root / block / qcow2.c @ 99cce9fa
History | View | Annotate | Download (40.5 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 |
#include "qemu-common.h" |
25 |
#include "block_int.h" |
26 |
#include "module.h" |
27 |
#include <zlib.h> |
28 |
#include "aes.h" |
29 |
#include "block/qcow2.h" |
30 |
#include "qemu-error.h" |
31 |
#include "qerror.h" |
32 |
|
33 |
/*
|
34 |
Differences with QCOW:
|
35 |
|
36 |
- Support for multiple incremental snapshots.
|
37 |
- Memory management by reference counts.
|
38 |
- Clusters which have a reference count of one have the bit
|
39 |
QCOW_OFLAG_COPIED to optimize write performance.
|
40 |
- Size of compressed clusters is stored in sectors to reduce bit usage
|
41 |
in the cluster offsets.
|
42 |
- Support for storing additional data (such as the VM state) in the
|
43 |
snapshots.
|
44 |
- If a backing store is used, the cluster size is not constrained
|
45 |
(could be backported to QCOW).
|
46 |
- L2 tables have always a size of one cluster.
|
47 |
*/
|
48 |
|
49 |
|
50 |
typedef struct { |
51 |
uint32_t magic; |
52 |
uint32_t len; |
53 |
} QCowExtension; |
54 |
#define QCOW2_EXT_MAGIC_END 0 |
55 |
#define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA |
56 |
|
57 |
static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename) |
58 |
{ |
59 |
const QCowHeader *cow_header = (const void *)buf; |
60 |
|
61 |
if (buf_size >= sizeof(QCowHeader) && |
62 |
be32_to_cpu(cow_header->magic) == QCOW_MAGIC && |
63 |
be32_to_cpu(cow_header->version) >= QCOW_VERSION) |
64 |
return 100; |
65 |
else
|
66 |
return 0; |
67 |
} |
68 |
|
69 |
|
70 |
/*
|
71 |
* read qcow2 extension and fill bs
|
72 |
* start reading from start_offset
|
73 |
* finish reading upon magic of value 0 or when end_offset reached
|
74 |
* unknown magic is skipped (future extension this version knows nothing about)
|
75 |
* return 0 upon success, non-0 otherwise
|
76 |
*/
|
77 |
static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset, |
78 |
uint64_t end_offset) |
79 |
{ |
80 |
QCowExtension ext; |
81 |
uint64_t offset; |
82 |
|
83 |
#ifdef DEBUG_EXT
|
84 |
printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
|
85 |
#endif
|
86 |
offset = start_offset; |
87 |
while (offset < end_offset) {
|
88 |
|
89 |
#ifdef DEBUG_EXT
|
90 |
/* Sanity check */
|
91 |
if (offset > s->cluster_size)
|
92 |
printf("qcow2_read_extension: suspicious offset %lu\n", offset);
|
93 |
|
94 |
printf("attemting to read extended header in offset %lu\n", offset);
|
95 |
#endif
|
96 |
|
97 |
if (bdrv_pread(bs->file, offset, &ext, sizeof(ext)) != sizeof(ext)) { |
98 |
fprintf(stderr, "qcow2_read_extension: ERROR: "
|
99 |
"pread fail from offset %" PRIu64 "\n", |
100 |
offset); |
101 |
return 1; |
102 |
} |
103 |
be32_to_cpus(&ext.magic); |
104 |
be32_to_cpus(&ext.len); |
105 |
offset += sizeof(ext);
|
106 |
#ifdef DEBUG_EXT
|
107 |
printf("ext.magic = 0x%x\n", ext.magic);
|
108 |
#endif
|
109 |
switch (ext.magic) {
|
110 |
case QCOW2_EXT_MAGIC_END:
|
111 |
return 0; |
112 |
|
113 |
case QCOW2_EXT_MAGIC_BACKING_FORMAT:
|
114 |
if (ext.len >= sizeof(bs->backing_format)) { |
115 |
fprintf(stderr, "ERROR: ext_backing_format: len=%u too large"
|
116 |
" (>=%zu)\n",
|
117 |
ext.len, sizeof(bs->backing_format));
|
118 |
return 2; |
119 |
} |
120 |
if (bdrv_pread(bs->file, offset , bs->backing_format,
|
121 |
ext.len) != ext.len) |
122 |
return 3; |
123 |
bs->backing_format[ext.len] = '\0';
|
124 |
#ifdef DEBUG_EXT
|
125 |
printf("Qcow2: Got format extension %s\n", bs->backing_format);
|
126 |
#endif
|
127 |
offset = ((offset + ext.len + 7) & ~7); |
128 |
break;
|
129 |
|
130 |
default:
|
131 |
/* unknown magic -- just skip it */
|
132 |
offset = ((offset + ext.len + 7) & ~7); |
133 |
break;
|
134 |
} |
135 |
} |
136 |
|
137 |
return 0; |
138 |
} |
139 |
|
140 |
|
141 |
static int qcow2_open(BlockDriverState *bs, int flags) |
142 |
{ |
143 |
BDRVQcowState *s = bs->opaque; |
144 |
int len, i, ret = 0; |
145 |
QCowHeader header; |
146 |
uint64_t ext_end; |
147 |
bool writethrough;
|
148 |
|
149 |
ret = bdrv_pread(bs->file, 0, &header, sizeof(header)); |
150 |
if (ret < 0) { |
151 |
goto fail;
|
152 |
} |
153 |
be32_to_cpus(&header.magic); |
154 |
be32_to_cpus(&header.version); |
155 |
be64_to_cpus(&header.backing_file_offset); |
156 |
be32_to_cpus(&header.backing_file_size); |
157 |
be64_to_cpus(&header.size); |
158 |
be32_to_cpus(&header.cluster_bits); |
159 |
be32_to_cpus(&header.crypt_method); |
160 |
be64_to_cpus(&header.l1_table_offset); |
161 |
be32_to_cpus(&header.l1_size); |
162 |
be64_to_cpus(&header.refcount_table_offset); |
163 |
be32_to_cpus(&header.refcount_table_clusters); |
164 |
be64_to_cpus(&header.snapshots_offset); |
165 |
be32_to_cpus(&header.nb_snapshots); |
166 |
|
167 |
if (header.magic != QCOW_MAGIC) {
|
168 |
ret = -EINVAL; |
169 |
goto fail;
|
170 |
} |
171 |
if (header.version != QCOW_VERSION) {
|
172 |
char version[64]; |
173 |
snprintf(version, sizeof(version), "QCOW version %d", header.version); |
174 |
qerror_report(QERR_UNKNOWN_BLOCK_FORMAT_FEATURE, |
175 |
bs->device_name, "qcow2", version);
|
176 |
ret = -ENOTSUP; |
177 |
goto fail;
|
178 |
} |
179 |
if (header.cluster_bits < MIN_CLUSTER_BITS ||
|
180 |
header.cluster_bits > MAX_CLUSTER_BITS) { |
181 |
ret = -EINVAL; |
182 |
goto fail;
|
183 |
} |
184 |
if (header.crypt_method > QCOW_CRYPT_AES) {
|
185 |
ret = -EINVAL; |
186 |
goto fail;
|
187 |
} |
188 |
