root / block / qcow2.c @ 6cfcb9b8
History | View | Annotate | Download (52.7 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/block_int.h" |
26 |
#include "qemu/module.h" |
27 |
#include <zlib.h> |
28 |
#include "qemu/aes.h" |
29 |
#include "block/qcow2.h" |
30 |
#include "qemu/error-report.h" |
31 |
#include "qapi/qmp/qerror.h" |
32 |
#include "qapi/qmp/qbool.h" |
33 |
#include "trace.h" |
34 |
|
35 |
/*
|
36 |
Differences with QCOW:
|
37 |
|
38 |
- Support for multiple incremental snapshots.
|
39 |
- Memory management by reference counts.
|
40 |
- Clusters which have a reference count of one have the bit
|
41 |
QCOW_OFLAG_COPIED to optimize write performance.
|
42 |
- Size of compressed clusters is stored in sectors to reduce bit usage
|
43 |
in the cluster offsets.
|
44 |
- Support for storing additional data (such as the VM state) in the
|
45 |
snapshots.
|
46 |
- If a backing store is used, the cluster size is not constrained
|
47 |
(could be backported to QCOW).
|
48 |
- L2 tables have always a size of one cluster.
|
49 |
*/
|
50 |
|
51 |
|
52 |
typedef struct { |
53 |
uint32_t magic; |
54 |
uint32_t len; |
55 |
} QCowExtension; |
56 |
|
57 |
#define QCOW2_EXT_MAGIC_END 0 |
58 |
#define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA |
59 |
#define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857 |
60 |
|
61 |
static int qcow2_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) >= 2)
|
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 qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset, |
82 |
uint64_t end_offset, void **p_feature_table)
|
83 |
{ |
84 |
BDRVQcowState *s = bs->opaque; |
85 |
QCowExtension ext; |
86 |
uint64_t offset; |
87 |
int ret;
|
88 |
|
89 |
#ifdef DEBUG_EXT
|
90 |
printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
|
91 |
#endif
|
92 |
offset = start_offset; |
93 |
while (offset < end_offset) {
|
94 |
|
95 |
#ifdef DEBUG_EXT
|
96 |
/* Sanity check */
|
97 |
if (offset > s->cluster_size)
|
98 |
printf("qcow2_read_extension: suspicious offset %lu\n", offset);
|
99 |
|
100 |
printf("attempting to read extended header in offset %lu\n", offset);
|
101 |
#endif
|
102 |
|
103 |
if (bdrv_pread(bs->file, offset, &ext, sizeof(ext)) != sizeof(ext)) { |
104 |
fprintf(stderr, "qcow2_read_extension: ERROR: "
|
105 |
"pread fail from offset %" PRIu64 "\n", |
106 |
offset); |
107 |
return 1; |
108 |
} |
109 |
be32_to_cpus(&ext.magic); |
110 |
be32_to_cpus(&ext.len); |
111 |
offset += sizeof(ext);
|
112 |
#ifdef DEBUG_EXT
|
113 |
printf("ext.magic = 0x%x\n", ext.magic);
|
114 |
#endif
|
115 |
if (ext.len > end_offset - offset) {
|
116 |
error_report("Header extension too large");
|
117 |
return -EINVAL;
|
118 |
} |
119 |
|
120 |
switch (ext.magic) {
|
121 |
case QCOW2_EXT_MAGIC_END:
|
122 |
return 0; |
123 |
|
124 |
case QCOW2_EXT_MAGIC_BACKING_FORMAT:
|
125 |
if (ext.len >= sizeof(bs->backing_format)) { |
126 |
fprintf(stderr, "ERROR: ext_backing_format: len=%u too large"
|
127 |
" (>=%zu)\n",
|
128 |
ext.len, sizeof(bs->backing_format));
|
129 |
return 2; |
130 |
} |
131 |
if (bdrv_pread(bs->file, offset , bs->backing_format,
|
132 |
ext.len) != ext.len) |
133 |
return 3; |
134 |
bs->backing_format[ext.len] = '\0';
|
135 |
#ifdef DEBUG_EXT
|
136 |
printf("Qcow2: Got format extension %s\n", bs->backing_format);
|
137 |
#endif
|
138 |
break;
|
139 |
|
140 |
case QCOW2_EXT_MAGIC_FEATURE_TABLE:
|
141 |
if (p_feature_table != NULL) { |
142 |
void* feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature)); |
143 |
ret = bdrv_pread(bs->file, offset , feature_table, ext.len); |
144 |
if (ret < 0) { |
145 |
return ret;
|
146 |
} |
147 |
|
148 |
*p_feature_table = feature_table; |
149 |
} |
150 |
break;
|
151 |
|
152 |
default:
|
153 |
/* unknown magic - save it in case we need to rewrite the header */
|
154 |
{ |
155 |
Qcow2UnknownHeaderExtension *uext; |
156 |
|
157 |
uext = g_malloc0(sizeof(*uext) + ext.len);
|
158 |
uext->magic = ext.magic; |
159 |
uext->len = ext.len; |
160 |
QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next); |
161 |
|
162 |
ret = bdrv_pread(bs->file, offset , uext->data, uext->len); |
163 |
if (ret < 0) { |
164 |
return ret;
|
165 |
} |
166 |
} |
167 |
break;
|
168 |
} |
169 |
|
170 |
offset += ((ext.len + 7) & ~7); |
171 |
} |
172 |
|
173 |
return 0; |
174 |
} |
175 |
|
176 |
static void cleanup_unknown_header_ext(BlockDriverState *bs) |
177 |
{ |
178 |
BDRVQcowState *s = bs->opaque; |
179 |
Qcow2UnknownHeaderExtension *uext, *next; |
180 |
|
181 |
QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) { |
182 |
QLIST_REMOVE(uext, next); |
183 |
g_free(uext); |
184 |
} |
185 |
} |
186 |
|
187 |
static void GCC_FMT_ATTR(2, 3) report_unsupported(BlockDriverState *bs, |
188 |
const char *fmt, ...) |
189 |
{ |
190 |
char msg[64]; |
191 |
va_list ap; |
192 |
|
193 |
va_start(ap, fmt); |
194 |
vsnprintf(msg, sizeof(msg), fmt, ap);
|
195 |
va_end(ap); |
196 |
|
197 |
qerror_report(QERR_UNKNOWN_BLOCK_FORMAT_FEATURE, |
198 |
bs->device_name, "qcow2", msg);
|
199 |
} |
200 |
|
201 |
static void report_unsupported_feature(BlockDriverState *bs, |
202 |
Qcow2Feature *table, uint64_t mask) |
203 |
{ |
204 |
while (table && table->name[0] != '\0') { |
205 |
if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) {
|
206 |
if (mask & (1 << table->bit)) { |
207 |
report_unsupported(bs, "%.46s",table->name);
|
208 |
mask &= ~(1 << table->bit);
|
209 |
} |
210 |
} |
211 |
table++; |
212 |
} |
213 |
|
214 |
if (mask) {
|
215 |
report_unsupported(bs, "Unknown incompatible feature: %" PRIx64, mask);
|
216 |
} |
217 |
} |
218 |
|
219 |
/*
|
220 |
* Sets the dirty bit and flushes afterwards if necessary.
|
221 |
*
|
222 |
* The incompatible_features bit is only set if the image file header was
|
223 |
* updated successfully. Therefore it is not required to check the return
|
224 |
* value of this function.
|
225 |
*/
|
226 |
int qcow2_mark_dirty(BlockDriverState *bs)
|
227 |
{ |
228 |
BDRVQcowState *s = bs->opaque; |
229 |
uint64_t val; |
230 |
int ret;
|
231 |
|
232 |
assert(s->qcow_version >= 3);
|
233 |
|
234 |
if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
|
235 |
return 0; /* already dirty */ |
236 |
} |
237 |
|
238 |
val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY); |
239 |
ret = bdrv_pwrite(bs->file, offsetof(QCowHeader, incompatible_features), |
240 |
&val, sizeof(val));
|
241 |
if (ret < 0) { |
242 |
return ret;
|
243 |
} |
244 |
ret = bdrv_flush(bs->file); |
245 |
if (ret < 0) { |
246 |
return ret;
|
247 |
} |
248 |
|
249 |
/* Only treat image as dirty if the header was updated successfully */
|
250 |
s->incompatible_features |= QCOW2_INCOMPAT_DIRTY; |
251 |
return 0; |
252 |
} |
253 |
|
254 |
/*
|
255 |
* Clears the dirty bit and flushes before if necessary. Only call this
|
256 |
* function when there are no pending requests, it does not guard against
|
257 |
* concurrent requests dirtying the image.
