qed: fix use-after-free during l2 cache commit
[qemu.git] / block / qcow2-cluster.c
blob882f50a80b64137b75221b8c9fd8e1e6761b8641
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.
25 #include <zlib.h>
27 #include "qemu-common.h"
28 #include "block_int.h"
29 #include "block/qcow2.h"
31 int qcow2_grow_l1_table(BlockDriverState *bs, int min_size, bool exact_size)
33 BDRVQcowState *s = bs->opaque;
34 int new_l1_size, new_l1_size2, ret, i;
35 uint64_t *new_l1_table;
36 int64_t new_l1_table_offset;
37 uint8_t data[12];
39 if (min_size <= s->l1_size)
40 return 0;
42 if (exact_size) {
43 new_l1_size = min_size;
44 } else {
45 /* Bump size up to reduce the number of times we have to grow */
46 new_l1_size = s->l1_size;
47 if (new_l1_size == 0) {
48 new_l1_size = 1;
50 while (min_size > new_l1_size) {
51 new_l1_size = (new_l1_size * 3 + 1) / 2;
55 #ifdef DEBUG_ALLOC2
56 printf("grow l1_table from %d to %d\n", s->l1_size, new_l1_size);
57 #endif
59 new_l1_size2 = sizeof(uint64_t) * new_l1_size;
60 new_l1_table = qemu_mallocz(align_offset(new_l1_size2, 512));
61 memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t));
63 /* write new table (align to cluster) */
64 BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ALLOC_TABLE);
65 new_l1_table_offset = qcow2_alloc_clusters(bs, new_l1_size2);
66 if (new_l1_table_offset < 0) {
67 qemu_free(new_l1_table);
68 return new_l1_table_offset;
71 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
72 if (ret < 0) {
73 goto fail;
76 BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_WRITE_TABLE);
77 for(i = 0; i < s->l1_size; i++)
78 new_l1_table[i] = cpu_to_be64(new_l1_table[i]);
79 ret = bdrv_pwrite_sync(bs->file, new_l1_table_offset, new_l1_table, new_l1_size2);
80 if (ret < 0)
81 goto fail;
82 for(i = 0; i < s->l1_size; i++)
83 new_l1_table[i] = be64_to_cpu(new_l1_table[i]);
85 /* set new table */
86 BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ACTIVATE_TABLE);
87 cpu_to_be32w((uint32_t*)data, new_l1_size);
88 cpu_to_be64wu((uint64_t*)(data + 4), new_l1_table_offset);
89 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_size), data,sizeof(data));
90 if (ret < 0) {
91 goto fail;
93 qemu_free(s->l1_table);
94 qcow2_free_clusters(bs, s->l1_table_offset, s->l1_size * sizeof(uint64_t));
95 s->l1_table_offset = new_l1_table_offset;
96 s->l1_table = new_l1_table;
97 s->l1_size = new_l1_size;
98 return 0;
99 fail:
100 qemu_free(new_l1_table);
101 qcow2_free_clusters(bs, new_l1_table_offset, new_l1_size2);
102 return ret;
106 * l2_load
108 * Loads a L2 table into memory. If the table is in the cache, the cache
109 * is used; otherwise the L2 table is loaded from the image file.
111 * Returns a pointer to the L2 table on success, or NULL if the read from
112 * the image file failed.
115 static int l2_load(BlockDriverState *bs, uint64_t l2_offset,
116 uint64_t **l2_table)
118 BDRVQcowState *s = bs->opaque;
119 int ret;
121 ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset, (void**) l2_table);
123 return ret;
127 * Writes one sector of the L1 table to the disk (can't update single entries
128 * and we really don't want bdrv_pread to perform a read-modify-write)
130 #define L1_ENTRIES_PER_SECTOR (512 / 8)
131 static int write_l1_entry(BlockDriverState *bs, int l1_index)
133 BDRVQcowState *s = bs->opaque;
134 uint64_t buf[L1_ENTRIES_PER_SECTOR];
135 int l1_start_index;
136 int i, ret;
138 l1_start_index = l1_index & ~(L1_ENTRIES_PER_SECTOR - 1);
139 for (i = 0; i < L1_ENTRIES_PER_SECTOR; i++) {
140 buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]);
143 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
144 ret = bdrv_pwrite_sync(bs->file, s->l1_table_offset + 8 * l1_start_index,
145 buf, sizeof(buf));
146 if (ret < 0) {
147 return ret;
150 return 0;
154 * l2_allocate
156 * Allocate a new l2 entry in the file. If l1_index points to an already
157 * used entry in the L2 table (i.e. we are doing a copy on write for the L2
158 * table) copy the contents of the old L2 table into the newly allocated one.
