]> git.proxmox.com Git - qemu.git/blame - block-qcow2.c
remove bdrv_aio_read/bdrv_aio_write (Christoph Hellwig)
[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
9b80ddf3 48//#define DEBUG_EXT
5fafdf24 49
585f8587
FB
50#define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
51#define QCOW_VERSION 2
52
53#define QCOW_CRYPT_NONE 0
54#define QCOW_CRYPT_AES 1
55
095a9c58
AL
56#define QCOW_MAX_CRYPT_CLUSTERS 32
57
585f8587
FB
58/* indicate that the refcount of the referenced cluster is exactly one. */
59#define QCOW_OFLAG_COPIED (1LL << 63)
60/* indicate that the cluster is compressed (they never have the copied flag) */
61#define QCOW_OFLAG_COMPRESSED (1LL << 62)
62
63#define REFCOUNT_SHIFT 1 /* refcount size is 2 bytes */
64
585f8587
FB
65typedef struct QCowHeader {
66 uint32_t magic;
67 uint32_t version;
68 uint64_t backing_file_offset;
69 uint32_t backing_file_size;
70 uint32_t cluster_bits;
71 uint64_t size; /* in bytes */
72 uint32_t crypt_method;
73 uint32_t l1_size; /* XXX: save number of clusters instead ? */
74 uint64_t l1_table_offset;
75 uint64_t refcount_table_offset;
76 uint32_t refcount_table_clusters;
77 uint32_t nb_snapshots;
78 uint64_t snapshots_offset;
79} QCowHeader;
80
9b80ddf3
AL
81
82typedef struct {
83 uint32_t magic;
84 uint32_t len;
85} QCowExtension;
86#define QCOW_EXT_MAGIC_END 0
f965509c 87#define QCOW_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
9b80ddf3
AL
88
89
585f8587
FB
90typedef struct __attribute__((packed)) QCowSnapshotHeader {
91 /* header is 8 byte aligned */
92 uint64_t l1_table_offset;
93
94 uint32_t l1_size;
95 uint16_t id_str_size;
96 uint16_t name_size;
97
98 uint32_t date_sec;
99 uint32_t date_nsec;
100
101 uint64_t vm_clock_nsec;
102
103 uint32_t vm_state_size;
104 uint32_t extra_data_size; /* for extension */
105 /* extra data follows */
106 /* id_str follows */
107 /* name follows */
108} QCowSnapshotHeader;
109
110#define L2_CACHE_SIZE 16
111
112typedef struct QCowSnapshot {
113 uint64_t l1_table_offset;
114 uint32_t l1_size;
115 char *id_str;
116 char *name;
117 uint32_t vm_state_size;
118 uint32_t date_sec;
119 uint32_t date_nsec;
120 uint64_t vm_clock_nsec;
121} QCowSnapshot;
122
123typedef struct BDRVQcowState {
124 BlockDriverState *hd;
125 int cluster_bits;
126 int cluster_size;
127 int cluster_sectors;
128 int l2_bits;
129 int l2_size;
130 int l1_size;
131 int l1_vm_state_index;
132 int csize_shift;
133 int csize_mask;
134 uint64_t cluster_offset_mask;
135 uint64_t l1_table_offset;
136 uint64_t *l1_table;
137 uint64_t *l2_cache;
138 uint64_t l2_cache_offsets[L2_CACHE_SIZE];
139 uint32_t l2_cache_counts[L2_CACHE_SIZE];
140 uint8_t *cluster_cache;
141 uint8_t *cluster_data;
142 uint64_t cluster_cache_offset;
143
144 uint64_t *refcount_table;
145 uint64_t refcount_table_offset;
146 uint32_t refcount_table_size;
147 uint64_t refcount_block_cache_offset;
148 uint16_t *refcount_block_cache;
149 int64_t free_cluster_index;
150 int64_t free_byte_offset;
151
152 uint32_t crypt_method; /* current crypt method, 0 if no key yet */
153 uint32_t crypt_method_header;
154 AES_KEY aes_encrypt_key;
155 AES_KEY aes_decrypt_key;
156 uint64_t snapshots_offset;
157 int snapshots_size;
158 int nb_snapshots;
159 QCowSnapshot *snapshots;
160} BDRVQcowState;
161
162static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset);
5fafdf24 163static int qcow_read(BlockDriverState *bs, int64_t sector_num,
585f8587
FB
164 uint8_t *buf, int nb_sectors);
165static int qcow_read_snapshots(BlockDriverState *bs);
166static void qcow_free_snapshots(BlockDriverState *bs);
167static int refcount_init(BlockDriverState *bs);
168static void refcount_close(BlockDriverState *bs);
169static int get_refcount(BlockDriverState *bs, int64_t cluster_index);
5fafdf24 170static int update_cluster_refcount(BlockDriverState *bs,
585f8587
FB
171 int64_t cluster_index,
172 int addend);
5fafdf24
TS
173static void update_refcount(BlockDriverState *bs,
174 int64_t offset, int64_t length,
585f8587
FB
175 int addend);
176static int64_t alloc_clusters(BlockDriverState *bs, int64_t size);
177static int64_t alloc_bytes(BlockDriverState *bs, int size);
5fafdf24 178static void free_clusters(BlockDriverState *bs,
585f8587
FB
179 int64_t offset, int64_t size);
180#ifdef DEBUG_ALLOC
181static void check_refcounts(BlockDriverState *bs);
182#endif
183
184static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
185{
186 const QCowHeader *cow_header = (const void *)buf;
3b46e624 187
585f8587
FB
188 if (buf_size >= sizeof(QCowHeader) &&
189 be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
5fafdf24 190 be32_to_cpu(cow_header->version) == QCOW_VERSION)
585f8587
FB
191 return 100;
192 else
193 return 0;
194}
195
9b80ddf3
AL
196
197/*
198 * read qcow2 extension and fill bs
199 * start reading from start_offset
200 * finish reading upon magic of value 0 or when end_offset reached
201 * unknown magic is skipped (future extension this version knows nothing about)
202 * return 0 upon success, non-0 otherwise
203 */
204static int qcow_read_extensions(BlockDriverState *bs, uint64_t start_offset,
205 uint64_t end_offset)
206{
207 BDRVQcowState *s = bs->opaque;
208 QCowExtension ext;
209 uint64_t offset;
210
211#ifdef DEBUG_EXT
212 printf("qcow_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
213#endif
214 offset = start_offset;
215 while (offset < end_offset) {
216
217#ifdef DEBUG_EXT
218 /* Sanity check */
219 if (offset > s->cluster_size)
220 printf("qcow_handle_extension: suspicious offset %lu\n", offset);
221
222 printf("attemting to read extended header in offset %lu\n", offset);
223#endif
224
225 if (bdrv_pread(s->hd, offset, &ext, sizeof(ext)) != sizeof(ext)) {
4c978075
AL
226 fprintf(stderr, "qcow_handle_extension: ERROR: pread fail from offset %llu\n",
227 (unsigned long long)offset);
9b80ddf3
AL
228 return 1;
229 }
230 be32_to_cpus(&ext.magic);
231 be32_to_cpus(&ext.len);
232 offset += sizeof(ext);
233#ifdef DEBUG_EXT
234 printf("ext.magic = 0x%x\n", ext.magic);
235#endif
236 switch (ext.magic) {
237 case QCOW_EXT_MAGIC_END:
238 return 0;
f965509c
AL
239
240 case QCOW_EXT_MAGIC_BACKING_FORMAT:
241 if (ext.len >= sizeof(bs->backing_format)) {
242 fprintf(stderr, "ERROR: ext_backing_format: len=%u too large"
4c978075 243 " (>=%zu)\n",
f965509c
AL
244 ext.len, sizeof(bs->backing_format));
245 return 2;
246 }
247 if (bdrv_pread(s->hd, offset , bs->backing_format,
248 ext.len) != ext.len)
249 return 3;
250 bs->backing_format[ext.len] = '\0';
251#ifdef DEBUG_EXT
252 printf("Qcow2: Got format extension %s\n", bs->backing_format);
253#endif
254 offset += ((ext.len + 7) & ~7);
255 break;
256
9b80ddf3
AL
257 default:
258 /* unknown magic -- just skip it */
259 offset += ((ext.len + 7) & ~7);
260 break;
261 }
262 }
263
264 return 0;
265}
266
267
585f8587
FB
268static int qcow_open(BlockDriverState *bs, const char *filename, int flags)
269{
270 BDRVQcowState *s = bs->opaque;
271 int len, i, shift, ret;
272 QCowHeader header;
9b80ddf3 273 uint64_t ext_end;
585f8587 274
4dc822d7
AL
275 /* Performance is terrible right now with cache=writethrough due mainly
276 * to reference count updates. If the user does not explicitly specify
277 * a caching type, force to writeback caching.
278 */
279 if ((flags & BDRV_O_CACHE_DEF)) {
280 flags |= BDRV_O_CACHE_WB;
281 flags &= ~BDRV_O_CACHE_DEF;
282 }
b5eff355 283 ret = bdrv_file_open(&s->hd, filename, flags);
585f8587
FB
284 if (ret < 0)
285 return ret;
286 if (bdrv_pread(s->hd, 0, &header, sizeof(header)) != sizeof(header))
287 goto fail;
288 be32_to_cpus(&header.magic);
289 be32_to_cpus(&header.version);
290 be64_to_cpus(&header.backing_file_offset);
291 be32_to_cpus(&header.backing_file_size);
292 be64_to_cpus(&header.size);
293 be32_to_cpus(&header.cluster_bits);
294 be32_to_cpus(&header.crypt_method);
295 be64_to_cpus(&header.l1_table_offset);
296 be32_to_cpus(&header.l1_size);
297 be64_to_cpus(&header.refcount_table_offset);
298 be32_to_cpus(&header.refcount_table_clusters);
299 be64_to_cpus(&header.snapshots_offset);
300 be32_to_cpus(&header.nb_snapshots);
3b46e624 301
585f8587
FB
302 if (header.magic != QCOW_MAGIC || header.version != QCOW_VERSION)
303 goto fail;
5fafdf24
TS
304 if (header.size <= 1 ||
305 header.cluster_bits < 9 ||
585f8587
FB
306 header.cluster_bits > 16)
307 goto fail;
308 if (header.crypt_method > QCOW_CRYPT_AES)
309 goto fail;
310 s->crypt_method_header = header.crypt_method;
311 if (s->crypt_method_header)
312 bs->encrypted = 1;
313 s->cluster_bits = header.cluster_bits;
314 s->cluster_size = 1 << s->cluster_bits;
315 s->cluster_sectors = 1 << (s->cluster_bits - 9);
316 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
317 s->l2_size = 1 << s->l2_bits;
318 bs->total_sectors = header.size / 512;
319 s->csize_shift = (62 - (s->cluster_bits - 8));
320 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
321 s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
322 s->refcount_table_offset = header.refcount_table_offset;
5fafdf24 323 s->refcount_table_size =
585f8587
FB
324 header.refcount_table_clusters << (s->cluster_bits - 3);
325
326 s->snapshots_offset = header.snapshots_offset;
327 s->nb_snapshots = header.nb_snapshots;
328
329 /* read the level 1 table */
330 s->l1_size = header.l1_size;
331 shift = s->cluster_bits + s->l2_bits;
332 s->l1_vm_state_index = (header.size + (1LL << shift) - 1) >> shift;
333 /* the L1 table must contain at least enough entries to put
334 header.size bytes */
335 if (s->l1_size < s->l1_vm_state_index)
336 goto fail;
337 s->l1_table_offset = header.l1_table_offset;
338 s->l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
5fafdf24 339 if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, s->l1_size * sizeof(uint64_t)) !=
585f8587
FB
340 s->l1_size * sizeof(uint64_t))
341 goto fail;
342 for(i = 0;i < s->l1_size; i++) {
343 be64_to_cpus(&s->l1_table[i]);
344 }
345 /* alloc L2 cache */
346 s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
585f8587 347 s->cluster_cache = qemu_malloc(s->cluster_size);
585f8587 348 /* one more sector for decompressed data alignment */
095a9c58
AL
349 s->cluster_data = qemu_malloc(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
350 + 512);
585f8587 351 s->cluster_cache_offset = -1;
3b46e624 352
585f8587
FB
353 if (refcount_init(bs) < 0)
354 goto fail;
355
9b80ddf3
AL
356 /* read qcow2 extensions */
357 if (header.backing_file_offset)
358 ext_end = header.backing_file_offset;
359 else
360 ext_end = s->cluster_size;
361 if (qcow_read_extensions(bs, sizeof(header), ext_end))
362 goto fail;
363
585f8587
FB
364 /* read the backing file name */
365 if (header.backing_file_offset != 0) {
366 len = header.backing_file_size;
367 if (len > 1023)
368 len = 1023;
369 if (bdrv_pread(s->hd, header.backing_file_offset, bs->backing_file, len) != len)
370 goto fail;
371 bs->backing_file[len] = '\0';
372 }
373 if (qcow_read_snapshots(bs) < 0)
374 goto fail;
375
376#ifdef DEBUG_ALLOC
377 check_refcounts(bs);
378#endif
379 return 0;
380
381 fail:
382 qcow_free_snapshots(bs);
383 refcount_close(bs);
384 qemu_free(s->l1_table);
385 qemu_free(s->l2_cache);
386 qemu_free(s->cluster_cache);
387 qemu_free(s->cluster_data);
388 bdrv_delete(s->hd);
389 return -1;
390}
391
392static int qcow_set_key(BlockDriverState *bs, const char *key)
393{
394 BDRVQcowState *s = bs->opaque;
395 uint8_t keybuf[16];
396 int len, i;
3b46e624 397
585f8587
FB
398 memset(keybuf, 0, 16);
399 len = strlen(key);
400 if (len > 16)
401 len = 16;
402 /* XXX: we could compress the chars to 7 bits to increase
403 entropy */
404 for(i = 0;i < len;i++) {
405 keybuf[i] = key[i];
406 }
407 s->crypt_method = s->crypt_method_header;
408
409 if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
410 return -1;
411 if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
412 return -1;
413#if 0
414 /* test */
415 {
416 uint8_t in[16];
417 uint8_t out[16];
418 uint8_t tmp[16];
419 for(i=0;i<16;i++)
420 in[i] = i;
421 AES_encrypt(in, tmp, &s->aes_encrypt_key);
422 AES_decrypt(tmp, out, &s->aes_decrypt_key);
423 for(i = 0; i < 16; i++)
424 printf(" %02x", tmp[i]);
425 printf("\n");
426 for(i = 0; i < 16; i++)
427 printf(" %02x", out[i]);
428 printf("\n");
429 }
430#endif
