kvm: handle internal error
[qemu.git] / block / qcow2-cluster.c
blob639e05e98979c4e4f1158308c5b6bbe487ece478
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)
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 new_l1_size = s->l1_size;
40 if (min_size <= new_l1_size)
41 return 0;
42 if (new_l1_size == 0) {
43 new_l1_size = 1;
45 while (min_size > new_l1_size) {
46 new_l1_size = (new_l1_size * 3 + 1) / 2;
48 #ifdef DEBUG_ALLOC2
49 printf("grow l1_table from %d to %d\n", s->l1_size, new_l1_size);
50 #endif
52 new_l1_size2 = sizeof(uint64_t) * new_l1_size;
53 new_l1_table = qemu_mallocz(align_offset(new_l1_size2, 512));
54 memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t));
56 /* write new table (align to cluster) */
57 BLKDBG_EVENT(s->hd, BLKDBG_L1_GROW_ALLOC_TABLE);
58 new_l1_table_offset = qcow2_alloc_clusters(bs, new_l1_size2);
59 if (new_l1_table_offset < 0) {
60 qemu_free(new_l1_table);
61 return new_l1_table_offset;
64 BLKDBG_EVENT(s->hd, BLKDBG_L1_GROW_WRITE_TABLE);
65 for(i = 0; i < s->l1_size; i++)
66 new_l1_table[i] = cpu_to_be64(new_l1_table[i]);
67 ret = bdrv_pwrite(s->hd, new_l1_table_offset, new_l1_table, new_l1_size2);
68 if (ret != new_l1_size2)
69 goto fail;
70 for(i = 0; i < s->l1_size; i++)
71 new_l1_table[i] = be64_to_cpu(new_l1_table[i]);
73 /* set new table */
74 BLKDBG_EVENT(s->hd, BLKDBG_L1_GROW_ACTIVATE_TABLE);
75 cpu_to_be32w((uint32_t*)data, new_l1_size);
76 cpu_to_be64w((uint64_t*)(data + 4), new_l1_table_offset);
77 ret = bdrv_pwrite(s->hd, offsetof(QCowHeader, l1_size), data,sizeof(data));
78 if (ret != sizeof(data)) {
79 goto fail;
81 qemu_free(s->l1_table);
82 qcow2_free_clusters(bs, s->l1_table_offset, s->l1_size * sizeof(uint64_t));
83 s->l1_table_offset = new_l1_table_offset;
84 s->l1_table = new_l1_table;
85 s->l1_size = new_l1_size;
86 return 0;
87 fail:
88 qemu_free(new_l1_table);
89 qcow2_free_clusters(bs, new_l1_table_offset, new_l1_size2);
90 return ret < 0 ? ret : -EIO;
93 void qcow2_l2_cache_reset(BlockDriverState *bs)
95 BDRVQcowState *s = bs->opaque;
97 memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
98 memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
99 memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
102 static inline int l2_cache_new_entry(BlockDriverState *bs)
104 BDRVQcowState *s = bs->opaque;
105 uint32_t min_count;
106 int min_index, i;
108 /* find a new entry in the least used one */
109 min_index = 0;
110 min_count = 0xffffffff;
111 for(i = 0; i < L2_CACHE_SIZE; i++) {
112 if (s->l2_cache_counts[i] < min_count) {
113 min_count = s->l2_cache_counts[i];
114 min_index = i;
117 return min_index;
121 * seek_l2_table
123 * seek l2_offset in the l2_cache table
124 * if not found, return NULL,
125 * if found,
126 * increments the l2 cache hit count of the entry,
127 * if counter overflow, divide by two all counters
128 * return the pointer to the l2 cache entry
132 static uint64_t *seek_l2_table(BDRVQcowState *s, uint64_t l2_offset)
134 int i, j;
136 for(i = 0; i < L2_CACHE_SIZE; i++) {
137 if (l2_offset == s->l2_cache_offsets[i]) {
138 /* increment the hit count */
139 if (++s->l2_cache_counts[i] == 0xffffffff) {
140 for(j = 0; j < L2_CACHE_SIZE; j++) {
141 s->l2_cache_counts[j] >>= 1;
144 return s->l2_cache + (i << s->l2_bits);
147 return NULL;
151 * l2_load
153 * Loads a L2 table into memory. If the table is in the cache, the cache
154 * is used; otherwise the L2 table is loaded from the image file.