s->crypt_method_header = header.crypt_method; |
189 |
if (s->crypt_method_header) {
|
190 |
bs->encrypted = 1;
|
191 |
} |
192 |
s->cluster_bits = header.cluster_bits; |
193 |
s->cluster_size = 1 << s->cluster_bits;
|
194 |
s->cluster_sectors = 1 << (s->cluster_bits - 9); |
195 |
s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */ |
196 |
s->l2_size = 1 << s->l2_bits;
|
197 |
bs->total_sectors = header.size / 512;
|
198 |
s->csize_shift = (62 - (s->cluster_bits - 8)); |
199 |
s->csize_mask = (1 << (s->cluster_bits - 8)) - 1; |
200 |
s->cluster_offset_mask = (1LL << s->csize_shift) - 1; |
201 |
s->refcount_table_offset = header.refcount_table_offset; |
202 |
s->refcount_table_size = |
203 |
header.refcount_table_clusters << (s->cluster_bits - 3);
|
204 |
|
205 |
s->snapshots_offset = header.snapshots_offset; |
206 |
s->nb_snapshots = header.nb_snapshots; |
207 |
|
208 |
/* read the level 1 table */
|
209 |
s->l1_size = header.l1_size; |
210 |
s->l1_vm_state_index = size_to_l1(s, header.size); |
211 |
/* the L1 table must contain at least enough entries to put
|
212 |
header.size bytes */
|
213 |
if (s->l1_size < s->l1_vm_state_index) {
|
214 |
ret = -EINVAL; |
215 |
goto fail;
|
216 |
} |
217 |
s->l1_table_offset = header.l1_table_offset; |
218 |
if (s->l1_size > 0) { |
219 |
s->l1_table = qemu_mallocz( |
220 |
align_offset(s->l1_size * sizeof(uint64_t), 512)); |
221 |
ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table, |
222 |
s->l1_size * sizeof(uint64_t));
|
223 |
if (ret < 0) { |
224 |
goto fail;
|
225 |
} |
226 |
for(i = 0;i < s->l1_size; i++) { |
227 |
be64_to_cpus(&s->l1_table[i]); |
228 |
} |
229 |
} |
230 |
|
231 |
/* alloc L2 table/refcount block cache */
|
232 |
writethrough = ((flags & BDRV_O_CACHE_WB) == 0);
|
233 |
s->l2_table_cache = qcow2_cache_create(bs, L2_CACHE_SIZE, writethrough); |
234 |
s->refcount_block_cache = qcow2_cache_create(bs, REFCOUNT_CACHE_SIZE, |
235 |
writethrough); |
236 |
|
237 |
s->cluster_cache = qemu_malloc(s->cluster_size); |
238 |
/* one more sector for decompressed data alignment */
|
239 |
s->cluster_data = qemu_malloc(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size |
240 |
+ 512);
|
241 |
s->cluster_cache_offset = -1;
|
242 |
|
243 |
ret = qcow2_refcount_init(bs); |
244 |
if (ret != 0) { |
245 |
goto fail;
|
246 |
} |
247 |
|
248 |
QLIST_INIT(&s->cluster_allocs); |
249 |
|
250 |
/* read qcow2 extensions */
|
251 |
if (header.backing_file_offset) {
|
252 |
ext_end = header.backing_file_offset; |
253 |
} else {
|
254 |
ext_end = s->cluster_size; |
255 |
} |
256 |
if (qcow2_read_extensions(bs, sizeof(header), ext_end)) { |
257 |
ret = -EINVAL; |
258 |
goto fail;
|
259 |
} |
260 |
|
261 |
/* read the backing file name */
|
262 |
if (header.backing_file_offset != 0) { |
263 |
len = header.backing_file_size; |
264 |
if (len > 1023) { |
265 |
len = 1023;
|
266 |
} |
267 |
ret = bdrv_pread(bs->file, header.backing_file_offset, |
268 |
bs->backing_file, len); |
269 |
if (ret < 0) { |
270 |
goto fail;
|
271 |
} |
272 |
bs->backing_file[len] = '\0';
|
273 |
} |
274 |
if (qcow2_read_snapshots(bs) < 0) { |
275 |
ret = -EINVAL; |
276 |
goto fail;
|
277 |
} |
278 |
|
279 |
#ifdef DEBUG_ALLOC
|
280 |
qcow2_check_refcounts(bs); |
281 |
#endif
|
282 |
return ret;
|
283 |
|
284 |
fail:
|
285 |
qcow2_free_snapshots(bs); |
286 |
qcow2_refcount_close(bs); |
287 |
qemu_free(s->l1_table); |
288 |
if (s->l2_table_cache) {
|
289 |
qcow2_cache_destroy(bs, s->l2_table_cache); |
290 |
} |
291 |
qemu_free(s->cluster_cache); |
292 |
qemu_free(s->cluster_data); |
293 |
return ret;
|
294 |
} |
295 |
|
296 |
static int qcow2_set_key(BlockDriverState *bs, const char *key) |
297 |
{ |
298 |
BDRVQcowState *s = bs->opaque; |
299 |
uint8_t keybuf[16];
|
300 |
int len, i;
|
301 |
|
302 |
memset(keybuf, 0, 16); |
303 |
len = strlen(key); |
304 |
if (len > 16) |
305 |
len = 16;
|
306 |
/* XXX: we could compress the chars to 7 bits to increase
|
307 |
entropy */
|
308 |
for(i = 0;i < len;i++) { |
309 |
keybuf[i] = key[i]; |
310 |
} |
311 |
s->crypt_method = s->crypt_method_header; |
312 |
|
313 |
if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0) |
314 |
return -1; |
315 |
if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0) |
316 |
return -1; |
317 |
#if 0
|
318 |
/* test */
|
319 |
{
|
320 |
uint8_t in[16];
|
321 |
uint8_t out[16];
|
322 |
uint8_t tmp[16];
|
323 |
for(i=0;i<16;i++)
|
324 |
in[i] = i;
|
325 |
AES_encrypt(in, tmp, &s->aes_encrypt_key);
|
326 |
AES_decrypt(tmp, out, &s->aes_decrypt_key);
|
327 |
for(i = 0; i < 16; i++)
|
328 |
printf(" %02x", tmp[i]);
|
329 |
printf("\n");
|
330 |
for(i = 0; i < 16; i++)
|
331 |
printf(" %02x", out[i]);
|
332 |
printf("\n");
|
333 |
}
|
334 |
#endif
|
335 |
return 0; |
336 |
} |
337 |
|
338 |
static int qcow2_is_allocated(BlockDriverState *bs, int64_t sector_num, |
339 |
int nb_sectors, int *pnum) |
340 |
{ |
341 |
uint64_t cluster_offset; |
342 |
int ret;
|
343 |
|
344 |
*pnum = nb_sectors; |
345 |
/* FIXME We can get errors here, but the bdrv_is_allocated interface can't
|
346 |
* pass them on today */
|
347 |
ret = qcow2_get_cluster_offset(bs, sector_num << 9, pnum, &cluster_offset);
|
348 |
if (ret < 0) { |
349 |
*pnum = 0;
|
350 |
} |
351 |
|
352 |
return (cluster_offset != 0); |
353 |
} |
354 |
|
355 |
/* handle reading after the end of the backing file */
|
356 |
int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
|
357 |
int64_t sector_num, int nb_sectors)
|
358 |