431 return 0;
432}
433
434/* The crypt function is compatible with the linux cryptoloop
435 algorithm for < 4 GB images. NOTE: out_buf == in_buf is
436 supported */
437static void encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
438 uint8_t *out_buf, const uint8_t *in_buf,
439 int nb_sectors, int enc,
440 const AES_KEY *key)
441{
442 union {
443 uint64_t ll[2];
444 uint8_t b[16];
445 } ivec;
446 int i;
447
448 for(i = 0; i < nb_sectors; i++) {
449 ivec.ll[0] = cpu_to_le64(sector_num);
450 ivec.ll[1] = 0;
5fafdf24 451 AES_cbc_encrypt(in_buf, out_buf, 512, key,
585f8587
FB
452 ivec.b, enc);
453 sector_num++;
454 in_buf += 512;
455 out_buf += 512;
456 }
457}
458
459static int copy_sectors(BlockDriverState *bs, uint64_t start_sect,
460 uint64_t cluster_offset, int n_start, int n_end)
461{
462 BDRVQcowState *s = bs->opaque;
463 int n, ret;
464
465 n = n_end - n_start;
466 if (n <= 0)
467 return 0;
468 ret = qcow_read(bs, start_sect + n_start, s->cluster_data, n);
469 if (ret < 0)
470 return ret;
471 if (s->crypt_method) {
5fafdf24
TS
472 encrypt_sectors(s, start_sect + n_start,
473 s->cluster_data,
585f8587
FB
474 s->cluster_data, n, 1,
475 &s->aes_encrypt_key);
476 }
5fafdf24 477 ret = bdrv_write(s->hd, (cluster_offset >> 9) + n_start,
585f8587
FB
478 s->cluster_data, n);
479 if (ret < 0)
480 return ret;
481 return 0;
482}
483
484static void l2_cache_reset(BlockDriverState *bs)
485{
486 BDRVQcowState *s = bs->opaque;
487
488 memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
489 memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
490 memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
491}
492
493static inline int l2_cache_new_entry(BlockDriverState *bs)
494{
495 BDRVQcowState *s = bs->opaque;
496 uint32_t min_count;
497 int min_index, i;
498
499 /* find a new entry in the least used one */
500 min_index = 0;
501 min_count = 0xffffffff;
502 for(i = 0; i < L2_CACHE_SIZE; i++) {
503 if (s->l2_cache_counts[i] < min_count) {
504 min_count = s->l2_cache_counts[i];
505 min_index = i;
506 }
507 }
508 return min_index;
509}
510
511static int64_t align_offset(int64_t offset, int n)
512{
513 offset = (offset + n - 1) & ~(n - 1);
514 return offset;
515}
516
517static int grow_l1_table(BlockDriverState *bs, int min_size)
518{
519 BDRVQcowState *s = bs->opaque;
520 int new_l1_size, new_l1_size2, ret, i;
521 uint64_t *new_l1_table;
522 uint64_t new_l1_table_offset;
643e5399 523 uint8_t data[12];
585f8587
FB
524
525 new_l1_size = s->l1_size;
526 if (min_size <= new_l1_size)
527 return 0;
528 while (min_size > new_l1_size) {
529 new_l1_size = (new_l1_size * 3 + 1) / 2;
530 }
531#ifdef DEBUG_ALLOC2
532 printf("grow l1_table from %d to %d\n", s->l1_size, new_l1_size);
533#endif
534
535 new_l1_size2 = sizeof(uint64_t) * new_l1_size;
536 new_l1_table = qemu_mallocz(new_l1_size2);
585f8587
FB
537 memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t));
538
539 /* write new table (align to cluster) */
540 new_l1_table_offset = alloc_clusters(bs, new_l1_size2);
3b46e624 541
585f8587
FB
542 for(i = 0; i < s->l1_size; i++)
543 new_l1_table[i] = cpu_to_be64(new_l1_table[i]);
544 ret = bdrv_pwrite(s->hd, new_l1_table_offset, new_l1_table, new_l1_size2);
545 if (ret != new_l1_size2)
546 goto fail;
547 for(i = 0; i < s->l1_size; i++)
548 new_l1_table[i] = be64_to_cpu(new_l1_table[i]);
3b46e624 549
585f8587 550 /* set new table */
643e5399
AL
551 cpu_to_be32w((uint32_t*)data, new_l1_size);
552 cpu_to_be64w((uint64_t*)(data + 4), new_l1_table_offset);
553 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, l1_size), data,
554 sizeof(data)) != sizeof(data))
585f8587
FB
555 goto fail;
556 qemu_free(s->l1_table);
557 free_clusters(bs, s->l1_table_offset, s->l1_size * sizeof(uint64_t));
558 s->l1_table_offset = new_l1_table_offset;
559 s->l1_table = new_l1_table;
560 s->l1_size = new_l1_size;
561 return 0;
562 fail:
563 qemu_free(s->l1_table);
564 return -EIO;
565}
566
108534b9
AL
567/*
568 * seek_l2_table
585f8587 569 *
108534b9
AL
570 * seek l2_offset in the l2_cache table
571 * if not found, return NULL,
572 * if found,
573 * increments the l2 cache hit count of the entry,
574 * if counter overflow, divide by two all counters
575 * return the pointer to the l2 cache entry
585f8587 576 *
108534b9
AL
577 */
578
579static uint64_t *seek_l2_table(BDRVQcowState *s, uint64_t l2_offset)
580{
581 int i, j;
582
583 for(i = 0; i < L2_CACHE_SIZE; i++) {
584 if (l2_offset == s->l2_cache_offsets[i]) {
585 /* increment the hit count */
586 if (++s->l2_cache_counts[i] == 0xffffffff) {
587 for(j = 0; j < L2_CACHE_SIZE; j++) {
588 s->l2_cache_counts[j] >>= 1;
589 }
590 }
591 return s->l2_cache + (i << s->l2_bits);
592 }
593 }
594 return NULL;
595}
596
597/*
598 * l2_load
599 *
600 * Loads a L2 table into memory. If the table is in the cache, the cache
601 * is used; otherwise the L2 table is loaded from the image file.
602 *
603 * Returns a pointer to the L2 table on success, or NULL if the read from
604 * the image file failed.
605 */
606
607static uint64_t *l2_load(BlockDriverState *bs, uint64_t l2_offset)
608{
609 BDRVQcowState *s = bs->opaque;
610 int min_index;
611 uint64_t *l2_table;
612
613 /* seek if the table for the given offset is in the cache */
614
615 l2_table = seek_l2_table(s, l2_offset);
616 if (l2_table != NULL)
617 return l2_table;
618
619 /* not found: load a new entry in the least used one */
620
621 min_index = l2_cache_new_entry(bs);
622 l2_table = s->l2_cache + (min_index << s->l2_bits);
623 if (bdrv_pread(s->hd, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
624 s->l2_size * sizeof(uint64_t))
625 return NULL;
626 s->l2_cache_offsets[min_index] = l2_offset;
627 s->l2_cache_counts[min_index] = 1;
628
629 return l2_table;
630}
631
632/*
633 * l2_allocate
585f8587 634 *
108534b9
AL
635 * Allocate a new l2 entry in the file. If l1_index points to an already
636 * used entry in the L2 table (i.e. we are doing a copy on write for the L2
637 * table) copy the contents of the old L2 table into the newly allocated one.
638 * Otherwise the new table is initialized with zeros.
585f8587 639 *
585f8587 640 */
108534b9
AL
641
642static uint64_t *l2_allocate(BlockDriverState *bs, int l1_index)
643{
644 BDRVQcowState *s = bs->opaque;
645 int min_index;
646 uint64_t old_l2_offset, tmp;
647 uint64_t *l2_table, l2_offset;
648
649 old_l2_offset = s->l1_table[l1_index];
650
651 /* allocate a new l2 entry */
652
653 l2_offset = alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
654
655 /* update the L1 entry */
656
657 s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
658
659 tmp = cpu_to_be64(l2_offset | QCOW_OFLAG_COPIED);
660 if (bdrv_pwrite(s->hd, s->l1_table_offset + l1_index * sizeof(tmp),
661 &tmp, sizeof(tmp)) != sizeof(tmp))
662 return NULL;
663
664 /* allocate a new entry in the l2 cache */
665
666 min_index = l2_cache_new_entry(bs);
667 l2_table = s->l2_cache + (min_index << s->l2_bits);
668
669 if (old_l2_offset == 0) {
670 /* if there was no old l2 table, clear the new table */
671 memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
672 } else {
673 /* if there was an old l2 table, read it from the disk */
674 if (bdrv_pread(s->hd, old_l2_offset,
675 l2_table, s->l2_size * sizeof(uint64_t)) !=
676 s->l2_size * sizeof(uint64_t))
677 return NULL;
678 }
679 /* write the l2 table to the file */
680 if (bdrv_pwrite(s->hd, l2_offset,
681 l2_table, s->l2_size * sizeof(uint64_t)) !=
682 s->l2_size * sizeof(uint64_t))
683 return NULL;
684
685 /* update the l2 cache entry */
686
687 s->l2_cache_offsets[min_index] = l2_offset;
688 s->l2_cache_counts[min_index] = 1;
689
690 return l2_table;
691}
692
6db6c638
AL
693static int size_to_clusters(BDRVQcowState *s, int64_t size)
694{
695 return (size + (s->cluster_size - 1)) >> s->cluster_bits;
696}
697
698static int count_contiguous_clusters(uint64_t nb_clusters, int cluster_size,
ff4b91c2 699 uint64_t *l2_table, uint64_t start, uint64_t mask)
6db6c638
AL
700{
701 int i;
702 uint64_t offset = be64_to_cpu(l2_table[0]) & ~mask;
703
ab5ccbd6
AL
704 if (!offset)
705 return 0;
706
ff4b91c2 707 for (i = start; i < start + nb_clusters; i++)
6db6c638
AL
708 if (offset + i * cluster_size != (be64_to_cpu(l2_table[i]) & ~mask))
709 break;
710
ff4b91c2 711 return (i - start);
6db6c638
AL
712}
713
714static int count_contiguous_free_clusters(uint64_t nb_clusters, uint64_t *l2_table)
715{
716 int i = 0;
717
718 while(nb_clusters-- && l2_table[i] == 0)
719 i++;
720
721 return i;
722}
723
05203524
AL
724/*
725 * get_cluster_offset
726 *
727 * For a given offset of the disk image, return cluster offset in
728 * qcow2 file.
729 *
095a9c58
AL
730 * on entry, *num is the number of contiguous clusters we'd like to
731 * access following offset.
732 *
733 * on exit, *num is the number of contiguous clusters we can read.
734 *
05203524
AL
735 * Return 1, if the offset is found
736 * Return 0, otherwise.
737 *
738 */
739
095a9c58
AL
740static uint64_t get_cluster_offset(BlockDriverState *bs,
741 uint64_t offset, int *num)
05203524
AL
742{
743 BDRVQcowState *s = bs->opaque;
744 int l1_index, l2_index;
6db6c638
AL
745 uint64_t l2_offset, *l2_table, cluster_offset;
746 int l1_bits, c;
747 int index_in_cluster, nb_available, nb_needed, nb_clusters;
095a9c58
AL
748
749 index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1);
750 nb_needed = *num + index_in_cluster;
751
752 l1_bits = s->l2_bits + s->cluster_bits;
753
754 /* compute how many bytes there are between the offset and
6db6c638 755 * the end of the l1 entry
095a9c58
AL
756 */
757
758 nb_available = (1 << l1_bits) - (offset & ((1 << l1_bits) - 1));
759
760 /* compute the number of available sectors */
761
762 nb_available = (nb_available >> 9) + index_in_cluster;
763
f8de1660
AL
764 if (nb_needed > nb_available) {
765 nb_needed = nb_available;
766 }
767
095a9c58 768 cluster_offset = 0;
05203524
AL
769
770 /* seek the the l2 offset in the l1 table */
771
095a9c58 772 l1_index = offset >> l1_bits;
05203524 773 if (l1_index >= s->l1_size)
095a9c58 774 goto out;
05203524
AL
775
776 l2_offset = s->l1_table[l1_index];
777
778 /* seek the l2 table of the given l2 offset */
779
780 if (!l2_offset)
095a9c58 781 goto out;
05203524
AL
782
783 /* load the l2 table in memory */
784
785 l2_offset &= ~QCOW_OFLAG_COPIED;
786 l2_table = l2_load(bs, l2_offset);
787 if (l2_table == NULL)
768706a5 788 return 0;
05203524
AL
789
790 /* find the cluster offset for the given disk offset */
791
792 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
793 cluster_offset = be64_to_cpu(l2_table[l2_index]);
6db6c638 794 nb_clusters = size_to_clusters(s, nb_needed << 9);
095a9c58
AL
795
796 if (!cluster_offset) {
6db6c638
AL
797 /* how many empty clusters ? */
798 c = count_contiguous_free_clusters(nb_clusters, &l2_table[l2_index]);
095a9c58 799 } else {
6db6c638
AL
800 /* how many allocated clusters ? */
801 c = count_contiguous_clusters(nb_clusters, s->cluster_size,
ff4b91c2 802 &l2_table[l2_index], 0, QCOW_OFLAG_COPIED);
6db6c638 803 }
095a9c58 804
6db6c638 805 nb_available = (c * s->cluster_sectors);
095a9c58
AL
806out:
807 if (nb_available > nb_needed)
808 nb_available = nb_needed;
809
810 *num = nb_available - index_in_cluster;
811
6db6c638 812 return cluster_offset & ~QCOW_OFLAG_COPIED;
05203524
AL
813}
814
815/*
52d893ec 816 * free_any_clusters
05203524 817 *
52d893ec 818 * free clusters according to its type: compressed or not
05203524 819 *
52d893ec
AL
820 */
821
822static void free_any_clusters(BlockDriverState *bs,
095a9c58 823 uint64_t cluster_offset, int nb_clusters)
52d893ec
AL
824{
825 BDRVQcowState *s = bs->opaque;
826
52d893ec
AL
827 /* free the cluster */
828
829 if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
830 int nb_csectors;
831 nb_csectors = ((cluster_offset >> s->csize_shift) &
832 s->csize_mask) + 1;
833 free_clusters(bs, (cluster_offset & s->cluster_offset_mask) & ~511,
834 nb_csectors * 512);
835 return;
836 }
837
095a9c58
AL
838 free_clusters(bs, cluster_offset, nb_clusters << s->cluster_bits);
839
840 return;
52d893ec
AL
841}
842
843/*
844 * get_cluster_table
05203524 845 *
52d893ec
AL
846 * for a given disk offset, load (and allocate if needed)
847 * the l2 table.
