Clarify -soundhw documentation as to avoid hda confusion
[qemu/mini2440.git] / block-qcow2.c
blobc9d68af93f474987095f94188efa59d675e755f9
1 /*
2 * Block driver for the QCOW version 2 format
4 * Copyright (c) 2004-2006 Fabrice Bellard
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:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
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.
24 #include "qemu-common.h"
25 #include "block_int.h"
26 #include <zlib.h>
27 #include "aes.h"
28 #include <assert.h>
31 Differences with QCOW:
33 - Support for multiple incremental snapshots.
34 - Memory management by reference counts.
35 - Clusters which have a reference count of one have the bit
36 QCOW_OFLAG_COPIED to optimize write performance.
37 - Size of compressed clusters is stored in sectors to reduce bit usage
38 in the cluster offsets.
39 - Support for storing additional data (such as the VM state) in the
40 snapshots.
41 - If a backing store is used, the cluster size is not constrained
42 (could be backported to QCOW).
43 - L2 tables have always a size of one cluster.
46 //#define DEBUG_ALLOC
47 //#define DEBUG_ALLOC2
49 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
50 #define QCOW_VERSION 2
52 #define QCOW_CRYPT_NONE 0
53 #define QCOW_CRYPT_AES 1
55 #define QCOW_MAX_CRYPT_CLUSTERS 32
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)
62 #define REFCOUNT_SHIFT 1 /* refcount size is 2 bytes */
64 typedef struct QCowHeader {
65 uint32_t magic;
66 uint32_t version;
67 uint64_t backing_file_offset;
68 uint32_t backing_file_size;
69 uint32_t cluster_bits;
70 uint64_t size; /* in bytes */
71 uint32_t crypt_method;
72 uint32_t l1_size; /* XXX: save number of clusters instead ? */
73 uint64_t l1_table_offset;
74 uint64_t refcount_table_offset;
75 uint32_t refcount_table_clusters;
76 uint32_t nb_snapshots;
77 uint64_t snapshots_offset;
78 } QCowHeader;
80 typedef struct __attribute__((packed)) QCowSnapshotHeader {
81 /* header is 8 byte aligned */
82 uint64_t l1_table_offset;
84 uint32_t l1_size;
85 uint16_t id_str_size;
86 uint16_t name_size;
88 uint32_t date_sec;
89 uint32_t date_nsec;
91 uint64_t vm_clock_nsec;
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;
100 #define L2_CACHE_SIZE 16
102 typedef struct QCowSnapshot {
103 uint64_t l1_table_offset;
104 uint32_t l1_size;
105 char *id_str;
106 char *name;
107 uint32_t vm_state_size;
108 uint32_t date_sec;
109 uint32_t date_nsec;
110 uint64_t vm_clock_nsec;
111 } QCowSnapshot;
113 typedef struct BDRVQcowState {
114 BlockDriverState *hd;
115 int cluster_bits;
116 int cluster_size;
117 int cluster_sectors;
118 int l2_bits;
119 int l2_size;
120 int l1_size;
121 int l1_vm_state_index;
122 int csize_shift;
123 int csize_mask;
124 uint64_t cluster_offset_mask;
125 uint64_t l1_table_offset;
126 uint64_t *l1_table;
127 uint64_t *l2_cache;
128 uint64_t l2_cache_offsets[L2_CACHE_SIZE];
129 uint32_t l2_cache_counts[L2_CACHE_SIZE];
130 uint8_t *cluster_cache;
131 uint8_t *cluster_data;
132 uint64_t cluster_cache_offset;
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;
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;
152 static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset);
153 static int qcow_read(BlockDriverState *bs, int64_t sector_num,
154 uint8_t *buf, int nb_sectors);
155 static int qcow_read_snapshots(BlockDriverState *bs);
156 static void qcow_free_snapshots(BlockDriverState *bs);
157 static int refcount_init(BlockDriverState *bs);
158 static void refcount_close(BlockDriverState *bs);
159 static int get_refcount(BlockDriverState *bs, int64_t cluster_index);
160 static int update_cluster_refcount(BlockDriverState *bs,
161 int64_t cluster_index,
162 int addend);
163 static void update_refcount(BlockDriverState *bs,
164 int64_t offset, int64_t length,
165 int addend);
166 static int64_t alloc_clusters(BlockDriverState *bs, int64_t size);
167 static int64_t alloc_bytes(BlockDriverState *bs, int size);
168 static void free_clusters(BlockDriverState *bs,
169 int64_t offset, int64_t size);
170 #ifdef DEBUG_ALLOC
171 static void check_refcounts(BlockDriverState *bs);
172 #endif
174 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
176 const QCowHeader *cow_header = (const void *)buf;
178 if (buf_size >= sizeof(QCowHeader) &&
179 be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
180 be32_to_cpu(cow_header->version) == QCOW_VERSION)
181 return 100;
182 else
183 return 0;
186 static int qcow_open(BlockDriverState *bs, const char *filename, int flags)
188 BDRVQcowState *s = bs->opaque;
189 int len, i, shift, ret;
190 QCowHeader header;
192 ret = bdrv_file_open(&s->hd, filename, flags);
193 if (ret < 0)
194 return ret;
195 if (bdrv_pread(s->hd, 0, &header, sizeof(header)) != sizeof(header))
196 goto fail;
197 be32_to_cpus(&header.magic);
198 be32_to_cpus(&header.version);
199 be64_to_cpus(&header.backing_file_offset);
200 be32_to_cpus(&header.backing_file_size);
201 be64_to_cpus(&header.size);
202 be32_to_cpus(&header.cluster_bits);
203 be32_to_cpus(&header.crypt_method);
204 be64_to_cpus(&header.l1_table_offset);
205 be32_to_cpus(&header.l1_size);
206 be64_to_cpus(&header.refcount_table_offset);
207 be32_to_cpus(&header.refcount_table_clusters);
208 be64_to_cpus(&header.snapshots_offset);
209 be32_to_cpus(&header.nb_snapshots);
211 if (header.magic != QCOW_MAGIC || header.version != QCOW_VERSION)
212 goto fail;
213 if (header.size <= 1 ||
214 header.cluster_bits < 9 ||
215 header.cluster_bits > 16)
216 goto fail;
217 if (header.crypt_method > QCOW_CRYPT_AES)
218 goto fail;
219 s->crypt_method_header = header.crypt_method;
220 if (s->crypt_method_header)
221 bs->encrypted = 1;
222 s->cluster_bits = header.cluster_bits;
223 s->cluster_size = 1 << s->cluster_bits;
224 s->cluster_sectors = 1 << (s->cluster_bits - 9);
225 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
226 s->l2_size = 1 << s->l2_bits;
227 bs->total_sectors = header.size / 512;
228 s->csize_shift = (62 - (s->cluster_bits - 8));
229 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
230 s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
231 s->refcount_table_offset = header.refcount_table_offset;
232 s->refcount_table_size =
233 header.refcount_table_clusters << (s->cluster_bits - 3);
235 s->snapshots_offset = header.snapshots_offset;
236 s->nb_snapshots = header.nb_snapshots;
238 /* read the level 1 table */
239 s->l1_size = header.l1_size;
240 shift = s->cluster_bits + s->l2_bits;
241 s->l1_vm_state_index = (header.size + (1LL << shift) - 1) >> shift;
242 /* the L1 table must contain at least enough entries to put
243 header.size bytes */
244 if (s->l1_size < s->l1_vm_state_index)
245 goto fail;
246 s->l1_table_offset = header.l1_table_offset;
247 s->l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
248 if (!s->l1_table)
249 goto fail;
250 if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, s->l1_size * sizeof(uint64_t)) !=
251 s->l1_size * sizeof(uint64_t))
252 goto fail;
253 for(i = 0;i < s->l1_size; i++) {
254 be64_to_cpus(&s->l1_table[i]);
256 /* alloc L2 cache */
257 s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
258 if (!s->l2_cache)
259 goto fail;
260 s->cluster_cache = qemu_malloc(s->cluster_size);
261 if (!s->cluster_cache)
262 goto fail;
263 /* one more sector for decompressed data alignment */
264 s->cluster_data = qemu_malloc(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
265 + 512);
266 if (!s->cluster_data)
267 goto fail;
268 s->cluster_cache_offset = -1;
270 if (refcount_init(bs) < 0)
271 goto fail;
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';
282 if (qcow_read_snapshots(bs) < 0)
283 goto fail;
285 #ifdef DEBUG_ALLOC
286 check_refcounts(bs);
287 #endif
288 return 0;
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;
301 static int qcow_set_key(BlockDriverState *bs, const char *key)
303 BDRVQcowState *s = bs->opaque;
304 uint8_t keybuf[16];
305 int len, i;
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];
316 s->crypt_method = s->crypt_method_header;
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 */
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");
339 #endif
340 return 0;
343 /* The crypt function is compatible with the linux cryptoloop
344 algorithm for < 4 GB images. NOTE: out_buf == in_buf is
345 supported */
346 static void encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
347 uint8_t *out_buf, const uint8_t *in_buf,
348 int nb_sectors, int enc,
349 const AES_KEY *key)
351 union {
352 uint64_t ll[2];
353 uint8_t b[16];
354 } ivec;
355 int i;
357 for(i = 0; i < nb_sectors; i++) {
358 ivec.ll[0] = cpu_to_le64(sector_num);
359 ivec.ll[1] = 0;
360 AES_cbc_encrypt(in_buf, out_buf, 512, key,
361 ivec.b, enc);
362 sector_num++;
363 in_buf += 512;
364 out_buf += 512;
368 static int copy_sectors(BlockDriverState *bs, uint64_t start_sect,
369 uint64_t cluster_offset, int n_start, int n_end)
371 BDRVQcowState *s = bs->opaque;
372 int n, ret;
374 n = n_end - n_start;
375 if (n <= 0)
376 return 0;
377 ret = qcow_read(bs, start_sect + n_start, s->cluster_data, n);
378 if (ret < 0)
379 return ret;
380 if (s->crypt_method) {
381 encrypt_sectors(s, start_sect + n_start,
382 s->cluster_data,
383 s->cluster_data, n, 1,
384 &s->aes_encrypt_key);
386 ret = bdrv_write(s->hd, (cluster_offset >> 9) + n_start,
387 s->cluster_data, n);
388 if (ret < 0)
389 return ret;
390 return 0;
393 static void l2_cache_reset(BlockDriverState *bs)
395 BDRVQcowState *s = bs->opaque;
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));
402 static inline int l2_cache_new_entry(BlockDriverState *bs)
404 BDRVQcowState *s = bs->opaque;
405 uint32_t min_count;
406 int min_index, i;
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;
417 return min_index;
420 static int64_t align_offset(int64_t offset, int n)
422 offset = (offset + n - 1) & ~(n - 1);
423 return offset;
426 static int grow_l1_table(BlockDriverState *bs, int min_size)
428 BDRVQcowState *s = bs->opaque;
429 int new_l1_size, new_l1_size2, ret, i;
430 uint64_t *new_l1_table;
431 uint64_t new_l1_table_offset;
432 uint8_t data[12];
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;
440 #ifdef DEBUG_ALLOC2
441 printf("grow l1_table from %d to %d\n", s->l1_size, new_l1_size);
442 #endif
444 new_l1_size2 = sizeof(uint64_t) * new_l1_size;
445 new_l1_table = qemu_mallocz(new_l1_size2);
446 if (!new_l1_table)
447 return -ENOMEM;
448 memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t));
450 /* write new table (align to cluster) */
451 new_l1_table_offset = alloc_clusters(bs, new_l1_size2);
453 for(i = 0; i < s->l1_size; i++)
454 new_l1_table[i] = cpu_to_be64(new_l1_table[i]);
455 ret = bdrv_pwrite(s->hd, new_l1_table_offset, new_l1_table, new_l1_size2);
456 if (ret != new_l1_size2)
457 goto fail;
458 for(i = 0; i < s->l1_size; i++)
459 new_l1_table[i] = be64_to_cpu(new_l1_table[i]);
461 /* set new table */
462 cpu_to_be32w((uint32_t*)data, new_l1_size);
463 cpu_to_be64w((uint64_t*)(data + 4), new_l1_table_offset);
464 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, l1_size), data,
465 sizeof(data)) != sizeof(data))
466 goto fail;
467 qemu_free(s->l1_table);
468 free_clusters(bs, s->l1_table_offset, s->l1_size * sizeof(uint64_t));
469 s->l1_table_offset = new_l1_table_offset;
470 s->l1_table = new_l1_table;
471 s->l1_size = new_l1_size;
472 return 0;
473 fail:
474 qemu_free(s->l1_table);
475 return -EIO;
479 * seek_l2_table
481 * seek l2_offset in the l2_cache table
482 * if not found, return NULL,
483 * if found,
484 * increments the l2 cache hit count of the entry,
485 * if counter overflow, divide by two all counters
486 * return the pointer to the l2 cache entry
490 static uint64_t *seek_l2_table(BDRVQcowState *s, uint64_t l2_offset)
492 int i, j;
494 for(i = 0; i < L2_CACHE_SIZE; i++) {
495 if (l2_offset == s->l2_cache_offsets[i]) {
496 /* increment the hit count */
497 if (++s->l2_cache_counts[i] == 0xffffffff) {
498 for(j = 0; j < L2_CACHE_SIZE; j++) {
499 s->l2_cache_counts[j] >>= 1;
502 return s->l2_cache + (i << s->l2_bits);
505 return NULL;
509 * l2_load
511 * Loads a L2 table into memory. If the table is in the cache, the cache
512 * is used; otherwise the L2 table is loaded from the image file.
