT_CLUSTERS *
1380 s->cluster_size);
585f8587 1381 }
5fafdf24 1382 encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf,
ce1a14dc
PB
1383 acb->n, 1, &s->aes_encrypt_key);
1384 src_buf = acb->cluster_data;
585f8587 1385 } else {
ce1a14dc 1386 src_buf = acb->buf;
585f8587 1387 }
ce1a14dc 1388 acb->hd_aiocb = bdrv_aio_write(s->hd,
e976c6a1 1389 (acb->cluster_offset >> 9) + index_in_cluster,
5fafdf24 1390 src_buf, acb->n,
ce1a14dc
PB
1391 qcow_aio_write_cb, acb);
1392 if (acb->hd_aiocb == NULL)
585f8587
FB
1393 goto fail;
1394}
1395
ce1a14dc
PB
1396static BlockDriverAIOCB *qcow_aio_write(BlockDriverState *bs,
1397 int64_t sector_num, const uint8_t *buf, int nb_sectors,
1398 BlockDriverCompletionFunc *cb, void *opaque)
585f8587 1399{
585f8587 1400 BDRVQcowState *s = bs->opaque;
ce1a14dc 1401 QCowAIOCB *acb;
3b46e624 1402
585f8587
FB
1403 s->cluster_cache_offset = -1; /* disable compressed cache */
1404
ce1a14dc
PB
1405 acb = qcow_aio_setup(bs, sector_num, (uint8_t*)buf, nb_sectors, cb, opaque);
1406 if (!acb)
1407 return NULL;
3b46e624 1408
585f8587 1409 qcow_aio_write_cb(acb, 0);
ce1a14dc 1410 return &acb->common;
585f8587
FB
1411}
1412
ce1a14dc 1413static void qcow_aio_cancel(BlockDriverAIOCB *blockacb)
585f8587 1414{
ce1a14dc
PB
1415 QCowAIOCB *acb = (QCowAIOCB *)blockacb;
1416 if (acb->hd_aiocb)
1417 bdrv_aio_cancel(acb->hd_aiocb);
1418 qemu_aio_release(acb);
585f8587
FB
1419}
1420
1421static void qcow_close(BlockDriverState *bs)
1422{
1423 BDRVQcowState *s = bs->opaque;
1424 qemu_free(s->l1_table);
1425 qemu_free(s->l2_cache);
1426 qemu_free(s->cluster_cache);
1427 qemu_free(s->cluster_data);
1428 refcount_close(bs);
1429 bdrv_delete(s->hd);
1430}
1431
1432/* XXX: use std qcow open function ? */
1433typedef struct QCowCreateState {
1434 int cluster_size;
1435 int cluster_bits;
1436 uint16_t *refcount_block;
1437 uint64_t *refcount_table;
1438 int64_t l1_table_offset;
1439 int64_t refcount_table_offset;
1440 int64_t refcount_block_offset;
1441} QCowCreateState;
1442
1443static void create_refcount_update(QCowCreateState *s,
1444 int64_t offset, int64_t size)
1445{
1446 int refcount;
1447 int64_t start, last, cluster_offset;
1448 uint16_t *p;
1449
1450 start = offset & ~(s->cluster_size - 1);
1451 last = (offset + size - 1) & ~(s->cluster_size - 1);
5fafdf24 1452 for(cluster_offset = start; cluster_offset <= last;
585f8587
FB
1453 cluster_offset += s->cluster_size) {
1454 p = &s->refcount_block[cluster_offset >> s->cluster_bits];
1455 refcount = be16_to_cpu(*p);
1456 refcount++;
1457 *p = cpu_to_be16(refcount);
1458 }
1459}
1460
1461static int qcow_create(const char *filename, int64_t total_size,
1462 const char *backing_file, int flags)
1463{
1464 int fd, header_size, backing_filename_len, l1_size, i, shift, l2_bits;
1465 QCowHeader header;
1466 uint64_t tmp, offset;
1467 QCowCreateState s1, *s = &s1;
3b46e624 1468
585f8587
FB
1469 memset(s, 0, sizeof(*s));
1470
1471 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
1472 if (fd < 0)
1473 return -1;
1474 memset(&header, 0, sizeof(header));
1475 header.magic = cpu_to_be32(QCOW_MAGIC);
1476 header.version = cpu_to_be32(QCOW_VERSION);
1477 header.size = cpu_to_be64(total_size * 512);
1478 header_size = sizeof(header);
1479 backing_filename_len = 0;
1480 if (backing_file) {
1481 header.backing_file_offset = cpu_to_be64(header_size);
1482 backing_filename_len = strlen(backing_file);
1483 header.backing_file_size = cpu_to_be32(backing_filename_len);
1484 header_size += backing_filename_len;
1485 }
1486 s->cluster_bits = 12; /* 4 KB clusters */
1487 s->cluster_size = 1 << s->cluster_bits;
1488 header.cluster_bits = cpu_to_be32(s->cluster_bits);
1489 header_size = (header_size + 7) & ~7;
ec36ba14 1490 if (flags & BLOCK_FLAG_ENCRYPT) {
585f8587
FB
1491 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
1492 } else {
1493 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
1494 }
1495 l2_bits = s->cluster_bits - 3;
1496 shift = s->cluster_bits + l2_bits;
1497 l1_size = (((total_size * 512) + (1LL << shift) - 1) >> shift);
1498 offset = align_offset(header_size, s->cluster_size);
1499 s->l1_table_offset = offset;
1500 header.l1_table_offset = cpu_to_be64(s->l1_table_offset);
1501 header.l1_size = cpu_to_be32(l1_size);
15e6690a 1502 offset += align_offset(l1_size * sizeof(uint64_t), s->cluster_size);
585f8587
FB
1503
1504 s->refcount_table = qemu_mallocz(s->cluster_size);
585f8587 1505 s->refcount_block = qemu_mallocz(s->cluster_size);
3b46e624 1506
585f8587
FB
1507 s->refcount_table_offset = offset;
1508 header.refcount_table_offset = cpu_to_be64(offset);
1509 header.refcount_table_clusters = cpu_to_be32(1);
1510 offset += s->cluster_size;
1511
1512 s->refcount_table[0] = cpu_to_be64(offset);
1513 s->refcount_block_offset = offset;
1514 offset += s->cluster_size;
1515
1516 /* update refcounts */
1517 create_refcount_update(s, 0, header_size);
15e6690a 1518 create_refcount_update(s, s->l1_table_offset, l1_size * sizeof(uint64_t));
585f8587
FB
1519 create_refcount_update(s, s->refcount_table_offset, s->cluster_size);
