root / block / qcow2.c @ c142442b
History | View | Annotate | Download (30.2 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 |
#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 |
|
31 |
/*
|
32 |
Differences with QCOW:
|
33 |
|
34 |
- Support for multiple incremental snapshots.
|
35 |
- Memory management by reference counts.
|
36 |
- Clusters which have a reference count of one have the bit
|
37 |
QCOW_OFLAG_COPIED to optimize write performance.
|
38 |
- Size of compressed clusters is stored in sectors to reduce bit usage
|
39 |
in the cluster offsets.
|
40 |
- Support for storing additional data (such as the VM state) in the
|
41 |
snapshots.
|
42 |
- If a backing store is used, the cluster size is not constrained
|
43 |
(could be backported to QCOW).
|
44 |
- L2 tables have always a size of one cluster.
|
45 |
*/
|
46 |
|
47 |
//#define DEBUG_ALLOC
|
48 |
//#define DEBUG_ALLOC2
|
49 |
//#define DEBUG_EXT
|
50 |
|
51 |
|
52 |
typedef struct { |
53 |
uint32_t magic; |
54 |
uint32_t len; |
55 |
} QCowExtension; |
56 |
#define QCOW_EXT_MAGIC_END 0 |
57 |
#define QCOW_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA |
58 |
|
59 |
|
60 |
|
61 |
static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename) |
62 |
{ |
63 |
const QCowHeader *cow_header = (const void *)buf; |
64 |
|
65 |
if (buf_size >= sizeof(QCowHeader) && |
66 |
be32_to_cpu(cow_header->magic) == QCOW_MAGIC && |
67 |
be32_to_cpu(cow_header->version) == QCOW_VERSION) |
68 |
return 100; |
69 |
else
|
70 |
return 0; |
71 |
} |
72 |
|
73 |
|
74 |
/*
|
75 |
* read qcow2 extension and fill bs
|
76 |
* start reading from start_offset
|
77 |
* finish reading upon magic of value 0 or when end_offset reached
|
78 |
* unknown magic is skipped (future extension this version knows nothing about)
|
79 |
* return 0 upon success, non-0 otherwise
|
80 |
*/
|
81 |
static int qcow_read_extensions(BlockDriverState *bs, uint64_t start_offset, |
82 |
uint64_t end_offset) |
83 |
{ |
84 |
BDRVQcowState *s = bs->opaque; |
85 |
QCowExtension ext; |
86 |
uint64_t offset; |
87 |
|
88 |
#ifdef DEBUG_EXT
|
89 |
printf("qcow_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
|
90 |
#endif
|
91 |
offset = start_offset; |
92 |
while (offset < end_offset) {
|
93 |
|
94 |
#ifdef DEBUG_EXT
|
95 |
/* Sanity check */
|
96 |
if (offset > s->cluster_size)
|
97 |
printf("qcow_handle_extension: suspicious offset %lu\n", offset);
|
98 |
|
99 |
printf("attemting to read extended header in offset %lu\n", offset);
|
100 |
#endif
|
101 |
|
102 |
if (bdrv_pread(s->hd, offset, &ext, sizeof(ext)) != sizeof(ext)) { |
103 |
fprintf(stderr, "qcow_handle_extension: ERROR: pread fail from offset %llu\n",
|
104 |
(unsigned long long)offset); |
105 |
return 1; |
106 |
} |
107 |
be32_to_cpus(&ext.magic); |
108 |
be32_to_cpus(&ext.len); |
109 |
offset += sizeof(ext);
|
110 |
#ifdef DEBUG_EXT
|
111 |
printf("ext.magic = 0x%x\n", ext.magic);
|
112 |
#endif
|
113 |
switch (ext.magic) {
|
114 |
case QCOW_EXT_MAGIC_END:
|
115 |
return 0; |
116 |
|
117 |
case QCOW_EXT_MAGIC_BACKING_FORMAT:
|
118 |
if (ext.len >= sizeof(bs->backing_format)) { |
119 |
fprintf(stderr, "ERROR: ext_backing_format: len=%u too large"
|
120 |
" (>=%zu)\n",
|
121 |
ext.len, sizeof(bs->backing_format));
|
122 |
return 2; |
123 |
} |
124 |
if (bdrv_pread(s->hd, offset , bs->backing_format,
|
125 |
ext.len) != ext.len) |
126 |
return 3; |
127 |
bs->backing_format[ext.len] = '\0';
|
128 |
#ifdef DEBUG_EXT
|
129 |
printf("Qcow2: Got format extension %s\n", bs->backing_format);
|
130 |
#endif
|
131 |
offset += ((ext.len + 7) & ~7); |
132 |
break;
|
133 |
|
134 |
default:
|
135 |
/* unknown magic -- just skip it */
|
136 |
offset += ((ext.len + 7) & ~7); |
137 |
break;
|
138 |
} |
139 |
} |
140 |
|
141 |
return 0; |
142 |
} |
143 |
|
144 |
|
145 |
static int qcow_open(BlockDriverState *bs, const char *filename, int flags) |
146 |
{ |
147 |
BDRVQcowState *s = bs->opaque; |
148 |
int len, i, shift, ret;
|
149 |
QCowHeader header; |
150 |
uint64_t ext_end; |
151 |
|
152 |
/* Performance is terrible right now with cache=writethrough due mainly
|
153 |
* to reference count updates. If the user does not explicitly specify
|
154 |
* a caching type, force to writeback caching.
