[MISC] Updated maintainer
[qemu/mini2440/sniper_sniper_test.git] / block-qcow2.c
blobeec1255e6a46c0c7c37ceae683af17f37c8c5649
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;
147 int64_t highest_alloc; /* highest cluester allocated (in clusters) */
148 int64_t nc_free; /* num of free clusters below highest_alloc */
150 uint64_t snapshots_offset;
151 int snapshots_size;
152 int nb_snapshots;
153 QCowSnapshot *snapshots;
154 } BDRVQcowState;
156 static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset);
157 static int qcow_read(BlockDriverState *bs, int64_t sector_num,
158 uint8_t *buf, int nb_sectors);
159 static int qcow_read_snapshots(BlockDriverState *bs);
160 static void qcow_free_snapshots(BlockDriverState *bs);
161 static int refcount_init(BlockDriverState *bs);
162 static void refcount_close(BlockDriverState *bs);
163 static int get_refcount(BlockDriverState *bs, int64_t cluster_index);
164 static int update_cluster_refcount(BlockDriverState *bs,
165 int64_t cluster_index,
166 int addend);
167 static void update_refcount(BlockDriverState *bs,
168 int64_t offset, int64_t length,
169 int addend);
170 static int64_t alloc_clusters(BlockDriverState *bs, int64_t size);
171 static int64_t alloc_bytes(BlockDriverState *bs, int size);
172 static void free_clusters(BlockDriverState *bs,
173 int64_t offset, int64_t size);
174 #ifdef DEBUG_ALLOC
175 static void check_refcounts(BlockDriverState *bs);
176 #endif
177 static void scan_refcount(BlockDriverState *bs, int64_t *high, int64_t *free);
180 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
182 const QCowHeader *cow_header = (const void *)buf;
184 if (buf_size >= sizeof(QCowHeader) &&
185 be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
186 be32_to_cpu(cow_header->version) == QCOW_VERSION)
187 return 100;
188 else
189 return 0;
192 static int qcow_open(BlockDriverState *bs, const char *filename, int flags)
194 BDRVQcowState *s = bs->opaque;
195 int len, i, shift, ret;
196 QCowHeader header;
198 /* Performance is terrible right now with cache=writethrough due mainly
199 * to reference count updates. If the user does not explicitly specify
200 * a caching type, force to writeback caching.
202 if ((flags & BDRV_O_CACHE_DEF)) {
203 flags |= BDRV_O_CACHE_WB;
204 flags &= ~BDRV_O_CACHE_DEF;
206 ret = bdrv_file_open(&s->hd, filename, flags);
207 if (ret < 0)
208 return ret;
209 if (bdrv_pread(s->hd, 0, &header, sizeof(header)) != sizeof(header))
210 goto fail;
211 be32_to_cpus(&header.magic);
212 be32_to_cpus(&header.version);
213 be64_to_cpus(&header.backing_file_offset);
214 be32_to_cpus(&header.backing_file_size);
215 be64_to_cpus(&header.size);
216 be32_to_cpus(&header.cluster_bits);
217 be32_to_cpus(&header.crypt_method);
218 be64_to_cpus(&header.l1_table_offset);
219 be32_to_cpus(&header.l1_size);
220 be64_to_cpus(&header.refcount_table_offset);
221 be32_to_cpus(&header.refcount_table_clusters);
222 be64_to_cpus(&header.snapshots_offset);
223 be32_to_cpus(&header.nb_snapshots);
225 if (header.magic != QCOW_MAGIC || header.version != QCOW_VERSION)
226 goto fail;
227 if (header.size <= 1 ||
228 header.cluster_bits < 9 ||
229 header.cluster_bits > 16)
230 goto fail;
231 if (header.crypt_method > QCOW_CRYPT_AES)
232 goto fail;
233 s->crypt_method_header = header.crypt_method;
234 if (s->crypt_method_header)
235 bs->encrypted = 1;
236 s->cluster_bits = header.cluster_bits;
237 s->cluster_size = 1 << s->cluster_bits;
238 s->cluster_sectors = 1 << (s->cluster_bits - 9);
239 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
240 s->l2_size = 1 << s->l2_bits;
241 bs->total_sectors = header.size / 512;
242 s->csize_shift = (62 - (s->cluster_bits - 8));
243 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
244 s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
245 s->refcount_table_offset = header.refcount_table_offset;
246 s->refcount_table_size =
247 header.refcount_table_clusters << (s->cluster_bits - 3);
249 s->snapshots_offset = header.snapshots_offset;
250 s->nb_snapshots = header.nb_snapshots;
252 /* read the level 1 table */
253 s->l1_size = header.l1_size;
254 shift = s->cluster_bits + s->l2_bits;
255 s->l1_vm_state_index = (header.size + (1LL << shift) - 1) >> shift;
256 /* the L1 table must contain at least enough entries to put
257 header.size bytes */
258 if (s->l1_size < s->l1_vm_state_index)
259 goto fail;
260 s->l1_table_offset = header.l1_table_offset;
261 s->l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
262 if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, s->l1_size * sizeof(uint64_t)) !=
263 s->l1_size * sizeof(uint64_t))
264 goto fail;
265 for(i = 0;i < s->l1_size; i++) {
266 be64_to_cpus(&s->l1_table[i]);
268 /* alloc L2 cache */
269 s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
270 s->cluster_cache = qemu_malloc(s->cluster_size);
271 /* one more sector for decompressed data alignment */
272 s->cluster_data = qemu_malloc(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
273 + 512);
274 s->cluster_cache_offset = -1;
276 if (refcount_init(bs) < 0)
277 goto fail;
279 scan_refcount(bs, &s->highest_alloc, &s->nc_free);
281 /* read the backing file name */
282 if (header.backing_file_offset != 0) {
283 len = header.backing_file_size;
284 if (len > 1023)
285 len = 1023;
286 if (bdrv_pread(s->hd, header.backing_file_offset, bs->backing_file, len) != len)
287 goto fail;
288 bs->backing_file[len] = '\0';
290 if (qcow_read_snapshots(bs) < 0)
291 goto fail;
293 #ifdef DEBUG_ALLOC
294 check_refcounts(bs);
295 #endif
296 return 0;
298 fail:
299 qcow_free_snapshots(bs);
300 refcount_close(bs);
301 qemu_free(s->l1_table);
302 qemu_free(s->l2_cache);
303 qemu_free(s->cluster_cache);
304 qemu_free(s->cluster_data);
305 bdrv_delete(s->hd);
306 return -1;
309 static int qcow_set_key(BlockDriverState *bs, const char *key)
311 BDRVQcowState *s = bs->opaque;
312 uint8_t keybuf[16];
313 int len, i;
315 memset(keybuf, 0, 16);
316 len = strlen(key);
317 if (len > 16)
318 len = 16;
319 /* XXX: we could compress the chars to 7 bits to increase
320 entropy */
321 for(i = 0;i < len;i++) {
322 keybuf[i] = key[i];
324 s->crypt_method = s->crypt_method_header;
326 if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
327 return -1;
328 if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
329 return -1;
330 #if 0
331 /* test */
333 uint8_t in[16];
334 uint8_t out[16];
335 uint8_t tmp[16];
336 for(i=0;i<16;i++)
337 in[i] = i;
338 AES_encrypt(in, tmp, &s->aes_encrypt_key);
339 AES_decrypt(tmp, out, &s->aes_decrypt_key);
340 for(i = 0; i < 16; i++)
341 printf(" %02x", tmp[i]);
342 printf("\n");
343 for(i = 0; i < 16; i++)
344 printf(" %02x", out[i]);
345 printf("\n");
347 #endif
348 return 0;
351 /* The crypt function is compatible with the linux cryptoloop
352 algorithm for < 4 GB images. NOTE: out_buf == in_buf is
353 supported */
354 static void encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
355 uint8_t *out_buf, const uint8_t *in_buf,
356 int nb_sectors, int enc,
357 const AES_KEY *key)
359 union {
360 uint64_t ll[2];
361 uint8_t b[16];
362 } ivec;
363 int i;
365 for(i = 0; i < nb_sectors; i++) {
366 ivec.ll[0] = cpu_to_le64(sector_num);
367 ivec.ll[1] = 0;
368 AES_cbc_encrypt(in_buf, out_buf, 512, key,
369 ivec.b, enc);
370 sector_num++;
371 in_buf += 512;
372 out_buf += 512;
376 static int copy_sectors(BlockDriverState *bs, uint64_t start_sect,
377 uint64_t cluster_offset, int n_start, int n_end)
379 BDRVQcowState *s = bs->opaque;
380 int n, ret;
382 n = n_end - n_start;
383 if (n <= 0)
384 return 0;
385 ret = qcow_read(bs, start_sect + n_start, s->cluster_data, n);
386 if (ret < 0)
387 return ret;
388 if (s->crypt_method) {
389 encrypt_sectors(s, start_sect + n_start,
390 s->cluster_data,
391 s->cluster_data, n, 1,
392 &s->aes_encrypt_key);
394 ret = bdrv_write(s->hd, (cluster_offset >> 9) + n_start,
395 s->cluster_data, n);
396 if (ret < 0)
397 return ret;
398 return 0;
401 static void l2_cache_reset(BlockDriverState *bs)
403 BDRVQcowState *s = bs->opaque;
405 memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
406 memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
407 memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
410 static inline int l2_cache_new_entry(BlockDriverState *bs)
412 BDRVQcowState *s = bs->opaque;
413 uint32_t min_count;
414 int min_index, i;
416 /* find a new entry in the least used one */
417 min_index = 0;
418 min_count = 0xffffffff;
419 for(i = 0; i < L2_CACHE_SIZE; i++) {
420 if (s->l2_cache_counts[i] < min_count) {
421 min_count = s->l2_cache_counts[i];
422 min_index = i;
425 return min_index;
428 static int64_t align_offset(int64_t offset, int n)
430 offset = (offset + n - 1) & ~(n - 1);
431 return offset;
434 static int grow_l1_table(BlockDriverState *bs, int min_size)
436 BDRVQcowState *s = bs->opaque;
437 int new_l1_size, new_l1_size2, ret, i;
438 uint64_t *new_l1_table;
439 uint64_t new_l1_table_offset;
440 uint8_t data[12];
442 new_l1_size = s->l1_size;
443 if (min_size <= new_l1_size)
444 return 0;
445 while (min_size > new_l1_size) {
446 new_l1_size = (new_l1_size * 3 + 1) / 2;
448 #ifdef DEBUG_ALLOC2
449 printf("grow l1_table from %d to %d\n", s->l1_size, new_l1_size);
450 #endif
452 new_l1_size2 = sizeof(uint64_t) * new_l1_size;
453 new_l1_table = qemu_mallocz(new_l1_size2);
454 memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t));
456 /* write new table (align to cluster) */
457 new_l1_table_offset = alloc_clusters(bs, new_l1_size2);
459 for(i = 0; i < s->l1_size; i++)
460 new_l1_table[i] = cpu_to_be64(new_l1_table[i]);
461 ret = bdrv_pwrite(s->hd, new_l1_table_offset, new_l1_table, new_l1_size2);
462 if (ret != new_l1_size2)
463 goto fail;
464 for(i = 0; i < s->l1_size; i++)
465 new_l1_table[i] = be64_to_cpu(new_l1_table[i]);
467 /* set new table */
468 cpu_to_be32w((uint32_t*)data, new_l1_size);
469 cpu_to_be64w((uint64_t*)(data + 4), new_l1_table_offset);
470 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, l1_size), data,
471 sizeof(data)) != sizeof(data))
472 goto fail;
473 qemu_free(s->l1_table);
474 free_clusters(bs, s->l1_table_offset, s->l1_size * sizeof(uint64_t));
475 s->l1_table_offset = new_l1_table_offset;
476 s->l1_table = new_l1_table;
477 s->l1_size = new_l1_size;
478 return 0;
479 fail:
480 qemu_free(s->l1_table);
481 return -EIO;
485 * seek_l2_table
487 * seek l2_offset in the l2_cache table
488 * if not found, return NULL,
489 * if found,
490 * increments the l2 cache hit count of the entry,
491 * if counter overflow, divide by two all counters
492 * return the pointer to the l2 cache entry
496 static uint64_t *seek_l2_table(BDRVQcowState *s, uint64_t l2_offset)
498 int i, j;
500 for(i = 0; i < L2_CACHE_SIZE; i++) {
501 if (l2_offset == s->l2_cache_offsets[i]) {
502 /* increment the hit count */
503 if (++s->l2_cache_counts[i] == 0xffffffff) {
504 for(j = 0; j < L2_CACHE_SIZE; j++) {
505 s->l2_cache_counts[j] >>= 1;
508 return s->l2_cache + (i << s->l2_bits);
511 return NULL;
515 * l2_load
517 * Loads a L2 table into memory. If the table is in the cache, the cache
518 * is used; otherwise the L2 table is loaded from the image file.