{ |
359 |
int n1;
|
360 |
if ((sector_num + nb_sectors) <= bs->total_sectors)
|
361 |
return nb_sectors;
|
362 |
if (sector_num >= bs->total_sectors)
|
363 |
n1 = 0;
|
364 |
else
|
365 |
n1 = bs->total_sectors - sector_num; |
366 |
|
367 |
qemu_iovec_memset_skip(qiov, 0, 512 * (nb_sectors - n1), 512 * n1); |
368 |
|
369 |
return n1;
|
370 |
} |
371 |
|
372 |
typedef struct QCowAIOCB { |
373 |
BlockDriverAIOCB common; |
374 |
int64_t sector_num; |
375 |
QEMUIOVector *qiov; |
376 |
int remaining_sectors;
|
377 |
int cur_nr_sectors; /* number of sectors in current iteration */ |
378 |
uint64_t bytes_done; |
379 |
uint64_t cluster_offset; |
380 |
uint8_t *cluster_data; |
381 |
BlockDriverAIOCB *hd_aiocb; |
382 |
QEMUIOVector hd_qiov; |
383 |
QEMUBH *bh; |
384 |
QCowL2Meta l2meta; |
385 |
QLIST_ENTRY(QCowAIOCB) next_depend; |
386 |
} QCowAIOCB; |
387 |
|
388 |
static void qcow2_aio_cancel(BlockDriverAIOCB *blockacb) |
389 |
{ |
390 |
QCowAIOCB *acb = container_of(blockacb, QCowAIOCB, common); |
391 |
if (acb->hd_aiocb)
|
392 |
bdrv_aio_cancel(acb->hd_aiocb); |
393 |
qemu_aio_release(acb); |
394 |
} |
395 |
|
396 |
static AIOPool qcow2_aio_pool = {
|
397 |
.aiocb_size = sizeof(QCowAIOCB),
|
398 |
.cancel = qcow2_aio_cancel, |
399 |
}; |
400 |
|
401 |
static void qcow2_aio_read_cb(void *opaque, int ret); |
402 |
static void qcow2_aio_read_bh(void *opaque) |
403 |
{ |
404 |
QCowAIOCB *acb = opaque; |
405 |
qemu_bh_delete(acb->bh); |
406 |
acb->bh = NULL;
|
407 |
qcow2_aio_read_cb(opaque, 0);
|
408 |
} |
409 |
|
410 |
static int qcow2_schedule_bh(QEMUBHFunc *cb, QCowAIOCB *acb) |
411 |
{ |
412 |
if (acb->bh)
|
413 |
return -EIO;
|
414 |
|
415 |
acb->bh = qemu_bh_new(cb, acb); |
416 |
if (!acb->bh)
|
417 |
return -EIO;
|
418 |
|
419 |
qemu_bh_schedule(acb->bh); |
420 |
|
421 |
return 0; |
422 |
} |
423 |
|
424 |
static void qcow2_aio_read_cb(void *opaque, int ret) |
425 |
{ |
426 |
QCowAIOCB *acb = opaque; |
427 |
BlockDriverState *bs = acb->common.bs; |
428 |
BDRVQcowState *s = bs->opaque; |
429 |
int index_in_cluster, n1;
|
430 |
|
431 |
acb->hd_aiocb = NULL;
|
432 |
if (ret < 0) |
433 |
goto done;
|
434 |
|
435 |
/* post process the read buffer */
|
436 |
if (!acb->cluster_offset) {
|
437 |
/* nothing to do */
|
438 |
} else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) { |
439 |
/* nothing to do */
|
440 |
} else {
|
441 |
if (s->crypt_method) {
|
442 |
qcow2_encrypt_sectors(s, acb->sector_num, acb->cluster_data, |
443 |
acb->cluster_data, acb->cur_nr_sectors, 0, &s->aes_decrypt_key);
|
444 |
qemu_iovec_reset(&acb->hd_qiov); |
445 |
qemu_iovec_copy(&acb->hd_qiov, acb->qiov, acb->bytes_done, |
446 |
acb->cur_nr_sectors * 512);
|
447 |
qemu_iovec_from_buffer(&acb->hd_qiov, acb->cluster_data, |
448 |
512 * acb->cur_nr_sectors);
|
449 |
} |
450 |
} |
451 |
|
452 |
acb->remaining_sectors -= acb->cur_nr_sectors; |
453 |
acb->sector_num += acb->cur_nr_sectors; |
454 |
acb->bytes_done += acb->cur_nr_sectors * 512;
|
455 |
|
456 |
if (acb->remaining_sectors == 0) { |
457 |
/* request completed */
|
458 |
ret = 0;
|
459 |
goto done;
|
460 |
} |
461 |
|
462 |
/* prepare next AIO request */
|
463 |
acb->cur_nr_sectors = acb->remaining_sectors; |
464 |
if (s->crypt_method) {
|
465 |
acb->cur_nr_sectors = MIN(acb->cur_nr_sectors, |
466 |
QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors); |
467 |
} |
468 |
|
469 |
ret = qcow2_get_cluster_offset(bs, acb->sector_num << 9,
|
470 |
&acb->cur_nr_sectors, &acb->cluster_offset); |
471 |
if (ret < 0) { |
472 |
goto done;
|
473 |
} |
474 |
|
475 |
index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
|
476 |
|
477 |
qemu_iovec_reset(&acb->hd_qiov); |
478 |
qemu_iovec_copy(&acb->hd_qiov, acb->qiov, acb->bytes_done, |
479 |
acb->cur_nr_sectors * 512);
|
480 |
|
481 |
if (!acb->cluster_offset) {
|
482 |
|
483 |
if (bs->backing_hd) {
|
484 |
/* read from the base image */
|
485 |
n1 = qcow2_backing_read1(bs->backing_hd, &acb->hd_qiov, |
486 |
acb->sector_num, acb->cur_nr_sectors); |
487 |
if (n1 > 0) { |
488 |
BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO); |
489 |
acb->hd_aiocb = bdrv_aio_readv(bs->backing_hd, acb->sector_num, |
490 |
&acb->hd_qiov, n1, qcow2_aio_read_cb, acb); |
491 |
if (acb->hd_aiocb == NULL) { |
492 |
ret = -EIO; |
493 |
goto done;
|
494 |
} |
495 |
} else {
|
496 |
ret = qcow2_schedule_bh(qcow2_aio_read_bh, acb); |
497 |
if (ret < 0) |
498 |
goto done;
|
499 |
} |
500 |
} else {
|
501 |
/* Note: in this case, no need to wait */
|
502 |
qemu_iovec_memset(&acb->hd_qiov, 0, 512 * acb->cur_nr_sectors); |
503 |
ret = qcow2_schedule_bh(qcow2_aio_read_bh, acb); |
504 |
if (ret < 0) |
505 |
goto done;
|
506 |
} |
507 |
} else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) { |
508 |
/* add AIO support for compressed blocks ? */
|
509 |
ret = qcow2_decompress_cluster(bs, acb->cluster_offset); |
510 |
if (ret < 0) { |
511 |
goto done;
|
512 |
} |
513 |
|
514 |
qemu_iovec_from_buffer(&acb->hd_qiov, |
515 |
s->cluster_cache + index_in_cluster * 512,
|
516 |
512 * acb->cur_nr_sectors);
|
517 |
|
518 |
ret = qcow2_schedule_bh(qcow2_aio_read_bh, acb); |
519 |
if (ret < 0) |
520 |
goto done;
|
521 |
} else {
|
522 |
if ((acb->cluster_offset & 511) != 0) { |
523 |
ret = -EIO; |
524 |
goto done;
|
525 |
} |
526 |
|
527 |
if (s->crypt_method) {
|
528 |
/*
|
529 |
* For encrypted images, read everything into a temporary
|
530 |
* contiguous buffer on which the AES functions can work.