848 *
849 * the l2 table offset in the qcow2 file and the cluster index
850 * in the l2 table are given to the caller.
05203524
AL
851 *
852 */
853
52d893ec
AL
854static int get_cluster_table(BlockDriverState *bs, uint64_t offset,
855 uint64_t **new_l2_table,
856 uint64_t *new_l2_offset,
857 int *new_l2_index)
585f8587
FB
858{
859 BDRVQcowState *s = bs->opaque;
108534b9 860 int l1_index, l2_index, ret;
52d893ec 861 uint64_t l2_offset, *l2_table;
108534b9
AL
862
863 /* seek the the l2 offset in the l1 table */
3b46e624 864
585f8587
FB
865 l1_index = offset >> (s->l2_bits + s->cluster_bits);
866 if (l1_index >= s->l1_size) {
108534b9
AL
867 ret = grow_l1_table(bs, l1_index + 1);
868 if (ret < 0)
585f8587
FB
869 return 0;
870 }
871 l2_offset = s->l1_table[l1_index];
108534b9
AL
872
873 /* seek the l2 table of the given l2 offset */
874
05203524
AL
875 if (l2_offset & QCOW_OFLAG_COPIED) {
876 /* load the l2 table in memory */
877 l2_offset &= ~QCOW_OFLAG_COPIED;
878 l2_table = l2_load(bs, l2_offset);
879 if (l2_table == NULL)
585f8587 880 return 0;
05203524
AL
881 } else {
882 if (l2_offset)
883 free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t));
108534b9
AL
884 l2_table = l2_allocate(bs, l1_index);
885 if (l2_table == NULL)
585f8587 886 return 0;
108534b9 887 l2_offset = s->l1_table[l1_index] & ~QCOW_OFLAG_COPIED;
585f8587 888 }
108534b9
AL
889
890 /* find the cluster offset for the given disk offset */
891
585f8587 892 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
05203524 893
52d893ec
AL
894 *new_l2_table = l2_table;
895 *new_l2_offset = l2_offset;
896 *new_l2_index = l2_index;
897
898 return 1;
899}
900
901/*
902 * alloc_compressed_cluster_offset
903 *
904 * For a given offset of the disk image, return cluster offset in
905 * qcow2 file.
906 *
907 * If the offset is not found, allocate a new compressed cluster.
908 *
909 * Return the cluster offset if successful,
910 * Return 0, otherwise.
911 *
912 */
913
914static uint64_t alloc_compressed_cluster_offset(BlockDriverState *bs,
915 uint64_t offset,
916 int compressed_size)
917{
918 BDRVQcowState *s = bs->opaque;
919 int l2_index, ret;
920 uint64_t l2_offset, *l2_table, cluster_offset;
921 int nb_csectors;
922
923 ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
924 if (ret == 0)
925 return 0;
926
927 cluster_offset = be64_to_cpu(l2_table[l2_index]);
05203524
AL
928 if (cluster_offset & QCOW_OFLAG_COPIED)
929 return cluster_offset & ~QCOW_OFLAG_COPIED;
930
095a9c58
AL
931 if (cluster_offset)
932 free_any_clusters(bs, cluster_offset, 1);
108534b9 933
52d893ec
AL
934 cluster_offset = alloc_bytes(bs, compressed_size);
935 nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) -
936 (cluster_offset >> 9);
05203524 937
52d893ec
AL
938 cluster_offset |= QCOW_OFLAG_COMPRESSED |
939 ((uint64_t)nb_csectors << s->csize_shift);
05203524 940
52d893ec 941 /* update L2 table */
05203524 942
52d893ec 943 /* compressed clusters never have the copied flag */
05203524 944
52d893ec
AL
945 l2_table[l2_index] = cpu_to_be64(cluster_offset);
946 if (bdrv_pwrite(s->hd,
947 l2_offset + l2_index * sizeof(uint64_t),
948 l2_table + l2_index,
949 sizeof(uint64_t)) != sizeof(uint64_t))
950 return 0;
05203524 951
52d893ec
AL
952 return cluster_offset;
953}
05203524 954
e976c6a1
AL
955typedef struct QCowL2Meta
956{
957 uint64_t offset;
958 int n_start;
959 int nb_available;
960 int nb_clusters;
961} QCowL2Meta;
962
963static int alloc_cluster_link_l2(BlockDriverState *bs, uint64_t cluster_offset,
964 QCowL2Meta *m)
965{
966 BDRVQcowState *s = bs->opaque;
967 int i, j = 0, l2_index, ret;
968 uint64_t *old_cluster, start_sect, l2_offset, *l2_table;
969
970 if (m->nb_clusters == 0)
971 return 0;
972
3ec88e80 973 old_cluster = qemu_malloc(m->nb_clusters * sizeof(uint64_t));
e976c6a1
AL
974
975 /* copy content of unmodified sectors */
976 start_sect = (m->offset & ~(s->cluster_size - 1)) >> 9;
977 if (m->n_start) {
978 ret = copy_sectors(bs, start_sect, cluster_offset, 0, m->n_start);
979 if (ret < 0)
980 goto err;
981 }
982
983 if (m->nb_available & (s->cluster_sectors - 1)) {
984 uint64_t end = m->nb_available & ~(uint64_t)(s->cluster_sectors - 1);
985 ret = copy_sectors(bs, start_sect + end, cluster_offset + (end << 9),
986 m->nb_available - end, s->cluster_sectors);
987 if (ret < 0)
988 goto err;
989 }
990
991 ret = -EIO;
992 /* update L2 table */
993 if (!get_cluster_table(bs, m->offset, &l2_table, &l2_offset, &l2_index))
994 goto err;
995
996 for (i = 0; i < m->nb_clusters; i++) {
997 if(l2_table[l2_index + i] != 0)
998 old_cluster[j++] = l2_table[l2_index + i];
999
1000 l2_table[l2_index + i] = cpu_to_be64((cluster_offset +
1001 (i << s->cluster_bits)) | QCOW_OFLAG_COPIED);
1002 }
1003
1004 if (bdrv_pwrite(s->hd, l2_offset + l2_index * sizeof(uint64_t),
1005 l2_table + l2_index, m->nb_clusters * sizeof(uint64_t)) !=
1006 m->nb_clusters * sizeof(uint64_t))
1007 goto err;
1008
1009 for (i = 0; i < j; i++)
1010 free_any_clusters(bs, old_cluster[i], 1);
1011
1012 ret = 0;
1013err:
1014 qemu_free(old_cluster);
1015 return ret;
1016 }
1017
52d893ec
AL
1018/*
1019 * alloc_cluster_offset
1020 *
1021 * For a given offset of the disk image, return cluster offset in
1022 * qcow2 file.
1023 *
1024 * If the offset is not found, allocate a new cluster.
1025 *
1026 * Return the cluster offset if successful,
1027 * Return 0, otherwise.
1028 *
1029 */
1030
1031static uint64_t alloc_cluster_offset(BlockDriverState *bs,
1032 uint64_t offset,
095a9c58 1033 int n_start, int n_end,
e976c6a1 1034 int *num, QCowL2Meta *m)
52d893ec
AL
1035{
1036 BDRVQcowState *s = bs->opaque;
1037 int l2_index, ret;
1038 uint64_t l2_offset, *l2_table, cluster_offset;
e976c6a1 1039 int nb_clusters, i = 0;
52d893ec
AL
1040
1041 ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
1042 if (ret == 0)
1043 return 0;
1044
6db6c638
AL
1045 nb_clusters = size_to_clusters(s, n_end << 9);
1046
e976c6a1 1047 nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
095a9c58 1048
52d893ec 1049 cluster_offset = be64_to_cpu(l2_table[l2_index]);
52d893ec 1050
095a9c58
AL
1051 /* We keep all QCOW_OFLAG_COPIED clusters */
1052
1053 if (cluster_offset & QCOW_OFLAG_COPIED) {
6db6c638 1054 nb_clusters = count_contiguous_clusters(nb_clusters, s->cluster_size,
ff4b91c2 1055 &l2_table[l2_index], 0, 0);
095a9c58 1056
095a9c58 1057 cluster_offset &= ~QCOW_OFLAG_COPIED;
e976c6a1 1058 m->nb_clusters = 0;
095a9c58
AL
1059
1060 goto out;
1061 }
1062
1063 /* for the moment, multiple compressed clusters are not managed */
1064
1065 if (cluster_offset & QCOW_OFLAG_COMPRESSED)
1066 nb_clusters = 1;
1067
bc352085 1068 /* how many available clusters ? */
095a9c58 1069
bc352085 1070 while (i < nb_clusters) {
ab5ccbd6 1071 i += count_contiguous_clusters(nb_clusters - i, s->cluster_size,
ff4b91c2 1072 &l2_table[l2_index], i, 0);
ab5ccbd6
AL
1073
1074 if(be64_to_cpu(l2_table[l2_index + i]))
1075 break;
1076
6db6c638
AL
1077 i += count_contiguous_free_clusters(nb_clusters - i,
1078 &l2_table[l2_index + i]);
095a9c58 1079
6db6c638 1080 cluster_offset = be64_to_cpu(l2_table[l2_index + i]);
095a9c58 1081
6db6c638 1082 if ((cluster_offset & QCOW_OFLAG_COPIED) ||
bc352085 1083 (cluster_offset & QCOW_OFLAG_COMPRESSED))
6db6c638 1084 break;
095a9c58 1085 }
bc352085 1086 nb_clusters = i;
05203524
AL
1087
1088 /* allocate a new cluster */
1089
095a9c58 1090 cluster_offset = alloc_clusters(bs, nb_clusters * s->cluster_size);
05203524 1091
e976c6a1
AL
1092 /* save info needed for meta data update */
1093 m->offset = offset;
1094 m->n_start = n_start;
1095 m->nb_clusters = nb_clusters;
05203524 1096
095a9c58 1097out:
e976c6a1
AL
1098 m->nb_available = MIN(nb_clusters << (s->cluster_bits - 9), n_end);
1099
1100 *num = m->nb_available - n_start;
095a9c58 1101
585f8587
FB
1102 return cluster_offset;
1103}
1104
5fafdf24 1105static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
585f8587
FB
1106 int nb_sectors, int *pnum)
1107{
585f8587
FB
1108 uint64_t cluster_offset;
1109
095a9c58
AL
1110 *pnum = nb_sectors;
1111 cluster_offset = get_cluster_offset(bs, sector_num << 9, pnum);
1112
585f8587
FB
1113 return (cluster_offset != 0);
1114}
1115
1116static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
1117 const uint8_t *buf, int buf_size)
1118{
1119 z_stream strm1, *strm = &strm1;
1120 int ret, out_len;
1121
1122 memset(strm, 0, sizeof(*strm));
1123
1124 strm->next_in = (uint8_t *)buf;
1125 strm->avail_in = buf_size;
1126 strm->next_out = out_buf;
1127 strm->avail_out = out_buf_size;
1128
1129 ret = inflateInit2(strm, -12);
1130 if (ret != Z_OK)
1131 return -1;
1132 ret = inflate(strm, Z_FINISH);
1133 out_len = strm->next_out - out_buf;
1134 if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
1135 out_len != out_buf_size) {
1136 inflateEnd(strm);
1137 return -1;
1138 }
1139 inflateEnd(strm);
1140 return 0;
1141}
3b46e624 1142
585f8587
FB
1143static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset)
1144{
1145 int ret, csize, nb_csectors, sector_offset;
1146 uint64_t coffset;
1147
1148 coffset = cluster_offset & s->cluster_offset_mask;
1149 if (s->cluster_cache_offset != coffset) {
1150 nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
1151 sector_offset = coffset & 511;
1152 csize = nb_csectors * 512 - sector_offset;
1153 ret = bdrv_read(s->hd, coffset >> 9, s->cluster_data, nb_csectors);
1154 if (ret < 0) {
1155 return -1;
1156 }
1157 if (decompress_buffer(s->cluster_cache, s->cluster_size,
1158 s->cluster_data + sector_offset, csize) < 0) {
1159 return -1;
1160 }
1161 s->cluster_cache_offset = coffset;
1162 }
1163 return 0;
1164}
1165
a9465922 1166/* handle reading after the end of the backing file */
5fafdf24 1167static int backing_read1(BlockDriverState *bs,
a9465922
FB
1168 int64_t sector_num, uint8_t *buf, int nb_sectors)
1169{
1170 int n1;
1171 if ((sector_num + nb_sectors) <= bs->total_sectors)
1172 return nb_sectors;
1173 if (sector_num >= bs->total_sectors)
1174 n1 = 0;
1175 else
1176 n1 = bs->total_sectors - sector_num;
1177 memset(buf + n1 * 512, 0, 512 * (nb_sectors - n1));
1178 return n1;
1179}
1180
5fafdf24 1181static int qcow_read(BlockDriverState *bs, int64_t sector_num,
585f8587
FB
1182 uint8_t *buf, int nb_sectors)
1183{
1184 BDRVQcowState *s = bs->opaque;
a9465922 1185 int ret, index_in_cluster, n, n1;
585f8587 1186 uint64_t cluster_offset;
3b46e624 1187
585f8587 1188 while (nb_sectors > 0) {
095a9c58
AL
1189 n = nb_sectors;
1190 cluster_offset = get_cluster_offset(bs, sector_num << 9, &n);
