root / block / qcow2-cluster.c @ dc7588c1
History | View | Annotate | Download (43.2 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/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 |
if (s->qcow_version < 3) { |
458 |
return -EIO;
|
459 |
} |
460 |
c = count_contiguous_clusters(nb_clusters, s->cluster_size, |
461 |
&l2_table[l2_index], 0,
|
462 |
QCOW_OFLAG_COMPRESSED | QCOW_OFLAG_ZERO); |
463 |
*cluster_offset = 0;
|
464 |
break;
|
465 |
case QCOW2_CLUSTER_UNALLOCATED:
|
466 |
/* how many empty clusters ? */
|
467 |
c = count_contiguous_free_clusters(nb_clusters, &l2_table[l2_index]); |
468 |
*cluster_offset = 0;
|
469 |
break;
|
470 |
case QCOW2_CLUSTER_NORMAL:
|
471 |
/* how many allocated clusters ? */
|
472 |
c = count_contiguous_clusters(nb_clusters, s->cluster_size, |
473 |
&l2_table[l2_index], 0,
|
474 |
QCOW_OFLAG_COMPRESSED | QCOW_OFLAG_ZERO); |
475 |
*cluster_offset &= L2E_OFFSET_MASK; |
476 |
break;
|
477 |
default:
|
478 |
abort(); |
479 |
} |
480 |
|
481 |
qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
|
482 |
|
483 |
nb_available = (c * s->cluster_sectors); |
484 |
|
485 |
out:
|
486 |
if (nb_available > nb_needed)
|
487 |
nb_available = nb_needed; |
488 |
|
489 |
*num = nb_available - index_in_cluster; |
490 |
|
491 |
return ret;
|
492 |
} |
493 |
|
494 |
/*
|
495 |
* get_cluster_table
|
496 |
*
|
497 |
* for a given disk offset, load (and allocate if needed)
|
498 |
* the l2 table.
|
499 |
*
|
500 |
* the l2 table offset in the qcow2 file and the cluster index
|
501 |
* in the l2 table are given to the caller.
|
502 |
*
|
503 |
* Returns 0 on success, -errno in failure case
|
504 |
*/
|
505 |
static int get_cluster_table(BlockDriverState *bs, uint64_t offset, |
506 |
uint64_t **new_l2_table, |
507 |
int *new_l2_index)
|
508 |
{ |
509 |
BDRVQcowState *s = bs->opaque; |
510 |
unsigned int l1_index, l2_index; |
511 |
uint64_t l2_offset; |
512 |
uint64_t *l2_table = NULL;
|
513 |
int ret;
|
514 |
|
515 |
/* seek the the l2 offset in the l1 table */
|
516 |
|
517 |
l1_index = offset >> (s->l2_bits + s->cluster_bits); |
518 |
if (l1_index >= s->l1_size) {
|
519 |
ret = qcow2_grow_l1_table(bs, l1_index + 1, false); |
520 |
if (ret < 0) { |
521 |
return ret;
|
522 |
} |
523 |
} |
524 |
|
525 |
l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; |
526 |
|
527 |
/* seek the l2 table of the given l2 offset */
|
528 |
|
529 |
if (s->l1_table[l1_index] & QCOW_OFLAG_COPIED) {
|
530 |
/* load the l2 table in memory */
|
531 |
ret = l2_load(bs, l2_offset, &l2_table); |
532 |
if (ret < 0) { |
533 |
return ret;
|
534 |
} |
535 |
} else {
|
536 |
/* First allocate a new L2 table (and do COW if needed) */
|
537 |
ret = l2_allocate(bs, l1_index, &l2_table); |
538 |
if (ret < 0) { |
539 |
return ret;
|
540 |
} |
541 |
|
542 |
/* Then decrease the refcount of the old table */
|
543 |
if (l2_offset) {
|
544 |
qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t));
|
545 |
} |
546 |
} |
547 |
|
548 |
/* find the cluster offset for the given disk offset */
|
549 |
|
550 |
l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
|
551 |
|
552 |
*new_l2_table = l2_table; |
553 |
*new_l2_index = l2_index; |
554 |
|
555 |
return 0; |
556 |
} |
557 |
|
558 |
/*
|
559 |
* alloc_compressed_cluster_offset
|
560 |
*
|
561 |
* For a given offset of the disk image, return cluster offset in
|
562 |
* qcow2 file.
|
563 |
*
|
564 |
* If the offset is not found, allocate a new compressed cluster.
|
565 |
*
|
566 |
* Return the cluster offset if successful,
|
567 |
* Return 0, otherwise.