|
155 |
*/
|
156 |
if ((flags & BDRV_O_CACHE_DEF)) {
|
157 |
flags |= BDRV_O_CACHE_WB; |
158 |
flags &= ~BDRV_O_CACHE_DEF; |
159 |
} |
160 |
ret = bdrv_file_open(&s->hd, filename, flags); |
161 |
if (ret < 0) |
162 |
return ret;
|
163 |
if (bdrv_pread(s->hd, 0, &header, sizeof(header)) != sizeof(header)) |
164 |
goto fail;
|
165 |
be32_to_cpus(&header.magic); |
166 |
be32_to_cpus(&header.version); |
167 |
be64_to_cpus(&header.backing_file_offset); |
168 |
be32_to_cpus(&header.backing_file_size); |
169 |
be64_to_cpus(&header.size); |
170 |
be32_to_cpus(&header.cluster_bits); |
171 |
be32_to_cpus(&header.crypt_method); |
172 |
be64_to_cpus(&header.l1_table_offset); |
173 |
be32_to_cpus(&header.l1_size); |
174 |
be64_to_cpus(&header.refcount_table_offset); |
175 |
be32_to_cpus(&header.refcount_table_clusters); |
176 |
be64_to_cpus(&header.snapshots_offset); |
177 |
be32_to_cpus(&header.nb_snapshots); |
178 |
|
179 |
if (header.magic != QCOW_MAGIC || header.version != QCOW_VERSION)
|
180 |
goto fail;
|
181 |
if (header.size <= 1 || |
182 |
header.cluster_bits < MIN_CLUSTER_BITS || |
183 |
header.cluster_bits > MAX_CLUSTER_BITS) |
184 |
goto fail;
|
185 |
if (header.crypt_method > QCOW_CRYPT_AES)
|
186 |
goto fail;
|
187 |
s->crypt_method_header = header.crypt_method; |
188 |
if (s->crypt_method_header)
|
189 |
bs->encrypted = 1;
|
190 |
s->cluster_bits = header.cluster_bits; |
191 |
s->cluster_size = 1 << s->cluster_bits;
|
192 |
s->cluster_sectors = 1 << (s->cluster_bits - 9); |
193 |
s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */ |
194 |
s->l2_size = 1 << s->l2_bits;
|
195 |
bs->total_sectors = header.size / 512;
|
196 |
s->csize_shift = (62 - (s->cluster_bits - 8)); |
197 |
s->csize_mask = (1 << (s->cluster_bits - 8)) - 1; |
198 |
s->cluster_offset_mask = (1LL << s->csize_shift) - 1; |
199 |
s->refcount_table_offset = header.refcount_table_offset; |
200 |
s->refcount_table_size = |
201 |
header.refcount_table_clusters << (s->cluster_bits - 3);
|
202 |
|
203 |
s->snapshots_offset = header.snapshots_offset; |
204 |
s->nb_snapshots = header.nb_snapshots; |
205 |
|
206 |
/* read the level 1 table */
|
207 |
s->l1_size = header.l1_size; |
208 |
shift = s->cluster_bits + s->l2_bits; |
209 |
s->l1_vm_state_index = (header.size + (1LL << shift) - 1) >> shift; |
210 |
/* the L1 table must contain at least enough entries to put
|
211 |
header.size bytes */
|
212 |
if (s->l1_size < s->l1_vm_state_index)
|
213 |
goto fail;
|
214 |
s->l1_table_offset = header.l1_table_offset; |
215 |
s->l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
|
216 |
if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, s->l1_size * sizeof(uint64_t)) != |
217 |
s->l1_size * sizeof(uint64_t))
|
218 |
goto fail;
|
219 |
for(i = 0;i < s->l1_size; i++) { |
220 |
be64_to_cpus(&s->l1_table[i]); |
221 |
} |
222 |
/* alloc L2 cache */
|
223 |
s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
|
224 |
s->cluster_cache = qemu_malloc(s->cluster_size); |
225 |
/* one more sector for decompressed data alignment */
|
226 |
s->cluster_data = qemu_malloc(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size |
227 |
+ 512);
|
228 |
s->cluster_cache_offset = -1;
|
229 |
|
230 |
if (refcount_init(bs) < 0) |
231 |
goto fail;
|
232 |
|
233 |
/* read qcow2 extensions */
|
234 |
if (header.backing_file_offset)
|
235 |
ext_end = header.backing_file_offset; |
236 |
else
|
237 |
ext_end = s->cluster_size; |
238 |
if (qcow_read_extensions(bs, sizeof(header), ext_end)) |
239 |
goto fail;
|
240 |
|
241 |
/* read the backing file name */
|
242 |
if (header.backing_file_offset != 0) { |
243 |
len = header.backing_file_size; |
244 |
if (len > 1023) |
245 |
len = 1023;
|
246 |
if (bdrv_pread(s->hd, header.backing_file_offset, bs->backing_file, len) != len)
|
247 |
goto fail;
|
248 |
bs->backing_file[len] = '\0';
|
249 |
} |
250 |
if (qcow_read_snapshots(bs) < 0) |
251 |
goto fail;
|
252 |
|
253 |
#ifdef DEBUG_ALLOC
|
254 |
check_refcounts(bs); |
255 |
#endif
|
256 |
return 0; |
257 |
|
