Add specialized block driver scsi generic API (Avi Kivity)
[qemu/mini2440/sniper_sniper_test.git] / block-qcow2.c
blobafbf7feb56b87686817975f4738f42a239bdeed0
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 /* Performance is terrible right now with cache=writethrough due mainly
193 * to reference count updates. If the user does not explicitly specify
194 * a caching type, force to writeback caching.
196 if ((flags & BDRV_O_CACHE_DEF)) {
197 flags |= BDRV_O_CACHE_WB;
198 flags &= ~BDRV_O_CACHE_DEF;
200 ret = bdrv_file_open(&s->hd, filename, flags);
201 if (ret < 0)
202 return ret;
203 if (bdrv_pread(s->hd, 0, &header, sizeof(header)) != sizeof(header))
204 goto fail;
205 be32_to_cpus(&header.magic);
206 be32_to_cpus(&header.version);
207 be64_to_cpus(&header.backing_file_offset);
208 be32_to_cpus(&header.backing_file_size);
209 be64_to_cpus(&header.size);
210 be32_to_cpus(&header.cluster_bits);
211 be32_to_cpus(&header.crypt_method);
212 be64_to_cpus(&header.l1_table_offset);
213 be32_to_cpus(&header.l1_size);
214 be64_to_cpus(&header.refcount_table_offset);
215 be32_to_cpus(&header.refcount_table_clusters);
216 be64_to_cpus(&header.snapshots_offset);
217 be32_to_cpus(&header.nb_snapshots);
219 if (header.magic != QCOW_MAGIC || header.version != QCOW_VERSION)
220 goto fail;
221 if (header.size <= 1 ||
222 header.cluster_bits < 9 ||
223 header.cluster_bits > 16)
224 goto fail;
225 if (header.crypt_method > QCOW_CRYPT_AES)
226 goto fail;
227 s->crypt_method_header = header.crypt_method;
228 if (s->crypt_method_header)
229 bs->encrypted = 1;
230 s->cluster_bits = header.cluster_bits;
231 s->cluster_size = 1 << s->cluster_bits;
232 s->cluster_sectors = 1 << (s->cluster_bits - 9);
233 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
234 s->l2_size = 1 << s->l2_bits;
235 bs->total_sectors = header.size / 512;
236 s->csize_shift = (62 - (s->cluster_bits - 8));
237 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
238 s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
239 s->refcount_table_offset = header.refcount_table_offset;
240 s->refcount_table_size =
241 header.refcount_table_clusters << (s->cluster_bits - 3);
243 s->snapshots_offset = header.snapshots_offset;
244 s->nb_snapshots = header.nb_snapshots;
246 /* read the level 1 table */
247 s->l1_size = header.l1_size;
248 shift = s->cluster_bits + s->l2_bits;
249 s->l1_vm_state_index = (header.size + (1LL << shift) - 1) >> shift;
250 /* the L1 table must contain at least enough entries to put
251 header.size bytes */
252 if (s->l1_size < s->l1_vm_state_index)
253 goto fail;
254 s->l1_table_offset = header.l1_table_offset;
255 s->l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
256 if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, s->l1_size * sizeof(uint64_t)) !=
257 s->l1_size * sizeof(uint64_t))
258 goto fail;
259 for(i = 0;i < s->l1_size; i++) {
260 be64_to_cpus(&s->l1_table[i]);
262 /* alloc L2 cache */
263 s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
264 s->cluster_cache = qemu_malloc(s->cluster_size);
265 /* one more sector for decompressed data alignment */
266 s->cluster_data = qemu_malloc(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
267 + 512);
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 memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t));
448 /* write new table (align to cluster) */
449 new_l1_table_offset = alloc_clusters(bs, new_l1_size2);
451 for(i = 0; i < s->l1_size; i++)
452 new_l1_table[i] = cpu_to_be64(new_l1_table[i]);
453 ret = bdrv_pwrite(s->hd, new_l1_table_offset, new_l1_table, new_l1_size2);
454 if (ret != new_l1_size2)
455 goto fail;
456 for(i = 0; i < s->l1_size; i++)
457 new_l1_table[i] = be64_to_cpu(new_l1_table[i]);
459 /* set new table */
460 cpu_to_be32w((uint32_t*)data, new_l1_size);
461 cpu_to_be64w((uint64_t*)(data + 4), new_l1_table_offset);
462 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, l1_size), data,
463 sizeof(data)) != sizeof(data))
464 goto fail;
465 qemu_free(s->l1_table);
466 free_clusters(bs, s->l1_table_offset, s->l1_size * sizeof(uint64_t));
467 s->l1_table_offset = new_l1_table_offset;
468 s->l1_table = new_l1_table;
469 s->l1_size = new_l1_size;
470 return 0;
471 fail:
472 qemu_free(s->l1_table);
473 return -EIO;
477 * seek_l2_table
479 * seek l2_offset in the l2_cache table
480 * if not found, return NULL,
481 * if found,
482 * increments the l2 cache hit count of the entry,
483 * if counter overflow, divide by two all counters
484 * return the pointer to the l2 cache entry
488 static uint64_t *seek_l2_table(BDRVQcowState *s, uint64_t l2_offset)
490 int i, j;
492 for(i = 0; i < L2_CACHE_SIZE; i++) {
493 if (l2_offset == s->l2_cache_offsets[i]) {
494 /* increment the hit count */
495 if (++s->l2_cache_counts[i] == 0xffffffff) {
496 for(j = 0; j < L2_CACHE_SIZE; j++) {
497 s->l2_cache_counts[j] >>= 1;
500 return s->l2_cache + (i << s->l2_bits);
503 return NULL;
507 * l2_load
509 * Loads a L2 table into memory. If the table is in the cache, the cache
510 * is used; otherwise the L2 table is loaded from the image file.
512 * Returns a pointer to the L2 table on success, or NULL if the read from
513 * the image file failed.
516 static uint64_t *l2_load(BlockDriverState *bs, uint64_t l2_offset)
518 BDRVQcowState *s = bs->opaque;
519 int min_index;
520 uint64_t *l2_table;
522 /* seek if the table for the given offset is in the cache */
524 l2_table = seek_l2_table(s, l2_offset);
525 if (l2_table != NULL)
526 return l2_table;
528 /* not found: load a new entry in the least used one */
530 min_index = l2_cache_new_entry(bs);
531 l2_table = s->l2_cache + (min_index << s->l2_bits);
532 if (bdrv_pread(s->hd, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
533 s->l2_size * sizeof(uint64_t))
534 return NULL;
535 s->l2_cache_offsets[min_index] = l2_offset;
536 s->l2_cache_counts[min_index] = 1;
538 return l2_table;
542 * l2_allocate
544 * Allocate a new l2 entry in the file. If l1_index points to an already
545 * used entry in the L2 table (i.e. we are doing a copy on write for the L2
546 * table) copy the contents of the old L2 table into the newly allocated one.
547 * Otherwise the new table is initialized with zeros.