|
568 |
*
|
569 |
*/
|
570 |
|
571 |
uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs, |
572 |
uint64_t offset, |
573 |
int compressed_size)
|
574 |
{ |
575 |
BDRVQcowState *s = bs->opaque; |
576 |
int l2_index, ret;
|
577 |
uint64_t *l2_table; |
578 |
int64_t cluster_offset; |
579 |
int nb_csectors;
|
580 |
|
581 |
ret = get_cluster_table(bs, offset, &l2_table, &l2_index); |
582 |
if (ret < 0) { |
583 |
return 0; |
584 |
} |
585 |
|
586 |
/* Compression can't overwrite anything. Fail if the cluster was already
|
587 |
* allocated. */
|
588 |
cluster_offset = be64_to_cpu(l2_table[l2_index]); |
589 |
if (cluster_offset & L2E_OFFSET_MASK) {
|
590 |
qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
|
591 |
return 0; |
592 |
} |
593 |
|
594 |
cluster_offset = qcow2_alloc_bytes(bs, compressed_size); |
595 |
if (cluster_offset < 0) { |
596 |
qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
|
597 |
return 0; |
598 |
} |
599 |
|
600 |
nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) - |
601 |
(cluster_offset >> 9);
|
602 |
|
603 |
cluster_offset |= QCOW_OFLAG_COMPRESSED | |
604 |
((uint64_t)nb_csectors << s->csize_shift); |
605 |
|
606 |
/* update L2 table */
|
607 |
|
608 |
/* compressed clusters never have the copied flag */
|
609 |
|
610 |
BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED); |
611 |
qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); |
612 |
l2_table[l2_index] = cpu_to_be64(cluster_offset); |
613 |
ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
|
614 |
if (ret < 0) { |
615 |
return 0; |
616 |
} |
617 |
|
618 |
return cluster_offset;
|
619 |
} |
620 |
|
621 |
static int perform_cow(BlockDriverState *bs, QCowL2Meta *m, Qcow2COWRegion *r) |
622 |
{ |
623 |
BDRVQcowState *s = bs->opaque; |
624 |
int ret;
|
625 |
|
626 |
if (r->nb_sectors == 0) { |
627 |
return 0; |
628 |
} |
629 |
|
630 |
qemu_co_mutex_unlock(&s->lock); |
631 |
ret = copy_sectors(bs, m->offset / BDRV_SECTOR_SIZE, m->alloc_offset, |
632 |
r->offset / BDRV_SECTOR_SIZE, |
633 |
r->offset / BDRV_SECTOR_SIZE + r->nb_sectors); |
634 |
qemu_co_mutex_lock(&s->lock); |
635 |
|
636 |
if (ret < 0) { |
637 |
return ret;
|
638 |
} |
639 |
|
640 |
/*
|
641 |
* Before we update the L2 table to actually point to the new cluster, we
|
642 |
* need to be sure that the refcounts have been increased and COW was
|
643 |
* handled.
|
644 |
*/
|
645 |
qcow2_cache_depends_on_flush(s->l2_table_cache); |
646 |
|
647 |
return 0; |
648 |
} |
649 |
|
650 |
int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m)
|
651 |
{ |
652 |
BDRVQcowState *s = bs->opaque; |
653 |
int i, j = 0, l2_index, ret; |
654 |
uint64_t *old_cluster, *l2_table; |
655 |
uint64_t cluster_offset = m->alloc_offset; |
656 |
|
657 |
trace_qcow2_cluster_link_l2(qemu_coroutine_self(), m->nb_clusters); |
658 |
assert(m->nb_clusters > 0);
|
659 |
|
660 |
old_cluster = g_malloc(m->nb_clusters * sizeof(uint64_t));
|
661 |
|
662 |
/* copy content of unmodified sectors */
|
663 |
ret = perform_cow(bs, m, &m->cow_start); |
664 |
if (ret < 0) { |
665 |
goto err;
|
666 |
} |
667 |
|
668 |
ret = perform_cow(bs, m, &m->cow_end); |
669 |
if (ret < 0) { |
670 |
goto err;
|
671 |
} |
672 |
|
673 |
/* Update L2 table. */
|
674 |
if (s->use_lazy_refcounts) {
|
675 |
qcow2_mark_dirty(bs); |
676 |
} |
677 |
if (qcow2_need_accurate_refcounts(s)) {
|
678 |
qcow2_cache_set_dependency(bs, s->l2_table_cache, |
679 |
s->refcount_block_cache); |
680 |
} |
681 |
|
682 |
ret = get_cluster_table(bs, m->offset, &l2_table, &l2_index); |
683 |
if (ret < 0) { |
684 |
goto err;
|
685 |
} |
686 |
qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); |
687 |
|
688 |
for (i = 0; i < m->nb_clusters; i++) { |
689 |
/* if two concurrent writes happen to the same unallocated cluster
|
690 |
* each write allocates separate cluster and writes data concurrently.
|
691 |
* The first one to complete updates l2 table with pointer to its
|
692 |
* cluster the second one has to do RMW (which is done above by
|
693 |
* copy_sectors()), update l2 table with its cluster pointer and free
|
694 |
* old cluster. This is what this loop does */
|
695 |
if(l2_table[l2_index + i] != 0) |
696 |
old_cluster[j++] = l2_table[l2_index + i]; |
697 |
|
698 |
l2_table[l2_index + i] = cpu_to_be64((cluster_offset + |
699 |
(i << s->cluster_bits)) | QCOW_OFLAG_COPIED); |
700 |
} |
701 |
|
702 |
|
703 |
ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
|
704 |
if (ret < 0) { |
705 |
goto err;
|
706 |
} |
707 |
|
708 |
/*
|
709 |
* If this was a COW, we need to decrease the refcount of the old cluster.
|
710 |
* Also flush bs->file to get the right order for L2 and refcount update.