|
258 |
*/
|
259 |
static int qcow2_mark_clean(BlockDriverState *bs) |
260 |
{ |
261 |
BDRVQcowState *s = bs->opaque; |
262 |
|
263 |
if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
|
264 |
int ret = bdrv_flush(bs);
|
265 |
if (ret < 0) { |
266 |
return ret;
|
267 |
} |
268 |
|
269 |
s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY; |
270 |
return qcow2_update_header(bs);
|
271 |
} |
272 |
return 0; |
273 |
} |
274 |
|
275 |
static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result, |
276 |
BdrvCheckMode fix) |
277 |
{ |
278 |
int ret = qcow2_check_refcounts(bs, result, fix);
|
279 |
if (ret < 0) { |
280 |
return ret;
|
281 |
} |
282 |
|
283 |
if (fix && result->check_errors == 0 && result->corruptions == 0) { |
284 |
return qcow2_mark_clean(bs);
|
285 |
} |
286 |
return ret;
|
287 |
} |
288 |
|
289 |
static QemuOptsList qcow2_runtime_opts = {
|
290 |
.name = "qcow2",
|
291 |
.head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head), |
292 |
.desc = { |
293 |
{ |
294 |
.name = "lazy_refcounts",
|
295 |
.type = QEMU_OPT_BOOL, |
296 |
.help = "Postpone refcount updates",
|
297 |
}, |
298 |
{ /* end of list */ }
|
299 |
}, |
300 |
}; |
301 |
|
302 |
static int qcow2_open(BlockDriverState *bs, QDict *options, int flags) |
303 |
{ |
304 |
BDRVQcowState *s = bs->opaque; |
305 |
int len, i, ret = 0; |
306 |
QCowHeader header; |
307 |
QemuOpts *opts; |
308 |
Error *local_err = NULL;
|
309 |
uint64_t ext_end; |
310 |
uint64_t l1_vm_state_index; |
311 |
|
312 |
ret = bdrv_pread(bs->file, 0, &header, sizeof(header)); |
313 |
if (ret < 0) { |
314 |
goto fail;
|
315 |
} |
316 |
be32_to_cpus(&header.magic); |
317 |
be32_to_cpus(&header.version); |
318 |
be64_to_cpus(&header.backing_file_offset); |
319 |
be32_to_cpus(&header.backing_file_size); |
320 |
be64_to_cpus(&header.size); |
321 |
be32_to_cpus(&header.cluster_bits); |
322 |
be32_to_cpus(&header.crypt_method); |
323 |
be64_to_cpus(&header.l1_table_offset); |
324 |
be32_to_cpus(&header.l1_size); |
325 |
be64_to_cpus(&header.refcount_table_offset); |
326 |
be32_to_cpus(&header.refcount_table_clusters); |
327 |
be64_to_cpus(&header.snapshots_offset); |
328 |
be32_to_cpus(&header.nb_snapshots); |
329 |
|
330 |
if (header.magic != QCOW_MAGIC) {
|
331 |
ret = -EMEDIUMTYPE; |
332 |
goto fail;
|
333 |
} |
334 |
if (header.version < 2 || header.version > 3) { |
335 |
report_unsupported(bs, "QCOW version %d", header.version);
|
336 |
ret = -ENOTSUP; |
337 |
goto fail;
|
338 |
} |
339 |
|
340 |
s->qcow_version = header.version; |
341 |
|
342 |
/* Initialise version 3 header fields */
|
343 |
if (header.version == 2) { |
344 |
header.incompatible_features = 0;
|
345 |
header.compatible_features = 0;
|
346 |
header.autoclear_features = 0;
|
347 |
header.refcount_order = 4;
|
348 |
header.header_length = 72;
|
349 |
} else {
|
350 |
be64_to_cpus(&header.incompatible_features); |
351 |
be64_to_cpus(&header.compatible_features); |
352 |
be64_to_cpus(&header.autoclear_features); |
353 |
be32_to_cpus(&header.refcount_order); |
354 |
be32_to_cpus(&header.header_length); |
355 |
} |
356 |
|
357 |
if (header.header_length > sizeof(header)) { |
358 |
s->unknown_header_fields_size = header.header_length - sizeof(header);
|
359 |
s->unknown_header_fields = g_malloc(s->unknown_header_fields_size); |
360 |
ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields,
|
361 |
s->unknown_header_fields_size); |
362 |
if (ret < 0) { |
363 |
goto fail;
|
364 |
} |
365 |
} |
366 |
|
367 |
if (header.backing_file_offset) {
|
368 |
ext_end = header.backing_file_offset; |
369 |
} else {
|
370 |
ext_end = 1 << header.cluster_bits;
|
371 |
} |
372 |
|
373 |
/* Handle feature bits */
|
374 |
s->incompatible_features = header.incompatible_features; |
375 |
s->compatible_features = header.compatible_features; |
376 |
s->autoclear_features = header.autoclear_features; |
377 |
|
378 |
if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) {
|
379 |
void *feature_table = NULL; |
380 |
qcow2_read_extensions(bs, header.header_length, ext_end, |
381 |
&feature_table); |
382 |
report_unsupported_feature(bs, feature_table, |
383 |
s->incompatible_features & |
384 |
~QCOW2_INCOMPAT_MASK); |
385 |
ret = -ENOTSUP; |
386 |
goto fail;
|
387 |
} |
388 |
|
389 |
/* Check support for various header values */
|
390 |
if (header.refcount_order != 4) { |
391 |
report_unsupported(bs, "%d bit reference counts",
|
392 |
1 << header.refcount_order);
|
393 |
ret = -ENOTSUP; |
394 |
goto fail;
|
395 |
} |
396 |
|
397 |
if (header.cluster_bits < MIN_CLUSTER_BITS ||
|
398 |
header.cluster_bits > MAX_CLUSTER_BITS) { |
399 |
ret = -EINVAL; |
400 |
goto fail;
|
401 |
} |
402 |
if (header.crypt_method > QCOW_CRYPT_AES) {
|
403 |
ret = -EINVAL; |
404 |
goto fail;
|
405 |
} |
406 |
s->crypt_method_header = header.crypt_method; |
407 |
if (s->crypt_method_header) {
|
408 |
bs->encrypted = 1;
|
409 |
} |
410 |
s->cluster_bits = header.cluster_bits; |
411 |
s->cluster_size = 1 << s->cluster_bits;
|
412 |
s->cluster_sectors = 1 << (s->cluster_bits - 9); |
413 |
s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */ |
414 |
s->l2_size = 1 << s->l2_bits;
|
415 |
bs->total_sectors = header.size / 512;
|
416 |
s->csize_shift = (62 - (s->cluster_bits - 8)); |
417 |
s->csize_mask = (1 << (s->cluster_bits - 8)) - 1; |
418 |
s->cluster_offset_mask = (1LL << s->csize_shift) - 1; |
419 |
s->refcount_table_offset = header.refcount_table_offset; |
420 |
s->refcount_table_size = |
421 |
header.refcount_table_clusters << (s->cluster_bits - 3);
|
422 |
|
423 |
s->snapshots_offset = header.snapshots_offset; |
424 |
s->nb_snapshots = header.nb_snapshots; |
425 |
|
426 |
/* read the level 1 table */
|
427 |
s->l1_size = header.l1_size; |
428 |
|
429 |
l1_vm_state_index = size_to_l1(s, header.size); |
430 |
if (l1_vm_state_index > INT_MAX) {
|
431 |
ret = -EFBIG; |
432 |
goto fail;
|
433 |
} |
434 |
s->l1_vm_state_index = l1_vm_state_index; |
435 |
|
436 |
/* the L1 table must contain at least enough entries to put
|
437 |
header.size bytes */
|
438 |
if (s->l1_size < s->l1_vm_state_index) {
|
439 |
ret = -EINVAL; |
440 |
goto fail;
|
441 |
} |
442 |
s->l1_table_offset = header.l1_table_offset; |
443 |
if (s->l1_size > 0) { |
444 |
s->l1_table = g_malloc0( |
445 |
align_offset(s->l1_size * sizeof(uint64_t), 512)); |
446 |
ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table, |
447 |
s->l1_size * sizeof(uint64_t));
|
448 |
if (ret < 0) { |
449 |
goto fail;
|
450 |
} |
451 |
for(i = 0;i < s->l1_size; i++) { |
452 |
be64_to_cpus(&s->l1_table[i]); |
453 |
} |
454 |
} |
455 |
|
456 |
/* alloc L2 table/refcount block cache */
|
457 |
s->l2_table_cache = qcow2_cache_create(bs, L2_CACHE_SIZE); |
458 |
s->refcount_block_cache = qcow2_cache_create(bs, REFCOUNT_CACHE_SIZE); |
459 |
|
460 |
s->cluster_cache = g_malloc(s->cluster_size); |
461 |
/* one more sector for decompressed data alignment */
|
462 |
s->cluster_data = qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size |
463 |
+ 512);
|
464 |
s->cluster_cache_offset = -1;
|
465 |
s->flags = flags; |
466 |
|
467 |
ret = qcow2_refcount_init(bs); |
468 |
if (ret != 0) { |
469 |
goto fail;
|
470 |
} |
471 |
|
472 |
QLIST_INIT(&s->cluster_allocs); |
473 |
|
474 |
/* read qcow2 extensions */
|
475 |
if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL)) { |
476 |
ret = -EINVAL; |
477 |
goto fail;
|
478 |
} |
479 |
|
480 |
/* read the backing file name */
|
481 |
if (header.backing_file_offset != 0) { |
482 |
len = header.backing_file_size; |
483 |
if (len > 1023) { |
484 |
len = 1023;
|
485 |
} |
486 |
ret = bdrv_pread(bs->file, header.backing_file_offset, |