514 * Returns a pointer to the L2 table on success, or NULL if the read from
515 * the image file failed.
518 static uint64_t *l2_load(BlockDriverState *bs, uint64_t l2_offset)
520 BDRVQcowState *s = bs->opaque;
521 int min_index;
522 uint64_t *l2_table;
524 /* seek if the table for the given offset is in the cache */
526 l2_table = seek_l2_table(s, l2_offset);
527 if (l2_table != NULL)
528 return l2_table;
530 /* not found: load a new entry in the least used one */
532 min_index = l2_cache_new_entry(bs);
533 l2_table = s->l2_cache + (min_index << s->l2_bits);
534 if (bdrv_pread(s->hd, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
535 s->l2_size * sizeof(uint64_t))
536 return NULL;
537 s->l2_cache_offsets[min_index] = l2_offset;
538 s->l2_cache_counts[min_index] = 1;
540 return l2_table;
544 * l2_allocate
546 * Allocate a new l2 entry in the file. If l1_index points to an already
547 * used entry in the L2 table (i.e. we are doing a copy on write for the L2
548 * table) copy the contents of the old L2 table into the newly allocated one.
549 * Otherwise the new table is initialized with zeros.
553 static uint64_t *l2_allocate(BlockDriverState *bs, int l1_index)
555 BDRVQcowState *s = bs->opaque;
556 int min_index;
557 uint64_t old_l2_offset, tmp;
558 uint64_t *l2_table, l2_offset;
560 old_l2_offset = s->l1_table[l1_index];
562 /* allocate a new l2 entry */
564 l2_offset = alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
566 /* update the L1 entry */
568 s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
570 tmp = cpu_to_be64(l2_offset | QCOW_OFLAG_COPIED);
571 if (bdrv_pwrite(s->hd, s->l1_table_offset + l1_index * sizeof(tmp),
572 &tmp, sizeof(tmp)) != sizeof(tmp))
573 return NULL;
575 /* allocate a new entry in the l2 cache */
577 min_index = l2_cache_new_entry(bs);
578 l2_table = s->l2_cache + (min_index << s->l2_bits);
580 if (old_l2_offset == 0) {
581 /* if there was no old l2 table, clear the new table */
582 memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
583 } else {
584 /* if there was an old l2 table, read it from the disk */
585 if (bdrv_pread(s->hd, old_l2_offset,
586 l2_table, s->l2_size * sizeof(uint64_t)) !=
587 s->l2_size * sizeof(uint64_t))
588 return NULL;
590 /* write the l2 table to the file */
591 if (bdrv_pwrite(s->hd, l2_offset,
592 l2_table, s->l2_size * sizeof(uint64_t)) !=
593 s->l2_size * sizeof(uint64_t))
594 return NULL;
596 /* update the l2 cache entry */
598 s->l2_cache_offsets[min_index] = l2_offset;
599 s->l2_cache_counts[min_index] = 1;
601 return l2_table;
604 static int size_to_clusters(BDRVQcowState *s, int64_t size)
606 return (size + (s->cluster_size - 1)) >> s->cluster_bits;
609 static int count_contiguous_clusters(uint64_t nb_clusters, int cluster_size,
610 uint64_t *l2_table, uint64_t mask)
612 int i;
613 uint64_t offset = be64_to_cpu(l2_table[0]) & ~mask;
615 for (i = 0; i < nb_clusters; i++)
616 if (offset + i * cluster_size != (be64_to_cpu(l2_table[i]) & ~mask))
617 break;
619 return i;
622 static int count_contiguous_free_clusters(uint64_t nb_clusters, uint64_t *l2_table)
624 int i = 0;
626 while(nb_clusters-- && l2_table[i] == 0)
627 i++;
629 return i;
633 * get_cluster_offset
635 * For a given offset of the disk image, return cluster offset in
636 * qcow2 file.
638 * on entry, *num is the number of contiguous clusters we'd like to
639 * access following offset.
641 * on exit, *num is the number of contiguous clusters we can read.
643 * Return 1, if the offset is found
644 * Return 0, otherwise.
648 static uint64_t get_cluster_offset(BlockDriverState *bs,
649 uint64_t offset, int *num)
651 BDRVQcowState *s = bs->opaque;
652 int l1_index, l2_index;
653 uint64_t l2_offset, *l2_table, cluster_offset;
654 int l1_bits, c;
655 int index_in_cluster, nb_available, nb_needed, nb_clusters;
657 index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1);
658 nb_needed = *num + index_in_cluster;
660 l1_bits = s->l2_bits + s->cluster_bits;
662 /* compute how many bytes there are between the offset and
663 * the end of the l1 entry
666 nb_available = (1 << l1_bits) - (offset & ((1 << l1_bits) - 1));
668 /* compute the number of available sectors */
670 nb_available = (nb_available >> 9) + index_in_cluster;
672 cluster_offset = 0;
674 /* seek the the l2 offset in the l1 table */
676 l1_index = offset >> l1_bits;
677 if (l1_index >= s->l1_size)
678 goto out;
680 l2_offset = s->l1_table[l1_index];
682 /* seek the l2 table of the given l2 offset */
684 if (!l2_offset)
685 goto out;
687 /* load the l2 table in memory */
689 l2_offset &= ~QCOW_OFLAG_COPIED;
690 l2_table = l2_load(bs, l2_offset);
691 if (l2_table == NULL)
692 return 0;
694 /* find the cluster offset for the given disk offset */
696 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
697 cluster_offset = be64_to_cpu(l2_table[l2_index]);
698 nb_clusters = size_to_clusters(s, nb_needed << 9);
700 if (!cluster_offset) {
701 /* how many empty clusters ? */
702 c = count_contiguous_free_clusters(nb_clusters, &l2_table[l2_index]);
703 } else {
704 /* how many allocated clusters ? */
705 c = count_contiguous_clusters(nb_clusters, s->cluster_size,
706 &l2_table[l2_index], QCOW_OFLAG_COPIED);
709 nb_available = (c * s->cluster_sectors);
710 out:
711 if (nb_available > nb_needed)
712 nb_available = nb_needed;
714 *num = nb_available - index_in_cluster;
716 return cluster_offset & ~QCOW_OFLAG_COPIED;
720 * free_any_clusters
722 * free clusters according to its type: compressed or not
726 static void free_any_clusters(BlockDriverState *bs,
727 uint64_t cluster_offset, int nb_clusters)
729 BDRVQcowState *s = bs->opaque;
731 /* free the cluster */
733 if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
734 int nb_csectors;
735 nb_csectors = ((cluster_offset >> s->csize_shift) &
736 s->csize_mask) + 1;
737 free_clusters(bs, (cluster_offset & s->cluster_offset_mask) & ~511,
738 nb_csectors * 512);
739 return;
742 free_clusters(bs, cluster_offset, nb_clusters << s->cluster_bits);
744 return;
748 * get_cluster_table
750 * for a given disk offset, load (and allocate if needed)
751 * the l2 table.
753 * the l2 table offset in the qcow2 file and the cluster index
754 * in the l2 table are given to the caller.