|
531 |
*/
|
532 |
if (!acb->cluster_data) {
|
533 |
acb->cluster_data = |
534 |
qemu_mallocz(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); |
535 |
} |
536 |
|
537 |
assert(acb->cur_nr_sectors <= |
538 |
QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors); |
539 |
qemu_iovec_reset(&acb->hd_qiov); |
540 |
qemu_iovec_add(&acb->hd_qiov, acb->cluster_data, |
541 |
512 * acb->cur_nr_sectors);
|
542 |
} |
543 |
|
544 |
BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); |
545 |
acb->hd_aiocb = bdrv_aio_readv(bs->file, |
546 |
(acb->cluster_offset >> 9) + index_in_cluster,
|
547 |
&acb->hd_qiov, acb->cur_nr_sectors, |
548 |
qcow2_aio_read_cb, acb); |
549 |
if (acb->hd_aiocb == NULL) { |
550 |
ret = -EIO; |
551 |
goto done;
|
552 |
} |
553 |
} |
554 |
|
555 |
return;
|
556 |
done:
|
557 |
acb->common.cb(acb->common.opaque, ret); |
558 |
qemu_iovec_destroy(&acb->hd_qiov); |
559 |
qemu_aio_release(acb); |
560 |
} |
561 |
|
562 |
static QCowAIOCB *qcow2_aio_setup(BlockDriverState *bs, int64_t sector_num,
|
563 |
QEMUIOVector *qiov, int nb_sectors,
|
564 |
BlockDriverCompletionFunc *cb, |
565 |
void *opaque, int is_write) |
566 |
{ |
567 |
QCowAIOCB *acb; |
568 |
|
569 |
acb = qemu_aio_get(&qcow2_aio_pool, bs, cb, opaque); |
570 |
if (!acb)
|
571 |
return NULL; |
572 |
acb->hd_aiocb = NULL;
|
573 |
acb->sector_num = sector_num; |
574 |
acb->qiov = qiov; |
575 |
|
576 |
qemu_iovec_init(&acb->hd_qiov, qiov->niov); |
577 |
|
578 |
acb->bytes_done = 0;
|
579 |
acb->remaining_sectors = nb_sectors; |
580 |
acb->cur_nr_sectors = 0;
|
581 |
acb->cluster_offset = 0;
|
582 |
acb->l2meta.nb_clusters = 0;
|
583 |
QLIST_INIT(&acb->l2meta.dependent_requests); |
584 |
return acb;
|
585 |
} |
586 |
|
587 |
static BlockDriverAIOCB *qcow2_aio_readv(BlockDriverState *bs,
|
588 |
int64_t sector_num, |
589 |
QEMUIOVector *qiov, int nb_sectors,
|
590 |
BlockDriverCompletionFunc *cb, |
591 |
void *opaque)
|
592 |
{ |
593 |
QCowAIOCB *acb; |
594 |
|
595 |
acb = qcow2_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
|
596 |
if (!acb)
|
597 |
return NULL; |
598 |
|
599 |
qcow2_aio_read_cb(acb, 0);
|
600 |
return &acb->common;
|
601 |
} |
602 |
|
603 |
static void qcow2_aio_write_cb(void *opaque, int ret); |
604 |
|
605 |
static void run_dependent_requests(QCowL2Meta *m) |
606 |
{ |
607 |
QCowAIOCB *req; |
608 |
QCowAIOCB *next; |
609 |
|
610 |
/* Take the request off the list of running requests */
|
611 |
if (m->nb_clusters != 0) { |
612 |
QLIST_REMOVE(m, next_in_flight); |
613 |
} |
614 |
|
615 |
/* Restart all dependent requests */
|
616 |
QLIST_FOREACH_SAFE(req, &m->dependent_requests, next_depend, next) { |
617 |
qcow2_aio_write_cb(req, 0);
|
618 |
} |
619 |
|
620 |
/* Empty the list for the next part of the request */
|
621 |
QLIST_INIT(&m->dependent_requests); |
622 |
} |
623 |
|
624 |
static void qcow2_aio_write_cb(void *opaque, int ret) |
625 |
{ |
626 |
QCowAIOCB *acb = opaque; |
627 |
BlockDriverState *bs = acb->common.bs; |
628 |
BDRVQcowState *s = bs->opaque; |
629 |
int index_in_cluster;
|
630 |
int n_end;
|
631 |
|
632 |
acb->hd_aiocb = NULL;
|
633 |
|
634 |
if (ret >= 0) { |
635 |
ret = qcow2_alloc_cluster_link_l2(bs, &acb->l2meta); |
636 |
} |
637 |
|
638 |
run_dependent_requests(&acb->l2meta); |
639 |
|
640 |
if (ret < 0) |
641 |
goto done;
|
642 |
|
643 |
acb->remaining_sectors -= acb->cur_nr_sectors; |
644 |
acb->sector_num += acb->cur_nr_sectors; |
645 |
acb->bytes_done += acb->cur_nr_sectors * 512;
|
646 |
|
647 |
if (acb->remaining_sectors == 0) { |
648 |
/* request completed */
|
649 |
ret = 0;
|
650 |
goto done;
|
651 |
} |
652 |
|
653 |
index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
|
654 |
n_end = index_in_cluster + acb->remaining_sectors; |
655 |
if (s->crypt_method &&
|
656 |
n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors) |
657 |
n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors; |
658 |
|
659 |
ret = qcow2_alloc_cluster_offset(bs, acb->sector_num << 9,
|
660 |
index_in_cluster, n_end, &acb->cur_nr_sectors, &acb->l2meta); |
661 |
if (ret < 0) { |
662 |
goto done;
|
663 |
} |
664 |
|
665 |
acb->cluster_offset = acb->l2meta.cluster_offset; |
666 |
|
667 |
/* Need to wait for another request? If so, we are done for now. */
|
668 |
if (acb->l2meta.nb_clusters == 0 && acb->l2meta.depends_on != NULL) { |
669 |
QLIST_INSERT_HEAD(&acb->l2meta.depends_on->dependent_requests, |
670 |
acb, next_depend); |
671 |
return;
|
672 |
} |
673 |
|
674 |
assert((acb->cluster_offset & 511) == 0); |
675 |
|
676 |
qemu_iovec_reset(&acb->hd_qiov); |
677 |
qemu_iovec_copy(&acb->hd_qiov, acb->qiov, acb->bytes_done, |
678 |
acb->cur_nr_sectors * 512);
|
679 |
|
680 |
if (s->crypt_method) {
|
681 |
if (!acb->cluster_data) {
|
682 |
acb->cluster_data = qemu_mallocz(QCOW_MAX_CRYPT_CLUSTERS * |
683 |
s->cluster_size); |
684 |
} |
685 |
|
686 |
assert(acb->hd_qiov.size <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); |
687 |
qemu_iovec_to_buffer(&acb->hd_qiov, acb->cluster_data); |
688 |
|
689 |
qcow2_encrypt_sectors(s, acb->sector_num, acb->cluster_data, |
690 |
acb->cluster_data, acb->cur_nr_sectors, 1, &s->aes_encrypt_key);
|
691 |
|
692 |
qemu_iovec_reset(&acb->hd_qiov); |
693 |
qemu_iovec_add(&acb->hd_qiov, acb->cluster_data, |
694 |
acb->cur_nr_sectors * 512);
|
695 |
} |
696 |
|
697 |
BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); |
698 |
acb->hd_aiocb = bdrv_aio_writev(bs->file, |
699 |
(acb->cluster_offset >> 9) + index_in_cluster,
|
700 |
&acb->hd_qiov, acb->cur_nr_sectors, |
701 |
qcow2_aio_write_cb, acb); |
702 |
if (acb->hd_aiocb == NULL) { |
703 |
ret = -EIO; |
704 |
goto fail;
|
705 |
} |
706 |
|
707 |
return;
|
708 |
|
709 |
fail:
|
710 |
if (acb->l2meta.nb_clusters != 0) { |
711 |
QLIST_REMOVE(&acb->l2meta, next_in_flight); |
712 |
} |
713 |
done:
|
714 |
acb->common.cb(acb->common.opaque, ret); |