|
711 |
*/
|
712 |
if (j != 0) { |
713 |
for (i = 0; i < j; i++) { |
714 |
qcow2_free_any_clusters(bs, be64_to_cpu(old_cluster[i]), 1);
|
715 |
} |
716 |
} |
717 |
|
718 |
ret = 0;
|
719 |
err:
|
720 |
g_free(old_cluster); |
721 |
return ret;
|
722 |
} |
723 |
|
724 |
/*
|
725 |
* Returns the number of contiguous clusters that can be used for an allocating
|
726 |
* write, but require COW to be performed (this includes yet unallocated space,
|
727 |
* which must copy from the backing file)
|
728 |
*/
|
729 |
static int count_cow_clusters(BDRVQcowState *s, int nb_clusters, |
730 |
uint64_t *l2_table, int l2_index)
|
731 |
{ |
732 |
int i;
|
733 |
|
734 |
for (i = 0; i < nb_clusters; i++) { |
735 |
uint64_t l2_entry = be64_to_cpu(l2_table[l2_index + i]); |
736 |
int cluster_type = qcow2_get_cluster_type(l2_entry);
|
737 |
|
738 |
switch(cluster_type) {
|
739 |
case QCOW2_CLUSTER_NORMAL:
|
740 |
if (l2_entry & QCOW_OFLAG_COPIED) {
|
741 |
goto out;
|
742 |
} |
743 |
break;
|
744 |
case QCOW2_CLUSTER_UNALLOCATED:
|
745 |
case QCOW2_CLUSTER_COMPRESSED:
|
746 |
case QCOW2_CLUSTER_ZERO:
|
747 |
break;
|
748 |
default:
|
749 |
abort(); |
750 |
} |
751 |
} |
752 |
|
753 |
out:
|
754 |
assert(i <= nb_clusters); |
755 |
return i;
|
756 |
} |
757 |
|
758 |
/*
|
759 |
* Check if there already is an AIO write request in flight which allocates
|
760 |
* the same cluster. In this case we need to wait until the previous
|
761 |
* request has completed and updated the L2 table accordingly.
|
762 |
*
|
763 |
* Returns:
|
764 |
* 0 if there was no dependency. *cur_bytes indicates the number of
|
765 |
* bytes from guest_offset that can be read before the next
|
766 |
* dependency must be processed (or the request is complete)
|
767 |
*
|
768 |
* -EAGAIN if we had to wait for another request, previously gathered
|
769 |
* information on cluster allocation may be invalid now. The caller
|
770 |
* must start over anyway, so consider *cur_bytes undefined.
|
771 |
*/
|
772 |
static int handle_dependencies(BlockDriverState *bs, uint64_t guest_offset, |
773 |
uint64_t *cur_bytes, QCowL2Meta **m) |
774 |
{ |
775 |
BDRVQcowState *s = bs->opaque; |
776 |
QCowL2Meta *old_alloc; |
777 |
uint64_t bytes = *cur_bytes; |
778 |
|
779 |
QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) { |
780 |
|
781 |
uint64_t start = guest_offset; |
782 |
uint64_t end = start + bytes; |
783 |
uint64_t old_start = l2meta_cow_start(old_alloc); |
784 |
uint64_t old_end = l2meta_cow_end(old_alloc); |
785 |
|
786 |
if (end <= old_start || start >= old_end) {
|
787 |
/* No intersection */
|
788 |
} else {
|
789 |
if (start < old_start) {
|
790 |
/* Stop at the start of a running allocation */
|
791 |
bytes = old_start - start; |
792 |
} else {
|
793 |
bytes = 0;
|
794 |
} |
795 |
|
796 |
/* Stop if already an l2meta exists. After yielding, it wouldn't
|
797 |
* be valid any more, so we'd have to clean up the old L2Metas
|
798 |
* and deal with requests depending on them before starting to
|
799 |
* gather new ones. Not worth the trouble. */
|
800 |
if (bytes == 0 && *m) { |
801 |
*cur_bytes = 0;
|
802 |
return 0; |
803 |
} |
804 |
|
805 |
if (bytes == 0) { |
806 |
/* Wait for the dependency to complete. We need to recheck
|
807 |
* the free/allocated clusters when we continue. */
|
808 |
qemu_co_mutex_unlock(&s->lock); |
809 |
qemu_co_queue_wait(&old_alloc->dependent_requests); |
810 |
qemu_co_mutex_lock(&s->lock); |
811 |
return -EAGAIN;
|
812 |
} |
813 |
} |
814 |
} |
815 |
|
816 |
/* Make sure that existing clusters and new allocations are only used up to
|
817 |
* the next dependency if we shortened the request above */
|
818 |
*cur_bytes = bytes; |
819 |
|
820 |
return 0; |
821 |
} |
822 |
|
823 |
/*
|
824 |
* Checks how many already allocated clusters that don't require a copy on
|
825 |
* write there are at the given guest_offset (up to *bytes). If
|
826 |
* *host_offset is not zero, only physically contiguous clusters beginning at
|
827 |
* this host offset are counted.
|
828 |
*
|
829 |
* Note that guest_offset may not be cluster aligned. In this case, the
|
830 |
* returned *host_offset points to exact byte referenced by guest_offset and
|
831 |
* therefore isn't cluster aligned as well.
|
832 |
*
|
833 |
* Returns:
|
834 |
* 0: if no allocated clusters are available at the given offset.
|
835 |
* *bytes is normally unchanged. It is set to 0 if the cluster
|
836 |
* is allocated and doesn't need COW, but doesn't have the right
|
837 |
* physical offset.
|
838 |
*
|
839 |
* 1: if allocated clusters that don't require a COW are available at
|
840 |
* the requested offset. *bytes may have decreased and describes
|
841 |
* the length of the area that can be written to.