258 |
fail:
|
259 |
qcow_free_snapshots(bs); |
260 |
refcount_close(bs); |
261 |
qemu_free(s->l1_table); |
262 |
qemu_free(s->l2_cache); |
263 |
qemu_free(s->cluster_cache); |
264 |
qemu_free(s->cluster_data); |
265 |
bdrv_delete(s->hd); |
266 |
return -1; |
267 |
} |
268 |
|
269 |
static int qcow_set_key(BlockDriverState *bs, const char *key) |
270 |
{ |
271 |
BDRVQcowState *s = bs->opaque; |
272 |
uint8_t keybuf[16];
|
273 |
int len, i;
|
274 |
|
275 |
memset(keybuf, 0, 16); |
276 |
len = strlen(key); |
277 |
if (len > 16) |
278 |
len = 16;
|
279 |
/* XXX: we could compress the chars to 7 bits to increase
|
280 |
entropy */
|
281 |
for(i = 0;i < len;i++) { |
282 |
keybuf[i] = key[i]; |
283 |
} |
284 |
s->crypt_method = s->crypt_method_header; |
285 |
|
286 |
if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0) |
287 |
return -1; |
288 |
if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0) |
289 |
return -1; |
290 |
#if 0
|
291 |
/* test */
|
292 |
{
|
293 |
uint8_t in[16];
|
294 |
uint8_t out[16];
|
295 |
uint8_t tmp[16];
|
296 |
for(i=0;i<16;i++)
|
297 |
in[i] = i;
|
298 |
AES_encrypt(in, tmp, &s->aes_encrypt_key);
|
299 |
AES_decrypt(tmp, out, &s->aes_decrypt_key);
|
300 |
for(i = 0; i < 16; i++)
|
301 |
printf(" %02x", tmp[i]);
|
302 |
printf("\n");
|
303 |
for(i = 0; i < 16; i++)
|
304 |
printf(" %02x", out[i]);
|
305 |
printf("\n");
|
306 |
}
|
307 |
#endif
|
308 |
return 0; |
309 |
} |
310 |
|
311 |
static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num, |
312 |
int nb_sectors, int *pnum) |
313 |
{ |
314 |
uint64_t cluster_offset; |
315 |
|
316 |
*pnum = nb_sectors; |
317 |
cluster_offset = get_cluster_offset(bs, sector_num << 9, pnum);
|
318 |
|
319 |
return (cluster_offset != 0); |
320 |
} |
321 |
|
322 |
/* handle reading after the end of the backing file */
|
323 |
int backing_read1(BlockDriverState *bs,
|
324 |
int64_t sector_num, uint8_t *buf, int nb_sectors)
|
325 |
{ |
326 |
int n1;
|
327 |
if ((sector_num + nb_sectors) <= bs->total_sectors)
|
328 |
return nb_sectors;
|
329 |
if (sector_num >= bs->total_sectors)
|
330 |
n1 = 0;
|
331 |
else
|
332 |
n1 = bs->total_sectors - sector_num; |
333 |
memset(buf + n1 * 512, 0, 512 * (nb_sectors - n1)); |
334 |
return n1;
|
335 |
} |
336 |
|
337 |
typedef struct QCowAIOCB { |
338 |
BlockDriverAIOCB common; |
339 |
int64_t sector_num; |
340 |
QEMUIOVector *qiov; |
341 |
uint8_t *buf; |
342 |
void *orig_buf;
|
343 |
int nb_sectors;
|
344 |
int n;
|
345 |
uint64_t cluster_offset; |
346 |
uint8_t *cluster_data; |
347 |
BlockDriverAIOCB *hd_aiocb; |
348 |
struct iovec hd_iov;
|
349 |
QEMUIOVector hd_qiov; |
350 |
QEMUBH *bh; |
351 |
QCowL2Meta l2meta; |
352 |
} QCowAIOCB; |
353 |
|
354 |
static void qcow_aio_cancel(BlockDriverAIOCB *blockacb) |
355 |
{ |
356 |
QCowAIOCB *acb = (QCowAIOCB *)blockacb; |
357 |
if (acb->hd_aiocb)
|
358 |
bdrv_aio_cancel(acb->hd_aiocb); |
359 |
qemu_aio_release(acb); |
360 |
} |
361 |
|
362 |
static AIOPool qcow_aio_pool = {
|
363 |
.aiocb_size = sizeof(QCowAIOCB),
|
364 |
.cancel = qcow_aio_cancel, |
365 |
}; |
366 |
|
367 |
static void qcow_aio_read_cb(void *opaque, int ret); |
368 |
static void qcow_aio_read_bh(void *opaque) |
369 |
{ |
370 |
QCowAIOCB *acb = opaque; |
371 |
qemu_bh_delete(acb->bh); |
372 |
acb->bh = NULL;
|
373 |
qcow_aio_read_cb(opaque, 0);
|
374 |
} |
375 |
|
376 |
static int qcow_schedule_bh(QEMUBHFunc *cb, QCowAIOCB *acb) |
377 |
{ |
378 |
if (acb->bh)
|
379 |
return -EIO;
|
380 |
|
381 |
acb->bh = qemu_bh_new(cb, acb); |
382 |
if (!acb->bh)
|
383 |
return -EIO;
|
384 |
|
385 |
qemu_bh_schedule(acb->bh); |
386 |
|
387 |
return 0; |
388 |
} |
389 |
|
390 |
static void qcow_aio_read_cb(void *opaque, int ret) |
391 |
{ |
392 |
QCowAIOCB *acb = opaque; |
393 |
BlockDriverState *bs = acb->common.bs; |
394 |
BDRVQcowState *s = bs->opaque; |
395 |
int index_in_cluster, n1;
|
396 |
|
397 |
acb->hd_aiocb = NULL;
|
398 |
if (ret < 0) |
399 |
goto done;
|
400 |
|
401 |
/* post process the read buffer */
|
402 |
if (!acb->cluster_offset) {