715 |
qemu_iovec_destroy(&acb->hd_qiov); |
716 |
qemu_aio_release(acb); |
717 |
} |
718 |
|
719 |
static BlockDriverAIOCB *qcow2_aio_writev(BlockDriverState *bs,
|
720 |
int64_t sector_num, |
721 |
QEMUIOVector *qiov, int nb_sectors,
|
722 |
BlockDriverCompletionFunc *cb, |
723 |
void *opaque)
|
724 |
{ |
725 |
BDRVQcowState *s = bs->opaque; |
726 |
QCowAIOCB *acb; |
727 |
|
728 |
s->cluster_cache_offset = -1; /* disable compressed cache */ |
729 |
|
730 |
acb = qcow2_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
|
731 |
if (!acb)
|
732 |
return NULL; |
733 |
|
734 |
qcow2_aio_write_cb(acb, 0);
|
735 |
return &acb->common;
|
736 |
} |
737 |
|
738 |
static void qcow2_close(BlockDriverState *bs) |
739 |
{ |
740 |
BDRVQcowState *s = bs->opaque; |
741 |
qemu_free(s->l1_table); |
742 |
|
743 |
qcow2_cache_flush(bs, s->l2_table_cache); |
744 |
qcow2_cache_flush(bs, s->refcount_block_cache); |
745 |
|
746 |
qcow2_cache_destroy(bs, s->l2_table_cache); |
747 |
qcow2_cache_destroy(bs, s->refcount_block_cache); |
748 |
|
749 |
qemu_free(s->cluster_cache); |
750 |
qemu_free(s->cluster_data); |
751 |
qcow2_refcount_close(bs); |
752 |
} |
753 |
|
754 |
/*
|
755 |
* Updates the variable length parts of the qcow2 header, i.e. the backing file
|
756 |
* name and all extensions. qcow2 was not designed to allow such changes, so if
|
757 |
* we run out of space (we can only use the first cluster) this function may
|
758 |
* fail.
|
759 |
*
|
760 |
* Returns 0 on success, -errno in error cases.
|
761 |
*/
|
762 |
static int qcow2_update_ext_header(BlockDriverState *bs, |
763 |
const char *backing_file, const char *backing_fmt) |
764 |
{ |
765 |
size_t backing_file_len = 0;
|
766 |
size_t backing_fmt_len = 0;
|
767 |
BDRVQcowState *s = bs->opaque; |
768 |
QCowExtension ext_backing_fmt = {0, 0}; |
769 |
int ret;
|
770 |
|
771 |
/* Backing file format doesn't make sense without a backing file */
|
772 |
if (backing_fmt && !backing_file) {
|
773 |
return -EINVAL;
|
774 |
} |
775 |
|
776 |
/* Prepare the backing file format extension if needed */
|
777 |
if (backing_fmt) {
|
778 |
ext_backing_fmt.len = cpu_to_be32(strlen(backing_fmt)); |
779 |
ext_backing_fmt.magic = cpu_to_be32(QCOW2_EXT_MAGIC_BACKING_FORMAT); |
780 |
backing_fmt_len = ((sizeof(ext_backing_fmt)
|
781 |
+ strlen(backing_fmt) + 7) & ~7); |
782 |
} |
783 |
|
784 |
/* Check if we can fit the new header into the first cluster */
|
785 |
if (backing_file) {
|
786 |
backing_file_len = strlen(backing_file); |
787 |
} |
788 |
|
789 |
size_t header_size = sizeof(QCowHeader) + backing_file_len
|
790 |
+ backing_fmt_len; |
791 |
|
792 |
if (header_size > s->cluster_size) {
|
793 |
return -ENOSPC;
|
794 |
} |
795 |
|
796 |
/* Rewrite backing file name and qcow2 extensions */
|
797 |
size_t ext_size = header_size - sizeof(QCowHeader);
|
798 |
uint8_t buf[ext_size]; |
799 |
size_t offset = 0;
|
800 |
size_t backing_file_offset = 0;
|
801 |
|
802 |
if (backing_file) {
|
803 |
if (backing_fmt) {
|
804 |
int padding = backing_fmt_len -
|
805 |
(sizeof(ext_backing_fmt) + strlen(backing_fmt));
|
806 |
|
807 |
memcpy(buf + offset, &ext_backing_fmt, sizeof(ext_backing_fmt));
|
808 |
offset += sizeof(ext_backing_fmt);
|
809 |
|
810 |
memcpy(buf + offset, backing_fmt, strlen(backing_fmt)); |
811 |
offset += strlen(backing_fmt); |
812 |
|
813 |
memset(buf + offset, 0, padding);
|
814 |
offset += padding; |
815 |
} |
816 |
|
817 |
memcpy(buf + offset, backing_file, backing_file_len); |
818 |
backing_file_offset = sizeof(QCowHeader) + offset;
|
819 |
} |
820 |
|
821 |
ret = bdrv_pwrite_sync(bs->file, sizeof(QCowHeader), buf, ext_size);
|
822 |
if (ret < 0) { |
823 |
goto fail;
|
824 |
} |
825 |
|
826 |
/* Update header fields */
|
827 |
uint64_t be_backing_file_offset = cpu_to_be64(backing_file_offset); |
828 |
uint32_t be_backing_file_size = cpu_to_be32(backing_file_len); |
829 |
|
830 |
ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, backing_file_offset), |
831 |
&be_backing_file_offset, sizeof(uint64_t));
|
832 |
if (ret < 0) { |
833 |
goto fail;
|
834 |
} |
835 |
|
836 |
ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, backing_file_size), |
837 |
&be_backing_file_size, sizeof(uint32_t));
|
838 |
if (ret < 0) { |
839 |
goto fail;
|
840 |
} |
841 |
|
842 |
ret = 0;
|
843 |
fail:
|
844 |
return ret;
|
845 |
} |
846 |
|
847 |
static int qcow2_change_backing_file(BlockDriverState *bs, |
848 |
const char *backing_file, const char *backing_fmt) |
849 |
{ |
850 |
return qcow2_update_ext_header(bs, backing_file, backing_fmt);
|
851 |
} |
852 |
|
853 |
static int preallocate(BlockDriverState *bs) |
854 |
{ |
855 |
uint64_t nb_sectors; |
856 |
uint64_t offset; |
857 |
int num;
|
858 |
int ret;
|
859 |
QCowL2Meta meta; |
860 |
|
861 |
nb_sectors = bdrv_getlength(bs) >> 9;
|
862 |
offset = 0;
|
863 |
QLIST_INIT(&meta.dependent_requests); |
864 |
meta.cluster_offset = 0;
|
865 |
|
866 |
while (nb_sectors) {
|
867 |
num = MIN(nb_sectors, INT_MAX >> 9);
|
868 |
ret = qcow2_alloc_cluster_offset(bs, offset, 0, num, &num, &meta);
|
869 |
if (ret < 0) { |
870 |
return ret;
|
871 |
} |
872 |
|
873 |
ret = qcow2_alloc_cluster_link_l2(bs, &meta); |
874 |
if (ret < 0) { |
875 |
qcow2_free_any_clusters(bs, meta.cluster_offset, meta.nb_clusters); |
876 |
return ret;
|
877 |
} |
878 |
|
879 |
/* There are no dependent requests, but we need to remove our request
|
880 |
* from the list of in-flight requests */
|
881 |
run_dependent_requests(&meta); |
882 |
|
883 |
/* TODO Preallocate data if requested */
|
884 |
|
885 |
nb_sectors -= num; |
886 |
offset += num << 9;
|
887 |
} |
888 |
|
889 |
/*
|
890 |
* It is expected that the image file is large enough to actually contain
|
891 |
* all of the allocated clusters (otherwise we get failing reads after
|
892 |
* EOF). Extend the image to the last allocated sector.