1520 create_refcount_update(s, s->refcount_block_offset, s->cluster_size);
3b46e624 1521
585f8587
FB
1522 /* write all the data */
1523 write(fd, &header, sizeof(header));
1524 if (backing_file) {
1525 write(fd, backing_file, backing_filename_len);
1526 }
1527 lseek(fd, s->l1_table_offset, SEEK_SET);
1528 tmp = 0;
1529 for(i = 0;i < l1_size; i++) {
1530 write(fd, &tmp, sizeof(tmp));
1531 }
1532 lseek(fd, s->refcount_table_offset, SEEK_SET);
1533 write(fd, s->refcount_table, s->cluster_size);
3b46e624 1534
585f8587
FB
1535 lseek(fd, s->refcount_block_offset, SEEK_SET);
1536 write(fd, s->refcount_block, s->cluster_size);
1537
1538 qemu_free(s->refcount_table);
1539 qemu_free(s->refcount_block);
1540 close(fd);
1541 return 0;
585f8587
FB
1542}
1543
1544static int qcow_make_empty(BlockDriverState *bs)
1545{
1546#if 0
1547 /* XXX: not correct */
1548 BDRVQcowState *s = bs->opaque;
1549 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1550 int ret;
1551
1552 memset(s->l1_table, 0, l1_length);
1553 if (bdrv_pwrite(s->hd, s->l1_table_offset, s->l1_table, l1_length) < 0)
ac674887 1554 return -1;
585f8587
FB
1555 ret = bdrv_truncate(s->hd, s->l1_table_offset + l1_length);
1556 if (ret < 0)
1557 return ret;
3b46e624 1558
585f8587
FB
1559 l2_cache_reset(bs);
1560#endif
1561 return 0;
1562}
1563
1564/* XXX: put compressed sectors first, then all the cluster aligned
1565 tables to avoid losing bytes in alignment */
5fafdf24 1566static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
585f8587
FB
1567 const uint8_t *buf, int nb_sectors)
1568{
1569 BDRVQcowState *s = bs->opaque;
1570 z_stream strm;
1571 int ret, out_len;
1572 uint8_t *out_buf;
1573 uint64_t cluster_offset;
1574
1575 if (nb_sectors == 0) {
1576 /* align end of file to a sector boundary to ease reading with
1577 sector based I/Os */
1578 cluster_offset = bdrv_getlength(s->hd);
1579 cluster_offset = (cluster_offset + 511) & ~511;
1580 bdrv_truncate(s->hd, cluster_offset);
1581 return 0;
1582 }
1583
1584 if (nb_sectors != s->cluster_sectors)
1585 return -EINVAL;
1586
1587 out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
585f8587
FB
1588
1589 /* best compression, small window, no zlib header */
1590 memset(&strm, 0, sizeof(strm));
1591 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
5fafdf24 1592 Z_DEFLATED, -12,
585f8587
FB
1593 9, Z_DEFAULT_STRATEGY);
1594 if (ret != 0) {
1595 qemu_free(out_buf);
1596 return -1;
1597 }
1598
1599 strm.avail_in = s->cluster_size;
1600 strm.next_in = (uint8_t *)buf;
1601 strm.avail_out = s->cluster_size;
1602 strm.next_out = out_buf;
1603
1604 ret = deflate(&strm, Z_FINISH);
1605 if (ret != Z_STREAM_END && ret != Z_OK) {
1606 qemu_free(out_buf);
1607 deflateEnd(&strm);
1608 return -1;
1609 }
1610 out_len = strm.next_out - out_buf;
1611
1612 deflateEnd(&strm);
1613
1614 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1615 /* could not compress: write normal cluster */
1616 qcow_write(bs, sector_num, buf, s->cluster_sectors);
1617 } else {
52d893ec
AL
1618 cluster_offset = alloc_compressed_cluster_offset(bs, sector_num << 9,
1619 out_len);
1620 if (!cluster_offset)
1621 return -1;
585f8587
FB
1622 cluster_offset &= s->cluster_offset_mask;
1623 if (bdrv_pwrite(s->hd, cluster_offset, out_buf, out_len) != out_len) {
1624 qemu_free(out_buf);
1625 return -1;
1626 }
1627 }
3b46e624 1628
585f8587
FB
1629 qemu_free(out_buf);
1630 return 0;
1631}
1632
1633static void qcow_flush(BlockDriverState *bs)
1634{
1635 BDRVQcowState *s = bs->opaque;
1636 bdrv_flush(s->hd);
1637}
1638
1639static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1640{
1641 BDRVQcowState *s = bs->opaque;
1642 bdi->cluster_size = s->cluster_size;
5fafdf24 1643 bdi->vm_state_offset = (int64_t)s->l1_vm_state_index <<
585f8587
FB
1644 (s->cluster_bits + s->l2_bits);
1645 return 0;
1646}
1647
1648/*********************************************************/
1649/* snapshot support */
1650
1651/* update the refcounts of snapshots and the copied flag */
5fafdf24 1652static int update_snapshot_refcount(BlockDriverState *bs,
585f8587
FB
1653 int64_t l1_table_offset,
1654 int l1_size,
1655 int addend)
1656{
1657 BDRVQcowState *s = bs->opaque;
1658 uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated;
1659 int64_t old_offset, old_l2_offset;
1660 int l2_size, i, j, l1_modified, l2_modified, nb_csectors, refcount;
3b46e624 1661
585f8587
FB
1662 l2_cache_reset(bs);
1663
1664 l2_table = NULL;
1665 l1_table = NULL;
1666 l1_size2 = l1_size * sizeof(uint64_t);
1667 l1_allocated = 0;
1668 if (l1_table_offset != s->l1_table_offset) {
1669 l1_table = qemu_malloc(l1_size2);
585f8587 1670 l1_allocated = 1;
5fafdf24 1671 if (bdrv_pread(s->hd, l1_table_offset,
585f8587
FB
1672 l1_table, l1_size2) != l1_size2)
1673 goto fail;
1674 for(i = 0;i < l1_size; i++)
1675 be64_to_cpus(&l1_table[i]);
1676 } else {
1677 assert(l1_size == s->l1_size);
1678 l1_table = s->l1_table;
1679 l1_allocated = 0;
1680 }
3b46e624 1681
585f8587
FB
1682 l2_size = s->l2_size * sizeof(uint64_t);
1683 l2_table = qemu_malloc(l2_size);
585f8587
FB
