root / block / qcow2-cluster.c @ 60651f90
History | View | Annotate | Download (34.5 kB)
1 |
/*
|
---|---|
2 |
* Block driver for the QCOW version 2 format
|
3 |
*
|
4 |
* Copyright (c) 2004-2006 Fabrice Bellard
|
5 |
*
|
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:
|
12 |
*
|
13 |
* The above copyright notice and this permission notice shall be included in
|
14 |
* all copies or substantial portions of the Software.
|
15 |
*
|
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.
|
23 |
*/
|
24 |
|
25 |
#include <zlib.h> |
26 |
|
27 |
#include "qemu-common.h" |
28 |
#include "block_int.h" |
29 |
#include "block/qcow2.h" |
30 |
#include "trace.h" |
31 |
|
32 |
int qcow2_grow_l1_table(BlockDriverState *bs, int min_size, bool exact_size) |
33 |
{ |
34 |
BDRVQcowState *s = bs->opaque; |
35 |
int new_l1_size, new_l1_size2, ret, i;
|
36 |
uint64_t *new_l1_table; |
37 |
int64_t new_l1_table_offset; |
38 |
uint8_t data[12];
|
39 |
|
40 |
if (min_size <= s->l1_size)
|
41 |
return 0; |
42 |
|
43 |
if (exact_size) {
|
44 |
new_l1_size = min_size; |
45 |
} else {
|
46 |
/* Bump size up to reduce the number of times we have to grow */
|
47 |
new_l1_size = s->l1_size; |
48 |
if (new_l1_size == 0) { |
49 |
new_l1_size = 1;
|
50 |
} |
51 |
while (min_size > new_l1_size) {
|
52 |
new_l1_size = (new_l1_size * 3 + 1) / 2; |
53 |
} |
54 |
} |
55 |
|
56 |
#ifdef DEBUG_ALLOC2
|
57 |
fprintf(stderr, "grow l1_table from %d to %d\n", s->l1_size, new_l1_size);
|
58 |
#endif
|
59 |
|
60 |
new_l1_size2 = sizeof(uint64_t) * new_l1_size;
|
61 |
new_l1_table = g_malloc0(align_offset(new_l1_size2, 512));
|
62 |
memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t));
|
63 |
|
64 |
/* write new table (align to cluster) */
|
65 |
BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ALLOC_TABLE); |
66 |
new_l1_table_offset = qcow2_alloc_clusters(bs, new_l1_size2); |
67 |
if (new_l1_table_offset < 0) { |
68 |
g_free(new_l1_table); |
69 |
return new_l1_table_offset;
|
70 |
} |
71 |
|
72 |
ret = qcow2_cache_flush(bs, s->refcount_block_cache); |
73 |
if (ret < 0) { |
74 |
goto fail;
|
75 |
} |
76 |
|
77 |
BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_WRITE_TABLE); |
78 |
for(i = 0; i < s->l1_size; i++) |
79 |
new_l1_table[i] = cpu_to_be64(new_l1_table[i]); |
80 |
ret = bdrv_pwrite_sync(bs->file, new_l1_table_offset, new_l1_table, new_l1_size2); |
81 |
if (ret < 0) |
82 |
goto fail;
|
83 |
for(i = 0; i < s->l1_size; i++) |
84 |
new_l1_table[i] = be64_to_cpu(new_l1_table[i]); |
85 |
|
86 |
/* set new table */
|
87 |
BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ACTIVATE_TABLE); |
88 |
cpu_to_be32w((uint32_t*)data, new_l1_size); |
89 |
cpu_to_be64wu((uint64_t*)(data + 4), new_l1_table_offset);
|
90 |
ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_size), data,sizeof(data));
|
91 |
if (ret < 0) { |
92 |
goto fail;
|
93 |
} |
94 |
g_free(s->l1_table); |
95 |
qcow2_free_clusters(bs, s->l1_table_offset, s->l1_size * sizeof(uint64_t));
|
96 |
s->l1_table_offset = new_l1_table_offset; |
97 |
s->l1_table = new_l1_table; |
98 |
s->l1_size = new_l1_size; |
99 |
return 0; |
100 |
fail:
|
101 |
g_free(new_l1_table); |
102 |
qcow2_free_clusters(bs, new_l1_table_offset, new_l1_size2); |
103 |
return ret;
|
104 |
} |
105 |
|
106 |
/*
|
107 |
* l2_load
|
108 |
*
|
109 |
* Loads a L2 table into memory. If the table is in the cache, the cache
|
110 |
* is used; otherwise the L2 table is loaded from the image file.
|
111 |
*
|
112 |
* Returns a pointer to the L2 table on success, or NULL if the read from
|
113 |
* the image file failed.
|
114 |
*/
|
115 |
|
116 |
static int l2_load(BlockDriverState *bs, uint64_t l2_offset, |
117 |
uint64_t **l2_table) |
118 |
{ |
119 |
BDRVQcowState *s = bs->opaque; |
120 |
int ret;
|
121 |
|
122 |
ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset, (void**) l2_table);
|
123 |
|
124 |
return ret;
|
125 |
} |
126 |
|
127 |
/*
|
128 |
* Writes one sector of the L1 table to the disk (can't update single entries
|
129 |
* and we really don't want bdrv_pread to perform a read-modify-write)
|
130 |
*/
|
131 |
#define L1_ENTRIES_PER_SECTOR (512 / 8) |
132 |
static int write_l1_entry(BlockDriverState *bs, int l1_index) |
133 |
{ |
134 |
BDRVQcowState *s = bs->opaque; |
135 |
uint64_t buf[L1_ENTRIES_PER_SECTOR]; |
136 |
int l1_start_index;
|
137 |
int i, ret;
|
138 |
|
139 |
l1_start_index = l1_index & ~(L1_ENTRIES_PER_SECTOR - 1);
|
140 |
for (i = 0; i < L1_ENTRIES_PER_SECTOR; i++) { |
141 |
buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]); |
142 |
} |
143 |
|
144 |
BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); |
145 |
ret = bdrv_pwrite_sync(bs->file, s->l1_table_offset + 8 * l1_start_index,
|
146 |
buf, sizeof(buf));
|
147 |
if (ret < 0) { |
148 |
return ret;
|
149 |
} |
150 |
|
151 |
return 0; |
152 |
} |
153 |
|
154 |
/*
|
155 |
* l2_allocate
|
156 |
*
|
157 |
* Allocate a new l2 entry in the file. If l1_index points to an already
|
158 |
* used entry in the L2 table (i.e. we are doing a copy on write for the L2
|
159 |
* table) copy the contents of the old L2 table into the newly allocated one.