585f8587 1191 index_in_cluster = sector_num & (s->cluster_sectors - 1);
585f8587
FB
1192 if (!cluster_offset) {
1193 if (bs->backing_hd) {
1194 /* read from the base image */
a9465922
FB
1195 n1 = backing_read1(bs->backing_hd, sector_num, buf, n);
1196 if (n1 > 0) {
1197 ret = bdrv_read(bs->backing_hd, sector_num, buf, n1);
1198 if (ret < 0)
1199 return -1;
1200 }
585f8587
FB
1201 } else {
1202 memset(buf, 0, 512 * n);
1203 }
1204 } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
1205 if (decompress_cluster(s, cluster_offset) < 0)
1206 return -1;
1207 memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n);
1208 } else {
1209 ret = bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
5fafdf24 1210 if (ret != n * 512)
585f8587
FB
1211 return -1;
1212 if (s->crypt_method) {
5fafdf24 1213 encrypt_sectors(s, sector_num, buf, buf, n, 0,
585f8587
FB
1214 &s->aes_decrypt_key);
1215 }
1216 }
1217 nb_sectors -= n;
1218 sector_num += n;
1219 buf += n * 512;
1220 }
1221 return 0;
1222}
1223
5fafdf24 1224static int qcow_write(BlockDriverState *bs, int64_t sector_num,
585f8587
FB
1225 const uint8_t *buf, int nb_sectors)
1226{
1227 BDRVQcowState *s = bs->opaque;
1228 int ret, index_in_cluster, n;
1229 uint64_t cluster_offset;
095a9c58 1230 int n_end;
e976c6a1 1231 QCowL2Meta l2meta;
3b46e624 1232
585f8587
FB
1233 while (nb_sectors > 0) {
1234 index_in_cluster = sector_num & (s->cluster_sectors - 1);
095a9c58
AL
1235 n_end = index_in_cluster + nb_sectors;
1236 if (s->crypt_method &&
1237 n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
1238 n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
52d893ec 1239 cluster_offset = alloc_cluster_offset(bs, sector_num << 9,
05203524 1240 index_in_cluster,
e976c6a1 1241 n_end, &n, &l2meta);
585f8587
FB
1242 if (!cluster_offset)
1243 return -1;
1244 if (s->crypt_method) {
1245 encrypt_sectors(s, sector_num, s->cluster_data, buf, n, 1,
1246 &s->aes_encrypt_key);
5fafdf24 1247 ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512,
585f8587
FB
1248 s->cluster_data, n * 512);
1249 } else {
1250 ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
1251 }
e976c6a1
AL
1252 if (ret != n * 512 || alloc_cluster_link_l2(bs, cluster_offset, &l2meta) < 0) {
1253 free_any_clusters(bs, cluster_offset, l2meta.nb_clusters);
585f8587 1254 return -1;
e976c6a1 1255 }
585f8587
FB
1256 nb_sectors -= n;
1257 sector_num += n;
1258 buf += n * 512;
1259 }
1260 s->cluster_cache_offset = -1; /* disable compressed cache */
1261 return 0;
1262}
1263
ce1a14dc
PB
1264typedef struct QCowAIOCB {
1265 BlockDriverAIOCB common;
585f8587
FB
1266 int64_t sector_num;
1267 uint8_t *buf;
1268 int nb_sectors;
1269 int n;
1270 uint64_t cluster_offset;
5fafdf24 1271 uint8_t *cluster_data;
585f8587 1272 BlockDriverAIOCB *hd_aiocb;
c87c0672
AL
1273 struct iovec hd_iov;
1274 QEMUIOVector hd_qiov;
1490791f 1275 QEMUBH *bh;
e976c6a1 1276 QCowL2Meta l2meta;
585f8587
FB
1277} QCowAIOCB;
1278
1490791f
AL
1279static void qcow_aio_read_cb(void *opaque, int ret);
1280static void qcow_aio_read_bh(void *opaque)
1281{
1282 QCowAIOCB *acb = opaque;
1283 qemu_bh_delete(acb->bh);
1284 acb->bh = NULL;
1285 qcow_aio_read_cb(opaque, 0);
1286}
1287
a32ef786
AL
1288static int qcow_schedule_bh(QEMUBHFunc *cb, QCowAIOCB *acb)
1289{
1290 if (acb->bh)
1291 return -EIO;
1292
1293 acb->bh = qemu_bh_new(cb, acb);
1294 if (!acb->bh)
1295 return -EIO;
1296
1297 qemu_bh_schedule(acb->bh);
1298
1299 return 0;
1300}
1301
585f8587
FB
1302static void qcow_aio_read_cb(void *opaque, int ret)
1303{
ce1a14dc
PB
1304 QCowAIOCB *acb = opaque;
1305 BlockDriverState *bs = acb->common.bs;
585f8587 1306 BDRVQcowState *s = bs->opaque;
a9465922 1307 int index_in_cluster, n1;
585f8587 1308
ce1a14dc 1309 acb->hd_aiocb = NULL;
585f8587 1310 if (ret < 0) {
ac674887 1311fail:
ce1a14dc
PB
1312 acb->common.cb(acb->common.opaque, ret);
1313 qemu_aio_release(acb);
585f8587
FB
1314 return;
1315 }
1316
585f8587 1317 /* post process the read buffer */
ce1a14dc 1318 if (!acb->cluster_offset) {
585f8587 1319 /* nothing to do */
ce1a14dc 1320 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
585f8587
FB
1321 /* nothing to do */
1322 } else {
1323 if (s->crypt_method) {
5fafdf24
TS
1324 encrypt_sectors(s, acb->sector_num, acb->buf, acb->buf,
1325 acb->n, 0,
585f8587
FB
1326 &s->aes_decrypt_key);
1327 }
1328 }
1329
ce1a14dc
PB
1330 acb->nb_sectors -= acb->n;
1331 acb->sector_num += acb->n;
1332 acb->buf += acb->n * 512;
585f8587 1333
ce1a14dc 1334 if (acb->nb_sectors == 0) {
585f8587 1335 /* request completed */
ce1a14dc
PB
1336 acb->common.cb(acb->common.opaque, 0);
1337 qemu_aio_release(acb);
585f8587
FB
1338 return;
1339 }
3b46e624 1340
585f8587 1341 /* prepare next AIO request */
095a9c58
AL
1342 acb->n = acb->nb_sectors;
1343 acb->cluster_offset = get_cluster_offset(bs, acb->sector_num << 9, &acb->n);
ce1a14dc 1344 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
ce1a14dc
PB
1345
1346 if (!acb->cluster_offset) {
585f8587
FB
1347 if (bs->backing_hd) {
1348 /* read from the base image */
5fafdf24 1349 n1 = backing_read1(bs->backing_hd, acb->sector_num,
ce1a14dc 1350 acb->buf, acb->n);
a9465922 1351 if (n1 > 0) {
c87c0672
AL
1352 acb->hd_iov.iov_base = acb->buf;
1353 acb->hd_iov.iov_len = acb->n * 512;
1354 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
1355 acb->hd_aiocb = bdrv_aio_readv(bs->backing_hd, acb->sector_num,
1356 &acb->hd_qiov, acb->n,
1357 qcow_aio_read_cb, acb);
ce1a14dc 1358 if (acb->hd_aiocb == NULL)
a9465922
FB
1359 goto fail;
1360 } else {
a32ef786
AL
1361 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
1362 if (ret < 0)
ac674887 1363 goto fail;
a9465922 1364 }
585f8587
FB
1365 } else {
1366 /* Note: in this case, no need to wait */
ce1a14dc 1367 memset(acb->buf, 0, 512 * acb->n);
a32ef786
AL
1368 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
1369 if (ret < 0)
ac674887 1370 goto fail;
585f8587 1371 }
ce1a14dc 1372 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
585f8587 1373 /* add AIO support for compressed blocks ? */
ce1a14dc 1374 if (decompress_cluster(s, acb->cluster_offset) < 0)
585f8587 1375 goto fail;
5fafdf24 1376 memcpy(acb->buf,
ce1a14dc 1377 s->cluster_cache + index_in_cluster * 512, 512 * acb->n);
a32ef786
AL
1378 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
1379 if (ret < 0)
ac674887 1380 goto fail;
585f8587 1381 } else {
ce1a14dc 1382 if ((acb->cluster_offset & 511) != 0) {
585f8587
FB
1383 ret = -EIO;
1384 goto fail;
1385 }
c87c0672
AL
1386
1387 acb->hd_iov.iov_base = acb->buf;
1388 acb->hd_iov.iov_len = acb->n * 512;
1389 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
1390 acb->hd_aiocb = bdrv_aio_readv(s->hd,
5fafdf24 1391 (acb->cluster_offset >> 9) + index_in_cluster,
c87c0672 1392 &acb->hd_qiov, acb->n, qcow_aio_read_cb, acb);
ce1a14dc 1393 if (acb->hd_aiocb == NULL)
585f8587
FB
1394 goto fail;
1395 }
1396}
1397
ce1a14dc
PB
1398static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs,
1399 int64_t sector_num, uint8_t *buf, int nb_sectors,
1400 BlockDriverCompletionFunc *cb, void *opaque)
585f8587 1401{
ce1a14dc
PB
1402 QCowAIOCB *acb;
1403
1404 acb = qemu_aio_get(bs, cb, opaque);
1405 if (!acb)
1406 return NULL;
1407 acb->hd_aiocb = NULL;
1408 acb->sector_num = sector_num;
1409 acb->buf = buf;
1410 acb->nb_sectors = nb_sectors;
1411 acb->n = 0;
1412 acb->cluster_offset = 0;
e976c6a1 1413 acb->l2meta.nb_clusters = 0;
ce1a14dc
PB
1414 return acb;
1415}
1416
1417static BlockDriverAIOCB *qcow_aio_read(BlockDriverState *bs,
1418 int64_t sector_num, uint8_t *buf, int nb_sectors,
1419 BlockDriverCompletionFunc *cb, void *opaque)
1420{
1421 QCowAIOCB *acb;
1422
1423 acb = qcow_aio_setup(bs, sector_num, buf, nb_sectors, cb, opaque);
1424 if (!acb)
1425 return NULL;
585f8587
FB
1426
1427 qcow_aio_read_cb(acb, 0);
ce1a14dc 1428 return &acb->common;
585f8587
FB
1429}
1430
1431static void qcow_aio_write_cb(void *opaque, int ret)
1432{
ce1a14dc
PB
1433 QCowAIOCB *acb = opaque;
1434 BlockDriverState *bs = acb->common.bs;
585f8587 1435 BDRVQcowState *s = bs->opaque;
585f8587 1436 int index_in_cluster;
585f8587 1437 const uint8_t *src_buf;
095a9c58 1438 int n_end;
ce1a14dc
PB
1439
1440 acb->hd_aiocb = NULL;
1441
585f8587
FB
1442 if (ret < 0) {
1443 fail:
ce1a14dc
PB
1444 acb->common.cb(acb->common.opaque, ret);
1445 qemu_aio_release(acb);
585f8587
FB
1446 return;
1447 }
1448
e976c6a1
AL
1449 if (alloc_cluster_link_l2(bs, acb->cluster_offset, &acb->l2meta) < 0) {
1450 free_any_clusters(bs, acb->cluster_offset, acb->l2meta.nb_clusters);
1451 goto fail;
1452 }
1453
ce1a14dc
PB
1454 acb->nb_sectors -= acb->n;
1455 acb->sector_num += acb->n;
1456 acb->buf += acb->n * 512;
585f8587 1457
ce1a14dc 1458 if (acb->nb_sectors == 0) {
585f8587 1459 /* request completed */
ce1a14dc
PB
1460 acb->common.cb(acb->common.opaque, 0);
1461 qemu_aio_release(acb);
585f8587
FB
1462 return;
1463 }
3b46e624 1464
ce1a14dc 1465 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
095a9c58
AL
1466 n_end = index_in_cluster + acb->nb_sectors;
1467 if (s->crypt_method &&
1468 n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
1469 n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
1470
e976c6a1 1471 acb->cluster_offset = alloc_cluster_offset(bs, acb->sector_num << 9,
05203524 1472 index_in_cluster,
e976c6a1
AL
1473 n_end, &acb->n, &acb->l2meta);
1474 if (!acb->cluster_offset || (acb->cluster_offset & 511) != 0) {
585f8587
FB
1475 ret = -EIO;
1476 goto fail;
1477 }
1478 if (s->crypt_method) {
ce1a14dc 1479 if (!acb->cluster_data) {
095a9c58
AL
1480 acb->cluster_data = qemu_mallocz(QCOW_MAX_CRYPT_CLUSTERS *
1481 s->cluster_size);
585f8587 1482 }
5fafdf24 1483 encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf,
ce1a14dc
PB
1484 acb->n, 1, &s->aes_encrypt_key);
1485 src_buf = acb->cluster_data;
585f8587 1486 } else {
ce1a14dc 1487 src_buf = acb->buf;
585f8587 1488 }
c87c0672
AL
1489 acb->hd_iov.iov_base = (void *)src_buf;
1490 acb->hd_iov.iov_len = acb->n * 512;
1491 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
1492 acb->hd_aiocb = bdrv_aio_writev(s->hd,
1493 (acb->cluster_offset >> 9) + index_in_cluster,
1494 &acb->hd_qiov, acb->n,
1495 qcow_aio_write_cb, acb);
ce1a14dc 1496 if (acb->hd_aiocb == NULL)
585f8587
FB
1497 goto fail;
1498}
1499
ce1a14dc
PB
1500static BlockDriverAIOCB *qcow_aio_write(BlockDriverState *bs,
1501 int64_t sector_num, const uint8_t *buf, int nb_sectors,
1502 BlockDriverCompletionFunc *cb, void *opaque)
585f8587 1503{
585f8587 1504 BDRVQcowState *s = bs->opaque;
ce1a14dc 1505 QCowAIOCB *acb;