487 |
bs->backing_file, len); |
488 |
if (ret < 0) { |
489 |
goto fail;
|
490 |
} |
491 |
bs->backing_file[len] = '\0';
|
492 |
} |
493 |
|
494 |
ret = qcow2_read_snapshots(bs); |
495 |
if (ret < 0) { |
496 |
goto fail;
|
497 |
} |
498 |
|
499 |
/* Clear unknown autoclear feature bits */
|
500 |
if (!bs->read_only && s->autoclear_features != 0) { |
501 |
s->autoclear_features = 0;
|
502 |
ret = qcow2_update_header(bs); |
503 |
if (ret < 0) { |
504 |
goto fail;
|
505 |
} |
506 |
} |
507 |
|
508 |
/* Initialise locks */
|
509 |
qemu_co_mutex_init(&s->lock); |
510 |
|
511 |
/* Repair image if dirty */
|
512 |
if (!(flags & BDRV_O_CHECK) && !bs->read_only &&
|
513 |
(s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) { |
514 |
BdrvCheckResult result = {0};
|
515 |
|
516 |
ret = qcow2_check(bs, &result, BDRV_FIX_ERRORS); |
517 |
if (ret < 0) { |
518 |
goto fail;
|
519 |
} |
520 |
} |
521 |
|
522 |
/* Enable lazy_refcounts according to image and command line options */
|
523 |
opts = qemu_opts_create_nofail(&qcow2_runtime_opts); |
524 |
qemu_opts_absorb_qdict(opts, options, &local_err); |
525 |
if (error_is_set(&local_err)) {
|
526 |
qerror_report_err(local_err); |
527 |
error_free(local_err); |
528 |
ret = -EINVAL; |
529 |
goto fail;
|
530 |
} |
531 |
|
532 |
s->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS, |
533 |
(s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS)); |
534 |
|
535 |
qemu_opts_del(opts); |
536 |
|
537 |
if (s->use_lazy_refcounts && s->qcow_version < 3) { |
538 |
qerror_report(ERROR_CLASS_GENERIC_ERROR, "Lazy refcounts require "
|
539 |
"a qcow2 image with at least qemu 1.1 compatibility level");
|
540 |
ret = -EINVAL; |
541 |
goto fail;
|
542 |
} |
543 |
|
544 |
#ifdef DEBUG_ALLOC
|
545 |
{ |
546 |
BdrvCheckResult result = {0};
|
547 |
qcow2_check_refcounts(bs, &result, 0);
|
548 |
} |
549 |
#endif
|
550 |
return ret;
|
551 |
|
552 |
fail:
|
553 |
g_free(s->unknown_header_fields); |
554 |
cleanup_unknown_header_ext(bs); |
555 |
qcow2_free_snapshots(bs); |
556 |
qcow2_refcount_close(bs); |
557 |
g_free(s->l1_table); |
558 |
if (s->l2_table_cache) {
|
559 |
qcow2_cache_destroy(bs, s->l2_table_cache); |
560 |
} |
561 |
g_free(s->cluster_cache); |
562 |
qemu_vfree(s->cluster_data); |
563 |
return ret;
|
564 |
} |
565 |
|
566 |
static int qcow2_set_key(BlockDriverState *bs, const char *key) |
567 |
{ |
568 |
BDRVQcowState *s = bs->opaque; |
569 |
uint8_t keybuf[16];
|
570 |
int len, i;
|
571 |
|
572 |
memset(keybuf, 0, 16); |
573 |
len = strlen(key); |
574 |
if (len > 16) |
575 |
len = 16;
|
576 |
/* XXX: we could compress the chars to 7 bits to increase
|
577 |
entropy */
|
578 |
for(i = 0;i < len;i++) { |
579 |
keybuf[i] = key[i]; |
580 |
} |
581 |
s->crypt_method = s->crypt_method_header; |
582 |
|
583 |
if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0) |
584 |
return -1; |
585 |
if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0) |
586 |
return -1; |
587 |
#if 0
|
588 |
/* test */
|
589 |
{
|
590 |
uint8_t in[16];
|
591 |
uint8_t out[16];
|
592 |
uint8_t tmp[16];
|
593 |
for(i=0;i<16;i++)
|
594 |
in[i] = i;
|
595 |
AES_encrypt(in, tmp, &s->aes_encrypt_key);
|
596 |
AES_decrypt(tmp, out, &s->aes_decrypt_key);
|
597 |
for(i = 0; i < 16; i++)
|
598 |
printf(" %02x", tmp[i]);
|
599 |
printf("\n");
|
600 |
for(i = 0; i < 16; i++)
|
601 |
printf(" %02x", out[i]);
|
602 |
printf("\n");
|
603 |
}
|
604 |
#endif
|
605 |
return 0; |
606 |
} |
607 |
|
608 |
/* We have nothing to do for QCOW2 reopen, stubs just return
|
609 |
* success */
|
610 |
static int qcow2_reopen_prepare(BDRVReopenState *state, |
611 |
BlockReopenQueue *queue, Error **errp) |
612 |
{ |
613 |
return 0; |
614 |
} |
615 |
|
616 |
static int coroutine_fn qcow2_co_is_allocated(BlockDriverState *bs, |
617 |
int64_t sector_num, int nb_sectors, int *pnum) |
618 |
{ |
619 |
BDRVQcowState *s = bs->opaque; |
620 |
uint64_t cluster_offset; |
621 |
int ret;
|
622 |
|
623 |
*pnum = nb_sectors; |
624 |
/* FIXME We can get errors here, but the bdrv_co_is_allocated interface
|
625 |
* can't pass them on today */
|
626 |
qemu_co_mutex_lock(&s->lock); |
627 |
ret = qcow2_get_cluster_offset(bs, sector_num << 9, pnum, &cluster_offset);
|
628 |
qemu_co_mutex_unlock(&s->lock); |
629 |
if (ret < 0) { |
630 |
*pnum = 0;
|
631 |
} |
632 |
|
633 |
return (cluster_offset != 0) || (ret == QCOW2_CLUSTER_ZERO); |
634 |
} |
635 |
|
636 |
/* handle reading after the end of the backing file */
|
637 |
int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
|
638 |
int64_t sector_num, int nb_sectors)
|
639 |
{ |
640 |
int n1;
|
641 |
if ((sector_num + nb_sectors) <= bs->total_sectors)
|
642 |
return nb_sectors;
|
643 |
if (sector_num >= bs->total_sectors)
|
644 |
n1 = 0;
|
645 |
else
|
646 |
n1 = bs->total_sectors - sector_num; |
647 |
|
648 |
qemu_iovec_memset(qiov, 512 * n1, 0, 512 * (nb_sectors - n1)); |
649 |
|
650 |
return n1;
|
651 |
} |
652 |
|
653 |
static coroutine_fn int qcow2_co_readv(BlockDriverState *bs, int64_t sector_num, |
654 |
int remaining_sectors, QEMUIOVector *qiov)
|
655 |
{ |
656 |
BDRVQcowState *s = bs->opaque; |
657 |
int index_in_cluster, n1;
|
658 |
int ret;
|
659 |
int cur_nr_sectors; /* number of sectors in current iteration */ |
660 |
uint64_t cluster_offset = 0;
|
661 |
uint64_t bytes_done = 0;
|
662 |
QEMUIOVector hd_qiov; |
663 |
uint8_t *cluster_data = NULL;
|
664 |
|
665 |
qemu_iovec_init(&hd_qiov, qiov->niov); |
666 |
|
667 |
qemu_co_mutex_lock(&s->lock); |
668 |
|
669 |
while (remaining_sectors != 0) { |
670 |
|
671 |
/* prepare next request */
|
672 |
cur_nr_sectors = remaining_sectors; |
673 |
if (s->crypt_method) {
|
674 |
cur_nr_sectors = MIN(cur_nr_sectors, |
675 |
QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors); |
676 |
} |
677 |
|
678 |
ret = qcow2_get_cluster_offset(bs, sector_num << 9,
|
679 |
&cur_nr_sectors, &cluster_offset); |
680 |
if (ret < 0) { |
681 |
goto fail;
|
682 |
} |
683 |
|
684 |
index_in_cluster = sector_num & (s->cluster_sectors - 1);
|
685 |
|
686 |
qemu_iovec_reset(&hd_qiov); |
687 |
qemu_iovec_concat(&hd_qiov, qiov, bytes_done, |
688 |
cur_nr_sectors * 512);
|
689 |
|
690 |
switch (ret) {
|
691 |
case QCOW2_CLUSTER_UNALLOCATED:
|
692 |
|
693 |
if (bs->backing_hd) {
|
694 |
/* read from the base image */
|
695 |
n1 = qcow2_backing_read1(bs->backing_hd, &hd_qiov, |
696 |
sector_num, cur_nr_sectors); |
697 |
if (n1 > 0) { |
698 |
BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO); |
699 |
qemu_co_mutex_unlock(&s->lock); |
700 |
ret = bdrv_co_readv(bs->backing_hd, sector_num, |
701 |
n1, &hd_qiov); |
702 |
qemu_co_mutex_lock(&s->lock); |
703 |
if (ret < 0) { |
704 |
goto fail;
|
705 |
} |
706 |
} |
707 |
} else {
|
708 |
/* Note: in this case, no need to wait */
|
709 |
qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors); |
710 |
} |
711 |
break;
|
712 |
|
713 |
case QCOW2_CLUSTER_ZERO:
|
714 |
qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors); |
715 |
break;
|
716 |
|
717 |
case QCOW2_CLUSTER_COMPRESSED:
|
718 |
/* add AIO support for compressed blocks ? */
|
719 |
ret = qcow2_decompress_cluster(bs, cluster_offset); |
720 |
if (ret < 0) { |
721 |
goto fail;
|
722 |
} |
723 |
|
724 |
qemu_iovec_from_buf(&hd_qiov, 0,
|
725 |
s->cluster_cache + index_in_cluster * 512,
|
726 |
512 * cur_nr_sectors);
|
727 |
break;
|
728 |
|
729 |
case QCOW2_CLUSTER_NORMAL:
|
730 |
if ((cluster_offset & 511) != 0) { |
731 |
ret = -EIO; |
732 |
goto fail;
|
733 |
} |
734 |
|
735 |
if (s->crypt_method) {
|
736 |
/*
|
737 |
* For encrypted images, read everything into a temporary
|
738 |
* contiguous buffer on which the AES functions can work.