551 static uint64_t *l2_allocate(BlockDriverState *bs, int l1_index)
553 BDRVQcowState *s = bs->opaque;
554 int min_index;
555 uint64_t old_l2_offset, tmp;
556 uint64_t *l2_table, l2_offset;
558 old_l2_offset = s->l1_table[l1_index];
560 /* allocate a new l2 entry */
562 l2_offset = alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
564 /* update the L1 entry */
566 s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
568 tmp = cpu_to_be64(l2_offset | QCOW_OFLAG_COPIED);
569 if (bdrv_pwrite(s->hd, s->l1_table_offset + l1_index * sizeof(tmp),
570 &tmp, sizeof(tmp)) != sizeof(tmp))
571 return NULL;
573 /* allocate a new entry in the l2 cache */
575 min_index = l2_cache_new_entry(bs);
576 l2_table = s->l2_cache + (min_index << s->l2_bits);
578 if (old_l2_offset == 0) {
579 /* if there was no old l2 table, clear the new table */
580 memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
581 } else {
582 /* if there was an old l2 table, read it from the disk */
583 if (bdrv_pread(s->hd, old_l2_offset,
584 l2_table, s->l2_size * sizeof(uint64_t)) !=
585 s->l2_size * sizeof(uint64_t))
586 return NULL;
588 /* write the l2 table to the file */
589 if (bdrv_pwrite(s->hd, l2_offset,
590 l2_table, s->l2_size * sizeof(uint64_t)) !=
591 s->l2_size * sizeof(uint64_t))
592 return NULL;
594 /* update the l2 cache entry */
596 s->l2_cache_offsets[min_index] = l2_offset;
597 s->l2_cache_counts[min_index] = 1;
599 return l2_table;
602 static int size_to_clusters(BDRVQcowState *s, int64_t size)
604 return (size + (s->cluster_size - 1)) >> s->cluster_bits;
607 static int count_contiguous_clusters(uint64_t nb_clusters, int cluster_size,
608 uint64_t *l2_table, uint64_t start, uint64_t mask)
610 int i;
611 uint64_t offset = be64_to_cpu(l2_table[0]) & ~mask;
613 if (!offset)
614 return 0;
616 for (i = start; i < start + nb_clusters; i++)
617 if (offset + i * cluster_size != (be64_to_cpu(l2_table[i]) & ~mask))
618 break;
620 return (i - start);
623 static int count_contiguous_free_clusters(uint64_t nb_clusters, uint64_t *l2_table)
625 int i = 0;
627 while(nb_clusters-- && l2_table[i] == 0)
628 i++;
630 return i;
634 * get_cluster_offset
636 * For a given offset of the disk image, return cluster offset in
637 * qcow2 file.
639 * on entry, *num is the number of contiguous clusters we'd like to
640 * access following offset.
642 * on exit, *num is the number of contiguous clusters we can read.
644 * Return 1, if the offset is found
645 * Return 0, otherwise.
649 static uint64_t get_cluster_offset(BlockDriverState *bs,
650 uint64_t offset, int *num)
652 BDRVQcowState *s = bs->opaque;
653 int l1_index, l2_index;
654 uint64_t l2_offset, *l2_table, cluster_offset;
655 int l1_bits, c;
656 int index_in_cluster, nb_available, nb_needed, nb_clusters;
658 index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1);
659 nb_needed = *num + index_in_cluster;
661 l1_bits = s->l2_bits + s->cluster_bits;
663 /* compute how many bytes there are between the offset and
664 * the end of the l1 entry
667 nb_available = (1 << l1_bits) - (offset & ((1 << l1_bits) - 1));
669 /* compute the number of available sectors */
671 nb_available = (nb_available >> 9) + index_in_cluster;
673 cluster_offset = 0;
675 /* seek the the l2 offset in the l1 table */
677 l1_index = offset >> l1_bits;
678 if (l1_index >= s->l1_size)
679 goto out;
681 l2_offset = s->l1_table[l1_index];
683 /* seek the l2 table of the given l2 offset */
685 if (!l2_offset)
686 goto out;
688 /* load the l2 table in memory */
690 l2_offset &= ~QCOW_OFLAG_COPIED;
691 l2_table = l2_load(bs, l2_offset);
692 if (l2_table == NULL)
693 return 0;
695 /* find the cluster offset for the given disk offset */
697 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
698 cluster_offset = be64_to_cpu(l2_table[l2_index]);
699 nb_clusters = size_to_clusters(s, nb_needed << 9);
701 if (!cluster_offset) {
702 /* how many empty clusters ? */
703 c = count_contiguous_free_clusters(nb_clusters, &l2_table[l2_index]);
704 } else {
705 /* how many allocated clusters ? */
706 c = count_contiguous_clusters(nb_clusters, s->cluster_size,
707 &l2_table[l2_index], 0, QCOW_OFLAG_COPIED);
710 nb_available = (c * s->cluster_sectors);
711 out:
712 if (nb_available > nb_needed)
713 nb_available = nb_needed;
715 *num = nb_available - index_in_cluster;
717 return cluster_offset & ~QCOW_OFLAG_COPIED;
721 * free_any_clusters
723 * free clusters according to its type: compressed or not
727 static void free_any_clusters(BlockDriverState *bs,
728 uint64_t cluster_offset, int nb_clusters)
730 BDRVQcowState *s = bs->opaque;
732 /* free the cluster */
734 if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
735 int nb_csectors;
736 nb_csectors = ((cluster_offset >> s->csize_shift) &
737 s->csize_mask) + 1;
738 free_clusters(bs, (cluster_offset & s->cluster_offset_mask) & ~511,
739 nb_csectors * 512);
740 return;
743 free_clusters(bs, cluster_offset, nb_clusters << s->cluster_bits);
745 return;
749 * get_cluster_table
751 * for a given disk offset, load (and allocate if needed)
752 * the l2 table.
754 * the l2 table offset in the qcow2 file and the cluster index
755 * in the l2 table are given to the caller.
759 static int get_cluster_table(BlockDriverState *bs, uint64_t offset,
760 uint64_t **new_l2_table,
761 uint64_t *new_l2_offset,
762 int *new_l2_index)
764 BDRVQcowState *s = bs->opaque;
765 int l1_index, l2_index, ret;
766 uint64_t l2_offset, *l2_table;
768 /* seek the the l2 offset in the l1 table */
770 l1_index = offset >> (s->l2_bits + s->cluster_bits);
771 if (l1_index >= s->l1_size) {
772 ret = grow_l1_table(bs, l1_index + 1);
773 if (ret < 0)
774 return 0;
776 l2_offset = s->l1_table[l1_index];
778 /* seek the l2 table of the given l2 offset */
780 if (l2_offset & QCOW_OFLAG_COPIED) {
781 /* load the l2 table in memory */
782 l2_offset &= ~QCOW_OFLAG_COPIED;
783 l2_table = l2_load(bs, l2_offset);
784 if (l2_table == NULL)
785 return 0;
786 } else {
787 if (l2_offset)
788 free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t));
789 l2_table = l2_allocate(bs, l1_index);
790 if (l2_table == NULL)
791 return 0;
792 l2_offset = s->l1_table[l1_index] & ~QCOW_OFLAG_COPIED;
795 /* find the cluster offset for the given disk offset */
797 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
799 *new_l2_table = l2_table;
800 *new_l2_offset = l2_offset;
801 *new_l2_index = l2_index;
803 return 1;
807 * alloc_compressed_cluster_offset
809 * For a given offset of the disk image, return cluster offset in
810 * qcow2 file.
812 * If the offset is not found, allocate a new compressed cluster.
814 * Return the cluster offset if successful,
815 * Return 0, otherwise.