520 * Returns a pointer to the L2 table on success, or NULL if the read from
521 * the image file failed.
524 static uint64_t *l2_load(BlockDriverState *bs, uint64_t l2_offset)
526 BDRVQcowState *s = bs->opaque;
527 int min_index;
528 uint64_t *l2_table;
530 /* seek if the table for the given offset is in the cache */
532 l2_table = seek_l2_table(s, l2_offset);
533 if (l2_table != NULL)
534 return l2_table;
536 /* not found: load a new entry in the least used one */
538 min_index = l2_cache_new_entry(bs);
539 l2_table = s->l2_cache + (min_index << s->l2_bits);
540 if (bdrv_pread(s->hd, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
541 s->l2_size * sizeof(uint64_t))
542 return NULL;
543 s->l2_cache_offsets[min_index] = l2_offset;
544 s->l2_cache_counts[min_index] = 1;
546 return l2_table;
550 * l2_allocate
552 * Allocate a new l2 entry in the file. If l1_index points to an already
553 * used entry in the L2 table (i.e. we are doing a copy on write for the L2
554 * table) copy the contents of the old L2 table into the newly allocated one.
555 * Otherwise the new table is initialized with zeros.
559 static uint64_t *l2_allocate(BlockDriverState *bs, int l1_index)
561 BDRVQcowState *s = bs->opaque;
562 int min_index;
563 uint64_t old_l2_offset, tmp;
564 uint64_t *l2_table, l2_offset;
566 old_l2_offset = s->l1_table[l1_index];
568 /* allocate a new l2 entry */
570 l2_offset = alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
572 /* update the L1 entry */
574 s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
576 tmp = cpu_to_be64(l2_offset | QCOW_OFLAG_COPIED);
577 if (bdrv_pwrite(s->hd, s->l1_table_offset + l1_index * sizeof(tmp),
578 &tmp, sizeof(tmp)) != sizeof(tmp))
579 return NULL;
581 /* allocate a new entry in the l2 cache */
583 min_index = l2_cache_new_entry(bs);
584 l2_table = s->l2_cache + (min_index << s->l2_bits);
586 if (old_l2_offset == 0) {
587 /* if there was no old l2 table, clear the new table */
588 memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
589 } else {
590 /* if there was an old l2 table, read it from the disk */
591 if (bdrv_pread(s->hd, old_l2_offset,
592 l2_table, s->l2_size * sizeof(uint64_t)) !=
593 s->l2_size * sizeof(uint64_t))
594 return NULL;
596 /* write the l2 table to the file */
597 if (bdrv_pwrite(s->hd, l2_offset,
598 l2_table, s->l2_size * sizeof(uint64_t)) !=
599 s->l2_size * sizeof(uint64_t))
600 return NULL;
602 /* update the l2 cache entry */
604 s->l2_cache_offsets[min_index] = l2_offset;
605 s->l2_cache_counts[min_index] = 1;
607 return l2_table;
610 static int size_to_clusters(BDRVQcowState *s, int64_t size)
612 return (size + (s->cluster_size - 1)) >> s->cluster_bits;
615 static int count_contiguous_clusters(uint64_t nb_clusters, int cluster_size,
616 uint64_t *l2_table, uint64_t start, uint64_t mask)
618 int i;
619 uint64_t offset = be64_to_cpu(l2_table[0]) & ~mask;
621 if (!offset)
622 return 0;
624 for (i = start; i < start + nb_clusters; i++)
625 if (offset + i * cluster_size != (be64_to_cpu(l2_table[i]) & ~mask))
626 break;
628 return (i - start);
631 static int count_contiguous_free_clusters(uint64_t nb_clusters, uint64_t *l2_table)
633 int i = 0;
635 while(nb_clusters-- && l2_table[i] == 0)
636 i++;
638 return i;
642 * get_cluster_offset
644 * For a given offset of the disk image, return cluster offset in
645 * qcow2 file.
647 * on entry, *num is the number of contiguous clusters we'd like to
648 * access following offset.
650 * on exit, *num is the number of contiguous clusters we can read.
652 * Return 1, if the offset is found
653 * Return 0, otherwise.
657 static uint64_t get_cluster_offset(BlockDriverState *bs,
658 uint64_t offset, int *num)
660 BDRVQcowState *s = bs->opaque;
661 int l1_index, l2_index;
662 uint64_t l2_offset, *l2_table, cluster_offset;
663 int l1_bits, c;
664 int index_in_cluster, nb_available, nb_needed, nb_clusters;
666 index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1);
667 nb_needed = *num + index_in_cluster;
669 l1_bits = s->l2_bits + s->cluster_bits;
671 /* compute how many bytes there are between the offset and
672 * the end of the l1 entry
675 nb_available = (1 << l1_bits) - (offset & ((1 << l1_bits) - 1));
677 /* compute the number of available sectors */
679 nb_available = (nb_available >> 9) + index_in_cluster;
681 cluster_offset = 0;
683 /* seek the the l2 offset in the l1 table */
685 l1_index = offset >> l1_bits;
686 if (l1_index >= s->l1_size)
687 goto out;
689 l2_offset = s->l1_table[l1_index];
691 /* seek the l2 table of the given l2 offset */
693 if (!l2_offset)
694 goto out;
696 /* load the l2 table in memory */
698 l2_offset &= ~QCOW_OFLAG_COPIED;
699 l2_table = l2_load(bs, l2_offset);
700 if (l2_table == NULL)
701 return 0;
703 /* find the cluster offset for the given disk offset */
705 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
706 cluster_offset = be64_to_cpu(l2_table[l2_index]);
707 nb_clusters = size_to_clusters(s, nb_needed << 9);
709 if (!cluster_offset) {
710 /* how many empty clusters ? */
711 c = count_contiguous_free_clusters(nb_clusters, &l2_table[l2_index]);
712 } else {
713 /* how many allocated clusters ? */
714 c = count_contiguous_clusters(nb_clusters, s->cluster_size,
715 &l2_table[l2_index], 0, QCOW_OFLAG_COPIED);
718 nb_available = (c * s->cluster_sectors);
719 out:
720 if (nb_available > nb_needed)
721 nb_available = nb_needed;
723 *num = nb_available - index_in_cluster;
725 return cluster_offset & ~QCOW_OFLAG_COPIED;
729 * free_any_clusters
731 * free clusters according to its type: compressed or not
735 static void free_any_clusters(BlockDriverState *bs,
736 uint64_t cluster_offset, int nb_clusters)
738 BDRVQcowState *s = bs->opaque;
740 /* free the cluster */
742 if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
743 int nb_csectors;
744 nb_csectors = ((cluster_offset >> s->csize_shift) &
745 s->csize_mask) + 1;
746 free_clusters(bs, (cluster_offset & s->cluster_offset_mask) & ~511,
747 nb_csectors * 512);
748 return;
751 free_clusters(bs, cluster_offset, nb_clusters << s->cluster_bits);
753 return;
757 * get_cluster_table
759 * for a given disk offset, load (and allocate if needed)
760 * the l2 table.
762 * the l2 table offset in the qcow2 file and the cluster index
763 * in the l2 table are given to the caller.
767 static int get_cluster_table(BlockDriverState *bs, uint64_t offset,
768 uint64_t **new_l2_table,
769 uint64_t *new_l2_offset,
770 int *new_l2_index)
772 BDRVQcowState *s = bs->opaque;
773 int l1_index, l2_index, ret;
774 uint64_t l2_offset, *l2_table;
776 /* seek the the l2 offset in the l1 table */
778 l1_index = offset >> (s->l2_bits + s->cluster_bits);
779 if (l1_index >= s->l1_size) {
780 ret = grow_l1_table(bs, l1_index + 1);
781 if (ret < 0)
782 return 0;
784 l2_offset = s->l1_table[l1_index];
786 /* seek the l2 table of the given l2 offset */
788 if (l2_offset & QCOW_OFLAG_COPIED) {
789 /* load the l2 table in memory */
790 l2_offset &= ~QCOW_OFLAG_COPIED;
791 l2_table = l2_load(bs, l2_offset);
792 if (l2_table == NULL)
793 return 0;
794 } else {
795 if (l2_offset)
796 free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t));
797 l2_table = l2_allocate(bs, l1_index);
798 if (l2_table == NULL)
799 return 0;
800 l2_offset = s->l1_table[l1_index] & ~QCOW_OFLAG_COPIED;
803 /* find the cluster offset for the given disk offset */
805 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
807 *new_l2_table = l2_table;
808 *new_l2_offset = l2_offset;
809 *new_l2_index = l2_index;
811 return 1;
815 * alloc_compressed_cluster_offset
817 * For a given offset of the disk image, return cluster offset in
818 * qcow2 file.