758 static int get_cluster_table(BlockDriverState *bs, uint64_t offset,
759 uint64_t **new_l2_table,
760 uint64_t *new_l2_offset,
761 int *new_l2_index)
763 BDRVQcowState *s = bs->opaque;
764 int l1_index, l2_index, ret;
765 uint64_t l2_offset, *l2_table;
767 /* seek the the l2 offset in the l1 table */
769 l1_index = offset >> (s->l2_bits + s->cluster_bits);
770 if (l1_index >= s->l1_size) {
771 ret = grow_l1_table(bs, l1_index + 1);
772 if (ret < 0)
773 return 0;
775 l2_offset = s->l1_table[l1_index];
777 /* seek the l2 table of the given l2 offset */
779 if (l2_offset & QCOW_OFLAG_COPIED) {
780 /* load the l2 table in memory */
781 l2_offset &= ~QCOW_OFLAG_COPIED;
782 l2_table = l2_load(bs, l2_offset);
783 if (l2_table == NULL)
784 return 0;
785 } else {
786 if (l2_offset)
787 free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t));
788 l2_table = l2_allocate(bs, l1_index);
789 if (l2_table == NULL)
790 return 0;
791 l2_offset = s->l1_table[l1_index] & ~QCOW_OFLAG_COPIED;
794 /* find the cluster offset for the given disk offset */
796 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
798 *new_l2_table = l2_table;
799 *new_l2_offset = l2_offset;
800 *new_l2_index = l2_index;
802 return 1;
806 * alloc_compressed_cluster_offset
808 * For a given offset of the disk image, return cluster offset in
809 * qcow2 file.
811 * If the offset is not found, allocate a new compressed cluster.
813 * Return the cluster offset if successful,
814 * Return 0, otherwise.
818 static uint64_t alloc_compressed_cluster_offset(BlockDriverState *bs,
819 uint64_t offset,
820 int compressed_size)
822 BDRVQcowState *s = bs->opaque;
823 int l2_index, ret;
824 uint64_t l2_offset, *l2_table, cluster_offset;
825 int nb_csectors;
827 ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
828 if (ret == 0)
829 return 0;
831 cluster_offset = be64_to_cpu(l2_table[l2_index]);
832 if (cluster_offset & QCOW_OFLAG_COPIED)
833 return cluster_offset & ~QCOW_OFLAG_COPIED;
835 if (cluster_offset)
836 free_any_clusters(bs, cluster_offset, 1);
838 cluster_offset = alloc_bytes(bs, compressed_size);
839 nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) -
840 (cluster_offset >> 9);
842 cluster_offset |= QCOW_OFLAG_COMPRESSED |
843 ((uint64_t)nb_csectors << s->csize_shift);
845 /* update L2 table */
847 /* compressed clusters never have the copied flag */
849 l2_table[l2_index] = cpu_to_be64(cluster_offset);
850 if (bdrv_pwrite(s->hd,
851 l2_offset + l2_index * sizeof(uint64_t),
852 l2_table + l2_index,
853 sizeof(uint64_t)) != sizeof(uint64_t))
854 return 0;
856 return cluster_offset;
859 typedef struct QCowL2Meta
861 uint64_t offset;
862 int n_start;
863 int nb_available;
864 int nb_clusters;
865 } QCowL2Meta;
867 static int alloc_cluster_link_l2(BlockDriverState *bs, uint64_t cluster_offset,
868 QCowL2Meta *m)
870 BDRVQcowState *s = bs->opaque;
871 int i, j = 0, l2_index, ret;
872 uint64_t *old_cluster, start_sect, l2_offset, *l2_table;
874 if (m->nb_clusters == 0)
875 return 0;
877 if (!(old_cluster = qemu_malloc(m->nb_clusters * sizeof(uint64_t))))
878 return -ENOMEM;
880 /* copy content of unmodified sectors */
881 start_sect = (m->offset & ~(s->cluster_size - 1)) >> 9;
882 if (m->n_start) {
883 ret = copy_sectors(bs, start_sect, cluster_offset, 0, m->n_start);
884 if (ret < 0)
885 goto err;
888 if (m->nb_available & (s->cluster_sectors - 1)) {
889 uint64_t end = m->nb_available & ~(uint64_t)(s->cluster_sectors - 1);
890 ret = copy_sectors(bs, start_sect + end, cluster_offset + (end << 9),
891 m->nb_available - end, s->cluster_sectors);
892 if (ret < 0)
893 goto err;
896 ret = -EIO;
897 /* update L2 table */
898 if (!get_cluster_table(bs, m->offset, &l2_table, &l2_offset, &l2_index))
899 goto err;
901 for (i = 0; i < m->nb_clusters; i++) {
902 if(l2_table[l2_index + i] != 0)
903 old_cluster[j++] = l2_table[l2_index + i];
905 l2_table[l2_index + i] = cpu_to_be64((cluster_offset +
906 (i << s->cluster_bits)) | QCOW_OFLAG_COPIED);
909 if (bdrv_pwrite(s->hd, l2_offset + l2_index * sizeof(uint64_t),
910 l2_table + l2_index, m->nb_clusters * sizeof(uint64_t)) !=
911 m->nb_clusters * sizeof(uint64_t))
912 goto err;
914 for (i = 0; i < j; i++)
915 free_any_clusters(bs, old_cluster[i], 1);
917 ret = 0;
918 err:
919 qemu_free(old_cluster);
920 return ret;
924 * alloc_cluster_offset
926 * For a given offset of the disk image, return cluster offset in
927 * qcow2 file.
929 * If the offset is not found, allocate a new cluster.
931 * Return the cluster offset if successful,
932 * Return 0, otherwise.
936 static uint64_t alloc_cluster_offset(BlockDriverState *bs,
937 uint64_t offset,
938 int n_start, int n_end,
939 int *num, QCowL2Meta *m)
941 BDRVQcowState *s = bs->opaque;
942 int l2_index, ret;
943 uint64_t l2_offset, *l2_table, cluster_offset;
944 int nb_clusters, i = 0;
946 ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
947 if (ret == 0)
948 return 0;
950 nb_clusters = size_to_clusters(s, n_end << 9);
952 nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
954 cluster_offset = be64_to_cpu(l2_table[l2_index]);
956 /* We keep all QCOW_OFLAG_COPIED clusters */
958 if (cluster_offset & QCOW_OFLAG_COPIED) {
959 nb_clusters = count_contiguous_clusters(nb_clusters, s->cluster_size,
960 &l2_table[l2_index], 0);
962 cluster_offset &= ~QCOW_OFLAG_COPIED;
963 m->nb_clusters = 0;
965 goto out;
968 /* for the moment, multiple compressed clusters are not managed */
970 if (cluster_offset & QCOW_OFLAG_COMPRESSED)
971 nb_clusters = 1;
973 /* how many available clusters ? */
975 while (i < nb_clusters) {
976 i += count_contiguous_free_clusters(nb_clusters - i,
977 &l2_table[l2_index + i]);
979 cluster_offset = be64_to_cpu(l2_table[l2_index + i]);
981 if ((cluster_offset & QCOW_OFLAG_COPIED) ||
982 (cluster_offset & QCOW_OFLAG_COMPRESSED))
983 break;
985 i += count_contiguous_clusters(nb_clusters - i, s->cluster_size,
986 &l2_table[l2_index + i], 0);
988 if(be64_to_cpu(l2_table[l2_index + i]))
989 break;
991 nb_clusters = i;
993 /* allocate a new cluster */
995 cluster_offset = alloc_clusters(bs, nb_clusters * s->cluster_size);
997 /* save info needed for meta data update */
998 m->offset = offset;
999 m->n_start = n_start;
1000 m->nb_clusters = nb_clusters;
1002 out:
1003 m->nb_available = MIN(nb_clusters << (s->cluster_bits - 9), n_end);
1005 *num = m->nb_available - n_start;
1007 return cluster_offset;
1010 static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
1011 int nb_sectors, int *pnum)
1013 uint64_t cluster_offset;
1015 *pnum = nb_sectors;
1016 cluster_offset = get_cluster_offset(bs, sector_num << 9, pnum);
1018 return (cluster_offset != 0);
1021 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
1022 const uint8_t *buf, int buf_size)
1024 z_stream strm1, *strm = &strm1;
1025 int ret, out_len;
1027 memset(strm, 0, sizeof(*strm));
1029 strm->next_in = (uint8_t *)buf;
1030 strm->avail_in = buf_size;
1031 strm->next_out = out_buf;
1032 strm->avail_out = out_buf_size;
1034 ret = inflateInit2(strm, -12);
1035 if (ret != Z_OK)
1036 return -1;
1037 ret = inflate(strm, Z_FINISH);
1038 out_len = strm->next_out - out_buf;
1039 if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
1040 out_len != out_buf_size) {
1041 inflateEnd(strm);
1042 return -1;
1044 inflateEnd(strm);
1045 return 0;
1048 static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset)
1050 int ret, csize, nb_csectors, sector_offset;
1051 uint64_t coffset;
1053 coffset = cluster_offset & s->cluster_offset_mask;
1054 if (s->cluster_cache_offset != coffset) {
1055 nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
1056 sector_offset = coffset & 511;
1057 csize = nb_csectors * 512 - sector_offset;
1058 ret = bdrv_read(s->hd, coffset >> 9, s->cluster_data, nb_csectors);
1059 if (ret < 0) {
1060 return -1;
1062 if (decompress_buffer(s->cluster_cache, s->cluster_size,
1063 s->cluster_data + sector_offset, csize) < 0) {
1064 return -1;
1066 s->cluster_cache_offset = coffset;
1068 return 0;
1071 /* handle reading after the end of the backing file */
1072 static int backing_read1(BlockDriverState *bs,
1073 int64_t sector_num, uint8_t *buf, int nb_sectors)
1075 int n1;
1076 if ((sector_num + nb_sectors) <= bs->total_sectors)
1077 return nb_sectors;
1078 if (sector_num >= bs->total_sectors)
1079 n1 = 0;
1080 else
1081 n1 = bs->total_sectors - sector_num;
1082 memset(buf + n1 * 512, 0, 512 * (nb_sectors - n1));
1083 return n1;
1086 static int qcow_read(BlockDriverState *bs, int64_t sector_num,
1087 uint8_t *buf, int nb_sectors)
1089 BDRVQcowState *s = bs->opaque;
1090 int ret, index_in_cluster, n, n1;
1091 uint64_t cluster_offset;
1093 while (nb_sectors > 0) {
1094 n = nb_sectors;
1095 cluster_offset = get_cluster_offset(bs, sector_num << 9, &n);