3b46e624 1506
585f8587
FB
1507 s->cluster_cache_offset = -1; /* disable compressed cache */
1508
ce1a14dc
PB
1509 acb = qcow_aio_setup(bs, sector_num, (uint8_t*)buf, nb_sectors, cb, opaque);
1510 if (!acb)
1511 return NULL;
3b46e624 1512
585f8587 1513 qcow_aio_write_cb(acb, 0);
ce1a14dc 1514 return &acb->common;
585f8587
FB
1515}
1516
ce1a14dc 1517static void qcow_aio_cancel(BlockDriverAIOCB *blockacb)
585f8587 1518{
ce1a14dc
PB
1519 QCowAIOCB *acb = (QCowAIOCB *)blockacb;
1520 if (acb->hd_aiocb)
1521 bdrv_aio_cancel(acb->hd_aiocb);
1522 qemu_aio_release(acb);
585f8587
FB
1523}
1524
1525static void qcow_close(BlockDriverState *bs)
1526{
1527 BDRVQcowState *s = bs->opaque;
1528 qemu_free(s->l1_table);
1529 qemu_free(s->l2_cache);
1530 qemu_free(s->cluster_cache);
1531 qemu_free(s->cluster_data);
1532 refcount_close(bs);
1533 bdrv_delete(s->hd);
1534}
1535
1536/* XXX: use std qcow open function ? */
1537typedef struct QCowCreateState {
1538 int cluster_size;
1539 int cluster_bits;
1540 uint16_t *refcount_block;
1541 uint64_t *refcount_table;
1542 int64_t l1_table_offset;
1543 int64_t refcount_table_offset;
1544 int64_t refcount_block_offset;
1545} QCowCreateState;
1546
1547static void create_refcount_update(QCowCreateState *s,
1548 int64_t offset, int64_t size)
1549{
1550 int refcount;
1551 int64_t start, last, cluster_offset;
1552 uint16_t *p;
1553
1554 start = offset & ~(s->cluster_size - 1);
1555 last = (offset + size - 1) & ~(s->cluster_size - 1);
5fafdf24 1556 for(cluster_offset = start; cluster_offset <= last;
585f8587
FB
1557 cluster_offset += s->cluster_size) {
1558 p = &s->refcount_block[cluster_offset >> s->cluster_bits];
1559 refcount = be16_to_cpu(*p);
1560 refcount++;
1561 *p = cpu_to_be16(refcount);
1562 }
1563}
1564
f965509c
AL
1565static int qcow_create2(const char *filename, int64_t total_size,
1566 const char *backing_file, const char *backing_format,
1567 int flags)
585f8587 1568{
f965509c 1569
585f8587 1570 int fd, header_size, backing_filename_len, l1_size, i, shift, l2_bits;
2d2431f0 1571 int ref_clusters, backing_format_len = 0;
585f8587
FB
1572 QCowHeader header;
1573 uint64_t tmp, offset;
1574 QCowCreateState s1, *s = &s1;
f965509c
AL
1575 QCowExtension ext_bf = {0, 0};
1576
3b46e624 1577
585f8587
FB
1578 memset(s, 0, sizeof(*s));
1579
1580 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
1581 if (fd < 0)
1582 return -1;
1583 memset(&header, 0, sizeof(header));
1584 header.magic = cpu_to_be32(QCOW_MAGIC);
1585 header.version = cpu_to_be32(QCOW_VERSION);
1586 header.size = cpu_to_be64(total_size * 512);
1587 header_size = sizeof(header);
1588 backing_filename_len = 0;
1589 if (backing_file) {
f965509c
AL
1590 if (backing_format) {
1591 ext_bf.magic = QCOW_EXT_MAGIC_BACKING_FORMAT;
1592 backing_format_len = strlen(backing_format);
1593 ext_bf.len = (backing_format_len + 7) & ~7;
1594 header_size += ((sizeof(ext_bf) + ext_bf.len + 7) & ~7);
1595 }
585f8587
FB
1596 header.backing_file_offset = cpu_to_be64(header_size);
1597 backing_filename_len = strlen(backing_file);
1598 header.backing_file_size = cpu_to_be32(backing_filename_len);
1599 header_size += backing_filename_len;
1600 }
1601 s->cluster_bits = 12; /* 4 KB clusters */
1602 s->cluster_size = 1 << s->cluster_bits;
1603 header.cluster_bits = cpu_to_be32(s->cluster_bits);
1604 header_size = (header_size + 7) & ~7;
ec36ba14 1605 if (flags & BLOCK_FLAG_ENCRYPT) {
585f8587
FB
1606 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
1607 } else {
1608 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
1609 }
1610 l2_bits = s->cluster_bits - 3;
1611 shift = s->cluster_bits + l2_bits;
1612 l1_size = (((total_size * 512) + (1LL << shift) - 1) >> shift);
1613 offset = align_offset(header_size, s->cluster_size);
1614 s->l1_table_offset = offset;
1615 header.l1_table_offset = cpu_to_be64(s->l1_table_offset);
1616 header.l1_size = cpu_to_be32(l1_size);
15e6690a 1617 offset += align_offset(l1_size * sizeof(uint64_t), s->cluster_size);
585f8587
FB
1618
1619 s->refcount_table = qemu_mallocz(s->cluster_size);
3b46e624 1620
585f8587
FB
1621 s->refcount_table_offset = offset;
1622 header.refcount_table_offset = cpu_to_be64(offset);
1623 header.refcount_table_clusters = cpu_to_be32(1);
1624 offset += s->cluster_size;
585f8587 1625 s->refcount_block_offset = offset;
2d2431f0
AL
1626
1627 /* count how many refcount blocks needed */
1628 tmp = offset >> s->cluster_bits;
1629 ref_clusters = (tmp >> (s->cluster_bits - REFCOUNT_SHIFT)) + 1;
1630 for (i=0; i < ref_clusters; i++) {
1631 s->refcount_table[i] = cpu_to_be64(offset);
1632 offset += s->cluster_size;
1633 }
1634
1635 s->refcount_block = qemu_mallocz(ref_clusters * s->cluster_size);
585f8587
FB
1636
1637 /* update refcounts */
1638 create_refcount_update(s, 0, header_size);
15e6690a 1639 create_refcount_update(s, s->l1_table_offset, l1_size * sizeof(uint64_t));
585f8587 1640 create_refcount_update(s, s->refcount_table_offset, s->cluster_size);
2d2431f0 1641 create_refcount_update(s, s->refcount_block_offset, ref_clusters * s->cluster_size);
3b46e624 1642
585f8587
FB
1643 /* write all the data */
1644 write(fd, &header, sizeof(header));
1645 if (backing_file) {
f965509c
AL
1646 if (backing_format_len) {
1647 char zero[16];
1648 int d = ext_bf.len - backing_format_len;
1649
1650 memset(zero, 0, sizeof(zero));
1651 cpu_to_be32s(&ext_bf.magic);
1652 cpu_to_be32s(&ext_bf.len);
1653 write(fd, &ext_bf, sizeof(ext_bf));
1654 write(fd, backing_format, backing_format_len);
1655 if (d>0) {
1656 write(fd, zero, d);
1657 }
1658 }
585f8587
FB
1659 write(fd, backing_file, backing_filename_len);
1660 }
1661 lseek(fd, s->l1_table_offset, SEEK_SET);
1662 tmp = 0;
1663 for(i = 0;i < l1_size; i++) {
1664 write(fd, &tmp, sizeof(tmp));
1665 }
1666 lseek(fd, s->refcount_table_offset, SEEK_SET);
1667 write(fd, s->refcount_table, s->cluster_size);
3b46e624 1668
585f8587 1669 lseek(fd, s->refcount_block_offset, SEEK_SET);
2d2431f0 1670 write(fd, s->refcount_block, ref_clusters * s->cluster_size);
585f8587
FB
1671
1672 qemu_free(s->refcount_table);
1673 qemu_free(s->refcount_block);
1674 close(fd);
1675 return 0;
585f8587
FB
1676}
1677
f965509c
AL
1678static int qcow_create(const char *filename, int64_t total_size,
1679 const char *backing_file, int flags)
1680{
1681 return qcow_create2(filename, total_size, backing_file, NULL, flags);
1682}
1683
585f8587
FB
1684static int qcow_make_empty(BlockDriverState *bs)
1685{
1686#if 0
1687 /* XXX: not correct */
1688 BDRVQcowState *s = bs->opaque;
1689 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1690 int ret;
1691
1692 memset(s->l1_table, 0, l1_length);
1693 if (bdrv_pwrite(s->hd, s->l1_table_offset, s->l1_table, l1_length) < 0)
ac674887 1694 return -1;
585f8587
FB
1695 ret = bdrv_truncate(s->hd, s->l1_table_offset + l1_length);
1696 if (ret < 0)
1697 return ret;
3b46e624 1698
585f8587
FB
1699 l2_cache_reset(bs);
1700#endif
1701 return 0;
1702}
1703
1704/* XXX: put compressed sectors first, then all the cluster aligned
1705 tables to avoid losing bytes in alignment */
5fafdf24 1706static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
585f8587
FB
1707 const uint8_t *buf, int nb_sectors)
1708{
1709 BDRVQcowState *s = bs->opaque;
1710 z_stream strm;
1711 int ret, out_len;
1712 uint8_t *out_buf;
1713 uint64_t cluster_offset;
1714
1715 if (nb_sectors == 0) {
1716 /* align end of file to a sector boundary to ease reading with
1717 sector based I/Os */
1718 cluster_offset = bdrv_getlength(s->hd);
1719 cluster_offset = (cluster_offset + 511) & ~511;
1720 bdrv_truncate(s->hd, cluster_offset);
1721 return 0;
1722 }
1723
1724 if (nb_sectors != s->cluster_sectors)
1725 return -EINVAL;
1726
1727 out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
585f8587
FB
1728
1729 /* best compression, small window, no zlib header */
1730 memset(&strm, 0, sizeof(strm));
1731 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
5fafdf24 1732 Z_DEFLATED, -12,
585f8587
FB
1733 9, Z_DEFAULT_STRATEGY);
1734 if (ret != 0) {
1735 qemu_free(out_buf);
1736 return -1;
1737 }
1738
1739 strm.avail_in = s->cluster_size;
1740 strm.next_in = (uint8_t *)buf;
1741 strm.avail_out = s->cluster_size;
1742 strm.next_out = out_buf;
1743
1744 ret = deflate(&strm, Z_FINISH);
1745 if (ret != Z_STREAM_END && ret != Z_OK) {
1746 qemu_free(out_buf);
1747 deflateEnd(&strm);
1748 return -1;
1749 }
1750 out_len = strm.next_out - out_buf;
1751
1752 deflateEnd(&strm);
1753
1754 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1755 /* could not compress: write normal cluster */
1756 qcow_write(bs, sector_num, buf, s->cluster_sectors);
1757 } else {
52d893ec
AL
1758 cluster_offset = alloc_compressed_cluster_offset(bs, sector_num << 9,
1759 out_len);
1760 if (!cluster_offset)
1761 return -1;
585f8587
FB
1762 cluster_offset &= s->cluster_offset_mask;
1763 if (bdrv_pwrite(s->hd, cluster_offset, out_buf, out_len) != out_len) {
1764 qemu_free(out_buf);
1765 return -1;
1766 }
1767 }
3b46e624 1768
585f8587
FB
1769 qemu_free(out_buf);
1770 return 0;
1771}
1772
1773static void qcow_flush(BlockDriverState *bs)
1774{
1775 BDRVQcowState *s = bs->opaque;
1776 bdrv_flush(s->hd);
1777}
1778
1779static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1780{
1781 BDRVQcowState *s = bs->opaque;
1782 bdi->cluster_size = s->cluster_size;
5fafdf24 1783 bdi->vm_state_offset = (int64_t)s->l1_vm_state_index <<
585f8587
FB
1784 (s->cluster_bits + s->l2_bits);
1785 return 0;
1786}
1787
1788/*********************************************************/
1789/* snapshot support */
1790
1791/* update the refcounts of snapshots and the copied flag */
5fafdf24 1792static int update_snapshot_refcount(BlockDriverState *bs,
585f8587
FB
1793 int64_t l1_table_offset,
1794 int l1_size,
1795 int addend)
1796{
1797 BDRVQcowState *s = bs->opaque;
1798 uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated;
1799 int64_t old_offset, old_l2_offset;
1800 int l2_size, i, j, l1_modified, l2_modified, nb_csectors, refcount;
3b46e624 1801
585f8587
FB
1802 l2_cache_reset(bs);
1803
1804 l2_table = NULL;
1805 l1_table = NULL;
1806 l1_size2 = l1_size * sizeof(uint64_t);
1807 l1_allocated = 0;
1808 if (l1_table_offset != s->l1_table_offset) {
1809 l1_table = qemu_malloc(l1_size2);
585f8587 1810 l1_allocated = 1;
5fafdf24 1811 if (bdrv_pread(s->hd, l1_table_offset,
585f8587
FB
1812 l1_table, l1_size2) != l1_size2)
1813 goto fail;
1814 for(i = 0;i < l1_size; i++)
1815 be64_to_cpus(&l1_table[i]);
1816 } else {
1817 assert(l1_size == s->l1_size);