|
739 |
*/
|
740 |
if (!cluster_data) {
|
741 |
cluster_data = |
742 |
qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); |
743 |
} |
744 |
|
745 |
assert(cur_nr_sectors <= |
746 |
QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors); |
747 |
qemu_iovec_reset(&hd_qiov); |
748 |
qemu_iovec_add(&hd_qiov, cluster_data, |
749 |
512 * cur_nr_sectors);
|
750 |
} |
751 |
|
752 |
BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); |
753 |
qemu_co_mutex_unlock(&s->lock); |
754 |
ret = bdrv_co_readv(bs->file, |
755 |
(cluster_offset >> 9) + index_in_cluster,
|
756 |
cur_nr_sectors, &hd_qiov); |
757 |
qemu_co_mutex_lock(&s->lock); |
758 |
if (ret < 0) { |
759 |
goto fail;
|
760 |
} |
761 |
if (s->crypt_method) {
|
762 |
qcow2_encrypt_sectors(s, sector_num, cluster_data, |
763 |
cluster_data, cur_nr_sectors, 0, &s->aes_decrypt_key);
|
764 |
qemu_iovec_from_buf(qiov, bytes_done, |
765 |
cluster_data, 512 * cur_nr_sectors);
|
766 |
} |
767 |
break;
|
768 |
|
769 |
default:
|
770 |
g_assert_not_reached(); |
771 |
ret = -EIO; |
772 |
goto fail;
|
773 |
} |
774 |
|
775 |
remaining_sectors -= cur_nr_sectors; |
776 |
sector_num += cur_nr_sectors; |
777 |
bytes_done += cur_nr_sectors * 512;
|
778 |
} |
779 |
ret = 0;
|
780 |
|
781 |
fail:
|
782 |
qemu_co_mutex_unlock(&s->lock); |
783 |
|
784 |
qemu_iovec_destroy(&hd_qiov); |
785 |
qemu_vfree(cluster_data); |
786 |
|
787 |
return ret;
|
788 |
} |
789 |
|
790 |
static coroutine_fn int qcow2_co_writev(BlockDriverState *bs, |
791 |
int64_t sector_num, |
792 |
int remaining_sectors,
|
793 |
QEMUIOVector *qiov) |
794 |
{ |
795 |
BDRVQcowState *s = bs->opaque; |
796 |
int index_in_cluster;
|
797 |
int n_end;
|
798 |
int ret;
|
799 |
int cur_nr_sectors; /* number of sectors in current iteration */ |
800 |
uint64_t cluster_offset; |
801 |
QEMUIOVector hd_qiov; |
802 |
uint64_t bytes_done = 0;
|
803 |
uint8_t *cluster_data = NULL;
|
804 |
QCowL2Meta *l2meta = NULL;
|
805 |
|
806 |
trace_qcow2_writev_start_req(qemu_coroutine_self(), sector_num, |
807 |
remaining_sectors); |
808 |
|
809 |
qemu_iovec_init(&hd_qiov, qiov->niov); |
810 |
|
811 |
s->cluster_cache_offset = -1; /* disable compressed cache */ |
812 |
|
813 |
qemu_co_mutex_lock(&s->lock); |
814 |
|
815 |
while (remaining_sectors != 0) { |
816 |
|
817 |
l2meta = NULL;
|
818 |
|
819 |
trace_qcow2_writev_start_part(qemu_coroutine_self()); |
820 |
index_in_cluster = sector_num & (s->cluster_sectors - 1);
|
821 |
n_end = index_in_cluster + remaining_sectors; |
822 |
if (s->crypt_method &&
|
823 |
n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors) { |
824 |
n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors; |
825 |
} |
826 |
|
827 |
ret = qcow2_alloc_cluster_offset(bs, sector_num << 9,
|
828 |
index_in_cluster, n_end, &cur_nr_sectors, &cluster_offset, &l2meta); |
829 |
if (ret < 0) { |
830 |
goto fail;
|
831 |
} |
832 |
|
833 |
assert((cluster_offset & 511) == 0); |
834 |
|
835 |
qemu_iovec_reset(&hd_qiov); |
836 |
qemu_iovec_concat(&hd_qiov, qiov, bytes_done, |
837 |
cur_nr_sectors * 512);
|
838 |
|
839 |
if (s->crypt_method) {
|
840 |
if (!cluster_data) {
|
841 |
cluster_data = qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS * |
842 |
s->cluster_size); |
843 |
} |
844 |
|
845 |
assert(hd_qiov.size <= |
846 |
QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); |
847 |
qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size);
|
848 |
|
849 |
qcow2_encrypt_sectors(s, sector_num, cluster_data, |
850 |
cluster_data, cur_nr_sectors, 1, &s->aes_encrypt_key);
|
851 |
|
852 |
qemu_iovec_reset(&hd_qiov); |
853 |
qemu_iovec_add(&hd_qiov, cluster_data, |
854 |
cur_nr_sectors * 512);
|
855 |
} |
856 |
|
857 |
qemu_co_mutex_unlock(&s->lock); |
858 |
BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); |
859 |
trace_qcow2_writev_data(qemu_coroutine_self(), |
860 |
(cluster_offset >> 9) + index_in_cluster);
|
861 |
ret = bdrv_co_writev(bs->file, |
862 |
(cluster_offset >> 9) + index_in_cluster,
|
863 |
cur_nr_sectors, &hd_qiov); |
864 |
qemu_co_mutex_lock(&s->lock); |
865 |
if (ret < 0) { |
866 |
goto fail;
|
867 |
} |
868 |
|
869 |
while (l2meta != NULL) { |
870 |
QCowL2Meta *next; |
871 |
|
872 |
ret = qcow2_alloc_cluster_link_l2(bs, l2meta); |
873 |
if (ret < 0) { |
874 |
goto fail;
|
875 |
} |
876 |
|
877 |
/* Take the request off the list of running requests */
|
878 |
if (l2meta->nb_clusters != 0) { |
879 |
QLIST_REMOVE(l2meta, next_in_flight); |
880 |
} |
881 |
|
882 |
qemu_co_queue_restart_all(&l2meta->dependent_requests); |
883 |
|
884 |
next = l2meta->next; |
885 |
g_free(l2meta); |
886 |
l2meta = next; |
887 |
} |
888 |
|
889 |
remaining_sectors -= cur_nr_sectors; |
890 |
sector_num += cur_nr_sectors; |
891 |
bytes_done += cur_nr_sectors * 512;
|
892 |
trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_nr_sectors); |
893 |
} |
894 |
ret = 0;
|
895 |
|
896 |
fail:
|
897 |
qemu_co_mutex_unlock(&s->lock); |
898 |
|
899 |
while (l2meta != NULL) { |
900 |
QCowL2Meta *next; |
901 |
|
902 |
if (l2meta->nb_clusters != 0) { |
903 |
QLIST_REMOVE(l2meta, next_in_flight); |
904 |
} |
905 |
qemu_co_queue_restart_all(&l2meta->dependent_requests); |
906 |
|
907 |
next = l2meta->next; |
908 |
g_free(l2meta); |
909 |
l2meta = next; |
910 |
} |
911 |
|
912 |
qemu_iovec_destroy(&hd_qiov); |
913 |
qemu_vfree(cluster_data); |
914 |
trace_qcow2_writev_done_req(qemu_coroutine_self(), ret); |
915 |
|
916 |
return ret;
|
917 |
} |
918 |
|
919 |
static void qcow2_close(BlockDriverState *bs) |
920 |
{ |
921 |
BDRVQcowState *s = bs->opaque; |
922 |
g_free(s->l1_table); |
923 |
|
924 |
qcow2_cache_flush(bs, s->l2_table_cache); |
925 |
qcow2_cache_flush(bs, s->refcount_block_cache); |
926 |
|
927 |
qcow2_mark_clean(bs); |
928 |
|
929 |
qcow2_cache_destroy(bs, s->l2_table_cache); |
930 |
qcow2_cache_destroy(bs, s->refcount_block_cache); |
931 |
|
932 |
g_free(s->unknown_header_fields); |
933 |
cleanup_unknown_header_ext(bs); |
934 |
|
935 |
g_free(s->cluster_cache); |
936 |
qemu_vfree(s->cluster_data); |
937 |
qcow2_refcount_close(bs); |
938 |
qcow2_free_snapshots(bs); |
939 |
} |
940 |
|
941 |
static void qcow2_invalidate_cache(BlockDriverState *bs) |
942 |
{ |
943 |
BDRVQcowState *s = bs->opaque; |
944 |
int flags = s->flags;
|
945 |
AES_KEY aes_encrypt_key; |
946 |
AES_KEY aes_decrypt_key; |
947 |
uint32_t crypt_method = 0;
|
948 |
QDict *options; |
949 |
|
950 |
/*
|
951 |
* Backing files are read-only which makes all of their metadata immutable,
|
952 |
* that means we don't have to worry about reopening them here.