|
403 |
/* nothing to do */
|
404 |
} else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) { |
405 |
/* nothing to do */
|
406 |
} else {
|
407 |
if (s->crypt_method) {
|
408 |
encrypt_sectors(s, acb->sector_num, acb->buf, acb->buf, |
409 |
acb->n, 0,
|
410 |
&s->aes_decrypt_key); |
411 |
} |
412 |
} |
413 |
|
414 |
acb->nb_sectors -= acb->n; |
415 |
acb->sector_num += acb->n; |
416 |
acb->buf += acb->n * 512;
|
417 |
|
418 |
if (acb->nb_sectors == 0) { |
419 |
/* request completed */
|
420 |
ret = 0;
|
421 |
goto done;
|
422 |
} |
423 |
|
424 |
/* prepare next AIO request */
|
425 |
acb->n = acb->nb_sectors; |
426 |
acb->cluster_offset = get_cluster_offset(bs, acb->sector_num << 9, &acb->n);
|
427 |
index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
|
428 |
|
429 |
if (!acb->cluster_offset) {
|
430 |
if (bs->backing_hd) {
|
431 |
/* read from the base image */
|
432 |
n1 = backing_read1(bs->backing_hd, acb->sector_num, |
433 |
acb->buf, acb->n); |
434 |
if (n1 > 0) { |
435 |
acb->hd_iov.iov_base = (void *)acb->buf;
|
436 |
acb->hd_iov.iov_len = acb->n * 512;
|
437 |
qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
|
438 |
acb->hd_aiocb = bdrv_aio_readv(bs->backing_hd, acb->sector_num, |
439 |
&acb->hd_qiov, acb->n, |
440 |
qcow_aio_read_cb, acb); |
441 |
if (acb->hd_aiocb == NULL) |
442 |
goto done;
|
443 |
} else {
|
444 |
ret = qcow_schedule_bh(qcow_aio_read_bh, acb); |
445 |
if (ret < 0) |
446 |
goto done;
|
447 |
} |
448 |
} else {
|
449 |
/* Note: in this case, no need to wait */
|
450 |
memset(acb->buf, 0, 512 * acb->n); |
451 |
ret = qcow_schedule_bh(qcow_aio_read_bh, acb); |
452 |
if (ret < 0) |
453 |
goto done;
|
454 |
} |
455 |
} else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) { |
456 |
/* add AIO support for compressed blocks ? */
|
457 |
if (decompress_cluster(s, acb->cluster_offset) < 0) |
458 |
goto done;
|
459 |
memcpy(acb->buf, |
460 |
s->cluster_cache + index_in_cluster * 512, 512 * acb->n); |
461 |
ret = qcow_schedule_bh(qcow_aio_read_bh, acb); |
462 |
if (ret < 0) |
463 |
goto done;
|
464 |
} else {
|
465 |
if ((acb->cluster_offset & 511) != 0) { |
466 |
ret = -EIO; |
467 |
goto done;
|
468 |
} |
469 |
|
470 |
acb->hd_iov.iov_base = (void *)acb->buf;
|
471 |
acb->hd_iov.iov_len = acb->n * 512;
|
472 |
qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
|
473 |
acb->hd_aiocb = bdrv_aio_readv(s->hd, |
474 |
(acb->cluster_offset >> 9) + index_in_cluster,
|
475 |
&acb->hd_qiov, acb->n, qcow_aio_read_cb, acb); |
476 |
if (acb->hd_aiocb == NULL) |
477 |
goto done;
|
478 |
} |
479 |
|
480 |
return;
|
481 |
done:
|
482 |
if (acb->qiov->niov > 1) { |
483 |
qemu_iovec_from_buffer(acb->qiov, acb->orig_buf, acb->qiov->size); |
484 |
qemu_vfree(acb->orig_buf); |
485 |
} |
486 |
acb->common.cb(acb->common.opaque, ret); |
487 |
qemu_aio_release(acb); |
488 |
} |
489 |
|
490 |
static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs,
|
491 |
int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
|
492 |
BlockDriverCompletionFunc *cb, void *opaque, int is_write) |
493 |
{ |
494 |
QCowAIOCB *acb; |
495 |
|
496 |
acb = qemu_aio_get(&qcow_aio_pool, bs, cb, opaque); |
497 |
if (!acb)
|
498 |
return NULL; |
499 |
acb->hd_aiocb = NULL;
|
500 |
acb->sector_num = sector_num; |
501 |
acb->qiov = qiov; |
502 |
if (qiov->niov > 1) { |
503 |
acb->buf = acb->orig_buf = qemu_blockalign(bs, qiov->size); |
504 |
if (is_write)
|
505 |
qemu_iovec_to_buffer(qiov, acb->buf); |
506 |
} else {
|
507 |
acb->buf = (uint8_t *)qiov->iov->iov_base; |
508 |
} |
509 |
acb->nb_sectors = nb_sectors; |
510 |
acb->n = 0;
|
511 |
acb->cluster_offset = 0;
|
512 |
acb->l2meta.nb_clusters = 0;
|
513 |
return acb;
|
514 |
} |
515 |
|
516 |
static BlockDriverAIOCB *qcow_aio_readv(BlockDriverState *bs,
|
517 |
int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
|
518 |
BlockDriverCompletionFunc *cb, void *opaque)
|
519 |
{ |
520 |
QCowAIOCB *acb; |
521 |
|
522 |
acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
|
523 |