156 * Returns a pointer to the L2 table on success, or NULL if the read from
157 * the image file failed.
160 static uint64_t *l2_load(BlockDriverState *bs, uint64_t l2_offset)
162 BDRVQcowState *s = bs->opaque;
163 int min_index;
164 uint64_t *l2_table;
166 /* seek if the table for the given offset is in the cache */
168 l2_table = seek_l2_table(s, l2_offset);
169 if (l2_table != NULL)
170 return l2_table;
172 /* not found: load a new entry in the least used one */
174 min_index = l2_cache_new_entry(bs);
175 l2_table = s->l2_cache + (min_index << s->l2_bits);
177 BLKDBG_EVENT(s->hd, BLKDBG_L2_LOAD);
178 if (bdrv_pread(s->hd, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
179 s->l2_size * sizeof(uint64_t))
180 return NULL;
181 s->l2_cache_offsets[min_index] = l2_offset;
182 s->l2_cache_counts[min_index] = 1;
184 return l2_table;
188 * Writes one sector of the L1 table to the disk (can't update single entries
189 * and we really don't want bdrv_pread to perform a read-modify-write)
191 #define L1_ENTRIES_PER_SECTOR (512 / 8)
192 static int write_l1_entry(BDRVQcowState *s, int l1_index)
194 uint64_t buf[L1_ENTRIES_PER_SECTOR];
195 int l1_start_index;
196 int i, ret;
198 l1_start_index = l1_index & ~(L1_ENTRIES_PER_SECTOR - 1);
199 for (i = 0; i < L1_ENTRIES_PER_SECTOR; i++) {
200 buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]);
203 BLKDBG_EVENT(s->hd, BLKDBG_L1_UPDATE);
204 ret = bdrv_pwrite(s->hd, s->l1_table_offset + 8 * l1_start_index,
205 buf, sizeof(buf));
206 if (ret < 0) {
207 return ret;
210 return 0;
214 * l2_allocate
216 * Allocate a new l2 entry in the file. If l1_index points to an already
217 * used entry in the L2 table (i.e. we are doing a copy on write for the L2
218 * table) copy the contents of the old L2 table into the newly allocated one.
219 * Otherwise the new table is initialized with zeros.
223 static int l2_allocate(BlockDriverState *bs, int l1_index, uint64_t **table)
225 BDRVQcowState *s = bs->opaque;
226 int min_index;
227 uint64_t old_l2_offset;
228 uint64_t *l2_table;
229 int64_t l2_offset;
230 int ret;
232 old_l2_offset = s->l1_table[l1_index];
234 /* allocate a new l2 entry */
236 l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
237 if (l2_offset < 0) {
238 return l2_offset;
241 /* update the L1 entry */
243 s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
244 ret = write_l1_entry(s, l1_index);
245 if (ret < 0) {
246 return ret;
249 /* allocate a new entry in the l2 cache */
251 min_index = l2_cache_new_entry(bs);
252 l2_table = s->l2_cache + (min_index << s->l2_bits);
254 if (old_l2_offset == 0) {
255 /* if there was no old l2 table, clear the new table */
256 memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
257 } else {
258 /* if there was an old l2 table, read it from the disk */
259 BLKDBG_EVENT(s->hd, BLKDBG_L2_ALLOC_COW_READ);
260 ret = bdrv_pread(s->hd, old_l2_offset, l2_table,
261 s->l2_size * sizeof(uint64_t));
262 if (ret < 0) {
263 return ret;
266 /* write the l2 table to the file */
267 BLKDBG_EVENT(s->hd, BLKDBG_L2_ALLOC_WRITE);
268 ret = bdrv_pwrite(s->hd, l2_offset, l2_table,
269 s->l2_size * sizeof(uint64_t));
270 if (ret < 0) {
271 return ret;
274 /* update the l2 cache entry */
276 s->l2_cache_offsets[min_index] = l2_offset;
277 s->l2_cache_counts[min_index] = 1;