1096 index_in_cluster = sector_num & (s->cluster_sectors - 1);
1097 if (!cluster_offset) {
1098 if (bs->backing_hd) {
1099 /* read from the base image */
1100 n1 = backing_read1(bs->backing_hd, sector_num, buf, n);
1101 if (n1 > 0) {
1102 ret = bdrv_read(bs->backing_hd, sector_num, buf, n1);
1103 if (ret < 0)
1104 return -1;
1106 } else {
1107 memset(buf, 0, 512 * n);
1109 } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
1110 if (decompress_cluster(s, cluster_offset) < 0)
1111 return -1;
1112 memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n);
1113 } else {
1114 ret = bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
1115 if (ret != n * 512)
1116 return -1;
1117 if (s->crypt_method) {
1118 encrypt_sectors(s, sector_num, buf, buf, n, 0,
1119 &s->aes_decrypt_key);
1122 nb_sectors -= n;
1123 sector_num += n;
1124 buf += n * 512;
1126 return 0;
1129 static int qcow_write(BlockDriverState *bs, int64_t sector_num,
1130 const uint8_t *buf, int nb_sectors)
1132 BDRVQcowState *s = bs->opaque;
1133 int ret, index_in_cluster, n;
1134 uint64_t cluster_offset;
1135 int n_end;
1136 QCowL2Meta l2meta;
1138 while (nb_sectors > 0) {
1139 index_in_cluster = sector_num & (s->cluster_sectors - 1);
1140 n_end = index_in_cluster + nb_sectors;
1141 if (s->crypt_method &&
1142 n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
1143 n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
1144 cluster_offset = alloc_cluster_offset(bs, sector_num << 9,
1145 index_in_cluster,
1146 n_end, &n, &l2meta);
1147 if (!cluster_offset)
1148 return -1;
1149 if (s->crypt_method) {
1150 encrypt_sectors(s, sector_num, s->cluster_data, buf, n, 1,
1151 &s->aes_encrypt_key);
1152 ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512,
1153 s->cluster_data, n * 512);
1154 } else {
1155 ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
1157 if (ret != n * 512 || alloc_cluster_link_l2(bs, cluster_offset, &l2meta) < 0) {
1158 free_any_clusters(bs, cluster_offset, l2meta.nb_clusters);
1159 return -1;
1161 nb_sectors -= n;
1162 sector_num += n;
1163 buf += n * 512;
1165 s->cluster_cache_offset = -1; /* disable compressed cache */
1166 return 0;
1169 typedef struct QCowAIOCB {
1170 BlockDriverAIOCB common;
1171 int64_t sector_num;
1172 uint8_t *buf;
1173 int nb_sectors;
1174 int n;
1175 uint64_t cluster_offset;
1176 uint8_t *cluster_data;
1177 BlockDriverAIOCB *hd_aiocb;
1178 QEMUBH *bh;
1179 QCowL2Meta l2meta;
1180 } QCowAIOCB;
1182 static void qcow_aio_read_cb(void *opaque, int ret);
1183 static void qcow_aio_read_bh(void *opaque)
1185 QCowAIOCB *acb = opaque;
1186 qemu_bh_delete(acb->bh);
1187 acb->bh = NULL;
1188 qcow_aio_read_cb(opaque, 0);
1191 static int qcow_schedule_bh(QEMUBHFunc *cb, QCowAIOCB *acb)
1193 if (acb->bh)
1194 return -EIO;
1196 acb->bh = qemu_bh_new(cb, acb);
1197 if (!acb->bh)
1198 return -EIO;
1200 qemu_bh_schedule(acb->bh);
1202 return 0;
1205 static void qcow_aio_read_cb(void *opaque, int ret)
1207 QCowAIOCB *acb = opaque;
1208 BlockDriverState *bs = acb->common.bs;
1209 BDRVQcowState *s = bs->opaque;
1210 int index_in_cluster, n1;
1212 acb->hd_aiocb = NULL;
1213 if (ret < 0) {
1214 fail:
1215 acb->common.cb(acb->common.opaque, ret);
1216 qemu_aio_release(acb);
1217 return;
1220 /* post process the read buffer */
1221 if (!acb->cluster_offset) {
1222 /* nothing to do */
1223 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
1224 /* nothing to do */
1225 } else {
1226 if (s->crypt_method) {
1227 encrypt_sectors(s, acb->sector_num, acb->buf, acb->buf,
1228 acb->n, 0,
1229 &s->aes_decrypt_key);
1233 acb->nb_sectors -= acb->n;
1234 acb->sector_num += acb->n;
1235 acb->buf += acb->n * 512;
1237 if (acb->nb_sectors == 0) {
1238 /* request completed */
1239 acb->common.cb(acb->common.opaque, 0);
1240 qemu_aio_release(acb);
1241 return;
1244 /* prepare next AIO request */
1245 acb->n = acb->nb_sectors;
1246 acb->cluster_offset = get_cluster_offset(bs, acb->sector_num << 9, &acb->n);
1247 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
1249 if (!acb->cluster_offset) {
1250 if (bs->backing_hd) {
1251 /* read from the base image */
1252 n1 = backing_read1(bs->backing_hd, acb->sector_num,
1253 acb->buf, acb->n);
1254 if (n1 > 0) {
1255 acb->hd_aiocb = bdrv_aio_read(bs->backing_hd, acb->sector_num,
1256 acb->buf, acb->n, qcow_aio_read_cb, acb);
1257 if (acb->hd_aiocb == NULL)
1258 goto fail;
1259 } else {
1260 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
1261 if (ret < 0)
1262 goto fail;
1264 } else {
1265 /* Note: in this case, no need to wait */
1266 memset(acb->buf, 0, 512 * acb->n);
1267 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
1268 if (ret < 0)
1269 goto fail;
1271 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
1272 /* add AIO support for compressed blocks ? */
1273 if (decompress_cluster(s, acb->cluster_offset) < 0)
1274 goto fail;
1275 memcpy(acb->buf,
1276 s->cluster_cache + index_in_cluster * 512, 512 * acb->n);
1277 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
1278 if (ret < 0)
1279 goto fail;
1280 } else {
1281 if ((acb->cluster_offset & 511) != 0) {
1282 ret = -EIO;
1283 goto fail;
1285 acb->hd_aiocb = bdrv_aio_read(s->hd,
1286 (acb->cluster_offset >> 9) + index_in_cluster,
1287 acb->buf, acb->n, qcow_aio_read_cb, acb);
1288 if (acb->hd_aiocb == NULL)
1289 goto fail;
1293 static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs,
1294 int64_t sector_num, uint8_t *buf, int nb_sectors,
1295 BlockDriverCompletionFunc *cb, void *opaque)
1297 QCowAIOCB *acb;
1299 acb = qemu_aio_get(bs, cb, opaque);
1300 if (!acb)
1301 return NULL;
1302 acb->hd_aiocb = NULL;
1303 acb->sector_num = sector_num;
1304 acb->buf = buf;
1305 acb->nb_sectors = nb_sectors;
1306 acb->n = 0;
1307 acb->cluster_offset = 0;
1308 acb->l2meta.nb_clusters = 0;
1309 return acb;
1312 static BlockDriverAIOCB *qcow_aio_read(BlockDriverState *bs,
1313 int64_t sector_num, uint8_t *buf, int nb_sectors,
1314 BlockDriverCompletionFunc *cb, void *opaque)
1316 QCowAIOCB *acb;
1318 acb = qcow_aio_setup(bs, sector_num, buf, nb_sectors, cb, opaque);
1319 if (!acb)
1320 return NULL;
1322 qcow_aio_read_cb(acb, 0);
1323 return &acb->common;
1326 static void qcow_aio_write_cb(void *opaque, int ret)
1328 QCowAIOCB *acb = opaque;
1329 BlockDriverState *bs = acb->common.bs;
1330 BDRVQcowState *s = bs->opaque;
1331 int index_in_cluster;
1332 const uint8_t *src_buf;
1333 int n_end;
1335 acb->hd_aiocb = NULL;
1337 if (ret < 0) {
1338 fail:
1339 acb->common.cb(acb->common.opaque, ret);
1340 qemu_aio_release(acb);
1341 return;
1344 if (alloc_cluster_link_l2(bs, acb->cluster_offset, &acb->l2meta) < 0) {
1345 free_any_clusters(bs, acb->cluster_offset, acb->l2meta.nb_clusters);
1346 goto fail;
1349 acb->nb_sectors -= acb->n;
1350 acb->sector_num += acb->n;
1351 acb->buf += acb->n * 512;
1353 if (acb->nb_sectors == 0) {
1354 /* request completed */
1355 acb->common.cb(acb->common.opaque, 0);
1356 qemu_aio_release(acb);
1357 return;
1360 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
1361 n_end = index_in_cluster + acb->nb_sectors;
1362 if (s->crypt_method &&
1363 n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
1364 n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
1366 acb->cluster_offset = alloc_cluster_offset(bs, acb->sector_num << 9,
1367 index_in_cluster,
1368 n_end, &acb->n, &acb->l2meta);
1369 if (!acb->cluster_offset || (acb->cluster_offset & 511) != 0) {
1370 ret = -EIO;
1371 goto fail;
1373 if (s->crypt_method) {
1374 if (!acb->cluster_data) {
1375 acb->cluster_data = qemu_mallocz(QCOW_MAX_CRYPT_CLUSTERS *
1376 s->cluster_size);
1377 if (!acb->cluster_data) {
1378 ret = -ENOMEM;
1379 goto fail;
1382 encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf,
1383 acb->n, 1, &s->aes_encrypt_key);
1384 src_buf = acb->cluster_data;
1385 } else {
1386 src_buf = acb->buf;
1388 acb->hd_aiocb = bdrv_aio_write(s->hd,
1389 (acb->cluster_offset >> 9) + index_in_cluster,
1390 src_buf, acb->n,
1391 qcow_aio_write_cb, acb);
1392 if (acb->hd_aiocb == NULL)
1393 goto fail;
1396 static BlockDriverAIOCB *qcow_aio_write(BlockDriverState *bs,
1397 int64_t sector_num, const uint8_t *buf, int nb_sectors,
1398 BlockDriverCompletionFunc *cb, void *opaque)
1400 BDRVQcowState *s = bs->opaque;
1401 QCowAIOCB *acb;
1403 s->cluster_cache_offset = -1; /* disable compressed cache */
1405 acb = qcow_aio_setup(bs, sector_num, (uint8_t*)buf, nb_sectors, cb, opaque);
1406 if (!acb)
1407 return NULL;
1409 qcow_aio_write_cb(acb, 0);
1410 return &acb->common;
1413 static void qcow_aio_cancel(BlockDriverAIOCB *blockacb)
1415 QCowAIOCB *acb = (QCowAIOCB *)blockacb;
1416 if (acb->hd_aiocb)
1417 bdrv_aio_cancel(acb->hd_aiocb);
1418 qemu_aio_release(acb);
1421 static void qcow_close(BlockDriverState *bs)