if (!acb)
|
524 |
return NULL; |
525 |
|
526 |
qcow_aio_read_cb(acb, 0);
|
527 |
return &acb->common;
|
528 |
} |
529 |
|
530 |
static void qcow_aio_write_cb(void *opaque, int ret) |
531 |
{ |
532 |
QCowAIOCB *acb = opaque; |
533 |
BlockDriverState *bs = acb->common.bs; |
534 |
BDRVQcowState *s = bs->opaque; |
535 |
int index_in_cluster;
|
536 |
const uint8_t *src_buf;
|
537 |
int n_end;
|
538 |
|
539 |
acb->hd_aiocb = NULL;
|
540 |
|
541 |
if (ret < 0) |
542 |
goto done;
|
543 |
|
544 |
if (alloc_cluster_link_l2(bs, acb->cluster_offset, &acb->l2meta) < 0) { |
545 |
free_any_clusters(bs, acb->cluster_offset, acb->l2meta.nb_clusters); |
546 |
goto done;
|
547 |
} |
548 |
|
549 |
acb->nb_sectors -= acb->n; |
550 |
acb->sector_num += acb->n; |
551 |
acb->buf += acb->n * 512;
|
552 |
|
553 |
if (acb->nb_sectors == 0) { |
554 |
/* request completed */
|
555 |
ret = 0;
|
556 |
goto done;
|
557 |
} |
558 |
|
559 |
index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
|
560 |
n_end = index_in_cluster + acb->nb_sectors; |
561 |
if (s->crypt_method &&
|
562 |
n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors) |
563 |
n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors; |
564 |
|
565 |
acb->cluster_offset = alloc_cluster_offset(bs, acb->sector_num << 9,
|
566 |
index_in_cluster, |
567 |
n_end, &acb->n, &acb->l2meta); |
568 |
if (!acb->cluster_offset || (acb->cluster_offset & 511) != 0) { |
569 |
ret = -EIO; |
570 |
goto done;
|
571 |
} |
572 |
if (s->crypt_method) {
|
573 |
if (!acb->cluster_data) {
|
574 |
acb->cluster_data = qemu_mallocz(QCOW_MAX_CRYPT_CLUSTERS * |
575 |
s->cluster_size); |
576 |
} |
577 |
encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf, |
578 |
acb->n, 1, &s->aes_encrypt_key);
|
579 |
src_buf = acb->cluster_data; |
580 |
} else {
|
581 |
src_buf = acb->buf; |
582 |
} |
583 |
acb->hd_iov.iov_base = (void *)src_buf;
|
584 |
acb->hd_iov.iov_len = acb->n * 512;
|
585 |
qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
|
586 |
acb->hd_aiocb = bdrv_aio_writev(s->hd, |
587 |
(acb->cluster_offset >> 9) + index_in_cluster,
|
588 |
&acb->hd_qiov, acb->n, |
589 |
qcow_aio_write_cb, acb); |
590 |
if (acb->hd_aiocb == NULL) |
591 |
goto done;
|
592 |
|
593 |
return;
|
594 |
|
595 |
done:
|
596 |
if (acb->qiov->niov > 1) |
597 |
qemu_vfree(acb->orig_buf); |
598 |
acb->common.cb(acb->common.opaque, ret); |
599 |
qemu_aio_release(acb); |
600 |
} |
601 |
|
602 |
static BlockDriverAIOCB *qcow_aio_writev(BlockDriverState *bs,
|
603 |
int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
|
604 |
BlockDriverCompletionFunc *cb, void *opaque)
|
605 |
{ |
606 |
BDRVQcowState *s = bs->opaque; |
607 |
QCowAIOCB *acb; |
608 |
|
609 |
s->cluster_cache_offset = -1; /* disable compressed cache */ |
610 |
|
611 |
acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
|
612 |
if (!acb)
|
613 |
return NULL; |
614 |
|
615 |
qcow_aio_write_cb(acb, 0);
|
616 |
return &acb->common;
|
617 |
} |
618 |
|
619 |
static void qcow_close(BlockDriverState *bs) |
620 |
{ |
621 |
BDRVQcowState *s = bs->opaque; |
622 |
qemu_free(s->l1_table); |
623 |
qemu_free(s->l2_cache); |
624 |
qemu_free(s->cluster_cache); |
625 |
qemu_free(s->cluster_data); |
626 |
refcount_close(bs); |
627 |
bdrv_delete(s->hd); |
628 |
} |
629 |
|
630 |
static int get_bits_from_size(size_t size) |
631 |
{ |
632 |
int res = 0; |
633 |
|
634 |
if (size == 0) { |
635 |
return -1; |
636 |
} |
637 |
|
638 |
while (size != 1) { |
639 |
/* Not a power of two */
|
640 |
if (size & 1) { |
641 |
return -1; |
642 |
} |
643 |
|
644 |
size >>= 1;
|
645 |
res++; |
646 |
} |
647 |
|
648 |
return res;
|
649 |
} |
650 |
|
651 |
static int qcow_create2(const char *filename, int64_t total_size, |
652 |
const char *backing_file, const char *backing_format, |
653 |
int flags, size_t cluster_size)
|
654 |
{ |
655 |
|
656 |
int fd, header_size, backing_filename_len, l1_size, i, shift, l2_bits;
|
657 |
int ref_clusters, backing_format_len = 0; |
658 |
QCowHeader header; |
659 |
uint64_t tmp, offset; |