|
842 |
*
|
843 |
* -errno: in error cases
|
844 |
*/
|
845 |
static int handle_copied(BlockDriverState *bs, uint64_t guest_offset, |
846 |
uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m) |
847 |
{ |
848 |
BDRVQcowState *s = bs->opaque; |
849 |
int l2_index;
|
850 |
uint64_t cluster_offset; |
851 |
uint64_t *l2_table; |
852 |
unsigned int nb_clusters; |
853 |
unsigned int keep_clusters; |
854 |
int ret, pret;
|
855 |
|
856 |
trace_qcow2_handle_copied(qemu_coroutine_self(), guest_offset, *host_offset, |
857 |
*bytes); |
858 |
|
859 |
assert(*host_offset == 0 || offset_into_cluster(s, guest_offset)
|
860 |
== offset_into_cluster(s, *host_offset)); |
861 |
|
862 |
/*
|
863 |
* Calculate the number of clusters to look for. We stop at L2 table
|
864 |
* boundaries to keep things simple.
|
865 |
*/
|
866 |
nb_clusters = |
867 |
size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes); |
868 |
|
869 |
l2_index = offset_to_l2_index(s, guest_offset); |
870 |
nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); |
871 |
|
872 |
/* Find L2 entry for the first involved cluster */
|
873 |
ret = get_cluster_table(bs, guest_offset, &l2_table, &l2_index); |
874 |
if (ret < 0) { |
875 |
return ret;
|
876 |
} |
877 |
|
878 |
cluster_offset = be64_to_cpu(l2_table[l2_index]); |
879 |
|
880 |
/* Check how many clusters are already allocated and don't need COW */
|
881 |
if (qcow2_get_cluster_type(cluster_offset) == QCOW2_CLUSTER_NORMAL
|
882 |
&& (cluster_offset & QCOW_OFLAG_COPIED)) |
883 |
{ |
884 |
/* If a specific host_offset is required, check it */
|
885 |
bool offset_matches =
|
886 |
(cluster_offset & L2E_OFFSET_MASK) == *host_offset; |
887 |
|
888 |
if (*host_offset != 0 && !offset_matches) { |
889 |
*bytes = 0;
|
890 |
ret = 0;
|
891 |
goto out;
|
892 |
} |
893 |
|
894 |
/* We keep all QCOW_OFLAG_COPIED clusters */
|
895 |
keep_clusters = |
896 |
count_contiguous_clusters(nb_clusters, s->cluster_size, |
897 |
&l2_table[l2_index], 0,
|
898 |
QCOW_OFLAG_COPIED | QCOW_OFLAG_ZERO); |
899 |
assert(keep_clusters <= nb_clusters); |
900 |
|
901 |
*bytes = MIN(*bytes, |
902 |
keep_clusters * s->cluster_size |
903 |
- offset_into_cluster(s, guest_offset)); |
904 |
|
905 |
ret = 1;
|
906 |
} else {
|
907 |
ret = 0;
|
908 |
} |
909 |
|
910 |
/* Cleanup */
|
911 |
out:
|
912 |
pret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
|
913 |
if (pret < 0) { |
914 |
return pret;
|
915 |
} |
916 |
|
917 |
/* Only return a host offset if we actually made progress. Otherwise we
|
918 |
* would make requirements for handle_alloc() that it can't fulfill */
|
919 |
if (ret) {
|
920 |
*host_offset = (cluster_offset & L2E_OFFSET_MASK) |
921 |
+ offset_into_cluster(s, guest_offset); |
922 |
} |
923 |
|
924 |
return ret;
|
925 |
} |
926 |
|
927 |
/*
|
928 |
* Allocates new clusters for the given guest_offset.
|
929 |
*
|
930 |
* At most *nb_clusters are allocated, and on return *nb_clusters is updated to
|
931 |
* contain the number of clusters that have been allocated and are contiguous
|
932 |
* in the image file.
|
933 |
*
|
934 |
* If *host_offset is non-zero, it specifies the offset in the image file at
|
935 |
* which the new clusters must start. *nb_clusters can be 0 on return in this
|
936 |
* case if the cluster at host_offset is already in use. If *host_offset is
|
937 |
* zero, the clusters can be allocated anywhere in the image file.
|
938 |
*
|
939 |
* *host_offset is updated to contain the offset into the image file at which
|
940 |
* the first allocated cluster starts.
|
941 |
*
|
942 |
* Return 0 on success and -errno in error cases. -EAGAIN means that the
|
943 |
* function has been waiting for another request and the allocation must be
|
944 |
* restarted, but the whole request should not be failed.