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);
1432 /* XXX: use std qcow open function ? */
1433 typedef 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;
1443 static void create_refcount_update(QCowCreateState *s,
1444 int64_t offset, int64_t size)
1446 int refcount;
1447 int64_t start, last, cluster_offset;
1448 uint16_t *p;
1450 start = offset & ~(s->cluster_size - 1);
1451 last = (offset + size - 1) & ~(s->cluster_size - 1);
1452 for(cluster_offset = start; cluster_offset <= last;
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);
1461 static int qcow_create(const char *filename, int64_t total_size,
1462 const char *backing_file, int flags)
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;
1469 memset(s, 0, sizeof(*s));
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;
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;
1490 if (flags & BLOCK_FLAG_ENCRYPT) {
1491 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
1492 } else {
1493 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
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);
1502 offset += align_offset(l1_size * sizeof(uint64_t), s->cluster_size);
1504 s->refcount_table = qemu_mallocz(s->cluster_size);
1505 if (!s->refcount_table)
1506 goto fail;
1507 s->refcount_block = qemu_mallocz(s->cluster_size);
1508 if (!s->refcount_block)
1509 goto fail;
1511 s->refcount_table_offset = offset;
1512 header.refcount_table_offset = cpu_to_be64(offset);
1513 header.refcount_table_clusters = cpu_to_be32(1);
1514 offset += s->cluster_size;
1516 s->refcount_table[0] = cpu_to_be64(offset);
1517 s->refcount_block_offset = offset;
1518 offset += s->cluster_size;
1520 /* update refcounts */
1521 create_refcount_update(s, 0, header_size);
1522 create_refcount_update(s, s->l1_table_offset, l1_size * sizeof(uint64_t));
1523 create_refcount_update(s, s->refcount_table_offset, s->cluster_size);
1524 create_refcount_update(s, s->refcount_block_offset, s->cluster_size);
1526 /* write all the data */
1527 write(fd, &header, sizeof(header));
1528 if (backing_file) {
1529 write(fd, backing_file, backing_filename_len);
1531 lseek(fd, s->l1_table_offset, SEEK_SET);
1532 tmp = 0;
1533 for(i = 0;i < l1_size; i++) {
1534 write(fd, &tmp, sizeof(tmp));
1536 lseek(fd, s->refcount_table_offset, SEEK_SET);
1537 write(fd, s->refcount_table, s->cluster_size);
1539 lseek(fd, s->refcount_block_offset, SEEK_SET);
1540 write(fd, s->refcount_block, s->cluster_size);
1542 qemu_free(s->refcount_table);
1543 qemu_free(s->refcount_block);
1544 close(fd);
1545 return 0;
1546 fail:
1547 qemu_free(s->refcount_table);
1548 qemu_free(s->refcount_block);
1549 close(fd);
1550 return -ENOMEM;
1553 static int qcow_make_empty(BlockDriverState *bs)
1555 #if 0
1556 /* XXX: not correct */
1557 BDRVQcowState *s = bs->opaque;
1558 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1559 int ret;
1561 memset(s->l1_table, 0, l1_length);
1562 if (bdrv_pwrite(s->hd, s->l1_table_offset, s->l1_table, l1_length) < 0)
1563 return -1;
1564 ret = bdrv_truncate(s->hd, s->l1_table_offset + l1_length);
1565 if (ret < 0)
1566 return ret;
1568 l2_cache_reset(bs);
1569 #endif
1570 return 0;
1573 /* XXX: put compressed sectors first, then all the cluster aligned
1574 tables to avoid losing bytes in alignment */
1575 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
1576 const uint8_t *buf, int nb_sectors)
1578 BDRVQcowState *s = bs->opaque;
1579 z_stream strm;
1580 int ret, out_len;
1581 uint8_t *out_buf;
1582 uint64_t cluster_offset;
1584 if (nb_sectors == 0) {
1585 /* align end of file to a sector boundary to ease reading with
1586 sector based I/Os */
1587 cluster_offset = bdrv_getlength(s->hd);
1588 cluster_offset = (cluster_offset + 511) & ~511;
1589 bdrv_truncate(s->hd, cluster_offset);
1590 return 0;
1593 if (nb_sectors != s->cluster_sectors)
1594 return -EINVAL;
1596 out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
1597 if (!out_buf)
1598 return -ENOMEM;
1600 /* best compression, small window, no zlib header */
1601 memset(&strm, 0, sizeof(strm));
1602 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1603 Z_DEFLATED, -12,
1604 9, Z_DEFAULT_STRATEGY);
1605 if (ret != 0) {
1606 qemu_free(out_buf);
1607 return -1;
1610 strm.avail_in = s->cluster_size;
1611 strm.next_in = (uint8_t *)buf;
1612 strm.avail_out = s->cluster_size;
1613 strm.next_out = out_buf;
1615 ret = deflate(&strm, Z_FINISH);
1616 if (ret != Z_STREAM_END && ret != Z_OK) {
1617 qemu_free(out_buf);
1618 deflateEnd(&strm);
1619 return -1;
1621 out_len = strm.next_out - out_buf;
1623 deflateEnd(&strm);
1625 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1626 /* could not compress: write normal cluster */
1627 qcow_write(bs, sector_num, buf, s->cluster_sectors);
1628 } else {
1629 cluster_offset = alloc_compressed_cluster_offset(bs, sector_num << 9,
1630 out_len);
1631 if (!cluster_offset)
1632 return -1;
1633 cluster_offset &= s->cluster_offset_mask;
1634 if (bdrv_pwrite(s->hd, cluster_offset, out_buf, out_len) != out_len) {
1635 qemu_free(out_buf);
1636 return -1;
1640 qemu_free(out_buf);
1641 return 0;
1644 static void qcow_flush(BlockDriverState *bs)
1646 BDRVQcowState *s = bs->opaque;
1647 bdrv_flush(s->hd);
1650 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1652 BDRVQcowState *s = bs->opaque;
1653 bdi->cluster_size = s->cluster_size;
1654 bdi->vm_state_offset = (int64_t)s->l1_vm_state_index <<
1655 (s->cluster_bits + s->l2_bits);
1656 return 0;
1659 /*********************************************************/
1660 /* snapshot support */
1662 /* update the refcounts of snapshots and the copied flag */
1663 static int update_snapshot_refcount(BlockDriverState *bs,
1664 int64_t l1_table_offset,
1665 int l1_size,
1666 int addend)
1668 BDRVQcowState *s = bs->opaque;
1669 uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated;
1670 int64_t old_offset, old_l2_offset;
1671 int l2_size, i, j, l1_modified, l2_modified, nb_csectors, refcount;
1673 l2_cache_reset(bs);
1675 l2_table = NULL;
1676 l1_table = NULL;
1677 l1_size2 = l1_size * sizeof(uint64_t);
1678 l1_allocated = 0;
1679 if (l1_table_offset != s->l1_table_offset) {
1680 l1_table = qemu_malloc(l1_size2);
1681 if (!l1_table)
1682 goto fail;
1683 l1_allocated = 1;
1684 if (bdrv_pread(s->hd, l1_table_offset,
1685 l1_table, l1_size2) != l1_size2)
1686 goto fail;
1687 for(i = 0;i < l1_size; i++)
1688 be64_to_cpus(&l1_table[i]);
1689 } else {
1690 assert(l1_size == s->l1_size);
1691 l1_table = s->l1_table;
1692 l1_allocated = 0;
1695 l2_size = s->l2_size * sizeof(uint64_t);
1696 l2_table = qemu_malloc(l2_size);
1697 if (!l2_table)
1698 goto fail;
1699 l1_modified = 0;
1700 for(i = 0; i < l1_size; i++) {
1701 l2_offset = l1_table[i];
1702 if (l2_offset) {
1703 old_l2_offset = l2_offset;
1704 l2_offset &= ~QCOW_OFLAG_COPIED;
1705 l2_modified = 0;
1706 if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
1707 goto fail;
1708 for(j = 0; j < s->l2_size; j++) {
1709 offset = be64_to_cpu(l2_table[j]);
1710 if (offset != 0) {
1711 old_offset = offset;
1712 offset &= ~QCOW_OFLAG_COPIED;
1713 if (offset & QCOW_OFLAG_COMPRESSED) {
1714 nb_csectors = ((offset >> s->csize_shift) &
1715 s->csize_mask) + 1;
1716 if (addend != 0)
1717 update_refcount(bs, (offset & s->cluster_offset_mask) & ~511,
1718 nb_csectors * 512, addend);
1719 /* compressed clusters are never modified */
1720 refcount = 2;
1721 } else {
1722 if (addend != 0) {
1723 refcount = update_cluster_refcount(bs, offset >> s->cluster_bits, addend);
1724 } else {
1725 refcount = get_refcount(bs, offset >> s->cluster_bits);
1729 if (refcount == 1) {
1730 offset |= QCOW_OFLAG_COPIED;
1732 if (offset != old_offset) {
1733 l2_table[j] = cpu_to_be64(offset);
1734 l2_modified = 1;
1738 if (l2_modified) {
1739 if (bdrv_pwrite(s->hd,
1740 l2_offset, l2_table, l2_size) != l2_size)
1741 goto fail;
1744 if (addend != 0) {
1745 refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend);
1746 } else {
1747 refcount = get_refcount(bs, l2_offset >> s->cluster_bits);
1749 if (refcount == 1) {
1750 l2_offset |= QCOW_OFLAG_COPIED;
1752 if (l2_offset != old_l2_offset) {
1753 l1_table[i] = l2_offset;
1754 l1_modified = 1;
1758 if (l1_modified) {
1759 for(i = 0; i < l1_size; i++)
1760 cpu_to_be64s(&l1_table[i]);
1761 if (bdrv_pwrite(s->hd, l1_table_offset, l1_table,
1762 l1_size2) != l1_size2)
1763 goto fail;
1764 for(i = 0; i < l1_size; i++)
1765 be64_to_cpus(&l1_table[i]);
1767 if (l1_allocated)
1768 qemu_free(l1_table);
1769 qemu_free(l2_table);
1770 return 0;
1771 fail:
1772 if (l1_allocated)
1773 qemu_free(l1_table);
1774 qemu_free(l2_table);
1775 return -EIO;
1778 static void qcow_free_snapshots(BlockDriverState *bs)
1780 BDRVQcowState *s = bs->opaque;