159 * Otherwise the new table is initialized with zeros.
163 static int l2_allocate(BlockDriverState *bs, int l1_index, uint64_t **table)
165 BDRVQcowState *s = bs->opaque;
166 uint64_t old_l2_offset;
167 uint64_t *l2_table;
168 int64_t l2_offset;
169 int ret;
171 old_l2_offset = s->l1_table[l1_index];
173 /* allocate a new l2 entry */
175 l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
176 if (l2_offset < 0) {
177 return l2_offset;
180 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
181 if (ret < 0) {
182 goto fail;
185 /* allocate a new entry in the l2 cache */
187 ret = qcow2_cache_get_empty(bs, s->l2_table_cache, l2_offset, (void**) table);
188 if (ret < 0) {
189 return ret;
192 l2_table = *table;
194 if (old_l2_offset == 0) {
195 /* if there was no old l2 table, clear the new table */
196 memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
197 } else {
198 uint64_t* old_table;
200 /* if there was an old l2 table, read it from the disk */
201 BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_COW_READ);
202 ret = qcow2_cache_get(bs, s->l2_table_cache, old_l2_offset,
203 (void**) &old_table);
204 if (ret < 0) {
205 goto fail;
208 memcpy(l2_table, old_table, s->cluster_size);
210 ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &old_table);
211 if (ret < 0) {
212 goto fail;
216 /* write the l2 table to the file */
217 BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_WRITE);
219 qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
220 ret = qcow2_cache_flush(bs, s->l2_table_cache);
221 if (ret < 0) {
222 goto fail;
225 /* update the L1 entry */
226 s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
227 ret = write_l1_entry(bs, l1_index);
228 if (ret < 0) {
229 goto fail;
232 *table = l2_table;
233 return 0;
235 fail:
236 qcow2_cache_put(bs, s->l2_table_cache, (void**) table);
237 s->l1_table[l1_index] = old_l2_offset;
238 return ret;
241 static int count_contiguous_clusters(uint64_t nb_clusters, int cluster_size,
242 uint64_t *l2_table, uint64_t start, uint64_t mask)
244 int i;
245 uint64_t offset = be64_to_cpu(l2_table[0]) & ~mask;
247 if (!offset)
248 return 0;
250 for (i = start; i < start + nb_clusters; i++)
251 if (offset + (uint64_t) i * cluster_size != (be64_to_cpu(l2_table[i]) & ~mask))
252 break;
254 return (i - start);
257 static int count_contiguous_free_clusters(uint64_t nb_clusters, uint64_t *l2_table)
259 int i = 0;
261 while(nb_clusters-- && l2_table[i] == 0)
262 i++;
264 return i;
267 /* The crypt function is compatible with the linux cryptoloop
268 algorithm for < 4 GB images. NOTE: out_buf == in_buf is
269 supported */
270 void qcow2_encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
271 uint8_t *out_buf, const uint8_t *in_buf,
272 int nb_sectors, int enc,
273 const AES_KEY *key)
275 union {
276 uint64_t ll[2];
277 uint8_t b[16];
278 } ivec;