819 static uint64_t alloc_compressed_cluster_offset(BlockDriverState *bs,
820 uint64_t offset,
821 int compressed_size)
823 BDRVQcowState *s = bs->opaque;
824 int l2_index, ret;
825 uint64_t l2_offset, *l2_table, cluster_offset;
826 int nb_csectors;
828 ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
829 if (ret == 0)
830 return 0;
832 cluster_offset = be64_to_cpu(l2_table[l2_index]);
833 if (cluster_offset & QCOW_OFLAG_COPIED)
834 return cluster_offset & ~QCOW_OFLAG_COPIED;
836 if (cluster_offset)
837 free_any_clusters(bs, cluster_offset, 1);
839 cluster_offset = alloc_bytes(bs, compressed_size);
840 nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) -
841 (cluster_offset >> 9);
843 cluster_offset |= QCOW_OFLAG_COMPRESSED |
844 ((uint64_t)nb_csectors << s->csize_shift);
846 /* update L2 table */
848 /* compressed clusters never have the copied flag */
850 l2_table[l2_index] = cpu_to_be64(cluster_offset);
851 if (bdrv_pwrite(s->hd,
852 l2_offset + l2_index * sizeof(uint64_t),
853 l2_table + l2_index,
854 sizeof(uint64_t)) != sizeof(uint64_t))
855 return 0;
857 return cluster_offset;
860 typedef struct QCowL2Meta
862 uint64_t offset;
863 int n_start;
864 int nb_available;
865 int nb_clusters;
866 } QCowL2Meta;
868 static int alloc_cluster_link_l2(BlockDriverState *bs, uint64_t cluster_offset,
869 QCowL2Meta *m)
871 BDRVQcowState *s = bs->opaque;
872 int i, j = 0, l2_index, ret;
873 uint64_t *old_cluster, start_sect, l2_offset, *l2_table;
875 if (m->nb_clusters == 0)
876 return 0;
878 old_cluster = qemu_malloc(m->nb_clusters * sizeof(uint64_t));
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, 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_clusters(nb_clusters - i, s->cluster_size,
977 &l2_table[l2_index], i, 0);
979 if(be64_to_cpu(l2_table[l2_index + i]))
980 break;
982 i += count_contiguous_free_clusters(nb_clusters - i,
983 &l2_table[l2_index + i]);
985 cluster_offset = be64_to_cpu(l2_table[l2_index + i]);
987 if ((cluster_offset & QCOW_OFLAG_COPIED) ||
988 (cluster_offset & QCOW_OFLAG_COMPRESSED))
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);
1378 encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf,
1379 acb->n, 1, &s->aes_encrypt_key);
1380 src_buf = acb->cluster_data;
1381 } else {
1382 src_buf = acb->buf;
1384 acb->hd_aiocb = bdrv_aio_write(s->hd,
1385 (acb->cluster_offset >> 9) + index_in_cluster,
1386 src_buf, acb->n,
1387 qcow_aio_write_cb, acb);
1388 if (acb->hd_aiocb == NULL)
1389 goto fail;
1392 static BlockDriverAIOCB *qcow_aio_write(BlockDriverState *bs,
1393 int64_t sector_num, const uint8_t *buf, int nb_sectors,
1394 BlockDriverCompletionFunc *cb, void *opaque)
1396 BDRVQcowState *s = bs->opaque;
1397 QCowAIOCB *acb;
1399 s->cluster_cache_offset = -1; /* disable compressed cache */
1401 acb = qcow_aio_setup(bs, sector_num, (uint8_t*)buf, nb_sectors, cb, opaque);
1402 if (!acb)
1403 return NULL;
1405 qcow_aio_write_cb(acb, 0);
1406 return &acb->common;
1409 static void qcow_aio_cancel(BlockDriverAIOCB *blockacb)
1411 QCowAIOCB *acb = (QCowAIOCB *)blockacb;
1412 if (acb->hd_aiocb)
1413 bdrv_aio_cancel(acb->hd_aiocb);
1414 qemu_aio_release(acb);
1417 static void qcow_close(BlockDriverState *bs)
1419 BDRVQcowState *s = bs->opaque;
1420 qemu_free(s->l1_table);
1421 qemu_free(s->l2_cache);
1422 qemu_free(s->cluster_cache);
1423 qemu_free(s->cluster_data);
1424 refcount_close(bs);
1425 bdrv_delete(s->hd);
1428 /* XXX: use std qcow open function ? */
1429 typedef struct QCowCreateState {
1430 int cluster_size;
1431 int cluster_bits;
1432 uint16_t *refcount_block;
1433 uint64_t *refcount_table;
1434 int64_t l1_table_offset;
1435 int64_t refcount_table_offset;
1436 int64_t refcount_block_offset;
1437 } QCowCreateState;
1439 static void create_refcount_update(QCowCreateState *s,
1440 int64_t offset, int64_t size)
1442 int refcount;
1443 int64_t start, last, cluster_offset;
1444 uint16_t *p;
1446 start = offset & ~(s->cluster_size - 1);
1447 last = (offset + size - 1) & ~(s->cluster_size - 1);
1448 for(cluster_offset = start; cluster_offset <= last;
1449 cluster_offset += s->cluster_size) {
1450 p = &s->refcount_block[cluster_offset >> s->cluster_bits];
1451 refcount = be16_to_cpu(*p);
1452 refcount++;
1453 *p = cpu_to_be16(refcount);
1457 static int qcow_create(const char *filename, int64_t total_size,
1458 const char *backing_file, int flags)
1460 int fd, header_size, backing_filename_len, l1_size, i, shift, l2_bits;
1461 QCowHeader header;
1462 uint64_t tmp, offset;
1463 QCowCreateState s1, *s = &s1;
1465 memset(s, 0, sizeof(*s));
1467 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
1468 if (fd < 0)
1469 return -1;
1470 memset(&header, 0, sizeof(header));
1471 header.magic = cpu_to_be32(QCOW_MAGIC);
1472 header.version = cpu_to_be32(QCOW_VERSION);
1473 header.size = cpu_to_be64(total_size * 512);
1474 header_size = sizeof(header);
1475 backing_filename_len = 0;
1476 if (backing_file) {
1477 header.backing_file_offset = cpu_to_be64(header_size);
1478 backing_filename_len = strlen(backing_file);
1479 header.backing_file_size = cpu_to_be32(backing_filename_len);
1480 header_size += backing_filename_len;
1482 s->cluster_bits = 12; /* 4 KB clusters */
1483 s->cluster_size = 1 << s->cluster_bits;
1484 header.cluster_bits = cpu_to_be32(s->cluster_bits);
1485 header_size = (header_size + 7) & ~7;
1486 if (flags & BLOCK_FLAG_ENCRYPT) {
1487 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
1488 } else {
1489 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
1491 l2_bits = s->cluster_bits - 3;
1492 shift = s->cluster_bits + l2_bits;
1493 l1_size = (((total_size * 512) + (1LL << shift) - 1) >> shift);
1494 offset = align_offset(header_size, s->cluster_size);
1495 s->l1_table_offset = offset;
1496 header.l1_table_offset = cpu_to_be64(s->l1_table_offset);
1497 header.l1_size = cpu_to_be32(l1_size);
1498 offset += align_offset(l1_size * sizeof(uint64_t), s->cluster_size);
1500 s->refcount_table = qemu_mallocz(s->cluster_size);
1501 s->refcount_block = qemu_mallocz(s->cluster_size);
1503 s->refcount_table_offset = offset;
1504 header.refcount_table_offset = cpu_to_be64(offset);
1505 header.refcount_table_clusters = cpu_to_be32(1);
1506 offset += s->cluster_size;
1508 s->refcount_table[0] = cpu_to_be64(offset);
1509 s->refcount_block_offset = offset;
1510 offset += s->cluster_size;
1512 /* update refcounts */
1513 create_refcount_update(s, 0, header_size);
1514 create_refcount_update(s, s->l1_table_offset, l1_size * sizeof(uint64_t));