|
160 |
* Otherwise the new table is initialized with zeros.
|
161 |
*
|
162 |
*/
|
163 |
|
164 |
static int l2_allocate(BlockDriverState *bs, int l1_index, uint64_t **table) |
165 |
{ |
166 |
BDRVQcowState *s = bs->opaque; |
167 |
uint64_t old_l2_offset; |
168 |
uint64_t *l2_table; |
169 |
int64_t l2_offset; |
170 |
int ret;
|
171 |
|
172 |
old_l2_offset = s->l1_table[l1_index]; |
173 |
|
174 |
trace_qcow2_l2_allocate(bs, l1_index); |
175 |
|
176 |
/* allocate a new l2 entry */
|
177 |
|
178 |
l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
|
179 |
if (l2_offset < 0) { |
180 |
return l2_offset;
|
181 |
} |
182 |
|
183 |
ret = qcow2_cache_flush(bs, s->refcount_block_cache); |
184 |
if (ret < 0) { |
185 |
goto fail;
|
186 |
} |
187 |
|
188 |
/* allocate a new entry in the l2 cache */
|
189 |
|
190 |
trace_qcow2_l2_allocate_get_empty(bs, l1_index); |
191 |
ret = qcow2_cache_get_empty(bs, s->l2_table_cache, l2_offset, (void**) table);
|
192 |
if (ret < 0) { |
193 |
return ret;
|
194 |
} |
195 |
|
196 |
l2_table = *table; |
197 |
|
198 |
if ((old_l2_offset & L1E_OFFSET_MASK) == 0) { |
199 |
/* if there was no old l2 table, clear the new table */
|
200 |
memset(l2_table, 0, s->l2_size * sizeof(uint64_t)); |
201 |
} else {
|
202 |
uint64_t* old_table; |
203 |
|
204 |
/* if there was an old l2 table, read it from the disk */
|
205 |
BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_COW_READ); |
206 |
ret = qcow2_cache_get(bs, s->l2_table_cache, |
207 |
old_l2_offset & L1E_OFFSET_MASK, |
208 |
(void**) &old_table);
|
209 |
if (ret < 0) { |
210 |
goto fail;
|
211 |
} |
212 |
|
213 |
memcpy(l2_table, old_table, s->cluster_size); |
214 |
|
215 |
ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &old_table);
|
216 |
if (ret < 0) { |
217 |
goto fail;
|
218 |
} |
219 |
} |
220 |
|
221 |
/* write the l2 table to the file */
|
222 |
BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_WRITE); |
223 |
|
224 |
trace_qcow2_l2_allocate_write_l2(bs, l1_index); |
225 |
qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); |
226 |
ret = qcow2_cache_flush(bs, s->l2_table_cache); |
227 |
if (ret < 0) { |
228 |
goto fail;
|
229 |
} |
230 |
|
231 |
/* update the L1 entry */
|
232 |
trace_qcow2_l2_allocate_write_l1(bs, l1_index); |
233 |
s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED; |
234 |
ret = write_l1_entry(bs, l1_index); |
235 |
if (ret < 0) { |
236 |
goto fail;
|
237 |
} |
238 |
|
239 |
*table = l2_table; |
240 |
trace_qcow2_l2_allocate_done(bs, l1_index, 0);
|
241 |
return 0; |
242 |
|
243 |
fail:
|
244 |
trace_qcow2_l2_allocate_done(bs, l1_index, ret); |
245 |
qcow2_cache_put(bs, s->l2_table_cache, (void**) table);
|
246 |
s->l1_table[l1_index] = old_l2_offset; |
247 |
return ret;
|
248 |
} |
249 |
|
250 |
/*
|
251 |
* Checks how many clusters in a given L2 table are contiguous in the image
|
252 |
* file. As soon as one of the flags in the bitmask stop_flags changes compared
|
253 |
* to the first cluster, the search is stopped and the cluster is not counted
|
254 |
* as contiguous. (This allows it, for example, to stop at the first compressed
|
255 |
* cluster which may require a different handling)
|
256 |
*/
|
257 |
static int count_contiguous_clusters(uint64_t nb_clusters, int cluster_size, |
258 |
uint64_t *l2_table, uint64_t start, uint64_t stop_flags) |
259 |
{ |
260 |
int i;
|
261 |
uint64_t mask = stop_flags | L2E_OFFSET_MASK; |
262 |
uint64_t offset = be64_to_cpu(l2_table[0]) & mask;
|
263 |
|
264 |
if (!offset)
|
265 |
return 0; |
266 |
|
267 |
for (i = start; i < start + nb_clusters; i++) {
|
268 |
uint64_t l2_entry = be64_to_cpu(l2_table[i]) & mask; |
269 |
if (offset + (uint64_t) i * cluster_size != l2_entry) {
|
270 |
break;
|
271 |
} |
272 |
} |
273 |
|
274 |
return (i - start);
|
275 |
} |
276 |
|
277 |
static int count_contiguous_free_clusters(uint64_t nb_clusters, uint64_t *l2_table) |
278 |
{ |
279 |
int i;
|
280 |
|
281 |
for (i = 0; i < nb_clusters; i++) { |
282 |
int type = qcow2_get_cluster_type(be64_to_cpu(l2_table[i]));
|
283 |
|
284 |
if (type != QCOW2_CLUSTER_UNALLOCATED) {
|
285 |
break;
|
286 |
} |
287 |
} |
288 |
|
289 |
return i;
|
290 |
} |
291 |
|
292 |
/* The crypt function is compatible with the linux cryptoloop
|
293 |
algorithm for < 4 GB images. NOTE: out_buf == in_buf is
|
294 |
supported */
|
295 |
void qcow2_encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
|
296 |
uint8_t *out_buf, const uint8_t *in_buf,
|
297 |
int nb_sectors, int enc, |
298 |
const AES_KEY *key)
|
299 |
{ |
300 |
union {
|
301 |
uint64_t ll[2];
|
302 |
uint8_t b[16];
|
303 |
} ivec; |
304 |
int i;
|
305 |
|
306 |
for(i = 0; i < nb_sectors; i++) { |
307 |
ivec.ll[0] = cpu_to_le64(sector_num);
|
308 |
ivec.ll[1] = 0; |
309 |
AES_cbc_encrypt(in_buf, out_buf, 512, key,
|
310 |
ivec.b, enc); |
311 |
sector_num++; |
312 |
in_buf += 512;
|
313 |
out_buf += 512;
|
314 |
} |
315 |
} |
316 |
|
317 |
static int coroutine_fn copy_sectors(BlockDriverState *bs, |
318 |
uint64_t start_sect, |
319 |
uint64_t cluster_offset, |
320 |
int n_start, int n_end) |
321 |
{ |
322 |
BDRVQcowState *s = bs->opaque; |
323 |
QEMUIOVector qiov; |
324 |
struct iovec iov;
|
325 |
int n, ret;
|
326 |
|
327 |
/*
|
328 |
* If this is the last cluster and it is only partially used, we must only
|
329 |
* copy until the end of the image, or bdrv_check_request will fail for the
|
330 |
* bdrv_read/write calls below.