|
953 |
*/
|
954 |
|
955 |
if (s->crypt_method) {
|
956 |
crypt_method = s->crypt_method; |
957 |
memcpy(&aes_encrypt_key, &s->aes_encrypt_key, sizeof(aes_encrypt_key));
|
958 |
memcpy(&aes_decrypt_key, &s->aes_decrypt_key, sizeof(aes_decrypt_key));
|
959 |
} |
960 |
|
961 |
qcow2_close(bs); |
962 |
|
963 |
options = qdict_new(); |
964 |
qdict_put(options, QCOW2_OPT_LAZY_REFCOUNTS, |
965 |
qbool_from_int(s->use_lazy_refcounts)); |
966 |
|
967 |
memset(s, 0, sizeof(BDRVQcowState)); |
968 |
qcow2_open(bs, options, flags); |
969 |
|
970 |
QDECREF(options); |
971 |
|
972 |
if (crypt_method) {
|
973 |
s->crypt_method = crypt_method; |
974 |
memcpy(&s->aes_encrypt_key, &aes_encrypt_key, sizeof(aes_encrypt_key));
|
975 |
memcpy(&s->aes_decrypt_key, &aes_decrypt_key, sizeof(aes_decrypt_key));
|
976 |
} |
977 |
} |
978 |
|
979 |
static size_t header_ext_add(char *buf, uint32_t magic, const void *s, |
980 |
size_t len, size_t buflen) |
981 |
{ |
982 |
QCowExtension *ext_backing_fmt = (QCowExtension*) buf; |
983 |
size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7); |
984 |
|
985 |
if (buflen < ext_len) {
|
986 |
return -ENOSPC;
|
987 |
} |
988 |
|
989 |
*ext_backing_fmt = (QCowExtension) { |
990 |
.magic = cpu_to_be32(magic), |
991 |
.len = cpu_to_be32(len), |
992 |
}; |
993 |
memcpy(buf + sizeof(QCowExtension), s, len);
|
994 |
|
995 |
return ext_len;
|
996 |
} |
997 |
|
998 |
/*
|
999 |
* Updates the qcow2 header, including the variable length parts of it, i.e.
|
1000 |
* the backing file name and all extensions. qcow2 was not designed to allow
|
1001 |
* such changes, so if we run out of space (we can only use the first cluster)
|
1002 |
* this function may fail.
|
1003 |
*
|
1004 |
* Returns 0 on success, -errno in error cases.
|
1005 |
*/
|
1006 |
int qcow2_update_header(BlockDriverState *bs)
|
1007 |
{ |
1008 |
BDRVQcowState *s = bs->opaque; |
1009 |
QCowHeader *header; |
1010 |
char *buf;
|
1011 |
size_t buflen = s->cluster_size; |
1012 |
int ret;
|
1013 |
uint64_t total_size; |
1014 |
uint32_t refcount_table_clusters; |
1015 |
size_t header_length; |
1016 |
Qcow2UnknownHeaderExtension *uext; |
1017 |
|
1018 |
buf = qemu_blockalign(bs, buflen); |
1019 |
|
1020 |
/* Header structure */
|
1021 |
header = (QCowHeader*) buf; |
1022 |
|
1023 |
if (buflen < sizeof(*header)) { |
1024 |
ret = -ENOSPC; |
1025 |
goto fail;
|
1026 |
} |
1027 |
|
1028 |
header_length = sizeof(*header) + s->unknown_header_fields_size;
|
1029 |
total_size = bs->total_sectors * BDRV_SECTOR_SIZE; |
1030 |
refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
|
1031 |
|
1032 |
*header = (QCowHeader) { |
1033 |
/* Version 2 fields */
|
1034 |
.magic = cpu_to_be32(QCOW_MAGIC), |
1035 |
.version = cpu_to_be32(s->qcow_version), |
1036 |
.backing_file_offset = 0,
|
1037 |
.backing_file_size = 0,
|
1038 |
.cluster_bits = cpu_to_be32(s->cluster_bits), |
1039 |
.size = cpu_to_be64(total_size), |
1040 |
.crypt_method = cpu_to_be32(s->crypt_method_header), |
1041 |
.l1_size = cpu_to_be32(s->l1_size), |
1042 |
.l1_table_offset = cpu_to_be64(s->l1_table_offset), |
1043 |
.refcount_table_offset = cpu_to_be64(s->refcount_table_offset), |
1044 |
.refcount_table_clusters = cpu_to_be32(refcount_table_clusters), |
1045 |
.nb_snapshots = cpu_to_be32(s->nb_snapshots), |
1046 |
.snapshots_offset = cpu_to_be64(s->snapshots_offset), |
1047 |
|
1048 |
/* Version 3 fields */
|
1049 |
.incompatible_features = cpu_to_be64(s->incompatible_features), |
1050 |
.compatible_features = cpu_to_be64(s->compatible_features), |
1051 |
.autoclear_features = cpu_to_be64(s->autoclear_features), |
1052 |
.refcount_order = cpu_to_be32(3 + REFCOUNT_SHIFT),
|
1053 |
.header_length = cpu_to_be32(header_length), |
1054 |
}; |
1055 |
|
1056 |
/* For older versions, write a shorter header */
|
1057 |
switch (s->qcow_version) {
|
1058 |
case 2: |
1059 |
ret = offsetof(QCowHeader, incompatible_features); |
1060 |
break;
|
1061 |
case 3: |
1062 |
ret = sizeof(*header);
|
1063 |
break;
|
1064 |
default:
|
1065 |
ret = -EINVAL; |
1066 |
goto fail;
|
1067 |
} |
1068 |
|
1069 |
buf += ret; |
1070 |
buflen -= ret; |
1071 |
memset(buf, 0, buflen);
|
1072 |
|
1073 |
/* Preserve any unknown field in the header */
|
1074 |
if (s->unknown_header_fields_size) {
|
1075 |
if (buflen < s->unknown_header_fields_size) {
|
1076 |
ret = -ENOSPC; |
1077 |
goto fail;
|
1078 |
} |
1079 |
|
1080 |
memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size); |
1081 |
buf += s->unknown_header_fields_size; |
1082 |
buflen -= s->unknown_header_fields_size; |
1083 |
} |
1084 |
|
1085 |
/* Backing file format header extension */
|
1086 |
if (*bs->backing_format) {
|
1087 |
ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT, |
1088 |
bs->backing_format, strlen(bs->backing_format), |
1089 |
buflen); |
1090 |
if (ret < 0) { |
1091 |
goto fail;
|
1092 |
} |
1093 |
|
1094 |
buf += ret; |
1095 |
buflen -= ret; |
1096 |
} |
1097 |
|
1098 |
/* Feature table */
|
1099 |
Qcow2Feature features[] = { |
1100 |
{ |
1101 |
.type = QCOW2_FEAT_TYPE_INCOMPATIBLE, |
1102 |
.bit = QCOW2_INCOMPAT_DIRTY_BITNR, |
1103 |
.name = "dirty bit",
|
1104 |
}, |
1105 |
{ |
1106 |
.type = QCOW2_FEAT_TYPE_COMPATIBLE, |
1107 |
.bit = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR, |
1108 |
.name = "lazy refcounts",
|
1109 |
}, |
1110 |
}; |
1111 |
|
1112 |
ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE, |
1113 |
features, sizeof(features), buflen);
|
1114 |
if (ret < 0) { |
1115 |
goto fail;
|
1116 |
} |
1117 |
buf += ret; |
1118 |
buflen -= ret; |
1119 |
|
1120 |
/* Keep unknown header extensions */
|
1121 |
QLIST_FOREACH(uext, &s->unknown_header_ext, next) { |
1122 |
ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen); |
1123 |
if (ret < 0) { |
1124 |
goto fail;
|
1125 |
} |
1126 |
|
1127 |
buf += ret; |
1128 |
buflen -= ret; |
1129 |
} |
1130 |
|
1131 |
/* End of header extensions */
|
1132 |
ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen); |
1133 |
if (ret < 0) { |
1134 |
goto fail;
|
1135 |
} |
1136 |
|
1137 |
buf += ret; |
1138 |
buflen -= ret; |
1139 |
|
1140 |
/* Backing file name */
|
1141 |
if (*bs->backing_file) {
|
1142 |
size_t backing_file_len = strlen(bs->backing_file); |
1143 |
|
1144 |
if (buflen < backing_file_len) {
|
1145 |
ret = -ENOSPC; |
1146 |
goto fail;
|
1147 |
} |
1148 |
|
1149 |
/* Using strncpy is ok here, since buf is not NUL-terminated. */
|
1150 |
strncpy(buf, bs->backing_file, buflen); |
1151 |
|
1152 |
header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
|
1153 |
header->backing_file_size = cpu_to_be32(backing_file_len); |
1154 |
} |
1155 |
|
1156 |
/* Write the new header */
|
1157 |
ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size);
|
1158 |
if (ret < 0) { |
1159 |
goto fail;
|
1160 |
} |
1161 |
|
1162 |
ret = 0;
|
1163 |
fail:
|
1164 |
qemu_vfree(header); |
1165 |
return ret;
|
1166 |
} |
1167 |
|
1168 |
static int qcow2_change_backing_file(BlockDriverState *bs, |
1169 |
const char *backing_file, const char *backing_fmt) |
1170 |
{ |
1171 |
pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: ""); |
1172 |
pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: ""); |
1173 |
|
1174 |
return qcow2_update_header(bs);
|
1175 |
} |
1176 |
|
1177 |
static int preallocate(BlockDriverState *bs) |
1178 |
{ |
1179 |
uint64_t nb_sectors; |
1180 |
uint64_t offset; |
1181 |
uint64_t host_offset = 0;
|
1182 |
int num;
|
1183 |
int ret;
|
1184 |
QCowL2Meta *meta; |
1185 |
|
1186 |
nb_sectors = bdrv_getlength(bs) >> 9;
|
1187 |
offset = 0;
|
1188 |
|
1189 |
while (nb_sectors) {
|
1190 |
num = MIN(nb_sectors, INT_MAX >> 9);
|
1191 |
ret = qcow2_alloc_cluster_offset(bs, offset, 0, num, &num,
|
1192 |
&host_offset, &meta); |
1193 |
if (ret < 0) { |
1194 |
return ret;
|
1195 |
} |
1196 |
|
1197 |
ret = qcow2_alloc_cluster_link_l2(bs, meta); |
1198 |
if (ret < 0) { |
1199 |
qcow2_free_any_clusters(bs, meta->alloc_offset, meta->nb_clusters, |
1200 |
QCOW2_DISCARD_NEVER); |
1201 |
return ret;
|
1202 |
} |
1203 |
|
1204 |
/* There are no dependent requests, but we need to remove our request
|
1205 |
* from the list of in-flight requests */
|
1206 |
if (meta != NULL) { |
1207 |
QLIST_REMOVE(meta, next_in_flight); |
1208 |
} |
1209 |
|
1210 |
/* TODO Preallocate data if requested */
|
1211 |
|
1212 |
nb_sectors -= num; |
1213 |
offset += num << 9;
|
1214 |
} |
1215 |
|
1216 |
/*
|
1217 |
* It is expected that the image file is large enough to actually contain
|
1218 |
* all of the allocated clusters (otherwise we get failing reads after
|
1219 |
* EOF). Extend the image to the last allocated sector.