1515 create_refcount_update(s, s->refcount_table_offset, s->cluster_size);
1516 create_refcount_update(s, s->refcount_block_offset, s->cluster_size);
1518 /* write all the data */
1519 write(fd, &header, sizeof(header));
1520 if (backing_file) {
1521 write(fd, backing_file, backing_filename_len);
1523 lseek(fd, s->l1_table_offset, SEEK_SET);
1524 tmp = 0;
1525 for(i = 0;i < l1_size; i++) {
1526 write(fd, &tmp, sizeof(tmp));
1528 lseek(fd, s->refcount_table_offset, SEEK_SET);
1529 write(fd, s->refcount_table, s->cluster_size);
1531 lseek(fd, s->refcount_block_offset, SEEK_SET);
1532 write(fd, s->refcount_block, s->cluster_size);
1534 qemu_free(s->refcount_table);
1535 qemu_free(s->refcount_block);
1536 close(fd);
1537 return 0;
1540 static int qcow_make_empty(BlockDriverState *bs)
1542 #if 0
1543 /* XXX: not correct */
1544 BDRVQcowState *s = bs->opaque;
1545 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1546 int ret;
1548 memset(s->l1_table, 0, l1_length);
1549 if (bdrv_pwrite(s->hd, s->l1_table_offset, s->l1_table, l1_length) < 0)
1550 return -1;
1551 ret = bdrv_truncate(s->hd, s->l1_table_offset + l1_length);
1552 if (ret < 0)
1553 return ret;
1555 l2_cache_reset(bs);
1556 #endif
1557 return 0;
1560 /* XXX: put compressed sectors first, then all the cluster aligned
1561 tables to avoid losing bytes in alignment */
1562 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
1563 const uint8_t *buf, int nb_sectors)
1565 BDRVQcowState *s = bs->opaque;
1566 z_stream strm;
1567 int ret, out_len;
1568 uint8_t *out_buf;
1569 uint64_t cluster_offset;
1571 if (nb_sectors == 0) {
1572 /* align end of file to a sector boundary to ease reading with
1573 sector based I/Os */
1574 cluster_offset = bdrv_getlength(s->hd);
1575 cluster_offset = (cluster_offset + 511) & ~511;
1576 bdrv_truncate(s->hd, cluster_offset);
1577 return 0;
1580 if (nb_sectors != s->cluster_sectors)
1581 return -EINVAL;
1583 out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
1585 /* best compression, small window, no zlib header */
1586 memset(&strm, 0, sizeof(strm));
1587 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1588 Z_DEFLATED, -12,
1589 9, Z_DEFAULT_STRATEGY);
1590 if (ret != 0) {
1591 qemu_free(out_buf);
1592 return -1;
1595 strm.avail_in = s->cluster_size;
1596 strm.next_in = (uint8_t *)buf;
1597 strm.avail_out = s->cluster_size;
1598 strm.next_out = out_buf;
1600 ret = deflate(&strm, Z_FINISH);
1601 if (ret != Z_STREAM_END && ret != Z_OK) {
1602 qemu_free(out_buf);
1603 deflateEnd(&strm);
1604 return -1;
1606 out_len = strm.next_out - out_buf;
1608 deflateEnd(&strm);
1610 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1611 /* could not compress: write normal cluster */
1612 qcow_write(bs, sector_num, buf, s->cluster_sectors);
1613 } else {
1614 cluster_offset = alloc_compressed_cluster_offset(bs, sector_num << 9,
1615 out_len);
1616 if (!cluster_offset)
1617 return -1;
1618 cluster_offset &= s->cluster_offset_mask;
1619 if (bdrv_pwrite(s->hd, cluster_offset, out_buf, out_len) != out_len) {
1620 qemu_free(out_buf);
1621 return -1;
1625 qemu_free(out_buf);
1626 return 0;
1629 static void qcow_flush(BlockDriverState *bs)
1631 BDRVQcowState *s = bs->opaque;
1632 bdrv_flush(s->hd);
1635 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1637 BDRVQcowState *s = bs->opaque;
1638 bdi->cluster_size = s->cluster_size;
1639 bdi->vm_state_offset = (int64_t)s->l1_vm_state_index <<
1640 (s->cluster_bits + s->l2_bits);
1641 return 0;
1644 /*********************************************************/
1645 /* snapshot support */
1647 /* update the refcounts of snapshots and the copied flag */
1648 static int update_snapshot_refcount(BlockDriverState *bs,
1649 int64_t l1_table_offset,
1650 int l1_size,
1651 int addend)
1653 BDRVQcowState *s = bs->opaque;
1654 uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated;
1655 int64_t old_offset, old_l2_offset;
1656 int l2_size, i, j, l1_modified, l2_modified, nb_csectors, refcount;
1658 l2_cache_reset(bs);
1660 l2_table = NULL;
1661 l1_table = NULL;
1662 l1_size2 = l1_size * sizeof(uint64_t);
1663 l1_allocated = 0;
1664 if (l1_table_offset != s->l1_table_offset) {
1665 l1_table = qemu_malloc(l1_size2);
1666 l1_allocated = 1;
1667 if (bdrv_pread(s->hd, l1_table_offset,
1668 l1_table, l1_size2) != l1_size2)
1669 goto fail;
1670 for(i = 0;i < l1_size; i++)
1671 be64_to_cpus(&l1_table[i]);
1672 } else {
1673 assert(l1_size == s->l1_size);
1674 l1_table = s->l1_table;
1675 l1_allocated = 0;
1678 l2_size = s->l2_size * sizeof(uint64_t);
1679 l2_table = qemu_malloc(l2_size);
1680 l1_modified = 0;
1681 for(i = 0; i < l1_size; i++) {
1682 l2_offset = l1_table[i];
1683 if (l2_offset) {
1684 old_l2_offset = l2_offset;
1685 l2_offset &= ~QCOW_OFLAG_COPIED;
1686 l2_modified = 0;
1687 if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
1688 goto fail;
1689 for(j = 0; j < s->l2_size; j++) {
1690 offset = be64_to_cpu(l2_table[j]);
1691 if (offset != 0) {
1692 old_offset = offset;
1693 offset &= ~QCOW_OFLAG_COPIED;
1694 if (offset & QCOW_OFLAG_COMPRESSED) {
1695 nb_csectors = ((offset >> s->csize_shift) &
1696 s->csize_mask) + 1;
1697 if (addend != 0)
1698 update_refcount(bs, (offset & s->cluster_offset_mask) & ~511,
1699 nb_csectors * 512, addend);
1700 /* compressed clusters are never modified */
1701 refcount = 2;
1702 } else {
1703 if (addend != 0) {
1704 refcount = update_cluster_refcount(bs, offset >> s->cluster_bits, addend);
1705 } else {
1706 refcount = get_refcount(bs, offset >> s->cluster_bits);
1710 if (refcount == 1) {
1711 offset |= QCOW_OFLAG_COPIED;
1713 if (offset != old_offset) {
1714 l2_table[j] = cpu_to_be64(offset);
1715 l2_modified = 1;
1719 if (l2_modified) {
1720 if (bdrv_pwrite(s->hd,
1721 l2_offset, l2_table, l2_size) != l2_size)
1722 goto fail;
1725 if (addend != 0) {
1726 refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend);
1727 } else {
1728 refcount = get_refcount(bs, l2_offset >> s->cluster_bits);
1730 if (refcount == 1) {
1731 l2_offset |= QCOW_OFLAG_COPIED;
1733 if (l2_offset != old_l2_offset) {
1734 l1_table[i] = l2_offset;
1735 l1_modified = 1;
1739 if (l1_modified) {
1740 for(i = 0; i < l1_size; i++)
1741 cpu_to_be64s(&l1_table[i]);
1742 if (bdrv_pwrite(s->hd, l1_table_offset, l1_table,
1743 l1_size2) != l1_size2)
1744 goto fail;
1745 for(i = 0; i < l1_size; i++)
1746 be64_to_cpus(&l1_table[i]);
1748 if (l1_allocated)
1749 qemu_free(l1_table);
1750 qemu_free(l2_table);
1751 return 0;
1752 fail:
1753 if (l1_allocated)
1754 qemu_free(l1_table);