1684 l1_modified = 0;
1685 for(i = 0; i < l1_size; i++) {
1686 l2_offset = l1_table[i];
1687 if (l2_offset) {
1688 old_l2_offset = l2_offset;
1689 l2_offset &= ~QCOW_OFLAG_COPIED;
1690 l2_modified = 0;
1691 if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
1692 goto fail;
1693 for(j = 0; j < s->l2_size; j++) {
1694 offset = be64_to_cpu(l2_table[j]);
1695 if (offset != 0) {
1696 old_offset = offset;
1697 offset &= ~QCOW_OFLAG_COPIED;
1698 if (offset & QCOW_OFLAG_COMPRESSED) {
5fafdf24 1699 nb_csectors = ((offset >> s->csize_shift) &
585f8587
FB
1700 s->csize_mask) + 1;
1701 if (addend != 0)
1702 update_refcount(bs, (offset & s->cluster_offset_mask) & ~511,
1703 nb_csectors * 512, addend);
1704 /* compressed clusters are never modified */
5fafdf24 1705 refcount = 2;
585f8587
FB
1706 } else {
1707 if (addend != 0) {
1708 refcount = update_cluster_refcount(bs, offset >> s->cluster_bits, addend);
1709 } else {
1710 refcount = get_refcount(bs, offset >> s->cluster_bits);
1711 }
1712 }
1713
1714 if (refcount == 1) {
1715 offset |= QCOW_OFLAG_COPIED;
1716 }
1717 if (offset != old_offset) {
1718 l2_table[j] = cpu_to_be64(offset);
1719 l2_modified = 1;
1720 }
1721 }
1722 }
1723 if (l2_modified) {
5fafdf24 1724 if (bdrv_pwrite(s->hd,
585f8587
FB
1725 l2_offset, l2_table, l2_size) != l2_size)
1726 goto fail;
1727 }
1728
1729 if (addend != 0) {
1730 refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend);
1731 } else {
1732 refcount = get_refcount(bs, l2_offset >> s->cluster_bits);
1733 }
1734 if (refcount == 1) {
1735 l2_offset |= QCOW_OFLAG_COPIED;
1736 }
1737 if (l2_offset != old_l2_offset) {
1738 l1_table[i] = l2_offset;
1739 l1_modified = 1;
1740 }
1741 }
1742 }
1743 if (l1_modified) {
1744 for(i = 0; i < l1_size; i++)
1745 cpu_to_be64s(&l1_table[i]);
5fafdf24 1746 if (bdrv_pwrite(s->hd, l1_table_offset, l1_table,
585f8587
FB
1747 l1_size2) != l1_size2)
1748 goto fail;
1749 for(i = 0; i < l1_size; i++)
1750 be64_to_cpus(&l1_table[i]);
1751 }
1752 if (l1_allocated)
1753 qemu_free(l1_table);
1754 qemu_free(l2_table);
1755 return 0;
1756 fail:
1757 if (l1_allocated)
1758 qemu_free(l1_table);
1759 qemu_free(l2_table);
1760 return -EIO;
1761}
1762
1763static void qcow_free_snapshots(BlockDriverState *bs)
1764{
1765 BDRVQcowState *s = bs->opaque;
1766 int i;
1767
1768 for(i = 0; i < s->nb_snapshots; i++) {
1769 qemu_free(s->snapshots[i].name);
1770 qemu_free(s->snapshots[i].id_str);
1771 }
1772 qemu_free(s->snapshots);
1773 s->snapshots = NULL;
1774 s->nb_snapshots = 0;
1775}
1776
1777static int qcow_read_snapshots(BlockDriverState *bs)
1778{
1779 BDRVQcowState *s = bs->opaque;
1780 QCowSnapshotHeader h;
1781 QCowSnapshot *sn;
1782 int i, id_str_size, name_size;
1783 int64_t offset;
1784 uint32_t extra_data_size;
1785
63c75dcd 1786 if (!s->nb_snapshots) {
1787 s->snapshots = NULL;
1788 s->snapshots_size = 0;
1789 return 0;
1790 }
1791
585f8587
FB
1792 offset = s->snapshots_offset;
1793 s->snapshots = qemu_mallocz(s->nb_snapshots * sizeof(QCowSnapshot));
585f8587
FB
1794 for(i = 0; i < s->nb_snapshots; i++) {
1795 offset = align_offset(offset, 8);
1796 if (bdrv_pread(s->hd, offset, &h, sizeof(h)) != sizeof(h))
1797 goto fail;
1798 offset += sizeof(h);
1799 sn = s->snapshots + i;
1800 sn->l1_table_offset = be64_to_cpu(h.l1_table_offset);
1801 sn->l1_size = be32_to_cpu(h.l1_size);
1802 sn->vm_state_size = be32_to_cpu(h.vm_state_size);
1803 sn->date_sec = be32_to_cpu(h.date_sec);
1804 sn->date_nsec = be32_to_cpu(h.date_nsec);
1805 sn->vm_clock_nsec = be64_to_cpu(h.vm_clock_nsec);
1806 extra_data_size = be32_to_cpu(h.extra_data_size);
1807
1808 id_str_size = be16_to_cpu(h.id_str_size);
1809 name_size = be16_to_cpu(h.name_size);
1810
1811 offset += extra_data_size;
1812
1813 sn->id_str = qemu_malloc(id_str_size + 1);
585f8587
FB
1814 if (bdrv_pread(s->hd, offset, sn->id_str, id_str_size) != id_str_size)
1815 goto fail;
1816 offset += id_str_size;
1817 sn->id_str[id_str_size] = '\0';
1818
1819 sn->name = qemu_malloc(name_size + 1);
585f8587
FB
1820 if (bdrv_pread(s->hd, offset, sn->name, name_size) != name_size)
1821 goto fail;
1822 offset += name_size;
1823 sn->name[name_size] = '\0';
1824 }
1825 s->snapshots_size = offset - s->snapshots_offset;
1826 return 0;
1827 fail:
1828 qcow_free_snapshots(bs);
1829 return -1;
1830}
1831
1832/* add at the end of the file a new list of snapshots */
1833static int qcow_write_snapshots(BlockDriverState *bs)
1834{
1835 BDRVQcowState *s = bs->opaque;
1836 QCowSnapshot *sn;
1837 QCowSnapshotHeader h;
1838 int i, name_size, id_str_size, snapshots_size;
1839 uint64_t data64;
1840 uint32_t data32;
1841 int64_t offset, snapshots_offset;
1842
1843 /* compute the size of the snapshots */
1844 offset = 0;
1845 for(i = 0; i < s->nb_snapshots; i++) {
1846 sn = s->snapshots + i;
1847 offset = align_offset(offset, 8);
1848 offset += sizeof(h);
1849 offset += strlen(sn->id_str);
1850 offset += strlen(sn->name);
1851 }
1852 snapshots_size = offset;
1853
1854 snapshots_offset = alloc_clusters(bs, snapshots_size);
1855 offset = snapshots_offset;
3b46e624 1856
585f8587