1818 l1_table = s->l1_table;
1819 l1_allocated = 0;
1820 }
3b46e624 1821
585f8587
FB
1822 l2_size = s->l2_size * sizeof(uint64_t);
1823 l2_table = qemu_malloc(l2_size);
585f8587
FB
1824 l1_modified = 0;
1825 for(i = 0; i < l1_size; i++) {
1826 l2_offset = l1_table[i];
1827 if (l2_offset) {
1828 old_l2_offset = l2_offset;
1829 l2_offset &= ~QCOW_OFLAG_COPIED;
1830 l2_modified = 0;
1831 if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
1832 goto fail;
1833 for(j = 0; j < s->l2_size; j++) {
1834 offset = be64_to_cpu(l2_table[j]);
1835 if (offset != 0) {
1836 old_offset = offset;
1837 offset &= ~QCOW_OFLAG_COPIED;
1838 if (offset & QCOW_OFLAG_COMPRESSED) {
5fafdf24 1839 nb_csectors = ((offset >> s->csize_shift) &
585f8587
FB
1840 s->csize_mask) + 1;
1841 if (addend != 0)
1842 update_refcount(bs, (offset & s->cluster_offset_mask) & ~511,
1843 nb_csectors * 512, addend);
1844 /* compressed clusters are never modified */
5fafdf24 1845 refcount = 2;
585f8587
FB
1846 } else {
1847 if (addend != 0) {
1848 refcount = update_cluster_refcount(bs, offset >> s->cluster_bits, addend);
1849 } else {
1850 refcount = get_refcount(bs, offset >> s->cluster_bits);
1851 }
1852 }
1853
1854 if (refcount == 1) {
1855 offset |= QCOW_OFLAG_COPIED;
1856 }
1857 if (offset != old_offset) {
1858 l2_table[j] = cpu_to_be64(offset);
1859 l2_modified = 1;
1860 }
1861 }
1862 }
1863 if (l2_modified) {
5fafdf24 1864 if (bdrv_pwrite(s->hd,
585f8587
FB
1865 l2_offset, l2_table, l2_size) != l2_size)
1866 goto fail;
1867 }
1868
1869 if (addend != 0) {
1870 refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend);
1871 } else {
1872 refcount = get_refcount(bs, l2_offset >> s->cluster_bits);
1873 }
1874 if (refcount == 1) {
1875 l2_offset |= QCOW_OFLAG_COPIED;
1876 }
1877 if (l2_offset != old_l2_offset) {
1878 l1_table[i] = l2_offset;
1879 l1_modified = 1;
1880 }
1881 }
1882 }
1883 if (l1_modified) {
1884 for(i = 0; i < l1_size; i++)
1885 cpu_to_be64s(&l1_table[i]);
5fafdf24 1886 if (bdrv_pwrite(s->hd, l1_table_offset, l1_table,
585f8587
FB
1887 l1_size2) != l1_size2)
1888 goto fail;
1889 for(i = 0; i < l1_size; i++)
1890 be64_to_cpus(&l1_table[i]);
1891 }
1892 if (l1_allocated)
1893 qemu_free(l1_table);
1894 qemu_free(l2_table);
1895 return 0;
1896 fail:
1897 if (l1_allocated)
1898 qemu_free(l1_table);
1899 qemu_free(l2_table);
1900 return -EIO;
1901}
1902
1903static void qcow_free_snapshots(BlockDriverState *bs)
1904{
1905 BDRVQcowState *s = bs->opaque;
1906 int i;
1907
1908 for(i = 0; i < s->nb_snapshots; i++) {
1909 qemu_free(s->snapshots[i].name);
1910 qemu_free(s->snapshots[i].id_str);
1911 }
1912 qemu_free(s->snapshots);
1913 s->snapshots = NULL;
1914 s->nb_snapshots = 0;
1915}
1916
1917static int qcow_read_snapshots(BlockDriverState *bs)
1918{
1919 BDRVQcowState *s = bs->opaque;
1920 QCowSnapshotHeader h;
1921 QCowSnapshot *sn;
1922 int i, id_str_size, name_size;
1923 int64_t offset;
1924 uint32_t extra_data_size;
1925
63c75dcd 1926 if (!s->nb_snapshots) {
1927 s->snapshots = NULL;
1928 s->snapshots_size = 0;
1929 return 0;
1930 }
1931
585f8587
FB
1932 offset = s->snapshots_offset;
1933 s->snapshots = qemu_mallocz(s->nb_snapshots * sizeof(QCowSnapshot));
585f8587
FB
1934 for(i = 0; i < s->nb_snapshots; i++) {
1935 offset = align_offset(offset, 8);
1936 if (bdrv_pread(s->hd, offset, &h, sizeof(h)) != sizeof(h))
1937 goto fail;
1938 offset += sizeof(h);
1939 sn = s->snapshots + i;
1940 sn->l1_table_offset = be64_to_cpu(h.l1_table_offset);
1941 sn->l1_size = be32_to_cpu(h.l1_size);
1942 sn->vm_state_size = be32_to_cpu(h.vm_state_size);
1943 sn->date_sec = be32_to_cpu(h.date_sec);
1944 sn->date_nsec = be32_to_cpu(h.date_nsec);
1945 sn->vm_clock_nsec = be64_to_cpu(h.vm_clock_nsec);
1946 extra_data_size = be32_to_cpu(h.extra_data_size);
1947
1948 id_str_size = be16_to_cpu(h.id_str_size);
1949 name_size = be16_to_cpu(h.name_size);
1950
1951 offset += extra_data_size;
1952
1953 sn->id_str = qemu_malloc(id_str_size + 1);
585f8587
FB
1954 if (bdrv_pread(s->hd, offset, sn->id_str, id_str_size) != id_str_size)
1955 goto fail;
1956 offset += id_str_size;
1957 sn->id_str[id_str_size] = '\0';
1958
1959 sn->name = qemu_malloc(name_size + 1);
585f8587
FB
1960 if (bdrv_pread(s->hd, offset, sn->name, name_size) != name_size)
1961 goto fail;
1962 offset += name_size;
1963 sn->name[name_size] = '\0';
1964 }
1965 s->snapshots_size = offset - s->snapshots_offset;
1966 return 0;
1967 fail:
1968 qcow_free_snapshots(bs);
1969 return -1;
1970}
1971
1972/* add at the end of the file a new list of snapshots */
1973static int qcow_write_snapshots(BlockDriverState *bs)
1974{
1975 BDRVQcowState *s = bs->opaque;
1976 QCowSnapshot *sn;
1977 QCowSnapshotHeader h;
1978 int i, name_size, id_str_size, snapshots_size;
1979 uint64_t data64;
1980 uint32_t data32;
1981 int64_t offset, snapshots_offset;
1982
1983 /* compute the size of the snapshots */
1984 offset = 0;
1985 for(i = 0; i < s->nb_snapshots; i++) {
1986 sn = s->snapshots + i;
1987 offset = align_offset(offset, 8);
1988 offset += sizeof(h);
1989 offset += strlen(sn->id_str);
1990 offset += strlen(sn->name);
1991 }
1992 snapshots_size = offset;
1993
1994 snapshots_offset = alloc_clusters(bs, snapshots_size);
1995 offset = snapshots_offset;
3b46e624 1996
585f8587
FB
1997 for(i = 0; i < s->nb_snapshots; i++) {
1998 sn = s->snapshots + i;
1999 memset(&h, 0, sizeof(h));
2000 h.l1_table_offset = cpu_to_be64(sn->l1_table_offset);
2001 h.l1_size = cpu_to_be32(sn->l1_size);
2002 h.vm_state_size = cpu_to_be32(sn->vm_state_size);
2003 h.date_sec = cpu_to_be32(sn->date_sec);
2004 h.date_nsec = cpu_to_be32(sn->date_nsec);
2005 h.vm_clock_nsec = cpu_to_be64(sn->vm_clock_nsec);
3b46e624 2006
585f8587
FB
2007 id_str_size = strlen(sn->id_str);
2008 name_size = strlen(sn->name);
2009 h.id_str_size = cpu_to_be16(id_str_size);
2010 h.name_size = cpu_to_be16(name_size);
2011 offset = align_offset(offset, 8);
2012 if (bdrv_pwrite(s->hd, offset, &h, sizeof(h)) != sizeof(h))
2013 goto fail;
2014 offset += sizeof(h);
2015 if (bdrv_pwrite(s->hd, offset, sn->id_str, id_str_size) != id_str_size)
2016 goto fail;
2017 offset += id_str_size;
2018 if (bdrv_pwrite(s->hd, offset, sn->name, name_size) != name_size)
2019 goto fail;
2020 offset += name_size;
2021 }
2022
2023 /* update the various header fields */
2024 data64 = cpu_to_be64(snapshots_offset);
2025 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, snapshots_offset),
2026 &data64, sizeof(data64)) != sizeof(data64))
2027 goto fail;
2028 data32 = cpu_to_be32(s->nb_snapshots);
2029 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, nb_snapshots),
2030 &data32, sizeof(data32)) != sizeof(data32))
2031 goto fail;
2032
2033 /* free the old snapshot table */
2034 free_clusters(bs, s->snapshots_offset, s->snapshots_size);
2035 s->snapshots_offset = snapshots_offset;
2036 s->snapshots_size = snapshots_size;
2037 return 0;
2038 fail:
2039 return -1;
2040}
2041
2042static void find_new_snapshot_id(BlockDriverState *bs,
2043 char *id_str, int id_str_size)
2044{
2045 BDRVQcowState *s = bs->opaque;
2046 QCowSnapshot *sn;
2047 int i, id, id_max = 0;
2048
2049 for(i = 0; i < s->nb_snapshots; i++) {
2050 sn = s->snapshots + i;
2051 id = strtoul(sn->id_str, NULL, 10);
2052 if (id > id_max)
2053 id_max = id;
2054 }
2055 snprintf(id_str, id_str_size, "%d", id_max + 1);
2056}
2057
2058static int find_snapshot_by_id(BlockDriverState *bs, const char *id_str)
2059{
2060 BDRVQcowState *s = bs->opaque;
2061 int i;
2062
2063 for(i = 0; i < s->nb_snapshots; i++) {
2064 if (!strcmp(s->snapshots[i].id_str, id_str))
2065 return i;
2066 }
2067 return -1;
2068}
2069
2070static int find_snapshot_by_id_or_name(BlockDriverState *bs, const char *name)
2071{
2072 BDRVQcowState *s = bs->opaque;
2073 int i, ret;
3b46e624 2074
585f8587
FB
2075 ret = find_snapshot_by_id(bs, name);
2076 if (ret >= 0)
2077 return ret;
2078 for(i = 0; i < s->nb_snapshots; i++) {
2079 if (!strcmp(s->snapshots[i].name, name))
2080 return i;
2081 }
2082 return -1;
2083}
2084
2085/* if no id is provided, a new one is constructed */
5fafdf24 2086static int qcow_snapshot_create(BlockDriverState *bs,
585f8587
FB
2087 QEMUSnapshotInfo *sn_info)
2088{
2089 BDRVQcowState *s = bs->opaque;
2090 QCowSnapshot *snapshots1, sn1, *sn = &sn1;
2091 int i, ret;
2092 uint64_t *l1_table = NULL;
3b46e624 2093
585f8587
FB
2094 memset(sn, 0, sizeof(*sn));
2095
2096 if (sn_info->id_str[0] == '\0') {
2097 /* compute a new id */
2098 find_new_snapshot_id(bs, sn_info->id_str, sizeof(sn_info->id_str));
2099 }
2100
2101 /* check that the ID is unique */
2102 if (find_snapshot_by_id(bs, sn_info->id_str) >= 0)
2103 return -ENOENT;
2104
2105 sn->id_str = qemu_strdup(sn_info->id_str);
2106 if (!sn->id_str)
2107 goto fail;
2108 sn->name = qemu_strdup(sn_info->name);
2109 if (!sn->name)
2110 goto fail;
2111 sn->vm_state_size = sn_info->vm_state_size;
2112 sn->date_sec = sn_info->date_sec;
2113 sn->date_nsec = sn_info->date_nsec;
2114 sn->vm_clock_nsec = sn_info->vm_clock_nsec;
2115
2116 ret = update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1);
2117 if (ret < 0)
2118 goto fail;
2119
2120 /* create the L1 table of the snapshot */
2121 sn->l1_table_offset = alloc_clusters(bs, s->l1_size * sizeof(uint64_t));
2122 sn->l1_size = s->l1_size;
2123
2124 l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
585f8587
FB
2125 for(i = 0; i < s->l1_size; i++) {
2126 l1_table[i] = cpu_to_be64(s->l1_table[i]);
2127 }
2128 if (bdrv_pwrite(s->hd, sn->l1_table_offset,
5fafdf24 2129 l1_table, s->l1_size * sizeof(uint64_t)) !=
585f8587
FB
2130 (s->l1_size * sizeof(uint64_t)))
2131 goto fail;
2132 qemu_free(l1_table);
2133 l1_table = NULL;
2134
2135 snapshots1 = qemu_malloc((s->nb_snapshots + 1) * sizeof(QCowSnapshot));
63c75dcd 2136 if (s->snapshots) {
2137 memcpy(snapshots1, s->snapshots, s->nb_snapshots * sizeof(QCowSnapshot));
2138 qemu_free(s->snapshots);
2139 }
585f8587
FB
2140 s->snapshots = snapshots1;
2141 s->snapshots[s->nb_snapshots++] = *sn;
2142
2143 if (qcow_write_snapshots(bs) < 0)
2144 goto fail;
2145#ifdef DEBUG_ALLOC
2146 check_refcounts(bs);
2147#endif
2148 return 0;
2149 fail:
2150 qemu_free(sn->name);
2151 qemu_free(l1_table);
2152 return -1;
2153}
2154
2155/* copy the snapshot 'snapshot_name' into the current disk image */
5fafdf24 2156static int qcow_snapshot_goto(BlockDriverState *bs,
585f8587
FB
2157 const char *snapshot_id)
2158{
2159 BDRVQcowState *s = bs->opaque;
2160 QCowSnapshot *sn;
2161 int i, snapshot_index, l1_size2;
2162
