add a simple fork test
[qemu/linux-user.git] / block-qcow2.c
blob9aa7261e3fb908d4bddfdbe3a002e6c3cff55b40
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 (!s->l1_table)
257 goto fail;
258 if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, s->l1_size * sizeof(uint64_t)) !=
259 s->l1_size * sizeof(uint64_t))
260 goto fail;
261 for(i = 0;i < s->l1_size; i++) {
262 be64_to_cpus(&s->l1_table[i]);
264 /* alloc L2 cache */
265 s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
266 if (!s->l2_cache)
267 goto fail;
268 s->cluster_cache = qemu_malloc(s->cluster_size);
269 if (!s->cluster_cache)
270 goto fail;
271 /* one more sector for decompressed data alignment */
272 s->cluster_data = qemu_malloc(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
273 + 512);
274 if (!s->cluster_data)
275 goto fail;
276 s->cluster_cache_offset = -1;
278 if (refcount_init(bs) < 0)
279 goto fail;
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 if (!new_l1_table)
455 return -ENOMEM;
456 memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t));
458 /* write new table (align to cluster) */
459 new_l1_table_offset = alloc_clusters(bs, new_l1_size2);
461 for(i = 0; i < s->l1_size; i++)
462 new_l1_table[i] = cpu_to_be64(new_l1_table[i]);
463 ret = bdrv_pwrite(s->hd, new_l1_table_offset, new_l1_table, new_l1_size2);
464 if (ret != new_l1_size2)
465 goto fail;
466 for(i = 0; i < s->l1_size; i++)
467 new_l1_table[i] = be64_to_cpu(new_l1_table[i]);
469 /* set new table */
470 cpu_to_be32w((uint32_t*)data, new_l1_size);
471 cpu_to_be64w((uint64_t*)(data + 4), new_l1_table_offset);
472 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, l1_size), data,
473 sizeof(data)) != sizeof(data))
474 goto fail;
475 qemu_free(s->l1_table);
476 free_clusters(bs, s->l1_table_offset, s->l1_size * sizeof(uint64_t));
477 s->l1_table_offset = new_l1_table_offset;
478 s->l1_table = new_l1_table;
479 s->l1_size = new_l1_size;
480 return 0;
481 fail:
482 qemu_free(s->l1_table);
483 return -EIO;
487 * seek_l2_table
489 * seek l2_offset in the l2_cache table
490 * if not found, return NULL,
491 * if found,
492 * increments the l2 cache hit count of the entry,
493 * if counter overflow, divide by two all counters
494 * return the pointer to the l2 cache entry
498 static uint64_t *seek_l2_table(BDRVQcowState *s, uint64_t l2_offset)
500 int i, j;
502 for(i = 0; i < L2_CACHE_SIZE; i++) {
503 if (l2_offset == s->l2_cache_offsets[i]) {
504 /* increment the hit count */
505 if (++s->l2_cache_counts[i] == 0xffffffff) {
506 for(j = 0; j < L2_CACHE_SIZE; j++) {
507 s->l2_cache_counts[j] >>= 1;
510 return s->l2_cache + (i << s->l2_bits);
513 return NULL;
517 * l2_load
519 * Loads a L2 table into memory. If the table is in the cache, the cache
520 * is used; otherwise the L2 table is loaded from the image file.
522 * Returns a pointer to the L2 table on success, or NULL if the read from
523 * the image file failed.
526 static uint64_t *l2_load(BlockDriverState *bs, uint64_t l2_offset)
528 BDRVQcowState *s = bs->opaque;
529 int min_index;
530 uint64_t *l2_table;
532 /* seek if the table for the given offset is in the cache */
534 l2_table = seek_l2_table(s, l2_offset);
535 if (l2_table != NULL)
536 return l2_table;
538 /* not found: load a new entry in the least used one */
540 min_index = l2_cache_new_entry(bs);
541 l2_table = s->l2_cache + (min_index << s->l2_bits);
542 if (bdrv_pread(s->hd, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
543 s->l2_size * sizeof(uint64_t))
544 return NULL;
545 s->l2_cache_offsets[min_index] = l2_offset;
546 s->l2_cache_counts[min_index] = 1;
548 return l2_table;
552 * l2_allocate
554 * Allocate a new l2 entry in the file. If l1_index points to an already
555 * used entry in the L2 table (i.e. we are doing a copy on write for the L2
556 * table) copy the contents of the old L2 table into the newly allocated one.
557 * Otherwise the new table is initialized with zeros.
561 static uint64_t *l2_allocate(BlockDriverState *bs, int l1_index)
563 BDRVQcowState *s = bs->opaque;
564 int min_index;
565 uint64_t old_l2_offset, tmp;
566 uint64_t *l2_table, l2_offset;
568 old_l2_offset = s->l1_table[l1_index];
570 /* allocate a new l2 entry */
572 l2_offset = alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
574 /* update the L1 entry */
576 s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
578 tmp = cpu_to_be64(l2_offset | QCOW_OFLAG_COPIED);
579 if (bdrv_pwrite(s->hd, s->l1_table_offset + l1_index * sizeof(tmp),
580 &tmp, sizeof(tmp)) != sizeof(tmp))
581 return NULL;
583 /* allocate a new entry in the l2 cache */
585 min_index = l2_cache_new_entry(bs);
586 l2_table = s->l2_cache + (min_index << s->l2_bits);
588 if (old_l2_offset == 0) {
589 /* if there was no old l2 table, clear the new table */
590 memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
591 } else {
592 /* if there was an old l2 table, read it from the disk */
593 if (bdrv_pread(s->hd, old_l2_offset,
594 l2_table, s->l2_size * sizeof(uint64_t)) !=
595 s->l2_size * sizeof(uint64_t))
596 return NULL;
598 /* write the l2 table to the file */
599 if (bdrv_pwrite(s->hd, l2_offset,
600 l2_table, s->l2_size * sizeof(uint64_t)) !=
601 s->l2_size * sizeof(uint64_t))
602 return NULL;
604 /* update the l2 cache entry */
606 s->l2_cache_offsets[min_index] = l2_offset;
607 s->l2_cache_counts[min_index] = 1;
609 return l2_table;
612 static int size_to_clusters(BDRVQcowState *s, int64_t size)
614 return (size + (s->cluster_size - 1)) >> s->cluster_bits;
617 static int count_contiguous_clusters(uint64_t nb_clusters, int cluster_size,
618 uint64_t *l2_table, uint64_t start, uint64_t mask)
620 int i;
621 uint64_t offset = be64_to_cpu(l2_table[0]) & ~mask;
623 if (!offset)
624 return 0;
626 for (i = start; i < start + nb_clusters; i++)
627 if (offset + i * cluster_size != (be64_to_cpu(l2_table[i]) & ~mask))
628 break;
630 return (i - start);
633 static int count_contiguous_free_clusters(uint64_t nb_clusters, uint64_t *l2_table)
635 int i = 0;
637 while(nb_clusters-- && l2_table[i] == 0)
638 i++;
640 return i;
644 * get_cluster_offset
646 * For a given offset of the disk image, return cluster offset in
647 * qcow2 file.
649 * on entry, *num is the number of contiguous clusters we'd like to
650 * access following offset.
652 * on exit, *num is the number of contiguous clusters we can read.
654 * Return 1, if the offset is found
655 * Return 0, otherwise.
659 static uint64_t get_cluster_offset(BlockDriverState *bs,
660 uint64_t offset, int *num)
662 BDRVQcowState *s = bs->opaque;
663 int l1_index, l2_index;
664 uint64_t l2_offset, *l2_table, cluster_offset;
665 int l1_bits, c;
666 int index_in_cluster, nb_available, nb_needed, nb_clusters;
668 index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1);
669 nb_needed = *num + index_in_cluster;
671 l1_bits = s->l2_bits + s->cluster_bits;
673 /* compute how many bytes there are between the offset and
674 * the end of the l1 entry
677 nb_available = (1 << l1_bits) - (offset & ((1 << l1_bits) - 1));
679 /* compute the number of available sectors */
681 nb_available = (nb_available >> 9) + index_in_cluster;
683 cluster_offset = 0;
685 /* seek the the l2 offset in the l1 table */
687 l1_index = offset >> l1_bits;
688 if (l1_index >= s->l1_size)
689 goto out;
691 l2_offset = s->l1_table[l1_index];
693 /* seek the l2 table of the given l2 offset */
695 if (!l2_offset)
696 goto out;
698 /* load the l2 table in memory */
700 l2_offset &= ~QCOW_OFLAG_COPIED;
701 l2_table = l2_load(bs, l2_offset);
702 if (l2_table == NULL)
703 return 0;
705 /* find the cluster offset for the given disk offset */
707 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
708 cluster_offset = be64_to_cpu(l2_table[l2_index]);
709 nb_clusters = size_to_clusters(s, nb_needed << 9);
711 if (!cluster_offset) {
712 /* how many empty clusters ? */
713 c = count_contiguous_free_clusters(nb_clusters, &l2_table[l2_index]);
714 } else {
715 /* how many allocated clusters ? */
716 c = count_contiguous_clusters(nb_clusters, s->cluster_size,
717 &l2_table[l2_index], 0, QCOW_OFLAG_COPIED);
720 nb_available = (c * s->cluster_sectors);
721 out:
722 if (nb_available > nb_needed)
723 nb_available = nb_needed;
725 *num = nb_available - index_in_cluster;
727 return cluster_offset & ~QCOW_OFLAG_COPIED;
731 * free_any_clusters
733 * free clusters according to its type: compressed or not
737 static void free_any_clusters(BlockDriverState *bs,
738 uint64_t cluster_offset, int nb_clusters)
740 BDRVQcowState *s = bs->opaque;
742 /* free the cluster */
744 if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
745 int nb_csectors;
746 nb_csectors = ((cluster_offset >> s->csize_shift) &
747 s->csize_mask) + 1;
748 free_clusters(bs, (cluster_offset & s->cluster_offset_mask) & ~511,
749 nb_csectors * 512);
750 return;
753 free_clusters(bs, cluster_offset, nb_clusters << s->cluster_bits);
755 return;
759 * get_cluster_table
761 * for a given disk offset, load (and allocate if needed)
762 * the l2 table.