820 * If the offset is not found, allocate a new compressed cluster.
822 * Return the cluster offset if successful,
823 * Return 0, otherwise.
827 static uint64_t alloc_compressed_cluster_offset(BlockDriverState *bs,
828 uint64_t offset,
829 int compressed_size)
831 BDRVQcowState *s = bs->opaque;
832 int l2_index, ret;
833 uint64_t l2_offset, *l2_table, cluster_offset;
834 int nb_csectors;
836 ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
837 if (ret == 0)
838 return 0;
840 cluster_offset = be64_to_cpu(l2_table[l2_index]);
841 if (cluster_offset & QCOW_OFLAG_COPIED)
842 return cluster_offset & ~QCOW_OFLAG_COPIED;
844 if (cluster_offset)
845 free_any_clusters(bs, cluster_offset, 1);
847 cluster_offset = alloc_bytes(bs, compressed_size);
848 nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) -
849 (cluster_offset >> 9);
851 cluster_offset |= QCOW_OFLAG_COMPRESSED |
852 ((uint64_t)nb_csectors << s->csize_shift);
854 /* update L2 table */
856 /* compressed clusters never have the copied flag */
858 l2_table[l2_index] = cpu_to_be64(cluster_offset);
859 if (bdrv_pwrite(s->hd,
860 l2_offset + l2_index * sizeof(uint64_t),
861 l2_table + l2_index,
862 sizeof(uint64_t)) != sizeof(uint64_t))
863 return 0;
865 return cluster_offset;
868 typedef struct QCowL2Meta
870 uint64_t offset;
871 int n_start;
872 int nb_available;
873 int nb_clusters;
874 } QCowL2Meta;
876 static int alloc_cluster_link_l2(BlockDriverState *bs, uint64_t cluster_offset,
877 QCowL2Meta *m)
879 BDRVQcowState *s = bs->opaque;
880 int i, j = 0, l2_index, ret;
881 uint64_t *old_cluster, start_sect, l2_offset, *l2_table;
883 if (m->nb_clusters == 0)
884 return 0;
886 old_cluster = qemu_malloc(m->nb_clusters * sizeof(uint64_t));
888 /* copy content of unmodified sectors */
889 start_sect = (m->offset & ~(s->cluster_size - 1)) >> 9;
890 if (m->n_start) {
891 ret = copy_sectors(bs, start_sect, cluster_offset, 0, m->n_start);
892 if (ret < 0)
893 goto err;
896 if (m->nb_available & (s->cluster_sectors - 1)) {
897 uint64_t end = m->nb_available & ~(uint64_t)(s->cluster_sectors - 1);
898 ret = copy_sectors(bs, start_sect + end, cluster_offset + (end << 9),
899 m->nb_available - end, s->cluster_sectors);
900 if (ret < 0)
901 goto err;
904 ret = -EIO;
905 /* update L2 table */
906 if (!get_cluster_table(bs, m->offset, &l2_table, &l2_offset, &l2_index))
907 goto err;
909 for (i = 0; i < m->nb_clusters; i++) {
910 if(l2_table[l2_index + i] != 0)
911 old_cluster[j++] = l2_table[l2_index + i];
913 l2_table[l2_index + i] = cpu_to_be64((cluster_offset +
914 (i << s->cluster_bits)) | QCOW_OFLAG_COPIED);
917 if (bdrv_pwrite(s->hd, l2_offset + l2_index * sizeof(uint64_t),
918 l2_table + l2_index, m->nb_clusters * sizeof(uint64_t)) !=
919 m->nb_clusters * sizeof(uint64_t))
920 goto err;
922 for (i = 0; i < j; i++)
923 free_any_clusters(bs, old_cluster[i], 1);
925 ret = 0;
926 err:
927 qemu_free(old_cluster);
928 return ret;
932 * alloc_cluster_offset
934 * For a given offset of the disk image, return cluster offset in
935 * qcow2 file.
937 * If the offset is not found, allocate a new cluster.
939 * Return the cluster offset if successful,
940 * Return 0, otherwise.
944 static uint64_t alloc_cluster_offset(BlockDriverState *bs,
945 uint64_t offset,
946 int n_start, int n_end,
947 int *num, QCowL2Meta *m)
949 BDRVQcowState *s = bs->opaque;
950 int l2_index, ret;
951 uint64_t l2_offset, *l2_table, cluster_offset;
952 int nb_clusters, i = 0;
954 ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
955 if (ret == 0)
956 return 0;
958 nb_clusters = size_to_clusters(s, n_end << 9);
960 nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
962 cluster_offset = be64_to_cpu(l2_table[l2_index]);
964 /* We keep all QCOW_OFLAG_COPIED clusters */
966 if (cluster_offset & QCOW_OFLAG_COPIED) {
967 nb_clusters = count_contiguous_clusters(nb_clusters, s->cluster_size,
968 &l2_table[l2_index], 0, 0);
970 cluster_offset &= ~QCOW_OFLAG_COPIED;
971 m->nb_clusters = 0;
973 goto out;
976 /* for the moment, multiple compressed clusters are not managed */
978 if (cluster_offset & QCOW_OFLAG_COMPRESSED)
979 nb_clusters = 1;
981 /* how many available clusters ? */
983 while (i < nb_clusters) {
984 i += count_contiguous_clusters(nb_clusters - i, s->cluster_size,
985 &l2_table[l2_index], i, 0);
987 if(be64_to_cpu(l2_table[l2_index + i]))
988 break;
990 i += count_contiguous_free_clusters(nb_clusters - i,
991 &l2_table[l2_index + i]);
993 cluster_offset = be64_to_cpu(l2_table[l2_index + i]);
995 if ((cluster_offset & QCOW_OFLAG_COPIED) ||
996 (cluster_offset & QCOW_OFLAG_COMPRESSED))
997 break;
999 nb_clusters = i;
1001 /* allocate a new cluster */
1003 cluster_offset = alloc_clusters(bs, nb_clusters * s->cluster_size);
1005 /* save info needed for meta data update */
1006 m->offset = offset;
1007 m->n_start = n_start;
1008 m->nb_clusters = nb_clusters;
1010 out:
1011 m->nb_available = MIN(nb_clusters << (s->cluster_bits - 9), n_end);
1013 *num = m->nb_available - n_start;
1015 return cluster_offset;
1018 static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
1019 int nb_sectors, int *pnum)
1021 uint64_t cluster_offset;
1023 *pnum = nb_sectors;
1024 cluster_offset = get_cluster_offset(bs, sector_num << 9, pnum);
1026 return (cluster_offset != 0);
1029 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
1030 const uint8_t *buf, int buf_size)
1032 z_stream strm1, *strm = &strm1;
1033 int ret, out_len;
1035 memset(strm, 0, sizeof(*strm));
1037 strm->next_in = (uint8_t *)buf;
1038 strm->avail_in = buf_size;
1039 strm->next_out = out_buf;
1040 strm->avail_out = out_buf_size;
1042 ret = inflateInit2(strm, -12);
1043 if (ret != Z_OK)
1044 return -1;
1045 ret = inflate(strm, Z_FINISH);
1046 out_len = strm->next_out - out_buf;
1047 if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
1048 out_len != out_buf_size) {
1049 inflateEnd(strm);
1050 return -1;
1052 inflateEnd(strm);
1053 return 0;
1056 static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset)
1058 int ret, csize, nb_csectors, sector_offset;
1059 uint64_t coffset;
1061 coffset = cluster_offset & s->cluster_offset_mask;
1062 if (s->cluster_cache_offset != coffset) {
1063 nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
1064 sector_offset = coffset & 511;
1065 csize = nb_csectors * 512 - sector_offset;
1066 ret = bdrv_read(s->hd, coffset >> 9, s->cluster_data, nb_csectors);
1067 if (ret < 0) {
1068 return -1;
1070 if (decompress_buffer(s->cluster_cache, s->cluster_size,
1071 s->cluster_data + sector_offset, csize) < 0) {
1072 return -1;
1074 s->cluster_cache_offset = coffset;
1076 return 0;
1079 /* handle reading after the end of the backing file */
1080 static int backing_read1(BlockDriverState *bs,
1081 int64_t sector_num, uint8_t *buf, int nb_sectors)
1083 int n1;
1084 if ((sector_num + nb_sectors) <= bs->total_sectors)
1085 return nb_sectors;
1086 if (sector_num >= bs->total_sectors)
1087 n1 = 0;
1088 else
1089 n1 = bs->total_sectors - sector_num;
1090 memset(buf + n1 * 512, 0, 512 * (nb_sectors - n1));
1091 return n1;
1094 static int qcow_read(BlockDriverState *bs, int64_t sector_num,
1095 uint8_t *buf, int nb_sectors)
1097 BDRVQcowState *s = bs->opaque;
1098 int ret, index_in_cluster, n, n1;
1099 uint64_t cluster_offset;
1101 while (nb_sectors > 0) {
1102 n = nb_sectors;
1103 cluster_offset = get_cluster_offset(bs, sector_num << 9, &n);
1104 index_in_cluster = sector_num & (s->cluster_sectors - 1);
1105 if (!cluster_offset) {
1106 if (bs->backing_hd) {
1107 /* read from the base image */
1108 n1 = backing_read1(bs->backing_hd, sector_num, buf, n);
1109 if (n1 > 0) {
1110 ret = bdrv_read(bs->backing_hd, sector_num, buf, n1);
1111 if (ret < 0)
1112 return -1;
1114 } else {
1115 memset(buf, 0, 512 * n);
1117 } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
1118 if (decompress_cluster(s, cluster_offset) < 0)
1119 return -1;
1120 memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n);
1121 } else {
1122 ret = bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
1123 if (ret != n * 512)
1124 return -1;
1125 if (s->crypt_method) {
1126 encrypt_sectors(s, sector_num, buf, buf, n, 0,
1127 &s->aes_decrypt_key);
1130 nb_sectors -= n;
1131 sector_num += n;
1132 buf += n * 512;
1134 return 0;
1137 static int qcow_write(BlockDriverState *bs, int64_t sector_num,
1138 const uint8_t *buf, int nb_sectors)
1140 BDRVQcowState *s = bs->opaque;
1141 int ret, index_in_cluster, n;
1142 uint64_t cluster_offset;
1143 int n_end;
1144 QCowL2Meta l2meta;
1146 while (nb_sectors > 0) {
1147 index_in_cluster = sector_num & (s->cluster_sectors - 1);
1148 n_end = index_in_cluster + nb_sectors;
1149 if (s->crypt_method &&