2163 snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
2164 if (snapshot_index < 0)
2165 return -ENOENT;
2166 sn = &s->snapshots[snapshot_index];
2167
2168 if (update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, -1) < 0)
2169 goto fail;
2170
2171 if (grow_l1_table(bs, sn->l1_size) < 0)
2172 goto fail;
2173
2174 s->l1_size = sn->l1_size;
2175 l1_size2 = s->l1_size * sizeof(uint64_t);
2176 /* copy the snapshot l1 table to the current l1 table */
5fafdf24 2177 if (bdrv_pread(s->hd, sn->l1_table_offset,
585f8587
FB
2178 s->l1_table, l1_size2) != l1_size2)
2179 goto fail;
2180 if (bdrv_pwrite(s->hd, s->l1_table_offset,
2181 s->l1_table, l1_size2) != l1_size2)
2182 goto fail;
2183 for(i = 0;i < s->l1_size; i++) {
2184 be64_to_cpus(&s->l1_table[i]);
2185 }
2186
2187 if (update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1) < 0)
2188 goto fail;
2189
2190#ifdef DEBUG_ALLOC
2191 check_refcounts(bs);
2192#endif
2193 return 0;
2194 fail:
2195 return -EIO;
2196}
2197
2198static int qcow_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
2199{
2200 BDRVQcowState *s = bs->opaque;
2201 QCowSnapshot *sn;
2202 int snapshot_index, ret;
3b46e624 2203
585f8587
FB
2204 snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
2205 if (snapshot_index < 0)
2206 return -ENOENT;
2207 sn = &s->snapshots[snapshot_index];
2208
2209 ret = update_snapshot_refcount(bs, sn->l1_table_offset, sn->l1_size, -1);
2210 if (ret < 0)
2211 return ret;
2212 /* must update the copied flag on the current cluster offsets */
2213 ret = update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 0);
2214 if (ret < 0)
2215 return ret;
2216 free_clusters(bs, sn->l1_table_offset, sn->l1_size * sizeof(uint64_t));
2217
2218 qemu_free(sn->id_str);
2219 qemu_free(sn->name);
2220 memmove(sn, sn + 1, (s->nb_snapshots - snapshot_index - 1) * sizeof(*sn));
2221 s->nb_snapshots--;
2222 ret = qcow_write_snapshots(bs);
2223 if (ret < 0) {
2224 /* XXX: restore snapshot if error ? */
2225 return ret;
2226 }
2227#ifdef DEBUG_ALLOC
2228 check_refcounts(bs);
2229#endif
2230 return 0;
2231}
2232
5fafdf24 2233static int qcow_snapshot_list(BlockDriverState *bs,
585f8587
FB
2234 QEMUSnapshotInfo **psn_tab)
2235{
2236 BDRVQcowState *s = bs->opaque;
2237 QEMUSnapshotInfo *sn_tab, *sn_info;
2238 QCowSnapshot *sn;
2239 int i;
2240
2241 sn_tab = qemu_mallocz(s->nb_snapshots * sizeof(QEMUSnapshotInfo));
585f8587
FB
2242 for(i = 0; i < s->nb_snapshots; i++) {
2243 sn_info = sn_tab + i;
2244 sn = s->snapshots + i;
2245 pstrcpy(sn_info->id_str, sizeof(sn_info->id_str),
2246 sn->id_str);
2247 pstrcpy(sn_info->name, sizeof(sn_info->name),
2248 sn->name);
2249 sn_info->vm_state_size = sn->vm_state_size;
2250 sn_info->date_sec = sn->date_sec;
2251 sn_info->date_nsec = sn->date_nsec;
2252 sn_info->vm_clock_nsec = sn->vm_clock_nsec;
2253 }
2254 *psn_tab = sn_tab;
2255 return s->nb_snapshots;
585f8587
FB
2256}
2257
2258/*********************************************************/
2259/* refcount handling */
2260
2261static int refcount_init(BlockDriverState *bs)
2262{
2263 BDRVQcowState *s = bs->opaque;
2264 int ret, refcount_table_size2, i;
3b46e624 2265
585f8587 2266 s->refcount_block_cache = qemu_malloc(s->cluster_size);
585f8587
FB
2267 refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t);
2268 s->refcount_table = qemu_malloc(refcount_table_size2);
585f8587
FB
2269 if (s->refcount_table_size > 0) {
2270 ret = bdrv_pread(s->hd, s->refcount_table_offset,
2271 s->refcount_table, refcount_table_size2);
2272 if (ret != refcount_table_size2)
2273 goto fail;
2274 for(i = 0; i < s->refcount_table_size; i++)
2275 be64_to_cpus(&s->refcount_table[i]);
2276 }
2277 return 0;
2278 fail:
2279 return -ENOMEM;
2280}
2281
2282static void refcount_close(BlockDriverState *bs)
2283{
2284 BDRVQcowState *s = bs->opaque;
2285 qemu_free(s->refcount_block_cache);
2286 qemu_free(s->refcount_table);
2287}
2288
2289
5fafdf24 2290static int load_refcount_block(BlockDriverState *bs,
585f8587
FB
2291 int64_t refcount_block_offset)
2292{
2293 BDRVQcowState *s = bs->opaque;
2294 int ret;
5fafdf24 2295 ret = bdrv_pread(s->hd, refcount_block_offset, s->refcount_block_cache,
585f8587
FB
2296 s->cluster_size);
2297 if (ret != s->cluster_size)
2298 return -EIO;
2299 s->refcount_block_cache_offset = refcount_block_offset;
2300 return 0;
2301}
2302
2303static int get_refcount(BlockDriverState *bs, int64_t cluster_index)
2304{
2305 BDRVQcowState *s = bs->opaque;
2306 int refcount_table_index, block_index;
2307 int64_t refcount_block_offset;
2308
2309 refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
2310 if (refcount_table_index >= s->refcount_table_size)
2311 return 0;
2312 refcount_block_offset = s->refcount_table[refcount_table_index];
2313 if (!refcount_block_offset)
2314 return 0;
2315 if (refcount_block_offset != s->refcount_block_cache_offset) {
2316 /* better than nothing: return allocated if read error */
2317 if (load_refcount_block(bs, refcount_block_offset) < 0)
2318 return 1;
2319 }
5fafdf24 2320 block_index = cluster_index &
585f8587
FB
2321 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
2322 return be16_to_cpu(s->refcount_block_cache[block_index]);
2323}
2324
2325/* return < 0 if error */
2326static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size)
2327{
2328 BDRVQcowState *s = bs->opaque;
2329 int i, nb_clusters;
2330
6db6c638
AL
2331 nb_clusters = size_to_clusters(s, size);
2332retry:
2333 for(i = 0; i < nb_clusters; i++) {
2334 int64_t i = s->free_cluster_index++;
2335 if (get_refcount(bs, i) != 0)
2336 goto retry;
2337 }
585f8587 2338#ifdef DEBUG_ALLOC2
6db6c638
AL
2339 printf("alloc_clusters: size=%lld -> %lld\n",
2340 size,
2341 (s->free_cluster_index - nb_clusters) << s->cluster_bits);
585f8587 2342#endif
6db6c638 2343 return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
585f8587
FB
2344}
2345
2346static int64_t alloc_clusters(BlockDriverState *bs, int64_t size)
2347{
2348 int64_t offset;
2349
2350 offset = alloc_clusters_noref(bs, size);
2351 update_refcount(bs, offset, size, 1);
2352 return offset;
2353}
2354
2355/* only used to allocate compressed sectors. We try to allocate
2356 contiguous sectors. size must be <= cluster_size */
2357static int64_t alloc_bytes(BlockDriverState *bs, int size)
2358{
2359 BDRVQcowState *s = bs->opaque;
2360 int64_t offset, cluster_offset;
2361 int free_in_cluster;
3b46e624 2362
585f8587
FB
2363 assert(size > 0 && size <= s->cluster_size);
2364 if (s->free_byte_offset == 0) {
2365 s->free_byte_offset = alloc_clusters(bs, s->cluster_size);
2366 }
2367 redo:
5fafdf24 2368 free_in_cluster = s->cluster_size -
585f8587
FB
2369 (s->free_byte_offset & (s->cluster_size - 1));
2370 if (size <= free_in_cluster) {
2371 /* enough space in current cluster */
2372 offset = s->free_byte_offset;
2373 s->free_byte_offset += size;
2374 free_in_cluster -= size;
2375 if (free_in_cluster == 0)
2376 s->free_byte_offset = 0;
2377 if ((offset & (s->cluster_size - 1)) != 0)
2378 update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
2379 } else {
2380 offset = alloc_clusters(bs, s->cluster_size);
2381 cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1);
2382 if ((cluster_offset + s->cluster_size) == offset) {
2383 /* we are lucky: contiguous data */
2384 offset = s->free_byte_offset;
2385 update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
2386 s->free_byte_offset += size;
2387 } else {
2388 s->free_byte_offset = offset;
2389 goto redo;
2390 }
2391 }
2392 return offset;
2393}
2394
5fafdf24 2395static void free_clusters(BlockDriverState *bs,
585f8587
FB
2396 int64_t offset, int64_t size)
2397{
2398 update_refcount(bs, offset, size, -1);
2399}
2400
2401static int grow_refcount_table(BlockDriverState *bs, int min_size)
2402{
2403 BDRVQcowState *s = bs->opaque;
2404 int new_table_size, new_table_size2, refcount_table_clusters, i, ret;
2405 uint64_t *new_table;
2406 int64_t table_offset;
643e5399 2407 uint8_t data[12];
23be50f1
TS
2408 int old_table_size;
2409 int64_t old_table_offset;
585f8587
FB
2410
2411 if (min_size <= s->refcount_table_size)
2412 return 0;
2413 /* compute new table size */
2414 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
2415 for(;;) {
2416 if (refcount_table_clusters == 0) {
2417 refcount_table_clusters = 1;
2418 } else {
2419 refcount_table_clusters = (refcount_table_clusters * 3 + 1) / 2;
2420 }
2421 new_table_size = refcount_table_clusters << (s->cluster_bits - 3);
2422 if (min_size <= new_table_size)
2423 break;
2424 }
15e6690a
FB
2425#ifdef DEBUG_ALLOC2
2426 printf("grow_refcount_table from %d to %d\n",
2427 s->refcount_table_size,
2428 new_table_size);
2429#endif
585f8587
FB
2430 new_table_size2 = new_table_size * sizeof(uint64_t);
2431 new_table = qemu_mallocz(new_table_size2);
5fafdf24 2432 memcpy(new_table, s->refcount_table,
585f8587
FB
2433 s->refcount_table_size * sizeof(uint64_t));
2434 for(i = 0; i < s->refcount_table_size; i++)
2435 cpu_to_be64s(&new_table[i]);
2436 /* Note: we cannot update the refcount now to avoid recursion */
2437 table_offset = alloc_clusters_noref(bs, new_table_size2);
2438 ret = bdrv_pwrite(s->hd, table_offset, new_table, new_table_size2);
5fafdf24 2439 if (ret != new_table_size2)
585f8587
FB
2440 goto fail;
2441 for(i = 0; i < s->refcount_table_size; i++)
2442 be64_to_cpus(&new_table[i]);
2443
643e5399
AL
2444 cpu_to_be64w((uint64_t*)data, table_offset);
2445 cpu_to_be32w((uint32_t*)(data + 8), refcount_table_clusters);
585f8587 2446 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, refcount_table_offset),
643e5399 2447 data, sizeof(data)) != sizeof(data))
585f8587
FB
2448 goto fail;
2449 qemu_free(s->refcount_table);
23be50f1
TS
2450 old_table_offset = s->refcount_table_offset;
2451 old_table_size = s->refcount_table_size;
585f8587
FB
2452 s->refcount_table = new_table;
2453 s->refcount_table_size = new_table_size;
a4080ece 2454 s->refcount_table_offset = table_offset;
585f8587
FB
2455
2456 update_refcount(bs, table_offset, new_table_size2, 1);
23be50f1 2457 free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t));
585f8587
FB
2458 return 0;
2459 fail:
2460 free_clusters(bs, table_offset, new_table_size2);
2461 qemu_free(new_table);
2462 return -EIO;
2463}
2464
2465/* addend must be 1 or -1 */
2466/* XXX: cache several refcount block clusters ? */
5fafdf24 2467static int update_cluster_refcount(BlockDriverState *bs,
585f8587
FB
2468 int64_t cluster_index,
2469 int addend)
2470{
2471 BDRVQcowState *s = bs->opaque;