|
893 |
*/
|
894 |
if (meta.cluster_offset != 0) { |
895 |
uint8_t buf[512];
|
896 |
memset(buf, 0, 512); |
897 |
ret = bdrv_write(bs->file, (meta.cluster_offset >> 9) + num - 1, buf, 1); |
898 |
if (ret < 0) { |
899 |
return ret;
|
900 |
} |
901 |
} |
902 |
|
903 |
return 0; |
904 |
} |
905 |
|
906 |
static int qcow2_create2(const char *filename, int64_t total_size, |
907 |
const char *backing_file, const char *backing_format, |
908 |
int flags, size_t cluster_size, int prealloc, |
909 |
QEMUOptionParameter *options) |
910 |
{ |
911 |
/* Calulate cluster_bits */
|
912 |
int cluster_bits;
|
913 |
cluster_bits = ffs(cluster_size) - 1;
|
914 |
if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
|
915 |
(1 << cluster_bits) != cluster_size)
|
916 |
{ |
917 |
error_report( |
918 |
"Cluster size must be a power of two between %d and %dk\n",
|
919 |
1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10)); |
920 |
return -EINVAL;
|
921 |
} |
922 |
|
923 |
/*
|
924 |
* Open the image file and write a minimal qcow2 header.
|
925 |
*
|
926 |
* We keep things simple and start with a zero-sized image. We also
|
927 |
* do without refcount blocks or a L1 table for now. We'll fix the
|
928 |
* inconsistency later.
|
929 |
*
|
930 |
* We do need a refcount table because growing the refcount table means
|
931 |
* allocating two new refcount blocks - the seconds of which would be at
|
932 |
* 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
|
933 |
* size for any qcow2 image.
|
934 |
*/
|
935 |
BlockDriverState* bs; |
936 |
QCowHeader header; |
937 |
uint8_t* refcount_table; |
938 |
int ret;
|
939 |
|
940 |
ret = bdrv_create_file(filename, options); |
941 |
if (ret < 0) { |
942 |
return ret;
|
943 |
} |
944 |
|
945 |
ret = bdrv_file_open(&bs, filename, BDRV_O_RDWR); |
946 |
if (ret < 0) { |
947 |
return ret;
|
948 |
} |
949 |
|
950 |
/* Write the header */
|
951 |
memset(&header, 0, sizeof(header)); |
952 |
header.magic = cpu_to_be32(QCOW_MAGIC); |
953 |
header.version = cpu_to_be32(QCOW_VERSION); |
954 |
header.cluster_bits = cpu_to_be32(cluster_bits); |
955 |
header.size = cpu_to_be64(0);
|
956 |
header.l1_table_offset = cpu_to_be64(0);
|
957 |
header.l1_size = cpu_to_be32(0);
|
958 |
header.refcount_table_offset = cpu_to_be64(cluster_size); |
959 |
header.refcount_table_clusters = cpu_to_be32(1);
|
960 |
|
961 |
if (flags & BLOCK_FLAG_ENCRYPT) {
|
962 |
header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES); |
963 |
} else {
|
964 |
header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE); |
965 |
} |
966 |
|
967 |
ret = bdrv_pwrite(bs, 0, &header, sizeof(header)); |
968 |
if (ret < 0) { |
969 |
goto out;
|
970 |
} |
971 |
|
972 |
/* Write an empty refcount table */
|
973 |
refcount_table = qemu_mallocz(cluster_size); |
974 |
ret = bdrv_pwrite(bs, cluster_size, refcount_table, cluster_size); |
975 |
qemu_free(refcount_table); |
976 |
|
977 |
if (ret < 0) { |
978 |
goto out;
|
979 |
} |
980 |
|
981 |
bdrv_close(bs); |
982 |
|
983 |
/*
|
984 |
* And now open the image and make it consistent first (i.e. increase the
|
985 |
* refcount of the cluster that is occupied by the header and the refcount
|
986 |
* table)
|
987 |
*/
|
988 |
BlockDriver* drv = bdrv_find_format("qcow2");
|
989 |
assert(drv != NULL);
|
990 |
ret = bdrv_open(bs, filename, |
991 |
BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_FLUSH, drv); |
992 |
if (ret < 0) { |
993 |
goto out;
|
994 |
} |
995 |
|
996 |
ret = qcow2_alloc_clusters(bs, 2 * cluster_size);
|
997 |
if (ret < 0) { |
998 |
goto out;
|
999 |
|
1000 |
} else if (ret != 0) { |
1001 |
error_report("Huh, first cluster in empty image is already in use?");
|
1002 |
abort(); |
1003 |
} |
1004 |
|
1005 |
/* Okay, now that we have a valid image, let's give it the right size */
|
1006 |
ret = bdrv_truncate(bs, total_size * BDRV_SECTOR_SIZE); |
1007 |
if (ret < 0) { |
1008 |
goto out;
|
1009 |
} |
1010 |
|
1011 |
/* Want a backing file? There you go.*/
|
1012 |
if (backing_file) {
|
1013 |
ret = bdrv_change_backing_file(bs, backing_file, backing_format); |
1014 |
if (ret < 0) { |
1015 |
goto out;
|
1016 |
} |
1017 |
} |
1018 |
|
1019 |
/* And if we're supposed to preallocate metadata, do that now */
|
1020 |
if (prealloc) {
|
1021 |
ret = preallocate(bs); |
1022 |
if (ret < 0) { |
1023 |
goto out;
|
1024 |
} |
1025 |
} |
1026 |
|
1027 |
ret = 0;
|
1028 |
out:
|
1029 |
bdrv_delete(bs); |
1030 |
return ret;
|
1031 |
} |
1032 |
|
1033 |
static int qcow2_create(const char *filename, QEMUOptionParameter *options) |
1034 |
{ |
1035 |
const char *backing_file = NULL; |
1036 |
const char *backing_fmt = NULL; |
1037 |
uint64_t sectors = 0;
|
1038 |
int flags = 0; |
1039 |
size_t cluster_size = DEFAULT_CLUSTER_SIZE; |
1040 |
int prealloc = 0; |
1041 |
|
1042 |
/* Read out options */
|
1043 |
while (options && options->name) {
|
1044 |
if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
|
1045 |
sectors = options->value.n / 512;
|
1046 |
} else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) { |
1047 |
backing_file = options->value.s; |
1048 |
} else if (!strcmp(options->name, BLOCK_OPT_BACKING_FMT)) { |
1049 |
backing_fmt = options->value.s; |
1050 |
} else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) { |
1051 |
flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
|
1052 |
} else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) { |