279 *table = l2_table;
280 return 0;
283 static int count_contiguous_clusters(uint64_t nb_clusters, int cluster_size,
284 uint64_t *l2_table, uint64_t start, uint64_t mask)
286 int i;
287 uint64_t offset = be64_to_cpu(l2_table[0]) & ~mask;
289 if (!offset)
290 return 0;
292 for (i = start; i < start + nb_clusters; i++)
293 if (offset + (uint64_t) i * cluster_size != (be64_to_cpu(l2_table[i]) & ~mask))
294 break;
296 return (i - start);
299 static int count_contiguous_free_clusters(uint64_t nb_clusters, uint64_t *l2_table)
301 int i = 0;
303 while(nb_clusters-- && l2_table[i] == 0)
304 i++;
306 return i;
309 /* The crypt function is compatible with the linux cryptoloop
310 algorithm for < 4 GB images. NOTE: out_buf == in_buf is
311 supported */
312 void qcow2_encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
313 uint8_t *out_buf, const uint8_t *in_buf,
314 int nb_sectors, int enc,
315 const AES_KEY *key)
317 union {
318 uint64_t ll[2];
319 uint8_t b[16];
320 } ivec;
321 int i;
323 for(i = 0; i < nb_sectors; i++) {
324 ivec.ll[0] = cpu_to_le64(sector_num);
325 ivec.ll[1] = 0;
326 AES_cbc_encrypt(in_buf, out_buf, 512, key,
327 ivec.b, enc);
328 sector_num++;
329 in_buf += 512;
330 out_buf += 512;
335 static int qcow_read(BlockDriverState *bs, int64_t sector_num,
336 uint8_t *buf, int nb_sectors)
338 BDRVQcowState *s = bs->opaque;
339 int ret, index_in_cluster, n, n1;
340 uint64_t cluster_offset;
342 while (nb_sectors > 0) {
343 n = nb_sectors;
344 cluster_offset = qcow2_get_cluster_offset(bs, sector_num << 9, &n);
345 index_in_cluster = sector_num & (s->cluster_sectors - 1);
346 if (!cluster_offset) {
347 if (bs->backing_hd) {
348 /* read from the base image */
349 n1 = qcow2_backing_read1(bs->backing_hd, sector_num, buf, n);
350 if (n1 > 0) {
351 BLKDBG_EVENT(s->hd, BLKDBG_READ_BACKING);
352 ret = bdrv_read(bs->backing_hd, sector_num, buf, n1);
353 if (ret < 0)
354 return -1;
356 } else {
357 memset(buf, 0, 512 * n);
359 } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
360 if (qcow2_decompress_cluster(s, cluster_offset) < 0)
361 return -1;
362 memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n);
363 } else {
364 BLKDBG_EVENT(s->hd, BLKDBG_READ);
365 ret = bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
366 if (ret != n * 512)
367 return -1;
368 if (s->crypt_method) {
369 qcow2_encrypt_sectors(s, sector_num, buf, buf, n, 0,
370 &s->aes_decrypt_key);
373 nb_sectors -= n;
374 sector_num += n;
375 buf += n * 512;
377 return 0;
380 static int copy_sectors(BlockDriverState *bs, uint64_t start_sect,
381 uint64_t cluster_offset, int n_start, int n_end)
383 BDRVQcowState *s = bs->opaque;
384 int n, ret;
386 n = n_end - n_start;
387 if (n <= 0)
388 return 0;
389 BLKDBG_EVENT(s->hd, BLKDBG_COW_READ);
390 ret = qcow_read(bs, start_sect + n_start, s->cluster_data, n);
391 if (ret < 0)
392 return ret;
393 if (s->crypt_method) {
394 qcow2_encrypt_sectors(s, start_sect + n_start,
395 s->cluster_data,
396 s->cluster_data, n, 1,
397 &s->aes_encrypt_key);
399 BLKDBG_EVENT(s->hd, BLKDBG_COW_WRITE);
400 ret = bdrv_write(s->hd, (cluster_offset >> 9) + n_start,
401 s->cluster_data, n);
402 if (ret < 0)
403 return ret;
404 return 0;
409 * get_cluster_offset
411 * For a given offset of the disk image, return cluster offset in
412 * qcow2 file.