|
1220 |
*/
|
1221 |
if (host_offset != 0) { |
1222 |
uint8_t buf[512];
|
1223 |
memset(buf, 0, 512); |
1224 |
ret = bdrv_write(bs->file, (host_offset >> 9) + num - 1, buf, 1); |
1225 |
if (ret < 0) { |
1226 |
return ret;
|
1227 |
} |
1228 |
} |
1229 |
|
1230 |
return 0; |
1231 |
} |
1232 |
|
1233 |
static int qcow2_create2(const char *filename, int64_t total_size, |
1234 |
const char *backing_file, const char *backing_format, |
1235 |
int flags, size_t cluster_size, int prealloc, |
1236 |
QEMUOptionParameter *options, int version)
|
1237 |
{ |
1238 |
/* Calculate cluster_bits */
|
1239 |
int cluster_bits;
|
1240 |
cluster_bits = ffs(cluster_size) - 1;
|
1241 |
if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
|
1242 |
(1 << cluster_bits) != cluster_size)
|
1243 |
{ |
1244 |
error_report( |
1245 |
"Cluster size must be a power of two between %d and %dk",
|
1246 |
1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10)); |
1247 |
return -EINVAL;
|
1248 |
} |
1249 |
|
1250 |
/*
|
1251 |
* Open the image file and write a minimal qcow2 header.
|
1252 |
*
|
1253 |
* We keep things simple and start with a zero-sized image. We also
|
1254 |
* do without refcount blocks or a L1 table for now. We'll fix the
|
1255 |
* inconsistency later.
|
1256 |
*
|
1257 |
* We do need a refcount table because growing the refcount table means
|
1258 |
* allocating two new refcount blocks - the seconds of which would be at
|
1259 |
* 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
|
1260 |
* size for any qcow2 image.
|
1261 |
*/
|
1262 |
BlockDriverState* bs; |
1263 |
QCowHeader header; |
1264 |
uint8_t* refcount_table; |
1265 |
int ret;
|
1266 |
|
1267 |
ret = bdrv_create_file(filename, options); |
1268 |
if (ret < 0) { |
1269 |
return ret;
|
1270 |
} |
1271 |
|
1272 |
ret = bdrv_file_open(&bs, filename, NULL, BDRV_O_RDWR);
|
1273 |
if (ret < 0) { |
1274 |
return ret;
|
1275 |
} |
1276 |
|
1277 |
/* Write the header */
|
1278 |
memset(&header, 0, sizeof(header)); |
1279 |
header.magic = cpu_to_be32(QCOW_MAGIC); |
1280 |
header.version = cpu_to_be32(version); |
1281 |
header.cluster_bits = cpu_to_be32(cluster_bits); |
1282 |
header.size = cpu_to_be64(0);
|
1283 |
header.l1_table_offset = cpu_to_be64(0);
|
1284 |
header.l1_size = cpu_to_be32(0);
|
1285 |
header.refcount_table_offset = cpu_to_be64(cluster_size); |
1286 |
header.refcount_table_clusters = cpu_to_be32(1);
|
1287 |
header.refcount_order = cpu_to_be32(3 + REFCOUNT_SHIFT);
|
1288 |
header.header_length = cpu_to_be32(sizeof(header));
|
1289 |
|
1290 |
if (flags & BLOCK_FLAG_ENCRYPT) {
|
1291 |
header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES); |
1292 |
} else {
|
1293 |
header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE); |
1294 |
} |
1295 |
|
1296 |
if (flags & BLOCK_FLAG_LAZY_REFCOUNTS) {
|
1297 |
header.compatible_features |= |
1298 |
cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS); |
1299 |
} |
1300 |
|
1301 |
ret = bdrv_pwrite(bs, 0, &header, sizeof(header)); |
1302 |
if (ret < 0) { |
1303 |
goto out;
|
1304 |
} |
1305 |
|
1306 |
/* Write an empty refcount table */
|
1307 |
refcount_table = g_malloc0(cluster_size); |
1308 |
ret = bdrv_pwrite(bs, cluster_size, refcount_table, cluster_size); |
1309 |
g_free(refcount_table); |
1310 |
|
1311 |
if (ret < 0) { |
1312 |
goto out;
|
1313 |
} |
1314 |
|
1315 |
bdrv_close(bs); |
1316 |
|
1317 |
/*
|
1318 |
* And now open the image and make it consistent first (i.e. increase the
|
1319 |
* refcount of the cluster that is occupied by the header and the refcount
|
1320 |
* table)
|
1321 |
*/
|
1322 |
BlockDriver* drv = bdrv_find_format("qcow2");
|
1323 |
assert(drv != NULL);
|
1324 |
ret = bdrv_open(bs, filename, NULL,
|
1325 |
BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_FLUSH, drv); |
1326 |
if (ret < 0) { |
1327 |
goto out;
|
1328 |
} |
1329 |
|
1330 |
ret = qcow2_alloc_clusters(bs, 2 * cluster_size);
|
1331 |
if (ret < 0) { |
1332 |
goto out;
|
1333 |
|
1334 |
} else if (ret != 0) { |
1335 |
error_report("Huh, first cluster in empty image is already in use?");
|
1336 |
abort(); |
1337 |
} |
1338 |
|
1339 |
/* Okay, now that we have a valid image, let's give it the right size */
|
1340 |
ret = bdrv_truncate(bs, total_size * BDRV_SECTOR_SIZE); |
1341 |
if (ret < 0) { |
1342 |
goto out;
|
1343 |
} |
1344 |
|
1345 |
/* Want a backing file? There you go.*/
|
1346 |
if (backing_file) {
|
1347 |
ret = bdrv_change_backing_file(bs, backing_file, backing_format); |
1348 |
if (ret < 0) { |
1349 |
goto out;
|
1350 |
} |
1351 |
} |
1352 |
|
1353 |
/* And if we're supposed to preallocate metadata, do that now */
|
1354 |
if (prealloc) {
|
1355 |
BDRVQcowState *s = bs->opaque; |
1356 |
qemu_co_mutex_lock(&s->lock); |
1357 |
ret = preallocate(bs); |
1358 |
qemu_co_mutex_unlock(&s->lock); |
1359 |
if (ret < 0) { |
1360 |
goto out;
|
1361 |
} |
1362 |
} |
1363 |
|
1364 |
ret = 0;
|
1365 |
out:
|
1366 |
bdrv_delete(bs); |
1367 |
return ret;
|
1368 |
} |
1369 |
|
1370 |
static int qcow2_create(const char *filename, QEMUOptionParameter *options) |
1371 |
{ |
1372 |
const char *backing_file = NULL; |
1373 |
const char *backing_fmt = NULL; |
1374 |
uint64_t sectors = 0;
|
1375 |
int flags = 0; |
1376 |
size_t cluster_size = DEFAULT_CLUSTER_SIZE; |
1377 |
int prealloc = 0; |
1378 |
int version = 2; |
1379 |
|
1380 |
/* Read out options */
|
1381 |
while (options && options->name) {
|
1382 |
if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
|
1383 |
sectors = options->value.n / 512;
|
1384 |
} else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) { |
1385 |
backing_file = options->value.s; |
1386 |
} else if (!strcmp(options->name, BLOCK_OPT_BACKING_FMT)) { |
1387 |
backing_fmt = options->value.s; |
1388 |
} else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) { |
1389 |
flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
|
1390 |
} else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) { |
1391 |
if (options->value.n) {
|
1392 |
cluster_size = options->value.n; |
1393 |
} |
1394 |
} else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) { |
1395 |
if (!options->value.s || !strcmp(options->value.s, "off")) { |
1396 |
prealloc = 0;
|
1397 |
} else if (!strcmp(options->value.s, "metadata")) { |
1398 |
prealloc = 1;
|
1399 |
} else {
|
1400 |
fprintf(stderr, "Invalid preallocation mode: '%s'\n",
|
1401 |