1150 n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
1151 n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
1152 cluster_offset = alloc_cluster_offset(bs, sector_num << 9,
1153 index_in_cluster,
1154 n_end, &n, &l2meta);
1155 if (!cluster_offset)
1156 return -1;
1157 if (s->crypt_method) {
1158 encrypt_sectors(s, sector_num, s->cluster_data, buf, n, 1,
1159 &s->aes_encrypt_key);
1160 ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512,
1161 s->cluster_data, n * 512);
1162 } else {
1163 ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
1165 if (ret != n * 512 || alloc_cluster_link_l2(bs, cluster_offset, &l2meta) < 0) {
1166 free_any_clusters(bs, cluster_offset, l2meta.nb_clusters);
1167 return -1;
1169 nb_sectors -= n;
1170 sector_num += n;
1171 buf += n * 512;
1173 s->cluster_cache_offset = -1; /* disable compressed cache */
1174 return 0;
1177 typedef struct QCowAIOCB {
1178 BlockDriverAIOCB common;
1179 int64_t sector_num;
1180 uint8_t *buf;
1181 int nb_sectors;
1182 int n;
1183 uint64_t cluster_offset;
1184 uint8_t *cluster_data;
1185 BlockDriverAIOCB *hd_aiocb;
1186 QEMUBH *bh;
1187 QCowL2Meta l2meta;
1188 } QCowAIOCB;
1190 static void qcow_aio_read_cb(void *opaque, int ret);
1191 static void qcow_aio_read_bh(void *opaque)
1193 QCowAIOCB *acb = opaque;
1194 qemu_bh_delete(acb->bh);
1195 acb->bh = NULL;
1196 qcow_aio_read_cb(opaque, 0);
1199 static int qcow_schedule_bh(QEMUBHFunc *cb, QCowAIOCB *acb)
1201 if (acb->bh)
1202 return -EIO;
1204 acb->bh = qemu_bh_new(cb, acb);
1205 if (!acb->bh)
1206 return -EIO;
1208 qemu_bh_schedule(acb->bh);
1210 return 0;
1213 static void qcow_aio_read_cb(void *opaque, int ret)
1215 QCowAIOCB *acb = opaque;
1216 BlockDriverState *bs = acb->common.bs;
1217 BDRVQcowState *s = bs->opaque;
1218 int index_in_cluster, n1;
1220 acb->hd_aiocb = NULL;
1221 if (ret < 0) {
1222 fail:
1223 acb->common.cb(acb->common.opaque, ret);
1224 qemu_aio_release(acb);
1225 return;
1228 /* post process the read buffer */
1229 if (!acb->cluster_offset) {
1230 /* nothing to do */
1231 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
1232 /* nothing to do */
1233 } else {
1234 if (s->crypt_method) {
1235 encrypt_sectors(s, acb->sector_num, acb->buf, acb->buf,
1236 acb->n, 0,
1237 &s->aes_decrypt_key);
1241 acb->nb_sectors -= acb->n;
1242 acb->sector_num += acb->n;
1243 acb->buf += acb->n * 512;
1245 if (acb->nb_sectors == 0) {
1246 /* request completed */
1247 acb->common.cb(acb->common.opaque, 0);
1248 qemu_aio_release(acb);
1249 return;
1252 /* prepare next AIO request */
1253 acb->n = acb->nb_sectors;
1254 acb->cluster_offset = get_cluster_offset(bs, acb->sector_num << 9, &acb->n);
1255 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
1257 if (!acb->cluster_offset) {
1258 if (bs->backing_hd) {
1259 /* read from the base image */
1260 n1 = backing_read1(bs->backing_hd, acb->sector_num,
1261 acb->buf, acb->n);
1262 if (n1 > 0) {
1263 acb->hd_aiocb = bdrv_aio_read(bs->backing_hd, acb->sector_num,
1264 acb->buf, acb->n, qcow_aio_read_cb, acb);
1265 if (acb->hd_aiocb == NULL)
1266 goto fail;
1267 } else {
1268 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
1269 if (ret < 0)
1270 goto fail;
1272 } else {
1273 /* Note: in this case, no need to wait */
1274 memset(acb->buf, 0, 512 * acb->n);
1275 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
1276 if (ret < 0)
1277 goto fail;
1279 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
1280 /* add AIO support for compressed blocks ? */
1281 if (decompress_cluster(s, acb->cluster_offset) < 0)
1282 goto fail;
1283 memcpy(acb->buf,
1284 s->cluster_cache + index_in_cluster * 512, 512 * acb->n);
1285 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
1286 if (ret < 0)
1287 goto fail;
1288 } else {
1289 if ((acb->cluster_offset & 511) != 0) {
1290 ret = -EIO;
1291 goto fail;
1293 acb->hd_aiocb = bdrv_aio_read(s->hd,
1294 (acb->cluster_offset >> 9) + index_in_cluster,
1295 acb->buf, acb->n, qcow_aio_read_cb, acb);
1296 if (acb->hd_aiocb == NULL)
1297 goto fail;
1301 static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs,
1302 int64_t sector_num, uint8_t *buf, int nb_sectors,
1303 BlockDriverCompletionFunc *cb, void *opaque)
1305 QCowAIOCB *acb;
1307 acb = qemu_aio_get(bs, cb, opaque);
1308 if (!acb)
1309 return NULL;
1310 acb->hd_aiocb = NULL;
1311 acb->sector_num = sector_num;
1312 acb->buf = buf;
1313 acb->nb_sectors = nb_sectors;
1314 acb->n = 0;
1315 acb->cluster_offset = 0;
1316 acb->l2meta.nb_clusters = 0;
1317 return acb;
1320 static BlockDriverAIOCB *qcow_aio_read(BlockDriverState *bs,
1321 int64_t sector_num, uint8_t *buf, int nb_sectors,
1322 BlockDriverCompletionFunc *cb, void *opaque)
1324 QCowAIOCB *acb;
1326 acb = qcow_aio_setup(bs, sector_num, buf, nb_sectors, cb, opaque);
1327 if (!acb)
1328 return NULL;
1330 qcow_aio_read_cb(acb, 0);
1331 return &acb->common;
1334 static void qcow_aio_write_cb(void *opaque, int ret)
1336 QCowAIOCB *acb = opaque;
1337 BlockDriverState *bs = acb->common.bs;
1338 BDRVQcowState *s = bs->opaque;
1339 int index_in_cluster;
1340 const uint8_t *src_buf;
1341 int n_end;
1343 acb->hd_aiocb = NULL;
1345 if (ret < 0) {
1346 fail:
1347 acb->common.cb(acb->common.opaque, ret);
1348 qemu_aio_release(acb);
1349 return;
1352 if (alloc_cluster_link_l2(bs, acb->cluster_offset, &acb->l2meta) < 0) {
1353 free_any_clusters(bs, acb->cluster_offset, acb->l2meta.nb_clusters);
1354 goto fail;
1357 acb->nb_sectors -= acb->n;
1358 acb->sector_num += acb->n;
1359 acb->buf += acb->n * 512;
1361 if (acb->nb_sectors == 0) {
1362 /* request completed */
1363 acb->common.cb(acb->common.opaque, 0);
1364 qemu_aio_release(acb);
1365 return;
1368 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
1369 n_end = index_in_cluster + acb->nb_sectors;
1370 if (s->crypt_method &&
1371 n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
1372 n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
1374 acb->cluster_offset = alloc_cluster_offset(bs, acb->sector_num << 9,
1375 index_in_cluster,
1376 n_end, &acb->n, &acb->l2meta);
1377 if (!acb->cluster_offset || (acb->cluster_offset & 511) != 0) {
1378 ret = -EIO;
1379 goto fail;
1381 if (s->crypt_method) {
1382 if (!acb->cluster_data) {
1383 acb->cluster_data = qemu_mallocz(QCOW_MAX_CRYPT_CLUSTERS *
1384 s->cluster_size);
1386 encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf,
1387 acb->n, 1, &s->aes_encrypt_key);
1388 src_buf = acb->cluster_data;
1389 } else {
1390 src_buf = acb->buf;
1392 acb->hd_aiocb = bdrv_aio_write(s->hd,
1393 (acb->cluster_offset >> 9) + index_in_cluster,
1394 src_buf, acb->n,
1395 qcow_aio_write_cb, acb);
1396 if (acb->hd_aiocb == NULL)
1397 goto fail;
1400 static BlockDriverAIOCB *qcow_aio_write(BlockDriverState *bs,
1401 int64_t sector_num, const uint8_t *buf, int nb_sectors,
1402 BlockDriverCompletionFunc *cb, void *opaque)
1404 BDRVQcowState *s = bs->opaque;
1405 QCowAIOCB *acb;
1407 s->cluster_cache_offset = -1; /* disable compressed cache */
1409 acb = qcow_aio_setup(bs, sector_num, (uint8_t*)buf, nb_sectors, cb, opaque);
1410 if (!acb)
1411 return NULL;
1413 qcow_aio_write_cb(acb, 0);
1414 return &acb->common;
1417 static void qcow_aio_cancel(BlockDriverAIOCB *blockacb)
1419 QCowAIOCB *acb = (QCowAIOCB *)blockacb;
1420 if (acb->hd_aiocb)
1421 bdrv_aio_cancel(acb->hd_aiocb);
1422 qemu_aio_release(acb);
1425 static void qcow_close(BlockDriverState *bs)
1427 BDRVQcowState *s = bs->opaque;
1428 qemu_free(s->l1_table);
1429 qemu_free(s->l2_cache);
1430 qemu_free(s->cluster_cache);
1431 qemu_free(s->cluster_data);
1432 refcount_close(bs);
1433 bdrv_delete(s->hd);
1436 /* XXX: use std qcow open function ? */
1437 typedef struct QCowCreateState {
1438 int cluster_size;
1439 int cluster_bits;
1440 uint16_t *refcount_block;
1441 uint64_t *refcount_table;
1442 int64_t l1_table_offset;
1443 int64_t refcount_table_offset;
1444 int64_t refcount_block_offset;
1445 } QCowCreateState;
1447 static void create_refcount_update(QCowCreateState *s,
1448 int64_t offset, int64_t size)
1450 int refcount;
1451 int64_t start, last, cluster_offset;
1452 uint16_t *p;
1454 start = offset & ~(s->cluster_size - 1);
1455 last = (offset + size - 1) & ~(s->cluster_size - 1);
1456 for(cluster_offset = start; cluster_offset <= last;
1457 cluster_offset += s->cluster_size) {
1458 p = &s->refcount_block[cluster_offset >> s->cluster_bits];
1459 refcount = be16_to_cpu(*p);
1460 refcount++;
1461 *p = cpu_to_be16(refcount);
1465 static int qcow_create(const char *filename, int64_t total_size,
1466 const char *backing_file, int flags)
1468 int fd, header_size, backing_filename_len, l1_size, i, shift, l2_bits;
1469 QCowHeader header;
1470 uint64_t tmp, offset;
1471 QCowCreateState s1, *s = &s1;