764 * the l2 table offset in the qcow2 file and the cluster index
765 * in the l2 table are given to the caller.
769 static int get_cluster_table(BlockDriverState *bs, uint64_t offset,
770 uint64_t **new_l2_table,
771 uint64_t *new_l2_offset,
772 int *new_l2_index)
774 BDRVQcowState *s = bs->opaque;
775 int l1_index, l2_index, ret;
776 uint64_t l2_offset, *l2_table;
778 /* seek the the l2 offset in the l1 table */
780 l1_index = offset >> (s->l2_bits + s->cluster_bits);
781 if (l1_index >= s->l1_size) {
782 ret = grow_l1_table(bs, l1_index + 1);
783 if (ret < 0)
784 return 0;
786 l2_offset = s->l1_table[l1_index];
788 /* seek the l2 table of the given l2 offset */
790 if (l2_offset & QCOW_OFLAG_COPIED) {
791 /* load the l2 table in memory */
792 l2_offset &= ~QCOW_OFLAG_COPIED;
793 l2_table = l2_load(bs, l2_offset);
794 if (l2_table == NULL)
795 return 0;
796 } else {
797 if (l2_offset)
798 free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t));
799 l2_table = l2_allocate(bs, l1_index);
800 if (l2_table == NULL)
801 return 0;
802 l2_offset = s->l1_table[l1_index] & ~QCOW_OFLAG_COPIED;
805 /* find the cluster offset for the given disk offset */
807 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
809 *new_l2_table = l2_table;
810 *new_l2_offset = l2_offset;
811 *new_l2_index = l2_index;
813 return 1;
817 * alloc_compressed_cluster_offset
819 * For a given offset of the disk image, return cluster offset in
820 * qcow2 file.
822 * If the offset is not found, allocate a new compressed cluster.
824 * Return the cluster offset if successful,
825 * Return 0, otherwise.
829 static uint64_t alloc_compressed_cluster_offset(BlockDriverState *bs,
830 uint64_t offset,
831 int compressed_size)
833 BDRVQcowState *s = bs->opaque;
834 int l2_index, ret;
835 uint64_t l2_offset, *l2_table, cluster_offset;
836 int nb_csectors;
838 ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
839 if (ret == 0)
840 return 0;
842 cluster_offset = be64_to_cpu(l2_table[l2_index]);
843 if (cluster_offset & QCOW_OFLAG_COPIED)
844 return cluster_offset & ~QCOW_OFLAG_COPIED;
846 if (cluster_offset)
847 free_any_clusters(bs, cluster_offset, 1);
849 cluster_offset = alloc_bytes(bs, compressed_size);
850 nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) -
851 (cluster_offset >> 9);
853 cluster_offset |= QCOW_OFLAG_COMPRESSED |
854 ((uint64_t)nb_csectors << s->csize_shift);
856 /* update L2 table */
858 /* compressed clusters never have the copied flag */
860 l2_table[l2_index] = cpu_to_be64(cluster_offset);
861 if (bdrv_pwrite(s->hd,
862 l2_offset + l2_index * sizeof(uint64_t),
863 l2_table + l2_index,
864 sizeof(uint64_t)) != sizeof(uint64_t))
865 return 0;
867 return cluster_offset;
870 typedef struct QCowL2Meta
872 uint64_t offset;
873 int n_start;
874 int nb_available;
875 int nb_clusters;
876 } QCowL2Meta;
878 static int alloc_cluster_link_l2(BlockDriverState *bs, uint64_t cluster_offset,
879 QCowL2Meta *m)
881 BDRVQcowState *s = bs->opaque;
882 int i, j = 0, l2_index, ret;
883 uint64_t *old_cluster, start_sect, l2_offset, *l2_table;
885 if (m->nb_clusters == 0)
886 return 0;
888 if (!(old_cluster = qemu_malloc(m->nb_clusters * sizeof(uint64_t))))
889 return -ENOMEM;
891 /* copy content of unmodified sectors */
892 start_sect = (m->offset & ~(s->cluster_size - 1)) >> 9;
893 if (m->n_start) {
894 ret = copy_sectors(bs, start_sect, cluster_offset, 0, m->n_start);
895 if (ret < 0)
896 goto err;
899 if (m->nb_available & (s->cluster_sectors - 1)) {
900 uint64_t end = m->nb_available & ~(uint64_t)(s->cluster_sectors - 1);
901 ret = copy_sectors(bs, start_sect + end, cluster_offset + (end << 9),
902 m->nb_available - end, s->cluster_sectors);
903 if (ret < 0)
904 goto err;
907 ret = -EIO;
908 /* update L2 table */
909 if (!get_cluster_table(bs, m->offset, &l2_table, &l2_offset, &l2_index))
910 goto err;
912 for (i = 0; i < m->nb_clusters; i++) {
913 if(l2_table[l2_index + i] != 0)
914 old_cluster[j++] = l2_table[l2_index + i];
916 l2_table[l2_index + i] = cpu_to_be64((cluster_offset +
917 (i << s->cluster_bits)) | QCOW_OFLAG_COPIED);
920 if (bdrv_pwrite(s->hd, l2_offset + l2_index * sizeof(uint64_t),
921 l2_table + l2_index, m->nb_clusters * sizeof(uint64_t)) !=
922 m->nb_clusters * sizeof(uint64_t))
923 goto err;
925 for (i = 0; i < j; i++)
926 free_any_clusters(bs, old_cluster[i], 1);
928 ret = 0;
929 err:
930 qemu_free(old_cluster);
931 return ret;
935 * alloc_cluster_offset
937 * For a given offset of the disk image, return cluster offset in
938 * qcow2 file.
940 * If the offset is not found, allocate a new cluster.
942 * Return the cluster offset if successful,
943 * Return 0, otherwise.
947 static uint64_t alloc_cluster_offset(BlockDriverState *bs,
948 uint64_t offset,
949 int n_start, int n_end,
950 int *num, QCowL2Meta *m)
952 BDRVQcowState *s = bs->opaque;
953 int l2_index, ret;
954 uint64_t l2_offset, *l2_table, cluster_offset;
955 int nb_clusters, i = 0;
957 ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
958 if (ret == 0)
959 return 0;
961 nb_clusters = size_to_clusters(s, n_end << 9);
963 nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
965 cluster_offset = be64_to_cpu(l2_table[l2_index]);
967 /* We keep all QCOW_OFLAG_COPIED clusters */
969 if (cluster_offset & QCOW_OFLAG_COPIED) {
970 nb_clusters = count_contiguous_clusters(nb_clusters, s->cluster_size,
971 &l2_table[l2_index], 0, 0);
973 cluster_offset &= ~QCOW_OFLAG_COPIED;
974 m->nb_clusters = 0;
976 goto out;
979 /* for the moment, multiple compressed clusters are not managed */
981 if (cluster_offset & QCOW_OFLAG_COMPRESSED)
982 nb_clusters = 1;
984 /* how many available clusters ? */
986 while (i < nb_clusters) {
987 i += count_contiguous_clusters(nb_clusters - i, s->cluster_size,
988 &l2_table[l2_index], i, 0);
990 if(be64_to_cpu(l2_table[l2_index + i]))
991 break;
993 i += count_contiguous_free_clusters(nb_clusters - i,
994 &l2_table[l2_index + i]);
996 cluster_offset = be64_to_cpu(l2_table[l2_index + i]);
998 if ((cluster_offset & QCOW_OFLAG_COPIED) ||
999 (cluster_offset & QCOW_OFLAG_COMPRESSED))
1000 break;
1002 nb_clusters = i;
1004 /* allocate a new cluster */
1006 cluster_offset = alloc_clusters(bs, nb_clusters * s->cluster_size);
1008 /* save info needed for meta data update */
1009 m->offset = offset;
1010 m->n_start = n_start;
1011 m->nb_clusters = nb_clusters;
1013 out:
1014 m->nb_available = MIN(nb_clusters << (s->cluster_bits - 9), n_end);
1016 *num = m->nb_available - n_start;
1018 return cluster_offset;
1021 static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
1022 int nb_sectors, int *pnum)
1024 uint64_t cluster_offset;
1026 *pnum = nb_sectors;
1027 cluster_offset = get_cluster_offset(bs, sector_num << 9, pnum);
1029 return (cluster_offset != 0);
1032 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
1033 const uint8_t *buf, int buf_size)
1035 z_stream strm1, *strm = &strm1;
1036 int ret, out_len;
1038 memset(strm, 0, sizeof(*strm));
1040 strm->next_in = (uint8_t *)buf;
1041 strm->avail_in = buf_size;
1042 strm->next_out = out_buf;
1043 strm->avail_out = out_buf_size;
1045 ret = inflateInit2(strm, -12);
1046 if (ret != Z_OK)
1047 return -1;
1048 ret = inflate(strm, Z_FINISH);
1049 out_len = strm->next_out - out_buf;
1050 if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
1051 out_len != out_buf_size) {
1052 inflateEnd(strm);
1053 return -1;
1055 inflateEnd(strm);
1056 return 0;
1059 static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset)
1061 int ret, csize, nb_csectors, sector_offset;
1062 uint64_t coffset;
1064 coffset = cluster_offset & s->cluster_offset_mask;
1065 if (s->cluster_cache_offset != coffset) {
1066 nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