1053 |
if (options->value.n) {
|
1054 |
cluster_size = options->value.n; |
1055 |
} |
1056 |
} else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) { |
1057 |
if (!options->value.s || !strcmp(options->value.s, "off")) { |
1058 |
prealloc = 0;
|
1059 |
} else if (!strcmp(options->value.s, "metadata")) { |
1060 |
prealloc = 1;
|
1061 |
} else {
|
1062 |
fprintf(stderr, "Invalid preallocation mode: '%s'\n",
|
1063 |
options->value.s); |
1064 |
return -EINVAL;
|
1065 |
} |
1066 |
} |
1067 |
options++; |
1068 |
} |
1069 |
|
1070 |
if (backing_file && prealloc) {
|
1071 |
fprintf(stderr, "Backing file and preallocation cannot be used at "
|
1072 |
"the same time\n");
|
1073 |
return -EINVAL;
|
1074 |
} |
1075 |
|
1076 |
return qcow2_create2(filename, sectors, backing_file, backing_fmt, flags,
|
1077 |
cluster_size, prealloc, options); |
1078 |
} |
1079 |
|
1080 |
static int qcow2_make_empty(BlockDriverState *bs) |
1081 |
{ |
1082 |
#if 0
|
1083 |
/* XXX: not correct */
|
1084 |
BDRVQcowState *s = bs->opaque;
|
1085 |
uint32_t l1_length = s->l1_size * sizeof(uint64_t);
|
1086 |
int ret;
|
1087 |
|
1088 |
memset(s->l1_table, 0, l1_length);
|
1089 |
if (bdrv_pwrite(bs->file, s->l1_table_offset, s->l1_table, l1_length) < 0)
|
1090 |
return -1;
|
1091 |
ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length);
|
1092 |
if (ret < 0)
|
1093 |
return ret;
|
1094 |
|
1095 |
l2_cache_reset(bs);
|
1096 |
#endif
|
1097 |
return 0; |
1098 |
} |
1099 |
|
1100 |
static int qcow2_discard(BlockDriverState *bs, int64_t sector_num, |
1101 |
int nb_sectors)
|
1102 |
{ |
1103 |
return qcow2_discard_clusters(bs, sector_num << BDRV_SECTOR_BITS,
|
1104 |
nb_sectors); |
1105 |
} |
1106 |
|
1107 |
static int qcow2_truncate(BlockDriverState *bs, int64_t offset) |
1108 |
{ |
1109 |
BDRVQcowState *s = bs->opaque; |
1110 |
int ret, new_l1_size;
|
1111 |
|
1112 |
if (offset & 511) { |
1113 |
return -EINVAL;
|
1114 |
} |
1115 |
|
1116 |
/* cannot proceed if image has snapshots */
|
1117 |
if (s->nb_snapshots) {
|
1118 |
return -ENOTSUP;
|
1119 |
} |
1120 |
|
1121 |
/* shrinking is currently not supported */
|
1122 |
if (offset < bs->total_sectors * 512) { |
1123 |
return -ENOTSUP;
|
1124 |
} |
1125 |
|
1126 |
new_l1_size = size_to_l1(s, offset); |
1127 |
ret = qcow2_grow_l1_table(bs, new_l1_size, true);
|
1128 |
if (ret < 0) { |
1129 |
return ret;
|
1130 |
} |
1131 |
|
1132 |
/* write updated header.size */
|
1133 |
offset = cpu_to_be64(offset); |
1134 |
ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size), |
1135 |
&offset, sizeof(uint64_t));
|
1136 |
if (ret < 0) { |
1137 |
return ret;
|
1138 |
} |
1139 |
|
1140 |
s->l1_vm_state_index = new_l1_size; |
1141 |
return 0; |
1142 |
} |
1143 |
|
1144 |
/* XXX: put compressed sectors first, then all the cluster aligned
|
1145 |
tables to avoid losing bytes in alignment */
|
1146 |
static int qcow2_write_compressed(BlockDriverState *bs, int64_t sector_num, |
1147 |
const uint8_t *buf, int nb_sectors) |
1148 |
{ |
1149 |
BDRVQcowState *s = bs->opaque; |
1150 |
z_stream strm; |
1151 |
int ret, out_len;
|
1152 |
uint8_t *out_buf; |
1153 |
uint64_t cluster_offset; |
1154 |
|
1155 |
if (nb_sectors == 0) { |
1156 |
/* align end of file to a sector boundary to ease reading with
|
1157 |
sector based I/Os */
|
1158 |
cluster_offset = bdrv_getlength(bs->file); |
1159 |
cluster_offset = (cluster_offset + 511) & ~511; |
1160 |
bdrv_truncate(bs->file, cluster_offset); |
1161 |
return 0; |
1162 |
} |
1163 |
|
1164 |
if (nb_sectors != s->cluster_sectors)
|
1165 |
return -EINVAL;
|
1166 |
|
1167 |
out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128); |
1168 |
|
1169 |
/* best compression, small window, no zlib header */
|
1170 |
memset(&strm, 0, sizeof(strm)); |
1171 |
ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION, |
1172 |
Z_DEFLATED, -12,
|
1173 |
9, Z_DEFAULT_STRATEGY);
|
1174 |
if (ret != 0) { |
1175 |
qemu_free(out_buf); |
1176 |
return -1; |
1177 |
} |
1178 |
|
1179 |
strm.avail_in = s->cluster_size; |
1180 |
strm.next_in = (uint8_t *)buf; |
1181 |
strm.avail_out = s->cluster_size; |
1182 |
strm.next_out = out_buf; |
1183 |
|
1184 |
ret = deflate(&strm, Z_FINISH); |
1185 |
if (ret != Z_STREAM_END && ret != Z_OK) {
|
1186 |
qemu_free(out_buf); |
1187 |
deflateEnd(&strm); |
1188 |
return -1; |
1189 |
} |
1190 |
out_len = strm.next_out - out_buf; |
1191 |
|
1192 |
deflateEnd(&strm); |
1193 |
|
1194 |
if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
|
1195 |
/* could not compress: write normal cluster */
|
1196 |
bdrv_write(bs, sector_num, buf, s->cluster_sectors); |
1197 |
} else {
|
1198 |
cluster_offset = qcow2_alloc_compressed_cluster_offset(bs, |
1199 |
sector_num << 9, out_len);
|
1200 |
if (!cluster_offset)
|
1201 |
return -1; |
1202 |
cluster_offset &= s->cluster_offset_mask; |
1203 |
BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED); |
1204 |
if (bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len) != out_len) {
|
1205 |
qemu_free(out_buf); |
1206 |
return -1; |
1207 |
} |
1208 |
} |
1209 |
|
1210 |
qemu_free(out_buf); |
1211 |
return 0; |
1212 |
} |
1213 |
|
1214 |
static int qcow2_flush(BlockDriverState *bs) |
1215 |
{ |
1216 |
BDRVQcowState *s = bs->opaque; |
1217 |
int ret;
|
1218 |
|
1219 |
ret = qcow2_cache_flush(bs, s->l2_table_cache); |
1220 |
if (ret < 0) { |
1221 |
return ret;
|
1222 |
} |
1223 |
|
1224 |