options->value.s); |
1402 |
return -EINVAL;
|
1403 |
} |
1404 |
} else if (!strcmp(options->name, BLOCK_OPT_COMPAT_LEVEL)) { |
1405 |
if (!options->value.s || !strcmp(options->value.s, "0.10")) { |
1406 |
version = 2;
|
1407 |
} else if (!strcmp(options->value.s, "1.1")) { |
1408 |
version = 3;
|
1409 |
} else {
|
1410 |
fprintf(stderr, "Invalid compatibility level: '%s'\n",
|
1411 |
options->value.s); |
1412 |
return -EINVAL;
|
1413 |
} |
1414 |
} else if (!strcmp(options->name, BLOCK_OPT_LAZY_REFCOUNTS)) { |
1415 |
flags |= options->value.n ? BLOCK_FLAG_LAZY_REFCOUNTS : 0;
|
1416 |
} |
1417 |
options++; |
1418 |
} |
1419 |
|
1420 |
if (backing_file && prealloc) {
|
1421 |
fprintf(stderr, "Backing file and preallocation cannot be used at "
|
1422 |
"the same time\n");
|
1423 |
return -EINVAL;
|
1424 |
} |
1425 |
|
1426 |
if (version < 3 && (flags & BLOCK_FLAG_LAZY_REFCOUNTS)) { |
1427 |
fprintf(stderr, "Lazy refcounts only supported with compatibility "
|
1428 |
"level 1.1 and above (use compat=1.1 or greater)\n");
|
1429 |
return -EINVAL;
|
1430 |
} |
1431 |
|
1432 |
return qcow2_create2(filename, sectors, backing_file, backing_fmt, flags,
|
1433 |
cluster_size, prealloc, options, version); |
1434 |
} |
1435 |
|
1436 |
static int qcow2_make_empty(BlockDriverState *bs) |
1437 |
{ |
1438 |
#if 0
|
1439 |
/* XXX: not correct */
|
1440 |
BDRVQcowState *s = bs->opaque;
|
1441 |
uint32_t l1_length = s->l1_size * sizeof(uint64_t);
|
1442 |
int ret;
|
1443 |
|
1444 |
memset(s->l1_table, 0, l1_length);
|
1445 |
if (bdrv_pwrite(bs->file, s->l1_table_offset, s->l1_table, l1_length) < 0)
|
1446 |
return -1;
|
1447 |
ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length);
|
1448 |
if (ret < 0)
|
1449 |
return ret;
|
1450 |
|
1451 |
l2_cache_reset(bs);
|
1452 |
#endif
|
1453 |
return 0; |
1454 |
} |
1455 |
|
1456 |
static coroutine_fn int qcow2_co_write_zeroes(BlockDriverState *bs, |
1457 |
int64_t sector_num, int nb_sectors)
|
1458 |
{ |
1459 |
int ret;
|
1460 |
BDRVQcowState *s = bs->opaque; |
1461 |
|
1462 |
/* Emulate misaligned zero writes */
|
1463 |
if (sector_num % s->cluster_sectors || nb_sectors % s->cluster_sectors) {
|
1464 |
return -ENOTSUP;
|
1465 |
} |
1466 |
|
1467 |
/* Whatever is left can use real zero clusters */
|
1468 |
qemu_co_mutex_lock(&s->lock); |
1469 |
ret = qcow2_zero_clusters(bs, sector_num << BDRV_SECTOR_BITS, |
1470 |
nb_sectors); |
1471 |
qemu_co_mutex_unlock(&s->lock); |
1472 |
|
1473 |
return ret;
|
1474 |
} |
1475 |
|
1476 |
static coroutine_fn int qcow2_co_discard(BlockDriverState *bs, |
1477 |
int64_t sector_num, int nb_sectors)
|
1478 |
{ |
1479 |
int ret;
|
1480 |
BDRVQcowState *s = bs->opaque; |
1481 |
|
1482 |
qemu_co_mutex_lock(&s->lock); |
1483 |
ret = qcow2_discard_clusters(bs, sector_num << BDRV_SECTOR_BITS, |
1484 |
nb_sectors); |
1485 |
qemu_co_mutex_unlock(&s->lock); |
1486 |
return ret;
|
1487 |
} |
1488 |
|
1489 |
static int qcow2_truncate(BlockDriverState *bs, int64_t offset) |
1490 |
{ |
1491 |
BDRVQcowState *s = bs->opaque; |
1492 |
int64_t new_l1_size; |
1493 |
int ret;
|
1494 |
|
1495 |
if (offset & 511) { |
1496 |
error_report("The new size must be a multiple of 512");
|
1497 |
return -EINVAL;
|
1498 |
} |
1499 |
|
1500 |
/* cannot proceed if image has snapshots */
|
1501 |
if (s->nb_snapshots) {
|
1502 |
error_report("Can't resize an image which has snapshots");
|
1503 |
return -ENOTSUP;
|
1504 |
} |
1505 |
|
1506 |
/* shrinking is currently not supported */
|
1507 |
if (offset < bs->total_sectors * 512) { |
1508 |
error_report("qcow2 doesn't support shrinking images yet");
|
1509 |
return -ENOTSUP;
|
1510 |
} |
1511 |
|
1512 |
new_l1_size = size_to_l1(s, offset); |
1513 |
ret = qcow2_grow_l1_table(bs, new_l1_size, true);
|
1514 |
if (ret < 0) { |
1515 |
return ret;
|
1516 |
} |
1517 |
|
1518 |
/* write updated header.size */
|
1519 |
offset = cpu_to_be64(offset); |
1520 |
ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size), |
1521 |
&offset, sizeof(uint64_t));
|
1522 |
if (ret < 0) { |
1523 |
return ret;
|
1524 |
} |
1525 |
|
1526 |
s->l1_vm_state_index = new_l1_size; |
1527 |
return 0; |
1528 |
} |
1529 |
|
1530 |
/* XXX: put compressed sectors first, then all the cluster aligned
|
1531 |
tables to avoid losing bytes in alignment */
|
1532 |
static int qcow2_write_compressed(BlockDriverState *bs, int64_t sector_num, |
1533 |
const uint8_t *buf, int nb_sectors) |
1534 |
{ |
1535 |
BDRVQcowState *s = bs->opaque; |
1536 |
z_stream strm; |
1537 |
int ret, out_len;
|
1538 |
uint8_t *out_buf; |
1539 |
uint64_t cluster_offset; |
1540 |
|
1541 |
if (nb_sectors == 0) { |
1542 |
/* align end of file to a sector boundary to ease reading with
|
1543 |
sector based I/Os */
|
1544 |
cluster_offset = bdrv_getlength(bs->file); |
1545 |
cluster_offset = (cluster_offset + 511) & ~511; |
1546 |
bdrv_truncate(bs->file, cluster_offset); |
1547 |
return 0; |
1548 |
} |
1549 |
|
1550 |
if (nb_sectors != s->cluster_sectors) {
|
1551 |
ret = -EINVAL; |
1552 |
|
1553 |
/* Zero-pad last write if image size is not cluster aligned */
|
1554 |
if (sector_num + nb_sectors == bs->total_sectors &&
|
1555 |
nb_sectors < s->cluster_sectors) { |
1556 |
uint8_t *pad_buf = qemu_blockalign(bs, s->cluster_size); |
1557 |
memset(pad_buf, 0, s->cluster_size);
|
1558 |
memcpy(pad_buf, buf, nb_sectors * BDRV_SECTOR_SIZE); |
1559 |
ret = qcow2_write_compressed(bs, sector_num, |
1560 |
pad_buf, s->cluster_sectors); |
1561 |
qemu_vfree(pad_buf); |
1562 |
} |
1563 |
return ret;
|
1564 |
} |
1565 |
|
1566 |
out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128); |
1567 |
|
1568 |
/* best compression, small window, no zlib header */
|
1569 |
memset(&strm, 0, sizeof(strm)); |
1570 |
ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION, |
1571 |
Z_DEFLATED, -12,
|
1572 |
9, Z_DEFAULT_STRATEGY);
|
1573 |
if (ret != 0) { |
1574 |
ret = -EINVAL; |
1575 |
goto fail;
|
1576 |
} |
1577 |
|
1578 |
strm.avail_in = s->cluster_size; |
1579 |
strm.next_in = (uint8_t *)buf; |
1580 |
strm.avail_out = s->cluster_size; |
1581 |
strm.next_out = out_buf; |
1582 |
|
1583 |
ret = deflate(&strm, Z_FINISH); |
1584 |
if (ret != Z_STREAM_END && ret != Z_OK) {
|
1585 |
deflateEnd(&strm); |
1586 |
ret = -EINVAL; |
1587 |
goto fail;
|
1588 |
} |
1589 |
out_len = strm.next_out - out_buf; |
1590 |
|
1591 |
deflateEnd(&strm); |
1592 |
|
1593 |