1067 sector_offset = coffset & 511;
1068 csize = nb_csectors * 512 - sector_offset;
1069 ret = bdrv_read(s->hd, coffset >> 9, s->cluster_data, nb_csectors);
1070 if (ret < 0) {
1071 return -1;
1073 if (decompress_buffer(s->cluster_cache, s->cluster_size,
1074 s->cluster_data + sector_offset, csize) < 0) {
1075 return -1;
1077 s->cluster_cache_offset = coffset;
1079 return 0;
1082 /* handle reading after the end of the backing file */
1083 static int backing_read1(BlockDriverState *bs,
1084 int64_t sector_num, uint8_t *buf, int nb_sectors)
1086 int n1;
1087 if ((sector_num + nb_sectors) <= bs->total_sectors)
1088 return nb_sectors;
1089 if (sector_num >= bs->total_sectors)
1090 n1 = 0;
1091 else
1092 n1 = bs->total_sectors - sector_num;
1093 memset(buf + n1 * 512, 0, 512 * (nb_sectors - n1));
1094 return n1;
1097 static int qcow_read(BlockDriverState *bs, int64_t sector_num,
1098 uint8_t *buf, int nb_sectors)
1100 BDRVQcowState *s = bs->opaque;
1101 int ret, index_in_cluster, n, n1;
1102 uint64_t cluster_offset;
1104 while (nb_sectors > 0) {
1105 n = nb_sectors;
1106 cluster_offset = get_cluster_offset(bs, sector_num << 9, &n);
1107 index_in_cluster = sector_num & (s->cluster_sectors - 1);
1108 if (!cluster_offset) {
1109 if (bs->backing_hd) {
1110 /* read from the base image */
1111 n1 = backing_read1(bs->backing_hd, sector_num, buf, n);
1112 if (n1 > 0) {
1113 ret = bdrv_read(bs->backing_hd, sector_num, buf, n1);
1114 if (ret < 0)
1115 return -1;
1117 } else {
1118 memset(buf, 0, 512 * n);
1120 } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
1121 if (decompress_cluster(s, cluster_offset) < 0)
1122 return -1;
1123 memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n);
1124 } else {
1125 ret = bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
1126 if (ret != n * 512)
1127 return -1;
1128 if (s->crypt_method) {
1129 encrypt_sectors(s, sector_num, buf, buf, n, 0,
1130 &s->aes_decrypt_key);
1133 nb_sectors -= n;
1134 sector_num += n;
1135 buf += n * 512;
1137 return 0;
1140 static int qcow_write(BlockDriverState *bs, int64_t sector_num,
1141 const uint8_t *buf, int nb_sectors)
1143 BDRVQcowState *s = bs->opaque;
1144 int ret, index_in_cluster, n;
1145 uint64_t cluster_offset;
1146 int n_end;
1147 QCowL2Meta l2meta;
1149 while (nb_sectors > 0) {
1150 index_in_cluster = sector_num & (s->cluster_sectors - 1);
1151 n_end = index_in_cluster + nb_sectors;
1152 if (s->crypt_method &&
1153 n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
1154 n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
1155 cluster_offset = alloc_cluster_offset(bs, sector_num << 9,
1156 index_in_cluster,
1157 n_end, &n, &l2meta);
1158 if (!cluster_offset)
1159 return -1;
1160 if (s->crypt_method) {
1161 encrypt_sectors(s, sector_num, s->cluster_data, buf, n, 1,
1162 &s->aes_encrypt_key);
1163 ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512,
1164 s->cluster_data, n * 512);
1165 } else {
1166 ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
1168 if (ret != n * 512 || alloc_cluster_link_l2(bs, cluster_offset, &l2meta) < 0) {
1169 free_any_clusters(bs, cluster_offset, l2meta.nb_clusters);
1170 return -1;
1172 nb_sectors -= n;
1173 sector_num += n;
1174 buf += n * 512;
1176 s->cluster_cache_offset = -1; /* disable compressed cache */
1177 return 0;
1180 typedef struct QCowAIOCB {
1181 BlockDriverAIOCB common;
1182 int64_t sector_num;
1183 uint8_t *buf;
1184 int nb_sectors;
1185 int n;
1186 uint64_t cluster_offset;
1187 uint8_t *cluster_data;
1188 BlockDriverAIOCB *hd_aiocb;
1189 QEMUBH *bh;
1190 QCowL2Meta l2meta;
1191 } QCowAIOCB;
1193 static void qcow_aio_read_cb(void *opaque, int ret);
1194 static void qcow_aio_read_bh(void *opaque)
1196 QCowAIOCB *acb = opaque;
1197 qemu_bh_delete(acb->bh);
1198 acb->bh = NULL;
1199 qcow_aio_read_cb(opaque, 0);
1202 static int qcow_schedule_bh(QEMUBHFunc *cb, QCowAIOCB *acb)
1204 if (acb->bh)
1205 return -EIO;
1207 acb->bh = qemu_bh_new(cb, acb);
1208 if (!acb->bh)
1209 return -EIO;
1211 qemu_bh_schedule(acb->bh);
1213 return 0;
1216 static void qcow_aio_read_cb(void *opaque, int ret)
1218 QCowAIOCB *acb = opaque;
1219 BlockDriverState *bs = acb->common.bs;
1220 BDRVQcowState *s = bs->opaque;
1221 int index_in_cluster, n1;
1223 acb->hd_aiocb = NULL;
1224 if (ret < 0) {
1225 fail:
1226 acb->common.cb(acb->common.opaque, ret);
1227 qemu_aio_release(acb);
1228 return;
1231 /* post process the read buffer */
1232 if (!acb->cluster_offset) {
1233 /* nothing to do */
1234 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
1235 /* nothing to do */
1236 } else {
1237 if (s->crypt_method) {
1238 encrypt_sectors(s, acb->sector_num, acb->buf, acb->buf,
1239 acb->n, 0,
1240 &s->aes_decrypt_key);
1244 acb->nb_sectors -= acb->n;
1245 acb->sector_num += acb->n;
1246 acb->buf += acb->n * 512;
1248 if (acb->nb_sectors == 0) {
1249 /* request completed */
1250 acb->common.cb(acb->common.opaque, 0);
1251 qemu_aio_release(acb);
1252 return;
1255 /* prepare next AIO request */
1256 acb->n = acb->nb_sectors;
1257 acb->cluster_offset = get_cluster_offset(bs, acb->sector_num << 9, &acb->n);
1258 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
1260 if (!acb->cluster_offset) {
1261 if (bs->backing_hd) {
1262 /* read from the base image */
1263 n1 = backing_read1(bs->backing_hd, acb->sector_num,
1264 acb->buf, acb->n);
1265 if (n1 > 0) {
1266 acb->hd_aiocb = bdrv_aio_read(bs->backing_hd, acb->sector_num,
1267 acb->buf, acb->n, qcow_aio_read_cb, acb);
1268 if (acb->hd_aiocb == NULL)
1269 goto fail;
1270 } else {
1271 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
1272 if (ret < 0)
1273 goto fail;
1275 } else {
1276 /* Note: in this case, no need to wait */
1277 memset(acb->buf, 0, 512 * acb->n);
1278 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
1279 if (ret < 0)
1280 goto fail;
1282 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
1283 /* add AIO support for compressed blocks ? */
1284 if (decompress_cluster(s, acb->cluster_offset) < 0)
1285 goto fail;
1286 memcpy(acb->buf,
1287 s->cluster_cache + index_in_cluster * 512, 512 * acb->n);
1288 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
1289 if (ret < 0)
1290 goto fail;
1291 } else {
1292 if ((acb->cluster_offset & 511) != 0) {
1293 ret = -EIO;
1294 goto fail;
1296 acb->hd_aiocb = bdrv_aio_read(s->hd,
1297 (acb->cluster_offset >> 9) + index_in_cluster,
1298 acb->buf, acb->n, qcow_aio_read_cb, acb);
1299 if (acb->hd_aiocb == NULL)
1300 goto fail;
1304 static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs,
1305 int64_t sector_num, uint8_t *buf, int nb_sectors,
1306 BlockDriverCompletionFunc *cb, void *opaque)
1308 QCowAIOCB *acb;
1310 acb = qemu_aio_get(bs, cb, opaque);
1311 if (!acb)
1312 return NULL;
1313 acb->hd_aiocb = NULL;
1314 acb->sector_num = sector_num;
1315 acb->buf = buf;
1316 acb->nb_sectors = nb_sectors;
1317 acb->n = 0;
1318 acb->cluster_offset = 0;
1319 acb->l2meta.nb_clusters = 0;
1320 return acb;
1323 static BlockDriverAIOCB *qcow_aio_read(BlockDriverState *bs,
1324 int64_t sector_num, uint8_t *buf, int nb_sectors,
1325 BlockDriverCompletionFunc *cb, void *opaque)
1327 QCowAIOCB *acb;
1329 acb = qcow_aio_setup(bs, sector_num, buf, nb_sectors, cb, opaque);
1330 if (!acb)
1331 return NULL;
1333 qcow_aio_read_cb(acb, 0);
1334 return &acb->common;
1337 static void qcow_aio_write_cb(void *opaque, int ret)
1339 QCowAIOCB *acb = opaque;
1340 BlockDriverState *bs = acb->common.bs;
1341 BDRVQcowState *s = bs->opaque;
1342 int index_in_cluster;
1343 const uint8_t *src_buf;
1344 int n_end;
1346 acb->hd_aiocb = NULL;
1348 if (ret < 0) {
1349 fail:
1350 acb->common.cb(acb->common.opaque, ret);
1351 qemu_aio_release(acb);
1352 return;
1355 if (alloc_cluster_link_l2(bs, acb->cluster_offset, &acb->l2meta) < 0) {