|
331 |
*/
|
332 |
if (start_sect + n_end > bs->total_sectors) {
|
333 |
n_end = bs->total_sectors - start_sect; |
334 |
} |
335 |
|
336 |
n = n_end - n_start; |
337 |
if (n <= 0) { |
338 |
return 0; |
339 |
} |
340 |
|
341 |
iov.iov_len = n * BDRV_SECTOR_SIZE; |
342 |
iov.iov_base = qemu_blockalign(bs, iov.iov_len); |
343 |
|
344 |
qemu_iovec_init_external(&qiov, &iov, 1);
|
345 |
|
346 |
BLKDBG_EVENT(bs->file, BLKDBG_COW_READ); |
347 |
|
348 |
/* Call .bdrv_co_readv() directly instead of using the public block-layer
|
349 |
* interface. This avoids double I/O throttling and request tracking,
|
350 |
* which can lead to deadlock when block layer copy-on-read is enabled.
|
351 |
*/
|
352 |
ret = bs->drv->bdrv_co_readv(bs, start_sect + n_start, n, &qiov); |
353 |
if (ret < 0) { |
354 |
goto out;
|
355 |
} |
356 |
|
357 |
if (s->crypt_method) {
|
358 |
qcow2_encrypt_sectors(s, start_sect + n_start, |
359 |
iov.iov_base, iov.iov_base, n, 1,
|
360 |
&s->aes_encrypt_key); |
361 |
} |
362 |
|
363 |
BLKDBG_EVENT(bs->file, BLKDBG_COW_WRITE); |
364 |
ret = bdrv_co_writev(bs->file, (cluster_offset >> 9) + n_start, n, &qiov);
|
365 |
if (ret < 0) { |
366 |
goto out;
|
367 |
} |
368 |
|
369 |
ret = 0;
|
370 |
out:
|
371 |
qemu_vfree(iov.iov_base); |
372 |
return ret;
|
373 |
} |
374 |
|
375 |
|
376 |
/*
|
377 |
* get_cluster_offset
|
378 |
*
|
379 |
* For a given offset of the disk image, find the cluster offset in
|
380 |
* qcow2 file. The offset is stored in *cluster_offset.
|
381 |
*
|
382 |
* on entry, *num is the number of contiguous sectors we'd like to
|
383 |
* access following offset.
|
384 |
*
|
385 |
* on exit, *num is the number of contiguous sectors we can read.
|
386 |
*
|
387 |
* Returns the cluster type (QCOW2_CLUSTER_*) on success, -errno in error
|
388 |
* cases.
|
389 |
*/
|
390 |
int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
|
391 |
int *num, uint64_t *cluster_offset)
|
392 |
{ |
393 |
BDRVQcowState *s = bs->opaque; |
394 |
unsigned int l1_index, l2_index; |
395 |
uint64_t l2_offset, *l2_table; |
396 |
int l1_bits, c;
|
397 |
unsigned int index_in_cluster, nb_clusters; |
398 |
uint64_t nb_available, nb_needed; |
399 |
int ret;
|
400 |
|
401 |
index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1); |
402 |
nb_needed = *num + index_in_cluster; |
403 |
|
404 |
l1_bits = s->l2_bits + s->cluster_bits; |
405 |
|
406 |
/* compute how many bytes there are between the offset and
|
407 |
* the end of the l1 entry
|
408 |
*/
|
409 |
|
410 |
nb_available = (1ULL << l1_bits) - (offset & ((1ULL << l1_bits) - 1)); |
411 |
|
412 |
/* compute the number of available sectors */
|
413 |
|
414 |
nb_available = (nb_available >> 9) + index_in_cluster;
|
415 |
|
416 |
if (nb_needed > nb_available) {
|
417 |
nb_needed = nb_available; |
418 |
} |
419 |
|
420 |
*cluster_offset = 0;
|
421 |
|
422 |
/* seek the the l2 offset in the l1 table */
|
423 |
|
424 |
l1_index = offset >> l1_bits; |
425 |
if (l1_index >= s->l1_size) {
|
426 |
ret = QCOW2_CLUSTER_UNALLOCATED; |
427 |
goto out;
|
428 |
} |
429 |
|
430 |
l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; |
431 |
if (!l2_offset) {
|
432 |
ret = QCOW2_CLUSTER_UNALLOCATED; |
433 |
goto out;
|
434 |
} |
435 |
|
436 |
/* load the l2 table in memory */
|
437 |
|
438 |
ret = l2_load(bs, l2_offset, &l2_table); |
439 |
if (ret < 0) { |
440 |
return ret;
|
441 |
} |
442 |
|
443 |
/* find the cluster offset for the given disk offset */
|
444 |
|
445 |
l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
|
446 |
*cluster_offset = be64_to_cpu(l2_table[l2_index]); |
447 |
nb_clusters = size_to_clusters(s, nb_needed << 9);
|
448 |