1781 int i;
1783 for(i = 0; i < s->nb_snapshots; i++) {
1784 qemu_free(s->snapshots[i].name);
1785 qemu_free(s->snapshots[i].id_str);
1787 qemu_free(s->snapshots);
1788 s->snapshots = NULL;
1789 s->nb_snapshots = 0;
1792 static int qcow_read_snapshots(BlockDriverState *bs)
1794 BDRVQcowState *s = bs->opaque;
1795 QCowSnapshotHeader h;
1796 QCowSnapshot *sn;
1797 int i, id_str_size, name_size;
1798 int64_t offset;
1799 uint32_t extra_data_size;
1801 offset = s->snapshots_offset;
1802 s->snapshots = qemu_mallocz(s->nb_snapshots * sizeof(QCowSnapshot));
1803 if (!s->snapshots)
1804 goto fail;
1805 for(i = 0; i < s->nb_snapshots; i++) {
1806 offset = align_offset(offset, 8);
1807 if (bdrv_pread(s->hd, offset, &h, sizeof(h)) != sizeof(h))
1808 goto fail;
1809 offset += sizeof(h);
1810 sn = s->snapshots + i;
1811 sn->l1_table_offset = be64_to_cpu(h.l1_table_offset);
1812 sn->l1_size = be32_to_cpu(h.l1_size);
1813 sn->vm_state_size = be32_to_cpu(h.vm_state_size);
1814 sn->date_sec = be32_to_cpu(h.date_sec);
1815 sn->date_nsec = be32_to_cpu(h.date_nsec);
1816 sn->vm_clock_nsec = be64_to_cpu(h.vm_clock_nsec);
1817 extra_data_size = be32_to_cpu(h.extra_data_size);
1819 id_str_size = be16_to_cpu(h.id_str_size);
1820 name_size = be16_to_cpu(h.name_size);
1822 offset += extra_data_size;
1824 sn->id_str = qemu_malloc(id_str_size + 1);
1825 if (!sn->id_str)
1826 goto fail;
1827 if (bdrv_pread(s->hd, offset, sn->id_str, id_str_size) != id_str_size)
1828 goto fail;
1829 offset += id_str_size;
1830 sn->id_str[id_str_size] = '\0';
1832 sn->name = qemu_malloc(name_size + 1);
1833 if (!sn->name)
1834 goto fail;
1835 if (bdrv_pread(s->hd, offset, sn->name, name_size) != name_size)
1836 goto fail;
1837 offset += name_size;
1838 sn->name[name_size] = '\0';
1840 s->snapshots_size = offset - s->snapshots_offset;
1841 return 0;
1842 fail:
1843 qcow_free_snapshots(bs);
1844 return -1;
1847 /* add at the end of the file a new list of snapshots */
1848 static int qcow_write_snapshots(BlockDriverState *bs)
1850 BDRVQcowState *s = bs->opaque;
1851 QCowSnapshot *sn;
1852 QCowSnapshotHeader h;
1853 int i, name_size, id_str_size, snapshots_size;
1854 uint64_t data64;
1855 uint32_t data32;
1856 int64_t offset, snapshots_offset;
1858 /* compute the size of the snapshots */
1859 offset = 0;
1860 for(i = 0; i < s->nb_snapshots; i++) {
1861 sn = s->snapshots + i;
1862 offset = align_offset(offset, 8);
1863 offset += sizeof(h);
1864 offset += strlen(sn->id_str);
1865 offset += strlen(sn->name);
1867 snapshots_size = offset;
1869 snapshots_offset = alloc_clusters(bs, snapshots_size);
1870 offset = snapshots_offset;
1872 for(i = 0; i < s->nb_snapshots; i++) {
1873 sn = s->snapshots + i;
1874 memset(&h, 0, sizeof(h));
1875 h.l1_table_offset = cpu_to_be64(sn->l1_table_offset);
1876 h.l1_size = cpu_to_be32(sn->l1_size);
1877 h.vm_state_size = cpu_to_be32(sn->vm_state_size);
1878 h.date_sec = cpu_to_be32(sn->date_sec);
1879 h.date_nsec = cpu_to_be32(sn->date_nsec);
1880 h.vm_clock_nsec = cpu_to_be64(sn->vm_clock_nsec);
1882 id_str_size = strlen(sn->id_str);
1883 name_size = strlen(sn->name);
1884 h.id_str_size = cpu_to_be16(id_str_size);
1885 h.name_size = cpu_to_be16(name_size);
1886 offset = align_offset(offset, 8);
1887 if (bdrv_pwrite(s->hd, offset, &h, sizeof(h)) != sizeof(h))
1888 goto fail;
1889 offset += sizeof(h);
1890 if (bdrv_pwrite(s->hd, offset, sn->id_str, id_str_size) != id_str_size)
1891 goto fail;
1892 offset += id_str_size;
1893 if (bdrv_pwrite(s->hd, offset, sn->name, name_size) != name_size)
1894 goto fail;
1895 offset += name_size;
1898 /* update the various header fields */
1899 data64 = cpu_to_be64(snapshots_offset);
1900 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, snapshots_offset),
1901 &data64, sizeof(data64)) != sizeof(data64))
1902 goto fail;
1903 data32 = cpu_to_be32(s->nb_snapshots);
1904 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, nb_snapshots),
1905 &data32, sizeof(data32)) != sizeof(data32))
1906 goto fail;
1908 /* free the old snapshot table */
1909 free_clusters(bs, s->snapshots_offset, s->snapshots_size);
1910 s->snapshots_offset = snapshots_offset;
1911 s->snapshots_size = snapshots_size;
1912 return 0;
1913 fail:
1914 return -1;
1917 static void find_new_snapshot_id(BlockDriverState *bs,
1918 char *id_str, int id_str_size)
1920 BDRVQcowState *s = bs->opaque;
1921 QCowSnapshot *sn;
1922 int i, id, id_max = 0;
1924 for(i = 0; i < s->nb_snapshots; i++) {
1925 sn = s->snapshots + i;
1926 id = strtoul(sn->id_str, NULL, 10);
1927 if (id > id_max)
1928 id_max = id;
1930 snprintf(id_str, id_str_size, "%d", id_max + 1);
1933 static int find_snapshot_by_id(BlockDriverState *bs, const char *id_str)
1935 BDRVQcowState *s = bs->opaque;
1936 int i;
1938 for(i = 0; i < s->nb_snapshots; i++) {
1939 if (!strcmp(s->snapshots[i].id_str, id_str))
1940 return i;
1942 return -1;
1945 static int find_snapshot_by_id_or_name(BlockDriverState *bs, const char *name)
1947 BDRVQcowState *s = bs->opaque;
1948 int i, ret;
1950 ret = find_snapshot_by_id(bs, name);
1951 if (ret >= 0)
1952 return ret;
1953 for(i = 0; i < s->nb_snapshots; i++) {
1954 if (!strcmp(s->snapshots[i].name, name))
1955 return i;
1957 return -1;
1960 /* if no id is provided, a new one is constructed */
1961 static int qcow_snapshot_create(BlockDriverState *bs,
1962 QEMUSnapshotInfo *sn_info)
1964 BDRVQcowState *s = bs->opaque;
1965 QCowSnapshot *snapshots1, sn1, *sn = &sn1;
1966 int i, ret;
1967 uint64_t *l1_table = NULL;
1969 memset(sn, 0, sizeof(*sn));
1971 if (sn_info->id_str[0] == '\0') {
1972 /* compute a new id */
1973 find_new_snapshot_id(bs, sn_info->id_str, sizeof(sn_info->id_str));
1976 /* check that the ID is unique */
1977 if (find_snapshot_by_id(bs, sn_info->id_str) >= 0)
1978 return -ENOENT;
1980 sn->id_str = qemu_strdup(sn_info->id_str);
1981 if (!sn->id_str)
1982 goto fail;
1983 sn->name = qemu_strdup(sn_info->name);
1984 if (!sn->name)
1985 goto fail;
1986 sn->vm_state_size = sn_info->vm_state_size;
1987 sn->date_sec = sn_info->date_sec;
1988 sn->date_nsec = sn_info->date_nsec;
1989 sn->vm_clock_nsec = sn_info->vm_clock_nsec;
1991 ret = update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1);
1992 if (ret < 0)
1993 goto fail;
1995 /* create the L1 table of the snapshot */
1996 sn->l1_table_offset = alloc_clusters(bs, s->l1_size * sizeof(uint64_t));
1997 sn->l1_size = s->l1_size;
1999 l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
2000 if (!l1_table)
2001 goto fail;
2002 for(i = 0; i < s->l1_size; i++) {
2003 l1_table[i] = cpu_to_be64(s->l1_table[i]);
2005 if (bdrv_pwrite(s->hd, sn->l1_table_offset,
2006 l1_table, s->l1_size * sizeof(uint64_t)) !=
2007 (s->l1_size * sizeof(uint64_t)))
2008 goto fail;
2009 qemu_free(l1_table);
2010 l1_table = NULL;
2012 snapshots1 = qemu_malloc((s->nb_snapshots + 1) * sizeof(QCowSnapshot));
2013 if (!snapshots1)
2014 goto fail;
2015 memcpy(snapshots1, s->snapshots, s->nb_snapshots * sizeof(QCowSnapshot));
2016 s->snapshots = snapshots1;
2017 s->snapshots[s->nb_snapshots++] = *sn;
2019 if (qcow_write_snapshots(bs) < 0)
2020 goto fail;
2021 #ifdef DEBUG_ALLOC
2022 check_refcounts(bs);
2023 #endif
2024 return 0;
2025 fail:
2026 qemu_free(sn->name);
2027 qemu_free(l1_table);
2028 return -1;
2031 /* copy the snapshot 'snapshot_name' into the current disk image */
2032 static int qcow_snapshot_goto(BlockDriverState *bs,
2033 const char *snapshot_id)
2035 BDRVQcowState *s = bs->opaque;
2036 QCowSnapshot *sn;
2037 int i, snapshot_index, l1_size2;
2039 snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
2040 if (snapshot_index < 0)
2041 return -ENOENT;
2042 sn = &s->snapshots[snapshot_index];
2044 if (update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, -1) < 0)
2045 goto fail;
2047 if (grow_l1_table(bs, sn->l1_size) < 0)
2048 goto fail;
2050 s->l1_size = sn->l1_size;
2051 l1_size2 = s->l1_size * sizeof(uint64_t);
2052 /* copy the snapshot l1 table to the current l1 table */
2053 if (bdrv_pread(s->hd, sn->l1_table_offset,
2054 s->l1_table, l1_size2) != l1_size2)
2055 goto fail;
2056 if (bdrv_pwrite(s->hd, s->l1_table_offset,
2057 s->l1_table, l1_size2) != l1_size2)
2058 goto fail;
2059 for(i = 0;i < s->l1_size; i++) {
2060 be64_to_cpus(&s->l1_table[i]);
2063 if (update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1) < 0)
2064 goto fail;
2066 #ifdef DEBUG_ALLOC
2067 check_refcounts(bs);
2068 #endif
2069 return 0;
2070 fail:
2071 return -EIO;
2074 static int qcow_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
2076 BDRVQcowState *s = bs->opaque;
2077 QCowSnapshot *sn;
2078 int snapshot_index, ret;
2080 snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
2081 if (snapshot_index < 0)
2082 return -ENOENT;