279 int i;
281 for(i = 0; i < nb_sectors; i++) {
282 ivec.ll[0] = cpu_to_le64(sector_num);
283 ivec.ll[1] = 0;
284 AES_cbc_encrypt(in_buf, out_buf, 512, key,
285 ivec.b, enc);
286 sector_num++;
287 in_buf += 512;
288 out_buf += 512;
293 static int qcow2_read(BlockDriverState *bs, int64_t sector_num,
294 uint8_t *buf, int nb_sectors)
296 BDRVQcowState *s = bs->opaque;
297 int ret, index_in_cluster, n, n1;
298 uint64_t cluster_offset;
299 struct iovec iov;
300 QEMUIOVector qiov;
302 while (nb_sectors > 0) {
303 n = nb_sectors;
305 ret = qcow2_get_cluster_offset(bs, sector_num << 9, &n,
306 &cluster_offset);
307 if (ret < 0) {
308 return ret;
311 index_in_cluster = sector_num & (s->cluster_sectors - 1);
312 if (!cluster_offset) {
313 if (bs->backing_hd) {
314 /* read from the base image */
315 iov.iov_base = buf;
316 iov.iov_len = n * 512;
317 qemu_iovec_init_external(&qiov, &iov, 1);
319 n1 = qcow2_backing_read1(bs->backing_hd, &qiov, sector_num, n);
320 if (n1 > 0) {
321 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING);
322 ret = bdrv_read(bs->backing_hd, sector_num, buf, n1);
323 if (ret < 0)
324 return -1;
326 } else {
327 memset(buf, 0, 512 * n);
329 } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
330 if (qcow2_decompress_cluster(bs, cluster_offset) < 0)
331 return -1;
332 memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n);
333 } else {
334 BLKDBG_EVENT(bs->file, BLKDBG_READ);
335 ret = bdrv_pread(bs->file, cluster_offset + index_in_cluster * 512, buf, n * 512);
336 if (ret != n * 512)
337 return -1;
338 if (s->crypt_method) {
339 qcow2_encrypt_sectors(s, sector_num, buf, buf, n, 0,
340 &s->aes_decrypt_key);
343 nb_sectors -= n;
344 sector_num += n;
345 buf += n * 512;
347 return 0;
350 static int copy_sectors(BlockDriverState *bs, uint64_t start_sect,
351 uint64_t cluster_offset, int n_start, int n_end)
353 BDRVQcowState *s = bs->opaque;
354 int n, ret;
356 n = n_end - n_start;
357 if (n <= 0)
358 return 0;
359 BLKDBG_EVENT(bs->file, BLKDBG_COW_READ);
360 ret = qcow2_read(bs, start_sect + n_start, s->cluster_data, n);
361 if (ret < 0)
362 return ret;
363 if (s->crypt_method) {
364 qcow2_encrypt_sectors(s, start_sect + n_start,
365 s->cluster_data,
366 s->cluster_data, n, 1,
367 &s->aes_encrypt_key);
369 BLKDBG_EVENT(bs->file, BLKDBG_COW_WRITE);
370 ret = bdrv_write(bs->file, (cluster_offset >> 9) + n_start,
371 s->cluster_data, n);
372 if (ret < 0)
373 return ret;
374 return 0;
379 * get_cluster_offset
381 * For a given offset of the disk image, find the cluster offset in
382 * qcow2 file. The offset is stored in *cluster_offset.
384 * on entry, *num is the number of contiguous clusters we'd like to
385 * access following offset.
387 * on exit, *num is the number of contiguous clusters we can read.
389 * Return 0, if the offset is found
390 * Return -errno, otherwise.