FB
1857 for(i = 0; i < s->nb_snapshots; i++) {
1858 sn = s->snapshots + i;
1859 memset(&h, 0, sizeof(h));
1860 h.l1_table_offset = cpu_to_be64(sn->l1_table_offset);
1861 h.l1_size = cpu_to_be32(sn->l1_size);
1862 h.vm_state_size = cpu_to_be32(sn->vm_state_size);
1863 h.date_sec = cpu_to_be32(sn->date_sec);
1864 h.date_nsec = cpu_to_be32(sn->date_nsec);
1865 h.vm_clock_nsec = cpu_to_be64(sn->vm_clock_nsec);
3b46e624 1866
585f8587
FB
1867 id_str_size = strlen(sn->id_str);
1868 name_size = strlen(sn->name);
1869 h.id_str_size = cpu_to_be16(id_str_size);
1870 h.name_size = cpu_to_be16(name_size);
1871 offset = align_offset(offset, 8);
1872 if (bdrv_pwrite(s->hd, offset, &h, sizeof(h)) != sizeof(h))
1873 goto fail;
1874 offset += sizeof(h);
1875 if (bdrv_pwrite(s->hd, offset, sn->id_str, id_str_size) != id_str_size)
1876 goto fail;
1877 offset += id_str_size;
1878 if (bdrv_pwrite(s->hd, offset, sn->name, name_size) != name_size)
1879 goto fail;
1880 offset += name_size;
1881 }
1882
1883 /* update the various header fields */
1884 data64 = cpu_to_be64(snapshots_offset);
1885 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, snapshots_offset),
1886 &data64, sizeof(data64)) != sizeof(data64))
1887 goto fail;
1888 data32 = cpu_to_be32(s->nb_snapshots);
1889 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, nb_snapshots),
1890 &data32, sizeof(data32)) != sizeof(data32))
1891 goto fail;
1892
1893 /* free the old snapshot table */
1894 free_clusters(bs, s->snapshots_offset, s->snapshots_size);
1895 s->snapshots_offset = snapshots_offset;
1896 s->snapshots_size = snapshots_size;
1897 return 0;
1898 fail:
1899 return -1;
1900}
1901
1902static void find_new_snapshot_id(BlockDriverState *bs,
1903 char *id_str, int id_str_size)
1904{
1905 BDRVQcowState *s = bs->opaque;
1906 QCowSnapshot *sn;
1907 int i, id, id_max = 0;
1908
1909 for(i = 0; i < s->nb_snapshots; i++) {
1910 sn = s->snapshots + i;
1911 id = strtoul(sn->id_str, NULL, 10);
1912 if (id > id_max)
1913 id_max = id;
1914 }
1915 snprintf(id_str, id_str_size, "%d", id_max + 1);
1916}
1917
1918static int find_snapshot_by_id(BlockDriverState *bs, const char *id_str)
1919{
1920 BDRVQcowState *s = bs->opaque;
1921 int i;
1922
1923 for(i = 0; i < s->nb_snapshots; i++) {
1924 if (!strcmp(s->snapshots[i].id_str, id_str))
1925 return i;
1926 }
1927 return -1;
1928}
1929
1930static int find_snapshot_by_id_or_name(BlockDriverState *bs, const char *name)
1931{
1932 BDRVQcowState *s = bs->opaque;
1933 int i, ret;
3b46e624 1934
585f8587
FB
1935 ret = find_snapshot_by_id(bs, name);
1936 if (ret >= 0)
1937 return ret;
1938 for(i = 0; i < s->nb_snapshots; i++) {
1939 if (!strcmp(s->snapshots[i].name, name))
1940 return i;
1941 }
1942 return -1;
1943}
1944
1945/* if no id is provided, a new one is constructed */
5fafdf24 1946static int qcow_snapshot_create(BlockDriverState *bs,
585f8587
FB
1947 QEMUSnapshotInfo *sn_info)
1948{
1949 BDRVQcowState *s = bs->opaque;
1950 QCowSnapshot *snapshots1, sn1, *sn = &sn1;
1951 int i, ret;
1952 uint64_t *l1_table = NULL;
3b46e624 1953
585f8587
FB
1954 memset(sn, 0, sizeof(*sn));
1955
1956 if (sn_info->id_str[0] == '\0') {
1957 /* compute a new id */
1958 find_new_snapshot_id(bs, sn_info->id_str, sizeof(sn_info->id_str));
1959 }
1960
1961 /* check that the ID is unique */
1962 if (find_snapshot_by_id(bs, sn_info->id_str) >= 0)
1963 return -ENOENT;
1964
1965 sn->id_str = qemu_strdup(sn_info->id_str);
1966 if (!sn->id_str)
1967 goto fail;
1968 sn->name = qemu_strdup(sn_info->name);
1969 if (!sn->name)
1970 goto fail;
1971 sn->vm_state_size = sn_info->vm_state_size;
1972 sn->date_sec = sn_info->date_sec;
1973 sn->date_nsec = sn_info->date_nsec;
1974 sn->vm_clock_nsec = sn_info->vm_clock_nsec;
1975
1976 ret = update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1);
1977 if (ret < 0)
1978 goto fail;
1979
1980 /* create the L1 table of the snapshot */
1981 sn->l1_table_offset = alloc_clusters(bs, s->l1_size * sizeof(uint64_t));
1982 sn->l1_size = s->l1_size;
1983
1984 l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
585f8587
FB
1985 for(i = 0; i < s->l1_size; i++) {
1986 l1_table[i] = cpu_to_be64(s->l1_table[i]);
1987 }
1988 if (bdrv_pwrite(s->hd, sn->l1_table_offset,
5fafdf24 1989 l1_table, s->l1_size * sizeof(uint64_t)) !=
585f8587
FB
1990 (s->l1_size * sizeof(uint64_t)))
1991 goto fail;
1992 qemu_free(l1_table);
1993 l1_table = NULL;
1994
1995 snapshots1 = qemu_malloc((s->nb_snapshots + 1) * sizeof(QCowSnapshot));
63c75dcd 1996 if (s->snapshots) {
1997 memcpy(snapshots1, s->snapshots, s->nb_snapshots * sizeof(QCowSnapshot));
1998 qemu_free(s->snapshots);
1999 }
585f8587
FB
2000 s->snapshots = snapshots1;
2001 s->snapshots[s->nb_snapshots++] = *sn;
2002
2003 if (qcow_write_snapshots(bs) < 0)
2004 goto fail;
2005#ifdef DEBUG_ALLOC
2006 check_refcounts(bs);
2007#endif
2008 return 0;
2009 fail:
2010 qemu_free(sn->name);
2011 qemu_free(l1_table);
2012 return -1;
2013}
2014
2015/* copy the snapshot 'snapshot_name' into the current disk image */
5fafdf24 2016static int qcow_snapshot_goto(BlockDriverState *bs,