|
449 |
ret = qcow2_get_cluster_type(*cluster_offset); |
450 |
switch (ret) {
|
451 |
case QCOW2_CLUSTER_COMPRESSED:
|
452 |
/* Compressed clusters can only be processed one by one */
|
453 |
c = 1;
|
454 |
*cluster_offset &= L2E_COMPRESSED_OFFSET_SIZE_MASK; |
455 |
break;
|
456 |
case QCOW2_CLUSTER_ZERO:
|
457 |
c = count_contiguous_clusters(nb_clusters, s->cluster_size, |
458 |
&l2_table[l2_index], 0,
|
459 |
QCOW_OFLAG_COMPRESSED | QCOW_OFLAG_ZERO); |
460 |
*cluster_offset = 0;
|
461 |
break;
|
462 |
case QCOW2_CLUSTER_UNALLOCATED:
|
463 |
/* how many empty clusters ? */
|
464 |
c = count_contiguous_free_clusters(nb_clusters, &l2_table[l2_index]); |
465 |
*cluster_offset = 0;
|
466 |
break;
|
467 |
case QCOW2_CLUSTER_NORMAL:
|
468 |
/* how many allocated clusters ? */
|
469 |
c = count_contiguous_clusters(nb_clusters, s->cluster_size, |
470 |
&l2_table[l2_index], 0,
|
471 |
QCOW_OFLAG_COMPRESSED | QCOW_OFLAG_ZERO); |
472 |
*cluster_offset &= L2E_OFFSET_MASK; |
473 |
break;
|
474 |
} |
475 |
|
476 |
qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
|
477 |
|
478 |
nb_available = (c * s->cluster_sectors); |
479 |
|
480 |
out:
|
481 |
if (nb_available > nb_needed)
|
482 |
nb_available = nb_needed; |
483 |
|
484 |
*num = nb_available - index_in_cluster; |
485 |
|
486 |
return ret;
|
487 |
} |
488 |
|
489 |
/*
|
490 |
* get_cluster_table
|
491 |
*
|
492 |
* for a given disk offset, load (and allocate if needed)
|
493 |
* the l2 table.
|
494 |
*
|
495 |
* the l2 table offset in the qcow2 file and the cluster index
|
496 |
* in the l2 table are given to the caller.
|
497 |
*
|
498 |
* Returns 0 on success, -errno in failure case
|
499 |
*/
|
500 |
static int get_cluster_table(BlockDriverState *bs, uint64_t offset, |
501 |
uint64_t **new_l2_table, |
502 |
int *new_l2_index)
|
503 |
{ |
504 |
BDRVQcowState *s = bs->opaque; |
505 |
unsigned int l1_index, l2_index; |
506 |
uint64_t l2_offset; |
507 |
uint64_t *l2_table = NULL;
|
508 |
int ret;
|
509 |
|
510 |
/* seek the the l2 offset in the l1 table */
|
511 |
|
512 |
l1_index = offset >> (s->l2_bits + s->cluster_bits); |
513 |
if (l1_index >= s->l1_size) {
|
514 |
ret = qcow2_grow_l1_table(bs, l1_index + 1, false); |
515 |
if (ret < 0) { |
516 |
return ret;
|
517 |
} |
518 |
} |
519 |
|
520 |
l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; |
521 |
|
522 |
/* seek the l2 table of the given l2 offset */
|
523 |
|
524 |
if (s->l1_table[l1_index] & QCOW_OFLAG_COPIED) {
|
525 |
/* load the l2 table in memory */
|
526 |
ret = l2_load(bs, l2_offset, &l2_table); |
527 |
if (ret < 0) { |
528 |
return ret;
|
529 |
} |
530 |
} else {
|
531 |
/* First allocate a new L2 table (and do COW if needed) */
|
532 |
ret = l2_allocate(bs, l1_index, &l2_table); |
533 |
if (ret < 0) { |
534 |
return ret;
|
535 |
} |
536 |
|
537 |
/* Then decrease the refcount of the old table */
|
538 |
if (l2_offset) {
|
539 |
qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t));
|
540 |
} |
541 |
l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; |
542 |
} |
543 |
|
544 |
/* find the cluster offset for the given disk offset */
|
545 |
|
546 |
l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
|
547 |
|
548 |
*new_l2_table = l2_table; |
549 |
*new_l2_index = l2_index; |
550 |
|
551 |
return 0; |
552 |
} |
553 |
|
554 |
/*
|
555 |
* alloc_compressed_cluster_offset
|
556 |
*
|
557 |
* For a given offset of the disk image, return cluster offset in
|
558 |
* qcow2 file.
|
559 |
*
|
560 |
* If the offset is not found, allocate a new compressed cluster.
|
561 |
*
|
562 |
* Return the cluster offset if successful,
|
563 |
* Return 0, otherwise.