1356 free_any_clusters(bs, acb->cluster_offset, acb->l2meta.nb_clusters);
1357 goto fail;
1360 acb->nb_sectors -= acb->n;
1361 acb->sector_num += acb->n;
1362 acb->buf += acb->n * 512;
1364 if (acb->nb_sectors == 0) {
1365 /* request completed */
1366 acb->common.cb(acb->common.opaque, 0);
1367 qemu_aio_release(acb);
1368 return;
1371 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
1372 n_end = index_in_cluster + acb->nb_sectors;
1373 if (s->crypt_method &&
1374 n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
1375 n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
1377 acb->cluster_offset = alloc_cluster_offset(bs, acb->sector_num << 9,
1378 index_in_cluster,
1379 n_end, &acb->n, &acb->l2meta);
1380 if (!acb->cluster_offset || (acb->cluster_offset & 511) != 0) {
1381 ret = -EIO;
1382 goto fail;
1384 if (s->crypt_method) {
1385 if (!acb->cluster_data) {
1386 acb->cluster_data = qemu_mallocz(QCOW_MAX_CRYPT_CLUSTERS *
1387 s->cluster_size);
1388 if (!acb->cluster_data) {
1389 ret = -ENOMEM;
1390 goto fail;
1393 encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf,
1394 acb->n, 1, &s->aes_encrypt_key);
1395 src_buf = acb->cluster_data;
1396 } else {
1397 src_buf = acb->buf;
1399 acb->hd_aiocb = bdrv_aio_write(s->hd,
1400 (acb->cluster_offset >> 9) + index_in_cluster,
1401 src_buf, acb->n,
1402 qcow_aio_write_cb, acb);
1403 if (acb->hd_aiocb == NULL)
1404 goto fail;
1407 static BlockDriverAIOCB *qcow_aio_write(BlockDriverState *bs,
1408 int64_t sector_num, const uint8_t *buf, int nb_sectors,
1409 BlockDriverCompletionFunc *cb, void *opaque)
1411 BDRVQcowState *s = bs->opaque;
1412 QCowAIOCB *acb;
1414 s->cluster_cache_offset = -1; /* disable compressed cache */
1416 acb = qcow_aio_setup(bs, sector_num, (uint8_t*)buf, nb_sectors, cb, opaque);
1417 if (!acb)
1418 return NULL;
1420 qcow_aio_write_cb(acb, 0);
1421 return &acb->common;
1424 static void qcow_aio_cancel(BlockDriverAIOCB *blockacb)
1426 QCowAIOCB *acb = (QCowAIOCB *)blockacb;
1427 if (acb->hd_aiocb)
1428 bdrv_aio_cancel(acb->hd_aiocb);
1429 qemu_aio_release(acb);
1432 static void qcow_close(BlockDriverState *bs)
1434 BDRVQcowState *s = bs->opaque;
1435 qemu_free(s->l1_table);
1436 qemu_free(s->l2_cache);
1437 qemu_free(s->cluster_cache);
1438 qemu_free(s->cluster_data);
1439 refcount_close(bs);
1440 bdrv_delete(s->hd);
1443 /* XXX: use std qcow open function ? */
1444 typedef struct QCowCreateState {
1445 int cluster_size;
1446 int cluster_bits;
1447 uint16_t *refcount_block;
1448 uint64_t *refcount_table;
1449 int64_t l1_table_offset;
1450 int64_t refcount_table_offset;
1451 int64_t refcount_block_offset;
1452 } QCowCreateState;
1454 static void create_refcount_update(QCowCreateState *s,
1455 int64_t offset, int64_t size)
1457 int refcount;
1458 int64_t start, last, cluster_offset;
1459 uint16_t *p;
1461 start = offset & ~(s->cluster_size - 1);
1462 last = (offset + size - 1) & ~(s->cluster_size - 1);
1463 for(cluster_offset = start; cluster_offset <= last;
1464 cluster_offset += s->cluster_size) {
1465 p = &s->refcount_block[cluster_offset >> s->cluster_bits];
1466 refcount = be16_to_cpu(*p);
1467 refcount++;
1468 *p = cpu_to_be16(refcount);
1472 static int qcow_create(const char *filename, int64_t total_size,
1473 const char *backing_file, int flags)
1475 int fd, header_size, backing_filename_len, l1_size, i, shift, l2_bits;
1476 QCowHeader header;
1477 uint64_t tmp, offset;
1478 QCowCreateState s1, *s = &s1;
1480 memset(s, 0, sizeof(*s));
1482 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
1483 if (fd < 0)
1484 return -1;
1485 memset(&header, 0, sizeof(header));
1486 header.magic = cpu_to_be32(QCOW_MAGIC);
1487 header.version = cpu_to_be32(QCOW_VERSION);
1488 header.size = cpu_to_be64(total_size * 512);
1489 header_size = sizeof(header);
1490 backing_filename_len = 0;
1491 if (backing_file) {
1492 header.backing_file_offset = cpu_to_be64(header_size);
1493 backing_filename_len = strlen(backing_file);
1494 header.backing_file_size = cpu_to_be32(backing_filename_len);
1495 header_size += backing_filename_len;
1497 s->cluster_bits = 12; /* 4 KB clusters */
1498 s->cluster_size = 1 << s->cluster_bits;
1499 header.cluster_bits = cpu_to_be32(s->cluster_bits);
1500 header_size = (header_size + 7) & ~7;
1501 if (flags & BLOCK_FLAG_ENCRYPT) {
1502 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
1503 } else {
1504 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
1506 l2_bits = s->cluster_bits - 3;
1507 shift = s->cluster_bits + l2_bits;
1508 l1_size = (((total_size * 512) + (1LL << shift) - 1) >> shift);
1509 offset = align_offset(header_size, s->cluster_size);
1510 s->l1_table_offset = offset;
1511 header.l1_table_offset = cpu_to_be64(s->l1_table_offset);
1512 header.l1_size = cpu_to_be32(l1_size);
1513 offset += align_offset(l1_size * sizeof(uint64_t), s->cluster_size);
1515 s->refcount_table = qemu_mallocz(s->cluster_size);
1516 if (!s->refcount_table)
1517 goto fail;
1518 s->refcount_block = qemu_mallocz(s->cluster_size);
1519 if (!s->refcount_block)
1520 goto fail;
1522 s->refcount_table_offset = offset;
1523 header.refcount_table_offset = cpu_to_be64(offset);
1524 header.refcount_table_clusters = cpu_to_be32(1);
1525 offset += s->cluster_size;
1527 s->refcount_table[0] = cpu_to_be64(offset);
1528 s->refcount_block_offset = offset;
1529 offset += s->cluster_size;
1531 /* update refcounts */
1532 create_refcount_update(s, 0, header_size);
1533 create_refcount_update(s, s->l1_table_offset, l1_size * sizeof(uint64_t));
1534 create_refcount_update(s, s->refcount_table_offset, s->cluster_size);
1535 create_refcount_update(s, s->refcount_block_offset, s->cluster_size);
1537 /* write all the data */
1538 write(fd, &header, sizeof(header));
1539 if (backing_file) {
1540 write(fd, backing_file, backing_filename_len);
1542 lseek(fd, s->l1_table_offset, SEEK_SET);
1543 tmp = 0;
1544 for(i = 0;i < l1_size; i++) {
1545 write(fd, &tmp, sizeof(tmp));
1547 lseek(fd, s->refcount_table_offset, SEEK_SET);
1548 write(fd, s->refcount_table, s->cluster_size);
1550 lseek(fd, s->refcount_block_offset, SEEK_SET);
1551 write(fd, s->refcount_block, s->cluster_size);
1553 qemu_free(s->refcount_table);
1554 qemu_free(s->refcount_block);
1555 close(fd);
1556 return 0;
1557 fail:
1558 qemu_free(s->refcount_table);
1559 qemu_free(s->refcount_block);
1560 close(fd);
1561 return -ENOMEM;
1564 static int qcow_make_empty(BlockDriverState *bs)
1566 #if 0
1567 /* XXX: not correct */
1568 BDRVQcowState *s = bs->opaque;
1569 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1570 int ret;
1572 memset(s->l1_table, 0, l1_length);
1573 if (bdrv_pwrite(s->hd, s->l1_table_offset, s->l1_table, l1_length) < 0)
1574 return -1;
1575 ret = bdrv_truncate(s->hd, s->l1_table_offset + l1_length);
1576 if (ret < 0)
1577 return ret;
1579 l2_cache_reset(bs);
1580 #endif
1581 return 0;
1584 /* XXX: put compressed sectors first, then all the cluster aligned
1585 tables to avoid losing bytes in alignment */
1586 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
1587 const uint8_t *buf, int nb_sectors)
1589 BDRVQcowState *s = bs->opaque;
1590 z_stream strm;
1591 int ret, out_len;
1592 uint8_t *out_buf;
1593 uint64_t cluster_offset;
1595 if (nb_sectors == 0) {
1596 /* align end of file to a sector boundary to ease reading with
1597 sector based I/Os */
1598 cluster_offset = bdrv_getlength(s->hd);
1599 cluster_offset = (cluster_offset + 511) & ~511;
1600 bdrv_truncate(s->hd, cluster_offset);
1601 return 0;
1604 if (nb_sectors != s->cluster_sectors)
1605 return -EINVAL;
1607 out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
1608 if (!out_buf)
1609 return -ENOMEM;
1611 /* best compression, small window, no zlib header */