585f8587
FB
2017 const char *snapshot_id)
2018{
2019 BDRVQcowState *s = bs->opaque;
2020 QCowSnapshot *sn;
2021 int i, snapshot_index, l1_size2;
2022
2023 snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
2024 if (snapshot_index < 0)
2025 return -ENOENT;
2026 sn = &s->snapshots[snapshot_index];
2027
2028 if (update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, -1) < 0)
2029 goto fail;
2030
2031 if (grow_l1_table(bs, sn->l1_size) < 0)
2032 goto fail;
2033
2034 s->l1_size = sn->l1_size;
2035 l1_size2 = s->l1_size * sizeof(uint64_t);
2036 /* copy the snapshot l1 table to the current l1 table */
5fafdf24 2037 if (bdrv_pread(s->hd, sn->l1_table_offset,
585f8587
FB
2038 s->l1_table, l1_size2) != l1_size2)
2039 goto fail;
2040 if (bdrv_pwrite(s->hd, s->l1_table_offset,
2041 s->l1_table, l1_size2) != l1_size2)
2042 goto fail;
2043 for(i = 0;i < s->l1_size; i++) {
2044 be64_to_cpus(&s->l1_table[i]);
2045 }
2046
2047 if (update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1) < 0)
2048 goto fail;
2049
2050#ifdef DEBUG_ALLOC
2051 check_refcounts(bs);
2052#endif
2053 return 0;
2054 fail:
2055 return -EIO;
2056}
2057
2058static int qcow_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
2059{
2060 BDRVQcowState *s = bs->opaque;
2061 QCowSnapshot *sn;
2062 int snapshot_index, ret;
3b46e624 2063
585f8587
FB
2064 snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
2065 if (snapshot_index < 0)
2066 return -ENOENT;
2067 sn = &s->snapshots[snapshot_index];
2068
2069 ret = update_snapshot_refcount(bs, sn->l1_table_offset, sn->l1_size, -1);
2070 if (ret < 0)
2071 return ret;
2072 /* must update the copied flag on the current cluster offsets */
2073 ret = update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 0);
2074 if (ret < 0)
2075 return ret;
2076 free_clusters(bs, sn->l1_table_offset, sn->l1_size * sizeof(uint64_t));
2077
2078 qemu_free(sn->id_str);
2079 qemu_free(sn->name);
2080 memmove(sn, sn + 1, (s->nb_snapshots - snapshot_index - 1) * sizeof(*sn));
2081 s->nb_snapshots--;
2082 ret = qcow_write_snapshots(bs);
2083 if (ret < 0) {
2084 /* XXX: restore snapshot if error ? */
2085 return ret;
2086 }
2087#ifdef DEBUG_ALLOC
2088 check_refcounts(bs);
2089#endif
2090 return 0;
2091}
2092
5fafdf24 2093static int qcow_snapshot_list(BlockDriverState *bs,
585f8587
FB
2094 QEMUSnapshotInfo **psn_tab)
2095{
2096 BDRVQcowState *s = bs->opaque;
2097 QEMUSnapshotInfo *sn_tab, *sn_info;
2098 QCowSnapshot *sn;
2099 int i;
2100
2101 sn_tab = qemu_mallocz(s->nb_snapshots * sizeof(QEMUSnapshotInfo));
585f8587
FB
2102 for(i = 0; i < s->nb_snapshots; i++) {
2103 sn_info = sn_tab + i;
2104 sn = s->snapshots + i;
2105 pstrcpy(sn_info->id_str, sizeof(sn_info->id_str),
2106 sn->id_str);
2107 pstrcpy(sn_info->name, sizeof(sn_info->name),
2108 sn->name);
2109 sn_info->vm_state_size = sn->vm_state_size;
2110 sn_info->date_sec = sn->date_sec;
2111 sn_info->date_nsec = sn->date_nsec;
2112 sn_info->vm_clock_nsec = sn->vm_clock_nsec;
2113 }
2114 *psn_tab = sn_tab;
2115 return s->nb_snapshots;
585f8587
FB
2116}
2117
2118/*********************************************************/
2119/* refcount handling */
2120
2121static int refcount_init(BlockDriverState *bs)
2122{
2123 BDRVQcowState *s = bs->opaque;
2124 int ret, refcount_table_size2, i;
3b46e624 2125
585f8587 2126 s->refcount_block_cache = qemu_malloc(s->cluster_size);
585f8587
FB
2127 refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t);
2128 s->refcount_table = qemu_malloc(refcount_table_size2);
585f8587
FB
2129 if (s->refcount_table_size > 0) {
2130 ret = bdrv_pread(s->hd, s->refcount_table_offset,
2131 s->refcount_table, refcount_table_size2);
2132 if (ret != refcount_table_size2)
2133 goto fail;
2134 for(i = 0; i < s->refcount_table_size; i++)
2135 be64_to_cpus(&s->refcount_table[i]);
2136 }
2137 return 0;
2138 fail:
2139 return -ENOMEM;
2140}
2141
2142static void refcount_close(BlockDriverState *bs)
2143{
2144 BDRVQcowState *s = bs->opaque;
2145 qemu_free(s->refcount_block_cache);
2146 qemu_free(s->refcount_table);
2147}
2148
2149
5fafdf24 2150static int load_refcount_block(BlockDriverState *bs,
585f8587
FB
2151 int64_t refcount_block_offset)
2152{
2153 BDRVQcowState *s = bs->opaque;
2154 int ret;
5fafdf24 2155 ret = bdrv_pread(s->hd, refcount_block_offset, s->refcount_block_cache,
585f8587
FB
2156 s->cluster_size);
2157 if (ret != s->cluster_size)
2158 return -EIO;
2159 s->refcount_block_cache_offset = refcount_block_offset;
2160 return 0;
2161}
2162
2163static int get_refcount(BlockDriverState *bs, int64_t cluster_index)
2164{
2165 BDRVQcowState *s = bs->opaque;
2166 int refcount_table_index, block_index;
2167 int64_t refcount_block_offset;
2168
2169 refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
2170 if (refcount_table_index >= s->refcount_table_size)
2171 return 0;
2172 refcount_block_offset = s->refcount_table[refcount_table_index];
2173 if (!refcount_block_offset)
2174 return 0;
2175 if (refcount_block_offset != s->refcount_block_cache_offset) {
2176 /* better than nothing: return allocated if read error */