2083 sn = &s->snapshots[snapshot_index];
2085 ret = update_snapshot_refcount(bs, sn->l1_table_offset, sn->l1_size, -1);
2086 if (ret < 0)
2087 return ret;
2088 /* must update the copied flag on the current cluster offsets */
2089 ret = update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 0);
2090 if (ret < 0)
2091 return ret;
2092 free_clusters(bs, sn->l1_table_offset, sn->l1_size * sizeof(uint64_t));
2094 qemu_free(sn->id_str);
2095 qemu_free(sn->name);
2096 memmove(sn, sn + 1, (s->nb_snapshots - snapshot_index - 1) * sizeof(*sn));
2097 s->nb_snapshots--;
2098 ret = qcow_write_snapshots(bs);
2099 if (ret < 0) {
2100 /* XXX: restore snapshot if error ? */
2101 return ret;
2103 #ifdef DEBUG_ALLOC
2104 check_refcounts(bs);
2105 #endif
2106 return 0;
2109 static int qcow_snapshot_list(BlockDriverState *bs,
2110 QEMUSnapshotInfo **psn_tab)
2112 BDRVQcowState *s = bs->opaque;
2113 QEMUSnapshotInfo *sn_tab, *sn_info;
2114 QCowSnapshot *sn;
2115 int i;
2117 sn_tab = qemu_mallocz(s->nb_snapshots * sizeof(QEMUSnapshotInfo));
2118 if (!sn_tab)
2119 goto fail;
2120 for(i = 0; i < s->nb_snapshots; i++) {
2121 sn_info = sn_tab + i;
2122 sn = s->snapshots + i;
2123 pstrcpy(sn_info->id_str, sizeof(sn_info->id_str),
2124 sn->id_str);
2125 pstrcpy(sn_info->name, sizeof(sn_info->name),
2126 sn->name);
2127 sn_info->vm_state_size = sn->vm_state_size;
2128 sn_info->date_sec = sn->date_sec;
2129 sn_info->date_nsec = sn->date_nsec;
2130 sn_info->vm_clock_nsec = sn->vm_clock_nsec;
2132 *psn_tab = sn_tab;
2133 return s->nb_snapshots;
2134 fail:
2135 qemu_free(sn_tab);
2136 *psn_tab = NULL;
2137 return -ENOMEM;
2140 /*********************************************************/
2141 /* refcount handling */
2143 static int refcount_init(BlockDriverState *bs)
2145 BDRVQcowState *s = bs->opaque;
2146 int ret, refcount_table_size2, i;
2148 s->refcount_block_cache = qemu_malloc(s->cluster_size);
2149 if (!s->refcount_block_cache)
2150 goto fail;
2151 refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t);
2152 s->refcount_table = qemu_malloc(refcount_table_size2);
2153 if (!s->refcount_table)
2154 goto fail;
2155 if (s->refcount_table_size > 0) {
2156 ret = bdrv_pread(s->hd, s->refcount_table_offset,
2157 s->refcount_table, refcount_table_size2);
2158 if (ret != refcount_table_size2)
2159 goto fail;
2160 for(i = 0; i < s->refcount_table_size; i++)
2161 be64_to_cpus(&s->refcount_table[i]);
2163 return 0;
2164 fail:
2165 return -ENOMEM;
2168 static void refcount_close(BlockDriverState *bs)
2170 BDRVQcowState *s = bs->opaque;
2171 qemu_free(s->refcount_block_cache);
2172 qemu_free(s->refcount_table);
2176 static int load_refcount_block(BlockDriverState *bs,
2177 int64_t refcount_block_offset)
2179 BDRVQcowState *s = bs->opaque;
2180 int ret;
2181 ret = bdrv_pread(s->hd, refcount_block_offset, s->refcount_block_cache,
2182 s->cluster_size);
2183 if (ret != s->cluster_size)
2184 return -EIO;
2185 s->refcount_block_cache_offset = refcount_block_offset;
2186 return 0;
2189 static int get_refcount(BlockDriverState *bs, int64_t cluster_index)
2191 BDRVQcowState *s = bs->opaque;
2192 int refcount_table_index, block_index;
2193 int64_t refcount_block_offset;
2195 refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
2196 if (refcount_table_index >= s->refcount_table_size)
2197 return 0;
2198 refcount_block_offset = s->refcount_table[refcount_table_index];
2199 if (!refcount_block_offset)
2200 return 0;
2201 if (refcount_block_offset != s->refcount_block_cache_offset) {
2202 /* better than nothing: return allocated if read error */
2203 if (load_refcount_block(bs, refcount_block_offset) < 0)
2204 return 1;
2206 block_index = cluster_index &
2207 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
2208 return be16_to_cpu(s->refcount_block_cache[block_index]);
2211 /* return < 0 if error */
2212 static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size)
2214 BDRVQcowState *s = bs->opaque;
2215 int i, nb_clusters;
2217 nb_clusters = size_to_clusters(s, size);
2218 retry:
2219 for(i = 0; i < nb_clusters; i++) {
2220 int64_t i = s->free_cluster_index++;
2221 if (get_refcount(bs, i) != 0)
2222 goto retry;
2224 #ifdef DEBUG_ALLOC2
2225 printf("alloc_clusters: size=%lld -> %lld\n",
2226 size,
2227 (s->free_cluster_index - nb_clusters) << s->cluster_bits);
2228 #endif
2229 return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
2232 static int64_t alloc_clusters(BlockDriverState *bs, int64_t size)
2234 int64_t offset;
2236 offset = alloc_clusters_noref(bs, size);
2237 update_refcount(bs, offset, size, 1);
2238 return offset;
2241 /* only used to allocate compressed sectors. We try to allocate
2242 contiguous sectors. size must be <= cluster_size */
2243 static int64_t alloc_bytes(BlockDriverState *bs, int size)
2245 BDRVQcowState *s = bs->opaque;
2246 int64_t offset, cluster_offset;
2247 int free_in_cluster;
2249 assert(size > 0 && size <= s->cluster_size);
2250 if (s->free_byte_offset == 0) {
2251 s->free_byte_offset = alloc_clusters(bs, s->cluster_size);
2253 redo:
2254 free_in_cluster = s->cluster_size -
2255 (s->free_byte_offset & (s->cluster_size - 1));
2256 if (size <= free_in_cluster) {
2257 /* enough space in current cluster */
2258 offset = s->free_byte_offset;
2259 s->free_byte_offset += size;
2260 free_in_cluster -= size;
2261 if (free_in_cluster == 0)
2262 s->free_byte_offset = 0;
2263 if ((offset & (s->cluster_size - 1)) != 0)
2264 update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
2265 } else {
2266 offset = alloc_clusters(bs, s->cluster_size);
2267 cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1);
2268 if ((cluster_offset + s->cluster_size) == offset) {
2269 /* we are lucky: contiguous data */
2270 offset = s->free_byte_offset;
2271 update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
2272 s->free_byte_offset += size;
2273 } else {
2274 s->free_byte_offset = offset;
2275 goto redo;
2278 return offset;
2281 static void free_clusters(BlockDriverState *bs,
2282 int64_t offset, int64_t size)
2284 update_refcount(bs, offset, size, -1);
2287 static int grow_refcount_table(BlockDriverState *bs, int min_size)
2289 BDRVQcowState *s = bs->opaque;
2290 int new_table_size, new_table_size2, refcount_table_clusters, i, ret;
2291 uint64_t *new_table;
2292 int64_t table_offset;
2293 uint8_t data[12];
2294 int old_table_size;
2295 int64_t old_table_offset;
2297 if (min_size <= s->refcount_table_size)
2298 return 0;
2299 /* compute new table size */
2300 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
2301 for(;;) {
2302 if (refcount_table_clusters == 0) {
2303 refcount_table_clusters = 1;
2304 } else {
2305 refcount_table_clusters = (refcount_table_clusters * 3 + 1) / 2;
2307 new_table_size = refcount_table_clusters << (s->cluster_bits - 3);
2308 if (min_size <= new_table_size)
2309 break;
2311 #ifdef DEBUG_ALLOC2
2312 printf("grow_refcount_table from %d to %d\n",
2313 s->refcount_table_size,
2314 new_table_size);
2315 #endif
2316 new_table_size2 = new_table_size * sizeof(uint64_t);
2317 new_table = qemu_mallocz(new_table_size2);
2318 if (!new_table)
2319 return -ENOMEM;
2320 memcpy(new_table, s->refcount_table,
2321 s->refcount_table_size * sizeof(uint64_t));
2322 for(i = 0; i < s->refcount_table_size; i++)
2323 cpu_to_be64s(&new_table[i]);
2324 /* Note: we cannot update the refcount now to avoid recursion */
2325 table_offset = alloc_clusters_noref(bs, new_table_size2);
2326 ret = bdrv_pwrite(s->hd, table_offset, new_table, new_table_size2);
2327 if (ret != new_table_size2)
2328 goto fail;
2329 for(i = 0; i < s->refcount_table_size; i++)
2330 be64_to_cpus(&new_table[i]);
2332 cpu_to_be64w((uint64_t*)data, table_offset);
2333 cpu_to_be32w((uint32_t*)(data + 8), refcount_table_clusters);
2334 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, refcount_table_offset),
2335 data, sizeof(data)) != sizeof(data))
2336 goto fail;
2337 qemu_free(s->refcount_table);
2338 old_table_offset = s->refcount_table_offset;
2339 old_table_size = s->refcount_table_size;
2340 s->refcount_table = new_table;
2341 s->refcount_table_size = new_table_size;
2342 s->refcount_table_offset = table_offset;
2344 update_refcount(bs, table_offset, new_table_size2, 1);
2345 free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t));
2346 return 0;
2347 fail:
2348 free_clusters(bs, table_offset, new_table_size2);
2349 qemu_free(new_table);
2350 return -EIO;
2353 /* addend must be 1 or -1 */
2354 /* XXX: cache several refcount block clusters ? */
2355 static int update_cluster_refcount(BlockDriverState *bs,
2356 int64_t cluster_index,
2357 int addend)
2359 BDRVQcowState *s = bs->opaque;
2360 int64_t offset, refcount_block_offset;
2361 int ret, refcount_table_index, block_index, refcount;