1755 qemu_free(l2_table);
1756 return -EIO;
1759 static void qcow_free_snapshots(BlockDriverState *bs)
1761 BDRVQcowState *s = bs->opaque;
1762 int i;
1764 for(i = 0; i < s->nb_snapshots; i++) {
1765 qemu_free(s->snapshots[i].name);
1766 qemu_free(s->snapshots[i].id_str);
1768 qemu_free(s->snapshots);
1769 s->snapshots = NULL;
1770 s->nb_snapshots = 0;
1773 static int qcow_read_snapshots(BlockDriverState *bs)
1775 BDRVQcowState *s = bs->opaque;
1776 QCowSnapshotHeader h;
1777 QCowSnapshot *sn;
1778 int i, id_str_size, name_size;
1779 int64_t offset;
1780 uint32_t extra_data_size;
1782 if (!s->nb_snapshots) {
1783 s->snapshots = NULL;
1784 s->snapshots_size = 0;
1785 return 0;
1788 offset = s->snapshots_offset;
1789 s->snapshots = qemu_mallocz(s->nb_snapshots * sizeof(QCowSnapshot));
1790 for(i = 0; i < s->nb_snapshots; i++) {
1791 offset = align_offset(offset, 8);
1792 if (bdrv_pread(s->hd, offset, &h, sizeof(h)) != sizeof(h))
1793 goto fail;
1794 offset += sizeof(h);
1795 sn = s->snapshots + i;
1796 sn->l1_table_offset = be64_to_cpu(h.l1_table_offset);
1797 sn->l1_size = be32_to_cpu(h.l1_size);
1798 sn->vm_state_size = be32_to_cpu(h.vm_state_size);
1799 sn->date_sec = be32_to_cpu(h.date_sec);
1800 sn->date_nsec = be32_to_cpu(h.date_nsec);
1801 sn->vm_clock_nsec = be64_to_cpu(h.vm_clock_nsec);
1802 extra_data_size = be32_to_cpu(h.extra_data_size);
1804 id_str_size = be16_to_cpu(h.id_str_size);
1805 name_size = be16_to_cpu(h.name_size);
1807 offset += extra_data_size;
1809 sn->id_str = qemu_malloc(id_str_size + 1);
1810 if (bdrv_pread(s->hd, offset, sn->id_str, id_str_size) != id_str_size)
1811 goto fail;
1812 offset += id_str_size;
1813 sn->id_str[id_str_size] = '\0';
1815 sn->name = qemu_malloc(name_size + 1);
1816 if (bdrv_pread(s->hd, offset, sn->name, name_size) != name_size)
1817 goto fail;
1818 offset += name_size;
1819 sn->name[name_size] = '\0';
1821 s->snapshots_size = offset - s->snapshots_offset;
1822 return 0;
1823 fail:
1824 qcow_free_snapshots(bs);
1825 return -1;
1828 /* add at the end of the file a new list of snapshots */
1829 static int qcow_write_snapshots(BlockDriverState *bs)
1831 BDRVQcowState *s = bs->opaque;
1832 QCowSnapshot *sn;
1833 QCowSnapshotHeader h;
1834 int i, name_size, id_str_size, snapshots_size;
1835 uint64_t data64;
1836 uint32_t data32;
1837 int64_t offset, snapshots_offset;
1839 /* compute the size of the snapshots */
1840 offset = 0;
1841 for(i = 0; i < s->nb_snapshots; i++) {
1842 sn = s->snapshots + i;
1843 offset = align_offset(offset, 8);
1844 offset += sizeof(h);
1845 offset += strlen(sn->id_str);
1846 offset += strlen(sn->name);
1848 snapshots_size = offset;
1850 snapshots_offset = alloc_clusters(bs, snapshots_size);
1851 offset = snapshots_offset;
1853 for(i = 0; i < s->nb_snapshots; i++) {
1854 sn = s->snapshots + i;
1855 memset(&h, 0, sizeof(h));
1856 h.l1_table_offset = cpu_to_be64(sn->l1_table_offset);
1857 h.l1_size = cpu_to_be32(sn->l1_size);
1858 h.vm_state_size = cpu_to_be32(sn->vm_state_size);
1859 h.date_sec = cpu_to_be32(sn->date_sec);
1860 h.date_nsec = cpu_to_be32(sn->date_nsec);
1861 h.vm_clock_nsec = cpu_to_be64(sn->vm_clock_nsec);
1863 id_str_size = strlen(sn->id_str);
1864 name_size = strlen(sn->name);
1865 h.id_str_size = cpu_to_be16(id_str_size);
1866 h.name_size = cpu_to_be16(name_size);
1867 offset = align_offset(offset, 8);
1868 if (bdrv_pwrite(s->hd, offset, &h, sizeof(h)) != sizeof(h))
1869 goto fail;
1870 offset += sizeof(h);
1871 if (bdrv_pwrite(s->hd, offset, sn->id_str, id_str_size) != id_str_size)
1872 goto fail;
1873 offset += id_str_size;
1874 if (bdrv_pwrite(s->hd, offset, sn->name, name_size) != name_size)
1875 goto fail;
1876 offset += name_size;
1879 /* update the various header fields */
1880 data64 = cpu_to_be64(snapshots_offset);
1881 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, snapshots_offset),
1882 &data64, sizeof(data64)) != sizeof(data64))
1883 goto fail;
1884 data32 = cpu_to_be32(s->nb_snapshots);
1885 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, nb_snapshots),
1886 &data32, sizeof(data32)) != sizeof(data32))
1887 goto fail;
1889 /* free the old snapshot table */
1890 free_clusters(bs, s->snapshots_offset, s->snapshots_size);
1891 s->snapshots_offset = snapshots_offset;
1892 s->snapshots_size = snapshots_size;
1893 return 0;
1894 fail:
1895 return -1;
1898 static void find_new_snapshot_id(BlockDriverState *bs,
1899 char *id_str, int id_str_size)
1901 BDRVQcowState *s = bs->opaque;
1902 QCowSnapshot *sn;
1903 int i, id, id_max = 0;
1905 for(i = 0; i < s->nb_snapshots; i++) {
1906 sn = s->snapshots + i;
1907 id = strtoul(sn->id_str, NULL, 10);
1908 if (id > id_max)
1909 id_max = id;
1911 snprintf(id_str, id_str_size, "%d", id_max + 1);
1914 static int find_snapshot_by_id(BlockDriverState *bs, const char *id_str)
1916 BDRVQcowState *s = bs->opaque;
1917 int i;
1919 for(i = 0; i < s->nb_snapshots; i++) {
1920 if (!strcmp(s->snapshots[i].id_str, id_str))
1921 return i;
1923 return -1;
1926 static int find_snapshot_by_id_or_name(BlockDriverState *bs, const char *name)
1928 BDRVQcowState *s = bs->opaque;
1929 int i, ret;
1931 ret = find_snapshot_by_id(bs, name);
1932 if (ret >= 0)
1933 return ret;
1934 for(i = 0; i < s->nb_snapshots; i++) {
1935 if (!strcmp(s->snapshots[i].name, name))
1936 return i;
1938 return -1;
1941 /* if no id is provided, a new one is constructed */
1942 static int qcow_snapshot_create(BlockDriverState *bs,
1943 QEMUSnapshotInfo *sn_info)
1945 BDRVQcowState *s = bs->opaque;
1946 QCowSnapshot *snapshots1, sn1, *sn = &sn1;
1947 int i, ret;
1948 uint64_t *l1_table = NULL;
1950 memset(sn, 0, sizeof(*sn));
1952 if (sn_info->id_str[0] == '\0') {
1953 /* compute a new id */
1954 find_new_snapshot_id(bs, sn_info->id_str, sizeof(sn_info->id_str));
1957 /* check that the ID is unique */
1958 if (find_snapshot_by_id(bs, sn_info->id_str) >= 0)
1959 return -ENOENT;
1961 sn->id_str = qemu_strdup(sn_info->id_str);
1962 if (!sn->id_str)
1963 goto fail;
1964 sn->name = qemu_strdup(sn_info->name);
1965 if (!sn->name)
1966 goto fail;
1967 sn->vm_state_size = sn_info->vm_state_size;
1968 sn->date_sec = sn_info->date_sec;
1969 sn->date_nsec = sn_info->date_nsec;
1970 sn->vm_clock_nsec = sn_info->vm_clock_nsec;
1972 ret = update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1);
1973 if (ret < 0)
1974 goto fail;
1976 /* create the L1 table of the snapshot */
1977 sn->l1_table_offset = alloc_clusters(bs, s->l1_size * sizeof(uint64_t));
1978 sn->l1_size = s->l1_size;
1980 l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