660 |
QCowCreateState s1, *s = &s1; |
661 |
QCowExtension ext_bf = {0, 0}; |
662 |
|
663 |
|
664 |
memset(s, 0, sizeof(*s)); |
665 |
|
666 |
fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
|
667 |
if (fd < 0) |
668 |
return -1; |
669 |
memset(&header, 0, sizeof(header)); |
670 |
header.magic = cpu_to_be32(QCOW_MAGIC); |
671 |
header.version = cpu_to_be32(QCOW_VERSION); |
672 |
header.size = cpu_to_be64(total_size * 512);
|
673 |
header_size = sizeof(header);
|
674 |
backing_filename_len = 0;
|
675 |
if (backing_file) {
|
676 |
if (backing_format) {
|
677 |
ext_bf.magic = QCOW_EXT_MAGIC_BACKING_FORMAT; |
678 |
backing_format_len = strlen(backing_format); |
679 |
ext_bf.len = (backing_format_len + 7) & ~7; |
680 |
header_size += ((sizeof(ext_bf) + ext_bf.len + 7) & ~7); |
681 |
} |
682 |
header.backing_file_offset = cpu_to_be64(header_size); |
683 |
backing_filename_len = strlen(backing_file); |
684 |
header.backing_file_size = cpu_to_be32(backing_filename_len); |
685 |
header_size += backing_filename_len; |
686 |
} |
687 |
|
688 |
/* Cluster size */
|
689 |
s->cluster_bits = get_bits_from_size(cluster_size); |
690 |
if (s->cluster_bits < MIN_CLUSTER_BITS ||
|
691 |
s->cluster_bits > MAX_CLUSTER_BITS) |
692 |
{ |
693 |
fprintf(stderr, "Cluster size must be a power of two between "
|
694 |
"%d and %dk\n",
|
695 |
1 << MIN_CLUSTER_BITS,
|
696 |
1 << (MAX_CLUSTER_BITS - 10)); |
697 |
return -EINVAL;
|
698 |
} |
699 |
s->cluster_size = 1 << s->cluster_bits;
|
700 |
|
701 |
header.cluster_bits = cpu_to_be32(s->cluster_bits); |
702 |
header_size = (header_size + 7) & ~7; |
703 |
if (flags & BLOCK_FLAG_ENCRYPT) {
|
704 |
header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES); |
705 |
} else {
|
706 |
header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE); |
707 |
} |
708 |
l2_bits = s->cluster_bits - 3;
|
709 |
shift = s->cluster_bits + l2_bits; |
710 |
l1_size = (((total_size * 512) + (1LL << shift) - 1) >> shift); |
711 |
offset = align_offset(header_size, s->cluster_size); |
712 |
s->l1_table_offset = offset; |
713 |
header.l1_table_offset = cpu_to_be64(s->l1_table_offset); |
714 |
header.l1_size = cpu_to_be32(l1_size); |
715 |
offset += align_offset(l1_size * sizeof(uint64_t), s->cluster_size);
|
716 |
|
717 |
s->refcount_table = qemu_mallocz(s->cluster_size); |
718 |
|
719 |
s->refcount_table_offset = offset; |
720 |
header.refcount_table_offset = cpu_to_be64(offset); |
721 |
header.refcount_table_clusters = cpu_to_be32(1);
|
722 |
offset += s->cluster_size; |
723 |
s->refcount_block_offset = offset; |
724 |
|
725 |
/* count how many refcount blocks needed */
|
726 |
tmp = offset >> s->cluster_bits; |
727 |
ref_clusters = (tmp >> (s->cluster_bits - REFCOUNT_SHIFT)) + 1;
|
728 |
for (i=0; i < ref_clusters; i++) { |
729 |
s->refcount_table[i] = cpu_to_be64(offset); |
730 |
offset += s->cluster_size; |
731 |
} |
732 |
|
733 |
s->refcount_block = qemu_mallocz(ref_clusters * s->cluster_size); |
734 |
|
735 |
/* update refcounts */
|
736 |
create_refcount_update(s, 0, header_size);
|
737 |
create_refcount_update(s, s->l1_table_offset, l1_size * sizeof(uint64_t));
|
738 |
create_refcount_update(s, s->refcount_table_offset, s->cluster_size); |
739 |
create_refcount_update(s, s->refcount_block_offset, ref_clusters * s->cluster_size); |
740 |
|
741 |
/* write all the data */
|
742 |
write(fd, &header, sizeof(header));
|
743 |
if (backing_file) {
|
744 |
if (backing_format_len) {
|
745 |
char zero[16]; |
746 |
int d = ext_bf.len - backing_format_len;
|
747 |
|
748 |
memset(zero, 0, sizeof(zero)); |
749 |
cpu_to_be32s(&ext_bf.magic); |
750 |
cpu_to_be32s(&ext_bf.len); |
751 |
write(fd, &ext_bf, sizeof(ext_bf));
|
752 |
write(fd, backing_format, backing_format_len); |
753 |
if (d>0) { |
754 |
write(fd, zero, d); |
755 |
} |
756 |
} |
757 |
write(fd, backing_file, backing_filename_len); |
758 |
} |
759 |
lseek(fd, s->l1_table_offset, SEEK_SET); |
760 |
tmp = 0;
|
761 |
for(i = 0;i < l1_size; i++) { |
762 |
write(fd, &tmp, sizeof(tmp));