414 * on entry, *num is the number of contiguous clusters we'd like to
415 * access following offset.
417 * on exit, *num is the number of contiguous clusters we can read.
419 * Return 1, if the offset is found
420 * Return 0, otherwise.
424 uint64_t qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
425 int *num)
427 BDRVQcowState *s = bs->opaque;
428 unsigned int l1_index, l2_index;
429 uint64_t l2_offset, *l2_table, cluster_offset;
430 int l1_bits, c;
431 unsigned int index_in_cluster, nb_clusters;
432 uint64_t nb_available, nb_needed;
434 index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1);
435 nb_needed = *num + index_in_cluster;
437 l1_bits = s->l2_bits + s->cluster_bits;
439 /* compute how many bytes there are between the offset and
440 * the end of the l1 entry
443 nb_available = (1ULL << l1_bits) - (offset & ((1ULL << l1_bits) - 1));
445 /* compute the number of available sectors */
447 nb_available = (nb_available >> 9) + index_in_cluster;
449 if (nb_needed > nb_available) {
450 nb_needed = nb_available;
453 cluster_offset = 0;
455 /* seek the the l2 offset in the l1 table */
457 l1_index = offset >> l1_bits;
458 if (l1_index >= s->l1_size)
459 goto out;
461 l2_offset = s->l1_table[l1_index];
463 /* seek the l2 table of the given l2 offset */
465 if (!l2_offset)
466 goto out;
468 /* load the l2 table in memory */
470 l2_offset &= ~QCOW_OFLAG_COPIED;
471 l2_table = l2_load(bs, l2_offset);
472 if (l2_table == NULL)
473 return 0;
475 /* find the cluster offset for the given disk offset */
477 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
478 cluster_offset = be64_to_cpu(l2_table[l2_index]);
479 nb_clusters = size_to_clusters(s, nb_needed << 9);
481 if (!cluster_offset) {
482 /* how many empty clusters ? */
483 c = count_contiguous_free_clusters(nb_clusters, &l2_table[l2_index]);
484 } else {
485 /* how many allocated clusters ? */
486 c = count_contiguous_clusters(nb_clusters, s->cluster_size,
487 &l2_table[l2_index], 0, QCOW_OFLAG_COPIED);
490 nb_available = (c * s->cluster_sectors);
491 out:
492 if (nb_available > nb_needed)
493 nb_available = nb_needed;
495 *num = nb_available - index_in_cluster;
497 return cluster_offset & ~QCOW_OFLAG_COPIED;
501 * get_cluster_table
503 * for a given disk offset, load (and allocate if needed)
504 * the l2 table.
506 * the l2 table offset in the qcow2 file and the cluster index
507 * in the l2 table are given to the caller.