1981 for(i = 0; i < s->l1_size; i++) {
1982 l1_table[i] = cpu_to_be64(s->l1_table[i]);
1984 if (bdrv_pwrite(s->hd, sn->l1_table_offset,
1985 l1_table, s->l1_size * sizeof(uint64_t)) !=
1986 (s->l1_size * sizeof(uint64_t)))
1987 goto fail;
1988 qemu_free(l1_table);
1989 l1_table = NULL;
1991 snapshots1 = qemu_malloc((s->nb_snapshots + 1) * sizeof(QCowSnapshot));
1992 if (s->snapshots) {
1993 memcpy(snapshots1, s->snapshots, s->nb_snapshots * sizeof(QCowSnapshot));
1994 qemu_free(s->snapshots);
1996 s->snapshots = snapshots1;
1997 s->snapshots[s->nb_snapshots++] = *sn;
1999 if (qcow_write_snapshots(bs) < 0)
2000 goto fail;
2001 #ifdef DEBUG_ALLOC
2002 check_refcounts(bs);
2003 #endif
2004 return 0;
2005 fail:
2006 qemu_free(sn->name);
2007 qemu_free(l1_table);
2008 return -1;
2011 /* copy the snapshot 'snapshot_name' into the current disk image */
2012 static int qcow_snapshot_goto(BlockDriverState *bs,
2013 const char *snapshot_id)
2015 BDRVQcowState *s = bs->opaque;
2016 QCowSnapshot *sn;
2017 int i, snapshot_index, l1_size2;
2019 snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
2020 if (snapshot_index < 0)
2021 return -ENOENT;
2022 sn = &s->snapshots[snapshot_index];
2024 if (update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, -1) < 0)
2025 goto fail;
2027 if (grow_l1_table(bs, sn->l1_size) < 0)
2028 goto fail;
2030 s->l1_size = sn->l1_size;
2031 l1_size2 = s->l1_size * sizeof(uint64_t);
2032 /* copy the snapshot l1 table to the current l1 table */
2033 if (bdrv_pread(s->hd, sn->l1_table_offset,
2034 s->l1_table, l1_size2) != l1_size2)
2035 goto fail;
2036 if (bdrv_pwrite(s->hd, s->l1_table_offset,
2037 s->l1_table, l1_size2) != l1_size2)
2038 goto fail;
2039 for(i = 0;i < s->l1_size; i++) {
2040 be64_to_cpus(&s->l1_table[i]);
2043 if (update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1) < 0)
2044 goto fail;
2046 #ifdef DEBUG_ALLOC
2047 check_refcounts(bs);
2048 #endif
2049 return 0;
2050 fail:
2051 return -EIO;
2054 static int qcow_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
2056 BDRVQcowState *s = bs->opaque;
2057 QCowSnapshot *sn;
2058 int snapshot_index, ret;
2060 snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
2061 if (snapshot_index < 0)
2062 return -ENOENT;
2063 sn = &s->snapshots[snapshot_index];
2065 ret = update_snapshot_refcount(bs, sn->l1_table_offset, sn->l1_size, -1);
2066 if (ret < 0)
2067 return ret;
2068 /* must update the copied flag on the current cluster offsets */
2069 ret = update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 0);
2070 if (ret < 0)
2071 return ret;
2072 free_clusters(bs, sn->l1_table_offset, sn->l1_size * sizeof(uint64_t));
2074 qemu_free(sn->id_str);
2075 qemu_free(sn->name);
2076 memmove(sn, sn + 1, (s->nb_snapshots - snapshot_index - 1) * sizeof(*sn));
2077 s->nb_snapshots--;
2078 ret = qcow_write_snapshots(bs);
2079 if (ret < 0) {
2080 /* XXX: restore snapshot if error ? */
2081 return ret;
2083 #ifdef DEBUG_ALLOC
2084 check_refcounts(bs);
2085 #endif
2086 return 0;
2089 static int qcow_snapshot_list(BlockDriverState *bs,
2090 QEMUSnapshotInfo **psn_tab)
2092 BDRVQcowState *s = bs->opaque;
2093 QEMUSnapshotInfo *sn_tab, *sn_info;
2094 QCowSnapshot *sn;
2095 int i;
2097 sn_tab = qemu_mallocz(s->nb_snapshots * sizeof(QEMUSnapshotInfo));
2098 for(i = 0; i < s->nb_snapshots; i++) {
2099 sn_info = sn_tab + i;
2100 sn = s->snapshots + i;
2101 pstrcpy(sn_info->id_str, sizeof(sn_info->id_str),
2102 sn->id_str);
2103 pstrcpy(sn_info->name, sizeof(sn_info->name),
2104 sn->name);
2105 sn_info->vm_state_size = sn->vm_state_size;
2106 sn_info->date_sec = sn->date_sec;
2107 sn_info->date_nsec = sn->date_nsec;
2108 sn_info->vm_clock_nsec = sn->vm_clock_nsec;
2110 *psn_tab = sn_tab;
2111 return s->nb_snapshots;
2114 /*********************************************************/
2115 /* refcount handling */
2117 static int refcount_init(BlockDriverState *bs)
2119 BDRVQcowState *s = bs->opaque;
2120 int ret, refcount_table_size2, i;
2122 s->refcount_block_cache = qemu_malloc(s->cluster_size);
2123 refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t);
2124 s->refcount_table = qemu_malloc(refcount_table_size2);
2125 if (s->refcount_table_size > 0) {
2126 ret = bdrv_pread(s->hd, s->refcount_table_offset,
2127 s->refcount_table, refcount_table_size2);
2128 if (ret != refcount_table_size2)
2129 goto fail;
2130 for(i = 0; i < s->refcount_table_size; i++)
2131 be64_to_cpus(&s->refcount_table[i]);
2133 return 0;
2134 fail:
2135 return -ENOMEM;
2138 static void refcount_close(BlockDriverState *bs)
2140 BDRVQcowState *s = bs->opaque;
2141 qemu_free(s->refcount_block_cache);
2142 qemu_free(s->refcount_table);
2146 static int load_refcount_block(BlockDriverState *bs,
2147 int64_t refcount_block_offset)
2149 BDRVQcowState *s = bs->opaque;
2150 int ret;
2151 ret = bdrv_pread(s->hd, refcount_block_offset, s->refcount_block_cache,
2152 s->cluster_size);
2153 if (ret != s->cluster_size)
2154 return -EIO;
2155 s->refcount_block_cache_offset = refcount_block_offset;
2156 return 0;
2159 static int get_refcount(BlockDriverState *bs, int64_t cluster_index)
2161 BDRVQcowState *s = bs->opaque;
2162 int refcount_table_index, block_index;
2163 int64_t refcount_block_offset;
2165 refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
2166 if (refcount_table_index >= s->refcount_table_size)
2167 return 0;
2168 refcount_block_offset = s->refcount_table[refcount_table_index];
2169 if (!refcount_block_offset)
2170 return 0;
2171 if (refcount_block_offset != s->refcount_block_cache_offset) {
2172 /* better than nothing: return allocated if read error */
2173 if (load_refcount_block(bs, refcount_block_offset) < 0)
2174 return 1;
2176 block_index = cluster_index &
2177 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
2178 return be16_to_cpu(s->refcount_block_cache[block_index]);
2181 /* return < 0 if error */
2182 static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size)
2184 BDRVQcowState *s = bs->opaque;
2185 int i, nb_clusters;
2187 nb_clusters = size_to_clusters(s, size);
2188 retry:
2189 for(i = 0; i < nb_clusters; i++) {
2190 int64_t i = s->free_cluster_index++;
2191 if (get_refcount(bs, i) != 0)
2192 goto retry;
2194 #ifdef DEBUG_ALLOC2
2195 printf("alloc_clusters: size=%lld -> %lld\n",
2196 size,
2197 (s->free_cluster_index - nb_clusters) << s->cluster_bits);
2198 #endif
2199 return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
2202 static int64_t alloc_clusters(BlockDriverState *bs, int64_t size)