1612 memset(&strm, 0, sizeof(strm));
1613 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1614 Z_DEFLATED, -12,
1615 9, Z_DEFAULT_STRATEGY);
1616 if (ret != 0) {
1617 qemu_free(out_buf);
1618 return -1;
1621 strm.avail_in = s->cluster_size;
1622 strm.next_in = (uint8_t *)buf;
1623 strm.avail_out = s->cluster_size;
1624 strm.next_out = out_buf;
1626 ret = deflate(&strm, Z_FINISH);
1627 if (ret != Z_STREAM_END && ret != Z_OK) {
1628 qemu_free(out_buf);
1629 deflateEnd(&strm);
1630 return -1;
1632 out_len = strm.next_out - out_buf;
1634 deflateEnd(&strm);
1636 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1637 /* could not compress: write normal cluster */
1638 qcow_write(bs, sector_num, buf, s->cluster_sectors);
1639 } else {
1640 cluster_offset = alloc_compressed_cluster_offset(bs, sector_num << 9,
1641 out_len);
1642 if (!cluster_offset)
1643 return -1;
1644 cluster_offset &= s->cluster_offset_mask;
1645 if (bdrv_pwrite(s->hd, cluster_offset, out_buf, out_len) != out_len) {
1646 qemu_free(out_buf);
1647 return -1;
1651 qemu_free(out_buf);
1652 return 0;
1655 static void qcow_flush(BlockDriverState *bs)
1657 BDRVQcowState *s = bs->opaque;
1658 bdrv_flush(s->hd);
1661 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1663 BDRVQcowState *s = bs->opaque;
1664 bdi->cluster_size = s->cluster_size;
1665 bdi->vm_state_offset = (int64_t)s->l1_vm_state_index <<
1666 (s->cluster_bits + s->l2_bits);
1667 return 0;
1670 /*********************************************************/
1671 /* snapshot support */
1673 /* update the refcounts of snapshots and the copied flag */
1674 static int update_snapshot_refcount(BlockDriverState *bs,
1675 int64_t l1_table_offset,
1676 int l1_size,
1677 int addend)
1679 BDRVQcowState *s = bs->opaque;
1680 uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated;
1681 int64_t old_offset, old_l2_offset;
1682 int l2_size, i, j, l1_modified, l2_modified, nb_csectors, refcount;
1684 l2_cache_reset(bs);
1686 l2_table = NULL;
1687 l1_table = NULL;
1688 l1_size2 = l1_size * sizeof(uint64_t);
1689 l1_allocated = 0;
1690 if (l1_table_offset != s->l1_table_offset) {
1691 l1_table = qemu_malloc(l1_size2);
1692 if (!l1_table)
1693 goto fail;
1694 l1_allocated = 1;
1695 if (bdrv_pread(s->hd, l1_table_offset,
1696 l1_table, l1_size2) != l1_size2)
1697 goto fail;
1698 for(i = 0;i < l1_size; i++)
1699 be64_to_cpus(&l1_table[i]);
1700 } else {
1701 assert(l1_size == s->l1_size);
1702 l1_table = s->l1_table;
1703 l1_allocated = 0;
1706 l2_size = s->l2_size * sizeof(uint64_t);
1707 l2_table = qemu_malloc(l2_size);
1708 if (!l2_table)
1709 goto fail;
1710 l1_modified = 0;
1711 for(i = 0; i < l1_size; i++) {
1712 l2_offset = l1_table[i];
1713 if (l2_offset) {
1714 old_l2_offset = l2_offset;
1715 l2_offset &= ~QCOW_OFLAG_COPIED;
1716 l2_modified = 0;
1717 if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
1718 goto fail;
1719 for(j = 0; j < s->l2_size; j++) {
1720 offset = be64_to_cpu(l2_table[j]);
1721 if (offset != 0) {
1722 old_offset = offset;
1723 offset &= ~QCOW_OFLAG_COPIED;
1724 if (offset & QCOW_OFLAG_COMPRESSED) {
1725 nb_csectors = ((offset >> s->csize_shift) &
1726 s->csize_mask) + 1;
1727 if (addend != 0)
1728 update_refcount(bs, (offset & s->cluster_offset_mask) & ~511,
1729 nb_csectors * 512, addend);
1730 /* compressed clusters are never modified */
1731 refcount = 2;
1732 } else {
1733 if (addend != 0) {
1734 refcount = update_cluster_refcount(bs, offset >> s->cluster_bits, addend);
1735 } else {
1736 refcount = get_refcount(bs, offset >> s->cluster_bits);
1740 if (refcount == 1) {
1741 offset |= QCOW_OFLAG_COPIED;
1743 if (offset != old_offset) {
1744 l2_table[j] = cpu_to_be64(offset);
1745 l2_modified = 1;
1749 if (l2_modified) {
1750 if (bdrv_pwrite(s->hd,
1751 l2_offset, l2_table, l2_size) != l2_size)
1752 goto fail;
1755 if (addend != 0) {
1756 refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend);
1757 } else {
1758 refcount = get_refcount(bs, l2_offset >> s->cluster_bits);
1760 if (refcount == 1) {
1761 l2_offset |= QCOW_OFLAG_COPIED;
1763 if (l2_offset != old_l2_offset) {
1764 l1_table[i] = l2_offset;
1765 l1_modified = 1;
1769 if (l1_modified) {
1770 for(i = 0; i < l1_size; i++)
1771 cpu_to_be64s(&l1_table[i]);
1772 if (bdrv_pwrite(s->hd, l1_table_offset, l1_table,
1773 l1_size2) != l1_size2)
1774 goto fail;
1775 for(i = 0; i < l1_size; i++)
1776 be64_to_cpus(&l1_table[i]);
1778 if (l1_allocated)
1779 qemu_free(l1_table);
1780 qemu_free(l2_table);
1781 return 0;
1782 fail:
1783 if (l1_allocated)
1784 qemu_free(l1_table);
1785 qemu_free(l2_table);
1786 return -EIO;
1789 static void qcow_free_snapshots(BlockDriverState *bs)
1791 BDRVQcowState *s = bs->opaque;
1792 int i;
1794 for(i = 0; i < s->nb_snapshots; i++) {
1795 qemu_free(s->snapshots[i].name);
1796 qemu_free(s->snapshots[i].id_str);
1798 qemu_free(s->snapshots);
1799 s->snapshots = NULL;
1800 s->nb_snapshots = 0;
1803 static int qcow_read_snapshots(BlockDriverState *bs)
1805 BDRVQcowState *s = bs->opaque;
1806 QCowSnapshotHeader h;
1807 QCowSnapshot *sn;
1808 int i, id_str_size, name_size;
1809 int64_t offset;
1810 uint32_t extra_data_size;
1812 offset = s->snapshots_offset;
1813 s->snapshots = qemu_mallocz(s->nb_snapshots * sizeof(QCowSnapshot));
1814 if (!s->snapshots)
1815 goto fail;
1816 for(i = 0; i < s->nb_snapshots; i++) {
1817 offset = align_offset(offset, 8);
1818 if (bdrv_pread(s->hd, offset, &h, sizeof(h)) != sizeof(h))
1819 goto fail;
1820 offset += sizeof(h);
1821 sn = s->snapshots + i;
1822 sn->l1_table_offset = be64_to_cpu(h.l1_table_offset);
1823 sn->l1_size = be32_to_cpu(h.l1_size);
1824 sn->vm_state_size = be32_to_cpu(h.vm_state_size);
1825 sn->date_sec = be32_to_cpu(h.date_sec);
1826 sn->date_nsec = be32_to_cpu(h.date_nsec);
1827 sn->vm_clock_nsec = be64_to_cpu(h.vm_clock_nsec);
1828 extra_data_size = be32_to_cpu(h.extra_data_size);
1830 id_str_size = be16_to_cpu(h.id_str_size);
1831 name_size = be16_to_cpu(h.name_size);
1833 offset += extra_data_size;
1835 sn->id_str = qemu_malloc(id_str_size + 1);
1836 if (!sn->id_str)
1837 goto fail;
1838 if (bdrv_pread(s->hd, offset, sn->id_str, id_str_size) != id_str_size)
1839 goto fail;
1840 offset += id_str_size;
1841 sn->id_str[id_str_size] = '\0';
1843 sn->name = qemu_malloc(name_size + 1);
1844 if (!sn->name)
1845 goto fail;
1846 if (bdrv_pread(s->hd, offset, sn->name, name_size) != name_size)
1847 goto fail;
1848 offset += name_size;
1849 sn->name[name_size] = '\0';
1851 s->snapshots_size = offset - s->snapshots_offset;
1852 return 0;
1853 fail:
1854 qcow_free_snapshots(bs);
1855 return -1;
1858 /* add at the end of the file a new list of snapshots */
1859 static int qcow_write_snapshots(BlockDriverState *bs)
1861 BDRVQcowState *s = bs->opaque;
1862 QCowSnapshot *sn;
1863 QCowSnapshotHeader h;
1864 int i, name_size, id_str_size, snapshots_size;
1865 uint64_t data64;
1866 uint32_t data32;
1867 int64_t offset, snapshots_offset;
1869 /* compute the size of the snapshots */
1870 offset = 0;
1871 for(i = 0; i < s->nb_snapshots; i++) {
1872 sn = s->snapshots + i;
1873 offset = align_offset(offset, 8);
1874 offset += sizeof(h);
1875 offset += strlen(sn->id_str);
1876 offset += strlen(sn->name);
1878 snapshots_size = offset;
1880 snapshots_offset = alloc_clusters(bs, snapshots_size);
1881 offset = snapshots_offset;
1883 for(i = 0; i < s->nb_snapshots; i++) {
1884 sn = s->snapshots + i;
1885 memset(&h, 0, sizeof(h));
1886 h.l1_table_offset = cpu_to_be64(sn->l1_table_offset);
1887 h.l1_size = cpu_to_be32(sn->l1_size);
1888 h.vm_state_size = cpu_to_be32(sn->vm_state_size);
1889 h.date_sec = cpu_to_be32(sn->date_sec);
1890 h.date_nsec = cpu_to_be32(sn->date_nsec);
1891 h.vm_clock_nsec = cpu_to_be64(sn->vm_clock_nsec);