2472 int64_t offset, refcount_block_offset;
2473 int ret, refcount_table_index, block_index, refcount;
2474 uint64_t data64;
2475
2476 refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
2477 if (refcount_table_index >= s->refcount_table_size) {
2478 if (addend < 0)
2479 return -EINVAL;
2480 ret = grow_refcount_table(bs, refcount_table_index + 1);
2481 if (ret < 0)
2482 return ret;
2483 }
2484 refcount_block_offset = s->refcount_table[refcount_table_index];
2485 if (!refcount_block_offset) {
2486 if (addend < 0)
2487 return -EINVAL;
2488 /* create a new refcount block */
2489 /* Note: we cannot update the refcount now to avoid recursion */
2490 offset = alloc_clusters_noref(bs, s->cluster_size);
2491 memset(s->refcount_block_cache, 0, s->cluster_size);
2492 ret = bdrv_pwrite(s->hd, offset, s->refcount_block_cache, s->cluster_size);
2493 if (ret != s->cluster_size)
2494 return -EINVAL;
2495 s->refcount_table[refcount_table_index] = offset;
2496 data64 = cpu_to_be64(offset);
5fafdf24
TS
2497 ret = bdrv_pwrite(s->hd, s->refcount_table_offset +
2498 refcount_table_index * sizeof(uint64_t),
585f8587
FB
2499 &data64, sizeof(data64));
2500 if (ret != sizeof(data64))
2501 return -EINVAL;
2502
2503 refcount_block_offset = offset;
2504 s->refcount_block_cache_offset = offset;
2505 update_refcount(bs, offset, s->cluster_size, 1);
2506 } else {
2507 if (refcount_block_offset != s->refcount_block_cache_offset) {
2508 if (load_refcount_block(bs, refcount_block_offset) < 0)
2509 return -EIO;
2510 }
2511 }
2512 /* we can update the count and save it */
5fafdf24 2513 block_index = cluster_index &
585f8587
FB
2514 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
2515 refcount = be16_to_cpu(s->refcount_block_cache[block_index]);
2516 refcount += addend;
2517 if (refcount < 0 || refcount > 0xffff)
2518 return -EINVAL;
2519 if (refcount == 0 && cluster_index < s->free_cluster_index) {
2520 s->free_cluster_index = cluster_index;
2521 }
2522 s->refcount_block_cache[block_index] = cpu_to_be16(refcount);
5fafdf24
TS
2523 if (bdrv_pwrite(s->hd,
2524 refcount_block_offset + (block_index << REFCOUNT_SHIFT),
585f8587
FB
2525 &s->refcount_block_cache[block_index], 2) != 2)
2526 return -EIO;
2527 return refcount;
2528}
2529
5fafdf24
TS
2530static void update_refcount(BlockDriverState *bs,
2531 int64_t offset, int64_t length,
585f8587
FB
2532 int addend)
2533{
2534 BDRVQcowState *s = bs->opaque;
2535 int64_t start, last, cluster_offset;
2536
2537#ifdef DEBUG_ALLOC2
5fafdf24 2538 printf("update_refcount: offset=%lld size=%lld addend=%d\n",
585f8587
FB
2539 offset, length, addend);
2540#endif
2541 if (length <= 0)
2542 return;
2543 start = offset & ~(s->cluster_size - 1);
2544 last = (offset + length - 1) & ~(s->cluster_size - 1);
5fafdf24 2545 for(cluster_offset = start; cluster_offset <= last;
585f8587
FB
2546 cluster_offset += s->cluster_size) {
2547 update_cluster_refcount(bs, cluster_offset >> s->cluster_bits, addend);
2548 }
2549}
2550
2551#ifdef DEBUG_ALLOC
5fafdf24
TS
2552static void inc_refcounts(BlockDriverState *bs,
2553 uint16_t *refcount_table,
585f8587
FB
2554 int refcount_table_size,
2555 int64_t offset, int64_t size)
2556{
2557 BDRVQcowState *s = bs->opaque;
2558 int64_t start, last, cluster_offset;
2559 int k;
3b46e624 2560
585f8587
FB
2561 if (size <= 0)
2562 return;
2563
2564 start = offset & ~(s->cluster_size - 1);
2565 last = (offset + size - 1) & ~(s->cluster_size - 1);
5fafdf24 2566 for(cluster_offset = start; cluster_offset <= last;
585f8587
FB
2567 cluster_offset += s->cluster_size) {
2568 k = cluster_offset >> s->cluster_bits;
2569 if (k < 0 || k >= refcount_table_size) {
2570 printf("ERROR: invalid cluster offset=0x%llx\n", cluster_offset);
2571 } else {
2572 if (++refcount_table[k] == 0) {
2573 printf("ERROR: overflow cluster offset=0x%llx\n", cluster_offset);
2574 }
2575 }
2576 }
2577}
2578
5fafdf24
TS
2579static int check_refcounts_l1(BlockDriverState *bs,
2580 uint16_t *refcount_table,
585f8587
FB
2581 int refcount_table_size,
2582 int64_t l1_table_offset, int l1_size,
2583 int check_copied)
2584{
2585 BDRVQcowState *s = bs->opaque;
2586 uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2;
2587 int l2_size, i, j, nb_csectors, refcount;
2588
2589 l2_table = NULL;
2590 l1_size2 = l1_size * sizeof(uint64_t);
2591
2592 inc_refcounts(bs, refcount_table, refcount_table_size,
2593 l1_table_offset, l1_size2);
2594
2595 l1_table = qemu_malloc(l1_size2);
5fafdf24 2596 if (bdrv_pread(s->hd, l1_table_offset,
585f8587
FB
2597 l1_table, l1_size2) != l1_size2)
2598 goto fail;
2599 for(i = 0;i < l1_size; i++)
2600 be64_to_cpus(&l1_table[i]);
3b46e624 2601
585f8587
FB
2602 l2_size = s->l2_size * sizeof(uint64_t);
2603 l2_table = qemu_malloc(l2_size);
585f8587
FB
2604 for(i = 0; i < l1_size; i++) {
2605 l2_offset = l1_table[i];
2606 if (l2_offset) {
2607 if (check_copied) {
2608 refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED) >> s->cluster_bits);
2609 if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) {
2610 printf("ERROR OFLAG_COPIED: l2_offset=%llx refcount=%d\n",
2611 l2_offset, refcount);
2612 }
2613 }
2614 l2_offset &= ~QCOW_OFLAG_COPIED;
2615 if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
2616 goto fail;
2617 for(j = 0; j < s->l2_size; j++) {
2618 offset = be64_to_cpu(l2_table[j]);
2619 if (offset != 0) {
2620 if (offset & QCOW_OFLAG_COMPRESSED) {
2621 if (offset & QCOW_OFLAG_COPIED) {
2622 printf("ERROR: cluster %lld: copied flag must never be set for compressed clusters\n",
2623 offset >> s->cluster_bits);
2624 offset &= ~QCOW_OFLAG_COPIED;
2625 }
5fafdf24 2626 nb_csectors = ((offset >> s->csize_shift) &
585f8587
FB
2627 s->csize_mask) + 1;
2628 offset &= s->cluster_offset_mask;
5fafdf24 2629 inc_refcounts(bs, refcount_table,
585f8587
FB
2630 refcount_table_size,
2631 offset & ~511, nb_csectors * 512);
2632 } else {
2633 if (check_copied) {
2634 refcount = get_refcount(bs, (offset & ~QCOW_OFLAG_COPIED) >> s->cluster_bits);
2635 if ((refcount == 1) != ((offset & QCOW_OFLAG_COPIED) != 0)) {
2636 printf("ERROR OFLAG_COPIED: offset=%llx refcount=%d\n",
2637 offset, refcount);
2638 }
2639 }
2640 offset &= ~QCOW_OFLAG_COPIED;
5fafdf24 2641 inc_refcounts(bs, refcount_table,
585f8587
FB
2642 refcount_table_size,
2643 offset, s->cluster_size);
2644 }
2645 }
2646 }
5fafdf24 2647 inc_refcounts(bs, refcount_table,
585f8587
FB
2648 refcount_table_size,
2649 l2_offset,
2650 s->cluster_size);
2651 }
2652 }
2653 qemu_free(l1_table);
2654 qemu_free(l2_table);
2655 return 0;
2656 fail:
2657 printf("ERROR: I/O error in check_refcounts_l1\n");
2658 qemu_free(l1_table);
2659 qemu_free(l2_table);
2660 return -EIO;
2661}
2662
2663static void check_refcounts(BlockDriverState *bs)
2664{
2665 BDRVQcowState *s = bs->opaque;
2666 int64_t size;
2667 int nb_clusters, refcount1, refcount2, i;
2668 QCowSnapshot *sn;
2669 uint16_t *refcount_table;
2670
2671 size = bdrv_getlength(s->hd);
6db6c638 2672 nb_clusters = size_to_clusters(s, size);
585f8587 2673 refcount_table = qemu_mallocz(nb_clusters * sizeof(uint16_t));
15e6690a 2674
585f8587
FB
2675 /* header */
2676 inc_refcounts(bs, refcount_table, nb_clusters,
2677 0, s->cluster_size);
3b46e624 2678
585f8587
FB
2679 check_refcounts_l1(bs, refcount_table, nb_clusters,
2680 s->l1_table_offset, s->l1_size, 1);
2681
2682 /* snapshots */
2683 for(i = 0; i < s->nb_snapshots; i++) {
2684 sn = s->snapshots + i;
2685 check_refcounts_l1(bs, refcount_table, nb_clusters,
2686 sn->l1_table_offset, sn->l1_size, 0);
2687 }
2688 inc_refcounts(bs, refcount_table, nb_clusters,
2689 s->snapshots_offset, s->snapshots_size);
2690
2691 /* refcount data */
2692 inc_refcounts(bs, refcount_table, nb_clusters,
5fafdf24 2693 s->refcount_table_offset,
585f8587
FB
2694 s->refcount_table_size * sizeof(uint64_t));
2695 for(i = 0; i < s->refcount_table_size; i++) {
2696 int64_t offset;
2697 offset = s->refcount_table[i];
2698 if (offset != 0) {
2699 inc_refcounts(bs, refcount_table, nb_clusters,
2700 offset, s->cluster_size);
2701 }
2702 }
2703
2704 /* compare ref counts */
2705 for(i = 0; i < nb_clusters; i++) {
2706 refcount1 = get_refcount(bs, i);
2707 refcount2 = refcount_table[i];
2708 if (refcount1 != refcount2)
2709 printf("ERROR cluster %d refcount=%d reference=%d\n",
2710 i, refcount1, refcount2);
2711 }
2712
2713 qemu_free(refcount_table);
2714}
2715
2716#if 0
2717static void dump_refcounts(BlockDriverState *bs)
2718{
2719 BDRVQcowState *s = bs->opaque;
2720 int64_t nb_clusters, k, k1, size;
2721 int refcount;
2722
2723 size = bdrv_getlength(s->hd);
6db6c638 2724 nb_clusters = size_to_clusters(s, size);
585f8587
FB
2725 for(k = 0; k < nb_clusters;) {
2726 k1 = k;
2727 refcount = get_refcount(bs, k);
2728 k++;
2729 while (k < nb_clusters && get_refcount(bs, k) == refcount)
2730 k++;
2731 printf("%lld: refcount=%d nb=%lld\n", k, refcount, k - k1);
2732 }
2733}
2734#endif
2735#endif
2736
178e08a5
AL
2737static int qcow_put_buffer(BlockDriverState *bs, const uint8_t *buf,
2738 int64_t pos, int size)
2739{
2740 int growable = bs->growable;
2741
2742 bs->growable = 1;
2743 bdrv_pwrite(bs, pos, buf, size);
2744 bs->growable = growable;
2745
2746 return size;
2747}
2748
2749static int qcow_get_buffer(BlockDriverState *bs, uint8_t *buf,
2750 int64_t pos, int size)
2751{
2752 int growable = bs->growable;
2753 int ret;
2754
2755 bs->growable = 1;
2756 ret = bdrv_pread(bs, pos, buf, size);
2757 bs->growable = growable;
2758
2759 return ret;
2760}
2761
585f8587 2762BlockDriver bdrv_qcow2 = {
e60f469c
AJ
2763 .format_name = "qcow2",
2764 .instance_size = sizeof(BDRVQcowState),
2765 .bdrv_probe = qcow_probe,
2766 .bdrv_open = qcow_open,
2767 .bdrv_close = qcow_close,
2768 .bdrv_create = qcow_create,
2769 .bdrv_flush = qcow_flush,
2770 .bdrv_is_allocated = qcow_is_allocated,
2771 .bdrv_set_key = qcow_set_key,
2772 .bdrv_make_empty = qcow_make_empty,
2773
2774 .bdrv_aio_read = qcow_aio_read,
2775 .bdrv_aio_write = qcow_aio_write,
2776 .bdrv_aio_cancel = qcow_aio_cancel,
2777 .aiocb_size = sizeof(QCowAIOCB),
585f8587
FB
2778 .bdrv_write_compressed = qcow_write_compressed,
2779
2780 .bdrv_snapshot_create = qcow_snapshot_create,
e60f469c 2781 .bdrv_snapshot_goto = qcow_snapshot_goto,
585f8587 2782 .bdrv_snapshot_delete = qcow_snapshot_delete,
e60f469c
AJ
2783 .bdrv_snapshot_list = qcow_snapshot_list,
2784 .bdrv_get_info = qcow_get_info,
f965509c 2785
178e08a5
AL
2786 .bdrv_put_buffer = qcow_put_buffer,
2787 .bdrv_get_buffer = qcow_get_buffer,
2788
f965509c 2789 .bdrv_create2 = qcow_create2,
585f8587 2790};