1473 memset(s, 0, sizeof(*s));
1475 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
1476 if (fd < 0)
1477 return -1;
1478 memset(&header, 0, sizeof(header));
1479 header.magic = cpu_to_be32(QCOW_MAGIC);
1480 header.version = cpu_to_be32(QCOW_VERSION);
1481 header.size = cpu_to_be64(total_size * 512);
1482 header_size = sizeof(header);
1483 backing_filename_len = 0;
1484 if (backing_file) {
1485 header.backing_file_offset = cpu_to_be64(header_size);
1486 backing_filename_len = strlen(backing_file);
1487 header.backing_file_size = cpu_to_be32(backing_filename_len);
1488 header_size += backing_filename_len;
1490 s->cluster_bits = 12; /* 4 KB clusters */
1491 s->cluster_size = 1 << s->cluster_bits;
1492 header.cluster_bits = cpu_to_be32(s->cluster_bits);
1493 header_size = (header_size + 7) & ~7;
1494 if (flags & BLOCK_FLAG_ENCRYPT) {
1495 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
1496 } else {
1497 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
1499 l2_bits = s->cluster_bits - 3;
1500 shift = s->cluster_bits + l2_bits;
1501 l1_size = (((total_size * 512) + (1LL << shift) - 1) >> shift);
1502 offset = align_offset(header_size, s->cluster_size);
1503 s->l1_table_offset = offset;
1504 header.l1_table_offset = cpu_to_be64(s->l1_table_offset);
1505 header.l1_size = cpu_to_be32(l1_size);
1506 offset += align_offset(l1_size * sizeof(uint64_t), s->cluster_size);
1508 s->refcount_table = qemu_mallocz(s->cluster_size);
1509 s->refcount_block = qemu_mallocz(s->cluster_size);
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;
1548 static int qcow_make_empty(BlockDriverState *bs)
1550 #if 0
1551 /* XXX: not correct */
1552 BDRVQcowState *s = bs->opaque;
1553 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1554 int ret;
1556 memset(s->l1_table, 0, l1_length);
1557 if (bdrv_pwrite(s->hd, s->l1_table_offset, s->l1_table, l1_length) < 0)
1558 return -1;
1559 ret = bdrv_truncate(s->hd, s->l1_table_offset + l1_length);
1560 if (ret < 0)
1561 return ret;
1563 l2_cache_reset(bs);
1564 #endif
1565 return 0;
1568 /* XXX: put compressed sectors first, then all the cluster aligned
1569 tables to avoid losing bytes in alignment */
1570 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
1571 const uint8_t *buf, int nb_sectors)
1573 BDRVQcowState *s = bs->opaque;
1574 z_stream strm;
1575 int ret, out_len;
1576 uint8_t *out_buf;
1577 uint64_t cluster_offset;
1579 if (nb_sectors == 0) {
1580 /* align end of file to a sector boundary to ease reading with
1581 sector based I/Os */
1582 cluster_offset = bdrv_getlength(s->hd);
1583 cluster_offset = (cluster_offset + 511) & ~511;
1584 bdrv_truncate(s->hd, cluster_offset);
1585 return 0;
1588 if (nb_sectors != s->cluster_sectors)
1589 return -EINVAL;
1591 out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
1593 /* best compression, small window, no zlib header */
1594 memset(&strm, 0, sizeof(strm));
1595 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1596 Z_DEFLATED, -12,
1597 9, Z_DEFAULT_STRATEGY);
1598 if (ret != 0) {
1599 qemu_free(out_buf);
1600 return -1;
1603 strm.avail_in = s->cluster_size;
1604 strm.next_in = (uint8_t *)buf;
1605 strm.avail_out = s->cluster_size;
1606 strm.next_out = out_buf;
1608 ret = deflate(&strm, Z_FINISH);
1609 if (ret != Z_STREAM_END && ret != Z_OK) {
1610 qemu_free(out_buf);
1611 deflateEnd(&strm);
1612 return -1;
1614 out_len = strm.next_out - out_buf;
1616 deflateEnd(&strm);
1618 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1619 /* could not compress: write normal cluster */
1620 qcow_write(bs, sector_num, buf, s->cluster_sectors);
1621 } else {
1622 cluster_offset = alloc_compressed_cluster_offset(bs, sector_num << 9,
1623 out_len);
1624 if (!cluster_offset)
1625 return -1;
1626 cluster_offset &= s->cluster_offset_mask;
1627 if (bdrv_pwrite(s->hd, cluster_offset, out_buf, out_len) != out_len) {
1628 qemu_free(out_buf);
1629 return -1;
1633 qemu_free(out_buf);
1634 return 0;
1637 static void qcow_flush(BlockDriverState *bs)
1639 BDRVQcowState *s = bs->opaque;
1640 bdrv_flush(s->hd);
1643 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1645 BDRVQcowState *s = bs->opaque;
1646 bdi->cluster_size = s->cluster_size;
1647 bdi->vm_state_offset = (int64_t)s->l1_vm_state_index <<
1648 (s->cluster_bits + s->l2_bits);
1649 bdi->highest_alloc = s->highest_alloc << s->cluster_bits;
1650 bdi->num_free_bytes = s->nc_free << s->cluster_bits;
1651 return 0;
1654 /*********************************************************/
1655 /* snapshot support */
1657 /* update the refcounts of snapshots and the copied flag */
1658 static int update_snapshot_refcount(BlockDriverState *bs,
1659 int64_t l1_table_offset,
1660 int l1_size,
1661 int addend)
1663 BDRVQcowState *s = bs->opaque;
1664 uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated;
1665 int64_t old_offset, old_l2_offset;
1666 int l2_size, i, j, l1_modified, l2_modified, nb_csectors, refcount;
1668 l2_cache_reset(bs);
1670 l2_table = NULL;
1671 l1_table = NULL;
1672 l1_size2 = l1_size * sizeof(uint64_t);
1673 l1_allocated = 0;
1674 if (l1_table_offset != s->l1_table_offset) {
1675 l1_table = qemu_malloc(l1_size2);
1676 l1_allocated = 1;
1677 if (bdrv_pread(s->hd, l1_table_offset,
1678 l1_table, l1_size2) != l1_size2)
1679 goto fail;
1680 for(i = 0;i < l1_size; i++)
1681 be64_to_cpus(&l1_table[i]);
1682 } else {
1683 assert(l1_size == s->l1_size);
1684 l1_table = s->l1_table;
1685 l1_allocated = 0;
1688 l2_size = s->l2_size * sizeof(uint64_t);
1689 l2_table = qemu_malloc(l2_size);
1690 l1_modified = 0;
1691 for(i = 0; i < l1_size; i++) {
1692 l2_offset = l1_table[i];
1693 if (l2_offset) {
1694 old_l2_offset = l2_offset;
1695 l2_offset &= ~QCOW_OFLAG_COPIED;
1696 l2_modified = 0;
1697 if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
1698 goto fail;
1699 for(j = 0; j < s->l2_size; j++) {
1700 offset = be64_to_cpu(l2_table[j]);
1701 if (offset != 0) {
1702 old_offset = offset;
1703 offset &= ~QCOW_OFLAG_COPIED;
1704 if (offset & QCOW_OFLAG_COMPRESSED) {
1705 nb_csectors = ((offset >> s->csize_shift) &
1706 s->csize_mask) + 1;
1707 if (addend != 0)
1708 update_refcount(bs, (offset & s->cluster_offset_mask) & ~511,
1709 nb_csectors * 512, addend);
1710 /* compressed clusters are never modified */
1711 refcount = 2;
1712 } else {
1713 if (addend != 0) {
1714 refcount = update_cluster_refcount(bs, offset >> s->cluster_bits, addend);
1715 } else {
1716 refcount = get_refcount(bs, offset >> s->cluster_bits);
1720 if (refcount == 1) {
1721 offset |= QCOW_OFLAG_COPIED;
1723 if (offset != old_offset) {
1724 l2_table[j] = cpu_to_be64(offset);
1725 l2_modified = 1;
1729 if (l2_modified) {
1730 if (bdrv_pwrite(s->hd,
1731 l2_offset, l2_table, l2_size) != l2_size)
1732 goto fail;
1735 if (addend != 0) {
1736 refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend);
1737 } else {
1738 refcount = get_refcount(bs, l2_offset >> s->cluster_bits);
1740 if (refcount == 1) {
1741 l2_offset |= QCOW_OFLAG_COPIED;
1743 if (l2_offset != old_l2_offset) {
1744 l1_table[i] = l2_offset;
1745 l1_modified = 1;
1749 if (l1_modified) {
1750 for(i = 0; i < l1_size; i++)
1751 cpu_to_be64s(&l1_table[i]);
1752 if (bdrv_pwrite(s->hd, l1_table_offset, l1_table,
1753 l1_size2) != l1_size2)
1754 goto fail;
1755 for(i = 0; i < l1_size; i++)
1756 be64_to_cpus(&l1_table[i]);
1758 if (l1_allocated)
1759 qemu_free(l1_table);
1760 qemu_free(l2_table);
1761 return 0;
1762 fail:
1763 if (l1_allocated)
1764 qemu_free(l1_table);
1765 qemu_free(l2_table);
1766 return -EIO;
1769 static void qcow_free_snapshots(BlockDriverState *bs)
1771 BDRVQcowState *s = bs->opaque;
1772 int i;
1774 for(i = 0; i < s->nb_snapshots; i++) {
1775 qemu_free(s->snapshots[i].name);
1776 qemu_free(s->snapshots[i].id_str);
1778 qemu_free(s->snapshots);
1779 s->snapshots = NULL;
1780 s->nb_snapshots = 0;
1783 static int qcow_read_snapshots(BlockDriverState *bs)
1785 BDRVQcowState *s = bs->opaque;
1786 QCowSnapshotHeader h;
1787 QCowSnapshot *sn;
1788 int i, id_str_size, name_size;
1789 int64_t offset;
1790 uint32_t extra_data_size;
1792 if (!s->nb_snapshots) {
1793 s->snapshots = NULL;
1794 s->snapshots_size = 0;
1795 return 0;
1798 offset = s->snapshots_offset;
1799 s->snapshots = qemu_mallocz(s->nb_snapshots * sizeof(QCowSnapshot));
1800 for(i = 0; i < s->nb_snapshots; i++) {
1801 offset = align_offset(offset, 8);
1802 if (bdrv_pread(s->hd, offset, &h, sizeof(h)) != sizeof(h))
1803 goto fail;
1804 offset += sizeof(h);
1805 sn = s->snapshots + i;
1806 sn->l1_table_offset = be64_to_cpu(h.l1_table_offset);
1807 sn->l1_size = be32_to_cpu(h.l1_size);
1808 sn->vm_state_size = be32_to_cpu(h.vm_state_size);
1809 sn->date_sec = be32_to_cpu(h.date_sec);