|
763 |
} |
764 |
lseek(fd, s->refcount_table_offset, SEEK_SET); |
765 |
write(fd, s->refcount_table, s->cluster_size); |
766 |
|
767 |
lseek(fd, s->refcount_block_offset, SEEK_SET); |
768 |
write(fd, s->refcount_block, ref_clusters * s->cluster_size); |
769 |
|
770 |
qemu_free(s->refcount_table); |
771 |
qemu_free(s->refcount_block); |
772 |
close(fd); |
773 |
return 0; |
774 |
} |
775 |
|
776 |
static int qcow_create(const char *filename, QEMUOptionParameter *options) |
777 |
{ |
778 |
const char *backing_file = NULL; |
779 |
const char *backing_fmt = NULL; |
780 |
uint64_t sectors = 0;
|
781 |
int flags = 0; |
782 |
size_t cluster_size = 65536;
|
783 |
|
784 |
/* Read out options */
|
785 |
while (options && options->name) {
|
786 |
if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
|
787 |
sectors = options->value.n / 512;
|
788 |
} else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) { |
789 |
backing_file = options->value.s; |
790 |
} else if (!strcmp(options->name, BLOCK_OPT_BACKING_FMT)) { |
791 |
backing_fmt = options->value.s; |
792 |
} else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) { |
793 |
flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
|
794 |
} else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) { |
795 |
if (options->value.n) {
|
796 |
cluster_size = options->value.n; |
797 |
} |
798 |
} |
799 |
options++; |
800 |
} |
801 |
|
802 |
return qcow_create2(filename, sectors, backing_file, backing_fmt, flags,
|
803 |
cluster_size); |
804 |
} |
805 |
|
806 |
static int qcow_make_empty(BlockDriverState *bs) |
807 |
{ |
808 |
#if 0
|
809 |
/* XXX: not correct */
|
810 |
BDRVQcowState *s = bs->opaque;
|
811 |
uint32_t l1_length = s->l1_size * sizeof(uint64_t);
|
812 |
int ret;
|
813 |
|
814 |
memset(s->l1_table, 0, l1_length);
|
815 |
if (bdrv_pwrite(s->hd, s->l1_table_offset, s->l1_table, l1_length) < 0)
|
816 |
return -1;
|
817 |
ret = bdrv_truncate(s->hd, s->l1_table_offset + l1_length);
|
818 |
if (ret < 0)
|
819 |
return ret;
|
820 |
|
821 |
l2_cache_reset(bs);
|
822 |
#endif
|
823 |
return 0; |
824 |
} |
825 |
|
826 |
/* XXX: put compressed sectors first, then all the cluster aligned
|
827 |
tables to avoid losing bytes in alignment */
|
828 |
static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num, |
829 |
const uint8_t *buf, int nb_sectors) |
830 |
{ |
831 |
BDRVQcowState *s = bs->opaque; |
832 |
z_stream strm; |
833 |
int ret, out_len;
|
834 |
uint8_t *out_buf; |
835 |
uint64_t cluster_offset; |
836 |
|
837 |
if (nb_sectors == 0) { |
838 |
/* align end of file to a sector boundary to ease reading with
|
839 |
sector based I/Os */
|
840 |
cluster_offset = bdrv_getlength(s->hd); |
841 |
cluster_offset = (cluster_offset + 511) & ~511; |
842 |
bdrv_truncate(s->hd, cluster_offset); |
843 |
return 0; |
844 |
} |
845 |
|
846 |
if (nb_sectors != s->cluster_sectors)
|
847 |
return -EINVAL;
|
848 |
|
849 |
out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128); |
850 |
|
851 |
/* best compression, small window, no zlib header */
|
852 |
memset(&strm, 0, sizeof(strm)); |
853 |
ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION, |
854 |
Z_DEFLATED, -12,
|
855 |
9, Z_DEFAULT_STRATEGY);
|
856 |
if (ret != 0) { |
857 |
qemu_free(out_buf); |
858 |
return -1; |
859 |
} |
860 |
|
861 |
strm.avail_in = s->cluster_size; |
862 |
strm.next_in = (uint8_t *)buf; |
863 |
strm.avail_out = s->cluster_size; |
864 |
strm.next_out = out_buf; |
865 |
|
866 |
ret = deflate(&strm, Z_FINISH); |
867 |
if (ret != Z_STREAM_END && ret != Z_OK) {
|
868 |
qemu_free(out_buf); |
869 |
deflateEnd(&strm); |
870 |
return -1; |
871 |
} |
872 |
out_len = strm.next_out - out_buf; |
873 |
|
874 |
deflateEnd(&strm); |
875 |
|
876 |
if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
|
877 |
/* could not compress: write normal cluster */
|
878 |
bdrv_write(bs, sector_num, buf, s->cluster_sectors); |
879 |
} else {
|
880 |
cluster_offset = alloc_compressed_cluster_offset(bs, sector_num << 9,
|
881 |
out_len); |
882 |