2177 if (load_refcount_block(bs, refcount_block_offset) < 0)
2178 return 1;
2179 }
5fafdf24 2180 block_index = cluster_index &
585f8587
FB
2181 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
2182 return be16_to_cpu(s->refcount_block_cache[block_index]);
2183}
2184
2185/* return < 0 if error */
2186static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size)
2187{
2188 BDRVQcowState *s = bs->opaque;
2189 int i, nb_clusters;
2190
6db6c638
AL
2191 nb_clusters = size_to_clusters(s, size);
2192retry:
2193 for(i = 0; i < nb_clusters; i++) {
2194 int64_t i = s->free_cluster_index++;
2195 if (get_refcount(bs, i) != 0)
2196 goto retry;
2197 }
585f8587 2198#ifdef DEBUG_ALLOC2
6db6c638
AL
2199 printf("alloc_clusters: size=%lld -> %lld\n",
2200 size,
2201 (s->free_cluster_index - nb_clusters) << s->cluster_bits);
585f8587 2202#endif
6db6c638 2203 return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
585f8587
FB
2204}
2205
2206static int64_t alloc_clusters(BlockDriverState *bs, int64_t size)
2207{
2208 int64_t offset;
2209
2210 offset = alloc_clusters_noref(bs, size);
2211 update_refcount(bs, offset, size, 1);
2212 return offset;
2213}
2214
2215/* only used to allocate compressed sectors. We try to allocate
2216 contiguous sectors. size must be <= cluster_size */
2217static int64_t alloc_bytes(BlockDriverState *bs, int size)
2218{
2219 BDRVQcowState *s = bs->opaque;
2220 int64_t offset, cluster_offset;
2221 int free_in_cluster;
3b46e624 2222
585f8587
FB
2223 assert(size > 0 && size <= s->cluster_size);
2224 if (s->free_byte_offset == 0) {
2225 s->free_byte_offset = alloc_clusters(bs, s->cluster_size);
2226 }
2227 redo:
5fafdf24 2228 free_in_cluster = s->cluster_size -
585f8587
FB
2229 (s->free_byte_offset & (s->cluster_size - 1));
2230 if (size <= free_in_cluster) {
2231 /* enough space in current cluster */
2232 offset = s->free_byte_offset;
2233 s->free_byte_offset += size;
2234 free_in_cluster -= size;
2235 if (free_in_cluster == 0)
2236 s->free_byte_offset = 0;
2237 if ((offset & (s->cluster_size - 1)) != 0)
2238 update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
2239 } else {
2240 offset = alloc_clusters(bs, s->cluster_size);
2241 cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1);
2242 if ((cluster_offset + s->cluster_size) == offset) {
2243 /* we are lucky: contiguous data */
2244 offset = s->free_byte_offset;
2245 update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
2246 s->free_byte_offset += size;
2247 } else {
2248 s->free_byte_offset = offset;
2249 goto redo;
2250 }
2251 }
2252 return offset;
2253}
2254
5fafdf24 2255static void free_clusters(BlockDriverState *bs,
585f8587
FB
2256 int64_t offset, int64_t size)
2257{
2258 update_refcount(bs, offset, size, -1);
2259}
2260
2261static int grow_refcount_table(BlockDriverState *bs, int min_size)
2262{
2263 BDRVQcowState *s = bs->opaque;
2264 int new_table_size, new_table_size2, refcount_table_clusters, i, ret;
2265 uint64_t *new_table;
2266 int64_t table_offset;
643e5399 2267 uint8_t data[12];
23be50f1
TS
2268 int old_table_size;
2269 int64_t old_table_offset;
585f8587
FB
2270
2271 if (min_size <= s->refcount_table_size)
2272 return 0;
2273 /* compute new table size */
2274 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
2275 for(;;) {
2276 if (refcount_table_clusters == 0) {
2277 refcount_table_clusters = 1;
2278 } else {
2279 refcount_table_clusters = (refcount_table_clusters * 3 + 1) / 2;
2280 }
2281 new_table_size = refcount_table_clusters << (s->cluster_bits - 3);
2282 if (min_size <= new_table_size)
2283 break;
2284 }
15e6690a
FB
2285#ifdef DEBUG_ALLOC2
2286 printf("grow_refcount_table from %d to %d\n",
2287 s->refcount_table_size,
2288 new_table_size);
2289#endif
585f8587
FB
2290 new_table_size2 = new_table_size * sizeof(uint64_t);
2291 new_table = qemu_mallocz(new_table_size2);
5fafdf24 2292 memcpy(new_table, s->refcount_table,
585f8587
FB
2293 s->refcount_table_size * sizeof(uint64_t));
2294 for(i = 0; i < s->refcount_table_size; i++)
2295 cpu_to_be64s(&new_table[i]);
2296 /* Note: we cannot update the refcount now to avoid recursion */
2297 table_offset = alloc_clusters_noref(bs, new_table_size2);
2298 ret = bdrv_pwrite(s->hd, table_offset, new_table, new_table_size2);
5fafdf24 2299 if (ret != new_table_size2)
585f8587
FB
2300 goto fail;
2301 for(i = 0; i < s->refcount_table_size; i++)
2302 be64_to_cpus(&new_table[i]);
2303
643e5399
AL
2304 cpu_to_be64w((uint64_t*)data, table_offset);
2305 cpu_to_be32w((uint32_t*)(data + 8), refcount_table_clusters);
585f8587 2306 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, refcount_table_offset),
643e5399 2307 data, sizeof(data)) != sizeof(data))
585f8587
FB
2308 goto fail;
2309 qemu_free(s->refcount_table);
23be50f1
TS
2310 old_table_offset = s->refcount_table_offset;
2311 old_table_size = s->refcount_table_size;
585f8587
FB
2312 s->refcount_table = new_table;
2313 s->refcount_table_size = new_table_size;
a4080ece 2314 s->refcount_table_offset = table_offset;
585f8587
FB
2315
2316 update_refcount(bs, table_offset, new_table_size2, 1);