1810 sn->date_nsec = be32_to_cpu(h.date_nsec);
1811 sn->vm_clock_nsec = be64_to_cpu(h.vm_clock_nsec);
1812 extra_data_size = be32_to_cpu(h.extra_data_size);
1814 id_str_size = be16_to_cpu(h.id_str_size);
1815 name_size = be16_to_cpu(h.name_size);
1817 offset += extra_data_size;
1819 sn->id_str = qemu_malloc(id_str_size + 1);
1820 if (bdrv_pread(s->hd, offset, sn->id_str, id_str_size) != id_str_size)
1821 goto fail;
1822 offset += id_str_size;
1823 sn->id_str[id_str_size] = '\0';
1825 sn->name = qemu_malloc(name_size + 1);
1826 if (bdrv_pread(s->hd, offset, sn->name, name_size) != name_size)
1827 goto fail;
1828 offset += name_size;
1829 sn->name[name_size] = '\0';
1831 s->snapshots_size = offset - s->snapshots_offset;
1832 return 0;
1833 fail:
1834 qcow_free_snapshots(bs);
1835 return -1;
1838 /* add at the end of the file a new list of snapshots */
1839 static int qcow_write_snapshots(BlockDriverState *bs)
1841 BDRVQcowState *s = bs->opaque;
1842 QCowSnapshot *sn;
1843 QCowSnapshotHeader h;
1844 int i, name_size, id_str_size, snapshots_size;
1845 uint64_t data64;
1846 uint32_t data32;
1847 int64_t offset, snapshots_offset;
1849 /* compute the size of the snapshots */
1850 offset = 0;
1851 for(i = 0; i < s->nb_snapshots; i++) {
1852 sn = s->snapshots + i;
1853 offset = align_offset(offset, 8);
1854 offset += sizeof(h);
1855 offset += strlen(sn->id_str);
1856 offset += strlen(sn->name);
1858 snapshots_size = offset;
1860 snapshots_offset = alloc_clusters(bs, snapshots_size);
1861 offset = snapshots_offset;
1863 for(i = 0; i < s->nb_snapshots; i++) {
1864 sn = s->snapshots + i;
1865 memset(&h, 0, sizeof(h));
1866 h.l1_table_offset = cpu_to_be64(sn->l1_table_offset);
1867 h.l1_size = cpu_to_be32(sn->l1_size);
1868 h.vm_state_size = cpu_to_be32(sn->vm_state_size);
1869 h.date_sec = cpu_to_be32(sn->date_sec);
1870 h.date_nsec = cpu_to_be32(sn->date_nsec);
1871 h.vm_clock_nsec = cpu_to_be64(sn->vm_clock_nsec);
1873 id_str_size = strlen(sn->id_str);
1874 name_size = strlen(sn->name);
1875 h.id_str_size = cpu_to_be16(id_str_size);
1876 h.name_size = cpu_to_be16(name_size);
1877 offset = align_offset(offset, 8);
1878 if (bdrv_pwrite(s->hd, offset, &h, sizeof(h)) != sizeof(h))
1879 goto fail;
1880 offset += sizeof(h);
1881 if (bdrv_pwrite(s->hd, offset, sn->id_str, id_str_size) != id_str_size)
1882 goto fail;
1883 offset += id_str_size;
1884 if (bdrv_pwrite(s->hd, offset, sn->name, name_size) != name_size)
1885 goto fail;
1886 offset += name_size;
1889 /* update the various header fields */
1890 data64 = cpu_to_be64(snapshots_offset);
1891 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, snapshots_offset),
1892 &data64, sizeof(data64)) != sizeof(data64))
1893 goto fail;
1894 data32 = cpu_to_be32(s->nb_snapshots);
1895 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, nb_snapshots),
1896 &data32, sizeof(data32)) != sizeof(data32))
1897 goto fail;
1899 /* free the old snapshot table */
1900 free_clusters(bs, s->snapshots_offset, s->snapshots_size);
1901 s->snapshots_offset = snapshots_offset;
1902 s->snapshots_size = snapshots_size;
1903 return 0;
1904 fail:
1905 return -1;
1908 static void find_new_snapshot_id(BlockDriverState *bs,
1909 char *id_str, int id_str_size)
1911 BDRVQcowState *s = bs->opaque;
1912 QCowSnapshot *sn;
1913 int i, id, id_max = 0;
1915 for(i = 0; i < s->nb_snapshots; i++) {
1916 sn = s->snapshots + i;
1917 id = strtoul(sn->id_str, NULL, 10);
1918 if (id > id_max)
1919 id_max = id;
1921 snprintf(id_str, id_str_size, "%d", id_max + 1);
1924 static int find_snapshot_by_id(BlockDriverState *bs, const char *id_str)
1926 BDRVQcowState *s = bs->opaque;
1927 int i;
1929 for(i = 0; i < s->nb_snapshots; i++) {
1930 if (!strcmp(s->snapshots[i].id_str, id_str))
1931 return i;
1933 return -1;
1936 static int find_snapshot_by_id_or_name(BlockDriverState *bs, const char *name)
1938 BDRVQcowState *s = bs->opaque;
1939 int i, ret;
1941 ret = find_snapshot_by_id(bs, name);
1942 if (ret >= 0)
1943 return ret;
1944 for(i = 0; i < s->nb_snapshots; i++) {
1945 if (!strcmp(s->snapshots[i].name, name))
1946 return i;
1948 return -1;
1951 /* if no id is provided, a new one is constructed */
1952 static int qcow_snapshot_create(BlockDriverState *bs,
1953 QEMUSnapshotInfo *sn_info)
1955 BDRVQcowState *s = bs->opaque;
1956 QCowSnapshot *snapshots1, sn1, *sn = &sn1;
1957 int i, ret;
1958 uint64_t *l1_table = NULL;
1960 memset(sn, 0, sizeof(*sn));
1962 if (sn_info->id_str[0] == '\0') {
1963 /* compute a new id */
1964 find_new_snapshot_id(bs, sn_info->id_str, sizeof(sn_info->id_str));
1967 /* check that the ID is unique */
1968 if (find_snapshot_by_id(bs, sn_info->id_str) >= 0)
1969 return -ENOENT;
1971 sn->id_str = qemu_strdup(sn_info->id_str);
1972 if (!sn->id_str)
1973 goto fail;
1974 sn->name = qemu_strdup(sn_info->name);
1975 if (!sn->name)
1976 goto fail;
1977 sn->vm_state_size = sn_info->vm_state_size;
1978 sn->date_sec = sn_info->date_sec;
1979 sn->date_nsec = sn_info->date_nsec;
1980 sn->vm_clock_nsec = sn_info->vm_clock_nsec;
1982 ret = update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1);
1983 if (ret < 0)
1984 goto fail;
1986 /* create the L1 table of the snapshot */
1987 sn->l1_table_offset = alloc_clusters(bs, s->l1_size * sizeof(uint64_t));
1988 sn->l1_size = s->l1_size;
1990 l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
1991 for(i = 0; i < s->l1_size; i++) {
1992 l1_table[i] = cpu_to_be64(s->l1_table[i]);
1994 if (bdrv_pwrite(s->hd, sn->l1_table_offset,
1995 l1_table, s->l1_size * sizeof(uint64_t)) !=
1996 (s->l1_size * sizeof(uint64_t)))
1997 goto fail;
1998 qemu_free(l1_table);
1999 l1_table = NULL;
2001 snapshots1 = qemu_malloc((s->nb_snapshots + 1) * sizeof(QCowSnapshot));
2002 if (s->snapshots) {
2003 memcpy(snapshots1, s->snapshots, s->nb_snapshots * sizeof(QCowSnapshot));
2004 qemu_free(s->snapshots);
2006 s->snapshots = snapshots1;
2007 s->snapshots[s->nb_snapshots++] = *sn;
2009 if (qcow_write_snapshots(bs) < 0)
2010 goto fail;
2011 #ifdef DEBUG_ALLOC
2012 check_refcounts(bs);
2013 #endif
2014 return 0;
2015 fail:
2016 qemu_free(sn->name);
2017 qemu_free(l1_table);
2018 return -1;
2021 /* copy the snapshot 'snapshot_name' into the current disk image */
2022 static int qcow_snapshot_goto(BlockDriverState *bs,
2023 const char *snapshot_id)
2025 BDRVQcowState *s = bs->opaque;
2026 QCowSnapshot *sn;
2027 int i, snapshot_index, l1_size2;
2029 snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
2030 if (snapshot_index < 0)
2031 return -ENOENT;
2032 sn = &s->snapshots[snapshot_index];
2034 if (update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, -1) < 0)
2035 goto fail;
2037 if (grow_l1_table(bs, sn->l1_size) < 0)
2038 goto fail;
2040 s->l1_size = sn->l1_size;
2041 l1_size2 = s->l1_size * sizeof(uint64_t);
2042 /* copy the snapshot l1 table to the current l1 table */
2043 if (bdrv_pread(s->hd, sn->l1_table_offset,
2044 s->l1_table, l1_size2) != l1_size2)
2045 goto fail;
2046 if (bdrv_pwrite(s->hd, s->l1_table_offset,
2047 s->l1_table, l1_size2) != l1_size2)
2048 goto fail;
2049 for(i = 0;i < s->l1_size; i++) {
2050 be64_to_cpus(&s->l1_table[i]);
2053 if (update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1) < 0)
2054 goto fail;
2056 #ifdef DEBUG_ALLOC
2057 check_refcounts(bs);
2058 #endif
2059 return 0;
2060 fail:
2061 return -EIO;
2064 static int qcow_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
2066 BDRVQcowState *s = bs->opaque;
2067 QCowSnapshot *sn;
2068 int snapshot_index, ret;
2070 snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
2071 if (snapshot_index < 0)
2072 return -ENOENT;
2073 sn = &s->snapshots[snapshot_index];
2075 ret = update_snapshot_refcount(bs, sn->l1_table_offset, sn->l1_size, -1);
2076 if (ret < 0)
2077 return ret;
2078 /* must update the copied flag on the current cluster offsets */
2079 ret = update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 0);
2080 if (ret < 0)
2081 return ret;
2082 free_clusters(bs, sn->l1_table_offset, sn->l1_size * sizeof(uint64_t));
2084 qemu_free(sn->id_str);
2085 qemu_free(sn->name);
2086 memmove(sn, sn + 1, (s->nb_snapshots - snapshot_index - 1) * sizeof(*sn));
2087 s->nb_snapshots--;
2088 ret = qcow_write_snapshots(bs);
2089 if (ret < 0) {
2090 /* XXX: restore snapshot if error ? */
2091 return ret;
2093 #ifdef DEBUG_ALLOC
2094 check_refcounts(bs);
2095 #endif
2096 return 0;
2099 static int qcow_snapshot_list(BlockDriverState *bs,
2100 QEMUSnapshotInfo **psn_tab)
2102 BDRVQcowState *s = bs->opaque;
2103 QEMUSnapshotInfo *sn_tab, *sn_info;
2104 QCowSnapshot *sn;
2105 int i;
2107 sn_tab = qemu_mallocz(s->nb_snapshots * sizeof(QEMUSnapshotInfo));
2108 for(i = 0; i < s->nb_snapshots; i++) {
2109 sn_info = sn_tab + i;