2362 uint64_t data64;
2364 refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
2365 if (refcount_table_index >= s->refcount_table_size) {
2366 if (addend < 0)
2367 return -EINVAL;
2368 ret = grow_refcount_table(bs, refcount_table_index + 1);
2369 if (ret < 0)
2370 return ret;
2372 refcount_block_offset = s->refcount_table[refcount_table_index];
2373 if (!refcount_block_offset) {
2374 if (addend < 0)
2375 return -EINVAL;
2376 /* create a new refcount block */
2377 /* Note: we cannot update the refcount now to avoid recursion */
2378 offset = alloc_clusters_noref(bs, s->cluster_size);
2379 memset(s->refcount_block_cache, 0, s->cluster_size);
2380 ret = bdrv_pwrite(s->hd, offset, s->refcount_block_cache, s->cluster_size);
2381 if (ret != s->cluster_size)
2382 return -EINVAL;
2383 s->refcount_table[refcount_table_index] = offset;
2384 data64 = cpu_to_be64(offset);
2385 ret = bdrv_pwrite(s->hd, s->refcount_table_offset +
2386 refcount_table_index * sizeof(uint64_t),
2387 &data64, sizeof(data64));
2388 if (ret != sizeof(data64))
2389 return -EINVAL;
2391 refcount_block_offset = offset;
2392 s->refcount_block_cache_offset = offset;
2393 update_refcount(bs, offset, s->cluster_size, 1);
2394 } else {
2395 if (refcount_block_offset != s->refcount_block_cache_offset) {
2396 if (load_refcount_block(bs, refcount_block_offset) < 0)
2397 return -EIO;
2400 /* we can update the count and save it */
2401 block_index = cluster_index &
2402 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
2403 refcount = be16_to_cpu(s->refcount_block_cache[block_index]);
2404 refcount += addend;
2405 if (refcount < 0 || refcount > 0xffff)
2406 return -EINVAL;
2407 if (refcount == 0 && cluster_index < s->free_cluster_index) {
2408 s->free_cluster_index = cluster_index;
2410 s->refcount_block_cache[block_index] = cpu_to_be16(refcount);
2411 if (bdrv_pwrite(s->hd,
2412 refcount_block_offset + (block_index << REFCOUNT_SHIFT),
2413 &s->refcount_block_cache[block_index], 2) != 2)
2414 return -EIO;
2415 return refcount;
2418 static void update_refcount(BlockDriverState *bs,
2419 int64_t offset, int64_t length,
2420 int addend)
2422 BDRVQcowState *s = bs->opaque;
2423 int64_t start, last, cluster_offset;
2425 #ifdef DEBUG_ALLOC2
2426 printf("update_refcount: offset=%lld size=%lld addend=%d\n",
2427 offset, length, addend);
2428 #endif
2429 if (length <= 0)
2430 return;
2431 start = offset & ~(s->cluster_size - 1);
2432 last = (offset + length - 1) & ~(s->cluster_size - 1);
2433 for(cluster_offset = start; cluster_offset <= last;
2434 cluster_offset += s->cluster_size) {
2435 update_cluster_refcount(bs, cluster_offset >> s->cluster_bits, addend);
2439 #ifdef DEBUG_ALLOC
2440 static void inc_refcounts(BlockDriverState *bs,
2441 uint16_t *refcount_table,
2442 int refcount_table_size,
2443 int64_t offset, int64_t size)
2445 BDRVQcowState *s = bs->opaque;
2446 int64_t start, last, cluster_offset;
2447 int k;
2449 if (size <= 0)
2450 return;
2452 start = offset & ~(s->cluster_size - 1);
2453 last = (offset + size - 1) & ~(s->cluster_size - 1);
2454 for(cluster_offset = start; cluster_offset <= last;
2455 cluster_offset += s->cluster_size) {
2456 k = cluster_offset >> s->cluster_bits;
2457 if (k < 0 || k >= refcount_table_size) {
2458 printf("ERROR: invalid cluster offset=0x%llx\n", cluster_offset);
2459 } else {
2460 if (++refcount_table[k] == 0) {
2461 printf("ERROR: overflow cluster offset=0x%llx\n", cluster_offset);
2467 static int check_refcounts_l1(BlockDriverState *bs,
2468 uint16_t *refcount_table,
2469 int refcount_table_size,
2470 int64_t l1_table_offset, int l1_size,
2471 int check_copied)
2473 BDRVQcowState *s = bs->opaque;
2474 uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2;
2475 int l2_size, i, j, nb_csectors, refcount;
2477 l2_table = NULL;
2478 l1_size2 = l1_size * sizeof(uint64_t);
2480 inc_refcounts(bs, refcount_table, refcount_table_size,
2481 l1_table_offset, l1_size2);
2483 l1_table = qemu_malloc(l1_size2);
2484 if (!l1_table)
2485 goto fail;
2486 if (bdrv_pread(s->hd, l1_table_offset,
2487 l1_table, l1_size2) != l1_size2)
2488 goto fail;
2489 for(i = 0;i < l1_size; i++)
2490 be64_to_cpus(&l1_table[i]);
2492 l2_size = s->l2_size * sizeof(uint64_t);
2493 l2_table = qemu_malloc(l2_size);
2494 if (!l2_table)
2495 goto fail;
2496 for(i = 0; i < l1_size; i++) {
2497 l2_offset = l1_table[i];
2498 if (l2_offset) {
2499 if (check_copied) {
2500 refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED) >> s->cluster_bits);
2501 if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) {
2502 printf("ERROR OFLAG_COPIED: l2_offset=%llx refcount=%d\n",
2503 l2_offset, refcount);
2506 l2_offset &= ~QCOW_OFLAG_COPIED;
2507 if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
2508 goto fail;
2509 for(j = 0; j < s->l2_size; j++) {
2510 offset = be64_to_cpu(l2_table[j]);
2511 if (offset != 0) {
2512 if (offset & QCOW_OFLAG_COMPRESSED) {
2513 if (offset & QCOW_OFLAG_COPIED) {
2514 printf("ERROR: cluster %lld: copied flag must never be set for compressed clusters\n",
2515 offset >> s->cluster_bits);
2516 offset &= ~QCOW_OFLAG_COPIED;
2518 nb_csectors = ((offset >> s->csize_shift) &
2519 s->csize_mask) + 1;
2520 offset &= s->cluster_offset_mask;
2521 inc_refcounts(bs, refcount_table,
2522 refcount_table_size,
2523 offset & ~511, nb_csectors * 512);
2524 } else {
2525 if (check_copied) {
2526 refcount = get_refcount(bs, (offset & ~QCOW_OFLAG_COPIED) >> s->cluster_bits);
2527 if ((refcount == 1) != ((offset & QCOW_OFLAG_COPIED) != 0)) {
2528 printf("ERROR OFLAG_COPIED: offset=%llx refcount=%d\n",
2529 offset, refcount);
2532 offset &= ~QCOW_OFLAG_COPIED;
2533 inc_refcounts(bs, refcount_table,
2534 refcount_table_size,
2535 offset, s->cluster_size);
2539 inc_refcounts(bs, refcount_table,
2540 refcount_table_size,
2541 l2_offset,
2542 s->cluster_size);
2545 qemu_free(l1_table);
2546 qemu_free(l2_table);
2547 return 0;
2548 fail:
2549 printf("ERROR: I/O error in check_refcounts_l1\n");
2550 qemu_free(l1_table);
2551 qemu_free(l2_table);
2552 return -EIO;
2555 static void check_refcounts(BlockDriverState *bs)
2557 BDRVQcowState *s = bs->opaque;
2558 int64_t size;
2559 int nb_clusters, refcount1, refcount2, i;
2560 QCowSnapshot *sn;
2561 uint16_t *refcount_table;
2563 size = bdrv_getlength(s->hd);
2564 nb_clusters = size_to_clusters(s, size);
2565 refcount_table = qemu_mallocz(nb_clusters * sizeof(uint16_t));
2567 /* header */
2568 inc_refcounts(bs, refcount_table, nb_clusters,
2569 0, s->cluster_size);
2571 check_refcounts_l1(bs, refcount_table, nb_clusters,
2572 s->l1_table_offset, s->l1_size, 1);
2574 /* snapshots */
2575 for(i = 0; i < s->nb_snapshots; i++) {
2576 sn = s->snapshots + i;
2577 check_refcounts_l1(bs, refcount_table, nb_clusters,
2578 sn->l1_table_offset, sn->l1_size, 0);
2580 inc_refcounts(bs, refcount_table, nb_clusters,
2581 s->snapshots_offset, s->snapshots_size);
2583 /* refcount data */
2584 inc_refcounts(bs, refcount_table, nb_clusters,
2585 s->refcount_table_offset,
2586 s->refcount_table_size * sizeof(uint64_t));
2587 for(i = 0; i < s->refcount_table_size; i++) {
2588 int64_t offset;
2589 offset = s->refcount_table[i];
2590 if (offset != 0) {
2591 inc_refcounts(bs, refcount_table, nb_clusters,
2592 offset, s->cluster_size);
2596 /* compare ref counts */
2597 for(i = 0; i < nb_clusters; i++) {
2598 refcount1 = get_refcount(bs, i);
2599 refcount2 = refcount_table[i];
2600 if (refcount1 != refcount2)
2601 printf("ERROR cluster %d refcount=%d reference=%d\n",
2602 i, refcount1, refcount2);
2605 qemu_free(refcount_table);
2608 #if 0
2609 static void dump_refcounts(BlockDriverState *bs)
2611 BDRVQcowState *s = bs->opaque;
2612 int64_t nb_clusters, k, k1, size;
2613 int refcount;
2615 size = bdrv_getlength(s->hd);
2616 nb_clusters = size_to_clusters(s, size);
2617 for(k = 0; k < nb_clusters;) {
2618 k1 = k;
2619 refcount = get_refcount(bs, k);
2620 k++;
2621 while (k < nb_clusters && get_refcount(bs, k) == refcount)
2622 k++;
2623 printf("%lld: refcount=%d nb=%lld\n", k, refcount, k - k1);
2626 #endif
2627 #endif
2629 BlockDriver bdrv_qcow2 = {
2630 "qcow2",
2631 sizeof(BDRVQcowState),
2632 qcow_probe,
2633 qcow_open,
2634 NULL,
2635 NULL,
2636 qcow_close,
2637 qcow_create,
2638 qcow_flush,
2639 qcow_is_allocated,
2640 qcow_set_key,
2641 qcow_make_empty,
2643 .bdrv_aio_read = qcow_aio_read,
2644 .bdrv_aio_write = qcow_aio_write,
2645 .bdrv_aio_cancel = qcow_aio_cancel,
2646 .aiocb_size = sizeof(QCowAIOCB),
2647 .bdrv_write_compressed = qcow_write_compressed,
2649 .bdrv_snapshot_create = qcow_snapshot_create,
2650 .bdrv_snapshot_goto = qcow_snapshot_goto,
2651 .bdrv_snapshot_delete = qcow_snapshot_delete,
2652 .bdrv_snapshot_list = qcow_snapshot_list,
2653 .bdrv_get_info = qcow_get_info,