509 * Returns 0 on success, -errno in failure case
511 static int get_cluster_table(BlockDriverState *bs, uint64_t offset,
512 uint64_t **new_l2_table,
513 uint64_t *new_l2_offset,
514 int *new_l2_index)
516 BDRVQcowState *s = bs->opaque;
517 unsigned int l1_index, l2_index;
518 uint64_t l2_offset;
519 uint64_t *l2_table = NULL;
520 int ret;
522 /* seek the the l2 offset in the l1 table */
524 l1_index = offset >> (s->l2_bits + s->cluster_bits);
525 if (l1_index >= s->l1_size) {
526 ret = qcow2_grow_l1_table(bs, l1_index + 1);
527 if (ret < 0) {
528 return ret;
531 l2_offset = s->l1_table[l1_index];
533 /* seek the l2 table of the given l2 offset */
535 if (l2_offset & QCOW_OFLAG_COPIED) {
536 /* load the l2 table in memory */
537 l2_offset &= ~QCOW_OFLAG_COPIED;
538 l2_table = l2_load(bs, l2_offset);
539 if (l2_table == NULL) {
540 return -EIO;
542 } else {
543 if (l2_offset)
544 qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t));
545 ret = l2_allocate(bs, l1_index, &l2_table);
546 if (ret < 0) {
547 return ret;
549 l2_offset = s->l1_table[l1_index] & ~QCOW_OFLAG_COPIED;
552 /* find the cluster offset for the given disk offset */
554 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
556 *new_l2_table = l2_table;
557 *new_l2_offset = l2_offset;
558 *new_l2_index = l2_index;
560 return 0;
564 * alloc_compressed_cluster_offset
566 * For a given offset of the disk image, return cluster offset in
567 * qcow2 file.
569 * If the offset is not found, allocate a new compressed cluster.
571 * Return the cluster offset if successful,
572 * Return 0, otherwise.
576 uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs,
577 uint64_t offset,
578 int compressed_size)
580 BDRVQcowState *s = bs->opaque;
581 int l2_index, ret;
582 uint64_t l2_offset, *l2_table;
583 int64_t cluster_offset;
584 int nb_csectors;
586 ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
587 if (ret < 0) {
588 return 0;
591 cluster_offset = be64_to_cpu(l2_table[l2_index]);
592 if (cluster_offset & QCOW_OFLAG_COPIED)
593 return cluster_offset & ~QCOW_OFLAG_COPIED;
595 if (cluster_offset)
596 qcow2_free_any_clusters(bs, cluster_offset, 1);
598 cluster_offset = qcow2_alloc_bytes(bs, compressed_size);
599 if (cluster_offset < 0) {
600 return 0;
603 nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) -
604 (cluster_offset >> 9);
606 cluster_offset |= QCOW_OFLAG_COMPRESSED |
607 ((uint64_t)nb_csectors << s->csize_shift);
609 /* update L2 table */
611 /* compressed clusters never have the copied flag */
613 BLKDBG_EVENT(s->hd, BLKDBG_L2_UPDATE_COMPRESSED);
614 l2_table[l2_index] = cpu_to_be64(cluster_offset);
615 if (bdrv_pwrite(s->hd,
616 l2_offset + l2_index * sizeof(uint64_t),
617 l2_table + l2_index,
618 sizeof(uint64_t)) != sizeof(uint64_t))
619 return 0;
621 return cluster_offset;
625 * Write L2 table updates to disk, writing whole sectors to avoid a
626 * read-modify-write in bdrv_pwrite
628 #define L2_ENTRIES_PER_SECTOR (512 / 8)
629 static int write_l2_entries(BDRVQcowState *s, uint64_t *l2_table,
630 uint64_t l2_offset, int l2_index, int num)
632 int l2_start_index = l2_index & ~(L1_ENTRIES_PER_SECTOR - 1);
633 int start_offset = (8 * l2_index) & ~511;
634 int end_offset = (8 * (l2_index + num) + 511) & ~511;
635 size_t len = end_offset - start_offset;
636 int ret;
638 BLKDBG_EVENT(s->hd, BLKDBG_L2_UPDATE);
639 ret = bdrv_pwrite(s->hd, l2_offset + start_offset,
640 &l2_table[l2_start_index], len);
641 if (ret < 0) {
642 return ret;
645 return 0;
648 int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m)
650 BDRVQcowState *s = bs->opaque;
651 int i, j = 0, l2_index, ret;
652 uint64_t *old_cluster, start_sect, l2_offset, *l2_table;
653 uint64_t cluster_offset = m->cluster_offset;
655 if (m->nb_clusters == 0)
656 return 0;
658 old_cluster = qemu_malloc(m->nb_clusters * sizeof(uint64_t));
660 /* copy content of unmodified sectors */
661 start_sect = (m->offset & ~(s->cluster_size - 1)) >> 9;
662 if (m->n_start) {
663 ret = copy_sectors(bs, start_sect, cluster_offset, 0, m->n_start);
664 if (ret < 0)
665 goto err;
668 if (m->nb_available & (s->cluster_sectors - 1)) {
669 uint64_t end = m->nb_available & ~(uint64_t)(s->cluster_sectors - 1);
670 ret = copy_sectors(bs, start_sect + end, cluster_offset + (end << 9),
671 m->nb_available - end, s->cluster_sectors);
672 if (ret < 0)
673 goto err;
676 /* update L2 table */
677 ret = get_cluster_table(bs, m->offset, &l2_table, &l2_offset, &l2_index);
678 if (ret < 0) {
679 goto err;
682 for (i = 0; i < m->nb_clusters; i++) {
683 /* if two concurrent writes happen to the same unallocated cluster
684 * each write allocates separate cluster and writes data concurrently.