|
564 |
*
|
565 |
*/
|
566 |
|
567 |
uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs, |
568 |
uint64_t offset, |
569 |
int compressed_size)
|
570 |
{ |
571 |
BDRVQcowState *s = bs->opaque; |
572 |
int l2_index, ret;
|
573 |
uint64_t *l2_table; |
574 |
int64_t cluster_offset; |
575 |
int nb_csectors;
|
576 |
|
577 |
ret = get_cluster_table(bs, offset, &l2_table, &l2_index); |
578 |
if (ret < 0) { |
579 |
return 0; |
580 |
} |
581 |
|
582 |
/* Compression can't overwrite anything. Fail if the cluster was already
|
583 |
* allocated. */
|
584 |
cluster_offset = be64_to_cpu(l2_table[l2_index]); |
585 |
if (cluster_offset & L2E_OFFSET_MASK) {
|
586 |
qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
|
587 |
return 0; |
588 |
} |
589 |
|
590 |
cluster_offset = qcow2_alloc_bytes(bs, compressed_size); |
591 |
if (cluster_offset < 0) { |
592 |
qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
|
593 |
return 0; |
594 |
} |
595 |
|
596 |
nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) - |
597 |
(cluster_offset >> 9);
|
598 |
|
599 |
cluster_offset |= QCOW_OFLAG_COMPRESSED | |
600 |
((uint64_t)nb_csectors << s->csize_shift); |
601 |
|
602 |
/* update L2 table */
|
603 |
|
604 |
/* compressed clusters never have the copied flag */
|
605 |
|
606 |
BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED); |
607 |
qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); |
608 |
l2_table[l2_index] = cpu_to_be64(cluster_offset); |
609 |
ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
|
610 |
if (ret < 0) { |
611 |
return 0; |
612 |
} |
613 |
|
614 |
return cluster_offset;
|
615 |
} |
616 |
|
617 |
int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m)
|
618 |
{ |
619 |
BDRVQcowState *s = bs->opaque; |
620 |
int i, j = 0, l2_index, ret; |
621 |
uint64_t *old_cluster, start_sect, *l2_table; |
622 |
uint64_t cluster_offset = m->alloc_offset; |
623 |
bool cow = false; |
624 |
|
625 |
trace_qcow2_cluster_link_l2(qemu_coroutine_self(), m->nb_clusters); |
626 |
|
627 |
if (m->nb_clusters == 0) |
628 |
return 0; |
629 |
|
630 |
old_cluster = g_malloc(m->nb_clusters * sizeof(uint64_t));
|
631 |
|
632 |
/* copy content of unmodified sectors */
|
633 |
start_sect = (m->offset & ~(s->cluster_size - 1)) >> 9; |
634 |
if (m->n_start) {
|
635 |
cow = true;
|
636 |
qemu_co_mutex_unlock(&s->lock); |
637 |
ret = copy_sectors(bs, start_sect, cluster_offset, 0, m->n_start);
|
638 |
qemu_co_mutex_lock(&s->lock); |
639 |
if (ret < 0) |
640 |
goto err;
|
641 |
} |
642 |
|
643 |
if (m->nb_available & (s->cluster_sectors - 1)) { |
644 |
uint64_t end = m->nb_available & ~(uint64_t)(s->cluster_sectors - 1);
|
645 |
cow = true;
|
646 |
qemu_co_mutex_unlock(&s->lock); |
647 |
ret = copy_sectors(bs, start_sect + end, cluster_offset + (end << 9),
|
648 |
m->nb_available - end, s->cluster_sectors); |
649 |
qemu_co_mutex_lock(&s->lock); |
650 |
if (ret < 0) |
651 |
goto err;
|
652 |
} |
653 |
|
654 |
/*
|
655 |
* Update L2 table.
|
656 |
*
|
657 |
* Before we update the L2 table to actually point to the new cluster, we
|
658 |
* need to be sure that the refcounts have been increased and COW was
|
659 |
* handled.
|
660 |
*/
|
661 |
if (cow) {
|
662 |
qcow2_cache_depends_on_flush(s->l2_table_cache); |
663 |
} |
664 |
|
665 |
qcow2_cache_set_dependency(bs, s->l2_table_cache, s->refcount_block_cache); |
666 |
ret = get_cluster_table(bs, m->offset, &l2_table, &l2_index); |
667 |
if (ret < 0) { |
668 |
goto err;
|
669 |
} |
670 |
qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); |
671 |
|
672 |
for (i = 0; i < m->nb_clusters; i++) { |
673 |
/* if two concurrent writes happen to the same unallocated cluster
|
674 |
* each write allocates separate cluster and writes data concurrently.
|
675 |
* The first one to complete updates l2 table with pointer to its
|
676 |
* cluster the second one has to do RMW (which is done above by
|
677 |
* copy_sectors()), update l2 table with its cluster pointer and free
|
678 |
* old cluster. This is what this loop does */
|
679 |
if(l2_table[l2_index + i] != 0) |
680 |
old_cluster[j++] = l2_table[l2_index + i]; |
681 |
|
682 |
l2_table[l2_index + i] = cpu_to_be64((cluster_offset + |
683 |
(i << s->cluster_bits)) | QCOW_OFLAG_COPIED); |
684 |
} |
685 |
|
686 |
|
687 |
ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
|
688 |
if (ret < 0) { |
689 |
goto err;
|
690 |
} |
691 |
|
692 |
/*
|
693 |
* If this was a COW, we need to decrease the refcount of the old cluster.
|
694 |
* Also flush bs->file to get the right order for L2 and refcount update.
|
695 |
*/
|
696 |
if (j != 0) { |
697 |
for (i = 0; i < j; i++) { |
698 |
qcow2_free_any_clusters(bs, be64_to_cpu(old_cluster[i]), 1);
|
699 |
} |
700 |
} |
701 |
|
702 |
ret = 0;
|
703 |
err:
|
704 |
g_free(old_cluster); |
705 |
return ret;
|
706 |
} |
707 |
|
708 |
/*
|
709 |
* Returns the number of contiguous clusters that can be used for an allocating
|
710 |
* write, but require COW to be performed (this includes yet unallocated space,
|
711 |
* which must copy from the backing file)
|
712 |
*/
|
713 |
static int count_cow_clusters(BDRVQcowState *s, int nb_clusters, |
714 |
uint64_t *l2_table, int l2_index)
|
715 |
{ |
716 |
int i;
|
717 |
|
718 |
for (i = 0; i < nb_clusters; i++) { |
719 |
uint64_t l2_entry = be64_to_cpu(l2_table[l2_index + i]); |
720 |
int cluster_type = qcow2_get_cluster_type(l2_entry);
|
721 |
|
722 |
switch(cluster_type) {
|
723 |
case QCOW2_CLUSTER_NORMAL:
|
724 |
if (l2_entry & QCOW_OFLAG_COPIED) {
|
725 |
goto out;
|
726 |
} |
727 |
break;
|
728 |
case QCOW2_CLUSTER_UNALLOCATED:
|
729 |
case QCOW2_CLUSTER_COMPRESSED:
|
730 |
case QCOW2_CLUSTER_ZERO:
|
731 |
break;
|
732 |
default:
|
733 |
abort(); |
734 |
} |
735 |
} |
736 |
|
737 |
out:
|
738 |
assert(i <= nb_clusters); |
739 |
return i;
|
740 |
} |
741 |
|
742 |
/*
|
743 |
* Allocates new clusters for the given guest_offset.
|
744 |
*
|
745 |
* At most *nb_clusters are allocated, and on return *nb_clusters is updated to
|
746 |
* contain the number of clusters that have been allocated and are contiguous
|
747 |
* in the image file.