1893 id_str_size = strlen(sn->id_str);
1894 name_size = strlen(sn->name);
1895 h.id_str_size = cpu_to_be16(id_str_size);
1896 h.name_size = cpu_to_be16(name_size);
1897 offset = align_offset(offset, 8);
1898 if (bdrv_pwrite(s->hd, offset, &h, sizeof(h)) != sizeof(h))
1899 goto fail;
1900 offset += sizeof(h);
1901 if (bdrv_pwrite(s->hd, offset, sn->id_str, id_str_size) != id_str_size)
1902 goto fail;
1903 offset += id_str_size;
1904 if (bdrv_pwrite(s->hd, offset, sn->name, name_size) != name_size)
1905 goto fail;
1906 offset += name_size;
1909 /* update the various header fields */
1910 data64 = cpu_to_be64(snapshots_offset);
1911 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, snapshots_offset),
1912 &data64, sizeof(data64)) != sizeof(data64))
1913 goto fail;
1914 data32 = cpu_to_be32(s->nb_snapshots);
1915 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, nb_snapshots),
1916 &data32, sizeof(data32)) != sizeof(data32))
1917 goto fail;
1919 /* free the old snapshot table */
1920 free_clusters(bs, s->snapshots_offset, s->snapshots_size);
1921 s->snapshots_offset = snapshots_offset;
1922 s->snapshots_size = snapshots_size;
1923 return 0;
1924 fail:
1925 return -1;
1928 static void find_new_snapshot_id(BlockDriverState *bs,
1929 char *id_str, int id_str_size)
1931 BDRVQcowState *s = bs->opaque;
1932 QCowSnapshot *sn;
1933 int i, id, id_max = 0;
1935 for(i = 0; i < s->nb_snapshots; i++) {
1936 sn = s->snapshots + i;
1937 id = strtoul(sn->id_str, NULL, 10);
1938 if (id > id_max)
1939 id_max = id;
1941 snprintf(id_str, id_str_size, "%d", id_max + 1);
1944 static int find_snapshot_by_id(BlockDriverState *bs, const char *id_str)
1946 BDRVQcowState *s = bs->opaque;
1947 int i;
1949 for(i = 0; i < s->nb_snapshots; i++) {
1950 if (!strcmp(s->snapshots[i].id_str, id_str))
1951 return i;
1953 return -1;
1956 static int find_snapshot_by_id_or_name(BlockDriverState *bs, const char *name)
1958 BDRVQcowState *s = bs->opaque;
1959 int i, ret;
1961 ret = find_snapshot_by_id(bs, name);
1962 if (ret >= 0)
1963 return ret;
1964 for(i = 0; i < s->nb_snapshots; i++) {
1965 if (!strcmp(s->snapshots[i].name, name))
1966 return i;
1968 return -1;
1971 /* if no id is provided, a new one is constructed */
1972 static int qcow_snapshot_create(BlockDriverState *bs,
1973 QEMUSnapshotInfo *sn_info)
1975 BDRVQcowState *s = bs->opaque;
1976 QCowSnapshot *snapshots1, sn1, *sn = &sn1;
1977 int i, ret;
1978 uint64_t *l1_table = NULL;
1980 memset(sn, 0, sizeof(*sn));
1982 if (sn_info->id_str[0] == '\0') {
1983 /* compute a new id */
1984 find_new_snapshot_id(bs, sn_info->id_str, sizeof(sn_info->id_str));
1987 /* check that the ID is unique */
1988 if (find_snapshot_by_id(bs, sn_info->id_str) >= 0)
1989 return -ENOENT;
1991 sn->id_str = qemu_strdup(sn_info->id_str);
1992 if (!sn->id_str)
1993 goto fail;
1994 sn->name = qemu_strdup(sn_info->name);
1995 if (!sn->name)
1996 goto fail;
1997 sn->vm_state_size = sn_info->vm_state_size;
1998 sn->date_sec = sn_info->date_sec;
1999 sn->date_nsec = sn_info->date_nsec;
2000 sn->vm_clock_nsec = sn_info->vm_clock_nsec;
2002 ret = update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1);
2003 if (ret < 0)
2004 goto fail;
2006 /* create the L1 table of the snapshot */
2007 sn->l1_table_offset = alloc_clusters(bs, s->l1_size * sizeof(uint64_t));
2008 sn->l1_size = s->l1_size;
2010 l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
2011 if (!l1_table)
2012 goto fail;
2013 for(i = 0; i < s->l1_size; i++) {
2014 l1_table[i] = cpu_to_be64(s->l1_table[i]);
2016 if (bdrv_pwrite(s->hd, sn->l1_table_offset,
2017 l1_table, s->l1_size * sizeof(uint64_t)) !=
2018 (s->l1_size * sizeof(uint64_t)))
2019 goto fail;
2020 qemu_free(l1_table);
2021 l1_table = NULL;
2023 snapshots1 = qemu_malloc((s->nb_snapshots + 1) * sizeof(QCowSnapshot));
2024 if (!snapshots1)
2025 goto fail;
2026 memcpy(snapshots1, s->snapshots, s->nb_snapshots * sizeof(QCowSnapshot));
2027 qemu_free(s->snapshots);
2028 s->snapshots = snapshots1;
2029 s->snapshots[s->nb_snapshots++] = *sn;
2031 if (qcow_write_snapshots(bs) < 0)
2032 goto fail;
2033 #ifdef DEBUG_ALLOC
2034 check_refcounts(bs);
2035 #endif
2036 return 0;
2037 fail:
2038 qemu_free(sn->name);
2039 qemu_free(l1_table);
2040 return -1;
2043 /* copy the snapshot 'snapshot_name' into the current disk image */
2044 static int qcow_snapshot_goto(BlockDriverState *bs,
2045 const char *snapshot_id)
2047 BDRVQcowState *s = bs->opaque;
2048 QCowSnapshot *sn;
2049 int i, snapshot_index, l1_size2;
2051 snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
2052 if (snapshot_index < 0)
2053 return -ENOENT;
2054 sn = &s->snapshots[snapshot_index];
2056 if (update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, -1) < 0)
2057 goto fail;
2059 if (grow_l1_table(bs, sn->l1_size) < 0)
2060 goto fail;
2062 s->l1_size = sn->l1_size;
2063 l1_size2 = s->l1_size * sizeof(uint64_t);
2064 /* copy the snapshot l1 table to the current l1 table */
2065 if (bdrv_pread(s->hd, sn->l1_table_offset,
2066 s->l1_table, l1_size2) != l1_size2)
2067 goto fail;
2068 if (bdrv_pwrite(s->hd, s->l1_table_offset,
2069 s->l1_table, l1_size2) != l1_size2)
2070 goto fail;
2071 for(i = 0;i < s->l1_size; i++) {
2072 be64_to_cpus(&s->l1_table[i]);
2075 if (update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1) < 0)
2076 goto fail;
2078 #ifdef DEBUG_ALLOC
2079 check_refcounts(bs);
2080 #endif
2081 return 0;
2082 fail:
2083 return -EIO;
2086 static int qcow_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
2088 BDRVQcowState *s = bs->opaque;
2089 QCowSnapshot *sn;
2090 int snapshot_index, ret;
2092 snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
2093 if (snapshot_index < 0)
2094 return -ENOENT;
2095 sn = &s->snapshots[snapshot_index];
2097 ret = update_snapshot_refcount(bs, sn->l1_table_offset, sn->l1_size, -1);
2098 if (ret < 0)
2099 return ret;
2100 /* must update the copied flag on the current cluster offsets */
2101 ret = update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 0);
2102 if (ret < 0)
2103 return ret;
2104 free_clusters(bs, sn->l1_table_offset, sn->l1_size * sizeof(uint64_t));
2106 qemu_free(sn->id_str);
2107 qemu_free(sn->name);
2108 memmove(sn, sn + 1, (s->nb_snapshots - snapshot_index - 1) * sizeof(*sn));
2109 s->nb_snapshots--;
2110 ret = qcow_write_snapshots(bs);
2111 if (ret < 0) {
2112 /* XXX: restore snapshot if error ? */
2113 return ret;
2115 #ifdef DEBUG_ALLOC
2116 check_refcounts(bs);
2117 #endif
2118 return 0;
2121 static int qcow_snapshot_list(BlockDriverState *bs,
2122 QEMUSnapshotInfo **psn_tab)
2124 BDRVQcowState *s = bs->opaque;
2125 QEMUSnapshotInfo *sn_tab, *sn_info;
2126 QCowSnapshot *sn;
2127 int i;
2129 sn_tab = qemu_mallocz(s->nb_snapshots * sizeof(QEMUSnapshotInfo));
2130 if (!sn_tab)
2131 goto fail;
2132 for(i = 0; i < s->nb_snapshots; i++) {
2133 sn_info = sn_tab + i;
2134 sn = s->snapshots + i;
2135 pstrcpy(sn_info->id_str, sizeof(sn_info->id_str),
2136 sn->id_str);
2137 pstrcpy(sn_info->name, sizeof(sn_info->name),
2138 sn->name);
2139 sn_info->vm_state_size = sn->vm_state_size;
2140 sn_info->date_sec = sn->date_sec;
2141 sn_info->date_nsec = sn->date_nsec;
2142 sn_info->vm_clock_nsec = sn->vm_clock_nsec;
2144 *psn_tab = sn_tab;
2145 return s->nb_snapshots;
2146 fail:
2147 qemu_free(sn_tab);
2148 *psn_tab = NULL;
2149 return -ENOMEM;
2152 /*********************************************************/
2153 /* refcount handling */
2155 static int refcount_init(BlockDriverState *bs)
2157 BDRVQcowState *s = bs->opaque;
2158 int ret, refcount_table_size2, i;
2160 s->refcount_block_cache = qemu_malloc(s->cluster_size);
2161 if (!s->refcount_block_cache)
2162 goto fail;
2163 refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t);