|
945 |
*/
|
946 |
static int do_alloc_cluster_offset(BlockDriverState *bs, uint64_t guest_offset, |
947 |
uint64_t *host_offset, unsigned int *nb_clusters) |
948 |
{ |
949 |
BDRVQcowState *s = bs->opaque; |
950 |
|
951 |
trace_qcow2_do_alloc_clusters_offset(qemu_coroutine_self(), guest_offset, |
952 |
*host_offset, *nb_clusters); |
953 |
|
954 |
/* Allocate new clusters */
|
955 |
trace_qcow2_cluster_alloc_phys(qemu_coroutine_self()); |
956 |
if (*host_offset == 0) { |
957 |
int64_t cluster_offset = |
958 |
qcow2_alloc_clusters(bs, *nb_clusters * s->cluster_size); |
959 |
if (cluster_offset < 0) { |
960 |
return cluster_offset;
|
961 |
} |
962 |
*host_offset = cluster_offset; |
963 |
return 0; |
964 |
} else {
|
965 |
int ret = qcow2_alloc_clusters_at(bs, *host_offset, *nb_clusters);
|
966 |
if (ret < 0) { |
967 |
return ret;
|
968 |
} |
969 |
*nb_clusters = ret; |
970 |
return 0; |
971 |
} |
972 |
} |
973 |
|
974 |
/*
|
975 |
* Allocates new clusters for an area that either is yet unallocated or needs a
|
976 |
* copy on write. If *host_offset is non-zero, clusters are only allocated if
|
977 |
* the new allocation can match the specified host offset.
|
978 |
*
|
979 |
* Note that guest_offset may not be cluster aligned. In this case, the
|
980 |
* returned *host_offset points to exact byte referenced by guest_offset and
|
981 |
* therefore isn't cluster aligned as well.
|
982 |
*
|
983 |
* Returns:
|
984 |
* 0: if no clusters could be allocated. *bytes is set to 0,
|
985 |
* *host_offset is left unchanged.
|
986 |
*
|
987 |
* 1: if new clusters were allocated. *bytes may be decreased if the
|
988 |
* new allocation doesn't cover all of the requested area.
|
989 |
* *host_offset is updated to contain the host offset of the first
|
990 |
* newly allocated cluster.
|
991 |
*
|
992 |
* -errno: in error cases
|
993 |
*/
|
994 |
static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset, |
995 |
uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m) |
996 |
{ |
997 |
BDRVQcowState *s = bs->opaque; |
998 |
int l2_index;
|
999 |
uint64_t *l2_table; |
1000 |
uint64_t entry; |
1001 |
unsigned int nb_clusters; |
1002 |
int ret;
|
1003 |
|
1004 |
uint64_t alloc_cluster_offset; |
1005 |
|
1006 |
trace_qcow2_handle_alloc(qemu_coroutine_self(), guest_offset, *host_offset, |
1007 |
*bytes); |
1008 |
assert(*bytes > 0);
|
1009 |
|
1010 |
/*
|
1011 |
* Calculate the number of clusters to look for. We stop at L2 table
|
1012 |
* boundaries to keep things simple.
|
1013 |
*/
|
1014 |
nb_clusters = |
1015 |
size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes); |
1016 |
|
1017 |
l2_index = offset_to_l2_index(s, guest_offset); |
1018 |
nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); |
1019 |
|
1020 |
/* Find L2 entry for the first involved cluster */
|
1021 |
ret = get_cluster_table(bs, guest_offset, &l2_table, &l2_index); |
1022 |
if (ret < 0) { |
1023 |
return ret;
|
1024 |
} |
1025 |
|
1026 |
entry = be64_to_cpu(l2_table[l2_index]); |
1027 |
|
1028 |
/* For the moment, overwrite compressed clusters one by one */
|
1029 |
if (entry & QCOW_OFLAG_COMPRESSED) {
|
1030 |
nb_clusters = 1;
|
1031 |
} else {
|
1032 |
nb_clusters = count_cow_clusters(s, nb_clusters, l2_table, l2_index); |
1033 |
} |
1034 |
|
1035 |
/* This function is only called when there were no non-COW clusters, so if
|
1036 |
* we can't find any unallocated or COW clusters either, something is
|
1037 |
* wrong with our code. */
|
1038 |
assert(nb_clusters > 0);
|
1039 |
|
1040 |
ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
|
1041 |
if (ret < 0) { |
1042 |
return ret;
|
1043 |
} |
1044 |
|
1045 |
/* Allocate, if necessary at a given offset in the image file */
|
1046 |
alloc_cluster_offset = start_of_cluster(s, *host_offset); |
1047 |
ret = do_alloc_cluster_offset(bs, guest_offset, &alloc_cluster_offset, |
1048 |
&nb_clusters); |
1049 |
if (ret < 0) { |
1050 |
goto fail;
|
1051 |
} |
1052 |
|
1053 |
/* Can't extend contiguous allocation */
|
1054 |
if (nb_clusters == 0) { |
1055 |
*bytes = 0;
|
1056 |
return 0; |
1057 |
} |
1058 |
|
1059 |
/*
|
1060 |
* Save info needed for meta data update.
|
1061 |
*
|
1062 |
* requested_sectors: Number of sectors from the start of the first
|
1063 |
* newly allocated cluster to the end of the (possibly shortened
|
1064 |
* before) write request.
|
1065 |
*
|
1066 |
* avail_sectors: Number of sectors from the start of the first
|
1067 |
* newly allocated to the end of the last newly allocated cluster.