23be50f1 2317 free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t));
585f8587
FB
2318 return 0;
2319 fail:
2320 free_clusters(bs, table_offset, new_table_size2);
2321 qemu_free(new_table);
2322 return -EIO;
2323}
2324
2325/* addend must be 1 or -1 */
2326/* XXX: cache several refcount block clusters ? */
5fafdf24 2327static int update_cluster_refcount(BlockDriverState *bs,
585f8587
FB
2328 int64_t cluster_index,
2329 int addend)
2330{
2331 BDRVQcowState *s = bs->opaque;
2332 int64_t offset, refcount_block_offset;
2333 int ret, refcount_table_index, block_index, refcount;
2334 uint64_t data64;
2335
2336 refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
2337 if (refcount_table_index >= s->refcount_table_size) {
2338 if (addend < 0)
2339 return -EINVAL;
2340 ret = grow_refcount_table(bs, refcount_table_index + 1);
2341 if (ret < 0)
2342 return ret;
2343 }
2344 refcount_block_offset = s->refcount_table[refcount_table_index];
2345 if (!refcount_block_offset) {
2346 if (addend < 0)
2347 return -EINVAL;
2348 /* create a new refcount block */
2349 /* Note: we cannot update the refcount now to avoid recursion */
2350 offset = alloc_clusters_noref(bs, s->cluster_size);
2351 memset(s->refcount_block_cache, 0, s->cluster_size);
2352 ret = bdrv_pwrite(s->hd, offset, s->refcount_block_cache, s->cluster_size);
2353 if (ret != s->cluster_size)
2354 return -EINVAL;
2355 s->refcount_table[refcount_table_index] = offset;
2356 data64 = cpu_to_be64(offset);
5fafdf24
TS
2357 ret = bdrv_pwrite(s->hd, s->refcount_table_offset +
2358 refcount_table_index * sizeof(uint64_t),
585f8587
FB
2359 &data64, sizeof(data64));
2360 if (ret != sizeof(data64))
2361 return -EINVAL;
2362
2363 refcount_block_offset = offset;
2364 s->refcount_block_cache_offset = offset;
2365 update_refcount(bs, offset, s->cluster_size, 1);
2366 } else {
2367 if (refcount_block_offset != s->refcount_block_cache_offset) {
2368 if (load_refcount_block(bs, refcount_block_offset) < 0)
2369 return -EIO;
2370 }
2371 }
2372 /* we can update the count and save it */
5fafdf24 2373 block_index = cluster_index &
585f8587
FB
2374 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
2375 refcount = be16_to_cpu(s->refcount_block_cache[block_index]);
2376 refcount += addend;
2377 if (refcount < 0 || refcount > 0xffff)
2378 return -EINVAL;
2379 if (refcount == 0 && cluster_index < s->free_cluster_index) {
2380 s->free_cluster_index = cluster_index;
2381 }
2382 s->refcount_block_cache[block_index] = cpu_to_be16(refcount);
5fafdf24
TS
2383 if (bdrv_pwrite(s->hd,
2384 refcount_block_offset + (block_index << REFCOUNT_SHIFT),
585f8587
FB
2385 &s->refcount_block_cache[block_index], 2) != 2)
2386 return -EIO;
2387 return refcount;
2388}
2389
5fafdf24
TS
2390static void update_refcount(BlockDriverState *bs,
2391 int64_t offset, int64_t length,
585f8587
FB
2392 int addend)
2393{
2394 BDRVQcowState *s = bs->opaque;
2395 int64_t start, last, cluster_offset;
2396
2397#ifdef DEBUG_ALLOC2
5fafdf24 2398 printf("update_refcount: offset=%lld size=%lld addend=%d\n",
585f8587
FB
2399 offset, length, addend);
2400#endif
2401 if (length <= 0)
2402 return;
2403 start = offset & ~(s->cluster_size - 1);
2404 last = (offset + length - 1) & ~(s->cluster_size - 1);
5fafdf24 2405 for(cluster_offset = start; cluster_offset <= last;
585f8587
FB
2406 cluster_offset += s->cluster_size) {
2407 update_cluster_refcount(bs, cluster_offset >> s->cluster_bits, addend);
2408 }
2409}
2410
2411#ifdef DEBUG_ALLOC
5fafdf24
TS
2412static void inc_refcounts(BlockDriverState *bs,
2413 uint16_t *refcount_table,
585f8587
FB
2414 int refcount_table_size,
2415 int64_t offset, int64_t size)
2416{
2417 BDRVQcowState *s = bs->opaque;
2418 int64_t start, last, cluster_offset;
2419 int k;
3b46e624 2420
585f8587
FB
2421 if (size <= 0)
2422 return;
2423
2424 start = offset & ~(s->cluster_size - 1);
2425 last = (offset + size - 1) & ~(s->cluster_size - 1);
5fafdf24 2426 for(cluster_offset = start; cluster_offset <= last;
585f8587
FB
2427 cluster_offset += s->cluster_size) {
2428 k = cluster_offset >> s->cluster_bits;
2429 if (k < 0 || k >= refcount_table_size) {
2430 printf("ERROR: invalid cluster offset=0x%llx\n", cluster_offset);
2431 } else {
2432 if (++refcount_table[k] == 0) {
2433 printf("ERROR: overflow cluster offset=0x%llx\n", cluster_offset);
2434 }
2435 }
2436 }
2437}
2438
5fafdf24
TS
2439static int check_refcounts_l1(BlockDriverState *bs,
2440 uint16_t *refcount_table,
585f8587
FB
2441 int refcount_table_size,
2442 int64_t l1_table_offset, int l1_size,
2443 int check_copied)
2444{
2445 BDRVQcowState *s = bs->opaque;
2446 uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2;
2447 int l2_size, i, j, nb_csectors, refcount;
2448
2449 l2_table = NULL;
2450 l1_size2 = l1_size * sizeof(uint64_t);
2451
2452 inc_refcounts(bs, refcount_table, refcount_table_size,
2453 l1_table_offset, l1_size2);
2454
2455 l1_table = qemu_malloc(l1_size2);
5fafdf24 2456 if (bdrv_pread(s->hd, l1_table_offset,
585f8587
FB
2457 l1_table, l1_size2) != l1_size2)
2458 goto fail;