2164 s->refcount_table = qemu_malloc(refcount_table_size2);
2165 if (!s->refcount_table)
2166 goto fail;
2167 if (s->refcount_table_size > 0) {
2168 ret = bdrv_pread(s->hd, s->refcount_table_offset,
2169 s->refcount_table, refcount_table_size2);
2170 if (ret != refcount_table_size2)
2171 goto fail;
2172 for(i = 0; i < s->refcount_table_size; i++)
2173 be64_to_cpus(&s->refcount_table[i]);
2175 return 0;
2176 fail:
2177 return -ENOMEM;
2180 static void refcount_close(BlockDriverState *bs)
2182 BDRVQcowState *s = bs->opaque;
2183 qemu_free(s->refcount_block_cache);
2184 qemu_free(s->refcount_table);
2188 static int load_refcount_block(BlockDriverState *bs,
2189 int64_t refcount_block_offset)
2191 BDRVQcowState *s = bs->opaque;
2192 int ret;
2193 ret = bdrv_pread(s->hd, refcount_block_offset, s->refcount_block_cache,
2194 s->cluster_size);
2195 if (ret != s->cluster_size)
2196 return -EIO;
2197 s->refcount_block_cache_offset = refcount_block_offset;
2198 return 0;
2201 static int get_refcount(BlockDriverState *bs, int64_t cluster_index)
2203 BDRVQcowState *s = bs->opaque;
2204 int refcount_table_index, block_index;
2205 int64_t refcount_block_offset;
2207 refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
2208 if (refcount_table_index >= s->refcount_table_size)
2209 return 0;
2210 refcount_block_offset = s->refcount_table[refcount_table_index];
2211 if (!refcount_block_offset)
2212 return 0;
2213 if (refcount_block_offset != s->refcount_block_cache_offset) {
2214 /* better than nothing: return allocated if read error */
2215 if (load_refcount_block(bs, refcount_block_offset) < 0)
2216 return 1;
2218 block_index = cluster_index &
2219 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
2220 return be16_to_cpu(s->refcount_block_cache[block_index]);
2223 /* return < 0 if error */
2224 static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size)
2226 BDRVQcowState *s = bs->opaque;
2227 int i, nb_clusters;
2229 nb_clusters = size_to_clusters(s, size);
2230 retry:
2231 for(i = 0; i < nb_clusters; i++) {
2232 int64_t i = s->free_cluster_index++;
2233 if (get_refcount(bs, i) != 0)
2234 goto retry;
2236 #ifdef DEBUG_ALLOC2
2237 printf("alloc_clusters: size=%lld -> %lld\n",
2238 size,
2239 (s->free_cluster_index - nb_clusters) << s->cluster_bits);
2240 #endif
2241 return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
2244 static int64_t alloc_clusters(BlockDriverState *bs, int64_t size)
2246 int64_t offset;
2248 offset = alloc_clusters_noref(bs, size);
2249 update_refcount(bs, offset, size, 1);
2250 return offset;
2253 /* only used to allocate compressed sectors. We try to allocate
2254 contiguous sectors. size must be <= cluster_size */
2255 static int64_t alloc_bytes(BlockDriverState *bs, int size)
2257 BDRVQcowState *s = bs->opaque;
2258 int64_t offset, cluster_offset;
2259 int free_in_cluster;
2261 assert(size > 0 && size <= s->cluster_size);
2262 if (s->free_byte_offset == 0) {
2263 s->free_byte_offset = alloc_clusters(bs, s->cluster_size);
2265 redo:
2266 free_in_cluster = s->cluster_size -
2267 (s->free_byte_offset & (s->cluster_size - 1));
2268 if (size <= free_in_cluster) {
2269 /* enough space in current cluster */
2270 offset = s->free_byte_offset;
2271 s->free_byte_offset += size;
2272 free_in_cluster -= size;
2273 if (free_in_cluster == 0)
2274 s->free_byte_offset = 0;
2275 if ((offset & (s->cluster_size - 1)) != 0)
2276 update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
2277 } else {
2278 offset = alloc_clusters(bs, s->cluster_size);
2279 cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1);
2280 if ((cluster_offset + s->cluster_size) == offset) {
2281 /* we are lucky: contiguous data */
2282 offset = s->free_byte_offset;
2283 update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
2284 s->free_byte_offset += size;
2285 } else {
2286 s->free_byte_offset = offset;
2287 goto redo;
2290 return offset;
2293 static void free_clusters(BlockDriverState *bs,
2294 int64_t offset, int64_t size)
2296 update_refcount(bs, offset, size, -1);
2299 static int grow_refcount_table(BlockDriverState *bs, int min_size)
2301 BDRVQcowState *s = bs->opaque;
2302 int new_table_size, new_table_size2, refcount_table_clusters, i, ret;
2303 uint64_t *new_table;
2304 int64_t table_offset;
2305 uint8_t data[12];
2306 int old_table_size;
2307 int64_t old_table_offset;
2309 if (min_size <= s->refcount_table_size)
2310 return 0;
2311 /* compute new table size */
2312 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
2313 for(;;) {
2314 if (refcount_table_clusters == 0) {
2315 refcount_table_clusters = 1;
2316 } else {
2317 refcount_table_clusters = (refcount_table_clusters * 3 + 1) / 2;
2319 new_table_size = refcount_table_clusters << (s->cluster_bits - 3);
2320 if (min_size <= new_table_size)
2321 break;
2323 #ifdef DEBUG_ALLOC2
2324 printf("grow_refcount_table from %d to %d\n",
2325 s->refcount_table_size,
2326 new_table_size);
2327 #endif
2328 new_table_size2 = new_table_size * sizeof(uint64_t);
2329 new_table = qemu_mallocz(new_table_size2);
2330 if (!new_table)
2331 return -ENOMEM;
2332 memcpy(new_table, s->refcount_table,
2333 s->refcount_table_size * sizeof(uint64_t));
2334 for(i = 0; i < s->refcount_table_size; i++)
2335 cpu_to_be64s(&new_table[i]);
2336 /* Note: we cannot update the refcount now to avoid recursion */
2337 table_offset = alloc_clusters_noref(bs, new_table_size2);
2338 ret = bdrv_pwrite(s->hd, table_offset, new_table, new_table_size2);
2339 if (ret != new_table_size2)
2340 goto fail;
2341 for(i = 0; i < s->refcount_table_size; i++)
2342 be64_to_cpus(&new_table[i]);
2344 cpu_to_be64w((uint64_t*)data, table_offset);
2345 cpu_to_be32w((uint32_t*)(data + 8), refcount_table_clusters);
2346 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, refcount_table_offset),
2347 data, sizeof(data)) != sizeof(data))
2348 goto fail;
2349 qemu_free(s->refcount_table);
2350 old_table_offset = s->refcount_table_offset;
2351 old_table_size = s->refcount_table_size;
2352 s->refcount_table = new_table;
2353 s->refcount_table_size = new_table_size;
2354 s->refcount_table_offset = table_offset;
2356 update_refcount(bs, table_offset, new_table_size2, 1);
2357 free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t));
2358 return 0;
2359 fail:
2360 free_clusters(bs, table_offset, new_table_size2);
2361 qemu_free(new_table);
2362 return -EIO;
2365 /* addend must be 1 or -1 */
2366 /* XXX: cache several refcount block clusters ? */
2367 static int update_cluster_refcount(BlockDriverState *bs,
2368 int64_t cluster_index,
2369 int addend)
2371 BDRVQcowState *s = bs->opaque;
2372 int64_t offset, refcount_block_offset;
2373 int ret, refcount_table_index, block_index, refcount;
2374 uint64_t data64;
2376 refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
2377 if (refcount_table_index >= s->refcount_table_size) {
2378 if (addend < 0)
2379 return -EINVAL;
2380 ret = grow_refcount_table(bs, refcount_table_index + 1);
2381 if (ret < 0)
2382 return ret;
2384 refcount_block_offset = s->refcount_table[refcount_table_index];
2385 if (!refcount_block_offset) {
2386 if (addend < 0)
2387 return -EINVAL;
2388 /* create a new refcount block */
2389 /* Note: we cannot update the refcount now to avoid recursion */
2390 offset = alloc_clusters_noref(bs, s->cluster_size);
2391 memset(s->refcount_block_cache, 0, s->cluster_size);
2392 ret = bdrv_pwrite(s->hd, offset, s->refcount_block_cache, s->cluster_size);
2393 if (ret != s->cluster_size)
2394 return -EINVAL;
2395 s->refcount_table[refcount_table_index] = offset;
2396 data64 = cpu_to_be64(offset);
2397 ret = bdrv_pwrite(s->hd, s->refcount_table_offset +
2398 refcount_table_index * sizeof(uint64_t),
2399 &data64, sizeof(data64));