if (!cluster_offset)
|
883 |
return -1; |
884 |
cluster_offset &= s->cluster_offset_mask; |
885 |
if (bdrv_pwrite(s->hd, cluster_offset, out_buf, out_len) != out_len) {
|
886 |
qemu_free(out_buf); |
887 |
return -1; |
888 |
} |
889 |
} |
890 |
|
891 |
qemu_free(out_buf); |
892 |
return 0; |
893 |
} |
894 |
|
895 |
static void qcow_flush(BlockDriverState *bs) |
896 |
{ |
897 |
BDRVQcowState *s = bs->opaque; |
898 |
bdrv_flush(s->hd); |
899 |
} |
900 |
|
901 |
static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) |
902 |
{ |
903 |
BDRVQcowState *s = bs->opaque; |
904 |
bdi->cluster_size = s->cluster_size; |
905 |
bdi->vm_state_offset = (int64_t)s->l1_vm_state_index << |
906 |
(s->cluster_bits + s->l2_bits); |
907 |
return 0; |
908 |
} |
909 |
|
910 |
|
911 |
static int qcow_check(BlockDriverState *bs) |
912 |
{ |
913 |
return check_refcounts(bs);
|
914 |
} |
915 |
|
916 |
#if 0
|
917 |
static void dump_refcounts(BlockDriverState *bs)
|
918 |
{
|
919 |
BDRVQcowState *s = bs->opaque;
|
920 |
int64_t nb_clusters, k, k1, size;
|
921 |
int refcount;
|
922 |
|
923 |
size = bdrv_getlength(s->hd);
|
924 |
nb_clusters = size_to_clusters(s, size);
|
925 |
for(k = 0; k < nb_clusters;) {
|
926 |
k1 = k;
|
927 |
refcount = get_refcount(bs, k);
|
928 |
k++;
|
929 |
while (k < nb_clusters && get_refcount(bs, k) == refcount)
|
930 |
k++;
|
931 |
printf("%lld: refcount=%d nb=%lld\n", k, refcount, k - k1);
|
932 |
}
|
933 |
}
|
934 |
#endif
|
935 |
|
936 |
static int qcow_put_buffer(BlockDriverState *bs, const uint8_t *buf, |
937 |
int64_t pos, int size)
|
938 |
{ |
939 |
int growable = bs->growable;
|
940 |
|
941 |
bs->growable = 1;
|
942 |
bdrv_pwrite(bs, pos, buf, size); |
943 |
bs->growable = growable; |
944 |
|
945 |
return size;
|
946 |
} |
947 |
|
948 |
static int qcow_get_buffer(BlockDriverState *bs, uint8_t *buf, |
949 |
int64_t pos, int size)
|
950 |
{ |
951 |
int growable = bs->growable;
|
952 |
int ret;
|
953 |
|
954 |
bs->growable = 1;
|
955 |
ret = bdrv_pread(bs, pos, buf, size); |
956 |
bs->growable = growable; |
957 |
|
958 |
return ret;
|
959 |
} |
960 |
|
961 |
static QEMUOptionParameter qcow_create_options[] = {
|
962 |
{ |
963 |
.name = BLOCK_OPT_SIZE, |
964 |
.type = OPT_SIZE, |
965 |
.help = "Virtual disk size"
|
966 |
}, |
967 |
{ |
968 |
.name = BLOCK_OPT_BACKING_FILE, |
969 |
.type = OPT_STRING, |
970 |
.help = "File name of a base image"
|
971 |
}, |
972 |
{ |
973 |
.name = BLOCK_OPT_BACKING_FMT, |
974 |
.type = OPT_STRING, |
975 |
.help = "Image format of the base image"
|
976 |
}, |
977 |
{ |
978 |
.name = BLOCK_OPT_ENCRYPT, |
979 |
.type = OPT_FLAG, |
980 |
.help = "Encrypt the image"
|
981 |
}, |
982 |
{ |
983 |
.name = BLOCK_OPT_CLUSTER_SIZE, |
984 |
.type = OPT_SIZE, |
985 |
.help = "qcow2 cluster size"
|
986 |
}, |
987 |
{ NULL }
|
988 |
}; |
989 |
|
990 |
static BlockDriver bdrv_qcow2 = {
|
991 |
.format_name = "qcow2",
|
992 |
.instance_size = sizeof(BDRVQcowState),
|
993 |
.bdrv_probe = qcow_probe, |
994 |
.bdrv_open = qcow_open, |
995 |
.bdrv_close = qcow_close, |
996 |
.bdrv_create = qcow_create, |
997 |
.bdrv_flush = qcow_flush, |
998 |
.bdrv_is_allocated = qcow_is_allocated, |
999 |
.bdrv_set_key = qcow_set_key, |
1000 |
.bdrv_make_empty = qcow_make_empty, |
1001 |
|
1002 |
.bdrv_aio_readv = qcow_aio_readv, |
1003 |
.bdrv_aio_writev = qcow_aio_writev, |
1004 |
.bdrv_write_compressed = qcow_write_compressed, |
1005 |
|
1006 |
.bdrv_snapshot_create = qcow_snapshot_create, |
1007 |
.bdrv_snapshot_goto = qcow_snapshot_goto, |
1008 |
.bdrv_snapshot_delete = qcow_snapshot_delete, |
1009 |
.bdrv_snapshot_list = qcow_snapshot_list, |
1010 |
.bdrv_get_info = qcow_get_info, |
1011 |
|
1012 |
.bdrv_put_buffer = qcow_put_buffer, |
1013 |
.bdrv_get_buffer = qcow_get_buffer, |
1014 |
|
1015 |
.create_options = qcow_create_options, |
1016 |
.bdrv_check = qcow_check, |
1017 |
}; |
1018 |
|
1019 |
static void bdrv_qcow2_init(void) |
1020 |
{ |
1021 |
bdrv_register(&bdrv_qcow2); |
1022 |
} |
1023 |
|
1024 |
block_init(bdrv_qcow2_init); |