2459 for(i = 0;i < l1_size; i++)
2460 be64_to_cpus(&l1_table[i]);
3b46e624 2461
585f8587
FB
2462 l2_size = s->l2_size * sizeof(uint64_t);
2463 l2_table = qemu_malloc(l2_size);
585f8587
FB
2464 for(i = 0; i < l1_size; i++) {
2465 l2_offset = l1_table[i];
2466 if (l2_offset) {
2467 if (check_copied) {
2468 refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED) >> s->cluster_bits);
2469 if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) {
2470 printf("ERROR OFLAG_COPIED: l2_offset=%llx refcount=%d\n",
2471 l2_offset, refcount);
2472 }
2473 }
2474 l2_offset &= ~QCOW_OFLAG_COPIED;
2475 if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
2476 goto fail;
2477 for(j = 0; j < s->l2_size; j++) {
2478 offset = be64_to_cpu(l2_table[j]);
2479 if (offset != 0) {
2480 if (offset & QCOW_OFLAG_COMPRESSED) {
2481 if (offset & QCOW_OFLAG_COPIED) {
2482 printf("ERROR: cluster %lld: copied flag must never be set for compressed clusters\n",
2483 offset >> s->cluster_bits);
2484 offset &= ~QCOW_OFLAG_COPIED;
2485 }
5fafdf24 2486 nb_csectors = ((offset >> s->csize_shift) &
585f8587
FB
2487 s->csize_mask) + 1;
2488 offset &= s->cluster_offset_mask;
5fafdf24 2489 inc_refcounts(bs, refcount_table,
585f8587
FB
2490 refcount_table_size,
2491 offset & ~511, nb_csectors * 512);
2492 } else {
2493 if (check_copied) {
2494 refcount = get_refcount(bs, (offset & ~QCOW_OFLAG_COPIED) >> s->cluster_bits);
2495 if ((refcount == 1) != ((offset & QCOW_OFLAG_COPIED) != 0)) {
2496 printf("ERROR OFLAG_COPIED: offset=%llx refcount=%d\n",
2497 offset, refcount);
2498 }
2499 }
2500 offset &= ~QCOW_OFLAG_COPIED;
5fafdf24 2501 inc_refcounts(bs, refcount_table,
585f8587
FB
2502 refcount_table_size,
2503 offset, s->cluster_size);
2504 }
2505 }
2506 }
5fafdf24 2507 inc_refcounts(bs, refcount_table,
585f8587
FB
2508 refcount_table_size,
2509 l2_offset,
2510 s->cluster_size);
2511 }
2512 }
2513 qemu_free(l1_table);
2514 qemu_free(l2_table);
2515 return 0;
2516 fail:
2517 printf("ERROR: I/O error in check_refcounts_l1\n");
2518 qemu_free(l1_table);
2519 qemu_free(l2_table);
2520 return -EIO;
2521}
2522
2523static void check_refcounts(BlockDriverState *bs)
2524{
2525 BDRVQcowState *s = bs->opaque;
2526 int64_t size;
2527 int nb_clusters, refcount1, refcount2, i;
2528 QCowSnapshot *sn;
2529 uint16_t *refcount_table;
2530
2531 size = bdrv_getlength(s->hd);
6db6c638 2532 nb_clusters = size_to_clusters(s, size);
585f8587 2533 refcount_table = qemu_mallocz(nb_clusters * sizeof(uint16_t));
15e6690a 2534
585f8587
FB
2535 /* header */
2536 inc_refcounts(bs, refcount_table, nb_clusters,
2537 0, s->cluster_size);
3b46e624 2538
585f8587
FB
2539 check_refcounts_l1(bs, refcount_table, nb_clusters,
2540 s->l1_table_offset, s->l1_size, 1);
2541
2542 /* snapshots */
2543 for(i = 0; i < s->nb_snapshots; i++) {
2544 sn = s->snapshots + i;
2545 check_refcounts_l1(bs, refcount_table, nb_clusters,
2546 sn->l1_table_offset, sn->l1_size, 0);
2547 }
2548 inc_refcounts(bs, refcount_table, nb_clusters,
2549 s->snapshots_offset, s->snapshots_size);
2550
2551 /* refcount data */
2552 inc_refcounts(bs, refcount_table, nb_clusters,
5fafdf24 2553 s->refcount_table_offset,
585f8587
FB
2554 s->refcount_table_size * sizeof(uint64_t));
2555 for(i = 0; i < s->refcount_table_size; i++) {
2556 int64_t offset;
2557 offset = s->refcount_table[i];
2558 if (offset != 0) {
2559 inc_refcounts(bs, refcount_table, nb_clusters,
2560 offset, s->cluster_size);
2561 }
2562 }
2563
2564 /* compare ref counts */
2565 for(i = 0; i < nb_clusters; i++) {
2566 refcount1 = get_refcount(bs, i);
2567 refcount2 = refcount_table[i];
2568 if (refcount1 != refcount2)
2569 printf("ERROR cluster %d refcount=%d reference=%d\n",
2570 i, refcount1, refcount2);
2571 }
2572
2573 qemu_free(refcount_table);
2574}
2575
2576#if 0
2577static void dump_refcounts(BlockDriverState *bs)
2578{
2579 BDRVQcowState *s = bs->opaque;
2580 int64_t nb_clusters, k, k1, size;
2581 int refcount;
2582
2583 size = bdrv_getlength(s->hd);
6db6c638 2584 nb_clusters = size_to_clusters(s, size);
585f8587
FB
2585 for(k = 0; k < nb_clusters;) {
2586 k1 = k;
2587 refcount = get_refcount(bs, k);
2588 k++;
2589 while (k < nb_clusters && get_refcount(bs, k) == refcount)
2590 k++;
2591 printf("%lld: refcount=%d nb=%lld\n", k, refcount, k - k1);
2592 }
2593}
2594#endif
2595#endif
2596
2597BlockDriver bdrv_qcow2 = {
2598 "qcow2",
2599 sizeof(BDRVQcowState),
2600 qcow_probe,
2601 qcow_open,
2602 NULL,
2603 NULL,
2604 qcow_close,
2605 qcow_create,
2606 qcow_flush,
2607 qcow_is_allocated,
2608 qcow_set_key,
2609 qcow_make_empty,
2610
585f8587
FB
2611 .bdrv_aio_read = qcow_aio_read,
2612 .bdrv_aio_write = qcow_aio_write,
2613 .bdrv_aio_cancel = qcow_aio_cancel,
ce1a14dc 2614 .aiocb_size = sizeof(QCowAIOCB),
585f8587
FB
2615 .bdrv_write_compressed = qcow_write_compressed,
2616
2617 .bdrv_snapshot_create = qcow_snapshot_create,
2618 .bdrv_snapshot_goto = qcow_snapshot_goto,
2619 .bdrv_snapshot_delete = qcow_snapshot_delete,
2620 .bdrv_snapshot_list = qcow_snapshot_list,
2621 .bdrv_get_info = qcow_get_info,
2622};