|
1068 |
*
|
1069 |
* nb_sectors: The number of sectors from the start of the first
|
1070 |
* newly allocated cluster to the end of the area that the write
|
1071 |
* request actually writes to (excluding COW at the end)
|
1072 |
*/
|
1073 |
int requested_sectors =
|
1074 |
(*bytes + offset_into_cluster(s, guest_offset)) |
1075 |
>> BDRV_SECTOR_BITS; |
1076 |
int avail_sectors = nb_clusters
|
1077 |
<< (s->cluster_bits - BDRV_SECTOR_BITS); |
1078 |
int alloc_n_start = offset_into_cluster(s, guest_offset)
|
1079 |
>> BDRV_SECTOR_BITS; |
1080 |
int nb_sectors = MIN(requested_sectors, avail_sectors);
|
1081 |
QCowL2Meta *old_m = *m; |
1082 |
|
1083 |
*m = g_malloc0(sizeof(**m));
|
1084 |
|
1085 |
**m = (QCowL2Meta) { |
1086 |
.next = old_m, |
1087 |
|
1088 |
.alloc_offset = alloc_cluster_offset, |
1089 |
.offset = start_of_cluster(s, guest_offset), |
1090 |
.nb_clusters = nb_clusters, |
1091 |
.nb_available = nb_sectors, |
1092 |
|
1093 |
.cow_start = { |
1094 |
.offset = 0,
|
1095 |
.nb_sectors = alloc_n_start, |
1096 |
}, |
1097 |
.cow_end = { |
1098 |
.offset = nb_sectors * BDRV_SECTOR_SIZE, |
1099 |
.nb_sectors = avail_sectors - nb_sectors, |
1100 |
}, |
1101 |
}; |
1102 |
qemu_co_queue_init(&(*m)->dependent_requests); |
1103 |
QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight); |
1104 |
|
1105 |
*host_offset = alloc_cluster_offset + offset_into_cluster(s, guest_offset); |
1106 |
*bytes = MIN(*bytes, (nb_sectors * BDRV_SECTOR_SIZE) |
1107 |
- offset_into_cluster(s, guest_offset)); |
1108 |
assert(*bytes != 0);
|
1109 |
|
1110 |
return 1; |
1111 |
|
1112 |
fail:
|
1113 |
if (*m && (*m)->nb_clusters > 0) { |
1114 |
QLIST_REMOVE(*m, next_in_flight); |
1115 |
} |
1116 |
return ret;
|
1117 |
} |
1118 |
|
1119 |
/*
|
1120 |
* alloc_cluster_offset
|
1121 |
*
|
1122 |
* For a given offset on the virtual disk, find the cluster offset in qcow2
|
1123 |
* file. If the offset is not found, allocate a new cluster.
|
1124 |
*
|
1125 |
* If the cluster was already allocated, m->nb_clusters is set to 0 and
|
1126 |
* other fields in m are meaningless.
|
1127 |
*
|
1128 |
* If the cluster is newly allocated, m->nb_clusters is set to the number of
|
1129 |
* contiguous clusters that have been allocated. In this case, the other
|
1130 |
* fields of m are valid and contain information about the first allocated
|
1131 |
* cluster.
|
1132 |
*
|
1133 |
* If the request conflicts with another write request in flight, the coroutine
|
1134 |
* is queued and will be reentered when the dependency has completed.
|
1135 |
*
|
1136 |
* Return 0 on success and -errno in error cases
|
1137 |
*/
|
1138 |
int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset,
|
1139 |
int n_start, int n_end, int *num, uint64_t *host_offset, QCowL2Meta **m) |
1140 |
{ |
1141 |
BDRVQcowState *s = bs->opaque; |
1142 |
uint64_t start, remaining; |
1143 |
uint64_t cluster_offset; |
1144 |
uint64_t cur_bytes; |
1145 |
int ret;
|
1146 |
|
1147 |
trace_qcow2_alloc_clusters_offset(qemu_coroutine_self(), offset, |
1148 |
n_start, n_end); |
1149 |
|
1150 |
assert(n_start * BDRV_SECTOR_SIZE == offset_into_cluster(s, offset)); |
1151 |
offset = start_of_cluster(s, offset); |
1152 |
|
1153 |
again:
|
1154 |
start = offset + (n_start << BDRV_SECTOR_BITS); |
1155 |
remaining = (n_end - n_start) << BDRV_SECTOR_BITS; |
1156 |
cluster_offset = 0;
|
1157 |
*host_offset = 0;
|
1158 |
cur_bytes = 0;
|
1159 |
*m = NULL;
|
1160 |
|
1161 |
while (true) { |
1162 |
|
1163 |
if (!*host_offset) {
|
1164 |
*host_offset = start_of_cluster(s, cluster_offset); |
1165 |
} |
1166 |
|
1167 |
assert(remaining >= cur_bytes); |
1168 |
|
1169 |
start += cur_bytes; |
1170 |
remaining -= cur_bytes; |
1171 |
cluster_offset += cur_bytes; |
1172 |
|
1173 |
if (remaining == 0) { |
1174 |
break;
|
1175 |
} |
1176 |
|
1177 |
cur_bytes = remaining; |
1178 |
|
1179 |
/*
|
1180 |
* Now start gathering as many contiguous clusters as possible:
|
1181 |
*
|
1182 |
* 1. Check for overlaps with in-flight allocations
|
1183 |
*
|
1184 |
* a) Overlap not in the first cluster -> shorten this request and
|
1185 |
* let the caller handle the rest in its next loop iteration.
|
1186 |
*
|
1187 |
* b) Real overlaps of two requests. Yield and restart the search
|
1188 |
* for contiguous clusters (the situation could have changed
|
1189 |
* while we were sleeping)
|
1190 |
*
|
1191 |
* c) TODO: Request starts in the same cluster as the in-flight
|
1192 |
* allocation ends. Shorten the COW of the in-fight allocation,
|
1193 |
* set cluster_offset to write to the same cluster and set up
|
1194 |
* the right synchronisation between the in-flight request and
|
1195 |
* the new one.