if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
|
1594 |
/* could not compress: write normal cluster */
|
1595 |
ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors); |
1596 |
if (ret < 0) { |
1597 |
goto fail;
|
1598 |
} |
1599 |
} else {
|
1600 |
cluster_offset = qcow2_alloc_compressed_cluster_offset(bs, |
1601 |
sector_num << 9, out_len);
|
1602 |
if (!cluster_offset) {
|
1603 |
ret = -EIO; |
1604 |
goto fail;
|
1605 |
} |
1606 |
cluster_offset &= s->cluster_offset_mask; |
1607 |
BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED); |
1608 |
ret = bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len); |
1609 |
if (ret < 0) { |
1610 |
goto fail;
|
1611 |
} |
1612 |
} |
1613 |
|
1614 |
ret = 0;
|
1615 |
fail:
|
1616 |
g_free(out_buf); |
1617 |
return ret;
|
1618 |
} |
1619 |
|
1620 |
static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs) |
1621 |
{ |
1622 |
BDRVQcowState *s = bs->opaque; |
1623 |
int ret;
|
1624 |
|
1625 |
qemu_co_mutex_lock(&s->lock); |
1626 |
ret = qcow2_cache_flush(bs, s->l2_table_cache); |
1627 |
if (ret < 0) { |
1628 |
qemu_co_mutex_unlock(&s->lock); |
1629 |
return ret;
|
1630 |
} |
1631 |
|
1632 |
if (qcow2_need_accurate_refcounts(s)) {
|
1633 |
ret = qcow2_cache_flush(bs, s->refcount_block_cache); |
1634 |
if (ret < 0) { |
1635 |
qemu_co_mutex_unlock(&s->lock); |
1636 |
return ret;
|
1637 |
} |
1638 |
} |
1639 |
qemu_co_mutex_unlock(&s->lock); |
1640 |
|
1641 |
return 0; |
1642 |
} |
1643 |
|
1644 |
static int64_t qcow2_vm_state_offset(BDRVQcowState *s)
|
1645 |
{ |
1646 |
return (int64_t)s->l1_vm_state_index << (s->cluster_bits + s->l2_bits);
|
1647 |
} |
1648 |
|
1649 |
static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) |
1650 |
{ |
1651 |
BDRVQcowState *s = bs->opaque; |
1652 |
bdi->cluster_size = s->cluster_size; |
1653 |
bdi->vm_state_offset = qcow2_vm_state_offset(s); |
1654 |
return 0; |
1655 |
} |
1656 |
|
1657 |
#if 0
|
1658 |
static void dump_refcounts(BlockDriverState *bs)
|
1659 |
{
|
1660 |
BDRVQcowState *s = bs->opaque;
|
1661 |
int64_t nb_clusters, k, k1, size;
|
1662 |
int refcount;
|
1663 |
|
1664 |
size = bdrv_getlength(bs->file);
|
1665 |
nb_clusters = size_to_clusters(s, size);
|
1666 |
for(k = 0; k < nb_clusters;) {
|
1667 |
k1 = k;
|
1668 |
refcount = get_refcount(bs, k);
|
1669 |
k++;
|
1670 |
while (k < nb_clusters && get_refcount(bs, k) == refcount)
|
1671 |
k++;
|
1672 |
printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount,
|
1673 |
k - k1);
|
1674 |
}
|
1675 |
}
|
1676 |
#endif
|
1677 |
|
1678 |
static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, |
1679 |
int64_t pos) |
1680 |
{ |
1681 |
BDRVQcowState *s = bs->opaque; |
1682 |
int growable = bs->growable;
|
1683 |
int ret;
|
1684 |
|
1685 |
BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE); |
1686 |
bs->growable = 1;
|
1687 |
ret = bdrv_pwritev(bs, qcow2_vm_state_offset(s) + pos, qiov); |
1688 |
bs->growable = growable; |
1689 |
|
1690 |
return ret;
|
1691 |
} |
1692 |
|
1693 |
static int qcow2_load_vmstate(BlockDriverState *bs, uint8_t *buf, |
1694 |
int64_t pos, int size)
|
1695 |
{ |
1696 |
BDRVQcowState *s = bs->opaque; |
1697 |
int growable = bs->growable;
|
1698 |
int ret;
|
1699 |
|
1700 |
BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD); |
1701 |
bs->growable = 1;
|
1702 |
ret = bdrv_pread(bs, qcow2_vm_state_offset(s) + pos, buf, size); |
1703 |
bs->growable = growable; |
1704 |
|
1705 |
return ret;
|
1706 |
} |
1707 |
|
1708 |
static QEMUOptionParameter qcow2_create_options[] = {
|
1709 |
{ |
1710 |
.name = BLOCK_OPT_SIZE, |
1711 |
.type = OPT_SIZE, |
1712 |
.help = "Virtual disk size"
|
1713 |
}, |
1714 |
{ |
1715 |
.name = BLOCK_OPT_COMPAT_LEVEL, |
1716 |
.type = OPT_STRING, |
1717 |
.help = "Compatibility level (0.10 or 1.1)"
|
1718 |
}, |
1719 |
{ |
1720 |
.name = BLOCK_OPT_BACKING_FILE, |
1721 |
.type = OPT_STRING, |
1722 |
.help = "File name of a base image"
|
1723 |
}, |
1724 |
{ |
1725 |
.name = BLOCK_OPT_BACKING_FMT, |
1726 |
.type = OPT_STRING, |
1727 |
.help = "Image format of the base image"
|
1728 |
}, |
1729 |
{ |
1730 |
.name = BLOCK_OPT_ENCRYPT, |
1731 |
.type = OPT_FLAG, |
1732 |
.help = "Encrypt the image"
|
1733 |
}, |
1734 |
{ |
1735 |
.name = BLOCK_OPT_CLUSTER_SIZE, |
1736 |
.type = OPT_SIZE, |
1737 |
.help = "qcow2 cluster size",
|
1738 |
.value = { .n = DEFAULT_CLUSTER_SIZE }, |
1739 |
}, |
1740 |
{ |
1741 |
.name = BLOCK_OPT_PREALLOC, |
1742 |
.type = OPT_STRING, |
1743 |
.help = "Preallocation mode (allowed values: off, metadata)"
|
1744 |
}, |
1745 |
{ |
1746 |
.name = BLOCK_OPT_LAZY_REFCOUNTS, |
1747 |
.type = OPT_FLAG, |
1748 |
.help = "Postpone refcount updates",
|
1749 |
}, |
1750 |
{ NULL }
|
1751 |
}; |
1752 |
|
1753 |
static BlockDriver bdrv_qcow2 = {
|
1754 |
.format_name = "qcow2",
|
1755 |
.instance_size = sizeof(BDRVQcowState),
|
1756 |
.bdrv_probe = qcow2_probe, |
1757 |
.bdrv_open = qcow2_open, |
1758 |
.bdrv_close = qcow2_close, |
1759 |
.bdrv_reopen_prepare = qcow2_reopen_prepare, |
1760 |
.bdrv_create = qcow2_create, |
1761 |
.bdrv_co_is_allocated = qcow2_co_is_allocated, |
1762 |
.bdrv_set_key = qcow2_set_key, |
1763 |
.bdrv_make_empty = qcow2_make_empty, |
1764 |
|
1765 |
.bdrv_co_readv = qcow2_co_readv, |
1766 |
.bdrv_co_writev = qcow2_co_writev, |
1767 |
.bdrv_co_flush_to_os = qcow2_co_flush_to_os, |
1768 |
|
1769 |
.bdrv_co_write_zeroes = qcow2_co_write_zeroes, |
1770 |
.bdrv_co_discard = qcow2_co_discard, |
1771 |
.bdrv_truncate = qcow2_truncate, |
1772 |
.bdrv_write_compressed = qcow2_write_compressed, |
1773 |
|
1774 |
.bdrv_snapshot_create = qcow2_snapshot_create, |
1775 |
.bdrv_snapshot_goto = qcow2_snapshot_goto, |
1776 |
.bdrv_snapshot_delete = qcow2_snapshot_delete, |
1777 |
.bdrv_snapshot_list = qcow2_snapshot_list, |
1778 |
.bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp, |
1779 |
.bdrv_get_info = qcow2_get_info, |
1780 |
|
1781 |
.bdrv_save_vmstate = qcow2_save_vmstate, |
1782 |
.bdrv_load_vmstate = qcow2_load_vmstate, |
1783 |
|
1784 |
.bdrv_change_backing_file = qcow2_change_backing_file, |
1785 |
|
1786 |
.bdrv_invalidate_cache = qcow2_invalidate_cache, |
1787 |
|
1788 |
.create_options = qcow2_create_options, |
1789 |
.bdrv_check = qcow2_check, |
1790 |
}; |
1791 |
|
1792 |
static void bdrv_qcow2_init(void) |
1793 |
{ |
1794 |
bdrv_register(&bdrv_qcow2); |
1795 |
} |
1796 |
|
1797 |
block_init(bdrv_qcow2_init); |