2204 int64_t offset;
2206 offset = alloc_clusters_noref(bs, size);
2207 update_refcount(bs, offset, size, 1);
2208 return offset;
2211 /* only used to allocate compressed sectors. We try to allocate
2212 contiguous sectors. size must be <= cluster_size */
2213 static int64_t alloc_bytes(BlockDriverState *bs, int size)
2215 BDRVQcowState *s = bs->opaque;
2216 int64_t offset, cluster_offset;
2217 int free_in_cluster;
2219 assert(size > 0 && size <= s->cluster_size);
2220 if (s->free_byte_offset == 0) {
2221 s->free_byte_offset = alloc_clusters(bs, s->cluster_size);
2223 redo:
2224 free_in_cluster = s->cluster_size -
2225 (s->free_byte_offset & (s->cluster_size - 1));
2226 if (size <= free_in_cluster) {
2227 /* enough space in current cluster */
2228 offset = s->free_byte_offset;
2229 s->free_byte_offset += size;
2230 free_in_cluster -= size;
2231 if (free_in_cluster == 0)
2232 s->free_byte_offset = 0;
2233 if ((offset & (s->cluster_size - 1)) != 0)
2234 update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
2235 } else {
2236 offset = alloc_clusters(bs, s->cluster_size);
2237 cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1);
2238 if ((cluster_offset + s->cluster_size) == offset) {
2239 /* we are lucky: contiguous data */
2240 offset = s->free_byte_offset;
2241 update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
2242 s->free_byte_offset += size;
2243 } else {
2244 s->free_byte_offset = offset;
2245 goto redo;
2248 return offset;
2251 static void free_clusters(BlockDriverState *bs,
2252 int64_t offset, int64_t size)
2254 update_refcount(bs, offset, size, -1);
2257 static int grow_refcount_table(BlockDriverState *bs, int min_size)
2259 BDRVQcowState *s = bs->opaque;
2260 int new_table_size, new_table_size2, refcount_table_clusters, i, ret;
2261 uint64_t *new_table;
2262 int64_t table_offset;
2263 uint8_t data[12];
2264 int old_table_size;
2265 int64_t old_table_offset;
2267 if (min_size <= s->refcount_table_size)
2268 return 0;
2269 /* compute new table size */
2270 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
2271 for(;;) {
2272 if (refcount_table_clusters == 0) {
2273 refcount_table_clusters = 1;
2274 } else {
2275 refcount_table_clusters = (refcount_table_clusters * 3 + 1) / 2;
2277 new_table_size = refcount_table_clusters << (s->cluster_bits - 3);
2278 if (min_size <= new_table_size)
2279 break;
2281 #ifdef DEBUG_ALLOC2
2282 printf("grow_refcount_table from %d to %d\n",
2283 s->refcount_table_size,
2284 new_table_size);
2285 #endif
2286 new_table_size2 = new_table_size * sizeof(uint64_t);
2287 new_table = qemu_mallocz(new_table_size2);
2288 memcpy(new_table, s->refcount_table,
2289 s->refcount_table_size * sizeof(uint64_t));
2290 for(i = 0; i < s->refcount_table_size; i++)
2291 cpu_to_be64s(&new_table[i]);
2292 /* Note: we cannot update the refcount now to avoid recursion */
2293 table_offset = alloc_clusters_noref(bs, new_table_size2);
2294 ret = bdrv_pwrite(s->hd, table_offset, new_table, new_table_size2);
2295 if (ret != new_table_size2)
2296 goto fail;
2297 for(i = 0; i < s->refcount_table_size; i++)
2298 be64_to_cpus(&new_table[i]);
2300 cpu_to_be64w((uint64_t*)data, table_offset);
2301 cpu_to_be32w((uint32_t*)(data + 8), refcount_table_clusters);
2302 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, refcount_table_offset),
2303 data, sizeof(data)) != sizeof(data))
2304 goto fail;
2305 qemu_free(s->refcount_table);
2306 old_table_offset = s->refcount_table_offset;
2307 old_table_size = s->refcount_table_size;
2308 s->refcount_table = new_table;
2309 s->refcount_table_size = new_table_size;
2310 s->refcount_table_offset = table_offset;
2312 update_refcount(bs, table_offset, new_table_size2, 1);
2313 free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t));
2314 return 0;
2315 fail:
2316 free_clusters(bs, table_offset, new_table_size2);
2317 qemu_free(new_table);
2318 return -EIO;
2321 /* addend must be 1 or -1 */
2322 /* XXX: cache several refcount block clusters ? */
2323 static int update_cluster_refcount(BlockDriverState *bs,
2324 int64_t cluster_index,
2325 int addend)
2327 BDRVQcowState *s = bs->opaque;
2328 int64_t offset, refcount_block_offset;
2329 int ret, refcount_table_index, block_index, refcount;
2330 uint64_t data64;
2332 refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
2333 if (refcount_table_index >= s->refcount_table_size) {
2334 if (addend < 0)
2335 return -EINVAL;
2336 ret = grow_refcount_table(bs, refcount_table_index + 1);
2337 if (ret < 0)
2338 return ret;
2340 refcount_block_offset = s->refcount_table[refcount_table_index];
2341 if (!refcount_block_offset) {
2342 if (addend < 0)
2343 return -EINVAL;
2344 /* create a new refcount block */
2345 /* Note: we cannot update the refcount now to avoid recursion */
2346 offset = alloc_clusters_noref(bs, s->cluster_size);
2347 memset(s->refcount_block_cache, 0, s->cluster_size);
2348 ret = bdrv_pwrite(s->hd, offset, s->refcount_block_cache, s->cluster_size);
2349 if (ret != s->cluster_size)
2350 return -EINVAL;
2351 s->refcount_table[refcount_table_index] = offset;
2352 data64 = cpu_to_be64(offset);
2353 ret = bdrv_pwrite(s->hd, s->refcount_table_offset +
2354 refcount_table_index * sizeof(uint64_t),
2355 &data64, sizeof(data64));
2356 if (ret != sizeof(data64))
2357 return -EINVAL;
2359 refcount_block_offset = offset;
2360 s->refcount_block_cache_offset = offset;
2361 update_refcount(bs, offset, s->cluster_size, 1);
2362 } else {
2363 if (refcount_block_offset != s->refcount_block_cache_offset) {
2364 if (load_refcount_block(bs, refcount_block_offset) < 0)
2365 return -EIO;
2368 /* we can update the count and save it */
2369 block_index = cluster_index &
2370 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
2371 refcount = be16_to_cpu(s->refcount_block_cache[block_index]);
2372 refcount += addend;
2373 if (refcount < 0 || refcount > 0xffff)
2374 return -EINVAL;
2375 if (refcount == 0 && cluster_index < s->free_cluster_index) {
2376 s->free_cluster_index = cluster_index;
2378 s->refcount_block_cache[block_index] = cpu_to_be16(refcount);
2379 if (bdrv_pwrite(s->hd,
2380 refcount_block_offset + (block_index << REFCOUNT_SHIFT),
2381 &s->refcount_block_cache[block_index], 2) != 2)
2382 return -EIO;
2383 return refcount;
2386 static void update_refcount(BlockDriverState *bs,
2387 int64_t offset, int64_t length,
2388 int addend)
2390 BDRVQcowState *s = bs->opaque;
2391 int64_t start, last, cluster_offset;
2393 #ifdef DEBUG_ALLOC2
2394 printf("update_refcount: offset=%lld size=%lld addend=%d\n",