2400 if (ret != sizeof(data64))
2401 return -EINVAL;
2403 refcount_block_offset = offset;
2404 s->refcount_block_cache_offset = offset;
2405 update_refcount(bs, offset, s->cluster_size, 1);
2406 } else {
2407 if (refcount_block_offset != s->refcount_block_cache_offset) {
2408 if (load_refcount_block(bs, refcount_block_offset) < 0)
2409 return -EIO;
2412 /* we can update the count and save it */
2413 block_index = cluster_index &
2414 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
2415 refcount = be16_to_cpu(s->refcount_block_cache[block_index]);
2416 refcount += addend;
2417 if (refcount < 0 || refcount > 0xffff)
2418 return -EINVAL;
2419 if (refcount == 0 && cluster_index < s->free_cluster_index) {
2420 s->free_cluster_index = cluster_index;
2422 s->refcount_block_cache[block_index] = cpu_to_be16(refcount);
2423 if (bdrv_pwrite(s->hd,
2424 refcount_block_offset + (block_index << REFCOUNT_SHIFT),
2425 &s->refcount_block_cache[block_index], 2) != 2)
2426 return -EIO;
2427 return refcount;
2430 static void update_refcount(BlockDriverState *bs,
2431 int64_t offset, int64_t length,
2432 int addend)
2434 BDRVQcowState *s = bs->opaque;
2435 int64_t start, last, cluster_offset;
2437 #ifdef DEBUG_ALLOC2
2438 printf("update_refcount: offset=%lld size=%lld addend=%d\n",
2439 offset, length, addend);
2440 #endif
2441 if (length <= 0)
2442 return;
2443 start = offset & ~(s->cluster_size - 1);
2444 last = (offset + length - 1) & ~(s->cluster_size - 1);
2445 for(cluster_offset = start; cluster_offset <= last;
2446 cluster_offset += s->cluster_size) {
2447 update_cluster_refcount(bs, cluster_offset >> s->cluster_bits, addend);
2451 #ifdef DEBUG_ALLOC
2452 static void inc_refcounts(BlockDriverState *bs,
2453 uint16_t *refcount_table,
2454 int refcount_table_size,
2455 int64_t offset, int64_t size)
2457 BDRVQcowState *s = bs->opaque;
2458 int64_t start, last, cluster_offset;
2459 int k;
2461 if (size <= 0)
2462 return;
2464 start = offset & ~(s->cluster_size - 1);
2465 last = (offset + size - 1) & ~(s->cluster_size - 1);
2466 for(cluster_offset = start; cluster_offset <= last;
2467 cluster_offset += s->cluster_size) {
2468 k = cluster_offset >> s->cluster_bits;
2469 if (k < 0 || k >= refcount_table_size) {
2470 printf("ERROR: invalid cluster offset=0x%llx\n", cluster_offset);
2471 } else {
2472 if (++refcount_table[k] == 0) {
2473 printf("ERROR: overflow cluster offset=0x%llx\n", cluster_offset);
2479 static int check_refcounts_l1(BlockDriverState *bs,
2480 uint16_t *refcount_table,
2481 int refcount_table_size,
2482 int64_t l1_table_offset, int l1_size,
2483 int check_copied)
2485 BDRVQcowState *s = bs->opaque;
2486 uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2;
2487 int l2_size, i, j, nb_csectors, refcount;
2489 l2_table = NULL;
2490 l1_size2 = l1_size * sizeof(uint64_t);
2492 inc_refcounts(bs, refcount_table, refcount_table_size,
2493 l1_table_offset, l1_size2);
2495 l1_table = qemu_malloc(l1_size2);
2496 if (!l1_table)
2497 goto fail;
2498 if (bdrv_pread(s->hd, l1_table_offset,
2499 l1_table, l1_size2) != l1_size2)
2500 goto fail;
2501 for(i = 0;i < l1_size; i++)
2502 be64_to_cpus(&l1_table[i]);
2504 l2_size = s->l2_size * sizeof(uint64_t);
2505 l2_table = qemu_malloc(l2_size);
2506 if (!l2_table)
2507 goto fail;
2508 for(i = 0; i < l1_size; i++) {
2509 l2_offset = l1_table[i];
2510 if (l2_offset) {
2511 if (check_copied) {
2512 refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED) >> s->cluster_bits);
2513 if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) {
2514 printf("ERROR OFLAG_COPIED: l2_offset=%llx refcount=%d\n",
2515 l2_offset, refcount);
2518 l2_offset &= ~QCOW_OFLAG_COPIED;
2519 if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
2520 goto fail;
2521 for(j = 0; j < s->l2_size; j++) {
2522 offset = be64_to_cpu(l2_table[j]);
2523 if (offset != 0) {
2524 if (offset & QCOW_OFLAG_COMPRESSED) {
2525 if (offset & QCOW_OFLAG_COPIED) {
2526 printf("ERROR: cluster %lld: copied flag must never be set for compressed clusters\n",
2527 offset >> s->cluster_bits);
2528 offset &= ~QCOW_OFLAG_COPIED;
2530 nb_csectors = ((offset >> s->csize_shift) &
2531 s->csize_mask) + 1;
2532 offset &= s->cluster_offset_mask;
2533 inc_refcounts(bs, refcount_table,
2534 refcount_table_size,
2535 offset & ~511, nb_csectors * 512);
2536 } else {
2537 if (check_copied) {
2538 refcount = get_refcount(bs, (offset & ~QCOW_OFLAG_COPIED) >> s->cluster_bits);
2539 if ((refcount == 1) != ((offset & QCOW_OFLAG_COPIED) != 0)) {
2540 printf("ERROR OFLAG_COPIED: offset=%llx refcount=%d\n",
2541 offset, refcount);
2544 offset &= ~QCOW_OFLAG_COPIED;
2545 inc_refcounts(bs, refcount_table,
2546 refcount_table_size,
2547 offset, s->cluster_size);
2551 inc_refcounts(bs, refcount_table,
2552 refcount_table_size,
2553 l2_offset,
2554 s->cluster_size);
2557 qemu_free(l1_table);
2558 qemu_free(l2_table);
2559 return 0;
2560 fail:
2561 printf("ERROR: I/O error in check_refcounts_l1\n");
2562 qemu_free(l1_table);
2563 qemu_free(l2_table);
2564 return -EIO;
2567 static void check_refcounts(BlockDriverState *bs)
2569 BDRVQcowState *s = bs->opaque;
2570 int64_t size;
2571 int nb_clusters, refcount1, refcount2, i;
2572 QCowSnapshot *sn;
2573 uint16_t *refcount_table;
2575 size = bdrv_getlength(s->hd);
2576 nb_clusters = size_to_clusters(s, size);
2577 refcount_table = qemu_mallocz(nb_clusters * sizeof(uint16_t));
2579 /* header */
2580 inc_refcounts(bs, refcount_table, nb_clusters,
2581 0, s->cluster_size);
2583 check_refcounts_l1(bs, refcount_table, nb_clusters,
2584 s->l1_table_offset, s->l1_size, 1);
2586 /* snapshots */
2587 for(i = 0; i < s->nb_snapshots; i++) {
2588 sn = s->snapshots + i;
2589 check_refcounts_l1(bs, refcount_table, nb_clusters,
2590 sn->l1_table_offset, sn->l1_size, 0);
2592 inc_refcounts(bs, refcount_table, nb_clusters,
2593 s->snapshots_offset, s->snapshots_size);
2595 /* refcount data */
2596 inc_refcounts(bs, refcount_table, nb_clusters,
2597 s->refcount_table_offset,
2598 s->refcount_table_size * sizeof(uint64_t));
2599 for(i = 0; i < s->refcount_table_size; i++) {
2600 int64_t offset;
2601 offset = s->refcount_table[i];
2602 if (offset != 0) {
2603 inc_refcounts(bs, refcount_table, nb_clusters,
2604 offset, s->cluster_size);
2608 /* compare ref counts */
2609 for(i = 0; i < nb_clusters; i++) {
2610 refcount1 = get_refcount(bs, i);
2611 refcount2 = refcount_table[i];
2612 if (refcount1 != refcount2)
2613 printf("ERROR cluster %d refcount=%d reference=%d\n",
2614 i, refcount1, refcount2);
2617 qemu_free(refcount_table);
2620 #if 0
2621 static void dump_refcounts(BlockDriverState *bs)
2623 BDRVQcowState *s = bs->opaque;
2624 int64_t nb_clusters, k, k1, size;
2625 int refcount;
2627 size = bdrv_getlength(s->hd);
2628 nb_clusters = size_to_clusters(s, size);
2629 for(k = 0; k < nb_clusters;) {
2630 k1 = k;
2631 refcount = get_refcount(bs, k);
2632 k++;
2633 while (k < nb_clusters && get_refcount(bs, k) == refcount)
2634 k++;
2635 printf("%lld: refcount=%d nb=%lld\n", k, refcount, k - k1);
2638 #endif
2639 #endif
2641 BlockDriver bdrv_qcow2 = {
2642 "qcow2",
2643 sizeof(BDRVQcowState),
2644 qcow_probe,
2645 qcow_open,
2646 NULL,
2647 NULL,
2648 qcow_close,
2649 qcow_create,
2650 qcow_flush,
2651 qcow_is_allocated,
2652 qcow_set_key,
2653 qcow_make_empty,
2655 .bdrv_aio_read = qcow_aio_read,
2656 .bdrv_aio_write = qcow_aio_write,
2657 .bdrv_aio_cancel = qcow_aio_cancel,
2658 .aiocb_size = sizeof(QCowAIOCB),
2659 .bdrv_write_compressed = qcow_write_compressed,
2661 .bdrv_snapshot_create = qcow_snapshot_create,
2662 .bdrv_snapshot_goto = qcow_snapshot_goto,
2663 .bdrv_snapshot_delete = qcow_snapshot_delete,
2664 .bdrv_snapshot_list = qcow_snapshot_list,
2665 .bdrv_get_info = qcow_get_info,