|
748 |
*
|
749 |
* If *host_offset is non-zero, it specifies the offset in the image file at
|
750 |
* which the new clusters must start. *nb_clusters can be 0 on return in this
|
751 |
* case if the cluster at host_offset is already in use. If *host_offset is
|
752 |
* zero, the clusters can be allocated anywhere in the image file.
|
753 |
*
|
754 |
* *host_offset is updated to contain the offset into the image file at which
|
755 |
* the first allocated cluster starts.
|
756 |
*
|
757 |
* Return 0 on success and -errno in error cases. -EAGAIN means that the
|
758 |
* function has been waiting for another request and the allocation must be
|
759 |
* restarted, but the whole request should not be failed.
|
760 |
*/
|
761 |
static int do_alloc_cluster_offset(BlockDriverState *bs, uint64_t guest_offset, |
762 |
uint64_t *host_offset, unsigned int *nb_clusters) |
763 |
{ |
764 |
BDRVQcowState *s = bs->opaque; |
765 |
int64_t cluster_offset; |
766 |
QCowL2Meta *old_alloc; |
767 |
|
768 |
trace_qcow2_do_alloc_clusters_offset(qemu_coroutine_self(), guest_offset, |
769 |
*host_offset, *nb_clusters); |
770 |
|
771 |
/*
|
772 |
* Check if there already is an AIO write request in flight which allocates
|
773 |
* the same cluster. In this case we need to wait until the previous
|
774 |
* request has completed and updated the L2 table accordingly.
|
775 |
*/
|
776 |
QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) { |
777 |
|
778 |
uint64_t start = guest_offset >> s->cluster_bits; |
779 |
uint64_t end = start + *nb_clusters; |
780 |
uint64_t old_start = old_alloc->offset >> s->cluster_bits; |
781 |
uint64_t old_end = old_start + old_alloc->nb_clusters; |
782 |
|
783 |
if (end < old_start || start > old_end) {
|
784 |
/* No intersection */
|
785 |
} else {
|
786 |
if (start < old_start) {
|
787 |
/* Stop at the start of a running allocation */
|
788 |
*nb_clusters = old_start - start; |
789 |
} else {
|
790 |
*nb_clusters = 0;
|
791 |
} |
792 |
|
793 |
if (*nb_clusters == 0) { |
794 |
/* Wait for the dependency to complete. We need to recheck
|
795 |
* the free/allocated clusters when we continue. */
|
796 |
qemu_co_mutex_unlock(&s->lock); |
797 |
qemu_co_queue_wait(&old_alloc->dependent_requests); |
798 |
qemu_co_mutex_lock(&s->lock); |
799 |
return -EAGAIN;
|
800 |
} |
801 |
} |
802 |
} |
803 |
|
804 |
if (!*nb_clusters) {
|
805 |
abort(); |
806 |
} |
807 |
|
808 |
/* Allocate new clusters */
|
809 |
trace_qcow2_cluster_alloc_phys(qemu_coroutine_self()); |
810 |
if (*host_offset == 0) { |
811 |
cluster_offset = qcow2_alloc_clusters(bs, *nb_clusters * s->cluster_size); |
812 |
} else {
|
813 |
cluster_offset = *host_offset; |
814 |
*nb_clusters = qcow2_alloc_clusters_at(bs, cluster_offset, *nb_clusters); |
815 |
} |
816 |
|
817 |
if (cluster_offset < 0) { |
818 |
return cluster_offset;
|
819 |
} |
820 |
*host_offset = cluster_offset; |
821 |
return 0; |
822 |
} |
823 |
|
824 |
/*
|
825 |
* alloc_cluster_offset
|
826 |
*
|
827 |
* For a given offset on the virtual disk, find the cluster offset in qcow2
|
828 |
* file. If the offset is not found, allocate a new cluster.
|
829 |
*
|
830 |
* If the cluster was already allocated, m->nb_clusters is set to 0 and
|
831 |
* other fields in m are meaningless.
|
832 |
*
|
833 |
* If the cluster is newly allocated, m->nb_clusters is set to the number of
|
834 |
* contiguous clusters that have been allocated. In this case, the other
|
835 |
* fields of m are valid and contain information about the first allocated
|
836 |
* cluster.
|
837 |
*
|
838 |
* If the request conflicts with another write request in flight, the coroutine
|
839 |
* is queued and will be reentered when the dependency has completed.
|
840 |
*
|
841 |
* Return 0 on success and -errno in error cases
|
842 |
*/
|
843 |
int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset,
|
844 |
int n_start, int n_end, int *num, QCowL2Meta *m) |
845 |
{ |
846 |
BDRVQcowState *s = bs->opaque; |
847 |
int l2_index, ret, sectors;
|
848 |
uint64_t *l2_table; |
849 |
unsigned int nb_clusters, keep_clusters; |
850 |
uint64_t cluster_offset; |
851 |
|
852 |
trace_qcow2_alloc_clusters_offset(qemu_coroutine_self(), offset, |
853 |
n_start, n_end); |
854 |
|
855 |
/* Find L2 entry for the first involved cluster */
|
856 |
ret = get_cluster_table(bs, offset, &l2_table, &l2_index); |
857 |
if (ret < 0) { |
858 |
return ret;
|
859 |
} |
860 |
|
861 |
/*
|
862 |
* Calculate the number of clusters to look for. We stop at L2 table
|
863 |
* boundaries to keep things simple.
|
864 |
*/
|
865 |
again:
|
866 |
nb_clusters = MIN(size_to_clusters(s, n_end << BDRV_SECTOR_BITS), |
867 |
s->l2_size - l2_index); |
868 |
|
869 |
cluster_offset = be64_to_cpu(l2_table[l2_index]); |
870 |
|
871 |
/*
|
872 |
* Check how many clusters are already allocated and don't need COW, and how
|
873 |
* many need a new allocation.