ret = qcow2_cache_flush(bs, s->refcount_block_cache); |
1225 |
if (ret < 0) { |
1226 |
return ret;
|
1227 |
} |
1228 |
|
1229 |
return bdrv_flush(bs->file);
|
1230 |
} |
1231 |
|
1232 |
static BlockDriverAIOCB *qcow2_aio_flush(BlockDriverState *bs,
|
1233 |
BlockDriverCompletionFunc *cb, |
1234 |
void *opaque)
|
1235 |
{ |
1236 |
BDRVQcowState *s = bs->opaque; |
1237 |
int ret;
|
1238 |
|
1239 |
ret = qcow2_cache_flush(bs, s->l2_table_cache); |
1240 |
if (ret < 0) { |
1241 |
return NULL; |
1242 |
} |
1243 |
|
1244 |
ret = qcow2_cache_flush(bs, s->refcount_block_cache); |
1245 |
if (ret < 0) { |
1246 |
return NULL; |
1247 |
} |
1248 |
|
1249 |
return bdrv_aio_flush(bs->file, cb, opaque);
|
1250 |
} |
1251 |
|
1252 |
static int64_t qcow2_vm_state_offset(BDRVQcowState *s)
|
1253 |
{ |
1254 |
return (int64_t)s->l1_vm_state_index << (s->cluster_bits + s->l2_bits);
|
1255 |
} |
1256 |
|
1257 |
static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) |
1258 |
{ |
1259 |
BDRVQcowState *s = bs->opaque; |
1260 |
bdi->cluster_size = s->cluster_size; |
1261 |
bdi->vm_state_offset = qcow2_vm_state_offset(s); |
1262 |
return 0; |
1263 |
} |
1264 |
|
1265 |
|
1266 |
static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result) |
1267 |
{ |
1268 |
return qcow2_check_refcounts(bs, result);
|
1269 |
} |
1270 |
|
1271 |
#if 0
|
1272 |
static void dump_refcounts(BlockDriverState *bs)
|
1273 |
{
|
1274 |
BDRVQcowState *s = bs->opaque;
|
1275 |
int64_t nb_clusters, k, k1, size;
|
1276 |
int refcount;
|
1277 |
|
1278 |
size = bdrv_getlength(bs->file);
|
1279 |
nb_clusters = size_to_clusters(s, size);
|
1280 |
for(k = 0; k < nb_clusters;) {
|
1281 |
k1 = k;
|
1282 |
refcount = get_refcount(bs, k);
|
1283 |
k++;
|
1284 |
while (k < nb_clusters && get_refcount(bs, k) == refcount)
|
1285 |
k++;
|
1286 |
printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount,
|
1287 |
k - k1);
|
1288 |
}
|
1289 |
}
|
1290 |
#endif
|
1291 |
|
1292 |
static int qcow2_save_vmstate(BlockDriverState *bs, const uint8_t *buf, |
1293 |
int64_t pos, int size)
|
1294 |
{ |
1295 |
BDRVQcowState *s = bs->opaque; |
1296 |
int growable = bs->growable;
|
1297 |
int ret;
|
1298 |
|
1299 |
BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE); |
1300 |
bs->growable = 1;
|
1301 |
ret = bdrv_pwrite(bs, qcow2_vm_state_offset(s) + pos, buf, size); |
1302 |
bs->growable = growable; |
1303 |
|
1304 |
return ret;
|
1305 |
} |
1306 |
|
1307 |
static int qcow2_load_vmstate(BlockDriverState *bs, uint8_t *buf, |
1308 |
int64_t pos, int size)
|
1309 |
{ |
1310 |
BDRVQcowState *s = bs->opaque; |
1311 |
int growable = bs->growable;
|
1312 |
int ret;
|
1313 |
|
1314 |
BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD); |
1315 |
bs->growable = 1;
|
1316 |
ret = bdrv_pread(bs, qcow2_vm_state_offset(s) + pos, buf, size); |
1317 |
bs->growable = growable; |
1318 |
|
1319 |
return ret;
|
1320 |
} |
1321 |
|
1322 |
static QEMUOptionParameter qcow2_create_options[] = {
|
1323 |
{ |
1324 |
.name = BLOCK_OPT_SIZE, |
1325 |
.type = OPT_SIZE, |
1326 |
.help = "Virtual disk size"
|
1327 |
}, |
1328 |
{ |
1329 |
.name = BLOCK_OPT_BACKING_FILE, |
1330 |
.type = OPT_STRING, |
1331 |
.help = "File name of a base image"
|
1332 |
}, |
1333 |
{ |
1334 |
.name = BLOCK_OPT_BACKING_FMT, |
1335 |
.type = OPT_STRING, |
1336 |
.help = "Image format of the base image"
|
1337 |
}, |
1338 |
{ |
1339 |
.name = BLOCK_OPT_ENCRYPT, |
1340 |
.type = OPT_FLAG, |
1341 |
.help = "Encrypt the image"
|
1342 |
}, |
1343 |
{ |
1344 |
.name = BLOCK_OPT_CLUSTER_SIZE, |
1345 |
.type = OPT_SIZE, |
1346 |
.help = "qcow2 cluster size",
|
1347 |
.value = { .n = DEFAULT_CLUSTER_SIZE }, |
1348 |
}, |
1349 |
{ |
1350 |
.name = BLOCK_OPT_PREALLOC, |
1351 |
.type = OPT_STRING, |
1352 |
.help = "Preallocation mode (allowed values: off, metadata)"
|
1353 |
}, |
1354 |
{ NULL }
|
1355 |
}; |
1356 |
|
1357 |
static BlockDriver bdrv_qcow2 = {
|
1358 |
.format_name = "qcow2",
|
1359 |
.instance_size = sizeof(BDRVQcowState),
|
1360 |
.bdrv_probe = qcow2_probe, |
1361 |
.bdrv_open = qcow2_open, |
1362 |
.bdrv_close = qcow2_close, |
1363 |
.bdrv_create = qcow2_create, |
1364 |
.bdrv_flush = qcow2_flush, |
1365 |
.bdrv_is_allocated = qcow2_is_allocated, |
1366 |
.bdrv_set_key = qcow2_set_key, |
1367 |
.bdrv_make_empty = qcow2_make_empty, |
1368 |
|
1369 |
.bdrv_aio_readv = qcow2_aio_readv, |
1370 |
.bdrv_aio_writev = qcow2_aio_writev, |
1371 |
.bdrv_aio_flush = qcow2_aio_flush, |
1372 |
|
1373 |
.bdrv_discard = qcow2_discard, |
1374 |
.bdrv_truncate = qcow2_truncate, |
1375 |
.bdrv_write_compressed = qcow2_write_compressed, |
1376 |
|
1377 |
.bdrv_snapshot_create = qcow2_snapshot_create, |
1378 |
.bdrv_snapshot_goto = qcow2_snapshot_goto, |
1379 |
.bdrv_snapshot_delete = qcow2_snapshot_delete, |
1380 |
.bdrv_snapshot_list = qcow2_snapshot_list, |
1381 |
.bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp, |
1382 |
.bdrv_get_info = qcow2_get_info, |
1383 |
|
1384 |
.bdrv_save_vmstate = qcow2_save_vmstate, |
1385 |
.bdrv_load_vmstate = qcow2_load_vmstate, |
1386 |
|
1387 |
.bdrv_change_backing_file = qcow2_change_backing_file, |
1388 |
|
1389 |
.create_options = qcow2_create_options, |
1390 |
.bdrv_check = qcow2_check, |
1391 |
}; |
1392 |
|
1393 |
static void bdrv_qcow2_init(void) |
1394 |
{ |
1395 |
bdrv_register(&bdrv_qcow2); |
1396 |
} |
1397 |
|
1398 |
block_init(bdrv_qcow2_init); |