|
1196 |
*/
|
1197 |
ret = handle_dependencies(bs, start, &cur_bytes, m); |
1198 |
if (ret == -EAGAIN) {
|
1199 |
/* Currently handle_dependencies() doesn't yield if we already had
|
1200 |
* an allocation. If it did, we would have to clean up the L2Meta
|
1201 |
* structs before starting over. */
|
1202 |
assert(*m == NULL);
|
1203 |
goto again;
|
1204 |
} else if (ret < 0) { |
1205 |
return ret;
|
1206 |
} else if (cur_bytes == 0) { |
1207 |
break;
|
1208 |
} else {
|
1209 |
/* handle_dependencies() may have decreased cur_bytes (shortened
|
1210 |
* the allocations below) so that the next dependency is processed
|
1211 |
* correctly during the next loop iteration. */
|
1212 |
} |
1213 |
|
1214 |
/*
|
1215 |
* 2. Count contiguous COPIED clusters.
|
1216 |
*/
|
1217 |
ret = handle_copied(bs, start, &cluster_offset, &cur_bytes, m); |
1218 |
if (ret < 0) { |
1219 |
return ret;
|
1220 |
} else if (ret) { |
1221 |
continue;
|
1222 |
} else if (cur_bytes == 0) { |
1223 |
break;
|
1224 |
} |
1225 |
|
1226 |
/*
|
1227 |
* 3. If the request still hasn't completed, allocate new clusters,
|
1228 |
* considering any cluster_offset of steps 1c or 2.
|
1229 |
*/
|
1230 |
ret = handle_alloc(bs, start, &cluster_offset, &cur_bytes, m); |
1231 |
if (ret < 0) { |
1232 |
return ret;
|
1233 |
} else if (ret) { |
1234 |
continue;
|
1235 |
} else {
|
1236 |
assert(cur_bytes == 0);
|
1237 |
break;
|
1238 |
} |
1239 |
} |
1240 |
|
1241 |
*num = (n_end - n_start) - (remaining >> BDRV_SECTOR_BITS); |
1242 |
assert(*num > 0);
|
1243 |
assert(*host_offset != 0);
|
1244 |
|
1245 |
return 0; |
1246 |
} |
1247 |
|
1248 |
static int decompress_buffer(uint8_t *out_buf, int out_buf_size, |
1249 |
const uint8_t *buf, int buf_size) |
1250 |
{ |
1251 |
z_stream strm1, *strm = &strm1; |
1252 |
int ret, out_len;
|
1253 |
|
1254 |
memset(strm, 0, sizeof(*strm)); |
1255 |
|
1256 |
strm->next_in = (uint8_t *)buf; |
1257 |
strm->avail_in = buf_size; |
1258 |
strm->next_out = out_buf; |
1259 |
strm->avail_out = out_buf_size; |
1260 |
|
1261 |
ret = inflateInit2(strm, -12);
|
1262 |
if (ret != Z_OK)
|
1263 |
return -1; |
1264 |
ret = inflate(strm, Z_FINISH); |
1265 |
out_len = strm->next_out - out_buf; |
1266 |
if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
|
1267 |
out_len != out_buf_size) { |
1268 |
inflateEnd(strm); |
1269 |
return -1; |
1270 |
} |
1271 |
inflateEnd(strm); |
1272 |
return 0; |
1273 |
} |
1274 |
|
1275 |
int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
|
1276 |
{ |
1277 |
BDRVQcowState *s = bs->opaque; |
1278 |
int ret, csize, nb_csectors, sector_offset;
|
1279 |
uint64_t coffset; |
1280 |
|
1281 |
coffset = cluster_offset & s->cluster_offset_mask; |
1282 |
if (s->cluster_cache_offset != coffset) {
|
1283 |
nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
|
1284 |
sector_offset = coffset & 511;
|
1285 |
csize = nb_csectors * 512 - sector_offset;
|
1286 |
BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED); |
1287 |
ret = bdrv_read(bs->file, coffset >> 9, s->cluster_data, nb_csectors);
|
1288 |
if (ret < 0) { |
1289 |
return ret;
|
1290 |
} |
1291 |
if (decompress_buffer(s->cluster_cache, s->cluster_size,
|
1292 |
s->cluster_data + sector_offset, csize) < 0) {
|
1293 |
return -EIO;
|
1294 |
} |
1295 |
s->cluster_cache_offset = coffset; |
1296 |
} |
1297 |
return 0; |
1298 |
} |
1299 |
|
1300 |
/*
|
1301 |
* This discards as many clusters of nb_clusters as possible at once (i.e.
|
1302 |
* all clusters in the same L2 table) and returns the number of discarded
|
1303 |
* clusters.