|
874 |
*/
|
875 |
if (qcow2_get_cluster_type(cluster_offset) == QCOW2_CLUSTER_NORMAL
|
876 |
&& (cluster_offset & QCOW_OFLAG_COPIED)) |
877 |
{ |
878 |
/* We keep all QCOW_OFLAG_COPIED clusters */
|
879 |
keep_clusters = |
880 |
count_contiguous_clusters(nb_clusters, s->cluster_size, |
881 |
&l2_table[l2_index], 0,
|
882 |
QCOW_OFLAG_COPIED | QCOW_OFLAG_ZERO); |
883 |
assert(keep_clusters <= nb_clusters); |
884 |
nb_clusters -= keep_clusters; |
885 |
} else {
|
886 |
/* For the moment, overwrite compressed clusters one by one */
|
887 |
if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
|
888 |
nb_clusters = 1;
|
889 |
} else {
|
890 |
nb_clusters = count_cow_clusters(s, nb_clusters, l2_table, l2_index); |
891 |
} |
892 |
|
893 |
keep_clusters = 0;
|
894 |
cluster_offset = 0;
|
895 |
} |
896 |
|
897 |
cluster_offset &= L2E_OFFSET_MASK; |
898 |
|
899 |
/* If there is something left to allocate, do that now */
|
900 |
*m = (QCowL2Meta) { |
901 |
.cluster_offset = cluster_offset, |
902 |
.nb_clusters = 0,
|
903 |
}; |
904 |
qemu_co_queue_init(&m->dependent_requests); |
905 |
|
906 |
if (nb_clusters > 0) { |
907 |
uint64_t alloc_offset; |
908 |
uint64_t alloc_cluster_offset; |
909 |
uint64_t keep_bytes = keep_clusters * s->cluster_size; |
910 |
|
911 |
/* Calculate start and size of allocation */
|
912 |
alloc_offset = offset + keep_bytes; |
913 |
|
914 |
if (keep_clusters == 0) { |
915 |
alloc_cluster_offset = 0;
|
916 |
} else {
|
917 |
alloc_cluster_offset = cluster_offset + keep_bytes; |
918 |
} |
919 |
|
920 |
/* Allocate, if necessary at a given offset in the image file */
|
921 |
ret = do_alloc_cluster_offset(bs, alloc_offset, &alloc_cluster_offset, |
922 |
&nb_clusters); |
923 |
if (ret == -EAGAIN) {
|
924 |
goto again;
|
925 |
} else if (ret < 0) { |
926 |
goto fail;
|
927 |
} |
928 |
|
929 |
/* save info needed for meta data update */
|
930 |
if (nb_clusters > 0) { |
931 |
int requested_sectors = n_end - keep_clusters * s->cluster_sectors;
|
932 |
int avail_sectors = (keep_clusters + nb_clusters)
|
933 |
<< (s->cluster_bits - BDRV_SECTOR_BITS); |
934 |
|
935 |
*m = (QCowL2Meta) { |
936 |
.cluster_offset = keep_clusters == 0 ?
|
937 |
alloc_cluster_offset : cluster_offset, |
938 |
.alloc_offset = alloc_cluster_offset, |
939 |
.offset = alloc_offset, |
940 |
.n_start = keep_clusters == 0 ? n_start : 0, |
941 |
.nb_clusters = nb_clusters, |
942 |
.nb_available = MIN(requested_sectors, avail_sectors), |
943 |
}; |
944 |
qemu_co_queue_init(&m->dependent_requests); |
945 |
QLIST_INSERT_HEAD(&s->cluster_allocs, m, next_in_flight); |
946 |
} |
947 |
} |
948 |
|
949 |
/* Some cleanup work */
|
950 |
ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
|
951 |
if (ret < 0) { |
952 |
goto fail_put;
|
953 |
} |
954 |
|
955 |
sectors = (keep_clusters + nb_clusters) << (s->cluster_bits - 9);
|
956 |
if (sectors > n_end) {
|
957 |
sectors = n_end; |
958 |
} |
959 |
|
960 |
assert(sectors > n_start); |
961 |
*num = sectors - n_start; |
962 |
|
963 |
return 0; |
964 |
|
965 |
fail:
|
966 |
qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
|
967 |
fail_put:
|
968 |
if (m->nb_clusters > 0) { |
969 |
QLIST_REMOVE(m, next_in_flight); |
970 |
} |
971 |
return ret;
|
972 |
} |
973 |
|
974 |
static int decompress_buffer(uint8_t *out_buf, int out_buf_size, |
975 |
const uint8_t *buf, int buf_size) |
976 |
{ |
977 |
z_stream strm1, *strm = &strm1; |
978 |
int ret, out_len;
|
979 |
|
980 |
memset(strm, 0, sizeof(*strm)); |
981 |
|
982 |
strm->next_in = (uint8_t *)buf; |
983 |
strm->avail_in = buf_size; |
984 |
strm->next_out = out_buf; |
985 |
strm->avail_out = out_buf_size; |
986 |
|
987 |
ret = inflateInit2(strm, -12);
|
988 |
if (ret != Z_OK)
|
989 |
return -1; |
990 |
ret = inflate(strm, Z_FINISH); |
991 |
out_len = strm->next_out - out_buf; |
992 |
if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
|
993 |
out_len != out_buf_size) { |
994 |
inflateEnd(strm); |
995 |
return -1; |
996 |
} |
997 |
inflateEnd(strm); |
998 |
return 0; |
999 |
} |
1000 |
|
1001 |
int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
|
1002 |
{ |
1003 |
BDRVQcowState *s = bs->opaque; |
1004 |
int ret, csize, nb_csectors, sector_offset;
|
1005 |
uint64_t coffset; |
1006 |
|
1007 |
coffset = cluster_offset & s->cluster_offset_mask; |
1008 |
if (s->cluster_cache_offset != coffset) {
|
1009 |
nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
|
1010 |
sector_offset = coffset & 511;
|
1011 |
csize = nb_csectors * 512 - sector_offset;
|
1012 |
BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED); |
1013 |
ret = bdrv_read(bs->file, coffset >> 9, s->cluster_data, nb_csectors);
|
1014 |
if (ret < 0) { |
1015 |
return ret;
|
1016 |
} |
1017 |
if (decompress_buffer(s->cluster_cache, s->cluster_size,
|
1018 |
s->cluster_data + sector_offset, csize) < 0) {
|
1019 |
return -EIO;
|
1020 |
} |
1021 |
s->cluster_cache_offset = coffset; |
1022 |
} |
1023 |
return 0; |
1024 |
} |
1025 |
|
1026 |
/*
|
1027 |
* This discards as many clusters of nb_clusters as possible at once (i.e.