394 int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
395 int *num, uint64_t *cluster_offset)
397 BDRVQcowState *s = bs->opaque;
398 unsigned int l1_index, l2_index;
399 uint64_t l2_offset, *l2_table;
400 int l1_bits, c;
401 unsigned int index_in_cluster, nb_clusters;
402 uint64_t nb_available, nb_needed;
403 int ret;
405 index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1);
406 nb_needed = *num + index_in_cluster;
408 l1_bits = s->l2_bits + s->cluster_bits;
410 /* compute how many bytes there are between the offset and
411 * the end of the l1 entry
414 nb_available = (1ULL << l1_bits) - (offset & ((1ULL << l1_bits) - 1));
416 /* compute the number of available sectors */
418 nb_available = (nb_available >> 9) + index_in_cluster;
420 if (nb_needed > nb_available) {
421 nb_needed = nb_available;
424 *cluster_offset = 0;
426 /* seek the the l2 offset in the l1 table */
428 l1_index = offset >> l1_bits;
429 if (l1_index >= s->l1_size)
430 goto out;
432 l2_offset = s->l1_table[l1_index];
434 /* seek the l2 table of the given l2 offset */
436 if (!l2_offset)
437 goto out;
439 /* load the l2 table in memory */
441 l2_offset &= ~QCOW_OFLAG_COPIED;
442 ret = l2_load(bs, l2_offset, &l2_table);
443 if (ret < 0) {
444 return ret;
447 /* find the cluster offset for the given disk offset */
449 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
450 *cluster_offset = be64_to_cpu(l2_table[l2_index]);
451 nb_clusters = size_to_clusters(s, nb_needed << 9);
453 if (!*cluster_offset) {
454 /* how many empty clusters ? */
455 c = count_contiguous_free_clusters(nb_clusters, &l2_table[l2_index]);
456 } else {
457 /* how many allocated clusters ? */
458 c = count_contiguous_clusters(nb_clusters, s->cluster_size,
459 &l2_table[l2_index], 0, QCOW_OFLAG_COPIED);
462 qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
464 nb_available = (c * s->cluster_sectors);
465 out:
466 if (nb_available > nb_needed)
467 nb_available = nb_needed;
469 *num = nb_available - index_in_cluster;
471 *cluster_offset &=~QCOW_OFLAG_COPIED;
472 return 0;
476 * get_cluster_table
478 * for a given disk offset, load (and allocate if needed)
479 * the l2 table.
481 * the l2 table offset in the qcow2 file and the cluster index
482 * in the l2 table are given to the caller.
484 * Returns 0 on success, -errno in failure case
486 static int get_cluster_table(BlockDriverState *bs, uint64_t offset,
487 uint64_t **new_l2_table,
488 uint64_t *new_l2_offset,
489 int *new_l2_index)
491 BDRVQcowState *s = bs->opaque;
492 unsigned int l1_index, l2_index;
493 uint64_t l2_offset;
494 uint64_t *l2_table = NULL;
495 int ret;
497 /* seek the the l2 offset in the l1 table */
499 l1_index = offset >> (s->l2_bits + s->cluster_bits);
500 if (l1_index >= s->l1_size) {
501 ret = qcow2_grow_l1_table(bs, l1_index + 1, false);
502 if (ret < 0) {
503 return ret;
506 l2_offset = s->l1_table[l1_index];
508 /* seek the l2 table of the given l2 offset */
510 if (l2_offset & QCOW_OFLAG_COPIED) {
511 /* load the l2 table in memory */
512 l2_offset &= ~QCOW_OFLAG_COPIED;
513 ret = l2_load(bs, l2_offset, &l2_table);
514 if (ret < 0) {
515 return ret;
517 } else {
518 /* First allocate a new L2 table (and do COW if needed) */
519 ret = l2_allocate(bs, l1_index, &l2_table);
520 if (ret < 0) {
521 return ret;
524 /* Then decrease the refcount of the old table */
525 if (l2_offset) {
526 qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t));
528 l2_offset = s->l1_table[l1_index] & ~QCOW_OFLAG_COPIED;
531 /* find the cluster offset for the given disk offset */
533 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
535 *new_l2_table = l2_table;
536 *new_l2_offset = l2_offset;
537 *new_l2_index = l2_index;
539 return 0;
543 * alloc_compressed_cluster_offset
545 * For a given offset of the disk image, return cluster offset in
546 * qcow2 file.
548 * If the offset is not found, allocate a new compressed cluster.
550 * Return the cluster offset if successful,
551 * Return 0, otherwise.