|
1304 |
*/
|
1305 |
static int discard_single_l2(BlockDriverState *bs, uint64_t offset, |
1306 |
unsigned int nb_clusters) |
1307 |
{ |
1308 |
BDRVQcowState *s = bs->opaque; |
1309 |
uint64_t *l2_table; |
1310 |
int l2_index;
|
1311 |
int ret;
|
1312 |
int i;
|
1313 |
|
1314 |
ret = get_cluster_table(bs, offset, &l2_table, &l2_index); |
1315 |
if (ret < 0) { |
1316 |
return ret;
|
1317 |
} |
1318 |
|
1319 |
/* Limit nb_clusters to one L2 table */
|
1320 |
nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); |
1321 |
|
1322 |
for (i = 0; i < nb_clusters; i++) { |
1323 |
uint64_t old_offset; |
1324 |
|
1325 |
old_offset = be64_to_cpu(l2_table[l2_index + i]); |
1326 |
if ((old_offset & L2E_OFFSET_MASK) == 0) { |
1327 |
continue;
|
1328 |
} |
1329 |
|
1330 |
/* First remove L2 entries */
|
1331 |
qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); |
1332 |
l2_table[l2_index + i] = cpu_to_be64(0);
|
1333 |
|
1334 |
/* Then decrease the refcount */
|
1335 |
qcow2_free_any_clusters(bs, old_offset, 1);
|
1336 |
} |
1337 |
|
1338 |
ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
|
1339 |
if (ret < 0) { |
1340 |
return ret;
|
1341 |
} |
1342 |
|
1343 |
return nb_clusters;
|
1344 |
} |
1345 |
|
1346 |
int qcow2_discard_clusters(BlockDriverState *bs, uint64_t offset,
|
1347 |
int nb_sectors)
|
1348 |
{ |
1349 |
BDRVQcowState *s = bs->opaque; |
1350 |
uint64_t end_offset; |
1351 |
unsigned int nb_clusters; |
1352 |
int ret;
|
1353 |
|
1354 |
end_offset = offset + (nb_sectors << BDRV_SECTOR_BITS); |
1355 |
|
1356 |
/* Round start up and end down */
|
1357 |
offset = align_offset(offset, s->cluster_size); |
1358 |
end_offset &= ~(s->cluster_size - 1);
|
1359 |
|
1360 |
if (offset > end_offset) {
|
1361 |
return 0; |
1362 |
} |
1363 |
|
1364 |
nb_clusters = size_to_clusters(s, end_offset - offset); |
1365 |
|
1366 |
/* Each L2 table is handled by its own loop iteration */
|
1367 |
while (nb_clusters > 0) { |
1368 |
ret = discard_single_l2(bs, offset, nb_clusters); |
1369 |
if (ret < 0) { |
1370 |
return ret;
|
1371 |
} |
1372 |
|
1373 |
nb_clusters -= ret; |
1374 |
offset += (ret * s->cluster_size); |
1375 |
} |
1376 |
|
1377 |
return 0; |
1378 |
} |
1379 |
|
1380 |
/*
|
1381 |
* This zeroes as many clusters of nb_clusters as possible at once (i.e.
|
1382 |
* all clusters in the same L2 table) and returns the number of zeroed
|
1383 |
* clusters.
|
1384 |
*/
|
1385 |
static int zero_single_l2(BlockDriverState *bs, uint64_t offset, |
1386 |
unsigned int nb_clusters) |
1387 |
{ |
1388 |
BDRVQcowState *s = bs->opaque; |
1389 |
uint64_t *l2_table; |
1390 |
int l2_index;
|
1391 |
int ret;
|
1392 |
int i;
|
1393 |
|
1394 |
ret = get_cluster_table(bs, offset, &l2_table, &l2_index); |
1395 |
if (ret < 0) { |
1396 |
return ret;
|
1397 |
} |
1398 |
|
1399 |
/* Limit nb_clusters to one L2 table */
|
1400 |
nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); |
1401 |
|
1402 |
for (i = 0; i < nb_clusters; i++) { |
1403 |
uint64_t old_offset; |
1404 |
|
1405 |
old_offset = be64_to_cpu(l2_table[l2_index + i]); |
1406 |
|
1407 |
/* Update L2 entries */
|
1408 |
qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); |
1409 |
if (old_offset & QCOW_OFLAG_COMPRESSED) {
|
1410 |
l2_table[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO); |
1411 |
qcow2_free_any_clusters(bs, old_offset, 1);
|
1412 |
} else {
|
1413 |
l2_table[l2_index + i] |= cpu_to_be64(QCOW_OFLAG_ZERO); |
1414 |
} |
1415 |
} |
1416 |
|
1417 |
ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
|
1418 |
if (ret < 0) { |
1419 |
return ret;
|
1420 |
} |
1421 |
|
1422 |
return nb_clusters;
|
1423 |
} |
1424 |
|
1425 |
int qcow2_zero_clusters(BlockDriverState *bs, uint64_t offset, int nb_sectors) |
1426 |
{ |
1427 |
BDRVQcowState *s = bs->opaque; |
1428 |
unsigned int nb_clusters; |
1429 |
int ret;
|
1430 |
|
1431 |
/* The zero flag is only supported by version 3 and newer */
|
1432 |
if (s->qcow_version < 3) { |
1433 |
return -ENOTSUP;
|
1434 |
} |
1435 |
|
1436 |
/* Each L2 table is handled by its own loop iteration */
|
1437 |
nb_clusters = size_to_clusters(s, nb_sectors << BDRV_SECTOR_BITS); |
1438 |
|
1439 |
while (nb_clusters > 0) { |
1440 |
ret = zero_single_l2(bs, offset, nb_clusters); |
1441 |
if (ret < 0) { |
1442 |
return ret;
|
1443 |
} |
1444 |
|
1445 |
nb_clusters -= ret; |
1446 |
offset += (ret * s->cluster_size); |
1447 |
} |
1448 |
|
1449 |
return 0; |
1450 |
} |