2110 sn = s->snapshots + i;
2111 pstrcpy(sn_info->id_str, sizeof(sn_info->id_str),
2112 sn->id_str);
2113 pstrcpy(sn_info->name, sizeof(sn_info->name),
2114 sn->name);
2115 sn_info->vm_state_size = sn->vm_state_size;
2116 sn_info->date_sec = sn->date_sec;
2117 sn_info->date_nsec = sn->date_nsec;
2118 sn_info->vm_clock_nsec = sn->vm_clock_nsec;
2120 *psn_tab = sn_tab;
2121 return s->nb_snapshots;
2124 /*********************************************************/
2125 /* refcount handling */
2127 static int refcount_init(BlockDriverState *bs)
2129 BDRVQcowState *s = bs->opaque;
2130 int ret, refcount_table_size2, i;
2132 s->refcount_block_cache = qemu_malloc(s->cluster_size);
2133 refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t);
2134 s->refcount_table = qemu_malloc(refcount_table_size2);
2135 if (s->refcount_table_size > 0) {
2136 ret = bdrv_pread(s->hd, s->refcount_table_offset,
2137 s->refcount_table, refcount_table_size2);
2138 if (ret != refcount_table_size2)
2139 goto fail;
2140 for(i = 0; i < s->refcount_table_size; i++)
2141 be64_to_cpus(&s->refcount_table[i]);
2143 return 0;
2144 fail:
2145 return -ENOMEM;
2148 static void refcount_close(BlockDriverState *bs)
2150 BDRVQcowState *s = bs->opaque;
2151 qemu_free(s->refcount_block_cache);
2152 qemu_free(s->refcount_table);
2156 static int load_refcount_block(BlockDriverState *bs,
2157 int64_t refcount_block_offset)
2159 BDRVQcowState *s = bs->opaque;
2160 int ret;
2161 ret = bdrv_pread(s->hd, refcount_block_offset, s->refcount_block_cache,
2162 s->cluster_size);
2163 if (ret != s->cluster_size)
2164 return -EIO;
2165 s->refcount_block_cache_offset = refcount_block_offset;
2166 return 0;
2169 static void scan_refcount(BlockDriverState *bs, int64_t *high, int64_t *free)
2171 BDRVQcowState *s = bs->opaque;
2172 int64_t refcnt_index, cluster_index, cluster_end, h = 0, f = 0;
2173 int64_t tail = 0; /* do not count last consecutive free entries */
2175 for (refcnt_index=0; refcnt_index < s->refcount_table_size; refcnt_index++){
2176 if (s->refcount_table[refcnt_index] == 0) {
2177 f += 1 << (s->cluster_bits - REFCOUNT_SHIFT);
2178 tail += 1 << (s->cluster_bits - REFCOUNT_SHIFT);
2179 continue;
2181 cluster_index = refcnt_index << (s->cluster_bits - REFCOUNT_SHIFT);
2182 cluster_end = (refcnt_index + 1) << (s->cluster_bits - REFCOUNT_SHIFT);
2183 for ( ; cluster_index < cluster_end; cluster_index++) {
2184 if (get_refcount(bs, cluster_index) == 0) {
2185 f++;
2186 tail++;
2188 else {
2189 h = cluster_index;
2190 tail = 0;
2195 f -= tail;
2196 if (free)
2197 *free = f;
2198 if (high)
2199 *high = (h+1);
2202 static int get_refcount(BlockDriverState *bs, int64_t cluster_index)
2204 BDRVQcowState *s = bs->opaque;
2205 int refcount_table_index, block_index;
2206 int64_t refcount_block_offset;
2208 refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
2209 if (refcount_table_index >= s->refcount_table_size)
2210 return 0;
2211 refcount_block_offset = s->refcount_table[refcount_table_index];
2212 if (!refcount_block_offset)
2213 return 0;
2214 if (refcount_block_offset != s->refcount_block_cache_offset) {
2215 /* better than nothing: return allocated if read error */
2216 if (load_refcount_block(bs, refcount_block_offset) < 0)
2217 return 1;
2219 block_index = cluster_index &
2220 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
2221 return be16_to_cpu(s->refcount_block_cache[block_index]);
2224 /* return < 0 if error */
2225 static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size)
2227 BDRVQcowState *s = bs->opaque;
2228 int i, nb_clusters;
2230 nb_clusters = size_to_clusters(s, size);
2231 retry:
2232 for(i = 0; i < nb_clusters; i++) {
2233 int64_t i = s->free_cluster_index++;
2234 if (get_refcount(bs, i) != 0)
2235 goto retry;
2237 #ifdef DEBUG_ALLOC2
2238 printf("alloc_clusters: size=%lld -> %lld\n",
2239 size,
2240 (s->free_cluster_index - nb_clusters) << s->cluster_bits);
2241 #endif
2243 if (s->highest_alloc < s->free_cluster_index) {
2244 s->nc_free += (s->free_cluster_index - s->highest_alloc);
2245 s->highest_alloc = s->free_cluster_index;
2248 return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
2251 static int64_t alloc_clusters(BlockDriverState *bs, int64_t size)
2253 int64_t offset;
2255 offset = alloc_clusters_noref(bs, size);
2256 update_refcount(bs, offset, size, 1);
2257 return offset;
2260 /* only used to allocate compressed sectors. We try to allocate
2261 contiguous sectors. size must be <= cluster_size */
2262 static int64_t alloc_bytes(BlockDriverState *bs, int size)
2264 BDRVQcowState *s = bs->opaque;
2265 int64_t offset, cluster_offset;
2266 int free_in_cluster;
2268 assert(size > 0 && size <= s->cluster_size);
2269 if (s->free_byte_offset == 0) {
2270 s->free_byte_offset = alloc_clusters(bs, s->cluster_size);
2272 redo:
2273 free_in_cluster = s->cluster_size -
2274 (s->free_byte_offset & (s->cluster_size - 1));
2275 if (size <= free_in_cluster) {
2276 /* enough space in current cluster */
2277 offset = s->free_byte_offset;
2278 s->free_byte_offset += size;
2279 free_in_cluster -= size;
2280 if (free_in_cluster == 0)
2281 s->free_byte_offset = 0;
2282 if ((offset & (s->cluster_size - 1)) != 0)
2283 update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
2284 } else {
2285 offset = alloc_clusters(bs, s->cluster_size);
2286 cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1);
2287 if ((cluster_offset + s->cluster_size) == offset) {
2288 /* we are lucky: contiguous data */
2289 offset = s->free_byte_offset;
2290 update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
2291 s->free_byte_offset += size;
2292 } else {
2293 s->free_byte_offset = offset;
2294 goto redo;
2297 return offset;
2300 static void free_clusters(BlockDriverState *bs,
2301 int64_t offset, int64_t size)
2303 update_refcount(bs, offset, size, -1);
2306 static int grow_refcount_table(BlockDriverState *bs, int min_size)
2308 BDRVQcowState *s = bs->opaque;
2309 int new_table_size, new_table_size2, refcount_table_clusters, i, ret;
2310 uint64_t *new_table;
2311 int64_t table_offset;
2312 uint8_t data[12];
2313 int old_table_size;
2314 int64_t old_table_offset;
2316 if (min_size <= s->refcount_table_size)
2317 return 0;
2318 /* compute new table size */
2319 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
2320 for(;;) {
2321 if (refcount_table_clusters == 0) {
2322 refcount_table_clusters = 1;
2323 } else {
2324 refcount_table_clusters = (refcount_table_clusters * 3 + 1) / 2;
2326 new_table_size = refcount_table_clusters << (s->cluster_bits - 3);
2327 if (min_size <= new_table_size)
2328 break;
2330 #ifdef DEBUG_ALLOC2
2331 printf("grow_refcount_table from %d to %d\n",
2332 s->refcount_table_size,
2333 new_table_size);
2334 #endif
2335 new_table_size2 = new_table_size * sizeof(uint64_t);
2336 new_table = qemu_mallocz(new_table_size2);
2337 memcpy(new_table, s->refcount_table,
2338 s->refcount_table_size * sizeof(uint64_t));
2339 for(i = 0; i < s->refcount_table_size; i++)
2340 cpu_to_be64s(&new_table[i]);
2341 /* Note: we cannot update the refcount now to avoid recursion */
2342 table_offset = alloc_clusters_noref(bs, new_table_size2);
2343 ret = bdrv_pwrite(s->hd, table_offset, new_table, new_table_size2);
2344 if (ret != new_table_size2)
2345 goto fail;
2346 for(i = 0; i < s->refcount_table_size; i++)
2347 be64_to_cpus(&new_table[i]);
2349 cpu_to_be64w((uint64_t*)data, table_offset);
2350 cpu_to_be32w((uint32_t*)(data + 8), refcount_table_clusters);
2351 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, refcount_table_offset),
2352 data, sizeof(data)) != sizeof(data))
2353 goto fail;
2354 qemu_free(s->refcount_table);
2355 old_table_offset = s->refcount_table_offset;
2356 old_table_size = s->refcount_table_size;
2357 s->refcount_table = new_table;
2358 s->refcount_table_size = new_table_size;
2359 s->refcount_table_offset = table_offset;
2361 update_refcount(bs, table_offset, new_table_size2, 1);
2362 free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t));
2363 return 0;
2364 fail:
2365 free_clusters(bs, table_offset, new_table_size2);
2366 qemu_free(new_table);
2367 return -EIO;
2370 /* addend must be 1 or -1 */
2371 /* XXX: cache several refcount block clusters ? */
2372 static int update_cluster_refcount(BlockDriverState *bs,
2373 int64_t cluster_index,
2374 int addend)
2376 BDRVQcowState *s = bs->opaque;
2377 int64_t offset, refcount_block_offset;
2378 int ret, refcount_table_index, block_index, refcount;
2379 uint64_t data64;
2381 refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
2382 if (refcount_table_index >= s->refcount_table_size) {
2383 if (addend < 0)
2384 return -EINVAL;
2385 ret = grow_refcount_table(bs, refcount_table_index + 1);