685 * The first one to complete updates l2 table with pointer to its
686 * cluster the second one has to do RMW (which is done above by
687 * copy_sectors()), update l2 table with its cluster pointer and free
688 * old cluster. This is what this loop does */
689 if(l2_table[l2_index + i] != 0)
690 old_cluster[j++] = l2_table[l2_index + i];
692 l2_table[l2_index + i] = cpu_to_be64((cluster_offset +
693 (i << s->cluster_bits)) | QCOW_OFLAG_COPIED);
696 ret = write_l2_entries(s, l2_table, l2_offset, l2_index, m->nb_clusters);
697 if (ret < 0) {
698 goto err;
701 for (i = 0; i < j; i++)
702 qcow2_free_any_clusters(bs,
703 be64_to_cpu(old_cluster[i]) & ~QCOW_OFLAG_COPIED, 1);
705 ret = 0;
706 err:
707 qemu_free(old_cluster);
708 return ret;
712 * alloc_cluster_offset
714 * For a given offset of the disk image, return cluster offset in qcow2 file.
715 * If the offset is not found, allocate a new cluster.
717 * If the cluster was already allocated, m->nb_clusters is set to 0,
718 * m->depends_on is set to NULL and the other fields in m are meaningless.
720 * If the cluster is newly allocated, m->nb_clusters is set to the number of
721 * contiguous clusters that have been allocated. This may be 0 if the request
722 * conflict with another write request in flight; in this case, m->depends_on
723 * is set and the remaining fields of m are meaningless.
725 * If m->nb_clusters is non-zero, the other fields of m are valid and contain
726 * information about the first allocated cluster.
728 * Return 0 on success and -errno in error cases
730 int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset,
731 int n_start, int n_end, int *num, QCowL2Meta *m)
733 BDRVQcowState *s = bs->opaque;
734 int l2_index, ret;
735 uint64_t l2_offset, *l2_table;
736 int64_t cluster_offset;
737 unsigned int nb_clusters, i = 0;
738 QCowL2Meta *old_alloc;
740 ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
741 if (ret < 0) {
742 return ret;
745 nb_clusters = size_to_clusters(s, n_end << 9);
747 nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
749 cluster_offset = be64_to_cpu(l2_table[l2_index]);
751 /* We keep all QCOW_OFLAG_COPIED clusters */
753 if (cluster_offset & QCOW_OFLAG_COPIED) {
754 nb_clusters = count_contiguous_clusters(nb_clusters, s->cluster_size,
755 &l2_table[l2_index], 0, 0);
757 cluster_offset &= ~QCOW_OFLAG_COPIED;
758 m->nb_clusters = 0;
759 m->depends_on = NULL;
761 goto out;
764 /* for the moment, multiple compressed clusters are not managed */
766 if (cluster_offset & QCOW_OFLAG_COMPRESSED)
767 nb_clusters = 1;
769 /* how many available clusters ? */
771 while (i < nb_clusters) {
772 i += count_contiguous_clusters(nb_clusters - i, s->cluster_size,
773 &l2_table[l2_index], i, 0);
774 if ((i >= nb_clusters) || be64_to_cpu(l2_table[l2_index + i])) {
775 break;
778 i += count_contiguous_free_clusters(nb_clusters - i,
779 &l2_table[l2_index + i]);
780 if (i >= nb_clusters) {
781 break;
784 cluster_offset = be64_to_cpu(l2_table[l2_index + i]);
786 if ((cluster_offset & QCOW_OFLAG_COPIED) ||
787 (cluster_offset & QCOW_OFLAG_COMPRESSED))
788 break;
790 assert(i <= nb_clusters);
791 nb_clusters = i;
794 * Check if there already is an AIO write request in flight which allocates
795 * the same cluster. In this case we need to wait until the previous
796 * request has completed and updated the L2 table accordingly.