|
1028 |
* all clusters in the same L2 table) and returns the number of discarded
|
1029 |
* clusters.
|
1030 |
*/
|
1031 |
static int discard_single_l2(BlockDriverState *bs, uint64_t offset, |
1032 |
unsigned int nb_clusters) |
1033 |
{ |
1034 |
BDRVQcowState *s = bs->opaque; |
1035 |
uint64_t *l2_table; |
1036 |
int l2_index;
|
1037 |
int ret;
|
1038 |
int i;
|
1039 |
|
1040 |
ret = get_cluster_table(bs, offset, &l2_table, &l2_index); |
1041 |
if (ret < 0) { |
1042 |
return ret;
|
1043 |
} |
1044 |
|
1045 |
/* Limit nb_clusters to one L2 table */
|
1046 |
nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); |
1047 |
|
1048 |
for (i = 0; i < nb_clusters; i++) { |
1049 |
uint64_t old_offset; |
1050 |
|
1051 |
old_offset = be64_to_cpu(l2_table[l2_index + i]); |
1052 |
if ((old_offset & L2E_OFFSET_MASK) == 0) { |
1053 |
continue;
|
1054 |
} |
1055 |
|
1056 |
/* First remove L2 entries */
|
1057 |
qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); |
1058 |
l2_table[l2_index + i] = cpu_to_be64(0);
|
1059 |
|
1060 |
/* Then decrease the refcount */
|
1061 |
qcow2_free_any_clusters(bs, old_offset, 1);
|
1062 |
} |
1063 |
|
1064 |
ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
|
1065 |
if (ret < 0) { |
1066 |
return ret;
|
1067 |
} |
1068 |
|
1069 |
return nb_clusters;
|
1070 |
} |
1071 |
|
1072 |
int qcow2_discard_clusters(BlockDriverState *bs, uint64_t offset,
|
1073 |
int nb_sectors)
|
1074 |
{ |
1075 |
BDRVQcowState *s = bs->opaque; |
1076 |
uint64_t end_offset; |
1077 |
unsigned int nb_clusters; |
1078 |
int ret;
|
1079 |
|
1080 |
end_offset = offset + (nb_sectors << BDRV_SECTOR_BITS); |
1081 |
|
1082 |
/* Round start up and end down */
|
1083 |
offset = align_offset(offset, s->cluster_size); |
1084 |
end_offset &= ~(s->cluster_size - 1);
|
1085 |
|
1086 |
if (offset > end_offset) {
|
1087 |
return 0; |
1088 |
} |
1089 |
|
1090 |
nb_clusters = size_to_clusters(s, end_offset - offset); |
1091 |
|
1092 |
/* Each L2 table is handled by its own loop iteration */
|
1093 |
while (nb_clusters > 0) { |
1094 |
ret = discard_single_l2(bs, offset, nb_clusters); |
1095 |
if (ret < 0) { |
1096 |
return ret;
|
1097 |
} |
1098 |
|
1099 |
nb_clusters -= ret; |
1100 |
offset += (ret * s->cluster_size); |
1101 |
} |
1102 |
|
1103 |
return 0; |
1104 |
} |
1105 |
|
1106 |
/*
|
1107 |
* This zeroes as many clusters of nb_clusters as possible at once (i.e.
|
1108 |
* all clusters in the same L2 table) and returns the number of zeroed
|
1109 |
* clusters.
|
1110 |
*/
|
1111 |
static int zero_single_l2(BlockDriverState *bs, uint64_t offset, |
1112 |
unsigned int nb_clusters) |
1113 |
{ |
1114 |
BDRVQcowState *s = bs->opaque; |
1115 |
uint64_t *l2_table; |
1116 |
int l2_index;
|
1117 |
int ret;
|
1118 |
int i;
|
1119 |
|
1120 |
ret = get_cluster_table(bs, offset, &l2_table, &l2_index); |
1121 |
if (ret < 0) { |
1122 |
return ret;
|
1123 |
} |
1124 |
|
1125 |
/* Limit nb_clusters to one L2 table */
|
1126 |
nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); |
1127 |
|
1128 |
for (i = 0; i < nb_clusters; i++) { |
1129 |
uint64_t old_offset; |
1130 |
|
1131 |
old_offset = be64_to_cpu(l2_table[l2_index + i]); |
1132 |
|
1133 |
/* Update L2 entries */
|
1134 |
qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); |
1135 |
if (old_offset & QCOW_OFLAG_COMPRESSED) {
|
1136 |
l2_table[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO); |
1137 |
qcow2_free_any_clusters(bs, old_offset, 1);
|
1138 |
} else {
|
1139 |
l2_table[l2_index + i] |= cpu_to_be64(QCOW_OFLAG_ZERO); |
1140 |
} |
1141 |
} |
1142 |
|
1143 |
ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
|
1144 |
if (ret < 0) { |
1145 |
return ret;
|
1146 |
} |
1147 |
|
1148 |
return nb_clusters;
|
1149 |
} |
1150 |
|
1151 |
int qcow2_zero_clusters(BlockDriverState *bs, uint64_t offset, int nb_sectors) |
1152 |
{ |
1153 |
BDRVQcowState *s = bs->opaque; |
1154 |
unsigned int nb_clusters; |
1155 |
int ret;
|
1156 |
|
1157 |
/* The zero flag is only supported by version 3 and newer */
|
1158 |
if (s->qcow_version < 3) { |
1159 |
return -ENOTSUP;
|
1160 |
} |
1161 |
|
1162 |
/* Each L2 table is handled by its own loop iteration */
|
1163 |
nb_clusters = size_to_clusters(s, nb_sectors << BDRV_SECTOR_BITS); |
1164 |
|
1165 |
while (nb_clusters > 0) { |
1166 |
ret = zero_single_l2(bs, offset, nb_clusters); |
1167 |
if (ret < 0) { |
1168 |
return ret;
|
1169 |
} |
1170 |
|
1171 |
nb_clusters -= ret; |
1172 |
offset += (ret * s->cluster_size); |
1173 |
} |
1174 |
|
1175 |
return 0; |
1176 |
} |