555 uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs,
556 uint64_t offset,
557 int compressed_size)
559 BDRVQcowState *s = bs->opaque;
560 int l2_index, ret;
561 uint64_t l2_offset, *l2_table;
562 int64_t cluster_offset;
563 int nb_csectors;
565 ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
566 if (ret < 0) {
567 return 0;
570 cluster_offset = be64_to_cpu(l2_table[l2_index]);
571 if (cluster_offset & QCOW_OFLAG_COPIED)
572 return cluster_offset & ~QCOW_OFLAG_COPIED;
574 if (cluster_offset)
575 qcow2_free_any_clusters(bs, cluster_offset, 1);
577 cluster_offset = qcow2_alloc_bytes(bs, compressed_size);
578 if (cluster_offset < 0) {
579 qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
580 return 0;
583 nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) -
584 (cluster_offset >> 9);
586 cluster_offset |= QCOW_OFLAG_COMPRESSED |
587 ((uint64_t)nb_csectors << s->csize_shift);
589 /* update L2 table */
591 /* compressed clusters never have the copied flag */
593 BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED);
594 qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
595 l2_table[l2_index] = cpu_to_be64(cluster_offset);
596 ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
597 if (ret < 0) {
598 return 0;
601 return cluster_offset;
604 int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m)
606 BDRVQcowState *s = bs->opaque;
607 int i, j = 0, l2_index, ret;
608 uint64_t *old_cluster, start_sect, l2_offset, *l2_table;
609 uint64_t cluster_offset = m->cluster_offset;
610 bool cow = false;
612 if (m->nb_clusters == 0)
613 return 0;
615 old_cluster = qemu_malloc(m->nb_clusters * sizeof(uint64_t));
617 /* copy content of unmodified sectors */
618 start_sect = (m->offset & ~(s->cluster_size - 1)) >> 9;
619 if (m->n_start) {
620 cow = true;
621 ret = copy_sectors(bs, start_sect, cluster_offset, 0, m->n_start);
622 if (ret < 0)
623 goto err;
626 if (m->nb_available & (s->cluster_sectors - 1)) {
627 uint64_t end = m->nb_available & ~(uint64_t)(s->cluster_sectors - 1);
628 cow = true;
629 ret = copy_sectors(bs, start_sect + end, cluster_offset + (end << 9),
630 m->nb_available - end, s->cluster_sectors);
631 if (ret < 0)
632 goto err;
636 * Update L2 table.
638 * Before we update the L2 table to actually point to the new cluster, we
639 * need to be sure that the refcounts have been increased and COW was
640 * handled.
642 if (cow) {
643 qcow2_cache_depends_on_flush(s->l2_table_cache);
646 qcow2_cache_set_dependency(bs, s->l2_table_cache, s->refcount_block_cache);
647 ret = get_cluster_table(bs, m->offset, &l2_table, &l2_offset, &l2_index);
648 if (ret < 0) {
649 goto err;
651 qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
653 for (i = 0; i < m->nb_clusters; i++) {
654 /* if two concurrent writes happen to the same unallocated cluster
655 * each write allocates separate cluster and writes data concurrently.
656 * The first one to complete updates l2 table with pointer to its
657 * cluster the second one has to do RMW (which is done above by
658 * copy_sectors()), update l2 table with its cluster pointer and free
659 * old cluster. This is what this loop does */
660 if(l2_table[l2_index + i] != 0)
661 old_cluster[j++] = l2_table[l2_index + i];
663 l2_table[l2_index + i] = cpu_to_be64((cluster_offset +
664 (i << s->cluster_bits)) | QCOW_OFLAG_COPIED);
668 ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
669 if (ret < 0) {
670 goto err;
674 * If this was a COW, we need to decrease the refcount of the old cluster.
675 * Also flush bs->file to get the right order for L2 and refcount update.
677 if (j != 0) {
678 for (i = 0; i < j; i++) {
679 qcow2_free_any_clusters(bs,
680 be64_to_cpu(old_cluster[i]) & ~QCOW_OFLAG_COPIED, 1);
684 ret = 0;
685 err:
686 qemu_free(old_cluster);
687 return ret;
691 * alloc_cluster_offset
693 * For a given offset of the disk image, return cluster offset in qcow2 file.
694 * If the offset is not found, allocate a new cluster.