2386 if (ret < 0)
2387 return ret;
2389 refcount_block_offset = s->refcount_table[refcount_table_index];
2390 if (!refcount_block_offset) {
2391 if (addend < 0)
2392 return -EINVAL;
2393 /* create a new refcount block */
2394 /* Note: we cannot update the refcount now to avoid recursion */
2395 offset = alloc_clusters_noref(bs, s->cluster_size);
2396 memset(s->refcount_block_cache, 0, s->cluster_size);
2397 ret = bdrv_pwrite(s->hd, offset, s->refcount_block_cache, s->cluster_size);
2398 if (ret != s->cluster_size)
2399 return -EINVAL;
2400 s->refcount_table[refcount_table_index] = offset;
2401 data64 = cpu_to_be64(offset);
2402 ret = bdrv_pwrite(s->hd, s->refcount_table_offset +
2403 refcount_table_index * sizeof(uint64_t),
2404 &data64, sizeof(data64));
2405 if (ret != sizeof(data64))
2406 return -EINVAL;
2408 refcount_block_offset = offset;
2409 s->refcount_block_cache_offset = offset;
2410 update_refcount(bs, offset, s->cluster_size, 1);
2411 } else {
2412 if (refcount_block_offset != s->refcount_block_cache_offset) {
2413 if (load_refcount_block(bs, refcount_block_offset) < 0)
2414 return -EIO;
2417 /* we can update the count and save it */
2418 block_index = cluster_index &
2419 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
2420 refcount = be16_to_cpu(s->refcount_block_cache[block_index]);
2422 if (refcount == 1 && addend == -1)
2423 s->nc_free += 1;
2424 else if (refcount == 0 && addend == 1)
2425 s->nc_free -= 1;
2427 refcount += addend;
2428 if (refcount < 0 || refcount > 0xffff)
2429 return -EINVAL;
2430 if (refcount == 0 && cluster_index < s->free_cluster_index) {
2431 s->free_cluster_index = cluster_index;
2433 s->refcount_block_cache[block_index] = cpu_to_be16(refcount);
2434 if (bdrv_pwrite(s->hd,
2435 refcount_block_offset + (block_index << REFCOUNT_SHIFT),
2436 &s->refcount_block_cache[block_index], 2) != 2)
2437 return -EIO;
2438 return refcount;
2441 static void update_refcount(BlockDriverState *bs,
2442 int64_t offset, int64_t length,
2443 int addend)
2445 BDRVQcowState *s = bs->opaque;
2446 int64_t start, last, cluster_offset;
2448 #ifdef DEBUG_ALLOC2
2449 printf("update_refcount: offset=%lld size=%lld addend=%d\n",
2450 offset, length, addend);
2451 #endif
2452 if (length <= 0)
2453 return;
2454 start = offset & ~(s->cluster_size - 1);
2455 last = (offset + length - 1) & ~(s->cluster_size - 1);
2456 for(cluster_offset = start; cluster_offset <= last;
2457 cluster_offset += s->cluster_size) {
2458 update_cluster_refcount(bs, cluster_offset >> s->cluster_bits, addend);
2462 #ifdef DEBUG_ALLOC
2463 static void inc_refcounts(BlockDriverState *bs,
2464 uint16_t *refcount_table,
2465 int refcount_table_size,
2466 int64_t offset, int64_t size)
2468 BDRVQcowState *s = bs->opaque;
2469 int64_t start, last, cluster_offset;
2470 int k;
2472 if (size <= 0)
2473 return;
2475 start = offset & ~(s->cluster_size - 1);
2476 last = (offset + size - 1) & ~(s->cluster_size - 1);
2477 for(cluster_offset = start; cluster_offset <= last;
2478 cluster_offset += s->cluster_size) {
2479 k = cluster_offset >> s->cluster_bits;
2480 if (k < 0 || k >= refcount_table_size) {
2481 printf("ERROR: invalid cluster offset=0x%llx\n", cluster_offset);
2482 } else {
2483 if (++refcount_table[k] == 0) {
2484 printf("ERROR: overflow cluster offset=0x%llx\n", cluster_offset);
2490 static int check_refcounts_l1(BlockDriverState *bs,
2491 uint16_t *refcount_table,
2492 int refcount_table_size,
2493 int64_t l1_table_offset, int l1_size,
2494 int check_copied)
2496 BDRVQcowState *s = bs->opaque;
2497 uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2;
2498 int l2_size, i, j, nb_csectors, refcount;
2500 l2_table = NULL;
2501 l1_size2 = l1_size * sizeof(uint64_t);
2503 inc_refcounts(bs, refcount_table, refcount_table_size,
2504 l1_table_offset, l1_size2);
2506 l1_table = qemu_malloc(l1_size2);
2507 if (bdrv_pread(s->hd, l1_table_offset,
2508 l1_table, l1_size2) != l1_size2)
2509 goto fail;
2510 for(i = 0;i < l1_size; i++)
2511 be64_to_cpus(&l1_table[i]);
2513 l2_size = s->l2_size * sizeof(uint64_t);
2514 l2_table = qemu_malloc(l2_size);
2515 for(i = 0; i < l1_size; i++) {
2516 l2_offset = l1_table[i];
2517 if (l2_offset) {
2518 if (check_copied) {
2519 refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED) >> s->cluster_bits);
2520 if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) {
2521 printf("ERROR OFLAG_COPIED: l2_offset=%llx refcount=%d\n",
2522 l2_offset, refcount);
2525 l2_offset &= ~QCOW_OFLAG_COPIED;
2526 if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
2527 goto fail;
2528 for(j = 0; j < s->l2_size; j++) {
2529 offset = be64_to_cpu(l2_table[j]);
2530 if (offset != 0) {
2531 if (offset & QCOW_OFLAG_COMPRESSED) {
2532 if (offset & QCOW_OFLAG_COPIED) {
2533 printf("ERROR: cluster %lld: copied flag must never be set for compressed clusters\n",
2534 offset >> s->cluster_bits);
2535 offset &= ~QCOW_OFLAG_COPIED;
2537 nb_csectors = ((offset >> s->csize_shift) &
2538 s->csize_mask) + 1;
2539 offset &= s->cluster_offset_mask;
2540 inc_refcounts(bs, refcount_table,
2541 refcount_table_size,
2542 offset & ~511, nb_csectors * 512);
2543 } else {
2544 if (check_copied) {
2545 refcount = get_refcount(bs, (offset & ~QCOW_OFLAG_COPIED) >> s->cluster_bits);
2546 if ((refcount == 1) != ((offset & QCOW_OFLAG_COPIED) != 0)) {
2547 printf("ERROR OFLAG_COPIED: offset=%llx refcount=%d\n",
2548 offset, refcount);
2551 offset &= ~QCOW_OFLAG_COPIED;
2552 inc_refcounts(bs, refcount_table,
2553 refcount_table_size,
2554 offset, s->cluster_size);
2558 inc_refcounts(bs, refcount_table,
2559 refcount_table_size,
2560 l2_offset,
2561 s->cluster_size);
2564 qemu_free(l1_table);
2565 qemu_free(l2_table);
2566 return 0;
2567 fail:
2568 printf("ERROR: I/O error in check_refcounts_l1\n");
2569 qemu_free(l1_table);
2570 qemu_free(l2_table);
2571 return -EIO;
2574 static void check_refcounts(BlockDriverState *bs)
2576 BDRVQcowState *s = bs->opaque;
2577 int64_t size;
2578 int nb_clusters, refcount1, refcount2, i;
2579 QCowSnapshot *sn;
2580 uint16_t *refcount_table;
2582 size = bdrv_getlength(s->hd);
2583 nb_clusters = size_to_clusters(s, size);
2584 refcount_table = qemu_mallocz(nb_clusters * sizeof(uint16_t));
2586 /* header */
2587 inc_refcounts(bs, refcount_table, nb_clusters,
2588 0, s->cluster_size);
2590 check_refcounts_l1(bs, refcount_table, nb_clusters,
2591 s->l1_table_offset, s->l1_size, 1);
2593 /* snapshots */
2594 for(i = 0; i < s->nb_snapshots; i++) {
2595 sn = s->snapshots + i;
2596 check_refcounts_l1(bs, refcount_table, nb_clusters,
2597 sn->l1_table_offset, sn->l1_size, 0);
2599 inc_refcounts(bs, refcount_table, nb_clusters,
2600 s->snapshots_offset, s->snapshots_size);
2602 /* refcount data */
2603 inc_refcounts(bs, refcount_table, nb_clusters,
2604 s->refcount_table_offset,
2605 s->refcount_table_size * sizeof(uint64_t));
2606 for(i = 0; i < s->refcount_table_size; i++) {
2607 int64_t offset;
2608 offset = s->refcount_table[i];
2609 if (offset != 0) {
2610 inc_refcounts(bs, refcount_table, nb_clusters,
2611 offset, s->cluster_size);
2615 /* compare ref counts */
2616 for(i = 0; i < nb_clusters; i++) {
2617 refcount1 = get_refcount(bs, i);
2618 refcount2 = refcount_table[i];
2619 if (refcount1 != refcount2)
2620 printf("ERROR cluster %d refcount=%d reference=%d\n",
2621 i, refcount1, refcount2);
2624 qemu_free(refcount_table);
2627 #if 0
2628 static void dump_refcounts(BlockDriverState *bs)
2630 BDRVQcowState *s = bs->opaque;
2631 int64_t nb_clusters, k, k1, size;
2632 int refcount;
2634 size = bdrv_getlength(s->hd);
2635 nb_clusters = size_to_clusters(s, size);
2636 for(k = 0; k < nb_clusters;) {
2637 k1 = k;
2638 refcount = get_refcount(bs, k);
2639 k++;
2640 while (k < nb_clusters && get_refcount(bs, k) == refcount)
2641 k++;
2642 printf("%lld: refcount=%d nb=%lld\n", k, refcount, k - k1);
2645 #endif
2646 #endif
2648 BlockDriver bdrv_qcow2 = {
2649 .format_name = "qcow2",
2650 .instance_size = sizeof(BDRVQcowState),
2651 .bdrv_probe = qcow_probe,
2652 .bdrv_open = qcow_open,
2653 .bdrv_close = qcow_close,
2654 .bdrv_create = qcow_create,
2655 .bdrv_flush = qcow_flush,
2656 .bdrv_is_allocated = qcow_is_allocated,
2657 .bdrv_set_key = qcow_set_key,
2658 .bdrv_make_empty = qcow_make_empty,
2660 .bdrv_aio_read = qcow_aio_read,
2661 .bdrv_aio_write = qcow_aio_write,
2662 .bdrv_aio_cancel = qcow_aio_cancel,
2663 .aiocb_size = sizeof(QCowAIOCB),
2664 .bdrv_write_compressed = qcow_write_compressed,
2666 .bdrv_snapshot_create = qcow_snapshot_create,
2667 .bdrv_snapshot_goto = qcow_snapshot_goto,
2668 .bdrv_snapshot_delete = qcow_snapshot_delete,
2669 .bdrv_snapshot_list = qcow_snapshot_list,
2670 .bdrv_get_info = qcow_get_info,