2395 offset, length, addend);
2396 #endif
2397 if (length <= 0)
2398 return;
2399 start = offset & ~(s->cluster_size - 1);
2400 last = (offset + length - 1) & ~(s->cluster_size - 1);
2401 for(cluster_offset = start; cluster_offset <= last;
2402 cluster_offset += s->cluster_size) {
2403 update_cluster_refcount(bs, cluster_offset >> s->cluster_bits, addend);
2407 #ifdef DEBUG_ALLOC
2408 static void inc_refcounts(BlockDriverState *bs,
2409 uint16_t *refcount_table,
2410 int refcount_table_size,
2411 int64_t offset, int64_t size)
2413 BDRVQcowState *s = bs->opaque;
2414 int64_t start, last, cluster_offset;
2415 int k;
2417 if (size <= 0)
2418 return;
2420 start = offset & ~(s->cluster_size - 1);
2421 last = (offset + size - 1) & ~(s->cluster_size - 1);
2422 for(cluster_offset = start; cluster_offset <= last;
2423 cluster_offset += s->cluster_size) {
2424 k = cluster_offset >> s->cluster_bits;
2425 if (k < 0 || k >= refcount_table_size) {
2426 printf("ERROR: invalid cluster offset=0x%llx\n", cluster_offset);
2427 } else {
2428 if (++refcount_table[k] == 0) {
2429 printf("ERROR: overflow cluster offset=0x%llx\n", cluster_offset);
2435 static int check_refcounts_l1(BlockDriverState *bs,
2436 uint16_t *refcount_table,
2437 int refcount_table_size,
2438 int64_t l1_table_offset, int l1_size,
2439 int check_copied)
2441 BDRVQcowState *s = bs->opaque;
2442 uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2;
2443 int l2_size, i, j, nb_csectors, refcount;
2445 l2_table = NULL;
2446 l1_size2 = l1_size * sizeof(uint64_t);
2448 inc_refcounts(bs, refcount_table, refcount_table_size,
2449 l1_table_offset, l1_size2);
2451 l1_table = qemu_malloc(l1_size2);
2452 if (bdrv_pread(s->hd, l1_table_offset,
2453 l1_table, l1_size2) != l1_size2)
2454 goto fail;
2455 for(i = 0;i < l1_size; i++)
2456 be64_to_cpus(&l1_table[i]);
2458 l2_size = s->l2_size * sizeof(uint64_t);
2459 l2_table = qemu_malloc(l2_size);
2460 for(i = 0; i < l1_size; i++) {
2461 l2_offset = l1_table[i];
2462 if (l2_offset) {
2463 if (check_copied) {
2464 refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED) >> s->cluster_bits);
2465 if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) {
2466 printf("ERROR OFLAG_COPIED: l2_offset=%llx refcount=%d\n",
2467 l2_offset, refcount);
2470 l2_offset &= ~QCOW_OFLAG_COPIED;
2471 if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
2472 goto fail;
2473 for(j = 0; j < s->l2_size; j++) {
2474 offset = be64_to_cpu(l2_table[j]);
2475 if (offset != 0) {
2476 if (offset & QCOW_OFLAG_COMPRESSED) {
2477 if (offset & QCOW_OFLAG_COPIED) {
2478 printf("ERROR: cluster %lld: copied flag must never be set for compressed clusters\n",
2479 offset >> s->cluster_bits);
2480 offset &= ~QCOW_OFLAG_COPIED;
2482 nb_csectors = ((offset >> s->csize_shift) &
2483 s->csize_mask) + 1;
2484 offset &= s->cluster_offset_mask;
2485 inc_refcounts(bs, refcount_table,
2486 refcount_table_size,
2487 offset & ~511, nb_csectors * 512);
2488 } else {
2489 if (check_copied) {
2490 refcount = get_refcount(bs, (offset & ~QCOW_OFLAG_COPIED) >> s->cluster_bits);
2491 if ((refcount == 1) != ((offset & QCOW_OFLAG_COPIED) != 0)) {
2492 printf("ERROR OFLAG_COPIED: offset=%llx refcount=%d\n",
2493 offset, refcount);
2496 offset &= ~QCOW_OFLAG_COPIED;
2497 inc_refcounts(bs, refcount_table,
2498 refcount_table_size,
2499 offset, s->cluster_size);
2503 inc_refcounts(bs, refcount_table,
2504 refcount_table_size,
2505 l2_offset,
2506 s->cluster_size);
2509 qemu_free(l1_table);
2510 qemu_free(l2_table);
2511 return 0;
2512 fail:
2513 printf("ERROR: I/O error in check_refcounts_l1\n");
2514 qemu_free(l1_table);
2515 qemu_free(l2_table);
2516 return -EIO;
2519 static void check_refcounts(BlockDriverState *bs)
2521 BDRVQcowState *s = bs->opaque;
2522 int64_t size;
2523 int nb_clusters, refcount1, refcount2, i;
2524 QCowSnapshot *sn;
2525 uint16_t *refcount_table;
2527 size = bdrv_getlength(s->hd);
2528 nb_clusters = size_to_clusters(s, size);
2529 refcount_table = qemu_mallocz(nb_clusters * sizeof(uint16_t));
2531 /* header */
2532 inc_refcounts(bs, refcount_table, nb_clusters,
2533 0, s->cluster_size);
2535 check_refcounts_l1(bs, refcount_table, nb_clusters,
2536 s->l1_table_offset, s->l1_size, 1);
2538 /* snapshots */
2539 for(i = 0; i < s->nb_snapshots; i++) {
2540 sn = s->snapshots + i;
2541 check_refcounts_l1(bs, refcount_table, nb_clusters,
2542 sn->l1_table_offset, sn->l1_size, 0);
2544 inc_refcounts(bs, refcount_table, nb_clusters,
2545 s->snapshots_offset, s->snapshots_size);
2547 /* refcount data */
2548 inc_refcounts(bs, refcount_table, nb_clusters,
2549 s->refcount_table_offset,
2550 s->refcount_table_size * sizeof(uint64_t));
2551 for(i = 0; i < s->refcount_table_size; i++) {
2552 int64_t offset;
2553 offset = s->refcount_table[i];
2554 if (offset != 0) {
2555 inc_refcounts(bs, refcount_table, nb_clusters,
2556 offset, s->cluster_size);
2560 /* compare ref counts */
2561 for(i = 0; i < nb_clusters; i++) {
2562 refcount1 = get_refcount(bs, i);
2563 refcount2 = refcount_table[i];
2564 if (refcount1 != refcount2)
2565 printf("ERROR cluster %d refcount=%d reference=%d\n",
2566 i, refcount1, refcount2);
2569 qemu_free(refcount_table);
2572 #if 0
2573 static void dump_refcounts(BlockDriverState *bs)
2575 BDRVQcowState *s = bs->opaque;
2576 int64_t nb_clusters, k, k1, size;
2577 int refcount;
2579 size = bdrv_getlength(s->hd);
2580 nb_clusters = size_to_clusters(s, size);
2581 for(k = 0; k < nb_clusters;) {
2582 k1 = k;
2583 refcount = get_refcount(bs, k);
2584 k++;
2585 while (k < nb_clusters && get_refcount(bs, k) == refcount)
2586 k++;
2587 printf("%lld: refcount=%d nb=%lld\n", k, refcount, k - k1);
2590 #endif
2591 #endif
2593 BlockDriver bdrv_qcow2 = {
2594 .format_name = "qcow2",
2595 .instance_size = sizeof(BDRVQcowState),
2596 .bdrv_probe = qcow_probe,
2597 .bdrv_open = qcow_open,
2598 .bdrv_close = qcow_close,
2599 .bdrv_create = qcow_create,
2600 .bdrv_flush = qcow_flush,
2601 .bdrv_is_allocated = qcow_is_allocated,
2602 .bdrv_set_key = qcow_set_key,
2603 .bdrv_make_empty = qcow_make_empty,
2605 .bdrv_aio_read = qcow_aio_read,
2606 .bdrv_aio_write = qcow_aio_write,
2607 .bdrv_aio_cancel = qcow_aio_cancel,
2608 .aiocb_size = sizeof(QCowAIOCB),
2609 .bdrv_write_compressed = qcow_write_compressed,
2611 .bdrv_snapshot_create = qcow_snapshot_create,
2612 .bdrv_snapshot_goto = qcow_snapshot_goto,
2613 .bdrv_snapshot_delete = qcow_snapshot_delete,
2614 .bdrv_snapshot_list = qcow_snapshot_list,
2615 .bdrv_get_info = qcow_get_info,