696 * If the cluster was already allocated, m->nb_clusters is set to 0,
697 * m->depends_on is set to NULL and the other fields in m are meaningless.
699 * If the cluster is newly allocated, m->nb_clusters is set to the number of
700 * contiguous clusters that have been allocated. This may be 0 if the request
701 * conflict with another write request in flight; in this case, m->depends_on
702 * is set and the remaining fields of m are meaningless.
704 * If m->nb_clusters is non-zero, the other fields of m are valid and contain
705 * information about the first allocated cluster.
707 * Return 0 on success and -errno in error cases
709 int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset,
710 int n_start, int n_end, int *num, QCowL2Meta *m)
712 BDRVQcowState *s = bs->opaque;
713 int l2_index, ret;
714 uint64_t l2_offset, *l2_table;
715 int64_t cluster_offset;
716 unsigned int nb_clusters, i = 0;
717 QCowL2Meta *old_alloc;
719 ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
720 if (ret < 0) {
721 return ret;
724 nb_clusters = size_to_clusters(s, n_end << 9);
726 nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
728 cluster_offset = be64_to_cpu(l2_table[l2_index]);
730 /* We keep all QCOW_OFLAG_COPIED clusters */
732 if (cluster_offset & QCOW_OFLAG_COPIED) {
733 nb_clusters = count_contiguous_clusters(nb_clusters, s->cluster_size,
734 &l2_table[l2_index], 0, 0);
736 cluster_offset &= ~QCOW_OFLAG_COPIED;
737 m->nb_clusters = 0;
738 m->depends_on = NULL;
740 goto out;
743 /* for the moment, multiple compressed clusters are not managed */
745 if (cluster_offset & QCOW_OFLAG_COMPRESSED)
746 nb_clusters = 1;
748 /* how many available clusters ? */
750 while (i < nb_clusters) {
751 i += count_contiguous_clusters(nb_clusters - i, s->cluster_size,
752 &l2_table[l2_index], i, 0);
753 if ((i >= nb_clusters) || be64_to_cpu(l2_table[l2_index + i])) {
754 break;
757 i += count_contiguous_free_clusters(nb_clusters - i,
758 &l2_table[l2_index + i]);
759 if (i >= nb_clusters) {
760 break;
763 cluster_offset = be64_to_cpu(l2_table[l2_index + i]);
765 if ((cluster_offset & QCOW_OFLAG_COPIED) ||
766 (cluster_offset & QCOW_OFLAG_COMPRESSED))
767 break;
769 assert(i <= nb_clusters);
770 nb_clusters = i;
773 * Check if there already is an AIO write request in flight which allocates
774 * the same cluster. In this case we need to wait until the previous
775 * request has completed and updated the L2 table accordingly.
777 QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) {
779 uint64_t end_offset = offset + nb_clusters * s->cluster_size;
780 uint64_t old_offset = old_alloc->offset;
781 uint64_t old_end_offset = old_alloc->offset +
782 old_alloc->nb_clusters * s->cluster_size;
784 if (end_offset < old_offset || offset > old_end_offset) {
785 /* No intersection */
786 } else {
787 if (offset < old_offset) {
788 /* Stop at the start of a running allocation */
789 nb_clusters = (old_offset - offset) >> s->cluster_bits;
790 } else {
791 nb_clusters = 0;
794 if (nb_clusters == 0) {
795 /* Set dependency and wait for a callback */
796 m->depends_on = old_alloc;
797 m->nb_clusters = 0;
798 *num = 0;
800 goto out_wait_dependency;
805 if (!nb_clusters) {
806 abort();
809 QLIST_INSERT_HEAD(&s->cluster_allocs, m, next_in_flight);
811 /* allocate a new cluster */
813 cluster_offset = qcow2_alloc_clusters(bs, nb_clusters * s->cluster_size);
814 if (cluster_offset < 0) {
815 ret = cluster_offset;
816 goto fail;
819 /* save info needed for meta data update */
820 m->offset = offset;
821 m->n_start = n_start;
822 m->nb_clusters = nb_clusters;
824 out:
825 ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
826 if (ret < 0) {