798 QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) {
800 uint64_t end_offset = offset + nb_clusters * s->cluster_size;
801 uint64_t old_offset = old_alloc->offset;
802 uint64_t old_end_offset = old_alloc->offset +
803 old_alloc->nb_clusters * s->cluster_size;
805 if (end_offset < old_offset || offset > old_end_offset) {
806 /* No intersection */
807 } else {
808 if (offset < old_offset) {
809 /* Stop at the start of a running allocation */
810 nb_clusters = (old_offset - offset) >> s->cluster_bits;
811 } else {
812 nb_clusters = 0;
815 if (nb_clusters == 0) {
816 /* Set dependency and wait for a callback */
817 m->depends_on = old_alloc;
818 m->nb_clusters = 0;
819 *num = 0;
820 return 0;
825 if (!nb_clusters) {
826 abort();
829 QLIST_INSERT_HEAD(&s->cluster_allocs, m, next_in_flight);
831 /* allocate a new cluster */
833 cluster_offset = qcow2_alloc_clusters(bs, nb_clusters * s->cluster_size);
834 if (cluster_offset < 0) {
835 QLIST_REMOVE(m, next_in_flight);
836 return cluster_offset;
839 /* save info needed for meta data update */
840 m->offset = offset;
841 m->n_start = n_start;
842 m->nb_clusters = nb_clusters;
844 out:
845 m->nb_available = MIN(nb_clusters << (s->cluster_bits - 9), n_end);
846 m->cluster_offset = cluster_offset;
848 *num = m->nb_available - n_start;
850 return 0;
853 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
854 const uint8_t *buf, int buf_size)
856 z_stream strm1, *strm = &strm1;
857 int ret, out_len;
859 memset(strm, 0, sizeof(*strm));
861 strm->next_in = (uint8_t *)buf;
862 strm->avail_in = buf_size;
863 strm->next_out = out_buf;
864 strm->avail_out = out_buf_size;
866 ret = inflateInit2(strm, -12);
867 if (ret != Z_OK)
868 return -1;
869 ret = inflate(strm, Z_FINISH);
870 out_len = strm->next_out - out_buf;
871 if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
872 out_len != out_buf_size) {
873 inflateEnd(strm);
874 return -1;
876 inflateEnd(strm);
877 return 0;
880 int qcow2_decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset)
882 int ret, csize, nb_csectors, sector_offset;
883 uint64_t coffset;
885 coffset = cluster_offset & s->cluster_offset_mask;
886 if (s->cluster_cache_offset != coffset) {
887 nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
888 sector_offset = coffset & 511;
889 csize = nb_csectors * 512 - sector_offset;
890 BLKDBG_EVENT(s->hd, BLKDBG_READ_COMPRESSED);
891 ret = bdrv_read(s->hd, coffset >> 9, s->cluster_data, nb_csectors);
892 if (ret < 0) {
893 return -1;
895 if (decompress_buffer(s->cluster_cache, s->cluster_size,
896 s->cluster_data + sector_offset, csize) < 0) {
897 return -1;
899 s->cluster_cache_offset = coffset;
901 return 0;