827 goto fail_put;
830 m->nb_available = MIN(nb_clusters << (s->cluster_bits - 9), n_end);
831 m->cluster_offset = cluster_offset;
833 *num = m->nb_available - n_start;
835 return 0;
837 out_wait_dependency:
838 return qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
840 fail:
841 qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
842 fail_put:
843 QLIST_REMOVE(m, next_in_flight);
844 return ret;
847 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
848 const uint8_t *buf, int buf_size)
850 z_stream strm1, *strm = &strm1;
851 int ret, out_len;
853 memset(strm, 0, sizeof(*strm));
855 strm->next_in = (uint8_t *)buf;
856 strm->avail_in = buf_size;
857 strm->next_out = out_buf;
858 strm->avail_out = out_buf_size;
860 ret = inflateInit2(strm, -12);
861 if (ret != Z_OK)
862 return -1;
863 ret = inflate(strm, Z_FINISH);
864 out_len = strm->next_out - out_buf;
865 if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
866 out_len != out_buf_size) {
867 inflateEnd(strm);
868 return -1;
870 inflateEnd(strm);
871 return 0;
874 int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
876 BDRVQcowState *s = bs->opaque;
877 int ret, csize, nb_csectors, sector_offset;
878 uint64_t coffset;
880 coffset = cluster_offset & s->cluster_offset_mask;
881 if (s->cluster_cache_offset != coffset) {
882 nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
883 sector_offset = coffset & 511;
884 csize = nb_csectors * 512 - sector_offset;
885 BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
886 ret = bdrv_read(bs->file, coffset >> 9, s->cluster_data, nb_csectors);
887 if (ret < 0) {
888 return ret;
890 if (decompress_buffer(s->cluster_cache, s->cluster_size,
891 s->cluster_data + sector_offset, csize) < 0) {
892 return -EIO;
894 s->cluster_cache_offset = coffset;
896 return 0;
900 * This discards as many clusters of nb_clusters as possible at once (i.e.
901 * all clusters in the same L2 table) and returns the number of discarded
902 * clusters.
904 static int discard_single_l2(BlockDriverState *bs, uint64_t offset,
905 unsigned int nb_clusters)
907 BDRVQcowState *s = bs->opaque;
908 uint64_t l2_offset, *l2_table;
909 int l2_index;
910 int ret;
911 int i;
913 ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
914 if (ret < 0) {
915 return ret;
918 /* Limit nb_clusters to one L2 table */
919 nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
921 for (i = 0; i < nb_clusters; i++) {
922 uint64_t old_offset;
924 old_offset = be64_to_cpu(l2_table[l2_index + i]);
925 old_offset &= ~QCOW_OFLAG_COPIED;
927 if (old_offset == 0) {
928 continue;
931 /* First remove L2 entries */
932 qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
933 l2_table[l2_index + i] = cpu_to_be64(0);
935 /* Then decrease the refcount */
936 qcow2_free_any_clusters(bs, old_offset, 1);
939 ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
940 if (ret < 0) {
941 return ret;
944 return nb_clusters;
947 int qcow2_discard_clusters(BlockDriverState *bs, uint64_t offset,
948 int nb_sectors)
950 BDRVQcowState *s = bs->opaque;
951 uint64_t end_offset;
952 unsigned int nb_clusters;
953 int ret;
955 end_offset = offset + (nb_sectors << BDRV_SECTOR_BITS);
957 /* Round start up and end down */
958 offset = align_offset(offset, s->cluster_size);
959 end_offset &= ~(s->cluster_size - 1);
961 if (offset > end_offset) {
962 return 0;
965 nb_clusters = size_to_clusters(s, end_offset - offset);
967 /* Each L2 table is handled by its own loop iteration */
968 while (nb_clusters > 0) {
969 ret = discard_single_l2(bs, offset, nb_clusters);
970 if (ret < 0) {
971 return ret;
974 nb_clusters -= ret;
975 offset += (ret * s->cluster_size);
978 return 0;